Subjective Wellbeing as an Affective Construct: Theory
Development and Construction with Adolescents
By
Adrian John Tomyn Hons (Psych)
Submitted in fulfillment of the requirements for the award of
Doctor of Philosophy
Deakin University
February 2008
DEAKIN UNIVERSITY
CANDIDATE DECLARATION
I certify that this thesis entitled: Subjective Wellbeing as an Affective Construct:
Theory Development and Construction With Adolescents
submitted for the degree of: Doctor of Philosophy
Is the result of my own work and that where reference is made to the work of
others, due acknowledgment is given.
I also certify that any material in the thesis which has been accepted for a degree
or diploma by any university or institution is identified in the text.
Full Name..................................................................................................................
(Please Print)
Signed.........................................................................................................................
Date………………………………………………………………………………….
Acknowledgements
First and foremost I would like to thank my mentor, Professor Robert Cummins,
for three wonderful years of supervision. It was an honor and a privilege to work
under the guidance of such a passionate and devoted academic and I thank him for
all of the hard work and commitment he happily put into nurturing my academic
and professional development. Thanks also to my associate supervisor Professor
Mark Stokes for being such a great teacher, role model and friend. Thanks Mark
for always being there when I needed you. Many thanks to Ann-marie James for
all her assistance ‘behind-the-scenes’. Ann-marie was a great help and I
appreciate all her support. To my friends, in particular, my fellow PhD’s Jed, Matt,
Johann, Tilsa and Jaqi. My fondest memories will be of the good times and
laughter we shared together. And to my brothers, Luke and Justin, thanks for all
of your unconditional love, friendship and understanding. Finally, I would like to
thank my beautiful mother Chriatiane Tomyn for all the kindness, support,
encouragement and love that she has given me. Without you mum, I may not have
made it this far.
i
TABLE OF CONTENTS
EXECUTIVE SUMMARY ................................................................................... 1
CHAPTER 1: IMPORTANT ISSUES IN SWB RESEARCH .......................... 4
Conceptualising Affect: Russell’s Distinction Between Mood and
Emotion ................................................................................................ 4
Russell’s Circumplex Model of Affect ................................................. 6
Evidence for Russell’s Circumplex Model of Affect ........................... 7
Revisiting the Circumplex Model of Affect: An Unresolved Issue ... 10
Empirical Demonstration of the Importance of Core Affect to SWB
Studies ................................................................................................ 11
Global and Domain Specific Judgments of SWB .............................. 13
Global and Domain Specific Judgments of SWB: The Present View 16
THEORETICAL EXPLANATIONS FOR SWB.................................................. 18
Bottom-up versus top-down influences on SWB ............................... 18
Personality and Subjective Wellbeing ................................................ 18
Multiple Discrepancies Theory and Subjective Wellbeing ................ 20
The Role of Comparative Devices in SWB and Criticisms of MDT . 21
CHAPTER 2: STABILITY IN LIFE SATISFACTION JUDGEMENTS ..... 23
TOWARDS A HOMEOSTATIC MODEL OF SUBJECTIVE WELLBEING ... 23
Headey and Wearing’s Longitudinal Data ......................................... 23
Population and Individual Normative Ranges of Subjective
Wellbeing ........................................................................................... 25
Overlap Between the Western and Global Range of Subjective
Wellbeing ........................................................................................... 26
The Relationship Between Means and Standard Deviations .............. 28
Distribution of Normative Life Satisfaction Scores ........................... 29
Defeat of the 70%SM Resistance Line ............................................... 30
How Much Variation Does the System Allow? ................................. 32
Adaptation Level Theory .................................................................... 32
Adaptation: A Two-way Process ........................................................ 33
Adaptation and Subjective Wellbeing ................................................ 34
Adaptation and Subjective Wellbeing Set-points ............................... 37
Relationship Between Objective and Subjective Quality of Life ....... 37
Objective and Subjective Wellbeing and SWB Homeostasis ............ 38
Homeostatic Buffers ........................................................................... 41
External Buffers.................................................................................. 41
Income and Subjective Wellbeing: A Closer Look at the Interaction
between Person, Environment and SWB............................................ 41
Internal Buffers ................................................................................... 45
The Cognitive Buffer System and Positive Cognitive Biases
(PCB’s) ............................................................................................... 46
Biases of Self-esteem ......................................................................... 46
ii
Biases of Control ................................................................................ 48
Biases of Optimism ............................................................................ 51
SWB HOMEOSTASIS: A WORKING MODEL ................................................. 52
Set-point Robustness, Fragility and Depression ................................. 54
SUMMARY AND STUDY ONE AIMS .............................................................. 57
CHAPTER 3: STUDY ONE ............................................................................... 59
METHODOLOGY ................................................................................................ 59
Participants ......................................................................................... 59
Questionnaire ...................................................................................... 59
3.1
Major Dependent Variable and Other Variables ................................ 59
Life Satisfaction (LS) and Subjective Wellbeing (SWB) .... 59
Core Affect ........................................................................... 60
Self-esteem ........................................................................... 61
Optimism .............................................................................. 61
Perceived Control ................................................................. 61
Personality ............................................................................ 62
OTHER MEASURES ........................................................................................... 62
School Satisfaction ............................................................................. 62
Multiple Discrepancies Theory .......................................................... 62
PROCEDURE ....................................................................................................... 63
RESULTS .............................................................................................................. 64
3.2
Data Screening and Preliminary Analyses ......................................... 64
3.2.1
Missing Data ........................................................................ 64
3.2.2
Outliers ................................................................................. 65
3.2.3
Normality and Linearity ....................................................... 65
3.2.4
Multicollinearity and Singularity ......................................... 65
3.2.5
Sample Size .......................................................................... 66
3.3
Core Affective Adjectives as Predictors of SWB............................... 66
3.4
Further Analyses Using Core Affect Adjectives as Predictors of LS
and SWB ............................................................................................. 73
3.5
Predicting LS and SWB above Core Affect using the Buffer
Variables ............................................................................................. 76
3.5.1
Predicting LS using Core Affect and the Buffer Variables .. 76
3.6
Predicting SWB Using MDT.............................................................. 80
3.7
Predicting SWB with Personality ....................................................... 83
3.8
Predicting SWB Using Normative Data Divisions: Further
Analyses ............................................................................................. 84
3.8.1
Predicting SWB using cases > 70%SM and <70%SM ........ 85
iii
3.8.2
3.9
Predicting SWB using cases Between 45 and 69%SM ........ 90
Predicting SWB using MDT: Further Analyses Using the
Adolescent Sample ............................................................................. 92
3.10 Predicting SWB Using Normative Data Divisions: Conclusions....... 95
3.11 Comparative Analysis Against an Adult Data Set ............................. 95
Method ................................................................................................ 95
Participants ........................................................................... 95
Measures ............................................................................... 96
Data Screening and Preliminary Analysis ............................ 96
Results ................................................................................................ 96
3.12 Predicting SWB using MDT: Further Analyses Using the
Comparative Adult Data Set ............................................................. 103
3.12.1
Comparative Analysis Against an Adult Data: Summary
of Results ............................................................................ 106
3.13 Exploratory Analyses: Satisfaction with School as a Unique
Construct........................................................................................... 107
CHAPTER 4: STUDY 1 DISCUSSION .......................................................... 109
4.1
Hypothesis one: That the mean score for LS will approximate the
mean score for SWB and that the mean score for SWB will lie
within the Australian adult normative range .................................... 109
4.2
Hypotheses two and three: That adjectives located on the
pleasantness-unpleasantness axis of the Circumplex Model of
Affect will dominate and explain significant variance in LS and
SWB and that core affect will explain greater unique variance in
LS than in SWB ................................................................................ 111
4.3
Hypothesis four: That the buffer variables will explain significant
variance in SWB above core affect in a group of people
experiencing homeostatic challenge ................................................. 113
4.4
Exploratory analysis: Predicting SWB using cases between 45 and
69%SM ............................................................................................. 114
4.5
Hypothesis five: That core affect is driving the relationship
between SWB and related variables. ................................................ 116
4.6
Exploratory analyses: Personality and MDT as predictors of SWB 117
4.7
Comparative analyses against an adult data Set ............................... 118
4.8
Exploratory analyses: School satisfaction as a unique construct ..... 120
4.9
Summary........................................................................................... 123
4.10 Conclusions ...................................................................................... 125
CHAPTER 5: STUDY 2.................................................................................... 127
INTRODUCTION ............................................................................................... 127
iv
METHODOLOGY .............................................................................................. 128
Participants ....................................................................................... 128
Questionnaire .................................................................................... 128
5.1
Major Dependent Variable and Other Variables .............................. 128
5.2
Other Measures: School Index ......................................................... 128
5.3
Procedure .......................................................................................... 129
5.4
Data Analytic Strategy ..................................................................... 129
Assessing Goodness of Fit: Which Statistics to Interpret ................ 130
RESULTS ............................................................................................................ 131
5.5
5.6
Data Screening and Preliminary Analyses ....................................... 131
5.5.1
Missing Data ...................................................................... 131
5.5.2
Outliers ............................................................................... 131
5.5.3
Normality and Linearity ..................................................... 131
5.5.4
Multicollinearity and Singularity ....................................... 132
5.5.5
Sample Size ........................................................................ 132
Core Affective Adjectives as Predictors of SWB............................. 133
5.6.1
5.7
Further Analyses Using Core Affect Adjectives as
Predictors of LS and SWB ................................................. 142
Predicting LS and SWB above Core Affect using the Buffer
Variables ........................................................................................... 144
5.7.1
Predicting LS using Core Affect and the Buffer Variables 145
5.7.2
Predicting SWB using Core Affect and the Buffer
Variables............................................................................. 146
5.8
Predicting SWB Using MDT............................................................ 148
5.9
Predicting SWB with Personality ..................................................... 155
5.10 Predicting SWB Using Normative Data Divisions: Further
Analyses ........................................................................................... 157
5.10.1
Predicting SWB using cases between 45 and 69 ................ 157
5.11 Model Testing: Evaluating Model Fit .............................................. 163
5.11.1
An Affective-Cognitive Model .......................................... 164
5.11.2
A Personality Model for SWB ........................................... 167
5.11.3
The Multiple Discrepancies Theory Model ....................... 170
5.12 A Comparison of the SEM models ................................................... 173
5.13 Exploratory Analyses: School Satisfaction as a Unique Construct .. 174
5.14 Satisfaction with School as a Unique Construct: Exploratory
Analysis Using Combined Data ....................................................... 177
5.15 Additional Exploratory Analyses: Predicting School Satisfaction ... 179
v
5.16 Study 2 Results: Summary ............................................................... 181
CHAPTER 6: STUDY 2 DISCUSSION .......................................................... 182
6.1
Hypothesis one: That the mean score for LS will approximate the
mean score for SWB and that the mean score for SWB will lie
within the Australian adult normative range .................................... 182
6.2
Hypotheses two and three: That adjectives happy, content and alert
will explain significant variance in SWB and that core affect will
explain greater unique variance in LS than in SWB when the
system is functioning normally ........................................................ 183
6.3
Hypothesis four: That in circumstances of homeostatic challenge,
but not in homeostatic rest, primary control and secondary control
will explain variance in SWB beyond core affect ............................ 186
6.4
Hypothesis five: That core affect is driving the relationship
between SWB and related variables. ................................................ 189
6.5
Exploratory analyses: Personality and MDT as predictors of SWB 189
6.6
Model testing using SEM in AMOS ................................................ 192
6.7
Hypothesis six: That in confirmation of study 1, only the three
domains of health, safety and relationships will contribute unique
variance to LS. .................................................................................. 193
6.8
Exploratory analyses: Satisfaction with school as a unique
construct ........................................................................................... 195
6.9
Exploratory analyses: Predicting satisfaction with school ............... 196
6.10 Summary and Conclusions ............................................................... 198
CHAPTER 7: A RE-ANALYSES OF PUBLISHED RESEARCH
INVESTIGATING THE IMPORTANCE OF AFFECT TO
MEASURES OF SUBJECTIVE WELLBEING ................................... 202
INTRODUCTION ............................................................................................... 202
General Method ................................................................................ 202
Conversion of Data to Percentage of Scale Maximum (%SM) ........ 203
REANALYSIS ONE: HEADEY AND WEARING (1989) ............................... 205
METHODOLOGY .............................................................................................. 207
Participants ....................................................................................... 207
7.1
Major Dependent Variable and Other Variables .............................. 208
Subjective Wellbeing Headey & Wearing (SWB-HW) ..... 208
Affect .................................................................................. 208
Personality .......................................................................... 208
Life Events ......................................................................... 209
RESULTS ............................................................................................................ 209
7.2
Headey and Wearing’s Dynamic Equilibrium Model ...................... 209
vi
7.3
Re-analysis Incorporating Affect ..................................................... 213
7.4
Summary........................................................................................... 217
REANALYSIS TWO: VITTERSO (2001) ......................................................... 218
METHODOLOGY .............................................................................................. 220
Participants ....................................................................................... 220
7.5
Major Dependent Variable and Other Variables .............................. 220
Overall Life Satisfaction (LS) ............................................ 220
Personality .......................................................................... 221
Positive Affect .................................................................... 221
RESULTS ............................................................................................................ 221
7.6
Re-constructing Vitterso’s Model (T1) ............................................ 222
7.7
Re-analysis Incorporating Affect (T1 data) ...................................... 223
7.8
A Comparison of the fit Statistics for the Personality and
Affectively-Driven Models of SWL (T1 data) ................................. 224
7.9
Summary of Model Comparisons (T1)............................................. 225
7.10 Re-constructing Vitterso’s Model (T2 Data) .................................... 226
7.11 Re-analysis Incorporating Affect (T2 data) ...................................... 226
7.12 A Comparison of the fit Statistics for the Personality and
Affectively Driven Models of SWL (T2 data) ................................. 227
7.13 Summary of Model Comparisons (T2)............................................. 228
7.14 Summary........................................................................................... 228
REANALYSIS THREE: VITTERSO AND NILSEN (2002) ............................ 230
METHODOLOGY .............................................................................................. 231
Participants ....................................................................................... 231
7.15 Major Dependent Variable and Other Variables .............................. 232
Satisfaction with Life (SWL) ............................................. 232
Personality .......................................................................... 232
Positive Affect .................................................................... 232
Negative Affect .................................................................. 233
RESULTS ............................................................................................................ 233
7.16 Vitterso and Nilsen’s Model ............................................................. 233
7.17 Re-analysis Incorporating Affect ..................................................... 237
7.18 A Comparison of fit Statistics for the Personality and Affectivelydriven Models of SWL ..................................................................... 238
7.19 Summary........................................................................................... 239
REANALYSIS FOUR: LIBRAN (2006) ............................................................ 240
METHODOLOGY .............................................................................................. 242
vii
Participants ....................................................................................... 242
7.20 Major Dependent Variable and Other Variables .............................. 242
Overall Life Satisfaction (SWL) ........................................ 242
Personality .......................................................................... 242
Positive and Negative Affect .............................................. 242
RESULTS ............................................................................................................ 243
7.21 A Personality-driven Model of Satisfaction With Life .................... 243
7.22 Re-analysis Incorporating Affect ..................................................... 245
7.23 A Comparison of Model fit Statistics for the Personality and
Affectively-driven Models of SWL.................................................. 246
7.24 Summary........................................................................................... 249
REANALYSIS FIVE: ZHENG, SANG AND LIN (2004) ................................. 251
METHODOLOGY .............................................................................................. 252
Participants ....................................................................................... 252
7.25 Major Dependent Variable and Other Variables .............................. 253
Overall Life Satisfaction (LS) ............................................ 253
Personality .......................................................................... 253
Positive Affect .................................................................... 253
Negative Affect .................................................................. 253
Cognitive and Social Orientations...................................... 254
RESULTS ............................................................................................................ 254
7.26 Zheng et al’s., Model of Subjective Wellbeing ................................ 254
7.27 Re-analysis Incorporating Affect ..................................................... 257
7.28 A Comparison of the Fit Statistics for all Three Models of
Satisfaction with Life ....................................................................... 258
7.29 Summary........................................................................................... 260
CHAPTER 8: STUDY 3 DISCUSSION .......................................................... 261
CHAPTER 9: SUMMARY AND OVERVIEW ............................................. 264
Theoretical Overview and Hypotheses ............................................. 264
Methodology..................................................................................... 266
Results and Implications................................................................... 267
Normative Data .................................................................. 267
Core Affect ......................................................................... 267
Homeostasis Theory ........................................................... 268
The Relationship Between SWB and Related Variables.... 268
Model Testing Using Structural Equation Modelling in
AMOS ................................................................................ 270
School Satisfaction as a Unique Construct ........................ 270
Re-analyses of Past Research Papers ................................. 271
Directions for Future Research ......................................................... 272
viii
Conclusion ........................................................................................ 273
REFERENCES .................................................................................................. 274
APPENDIX A: YOUNG AUSTRALIAN WELLBEING INDEX (STUDY
1 QUESTIONNAIRE) ............................................................................. 290
APPENDIX B: YOUNG AUSTRALIAN WELLBEING INDEX (STUDY
2 QUESTIONNAIRE) ............................................................................. 293
ix
INDEX OF FIGURES
Figure 1:
Russell’s Circumplex Model of Affect. ................................................ 7
Figure 2:
The frequency distribution of three hypothetical groups with life
satisfaction mean scores > 70%SM (reprint from Cummins, 2003) .. 30
Figure 3:
The frequency distribution of four hypothetical groups life
satisfaction scores lie slightly above, or below, 70%SM
(reproduced from Cummins, 2002) .................................................... 31
Figure 4:
The relationship between subjective and objective quality of life
(from Cummins, Lau, & Davern, unpublished) ................................. 39
Figure 5: Subjective wellbeing vs. household income (reprint from Cummins,
Lau, Mellor, & Stokes, in press)......................................................... 42
Figure 6:
Income vs. happy and sad events (reprint from Cummins, Lau, &
Davern, unpublished) ......................................................................... 44
Figure 7:
A homeostatic model of subjective wellbeing (reproduced from
Cummins & Nistico, 2002)................................................................. 53
Figure 8:
The relationship between SWB and depression (reprint from
Cummins, Lau, & Davern, unpublished)............................................ 55
Figure 9:
A simplified Affective-Cognitive model of SWB ............................ 164
Figure 10: Affective-Cognitive model of SWB ................................................. 166
Figure 11: SEM of a Personality-driven model of SWB ................................... 169
Figure 12: Multiple discrepancies theory model of SWB.................................. 172
Figure 13: SEM of Headey and Wearing’s Dynamic Equilibrium Model ........ 211
Figure 14: An Affectively-driven model of SWB-HW ..................................... 214
Figure 15: A simplified, Affectively-driven model of SWB-HW ..................... 216
Figure 16: Predicting SWL with personality (Step 1; T1 data) ......................... 223
Figure 17: Predicting SWL with PA above personality (Step 2; T1 data) ......... 224
Figure 18: Predicting SWL with personality (Step 1; T2 data) ......................... 226
Figure 19: Predicting SWL with PA above personality (Step 2; T2 data) ......... 227
Figure 20: A simplified version of Vitterso and Nilsen’s Personality-driven
model of satisfaction with life .......................................................... 236
Figure 21: An Affectively-driven model of satisfaction with life ..................... 237
x
Figure 22: A Personality-driven model of satisfaction with life ........................ 244
Figure 23: An Affectively-driven model of satisfaction with life ..................... 245
Figure 24: An alternate, Affectively-driven model of satisfaction with life (b) 248
Figure 25: Predicting satisfaction with life with personality and present-future
orientation ......................................................................................... 256
Figure 26: An affectively-driven model of satisfaction with Life (a) ................ 257
Figure 27: An Affectively-driven model of satisfaction with life (b) ................ 258
Figure 28: An Affective-model for SWB demonstrating the proposed influence
of core affect on SWB and related variables .................................... 269
xi
INDEX OF TABLES
Table 1:
Means and standard deviations for countries ranked 1-5 .................. 28
Table 2:
Means, standard deviations and correlations between variables
(N=146) .............................................................................................. 68
Table 3:
Predicting LS by nine affect adjectives ............................................. 69
Table 4:
Predicting SWB using nine affective adjectives. .............................. 70
Table 5:
Means, standard deviations and correlations between variables
(n=139 >50%SM) ............................................................................... 71
Table 6:
Predicting LS by nine affect adjectives (n = 139 >50%SM) ............. 72
Table 7:
Predicting SWB using nine affective adjectives (n = 139 >50%SM)
............................................................................................................ 72
Table 8:
Predicting life satisfaction using four affect adjectives ..................... 74
Table 9:
Predicting SWB using five affect adjectives ..................................... 75
Table 10:
Means, standard deviations and correlations between variables ....... 76
Table 11: Predicting LS using core affect and the buffer variables ...................... 77
Table 12:
Means, standard deviations and correlations between variables (n
= 146).................................................................................................. 78
Table 13:
Predicting SWB using core affect and the buffer variables .............. 79
Table 14:
Means, standard deviations and correlations between variables ....... 80
Table 15:
Predicting SWB with MDT ............................................................... 80
Table 16:
Means, standard deviations and correlations between variables (n
= 146).................................................................................................. 81
Table 17:
Predicting SWB with core affect, the buffer variables and MDT ..... 82
Table 18:
Means, standard deviations and correlations between variables ....... 83
Table 19:
Predicting SWB using the personality dimensions of extraversion
and emotional stability ....................................................................... 84
Table 20:
Means, standard deviations and correlations between variables
(SWB > 70%SM) ............................................................................... 86
Table 21:
Means, standard deviations and correlations between variables
(SWB < 70%SM) ............................................................................... 86
xii
Table 22:
Predicting SWB after core affect using the buffer variables (split
Cases) ................................................................................................. 88
Table 23:
Means, SD’s and correlations between variables (SWB between 45
and 69%SM) (n = 41) ......................................................................... 90
Table 24:
Predicting SWB using core affect and the buffer variables for
individuals with SWB between 45% and 69%SM (n = 41) ............... 91
Table 25:
Means, SD’s and correlations between variables .............................. 93
Table 26:
Predicting SWB After Core Affect Using MDT ............................... 94
Table 27:
Means, SD’s and correlations between variables: comparative
analysis (n=387) ................................................................................. 97
Table 28:
Predicting SWB using core affect and the buffer variables:
comparative analysis .......................................................................... 98
Table 29:
Means, SD’s and correlations between variables (SWB > 70%SM)
(n=261) ............................................................................................. 100
Table 30:
Means, SD’s and correlations between variables (SWB Between
45 and 69%SM) (n=261) .................................................................. 100
Table 31:
Predicting SWB using core affect and the buffer variables for
adults with SWB between 45 and 69%SM....................................... 102
Table 32:
Means, standard deviations and correlations between variables ..... 103
Table 33:
Predicting SWB after core affect using MDT ................................. 105
Table 34:
Means, standard deviations and correlations between variables ..... 107
Table 35:
Predicting LS using 7 PWI domains and satisfaction with school .. 108
Table 36:
Means, standard deviations and correlations between variables
(N=205) ............................................................................................ 134
Table 37: Comparing study 1 and study 2 mean scores and standard deviations
for LS, SWB and affects ................................................................... 135
Table 38:
A comparison of means, standard deviations and correlations
between variables ............................................................................. 136
Table 39:
Predicting LS by nine affect adjectives ........................................... 137
Table 40:
Predicting SWB using nine affective adjectives. ............................ 138
Table 41:
Means, standard deviations and correlations between variables
(n=191 >50%SM) ............................................................................. 139
Table 42:
Predicting LS by nine affect adjectives (n = 191 >50%SM) ........... 140
xiii
Table 43: Predicting SWB using nine affective adjectives (n = 191 >50%SM)
.......................................................................................................... 141
Table 44:
Predicting life satisfaction using four affect adjectives ................... 142
Table 45:
Predicting SWB using four affect adjectives ................................... 143
Table 46:
Means, standard deviations and correlations between variables
(N=205) ............................................................................................ 145
Table 47:
Predicting LS using core affect and the buffer variables ................ 146
Table 48:
Means, standard deviations and correlations between variables (n
= 205)................................................................................................ 147
Table 49:
Predicting SWB using core Affect and the buffer variables ........... 148
Table 50:
Means, standard deviations and correlations between variables
(N=205) ............................................................................................ 149
Table 51:
Predicting SWB with MDT ............................................................. 149
Table 52:
Means, standard deviations and correlations between variables
(N=205) ............................................................................................ 150
Table 53:
Predicting SWB with MDT ............................................................. 151
Table 54:
Means, standard deviations and correlations between variables (N
= 205)................................................................................................ 153
Table 55:
Predicting SWB with core affect, the buffer variables and MDT ... 154
Table 56:
Means, standard deviations and correlations between variables ..... 155
Table 57:
Predicting SWB using the personality dimensions of extraversion
and emotional stability ..................................................................... 156
Table 58:
Means, standard deviations and correlations between variables
(SWB > 70%SM) (n = 133).............................................................. 158
Table 59:
Means, standard deviations and correlations between variables
(SWB Between 45 & 69%SM) (n=64) ............................................. 158
Table 60:
Predicting SWB after core affect using the buffer variables (split
cases) ................................................................................................ 160
Table 61:
Means, standard deviations and correlations between variables
(SWB Between 45 & 69%SM) (n=105) ........................................... 162
Table 62:
Predicting SWB using core affect and the buffer variables (n=105)163
Table 63:
Absolute and relative fit Indices for the Affective-Cognitive model
of SWB ............................................................................................. 165
xiv
Table 64:
Analysis of an Affective-Cognitive model of SWB (N=205) ......... 167
Table 65:
Absolute and relative fit iIndices for the Personality-Driven model
of SWB ............................................................................................. 168
Table 66:
Analysis of a Personality-driven model of SWB (N=205) .............. 170
Table 67:
Absolute and relative fit indices for MDT model of SWB ............. 171
Table 68:
Analysis of a multiple discrepancies model of SWB (N=205) ....... 173
Table 69:
Summary of absolute and relative fit indices for all models ........... 173
Table 70:
Means, standard deviations and correlations between variables ..... 175
Table 71:
Predicting LS using 7 PWI domains and satisfaction with school .. 176
Table 72:
Means, standard deviations and correlations between variables
(N=351) ............................................................................................ 177
Table 73:
Predicting LS using 7 PWI domains and satisfaction with school .. 178
Table 74:
Means, standard deviations and correlations between variables
(N=205) ............................................................................................ 179
Table 75:
Predicting satisfaction with school .................................................. 180
Table 76:
Means, standard deviations and correlations between variables (N
= 649)................................................................................................ 210
Table 77:
Absolute and relative fit indices for Headey and Wearing’s
Dynamic Equilibrium Model ............................................................ 212
Table 78:
Absolute and relative fit Indices for an Affectively-driven model
SWB-HW ......................................................................................... 215
Table 79:
Absolute and relative fit indices for an Affectively-driven model of
SWB-HW ......................................................................................... 216
Table 80:
Means, standard deviations and correlations between variables ..... 222
Table 81:
A comparison of model fit statistics (T1 data) ................................ 225
Table 82:
A comparison of model fit statistics (T2 data) ................................ 227
Table 83:
Means, standard deviations and correlations between variables (N
= 457)................................................................................................ 234
Table 84:
A comparison of model fit statistics ................................................ 238
Table 85:
Means, standard deviations and correlations between variables
(N=368) ............................................................................................ 244
xv
Table 86:
A comparison of model fit statistics ................................................ 246
Table 87:
A comparison of model fit statistics for all three models ................ 249
Table 88:
Means, standard deviations and correlations between variables (N
= 201)................................................................................................ 255
Table 89:
A comparison of model fit statistics for all three models ................ 259
1
EXECUTIVE SUMMARY
Within the literature, it is generally agreed that Subjective Wellbeing (SWB)
consists of both cognitive and affective components. However, the relationship
between cognition, affect and SWB is not entirely known. A recent study by
Davern, Cummins and Stokes (2007) has provided evidence that SWB is
essentially driven by affect, with cognitive discrepancies playing a significant, but
subsidiary role. The aim of this research is to replicate the Davern et al. study
using data from adolescents.
The first study determined that three affects, which represent the pleasantunpleasant axis and activation-deactivation axis of the Circumplex Model of
Affect, may be considered to represent core affect as initially conceived by
Russell (2003) and used by Davern et al., (2007). Russell defines core affect as a
“neurophysiological state consciously accessible as the simplest raw (nonreflective) feelings evident in moods and emotion” (p. 148). It was found that, the
three affect terms - happy, content and alert - explained 59% of the variability in
SWB. These results generally confirm those of Davern et al.
Further testing supported the notion that core affect also appears to be driving the
relationship between SWB and the related constructs of self-esteem, optimism,
perceived control, multiple discrerpancies theory (MDT; Michalos, 1985) and
personality. This implies that core affect, not personality, is the major cohesive
force underpinning these relationships.
Further analysis provided support for the idea of SWB homeostasis. According to
this theory, in a manner analogous to the homeostatic maintenance of body
temperature, SWB is actively controlled and maintained (Cummins & Nistico,
2002). Additionally, according to this theory, homeostasis acts to protect core
affect (which is argued to approximate the SWB set-point) so that each person can
maintain a positive level of wellbeing. In support of homeostasis theory, core
affect explained more unique variance in global life satisfaction (LS) than SWB
measured using the Personal Wellbeing Index (PWI) because LS is more abstract.
2
However, this effect was only observed when the sample was restricted to people
scoring in the operational range for homeostasis of >50 points on a 0-100 point
scale. This demonstrates the importance of understanding the sample
characteristics when engaging in such investigations.
A second prediction from SWB homeostasis is that, in challenged populations, the
buffer variables (control, self-esteem and optimism) will be activated in an
attempt to maintain and stabilise SWB within the set-point range. This was not
confirmed. Only primary control and secondary control contribute additional
variance above core affect in the homeostatically challenged situation. This may
be because primary and secondary control are relatively more independent of core
affect.
The aim of study 2 was to replicate these findings. The collective results were in
general agreement. Most importantly, the results confirmed that SWB is primarily
an affective construct with minor independent contribution from cognition. In this
second study, the same three adjectives (happy, content and alert) explained 57%
of variance in SWB. Furthermore, consistent with Davern et al., using structural
equation modeling (SEM) in AMOS, it was found that an Affective-Cognitive
model of SWB was the best fitting model, explaining 80% of variance. These
results overwhelmingly support core affect as the major component of SWB, with
MDT playing a significant, but subsidiary role. This again suggests that the
strength of the relationship between personality, SWB and MDT should be
revisited in the presence of suitable affective controls.
The combined data from both studies were also used to show that satisfaction with
school explained unique variance in global life satisfaction above all 7 existing
domains. This qualifies satisfaction with school as a unique construct. Further
analyses revealed that four domains - satisfaction with teachers at school, abilities
at school, safety at school and behaviour at school - predicted almost 50% of the
variance in school satisfaction. Thus, this research has identified a number of
important domains related to the overall school experience.
3
The aim of study 3 was to demonstrate the importance of affect to research in
subjective wellbeing studies generally. This involved a series of re-analyses of
past studies using Structural Equation Modeling (SEM) in AMOS. Results of all
five re-analyses demonstrated that in the presence of affect, the relationship
between SWB and related variables reduced considerably from that reported in the
original publications.
The major implication of this research is that core affect, as initially defined by
Russell (2003), is the driving force behind SWB and may also be responsible for
set-point stability in SWB ratings in Cummins’ homeostatic theory of SWB.
Furthermore, the data suggest that core affect may also be driving the relationship
between personality and SWB. Additionally, since core affect is driving both
personality and SWB, individual differences in set-point levels of core affect may
be causing personality and SWB to correlate.
In summary, this thesis proposes that under normal circumstances, SWB is driven
by core affect. Further, because core affect reflects an enduring and stable person
characteristic and one that is most likely biologically based, biology combines
with psychology to encourage the continued experience of pleasant core affect
across the life-cycle. Through the process of natural selection, it is possible that
humans have evolved to experience a level of core affect that is stable and
positive. Not only does this process provide the motivation for living, it is also
responsible for regulating stability in SWB set-points.
4
CHAPTER 1: IMPORTANT ISSUES IN SWB RESEARCH
Subjective Wellbeing (SWB) is a broad construct that comprises global judgments
of life satisfaction, domain-based satisfactions, and people’s emotional responses
(Diener, Suh, Lucas, & Smith, 1999). Life satisfaction is the term applied to the
response to the question ‘How satisfied are you with your life as a whole’?
Domain-based satisfaction judgments refer to emotional and cogntive judgments
of satisfaction in relation to more specific, identifiable aspects of a person’s life,
such as satisfaction with health. SWB is commonly measured through either
general life satisfaction or the average of domain-based satisfaction. Although
there is general agreement within the literature that SWB comprises both a
cognitive and affective component (e.g., Campbell, Converse, & Rodgers, 1976;
Diener & Diener, 1996; Steel & Ones, 2002; Veenhoven, 1994), the relationship
between cognition, affect and SWB is not well understood. However, recent
findings by Davern, Cummins and Stokes (2007) have suggested that SWB is
essentially driven by affects, with cognitive discrepancies playing an important
but subsidiary role.
Conceptualising Affect: Russell’s Distinction Between Mood and Emotion
Although there are numerous models of affect (e.g., Watson & Tellegen’s TwoFactor model of affect, 1985; Larson & Diener’s Self-reported Affect Circumplex,
1992), the model provided by James A. Russell has provided the most
comprehensive description of this construct. Although it often seems difficult to
discriminate between terms such as mood, emotion and affect, Russell (2003)
makes a distinction in relation to the presence of an object. His work builds on
that of Oatley and Johnson-Laird (1987) in positing that mood and emotion
represent object-less and object-directed versions, respectively, of the same
processes. According to Russell (2003), emotions have objects and are directed at
something or someone, thereby involving cognitive processes. For example,
‘Adrian was happy when all his friends came to his birthday party’. In comparison,
moods, commonly referred to as feelings, are relatively low intensity, diffuse and
enduring affective states that have no object or cause and consequently, little or no
cognitive content (such as simply feeling good or bad). Russell also discriminates
5
mood and emotion from another construct he calls core affect. He defines this as a
“neurophysiological state consciously accessible as the simplest raw (nonreflective) feelings evident in moods and emotion” (Russell, 2003, p. 148). He
further states that core affect as consciously experienced is not cognitive or
reflective, is always in the background, often having no object or cause; and
resides at the core of all emotion-laden occurrent events.
Evidence for Russell’s distinction between mood, emotion and affect is supported
by a series of experiments by Murphy and Zajonc (e.g., Murphy & Zajonc, 1987;
Murphy & Zajonc, 1993). These authors sought to examine whether or not
affective reactions may take place in the absence of cognitive input. They did this
by inducing non-conscious affect and then explored this state using affective
priming. To induce non-conscious affect, participants were shown Chinese
ideographs for a duration of 2 seconds and required to state whether they liked
each of them and whether the ideograph presented represented something good or
bad. These ideographs were affectively primed – that is, they were preceded – by
smiling or frowning faces either for 4msec (below the threshold of awareness) or
for a full second. Faces were used as primes in these trials because previous
research had suggested that faces act as strong interpersonal stimuli that may
cause people to respond positively or negatively even without having full visual
access (e.g., Dimberg, 1982). Murphy and Zajonc used 45 trials with two controls
– no primes at all and neutral primes (e.g., random polygons). The crucial trials
were 20 trials interspersed over the series. On 10 of these, a smiling face preceded
each of the 10 ideographs (different face for each ideograph); and on the other 10,
a frowning face preceded each of the 10 ideographs. In the two sets of trials,
ideographs were the same, so that when comparing the liking for ideographs
following smiling and frowning faces, they were comparing judgments made by
the same participants. Finally, with respect to timing, ideographs were presented
below the threshold of awareness in half these trials; and above the threshold of
awareness in the other half of these trials.
From their experiments, Murphy and Zajonc (1987, 1993) demonstrated that
inducing specific affective experiences outside the realm of consciousness
generated marked changes in mood. That is, their findings indicated clear
6
differences in liking for ideographs preceded by smiling or frowning faces. Those
ideographs preceded by smiling faces were rated considerably more likeable than
those preceded by frowning faces. The significance of this finding is that clear
differences in preference for the ideographs were observed without cognitive
involvement. Importantly, the faces were presented below the threshold of
awareness and therefore, they argue that there was no cognitive processing
associated with the observed changes in mood. Since there was no recognition of
the affective experience, it was concluded that there was no corresponding
cognition to which a person has access. The implication of this is that people have
access to and can report changes in affect, without reference to any preceding
object or event.
In the following section, research will be presented suggesting that self-reported
affect may be best represented as lying along the perimeter of a circle in twodimensional space.
Russell’s Circumplex Model of Affect
Since Wundt’s (1912/1924) early work on emotion, psychologists have been
interested in explaining the underlying structure of affective experience. Of all the
proposed models to emerge over the years, none has received more attention than
the ‘Circumplex Model of Affect’. In 1954, Guttman introduced the term
‘circumplex’ to describe the circular ordering of a data matrix. Although
Guttman’s circumplex was initially employed as a means to order data concerning
mental abilities, the term has since been used by a variety of researchers in order
to explain data gathered from a variety of disciplines. In 1980, Russell revisited
earlier work on affect by Woodworth and Schlosberg (1938) and Schlosberg
(1952), and theorised that affective states are “best represented as a circle in a
two-dimensional space” (Russell, 1980, p.1162). Russell proposed that at any
given moment, conscious experience is a blend of two dimensions of activationdeactivation and pleasure-displeasure. These are describable as a single point on
the perimeter of the circumplex. The vertical axis of the circumplex (arousal)
relates to the strength or intensity of the emotion and is anchored by terms
representing activation at 0o and deactivation at 180o. The horizontal axis of the
circumplex, described by other researchers as hedonic tone (Cropanzano, Weiss,
7
Hale, & Reb, 2003), or pleasantness-unpleasantness (Yik, Russell, & Feldman
Barrett, 1999), is anchored by terms representing pleasure at 90o and displeasure
at 270o. The perimeter of the circle denoting affective space is thought to be
bipolar, with affective antonyms falling 180o apart. The antonym ‘satisfaction’,
for example, is located at the pleasant pole of the pleasant-unpleasant axis
(Remington, Fabrigar, & Visser, 2000). The opposite of satisfaction,
dissatisfaction, is therefore located 180o apart on the unpleasant pole of this axis.
Thus, satisfaction and dissatisfaction assessments in SWB, for example, can be
directly related to the circumplex model. Figure 1 below provides a graphical
representation of Russell’s Circumplex Model of Affect.
ACTIVATION
Tense
Jittery
Excited
Ebullient
Elated
Happy
Upset
Distressed
DISPLEASURE
PLEASURE
Serene
Contented
Sad
Gloomy
Tired
Lethargic
Placid
Calm
DEACTIVATION
Figure 1:
Russell’s Circumplex Model of Affect.
Evidence for Russell’s Circumplex Model of Affect
In an attempt to gain support for his thesis that affective states are best represented
as a circle in a two-dimensional bipolar space, Russell (1980) conducted a series
of experiments examining a) how laypeople conceptualise affective states and b)
multivariate analyses of self-reported affect. The aim of the first type of study was
8
to explore participants’ perceptions of the relationship between affects and affect
categories. As part of this initial work, participants were asked to sort 28 words
chosen to represent the domain of affect into one of the eight categories depicted
in Figure 1. These eight categories or quadrants represent a means by which
affects can be conceptualised according to varying degrees of pleasure-displeasure
and activation-arousal. The 28 stimulus words consisted of words or phrases that
people used to describe their moods, feelings, temporary states, affect or emotions.
As predicted, most emotions were classified as belonging to one of the eight
categories and from this, Russell concluded that his model was a good estimate of
affective states and could account for the majority of affects.
Following this classification task, participants were asked to arrange each of the
eight affect categories around the perimeter of a circle. According to the
instructions, words that were placed opposite each other described opposite
feelings; whilst those placed adjacent to one another would describe feelings that
are most similar. At the conclusion of this task, all eight categories were plotted
on the perimeter of the circle and it was found that modal response positions
generally configured with the expected order. Next, Russell assigned scale
coordinates to each of the 28 emotions, however, he did not explain the procedure
for constructing these polar coordinates. Nonetheless, all 28 affect terms were
found to lie meaningfully along the circle; indicating a high degree of agreement
between participant responses. As anticipated, emotions thought to be bipolar
opposites were located 180o apart, whilst emotions considered similar were
positioned near to one another. According to Russell, results from this study
support his earlier hypothesis about the cognitive structure of affect. More
specifically, that affective space is bipolar with antonyms falling 180o apart. From
this, Russell went on to conclude that the horizontal and vertical axes are
interpretable as the proposed pleasure-displeasure and degree of arousal
dimensions and that affective space lacks ‘simple structure’. That is, instead of
clusters of synonyms congregating near the axes, these synonyms more or less
spread out around the perimeter of the circle in a continuous fashion.
To determine the minimum number of dimensions that could most adequately
describe all 28 emotions, Russell’s conducted a second study. Here, participants
9
were asked to sort all 28 emotions into groups of 4, 7, 10 and 13 affect terms in
successive trials. Instructions were to group together emotional states that were
most similar. The similarity of each pair of words was judged according to the
number of trials that each pair was grouped together, with the score for each trial,
weighted by the number of alternatives available in that sort. Using the GuttmanLingoes multidimensional scaling procedure (Lingoes, 1973), which provides a
geometric representation of the relationships between all 28 emotions by placing
them in geometric space, a final similarity matrix was formed by taking the mean
entry across subjects for each cell of the matrix. Fit scores indicated that a twodimensional solution provided the best fit to the data. In other words, that emotion
is best represented by the axes of pleasure and arousal.
The aim of Russell’s third study was to test the how well his proposed model
approximated the structure of self-reported affect. As part of this experiment,
participants completed Mehrabian and Russell’s (1974) state affect scales of
pleasure-displeasure, degree of arousal and dominance-submission. Each
dimension was assessed by six items using a semantic differential format. In order
to test his model, each of the 28 emotions was regressed onto the two main bipolar
scales of pleasure-displeasure and degree of arousal, allowing beta weights from
regression analyses to be used as coordinates. Results indicated that a similar
circumplex model to that found in his preceding studies had been reproduced. The
only notable differences were that the terms ‘depressed’, ‘sad’ and ‘gloomy’ were
rotated slightly towards the horizontal axis of pleasure-displeasure indicating that
these terms may comprise more displeasure than first thought. Correlations
between all 28 items ranged from .22 to .66 (p <.001), indicating a mixture of
small to large relationships amongst emotion terms. A principal component factor
analysis revealed that 45.8% of variance in the happy-sad and tense-relaxed
contrasting words could be accounted for by two major components or
dimensions (i.e., pleasant-unpleasant and level of arousal). A further three
components accounted for an additional 13.1% variance making a total of 58.9%
variance accounted for by these five components. Finally, orthogonal rotation on
the five components produced happy-sad contrasting words (e.g., happy, delighted
and pleased, with sad, depressed and miserable) as the first component; and tenserelaxed contrasting words as the second component (e.g., tense and frustrated,
10
with relaxed, calm, tranquil and at ease). The third component was labeled
‘sleepy’ and consisted of sleepy, tired and drowsy. Unlike the first two
components, the third component of sleepy was without bipolar opposites. Also
without bipolar opposites was the fourth component labeled ‘angry’ (consisting of
angry and annoyed); and the fifth component labeled ‘alarm’ (consisting of
alarmed, astonished, and afraid).
In summary, Russell’s circumplex offers a simple and empirically justified model
for affect that conforms to the laypersons’s understanding of this construct.
Furthermore, the model’s emphasis on the two major axes of pleasantnessunpleasantness and arousal-sleepiness, provides a means by which all emotion can
be classified – around a circle in two-dimensional space. However, although
subsequent researchers have provided empirical support for Russell’s model of
self-reported affect (e.g., Huelsman, Furr, & Nemanick, 2003; Yik, Russell, &
Feldman Barrett, 1999), it is not without its critics (i.e., Larsen & Diener, 1992;
Watson & Tellegen, 1985). First, Russell provides no rationale for inclusion of the
28 emotions used in all three studies. This is important because the choice of
terms can have a substantial impact on model fit. Second, Russell’s circumplex
model was unable to account for approximately 30% of the variance in selfreported affect. However, in defense, Russell (1980) argues that measurement
error and acquiescent responding can account for most of this missing variance.
Revisiting the Circumplex Model of Affect: An Unresolved Issue
One unresolved issue concerning the Circumplex Model of Affect is that it
conceptualises affect in such a way that implies both axes are equally strong. An
alternative to this, that affect may be best represented by an ellipse rather than a
circle, has been supported by Huelsman, Nemanick and Munz (1998). They found
that students’ commonsense notions of mood rely more on good moods and bad
moods than activation. In particular, they found that adjectives such as exhausted,
fatigued, tired, weary, worn out (affects representing tiredness) and aggravated,
agitated, hostile, irritable, upset and uptight (affects representing unpleasantactivated), were simply classified as variations of the same bad-mood theme. It
appeared that energy levels had been overlooked in their ratings of mood. In
further support of the suggestion that both axes of the circumplex are not equally
11
strong, studies by Feldman (1995a; 1995b) found that affect valance accounted for
more variance in mood ratings than did arousal.
In an attempt to shed some light on this issue, Davern et al., (2007) present some
evidence that the pleasant-unpleasant axis of the circumplex dominates. In one of
their studies, these authors plotted the location of 31 affect descriptors using the
CIRCUM program developed by Browne (1992) and Fabrigar, Visser and Browne
(1997). This program provides a polar angle between 0o and 360o for each affect,
enabling them to be plotted on a circle. Their results indicated that affects tended
to aggregate around the poles of the pleasant-unpleasant axis. Thus, while these
results generally support circumplex theory, there is some evidence of angular
dispersion of the affect towards the pleasant-unpleasant axis. The implication of
this finding is that affects associated with SWB appear to be more strongly
categorised by the hedonic than by the activation axis. This finding is consistent
with Huelsman, et al.,(1998) and Feldman (1995a; 1995b).
Empirical Demonstration of the Importance of Core Affect to SWB Studies
A ground-breaking study by Davern, et al., (2007) provides empirical support that
SWB is primarily an affective construct, with independent contribution from
cognition. These authors sought to determine which affect terms could account for
the greatest amount of variance in SWB. SWB was measured using the single
global item ‘How satisfied are you with your life as a whole?’ Due to the abstract,
non-specific nature of this question (to be discussed later in more detail) this item
reflects the ‘core’ of SWB and taps the target of the investigation – individual
SWB set-points. (Cummins, Eckersley, Pallant, Van Vugt, & Misajon, 2003)
Using 478 participants from the Australian Unity Wellbeing Index Longitudinal
Study (Survey 5, November 2002), Davern, et al., found that only six of the affect
terms contributed significant unique variance to the prediction of satisfaction with
life as a whole as: content (sr2 = .15), happy (sr2 = .12), energised (sr2 = .09),
satisfied (sr2 = .08), stressed (sr2 = -.08) and pleased (sr2 = .06). In a second
multiple regression analysis, five of these affects were found to explain 64% of
the variance in SWB measured using the Personal Wellbeing Index (PWI;
International Wellbeing Group, 2006) suggesting a strong affective component.
From these studies, Davern et al., placed their findings into the context of ‘core
12
affect’, as defined initially by Russell (2003) and propose this small group of
unique affect predictors as a measure of core affect.
In a second study, Davern, et al., (2007) used SEM to explore the relative strength
of core affect (content, happy and excited), in three separate models incorporating
cognition (multiple discrepancies theory; Michalos, 1985) and all five factors of
personality (NEO Personality Inventory; Costa & McCrae, 1992). Using an
Australian adult sample of 854 participants aged 18-86 years (M = 52.00; SD =
15.37), SEM was used to compare an Affective-Cognitive driven model of SWB
(PWI; International Wellbeing Group, 2006) with a personality driven model of
SWB and a discrepancy-driven model of SWB. According to model fit statistics,
the Affective-Cognitive model provided the best fit to the data, explaining 90
percent of the variance in SWB. Moreover, all of their models confirmed that the
relationship between SWB, core affect and discrepancies is far stronger than that
between personality and SWB. Thus, evidence suggests that core affect is the
driving force behind SWB and not personality as is generally reported in the
literature (e.g., Headey & Wearing, 1989, 1992; Vitterso, 2001; Vitterso & Nilsen,
2002). According to Davern et al., the major implication of this finding is that
core affect may be driving the relationship between personality and SWB and
since core affect is driving both personality and SWB, individual differences in
set-point levels of core affect may be causing personality and SWB to correlate.
This is a likely reason why SWB and personality often appear so related.
In summary, Davern, et al., conclude that under ideal conditions of measurement
and sample selection, SWB will be driven by core affect. Indeed, in their view,
SWB is regarded as predominately a measure of core affect. Further, because core
affect reflects an enduring and stable characteristic and one that is most likely
biologically based, biology combines with psychology to encourage the continued
experience of pleasant core affect across the life-cycle. Through the process of
natural selection, it is possible that humans have evolved to experience a level of
core affect that is stable and positive. According to these authors, not only does
this process provide the motivation for living, it is also responsible for regulating
the stability of SWB set-points.
13
The debate surrounding the issue of whether satisfaction judgements are affective
or cognitive evaluations of a person’s experience will now be discussed.
Global and Domain Specific Judgments of SWB
Early studies, such as that by Andrews and Withey (1976), acknowledge that
affect comprises an important component of SWB, separate from cognitive
judgments of life satisfaction. In the view of Schwarz and Strack (1999), life
satisfaction judgments evoke a host of heuristic strategies as parsimonious
indicators of wellbeing, for example, through the evaluation of current mood.
According to these authors, it would be impossible for a person who is asked to
evaluate their satisfaction with their life as a whole to sum positive and negative
life experiences, weight these according to importance and then arrive at an
accurate, cumulative estimate of life satisfaction (Schwartz & Strack, 1999).
Rather, they simply rely on the use of a heuristic directed to mood. However,
Schwartz and Strack also consider that the use of heuristics is more pronounced
when participants respond to more global aspects of life satisfaction than when
individuals are required to rate their level of satisfaction with more specific life
domains. As mentioned, domain-based satisfaction judgments, although still quite
abstract in nature, are more specific than global life satisfaction and are
characterised by more well-defined assessment criteria. These judgments are
relatively easier to make and involve the use of intra-individual and interindividual comparative processes (Schwarz & Strack, 1999; Schwarz, Strack,
Kommer, & Wagner, 1987).
The view that heuristics, such as evaluating current mood, influences life
satisfaction is not shared by all researchers in this area. According to Schimmack,
Diener and Oishi (2002), the challenge for any theory of life satisfaction is to
explain the high level of temporal stability in life satisfaction judgments
frequently observed within the SWB literature. These authors believe that if
people rely on chronically accessible sources, then their life satisfaction
judgments are based on the same sources in repeated life satisfaction judgments.
Schimmack et al. further add that, if satisfaction in these domains remains stable
over time, then life satisfaction judgments will also remain stable. In their view,
personality, in particular, extraversion and neuroticism may be related to global
14
life satisfaction in that they predict stability of these chronically accessible
sources. For example, as stated by Diener, Napa Scollon, Oishi, Dzokoto and Suh,
(2000), because global evaluations are often vague, they allow greater freedom for
people to project norms, self-beliefs and dispositional tendencies on assessment
items.
In support of their contention that life satisfaction and domain-based satisfactions
reflect stable rather than variable states, Eid and Diener (2004) conducted a study
using a sample of 280 college students tested on three occasions, with four weeks
between measurements. The time-lag was chosen for two main reasons 1. to make
the time-lag as short as possible so as to analyse short-term variability in SWB
whilst reducing variability due to life circumstances; and 2. to prevent participant
bias (e.g., carry over effects from memory of the previous trial). Included in their
questionnaire were Diener’s Satisfaction with Life Scale (SWLS; Diener,
Emmons, Larsen, & Griffin, 1985; Pavot & Diener, 1993), measures of affect
(assessing the frequency and intensity of 24 emotions belonging to 6 emotion
groups - love, fear, joy, anger, shame and sadness - and personality scales
designed to assess personality correlates of SWB and their relations with mood.
According to their results, no more than 9 – 17% of the observed variance in SWB
could be accounted for by occasion-specific variability; whereas 74% of variance
was due to stable inter-individual differences. Based on these data, Eid and Diener
concluded that more stable factors, such as personality, were responsible for
regulating SWB. Thus, these authors consider life satisfaction and domain-based
satisfactions as constructs reflecting stable, not transient states.
However, although a proponent for the pre-eminence of cognition in forming
SWB, Diener (2000) acknowledges that affect comprises an important component
of SWB and that researcher’s need to use measures of both pleasant and
unpleasant affect. Furthermore, he concedes that if people’s moods and emotions
are indicators of their current state of affairs, then affect must be considered a
basic constituent of SWB (Diener, 1984; Diener & Diener, 1996). Thus, Diener
considers SWB as a construct consisting of an affective component.
15
Veenhoven (1991) offers an alternate perspective in this controversy. In his view,
happiness is considered to be the degree to which an individual judges their
overall quality of life favourably. In other words, how well a person likes the life
they lead. Veenhoven further adds that happiness can also be called life
satisfaction; thus, for him, the terms happiness and life satisfaction are
interchangeable. Furthermore, according to Veenhoven, there is evidence that
happiness is a concept comprising two independent components. The first of these
components is hedonic level of affect, which is the degree to which pleasant
affective experiences generally outbalance unpleasant ones. The second
component is contentment and refers to a cognitive evaluation of the degree to
which an individual perceives their aspirations to be met. Together, these affective
and cognitive appraisals of life are sub-components of the broader construct
overall ‘happiness’.
In his 1991 paper, one of Veenhoven’s greatest criticisms of the literature is that
happiness is often considered ‘relative’. That is, happiness does not depend on
objective good, but rather on subjective comparison. In his view, multiple
discrepancies theory (MDT; Michalos, 1985) typifies this approach in that it
assumes people use several standards of comparison when evaluating their life.
For example, according to MDT (to be discussed in more detail later), life
satisfaction reflects the size of the ‘gap’ between what one has and wants – the
greater this perceived gap, the less satisfied a person will be and vice versa.
Veenhoven sees a number of weaknesses in this theory which suggests that SWB
depends on comparison. One of his main arguments is that many proponents of
discrepancy theory tend to equate standards of comparison with needs. By
definition, needs are bio-psychological prerequisites for functioning, which are
innate and are also referred to as ‘instincts’, ‘drives’ and basic ‘needs’ (e.g.,
McDougal, 1908; Wentholt, 1975). Thus, because all needs (e.g., for food, water,
sex and shelter) are a product of evolution and necessary for survival (Maslow,
1965), in Veenhoven’s view, to presume that nature has left gratification of such
an important need to the wisdom of conscious reasoning alone is not intuitive.
Rather, he argues that nature has safeguarded the need for life satisfaction by
linking it to pleasant affect (Veenhoven, 1991). Thus, pleasant affect signals
gratification and encourages ongoing adaptive behaviours such as seeking others
16
and being social. On the other hand, unpleasant affect signals threat and
automatically slows down action. In this view, life satisfaction is a product of an
evaluation of how one affectively feels; and since hedonic level of affect is linked
to gratification of basic bio-psychological needs, in contrast to ‘acquired’
standards of comparison, these innate needs do not adjust to any and all conditions
- happiness is not then relative.
In summary, both Diener and Veenhoven contend that SWB is a construct
consisting of both affective and cognitive components. However, whilst Diener
argues that cognition is the most important correlate, in Veenhoven’s view, the
former, affective component is most important and serves an evolutionary
advantage of maintaining the normal condition that is a positive sense of oneself
and crucial to the survival of the species.
Global and Domain Specific Judgments of SWB: The Present View
Similarly to Schwartz and Strack (1999) and Diener et al., it is argued that life
satisfaction judgments are based on easily accessible sources of information.
However, unlike Diener et al., the present thesis does not consider the broad,
general evaluation of satisfaction with life in general as a cognitive task per se due
to the abstract, non-specific nature of questioning. Rather, like Veenhoven (1991),
affect is regarded as an important component of SWB and that heuristics
(reporting on these general affective feelings towards one’s life) are thought to
drive satisfaction judgments. Additionally, it is proposed that domain-based
satisfactions, that comprise the measure of SWB, represent the ‘first-level
deconstruction’ of ‘satisfaction with life as a whole’ (International Wellbeing
Group, 2006). Together, the seven domains represent broad, semi-abstract areas of
life that describe the overall experience of life satisfaction. However, because the
domains are more specific than life satisfaction, they deviate somewhat from the
general positive mood state that is core affect. Thus, the more abstract the
satisfaction judgment, the greater the influence of core affect on these evaluations
since core affect specifically concerns the most abstract and personal view of the
self.
17
Despite the contention that core affect is the main driver of SWB, it is also
believed that cognition can exert some influence and that this influence will vary
between people. In this view, domain-based satisfaction judgments may be
influenced by a variety of factors in addition to core affect, for example, thoughts
and feelings associated with particular events, situations and experiences salient to
each domain. Furthermore, as satisfaction judgments become less abstract and
more specific (for example, when a person is asked to rate their level of
satisfaction with their car), the evaluation process will involve more cognition as
relevant thoughts and experiences associated with the more specific judgment take
effect. Under these circumstances, level of satisfaction becomes a composite of
core affect and the experience itself. In an empirical demonstration of this, data
obtained from the Australian Unity Wellbeing Index (Cummins, Woerner, Tomyn,
Gibson, & Knapp, May 2006) informs us that the mean satisfaction score for the
domain of Relationships is consistently greater than the mean score for all other
domains. Across 15 surveys, the mean satisfaction score for this domain has
ranged from 77.65%SM (Survey 13) to 81.39%SM (Survey 12). On the other
hand, the mean satisfaction score for the domain of community connectiveness is
considerably lower than most, if not all other domains. For example, the mean
satisfaction score for this domain has ranged from 68.66%SM (Survey 1) to
72.55%SM (Survey 12). The differences in mean satisfaction judgments for the
domains of relationships and community connectiveness highlight the fact that
some judgments involve greater cognition than others and not all are driven
entirely by stable levels of core affect.
In summary, Eid and Diener (2004) argue that all judgments of satisfaction are
driven to a large extent by dispositions, such as stable personality, mediated by
individual’s idiosyncratic thoughts, experiences and beliefs. Whilst it is agreed
that life satisfaction and subjective wellbeing consist of a cognitive component,
the present thesis contends that these wellbeing indicators are largely heuristically
driven by core affect. Furthermore, the more abstract the satisfaction judgment,
the greater the influence of core affect since core affect concerns the abstract
component of subjective wellbeing.
18
THEORETICAL EXPLANATIONS FOR SWB
Bottom-up versus top-down influences on SWB
Bottom-up influences on SWB include external events, situations, experiences
and demographic variables such as age, gender and socio-economic status. Such
influences were the major focus of early studies into the determinants of SWB.
However, early SWB researchers all but put an end to theory suggesting that these
factors are important. For example, Campbell et al., (1976) found that
demographic factors, such as age, sex and income, accounted for no more than
20% of the variance in SWB; whilst Andrews and Withey (1976) report no more
than 8% variance.
On the other hand, ‘top-down’ influences on SWB have received far more
attention and have been more extensively researched than have ‘bottom-up’
influences. Top-down theories stress the importance of personality and its direct
influence on SWB. Such theories assume a global tendency (derived from stable
personality characteristics) to experience life in a positive or negative manner
(Diener, 1984). In this way, these global tendencies will have a consistent and
ongoing influence over the interpretation of momentary events and situations.
Early studies, such as by Andrews and Withey (1976) and Campbell, et al., (1976),
provide evidence for top-down models in studies which show little change in
SWB on the basis of different combinations of reactions to specific life domains.
Furthermore, Headey and Wearing (1989; 1992) offer a top-down approach to
SWB in their Dynamic Equilibrium model. According to this model, positive and
negative life events can alter individual levels of SWB above and below its setpoint; however, deviations are usually short lived as stable personality
characteristics ensure that SWB will revert back to its equilibrium. Empirical
evidence in support of the link between personality and SWB is outlined in the
section below.
Personality and Subjective Wellbeing
Personality is one of the most widely studied and thoroughly examined variables
in quality of life and SWB studies There is a large body of research which
suggests that personality, in particular the traits of extraversion and neuroticism, is
19
one of the strongest and most consistent single predictors of SWB (e.g., DeNeeve
& Cooper, 1998; Emmons & Diener, 1985; Headey & Wearing, 1989, 1992;
Vitterso, 2001; Vitterso & Nilsen, 2002). In support of personality as an important
correlate of SWB, in their 1998 study, DeNeeve and Cooper conducted a metaanalysis of over 137 personality traits involving a total of 148 separate studies and
42,171 respondents. To be included in their investigation, research reports had to
consist of a valid measure of SWB (e.g., Deiner’s Satisfaction With Life Scale;
Diener, et al, 1985) and at least one personality measure (e.g., NEO Personality
Inventory; Costa & McCrae, 1992). Their results produced an overall weighted
correlation of .19 between personality and SWB. SWB was found to correlate -.22
with neuroticism and .17 with extraversion. Thus, it was concluded that
personality is a consistent and reliable predictor of SWB.
Contributing to this line of investigation, Vitterso (2001) presents data suggesting
that emotional stability is the more important of the personality traits. Using a
sample of 264 Norwegian high school students, the Norwegian Big Five Inventory
(Engvik, 1993) as a measure of personality and the Tripartite concept of
subjective wellbeing ((subjective wellbeing = satisfaction with life + (positive
affect – negative affect)), Vitterso found moderate correlations between emotional
stability and satisfaction with life (r = .39 and r = .47). Additionally, his regression
analyses indicated that emotional stability explained 38% of the average variance
in subjective wellbeing across two occasions whilst the average amount of
variance explained by extraversion was a low 7%. In another study, Vitterso and
Nilsen (2002) investigated the influence of the Five Factor Model of personality
on subjective wellbeing. Using a sample of 461 Norwegian students from two
studies and the Satisfaction With Life Scale (Diener et al., 1985) as a measure of
wellbeing, Vitterso again concluded that neuroticism is the more important factor,
explaining eight times more variance in subjective wellbeing than extraversion.
Personality also features in early models of Cummins homeostatic theory of SWB.
According to Cummins (2000a), personality is part of a highly integrated
management system and comprises the primary genetic component responsible
for determining the set-point range for SWB, as initially proposed by Headey and
Wearing (1989; 1992). Furthermore, Cummins and Nistico (2002) argue that
20
personality has a profound influence on SWB by mediating the relationship
between external experiences and SWB (Cummins et al., 2002). In their view, the
personality components most strongly associated with SWB maintenance are
extraversion and neuroticism.
It is important to note, however, that the more recent research by Davern, et al.,
(2007), detailed earlier, found that personality is not the driving force behind
SWB. In fact, their results virtually eliminated personality, measured using the 60
item Short Form of the Revised NEO Personality Inventory (Costa & McCrae,
1992), as contributing unique variance to the explanation of SWB in the presence
of suitable affective and cognitive variables. The implication of their research is
that there is a need to reinterpret earlier research supporting a link between
extraversion, neuroticism and SWB.
Multiple Discrepancies Theory and Subjective Wellbeing
There are many ‘gap’ approaches to subjective wellbeing, in which it is proposed
that SWB is formed by the degree of separation between a person’s perceived
aspirations or goals and what they currently have (e.g., Andrews & Robinson,
1991). Of these, the one that has received the most attention is multiple
discrepancies theory (Michalos, 1985); which integrates social comparison,
personal experience and the concept of ‘relative deprivation’ (Runciman, 1966).
MDT is based on the premise that wellbeing results from perceived gaps or
discrepancies between: 1. what one has and wants (self wants); 2. what relevant
others have (self-other); 3. the best one has had in the past (self-best); 4. expected
to have three years ago (self-progress); 5. expects to have after five years (selffuture); 6. deserves (self-deserves); and 7. needs (self-needs). According to MDT,
smaller perceived gaps are associated with higher wellbeing, whilst larger
perceived gaps are associated with lower levels of wellbeing.
In support of his theory, Michalos (1985) regressed 12 domains and 6
demographic variables against global life satisfaction and found that together,
these variables explained 53% variance. However, according to his data, selfesteem (β = .225) and social support (β = .212), not any of the perceived
discrepancies associated with the domain satisfaction judgments, were the
21
underlying drivers and strongest predictors of life satisfaction (Michalos, 1985).
For example, significant predictors and their associated standardised regression
coefficients were satisfaction with health (β = .168, p<.001); family relations (β
= .141, p<.005); paid employment (β = .180, p<.001); and education (β = .123,
p<.01). Thus, results from Michalos’ research only partially support the influence
of perceived discrepancies on SWB and suggests more important influences such
as self-esteem and social support. The present study will investigate these findings.
Nonetheless, MDT remains as one of the most comprehensive of the gap
approaches to SWB measurement and proposes that in order for a person to
maximise their wellbeing, they should aim to decrease the discrepancy between
what they have and want, choose a realistic reference group and remain optimistic,
especially in the face of adversity.
The Role of Comparative Devices in SWB and Criticisms of MDT
Although MDT appears to offer a simple explanation of subjective wellbeing, its
theoretical underpinnings fail to explain the observed population standard of 75%
satisfied frequently observed in the literature (e.g., Cummins, 1995; Headey &
Wearing, 1989). For example, MDT argues in favor of a dispositional mechanism
responsible for maintaining a ‘gap’ that results in consistent life satisfaction
judgments of approximately 75%SM. The idea that such a mechanism is
responsible for producing constant unmet needs seems unintuitive and the fact
that humans can consciously identify and direct behavior towards the attainment
of unmet needs and goals is further evidence that a ‘gap’ or ‘gaps’ would be
difficult to maintain, as pointed out by Cummins and Nistico (2002). Thus, a
cognitive mechanism for life satisfaction based on maintenance of needs that can
be objectively met seems unlikely. A somewhat similar reservation is shared by
Veenhoven (1991), as has been discussed.
Cummins and Nistico (2002) do add, however, that if such needs are abstract, they
can exist detached from reality and could regulate perceived satisfaction with life
if two conditions are met. The first condition is if needs are perceived as
approximately three quarters met. For example, the need for optimism regarding
the occurrence of a future event, such as being promoted at work. This need may
22
be partially fulfilled by holding optimistic cognitions such as ‘I am a hard
working and respected employee and it is just a matter of time before I receive a
promotion and achieve higher status within my organisation’. The second
condition, as outlined by Cummins and Nistico, is that needs are maintained under
homeostatic control such that the ‘gap’ between needs and their fulfillment must
be maintained within a narrow range. This suggestion lies within the realm of
subjective wellbeing homeostasis.
23
CHAPTER 2: STABILITY IN LIFE SATISFACTION JUDGEMENTS
TOWARDS A HOMEOSTATIC MODEL OF SUBJECTIVE WELLBEING
One of the more interesting and important findings in SWB literature is that life
satisfaction is not free to vary over the entire range of values offered by a
particular measurement instrument. Rather, research has consistently found that
life satisfaction is held around a ‘set-point’ and is remarkably stable across time
(e.g., Cummins, 1995; Eid & Diener, 2004; Headey & Wearing, 1989, 1992;
Schimmack, Diener, & Oishi, 2002). SWB stability has been observed at both the
level of individuals (e.g., Diener & Diener, 1996; Hanestad & Albrektsen, 1992;
Headey & Wearing, 1989) and populations. In fact, 16 surveys conducted on the
Australian population over the years 2001-2007 revealed that the mean level of
SWB has varied by just 3.0 percentage points (Cummins et al., 2006). Thus,
according to these data, the mean SWB of the Australian population for any
survey can be predicted with 95% certainty to lie within the range of 73.43 to
76.43 points.
Headey and Wearing’s Longitudinal Data
The first authors to suggest that individuals may have a ‘set-point’ level of
wellbeing were Headey and Wearing (1989, 1992). Inspiration for their work
came from the discovery that most quality of life research to date had reported
that the majority of individuals living in Western industrialised nations (with the
exception of Black South Africans) have levels of wellbeing that are generally
stable and positive. Disenchanted with previous quality of life surveys that they
considered “one-offs” (cross-sectional) (p.4), Heading and Wearing’s main
objective was to go beyond merely taking ‘snapshots’ of the characteristics of
happy and unhappy people and detail sources of psychological wellbeing and the
process of change.
Their interest turned to ‘set-point’ theory following data obtained from the
Victorian Quality of Life Panel Study, which initially consisted of 942
participants interviewed on five occasions in the 1980’s (81’, 83’, 85’, 87’, 89’).
The goal of the panel study was to obtain repeated measures of wellbeing in order
24
to understand how people cope with change. In addition to measures of life
satisfaction and positive and negative affect, panel members were asked to
complete a life-events inventory, detailing the occurrence of events, happenings
and experiences in the previous two years of their lives.
One of the more influential findings from the Victorian Quality of Life Panel
Study was that life satisfaction amongst a majority of participants was quite stable
across the years 1981 through 1989 (r = .47). In fact, data showed that at the
conclusion of the eight year testing period, only 27.4% of participants had shifted
by over one standard deviation. In comparison, 32.5% had shifted for positive
affect and 27.9% for negative affect. Additionally, results support the importance
of personality factors and Headey and Wearing speculated that personality was
integral to the maintenance of life satisfaction. For example, it was found that the
personality dimensions of extraversion and neuroticism accounted for a
significant portion of the variance in life satisfaction. Further, extraversion was
found to correlate at .26 with life satisfaction, .22 with positive affect, -.17 with
anxiety and -.23 with depression. Neuroticism, on the other hand, was found to
correlate at -.34 with life satisfaction, -.12 with positive affect, .40 with anxiety
and .46 with depression.
Headey and Wearing’s panel study led to the development of their Dynamic
Equilibrium model, which proposes that, although personality characteristics
provide the stable set-point for wellbeing, unusual circumstances have the
potential to alter individual set-points above or below equilibrium. According to
this theory, such deviations from set-point are usually short lived because
dispositional influences on SWB, such as genetics and stable personality factors,
play an important equilibrating function that ensures under most circumstances,
SWB reverts back to baseline levels. Thus, in the absence of particularly positive
or negative events, individual levels of SWB are likely to remain relatively stable
across time.
Although Headey and Wearing’s Dynamic Equilibrium model offers some
insights into the nature of subjective wellbeing, their research is not without flaws.
For one, although life satisfaction amongst participants was found to be generally
25
quite stable over the eight years of testing, moderate levels of change in wellbeing
and psychological distress amongst some participants suggest that these measures
may have been sensitive to the instruments used. For example, life satisfaction
was measured by 6 items on a 9-point ‘delighted-terrible’ scale (Andrews &
Withey, 1976). Items such as ‘how do you feel about how exiting your life is’?
and ‘how do you feel about the extent to which you are succeeding and getting
ahead in life’? were originally conceived as state measures of life satisfaction
(Andrews & Withey, 1976) rather than wholly stable traits, as was the target of
the investigation. As a consequence, instability observed amongst some
respondents may be attributed to the fact that current states were measured, not
traits, as reported.
Furthermore, Headey and Wearing’s Dynamic Equilibrium model is incomplete in
that it fails to provide details as to the psychological processing operating within
individuals that maintains SWB at a stable, positive level. In 1995, Cummins
addressed this issue with his homeostatic theory of subjective wellbeing.
Cummins built on the work of Headey and Wearing by extending their theory of
personality to incorporate the variables that relate to satisfaction with the self (e.g.,
self-esteem, optimism and perceived control). Similarly to Headey and Wearing’s
model, SWB homeostasis takes into account the impact of life events on a
hypothetical management system.
Population and Individual Normative Ranges of Subjective Wellbeing
In 1995, Cummins published an article with compelling evidence suggesting that
SWB is actively controlled and maintained by a set of psychological devices,
analogous to the physiological, homeostatic maintenance of body temperature.
Inspired by research suggesting that SWB is stable and positive, Cummins (1995)
sought to investigate this phenomena in what has been one of the most influential
studies of his career. As part of this research, he combined data from 16 unrelated
studies conducted in Western nations that described population satisfaction.
Inclusion of these studies was based on a strict criterion: they had to originate
from countries comparable on culture and socio-economic status, include
participants between the ages of 17-65 and consist of 200 or more participants.
Finally, all studies had to include some variant of the single global item ‘All
26
things considered, how satisfied are you with your life as a whole?’ In order to
compare findings across studies, scores were converted to a Percentage of Scale
Maximum (%SM). Empirical demonstration of homeostasis is dependant upon
this statistic, which represents the conversion of any Likert Scale score into a
standard form that ranges from 0 -100. This conversion is made using the formula
[Likert score – 1/Number of points on the Likert scale – 1]100 and permits
comparison between studies using different scales of measurement.
Consistent with Headey & Wearing (1989, 1992), who found that on average,
people report being three-quarters satisfied with their lives, results of this study
revealed that, using population mean scores as data, life satisfaction yielded a
mean score of 75.02% of the scale maximum score (% SM) with a standard
deviation of just 2.74 percentage points. To examine this finding in greater detail,
Cummins (1998) conducted another study, although this time, combined data
from 47 non-Western and English speaking nations. Using a similar inclusion
criterion to his 1995 study, Cummins reported that the mean life satisfaction score
for these nations was approximately 70%SM (SD = 5.0). Based on the assumption
that two standard deviations represent the normal range, Cummins found
considerable overlap between the Western range (70 – 80) and the global range
(60-80).
Overlap Between the Western and Global Range of Subjective Wellbeing
The fact that the Western range for subjective wellbeing is contained within the
world range informs us about several important issues. First, Western and world
mean life satisfaction scores focus attention on ceiling and floor effects at the
upper (80%SM) and lower (60%SM) range for SWB. A ceiling effect results in a
reduction in sample variance as mean life satisfaction scores approach the upper
limit. Cummins (1998) presents data supporting this deduction – as mean life
satisfaction scores for a population increase, corresponding standard deviations
tend to decrease. For example, combining data from all countries reviewed in his
study yields a mean life satisfaction score of 75%SM with a standard deviation of
5.0%SM. However, if data are restricted to English-speaking Western countries
and Norway, the standard deviation reduces to 2.5%SM and to just 2.3%SM when
highest ranking nations are included.
27
This ceiling effect (at 80%SM) has important implications for policy makers,
health care professionals and those concerned with increasing wellbeing amongst
individuals and populations. The fact that no population has mean life satisfaction
scores consistently above 80%SM implies that in order to increase mean levels of
wellbeing amongst a population, people at the lower end of the SWB spectrum
must be targeted because these are a sub-group that can experience substantial
changes in wellbeing. Ceiling effects at approximately 80%SM suggest that SWB
amongst people in the upper range cannot be improved, let alone improved and
sustained over long periods. Concerted efforts should therefore be directed at
people in the lower range – people who can improve their wellbeing once
circumstances change for the better.
Data from these studies also indicate that even amongst those countries with the
greatest number of individuals in the lower range for SWB, mean life satisfaction
amongst these countries is still above the scale midpoint (50%SM). Floor effects,
found at approximately 60%SM, as well as ceiling effects just mentioned, are
consistent with the idea that SWB is under the active control of an internal,
psychological homeostatic system that ensures that under most circumstances,
people are generally quite satisfied with their lives.
The idea that subjective wellbeing is under homeostatic control seems intuitive.
Subjective wellbeing homeostasis, conceived by Cummins as an evolutionary
survival mechanism, ensures that under diverse, often unpredictable and aversive
circumstances, humans are able to adapt to their environment and remain
sufficiently positive and motivated – a necessity for the survival of a species.
So far, this discussion has only involved the use of mean population scores used
as data. Individual life satisfaction scores show a much broader distribution.
According to Cummins et al., (2002), the scores from individuals, measured by
some variant of the single ‘life as a whole’ item, yields a normative mean that is
similar to the mean of the population mean scores at about 75 points, but the
standard deviation is much higher at about 18 points. Again, using two standard
deviations to denote the normative range, this creates a normative range for the
life satisfaction of individuals from approximately 40 – 100%SM (Cummins,
28
2003) for Western countries. This range has been extended to 30 – 100%SM to
representative the normative range for individuals from all countries (Cummins,
2003).
The Relationship Between Means and Standard Deviations
Cummins (2003) explored the relationship between group mean life satisfaction
scores and their corresponding standard deviations by grouping nations selected
for his study according to the level of life satisfaction. For example, the first group
(rank 1) comprised nations with the highest reported levels of life satisfaction,
such as Denmark, Iceland, Sweden and Finland. Means for these nations did not
differ significantly from one another and ranged from 80.70 to 79.10%SM.
Ranked second were nations with aggregate levels of life satisfaction below rank
1 that again did not significantly differ from one another. This was continued
through to rank 5. Means and standard deviations for countries ranked 1-5 are
presented in Table 1 below.
Table 1:
Means and standard deviations for countries ranked 1-5
Rank
N
Rank Mean
Mean + SD
Rank SD
Mean + SD
Rank Mean vs. Rank
Mean + SD
1
11
79.93 + 1.31
16.74 + 1.44
r = .64
2
16
77.18 + 1.26
17.46 + 1.23
r = .50
3
10
74.26 + 1.98
18.23 + 1.26
r = .63
4
10
69.59 + 2.17
20.31 + 2.66
r = -.38
5
15
64.96 + 4.04
20.23 + 3.67
r = .27
Total
62
73.01 + 6.09
18.59 + 2.68
r = -.41
The most interesting finding from Table 1 is that while mean life satisfaction
scores for nations ranked 1 – 5 differed by approximately 14.97%SM, variation
between corresponding standard deviations differed by just 3.49%SM.
29
Additionally, for the 11 nations within Rank 1, the average standard deviation was
only 1.44%SM and raised to just 3.67%SM for the average of those nations
ranked 5. Data from this study informs us that there is a high level of consistency
in the range of standard deviations over those groups with mean life satisfaction
scores above 71%SM. However, these data also show that studies reporting mean
life satisfaction scores below 70%SM also report corresponding standard
deviations greater than those with mean life satisfaction scores above 70%SM.
The finding that there is greater variation in life satisfaction scores as wellbeing
approaches 70 points can be interpreted that, on average, this value marks the
lower limit of the set-point range. Further, greater variation in life satisfaction
scores at the lower limit of the set-point range suggests that a homeostatic system
is fighting external forces to gain control of and regulate subjective wellbeing. At
this level of challenge, resilient individuals are still able to maintain homeostatic
control, whereas other less resilient people are experiencing homeostatic defeat. It
is this latter group that is responsible for the rise in within group-variance.
Distribution of Normative Life Satisfaction Scores
The relationship between population mean life satisfaction scores and their
corresponding standard deviations can be used to build a homeostatic model for
life satisfaction. In Figure 1 below (reproduced from Cummins, 2002), the
ordinate %SM represents the normative range of life satisfaction scores for
western populations that range from 80 to 70%SM. The ordinate is intersected at
70%SM by a line of ‘resistance’. This line approximates the lower band of the
proposed homeostatic range for life satisfaction. That is, 70%SM represents a
level for life satisfaction that the homeostatic system is actively defending
(Cummins, 2002). Figure 2 displays the distributions of three hypothetical groups
with mean scores greater than 70%SM.
30
Figure 2:
The frequency distribution of three hypothetical groups with life
satisfaction mean scores > 70%SM (reprint from Cummins, 2003)
The implications of the normative data presented in Figure 2 are as follows. Three
hypothetical studies are represented with mean life satisfaction scores varying
between 70 – 75%SM. Each distribution is positively skewed and displays an
increasing leptokurtic distribution as the mean increasingly approximates the line
of resistance. Over the range 81 – 70% SM, it appears that another force is
operating that actively limits the upward extension of the range at approximately
80%SM. In support of this, data from Cummins (2002) indicate that for the 11
studies comprising group rank 1, the range of mean life satisfaction scores extends
over the range of only 4.2%SM (77.9-82.1%SM). Thus, 80%SM may represent
the upper line of resistance which the system is defending. It is important to note
that two proposed lines of resistance (70 and 80% SM) correspond with the
normative range for life satisfaction as outlined by Cummins (1995).
Defeat of the 70%SM Resistance Line
According to homeostatic theory of subjective wellbeing, mean population scores
below 70%SM represent a defeat of the homeostatic system for a significant
proportion of the population (Cummins, 2002). The inability of the system to
31
maintain mean life satisfaction above the line of resistance causes the distribution
to collapse and become more platykurtic. When this occurs, two important things
happen to the standard deviations 1) they increase as each distribution extends
downwards; and 2) a negative correlation between the size of the mean and
standard deviation results. Figure 3 provides a graphical representation of the
relationships described.
Figure 3:
The frequency distribution of four hypothetical groups life satisfaction
scores lie slightly above, or below, 70%SM (reproduced from Cummins, 2002)
Cummins (2002) proposes that the above relationships hold as the sample mean
decreases from approximately 70 to 60%SM. However, below this point,
distributions become relatively uninfluenced by homeostasis and the relationship
between means and standard deviations becomes less predictable. This is
supported by correlations presented in Table 1. More specifically, below a mean
score of approximately 70%, the relationship between group means and standard
deviations decreases considerably.
32
The concept of homeostatic threshold is discussed in the section below.
How Much Variation Does the System Allow?
Consistent with Headey and Wearing (1989, 1992), who proposed that subjective
wellbeing is normally held within a narrow range around a set-point, homeostatic
theory asserts that subjective wellbeing is managed within a ‘set-point range’ for
each individual and is remarkably stable across time (Cummins et al., 2002).
Homeostasis goes hand in hand with the idea of threshold. As previously
discussed, studies by Cummins (1995, 1998, 2003) have supported the existence
of subjective wellbeing margins that are actively being defended by the system. If
SWB approaches the limits of its set-point range, the homeostatic system resists
further change and if the threshold is challenged, it acts to restore SWB back to
the normal range for each individual. Once the threshold is exceeded, then
homeostasis has been defeated and if the challenge remains, homeostasis may
never recover (for example, as in people who are persistently depressed).
Empirical evidence for this can be found through analysis of sample means. As
discussed, life satisfaction data contain floor and ceiling effects at approximately
70 and 80%SM for western populations and 60 and 80%SM as represented by
global data. It is suggested by Cummins (2002) that these scores represent the top
and bottom limits of the normative ranges of population life satisfaction and that
which is being defended by the homeostatic system.
However, characteristic of all homeostatic mechanisms, a persistent and noxious
stimulus can limit the system’s ability to maintain equilibrium. In terms of SWB,
the likelihood of permanent system defeat following an aversive event is
contingent upon the residual discomfort or lost functional status (Cummins et al.,
2002).
Helson’s (1965) Adaptation Level Theory and its role in subjective wellbeing
homeostasis is outlined in the section below.
Adaptation Level Theory
Central to subjective wellbeing homeostasis is the concept of adaptation (Helson,
1965). From the Latin ‘adaptare’, meaning to ‘adjust’, the term adaptation has its
33
origins in the field of biology and has been since extended to a variety of
disciplines, including psychology. It refers to a broad, general sense of adjustment
to the conditions under which a species must live in order to survive. The most
common conception of adaptation used by psychologists stems from
neurophysiology, where adaptation refers to the gradual decline in the rate of
receptor discharge following constant stimulation to one of the senses. However,
Helson extends this theory to assert that individual experiences, judgments,
attitudes, learning, intellectual and emotional behavior and interpersonal relations
all represent psychological modes of adaptation to environmental and organismic
forces. According to Andrews and Withey (1976), adaptation is fundamental to
maintaining a stable level of subjective wellbeing and is therefore essential to
understanding this construct.
Adaptation: A Two-way Process
The most common conception of adaptation is that it serves a desensitising
function. That is, to reduce stimulation/response following the presentation of a
constant or unchanging stimulus. However, Helson (1965) believes that
adaptation is as much a sensitising process as it is a desensitising one and argues
that a decline in response to some receptors following repeated stimuli is
associated with a heightened response on the part of other receptors. For example,
in physiological terms, adaptation to warmth makes us more sensitive to cold.
In the context of psychology, the basic tenant of ALT is that people use past levels
of stimulation (either positive or negative) to evaluate current levels of stimulation.
This judgment depends on the degree to which previous stimulation is higher or
lower than that to which the person has been accustomed. If stimulation is higher
than previously experienced, an immediate change in self-evaluation will occur.
However, several processes operate that diminish the impact of such change. The
first is an upward shift in adaptation because a new level of stimulation has been
added to that which the person has already become accustomed. The second is
habituation and refers to a decrease in response following presentation of a
constant or unchanging stimulus.
34
Adaptation and Subjective Wellbeing
ALT has been used by several authors to explain the remarkable stability of SWB
(Brickman & Campbell, 1971; Diener, 2000; Lykken & Tellegen, 1996). ALT
asserts that the impact of particularly positive and negative events on SWB is
usually short lived, as most people tend to adapt to changing circumstances. In
support of this, the classic study by Brickman, Coates and Janoff-Bulman (1978)
found that, following their respective life change events, lottery winners and
major accident victims did not significantly differ from controls in self-reported
levels of happiness or sadness just one to 12 months after the significant event.
Similarly, Suh, Diener and Fujita (1996) found that in less than three months, the
effects of major life events (e.g., being fired or promoted) no longer influenced
subjective wellbeing. More recently, Myers (2000) highlighted the fact that the
decades following World War II were associated with a mass increase in national
income amongst many Western nations; however, over this period, subjective
wellbeing has remained unchanged, even in the United States and other welldeveloped nations.
The explanation for these findings in terms of ALT is as follows: lottery winners,
those who were promoted and to a somewhat lesser extent, people reaping the
benefits of a growing economy, all experienced initial rises in wellbeing as their
lives took a sudden change for the better. However, such jubilation was short
lived as an increase shift in adaptation level resulted in the inability of simpler
pleasures to increase their subjective wellbeing. The result is a return to baseline
levels of life satisfaction as previous sources of pleasure fail to reach the new
adaptation level and consequently seem less appealing. In contrast, those who
suffered a debilitating accident (e.g., acquired paraplegia) experienced initial
drops in wellbeing following this life change event, causing their adaptation level
to fall. In time, however, new found pleasures that would have previously gone
unnoticed impact positively on wellbeing and produce at least partial, and in some
cases, a complete return to baseline levels of life satisfaction in the majority of
individuals. In sum, altered adaptation levels are responsible for assisting in
subjective wellbeing recovery for all groups.
35
On the basis of this reasoning, some researchers have come to the conclusion that
adaptation is so quick, complete and inevitable, that most long-term stability in
SWB can be accounted for by genetic predispositions (such as personality) rather
than life events (Lykken & Tellegen, 1996). In a similar vein, Brickman and
Campbell (1971) proposed that all people are destined to labor on a ‘hedonic
treadmill’. That is, although good and bad circumstances can alter momentary
evaluations of life satisfaction, baseline levels will eventually return as SWB is
determined more by stable factors such as personality and affect and not life
circumstances.
Although adaptation level theory appears to provide a simple explanation of how
individuals return to baseline levels of SWB following an event, questions still
remain. For example, although it has been found that adaptation does occur,
research tells us little about the completeness of adaptation. Also, it is not clear
whether people will return to original baseline levels of SWB and whether or not
new baselines are created following an event. Further, if there are individual
differences in adaptation, what factors contribute to these differences – are some
people predisposed to experience more or less complete adaptation? In reference
to the latter question, research suggests that happy individuals react more strongly
to pleasant stimuli and unhappy people to unpleasant stimuli (e.g., Gable, Reis, &
Elliot, 2000). According to Lucas, Clark, Georgellis and Diener (2003), if this is
true, then individuals with high baseline levels of SWB (happy people) should
react positively to particular marital transitions (e.g., getting married), whereas
those with low baseline levels of SWB (e.g, unhappy people) should react more
negatively to particular marital transitions (e.g., becoming widowed). The aim of
their study was to address these issues and in order to understand the process of
adaptation in more detail, Lucas et al., examined the long-term effects of maritalstatus change on SWB.
Their results provide support for a general process of adaptation following marital
transitions. From the outset, it was evident that people’s satisfaction with life
changed as a consequence of the respective transitions, but over time, adaptation
occurred. For the event of marriage, small increases in life satisfaction were noted
(approximately 10%SM) soon after the event. However, when baseline scores
36
were compared with people’s adaptation phase scores (scores some time after the
event had taken place), people were no more satisfied after marriage then they
were before marriage. For the event of widowhood, longer lasting effects on SWB
were evident. Results indicated that widows/widowers were less satisfied with life
after the loss of a spouse then they were before. According the Lucas et al., this
apparent decline in average satisfaction was due to strong initial reactions to their
loss, followed by relatively slow adaptation. Interestingly, widows and widowers
who did not remarry returned closest to their baseline levels of satisfaction (after
approximately 8 years) but were still significantly below baseline. This finding
contrasts recent Australian survey data that indicates that widows report generally
high levels of wellbeing (Cummins, et al., 2006). It appears that age, income and
number of children are factors that have a great impact on the wellbeing of
widows. However, there may also be marked differences between countries
contingent on the level of financial security afforded to such people.
The implications of Lucas et al’s., study for adaptation level theory are important.
First, results indicate that baseline levels of subjective wellbeing will not
necessarily return to equilibrium through the process of adaptation following a life
change event. Rather, the results show that there is some variability in the return
to SWB. Furthermore, Lucas et al., (2003) make the point that although adaptation
does occur, it can be slow and partial, and data indicate that some groups may
never return to baseline levels of wellbeing, as predicted by homeostatic defeat.
Second, these findings also expose deficits in Brickman and Campbell’s (1971)
‘hedonic treadmill’ analogy of life satisfaction. These authors suggest that it is
futile searching for ways to permanently increase life satisfaction amongst those
that are already satisfied because humankind is destined to ‘hedonic neutrality’.
That is, while it is possible to experience significant, brief positive changes in
wellbeing, adaptation ensures that wellbeing returns to set-point. The problem
with this conception of life satisfaction is that it does not account for data
indicating that some people, such as accident victims, can experience a return to
set-point levels of life satisfaction following an aversive event. More adequate
terms for Brickman and Campbell’s wellbeing analogy could include the ‘reversecycle hedonic treadmill’ or ‘hedonic cross-trainer’.
37
While adaptation level theory provides some explanation as to how SWB setpoints may be maintained, questions still remain as to the underlying cognitive
process involved. For example, what is it that actually adapts? Despite some
criticisms (e.g., Hunskaar & Vinsnes, 1991; Zautra & Reich, 1980), there is little
doubt that adaptation plays an important role in the process of subjective
wellbeing homeostasis, most likely at the level of met needs. Moreover, through
one means or another, adaptation to a changing environment does occur and is
involved in the process of subjective wellbeing set-point regulation, even in the
face of aversive life circumstances.
Adaptation and Subjective Wellbeing Set-points
A common feature of any homeostatic system is its failure when subject to
extreme circumstances. Under strongly aversive conditions, such as extreme
financial hardship, the capacity of this homeostatic mechanism to regulate
wellbeing will be greatly diminished. Under these circumstances, the relationship
between objective and subjective measures of wellbeing will be much greater.
This is because the determination of SWB will move from the homeostatic system
to the challenging agent as homeostasis is challenged. Thus, a non-linear
relationship exists between objective and subjective indicators of subjective
wellbeing.
Relationship Between Objective and Subjective Quality of Life
Recent work examining the reliability and validity of both objective and
subjective measures of life quality has found a weak relationship between these
two constructs (Diener, Suh, Lucas, & Smith, 1999; Headey & Wearing, 1992; Ng,
1997). A review by Cummins (2000b) supports this conclusion. In this review,
Cummins analysed data from 10 studies and found that the combined average
correlation between objective indicators (r = .315, SD = .051) and between their
comparative subjective indicators of wellbeing (r = .380, SD = .145) did not differ
from each other. However, both these average correlations were higher than the
average correlation between comparable objective and subjective indicators (r
= .120, SD = .082). From this review, Cummins concluded that objective and
subjective indicators are variables that interact within a system that maintains
38
subjective quality of life within a narrow range. Therefore, due to the capacity of
the homeostatic system to adapt to changing environmental circumstances,
objective and subjective indicators of well being are normally poorly correlated.
Objective and Subjective Wellbeing and SWB Homeostasis
According to Cummins et al., (2002) temporal stability in SWB depends on the
severity of the challenging agent and is generally restricted to two broad groups –
people with initial levels of SWB within the homeostatic range who experience no
event that challenges homeostasis and people who experience persistent
homeostatic defeat as a result of chronic aversive circumstances to which they
cannot adapt. Additionally, it could also be predicted that people who experience
the least stability are those who, at the time of initial measurement, were
experiencing homeostatic defeat as a result of some transitory event or
circumstances that could be accommodated over time.
Consistent with these predictions, Landua (1992) conducted a study exploring
issues related to the link between objective and subjective variables. He measured
overall life satisfaction on a 0 – 10 scale using and using longitudinal data from
12,000 participants over the age of 20 years, then grouped participants into
response categories as 0-4, 5-6, 7-8 and 9-10 on life satisfaction. Measures were
made at baseline and then again four years later. Results indicated that the
following percentages of participants remained in their initial category: 11% (0-4),
50% (5-6), 65% (7-8) and 61% (9-10). As predicted, the greatest stability was
found amongst groups of people reporting initial mean life satisfaction scores of at
least 70%SM. Additionally, lowest stability was found in groups of people
reporting initial mean life satisfaction scores in the <40%SM range. These results
are consistent with the notion that the low scoring group contains a high
proportion of individuals suffering homeostatic defeat as a consequence of some
form of depressive event or circumstance that was amenable to adaptation. The
fact that only 11% of individuals in the 0-4 group remained in this group some
four years later suggests that 89% of those experiencing homeostatic defeat at
baseline measurement were able to re-establish homeostatic control, at least to
some degree, and so move out of the lowest response category.
39
The concept of threshold allows for a further prediction of the relationship
between objective life circumstances and subjective wellbeing. More specifically,
that the correlation between SWB and extrinsic indicators will increase as SWB
moves outside its normative set-point range (Cummins et al., 2002). Thus, under
normal, unthreatening conditions, SWB is controlled by the homeostatic system
and SWB is expected to lie inside the upper and lower thresholds of the set-point
range (80 – 70%SM). However, when the strength of an extrinsic agent
(challenging environmental stimulus) exceeds the capacity of the homeostatic
system to adapt, control over SWB is assumed by that agent, such that a noxious
stimulus that exceeds the homeostatic threshold will defeat the system, resulting
in a fall in SWB. A curvilinear relationship between the strength of an external
stimulus and the value of SWB can be plotted. Figure 4 provides a graphical
representation of the process described.
Dominant Source of SWB Control
Set
point
Homeostasis
Defensive range
Challenging
conditions
80
Set
point
range
a
b
70
c
Upper
Threshold Lower
Threshold
SWB
Lower
Threshold
5a0
No
challenge
Very strong
challenge
Strength of challenging agent
Figure 4:
The relationship between subjective and objective quality of life (from
Cummins, Lau, & Davern, unpublished)
Figure 4 shows the changing relationship between the strength of a challenging
agent and SWB. Provided that circumstances do not challenge either threshold,
their variation will exert little influence on SWB. Under these circumstances, the
homeostatic system can comfortably accommodate fluctuations in the
environment such that SWB will be maintained within its set-point-range (in this
40
case, around 75 points). As external conditions become more challenging, the
level of SWB will vary within its set-point range and this will be determined by
the level of challenge and support. More specifically, when the level of support is
greater than the level of challenge, SWB will average above the set-point. On the
other hand, when the level of challenge is greater than the level of support that is
available, SWB will average lower than the set-point. Given that the magnitude of
fluctuation within the set-point range is approximately 10 points (Cummins, Lau,
& Davern, unpublished), the extent of fluctuation is expected to be quite modest.
This phase is shown in Figure 3 as ‘a’.
As the strength of a challenging agent intensifies, so to does the strength of the
homeostatic system in an attempt to maintain SWB within the normal set-point
range. This phase, where homeostasis holds the line of resistance in order to
prevent further loss below 70 points in depicted as phase ‘b’ in Figure 4. At this
point, SWB homeostasis is insensitive to changing levels of threat and regardless
of the level of threat, homeostasis will hold the line of resistance as long as the
system holds. However, at some point, a persistent and noxious threat will
overwhelm the system. When the strength of challenging agents exceeds the
homeostatic threshold, the homeostatic system will fail and in the process, forfeit
control to the external agent, resulting in a fall in SWB (see phase ‘c’ in Figure 4).
At this point, the value of SWB is sensitive to the agent that has assumed control
from the homeostatic system and as this threat increases, SWB will continue to
fall.
In summary, homeostatic theory asserts that under normal, unthreatening
conditions, SWB is controlled by the homeostatic system. As a consequence,
SWB will be steadily maintained within the normal range. However, under
conditions of threat, negative circumstances can challenge personal beliefs, thus
threatening SWB set-points. When the strength of negative influences exceeds the
homeostatic threshold, the homeostatic system will forfeit control to the external
agent, resulting in a drop in SWB. SWB homeostasis theory further posits that
SWB is maintained by stable forces, such as positive affectivity and a system of
buffers – both internal and external. A comprehensive discussion of the
homeostatic buffers is found in the section below.
41
Homeostatic Buffers
In an often changing and unpredictable environment, there will be threats to
individual SWB set-points. As a consequence, the homeostatic system will
constantly be required to defend these set-points as ongoing interaction with the
environment provides such challenges. When unexpected positive and negative
events threaten to shift SWB to abnormally high or low levels, the homeostatic
system calls on a number of defences that serve to maintain or restore set-point
equilibrium. The first of these, named external resources, assist in avoiding, or at
least assist to attenuate, negative environmental interactions. The role of second
type of defence, known as the internal buffers, is to ensure that harm associated
with any external threat is minimised such that it is incapable of diminishing the
normally positive sense of self that is SWB. The external buffers will be discussed
first.
External Buffers
External buffers refer to any resources in a person’s environment that can protect
an individual from the vast array of potentially negative events, situations and
experiences that life presents (Cummins, Lau, & Davern, unpublished). Examples
of these protective resources include supportive relationships, a financially secure
job, access to healthcare, child care facilities for working parents and a reliable car.
According to Cummins (2000b), the two most important external buffers for
people in general are money and relationships. To highlight the influence of
external buffers on SWB, the relationship between income and SWB will now be
discussed.
Income and Subjective Wellbeing: A Closer Look at the Interaction between
Person, Environment and SWB
As previously mentioned, early SWB researchers such as Campbell, Converse and
Rodgers (1976) and Andrews and Withey (1976), concluded that demographic
factors, such as income, exerted little influence on SWB. In fact, according to
Campbell et al., no more than 20% variance can be explained by combined
demographic factors (e.g., age, sex, income, race, education and marital status). In
a similar vein, Andrews and Withey report that these variables account for only
42
8% of total variance. In contrast to this view, contemporary SWB researchers,
such as Cummins (2000b) and Cummins, et al., (unpublished) argue that there
exists an intimate relationship between personal income and SWB. However, they
do not share the view that money can permantly shift SWB so as to create a
happier person. This is because SWB set-points are proposed to be under genetic
control and cannot be altered (Cummins, et al., 2003; Cummins et al.,
unpublished). Cummins (2000b) does, however, view income as a flexible
external resource that can protect SWB by preventing certain negative events from
occurring and by assisting in neutralizing the impact of potentially negative events.
Thus, the benefit of having money in terms of its capacity to protect SWB lies in
its ability to assist people to minimise challenges inherent within their
environment (Cummins, et al., unpublished). For example, people on high
incomes can afford to buy quality food and access to quality health care services
when needed. Likewise, good nutrition and access to medical services means that
wealthier people are less likely to suffer from debilitating illnesses associated with
malnutrition and are more likely to receive the best medical assistance when these
services are required. Poorer people, on the other hand, lack these resources and
are more susceptible to challenges in their environment. Thus, income can protect
against unhappiness and maintain SWB by providing resources that alleviate, and
in some cases, prevent harm or loss from occurring in the first place. Figure 5 is a
graphical representation demonstrating the influence of money on SWB.
Subjective
wellbeing
81
80
79
78
77
76
75
74
73
72
71
Total N  28,000
79.2
*
*
*
*
78.3
78.0
76.5
76.3
74.9
Normal Range
73.9
73.0
71.7
<$15
$15-30
$31-60
$61-90
$91-120
$121-150
$150+
Median
Household Income ($'000)
Figure 5:
Subjective wellbeing vs. household income (reprint from Cummins,
Lau, Mellor, & Stokes, in press)
43
Figure 5 presents the cumulative Australian Unity Wellbeing Index data on
household income based on approximately 28,000 respondents. As shown in
Figure 5, a steady increase in SWB can be observed from the lowest income
category (<AUS$ 15,000) up to AUS$ 91,000-120,000 - which is about three
times the median income. Beyond that level, further income increments have no
systematic effect to increase SWB. Data presented in Figure 5 are consistent with
the idea that SWB may be actively maintained within a managed system and that
money is a flexible resource that may defend SWB homeostasis. For example, at
the lower end of the income ranges, increases in household income are associated
with increases in SWB as more money buys more buffering in an environment
that threatens homeostasis. However, once a certain level of income is attained,
(approximately $91-120K), additional income is of little assistance because
additional income cannot ameliorate the kinds of challenges that may continue to
threaten homeostasis (Cummins, et al., in press). At this level of income,
challenges to homeostasis are unlikely to represent threats that can be negated by
money. Examples of these types of threats include relationship issues, anxieties
brought about by work and chronic health conditions.
Finally, according to Cummins (2000b), personal wealth can also influence SWB
in more subtle ways and he makes a number of predictions based on people’s
level of wealth. For example, people with money are generally more able to
arrange their environment in ways that ensure certain events and happenings occur
in a predictable fashion. For example, wealthy people can pay others to mow their
laws, clean their houses and look after their children and, through the process,
generate free time for relaxation and/or leisure type activities. Thus, having
money and more of it creates a greater sense of control and predictability over
one’s environment which will reinforce generally positive, stable levels of SWB.
A graphical representation of the power of money to impact a person’s day-to-day
experience of positive and negative events is presented in Figure 6.
44
Sad event
35
29.5
30
%
reporting
an
event
Happy event
32.4
28.2
26.1
26.0
25.1
24.8
25
22.9
20
15
21.3
17.6
10
<$15
$15-30
$30-60
$60-90
$90+
Household incom e ($'000)
Figure 6:
Income vs. happy and sad events (reprint from Cummins, Lau, &
Davern, unpublished)
Figure 6 represents an empirical demonstration of the power of money to
influence positive and negative experiences in a person’s life. The above data
form part of the Australian Unity Wellbeing Index: Report 14 (Cummins,
Woerner, Tomyn, Knapp, & Gibson, 2005). As part of this study, participants
were asked ‘Has something happened to you recently causing you to feel happier
or sadder than normal?’ As shown in Figure 6, people on very low incomes
(annual household income <15,000 per year) were more likely to have
experienced a sad event (32.4% of people) than a happy event (17.6% of people).
However, this trend disappeared at higher incomes. For example, of those who
live in households with an annual income greater than $90,000, 26% had
experienced a happy event compared with 25.7% who had experienced a sad
event. Interestingly, this differential influence decreases as a function of
increasing income, up to a gross household income of about $60-90K per year.
According to the data, at this level of income and beyond, people experience
relatively similar levels of positive and negative events.
In summary, external buffers include any resource in a person’s environment that
can prevent, or ameliorate, the potentially negative impact posed by a challenging
environment. In the present review, money was introduced as a flexible resource
and was argued to have an indirect influence on people’s wellbeing by buffering
against environmental challenges. The theory of homeostasis does not contend,
however, that personal income is synonymous with happiness. Rather, income
provides an individual with the opportunity to acquire resources that can optimise
45
and assist in maintaining the functioning of their SWB homeostatic system. Also,
as discussed, there is normally a weak correlation between objective indicators
such as income and SWB. Thus, consistent with this line of investigation, it is
argued that increases in personal wealth will not necessarily equate with
equivalent increases in SWB – at least not for those on above average incomes.
However, for low income earners, the relationship between more money and
SWB will be stronger. For example, giving poorer people more money will have
an indirect impact on their SWB by enabling them to acquire goods and services
that provide an adequate standard by which to live. Finally, data clearly indicate
the existence of ceiling effects at the upper end of the SWB spectrum such that
adaptation ensures that increases in wealth will not necessarily equate with longterm rises in SWB.
Internal Buffers
When external buffers are not present, or are incapable of preventing harm from
occurring, all is not lost. Humans have evolved a complex cognitive system and at
the heart of homeostasis resides a set of genetically programmed devices. These
devices, known as internal buffers, are brought into action when SWB is
threatened (Cummins & Nistico, 2002). Given that SWB set-points represent the
optimum setting for successful adaptation, the potential for disturbances (e.g.,
challenges to a persons subjective wellbeing) had to be controlled. According to
Cummins, et al., (unpublished), at the simplest level, these devices involve the
automatic processes of adaptation and habituation. These act over time to make us
less aware of daily challenges that threaten the normally positive view of the self.
Cummins et al., also suggest that these processes are assisted by the cognitive
buffers. The buffers use cognition to restructure reality in ways that aim to protect
SWB from the world of consciousness and maladaptive thought processes and
serve to minimise the impact of unavoidable negative experiences.
Self-esteem, optimism and perceived control have been grouped together by
Cummins and Nistico (2002) to represent three major aspects of cognition that
comprise satisfaction with the self and are important to SWB homeostasis setpoint regulation. Self-esteem, perceived control and optimism are among the most
researched cognitive protective factors and have been found to be related to SWB
46
(Cummins & Nistico). The importance of these constructs to SWB and their role
in SWB homeostasis will now be discussed.
The Cognitive Buffer System and Positive Cognitive Biases (PCB’s)
Homeostatic theory asserts the existence of several constructs involved in the
psychological process of SWB maintenance. Combined, these constructs form
what is known as the ‘cognitive-buffer system’ and include biases of self-esteem
(i.e. feelings of self worth) (Cummins & Nistico, 2002); control (i.e. perception
that that one can achieve desired outcomes through their own actions) (Thompson
et al., 1998); and optimism (i.e. the belief that one’s future is bright despite what
objective life circumstances may suggest) (Peterson, 2000). Biases of self-esteem,
control and optimism have several important features. First, they are non-specific.
That is, they do not relate to any specific skills, abilities or attributes to which a
person can evaluate themselves negatively against some external or comparative
criteria. Second, they relate only to abstract ideas that are detached from reality.
For example, the belief that one is better looking, smarter, luckier or more liked.
Finally, these biases are empirically unfalsifiable. For example, it cannot be
proven that one person is luckier than another person (Cummins & Nistico, 2002).
In this way, positive biases of self-esteem, control and optimism assist in the
maintenance of subjective wellbeing at a generally stable, positive level by
buffering the impact of potentially harmful transactions between a person and
their environment. A comprehensive discussion of each construct will follow.
Biases of Self-esteem
Self-esteem is one of the most widely cited variables in psychology and refers to
an individual’s sense of his or her value or worth, or the extent to which a person
values, approves of, appreciates, prizes, or likes themselves (Blascovich &
Tomaka, 1991). A popular conception of self-esteem is by Rosenberg (1965), who
defines it as “a favorable or unfavorable attitude toward the self ” (p. 15).
Satisfaction with the self is one of the strongest single predictors of SWB. In fact,
in a review of six studies, Cummins and Nistico (2002) found average correlations
ranging from 0.54 – 0.77 between these constructs. Numerous authors report
47
similar correlations between self-esteem and life satisfaction. For example,
Campbell (1981) found that self-esteem was the strongest predictor of life
satisfaction (r = .55) using a sample of adults in the United States. In a similar
vein, in a study involving college students from 31 nations (N = 13,118), Diener
and Diener (1995) report that at the individual level, self-esteem and LS correlate
at .47. Finally, according to Diener, Lucas, Oishi and Suh (1992), happy people
are more likely to report satisfaction with the self, whereas unhappy people are
more likely to report low satisfaction with the self. As part of this study, Diener et
al., asked participants to rate their level of happiness with eight domains – health,
finances, family, friends, recreation, religion, self and education. According to the
results, the participants who were the most happy rated the ‘self’ domain as their
best domain. In contrast, participants who were least happy rated the ‘self’ domain
as their worst domain. Thus, these authors found that a positive sense of self was
positively related to participants overall level of happiness.
In explanation of the link between self-esteem and SWB, Cummins, Gullone and
Lau (2002) propose that the key to self-esteem and its buffering effects for SWB
lie in self enhancement. More specifically, people with high self-esteem engage in
direct forms of self-enhancement. That is, they tend to associate themselves with
positive outcomes and positive identities, such that the self is rated more
favorably when compared to others.
Finally, it is important to note that similarly to the personality dimensions of
extraversion and neuroticism, self-esteem is regarded as being highly stable over
time (e.g. Block & Robbins, 1993; O’Malley & Bachman, 1979). However, selfesteem may be susceptible to short-time changes either due to intentional
intervention or life events.
In summary, research supports the contention that self-esteem is positively related
to life satisfaction and it has been suggested that the buffering effects of selfesteem on SWB lie in self enhancement.
48
Biases of Control
Perceived control is another component of Cummins’ cognitive buffer system and
is defined as “a generalized belief of an individual concerning the extent to which
he or she can control outcomes of importance” and “as a situational appraisal of
the possibilities of control in a specific stressful encounter” (Folkman, 1984, p.
839).
The relationship between perceived control and SWB is complex. The most
common conceptualisation of control is its division into internal and external
locus of control (Rotter, 1966). People with an internal locus of control generally
believe that events, situations and happenings are within their immediate control.
On the other hand, people with an external locus of control will generally perceive
events and outcomes to be caused by uncontrollable agents, for example, powerful
others or luck (Cummins et al, 2002). Control beliefs also refer to perceptions
people make about the degree to which they can protect themselves from
misfortune (Thompson, et al., 1998). Research suggests that such beliefs have
important consequences for individuals. For example, those with high levels of
perceived control over a stressful situation have been found to experience less
anxiety, tolerate more pain, perform better on tasks and have better emotional
outcomes (Thompson, 1981; Thompson & Spacapan, 1991). Furthermore, people
with high personal control beliefs have been found to cope better with chronic
illness (Affleck, Tennen, & Gershman, 1985) and display greater psychological
wellbeing (Remondet & Hansson, 1991). A lack of perceived control, on the other
hand, commonly associated with a variety of inward behaviors, has been found to
be associated with depression and a range of cognitive impairments (Peterson,
Maier, & Seligman, 1993).
Three types of control beliefs have also been identified – these are primary control,
secondary control and relinquished control. Primary control is considered the
most adaptive of the three and is characterised by efforts directed at changing ones
world to create a better ‘fit’ with their environment (Rothbaum, Weisz, & Snyder,
1982). Such primary control is evident in individuals who attempt to enhance
personal rewards by influencing existing realities, for example, people,
circumstances and events (Weisz, Rothbaum, & Blackburn, 1984). Thus, by
49
definition, primary control implies that a person will make active efforts to bring
about change.
In
secondary
control,
individuals
enhance
rewards
or
outcomes
by
accommodating to existing realities. In contrast to primary control, secondary
control involves adjusting ones thoughts and behavior to create a better fit with
the world as it presents (Weisz, et al., 1984). Secondary control is believed to be
most commonly exercised after attempts to gain primary control have failed
(Rothbaum et al., 1984) and includes a variety of strategies such as reattribution,
making sense of the situation and a belief in luck (Thompson et al., 1998).
Relinquished control is the third type of control belief and is associated with a
particular set of interrelated, problematic behaviors. These behaviors, often
directed inward, include passivity, withdrawal and submissiveness. According to
Rothbaum et al., (1982), uncontrollability theorists consider relinquished control
as an abandonment of attempts at gaining control and studies have found that
experiences designed to induce perceptions of uncontrollability lead to decreased
learning, decreased persistence, and depressed affect – key components of inward
behavior (Weisz, et al., 1984).
It has generally been found that a high sense of personal control is positively
related with SWB. For example, according to Peterson (1999), personal control is
often linked to wellbeing, whilst a lack of control to passivity, to the failure to
achieve goals and to illness. In a similar vein, Myers and Diener (1995) contend
that happy people are those who have a greater sense of personal control. These
authors also assert that individuals with strong personal control beliefs feel more
empowered, cope better with illness, achieve more goals and live more happily. In
further support of the link between perceptions of control and SWB, Cummins
and Nistico (2002), in a review of 9 studies, report average correlations ranging
from .35 and .53 between high perceptions of control and SWB. Thus, research
indicates that a high sense of perceived control is important for psychological
adjustment and wellbeing.
50
It has been suggested (e.g., Thompson et al., 1998), however, that secondary
control is less adaptive than primary control because of its association with
passivity and withdrawal type behaviors. It is also argued that this type of control
may actually be motivated by a need for high levels of primary control. According
to Rothbaum et al., (1982), controllability theorists fail to recognise that a need for
high levels of personal control may sometimes motivate behavior not typically
considered controlling. For example, attributions to powerful others may permit
vicarious control (a type of secondary control whereby an individual identifies
with another as being more powerful). Instead of challenging the more powerful
individual (as in primary control), submission and subservience to this person may
foster unity and strengthen ties, thus, enabling a weaker, less controlling person to
share in this power through association. Furthermore, within the literature,
Rothbaum et al., (1982) found that attributing outcomes to low ability, combined
with behaving in a passive and withdrawn manner, is often associated with
helplessness. However, in their view, behaving in this manner may reflect a form
of secondary control known as predictive secondary control (e.g., Averill, 1973;
Lazarus, 1966), which involves preparing oneself for future events so as to inhibit
unfulfillable expectations. Thus, it is believed that under certain circumstances,
secondary control strategies may be an adaptive form of control.
The Discrimination model (Thompson et al., 1998) offers an alternate view. This
proposes that the most adaptive control beliefs are context dependant, with
primary control considered most adaptive when a situation is controllable and
secondary control considered most adaptive when a situation is resistant to change.
If a person’s environment can be manipulated so that a rewarding outcome is
more probable, perceptions of control, which are likely to be correct, will
motivate behavior directed towards that outcome. Thus, in this event, a high level
of perceived control is adaptive as it will often lead to satisfying consequences
which in turn, reinforces SWB. On the other hand, in situations where actual
control is low, perceptions of control may lead to disappointment and failure as
persistent efforts to change ones situation are often in vain (Thompson et al.,
1998). In situations such as this, where actual control is low, embracing secondary
control beliefs, such as acceptance, may be more constructive in that they can help
re-establish an overall sense of personal control through alternative means. For
51
example, accepting a lack of control in some areas may reinforce control beliefs in
other areas where the probability of control may be more realistic and where
efforts are more likely to be rewarded.
In summary, the terms ‘primary control’ and ‘secondary control’ discriminate
between agents of influence. When the self is seen as the more powerful agent,
control beliefs will be primary and when agents more powerful than the self have
been acknowledged and understood, control beliefs will be secondary.
Relinquished control occurs when a person has lost all faith in his/her ability to
control happenings in his/her life. Research suggests that primary control is the
more adaptive form of perceived control, is positively related to SWB and is the
type of control that we should strive for. However, situations will arise when
attempts at changing ones circumstances are met with resistance. Under these
circumstances, exercising secondary control may be more adaptive. As stated by
Lazarus (1981), neither form of perceived control is thought to exist in pure form.
Further, according to this author, both processes are intertwined and require
negotiation and compromise.
Biases of Optimism
Optimism is the final component of Cummins’ cognitive buffer system and is
defined as a perception that the future will be to the perceiver’s advantage or for
his/her pleasure (Peterson, 2000). This trait also refers to the general expectancy
of favorable outcomes in one’s life (Scheier & Carver, 1985). It seems intuitive
that a positive outlook on life would be related to SWB. In fact, optimism has
been found to correlate with LS generally between .40 and .77 (e.g., Diener et al.,
1999; Lucas et al., 1996; Olason & Roger, 2001). Furthermore, according to De
Neve (1999), who conducted a Meta-analysis of 137 personality constructs, the
happiest people are those who characteristically explain their lives in optimistic,
adaptive ways. In a similar line of investigation, several authors have found that
optimism inversely relates to depression (e.g. Pyszczynski, Holt, & Greenberg,
1987; Scheier & Carver, 1985). This is consistent with the work of Abramson,
Seligman & Teasdale (1978) who claim that hopelessness about the future is the
genesis of depression.
52
There is also large body of research which supports the notion that dispositional
optimism is related to greater psychological adjustment and SWB. For example,
Aspinwall and Taylor (1992) found that amongst college students, expecting
positive future outcomes predicted better adjustment to the stressors of college life.
Additionally, optimism has been found to correlate with high self-esteem
(Peterson, 2000), happiness (Dember & Brooks, 1989) and is predictive of life
satisfaction (Cummins & Nistico, 2002; Lucas, Diener, & Suh, 1996; Scheier &
Carver, 1985).
According to Diener, Suh, Lucas and Smith (1999), the tendency to expect
favorable outcomes positively influences SWB via the attainment of positive
experiences. In support of this argument, a study by O’brien, VanEgeren and
Mumby (1995), found that optimists both underestimate their susceptibility to
health problems and report lower levels of stress and physical symptoms (despite
performing a similar number of preventative health care behaviors) than nonoptimists. Interestingly, even though optimists underestimate levels of personal
threat, they still engaged in frequent, adaptive and health promoting behaviors.
In summary, optimism has been positively linked to SWB and is associated with a
number of positive outcomes; most likely through expectation of favorable
outcomes. Optimism is an important component of Cummins’ homeostatic theory
of SWB and is argued to act as a buffer and assist in maintaining SWB during
times of challenge.
SWB HOMEOSTASIS: A WORKING MODEL
Homeostasis theory asserts that SWB is maintained by stable forces such as
positive affectivity and a system of cognitive buffers. As discussed, the buffer
variables include positive biases of self-esteem (i.e. feelings of self worth)
(Cummins & Nistico, 2002), perceived control (i.e. perception that that one can
achieve desired outcomes through their own actions) (Thompson et al., 1998) and
optimism (i.e. the belief that one’s future is bright despite what objective life
circumstances may suggest) (Peterson, 2000). These buffers interact with
experience from the environment, and, under normal, unthreatening conditions,
SWB is maintained by the homeostatic system at a steady level within the normal
53
range. Under conditions of threat, the emotional responses to events can challenge
SWB, thus threatening SWB set-points; but SWB will still be maintained within
its normal range as long as homeostasis holds. However, when the strength of
negative influences exceeds the homeostatic capacity to counter such challenge,
the homeostatic system will forfeit control to the external agent, resulting in a
drop in SWB below its set-point-range.
This model proposes a pathway between external, objective conditions and their
relationship with SWB. It outlines processing that may occur at each level
between external input and SWB (output) (Cummins & Nistico, 2002). A model
for subjective wellbeing homeostasis is presented in Figure 7.
Homeostatic system NOT under challenge
Homeostatic system under challenge
Subjective
Wellbeing
Affectivity &
Personality
Subjective
Wellbeing
Cognitive Buffers
Affectivity &
Personality
Perceived Met
Needs
Perceived Unmet
Needs
Successful
Adaptation
Compromised
Adaptation
Mild Extrinsic
Conditions
Aversive Extrinsic
Conditions
Figure 7:
Cognitive Buffers
A homeostatic model of subjective wellbeing (reproduced from
Cummins & Nistico, 2002)
According to this model, the nature of processing that occurs at each level is
determined by the strength of challenge from environmental pressures (such as
losing ones job) and intra-psychic challenge, such as from anxiety and the
challenge these conditions pose to the homeostatic threshold. Under conditions of
54
relatively low challenge (the left hand side of Figure 7), habituation or adaptation
will occur; and needs will be perceived as having been met and the buffer system
will remain relatively inactive. When some external agent threatens wellbeing
(e.g., when a person loses his/her job) the cognitive buffers will become activated
in an attempt to maintain SWB within the normal, adaptive range (right hand side
of Figure 7). Provided the level of challenge is not too severe and a person has
resources to help counter any loss, homeostasis will hold and a person will
maintain his/her wellbeing. During a strong challenge, however, threats against a
person’s SWB may be so severe that adaptation to the stressor is markedly
reduced. As a consequence, needs may be perceived as unmet, positive beliefs
about the self may be compromised and the result is a likely challenge to the
ability of the system to maintain SWB. When external circumstances have
become so overwhelming that the homeostatic system has been defeated, SWB
will fall below the normal set-point range and this is characterised as depression
(Cummins & Nistico, 2002)
It is important to note that in previous models of homeostasis theory (e.g.,
Cummins & Nistico, 2002) personality was believed to have a profound influence
on the second and third levels of processing by mediating the relationship between
external experiences and SWB. As mentioned previously, the personality
components most strongly associated with SWB maintenance are extraversion and
neuroticism. However, recent work (e.g., Davern, et al, 2007), suggests that core
affect, not personality, is driving SWB and is responsible for maintaining setpoint levels in SWB homeostasis (Cummins, et al., unpublished).
Set-point Robustness, Fragility and Depression
In summary, SWB can be categorised as a state that is normally positive, stable
and held within a narrow range of values for each individual. This set-point range
for each person is genetically determined and actively maintained by a
homeostatic system. However, when the level of challenge to homeostasis
exceeds its threshold, the system will fail and SWB will drop below the set-point
range. According to Cummins, et al., (unpublished), the homeostatic system can
be inherently robust, or it can be fragile, depending on a number of factors. First,
it is argued that everybody has a genetic set-point for SWB. That is, set-points
55
constitute an individual difference. Some people have a high set-point level while
others have low set-point. A person who has the luxury of a set-point far from the
danger zone is at a far lower risk of depression than a person with a set-point
closer to the danger zone. The relationship between SWB and depression,
measured using the Depression Sub-Scale of the Depression, Anxiety and Stress
Scale (DASS; Lovibond & Lovibond, 1995) is presented below in Figure 8. Data
are taken from Davern (2004).
80
79.7
77.7
78
76.0
76
74.4
74
72.0
72
PWI
70.9
71.0
70
68
66
64
65.0
63.3
62
60
0
1-2
3-4
5-6
7-8
9-10
11-12
13-14
15-16
Depression scores
Figure 8: The relationship between SWB and depression (reprint from Cummins,
Lau, & Davern, unpublished)
As shown in Figure 8, the pattern of results conforms to the homeostatic model
described above. Depression scores, represented by 7 items measuring symptoms
of low energy, loss of purpose, low self-esteem etc., are presented along the
horizontal axis and represent various levels of challenge to the homeostatic
system. At zero challenge, SWB (presented along the vertical axis) is at the upper
end of the set-point range. However, as the level of challenge increases from 0 – 8,
the value of SWB decreases as a linear function of challenge until it approximates
the lower homeostatic threshold of 70 points (phase ‘b’ in Figure 4). Consistent
with homeostasis theory, at the depression range of 9-12, homeostasis ‘holds-theline’ and SWB is maintained. When a persistent and noxious challenge to
homeostasis overwhelms the system (represented by a depression score of 13 or
greater), control over SWB passes from the system to the challenging agent. The
result is a drop in SWB below the set-point range.
56
Using group mean scores, the relationship between depression and SWB
conforms nicely to the model. However, matching SWB measured using the PWI
and depression scores for individuals is not as clear cut and represents a challenge
for researchers. Given that it is suspected that people have different set-point
levels for SWB, it is not possible to explicitly state that an individual score of say
60 points is representative of a person experiencing a particular level of
homeostatic challenge. For example, 60 points may represent a normal level of
SWB for a person with a low set-point, however, for a person with a high setpoint (say 85 points), 60 points is a likely indication that this person’s SWB has
been severely challenged, even defeated. Thus, Cummins, et al., (unpublished)
suggest caution when interpreting these data. However, several diagnostic
approximations can be made based on the following information:
1.
That each person has a genetically determined set point that lies
somewhere in the 55-95 point range (Cummins, et al., unpublished), with
the average set-point for populations at 75 points.
Then:
If an individual score lies above 70 points, it is likely that the homeostatic system
is functioning normally and the person is not depressed. This is because 70 points
is argued to represent the lower border line of resistance of the homeostatic range
for life satisfaction for the average person and that which the homeostatic system
is defending (Cummins, 2003). Thus, on an average level basis, scores above 70
points most likely represent a homeostatic system that is functioning normally.
People who score below 70 points, however, are likely to be experiencing a level
of challenge that has altered their wellbeing below their set-point. People with
SWB below 70 points are at greater risk of depression.
2.
If an individual score lies below 50 points, then the person is likely to be
depressed, regardless of whether he/she has a low set-point. According to
Cummins, et al., (unpublished), approximately 4.4% of individuals can be
categorised into this group.
57
3.
Although the diagnostic meaning of scores between 50 and 70 points is
uncertain, it can be predicted that as scores progressively fall below the 70
point line of resistance and increasingly approximate 50 points, the greater
the probability that a person is homeostatically challenged, even defeated,
rather than having a low set-point.
In summary, Cummins, et al., (unpublished) offer the following guidelines for the
interpretation of SWB scores: 70+ points = normal; 51-69 points = either a low
set-point or strong homeostatic challenge, even defeat for those with high setpoints (e.g., 80+ points); 50 points or less = homeostatic defeat and depression.
SUMMARY AND STUDY ONE AIMS
There is general agreement within the literature that subjective wellbeing is a
construct comprising both a cognitive and affective component (e.g., Campbell,
Converse, & Rodgers, 1976; Diener & Diener, 1996; Steel & Ones, 2002;
Veenhoven, 1994). However, a recent study by Davern, et al., (2007) has
suggested that subjective wellbeing is essentially driven by affects, with cognitive
discrepancies playing an important but subsidiary role. A major aim of study one
is to test Davern et al’s., finding that SWB is driven by core affect, as initially
defined by Russell (2003).
A second major issue concerns the remarkable stability in subjective wellbeing
frequently observed within the literature. This stability has prompted a number of
authors (e.g., Cummins, 1995, 1998; Eid & Diener, 2004; Headey & Wearing,
1989, 1992) to suggest that individuals may have a ‘set-point’ level of wellbeing.
In an attempt to explain stability in SWB, it has been proposed that personality is
the driving force behind satisfaction judgments (e.g., Brickman & Campbell, 1971;
DeNeeve & Cooper, 1998; Emmons & Diener, 1985; Headey & Wearing, 1989,
1992; Vitterso, 2001; Vitterso & Nilsen, 2002). However, research conducted by
Davern et al., suggests that contrary to previous research, personality is not the
driving force behind SWB. The second major aim of study one is to test the
relative contribution of affect and personality to SWB.
58
The present study aims to test various hypotheses made using homeostasis theory
(e.g., Cummins, 1998; Cummins & Nistico, 2002). According to this theory, in a
manner analogous to the homeostatic maintenance of body temperature, SWB is
actively controlled and maintained (Cummins & Nistico). Additionally, according
to this theory, homeostasis acts to protect core affect (which is argued to
approximate the SWB set-point) so that each person can maintain a positive level
of wellbeing. Consistent with SWB homeostasis, it is hypothesised that core affect
will explain greater unique variance in life satisfaction than in subjective
wellbeing because life satisfaction is more abstract and that the buffer variables
will explain significant variance in SWB above core affect in a group of people
experiencing homeostatic challenge. It is also hypothesised that the mean SWB
score will lie within the Australian adult normative range of 73.43 to 76.43%
(Cummins et al., 2006). Finally, it is hypothesised that core affect is driving the
relationship between SWB and the related constructs of perceived control,
optimism, self-esteem, extraversion, emotional stability and MDT.
59
CHAPTER 3: STUDY ONE
METHODOLOGY
Participants
The 146 participants were attending various high-schools in the Melbourne
metropolitan region and Geelong. This sample consisted of 66 males (45.2%) and
80 females (54.8%), representing 49 students in year 7 (33.6%), 20 students in
year 8 (13.7%), 19 students in year 9 (13.0%), 35 students in year 10 (24.0%), and
23 students in year 11 (15.8%). Participant’s ages ranged from 12 – 17, with a
mean age of 14.4 years.
Questionnaire
A 60-item paper and pencil questionnaire titled The Young Australian Wellbeing
Index (see Appendix A) was self-completed by each student, under conditions of
information privacy, in their regular classroom. Four items obtained demographic
information - these were age, sex, year level at school and post-code of residence;
and the remaining 49 items comprised 7 sub-scales measuring the constructs
described below. Students responded to all scale items using an 11 point, enddefined scale.
3.1
Major Dependent Variable and Other Variables
Life Satisfaction (LS) and Subjective Wellbeing (SWB)
The measure of life satisfaction (LS) asks ‘How satisfied are you with your life as
a whole’ (0 = Very Dissatisfied; 5 = Neutral; 10 = Very Satisfied) (see Appendix
A, item 5). Due to its high level of abstraction and non-specificity, the ‘life as a
whole’ item is the closest approximation to core affect of any question involving
‘satisfaction’.
The Personal Wellbeing Index-School Children (PWI-SC; Cummins & Lau, 2005)
has been employed as a second measure of SWB (see Appendix A, items 6-12).
The PWI-SC creates a composite variable, calculated by averaging life
60
satisfaction scores across 7 domains. The domains have been selected to represent
the first level deconstruction of satisfaction with ‘life as a whole’ (Cummins,
1996). Measured on the same scale as LS, domains include 1) the things you have
(standard of living); 2) health; 3) things you want to be good at (achieving in life);
4) getting on with people you know (personal relationships); 5) safety; 6) doing
things outside of home (community belonging); and 7) what may happen later on
in life (future security).
Items on the PWI-SC are based on the Personal Wellbeing Index – Adult (PWI-A,
International Wellbeing group, 2005) and have been modified in that several items
were reworded to increase understanding and relevance for high-school age
children. The International Wellbeing Group (2005) report good internal
reliability using the PWI-A (Cronbach’s  between .70 and .85). In the present
study, the corresponding reliability co-efficient of the PWI-SC was .83.
Core Affect
Core affect was measured using affect terms that represent the two dimensions of
the Circumplex Model of Affect (pleasant-unpleasant, activated-deactivated;
Russel, 2003) (see Appendix A, items 14-23). Inclusion of affective adjectives
was based on research suggesting that each dimension of the circumplex is best
represented by terms located closest to its four poles (0°, 90°, 180°, 270°) (Russell,
2003). Thus, representing the Pleasant pole (90°) are the adjectives happy and
content; representing the Unpleasant pole (270°) are discontent and unhappy;
representing the Activated pole (0°) are active, alert and excited; and representing
the Deactivated pole (180°) are sleepy and quiet. These 9 items include the three
items found by Davern, et al., (2007) to represent core affect (content, happy and
excited). Subsequent testing will allow the confirmation or extension of the
original core affect definition. Participants responded to each item as follows:
‘Please indicate how each of the following describes your feelings when you think
about your life as a whole: How… (insert affective adjective)… do you generally
feel?’ (0 = Not At All; 10-extremely).
61
Self-esteem
Rosenberg’s self-esteem scale (Rosenberg, 1989) was used to measure global selfesteem (see Appendix A, items 40-49). This scale consists of 10 items that require
participants to indicate their strength of agreement with statements such as ‘On
the whole, I am satisfied with myself’ (0 = Strongly Disagree; 5 = Neutral; 10 =
Strongly Agree). The scale, which is suitable for use with children (Cronbach’s 
= .73; Rosenberg, 1965), yielded a Cronbach’s  = .89 and a single composite
variable is computed.
Optimism
Optimism was measured using the Life Orientation Test – Revised (LOT-R;
Scheier, Carver, & Bridges, 1994). The LOT-R, which is suitable for use with
children (Scheier et al., 1994), is a 10-item measure of dispositional optimism and
pessimism. Consistent with the view that a bi-dimensional relationship exists
between optimism and pessimism (Chang, Maydeu-Olivares, & D’Zurilla, 1997),
only the three items measuring the construct optimism were included in the
present study (see Appendix A, items 28-30). These three items, which include ‘In
uncertain times, I usually expect the best’ (0 = Strongly Disagree; 5 = Neutral; 10
= Strongly Agree), demonstrated good internal reliability (Cronbach’s  = .89). A
composite variable was computed.
Perceived Control
The two dimensions of primary and secondary control strategies were measured
using The Personal Perceived Control Scale (PPCS; Holloway, 2003, unpublished
thesis). Three items measuring each dimension began with the statement ‘When
bad things happen to you, how do you cope with them…?’ (see Appendix A,
items 31-36). For example, ‘I ask others for help or advice’ (primary control), ‘I
remind myself that something good may come of it’ (secondary control).
Responses were anchored at either pole by ‘Never’ (0) and ‘Always’ (10).
A further 3 items measured relinquished control (see Appendix A, items 37-39).
However, Principal Component Factor Analysis conducted on all 9 control items
62
revealed that only the items measuring primary and secondary control formed
separate factors. Due to poor inter-item reliability of the 3 relinquished control
items, these were excluded from further analysis. Despite being the first time that
this scale has been used on a sample of adolescents, Chronbach’s alpha for the
primary and secondary control items were .77 and .74 respectively.
Personality
The personality dimensions of extraversion and neuroticism were measured using
four items from the Ten-Item-Personality-Inventory (TIPI; Gosling, Rentfrow, &
Swann Jr., 2003) (see Appendix A, items 24-27). This is a brief, reliable measure
of the Big Five Factors of personality (see Gosling et al., 2003). Two items
measuring each of the personality dimensions of extraversion and neuroticism
(reverse coded as emotional stability) were included since these are the two
personality constructs most related to subjective wellbeing (DeNeve & Cooper,
1998). Cronbach’s alpha for extraversion in the present sample was acceptable
(Cronbach’s  = .73) but was poor for stability (Cronbach’s  =.53).
OTHER MEASURES
School Satisfaction
As all research participants attend high-school, a single item measuring
satisfaction with school was included to determine whether this fulfilled the
criteria for a new domain within the PWI-SC (see Appendix A, item 13).
Multiple Discrepancies Theory
MDT was assessed using 7 items adapted from Michalos’ (1985) original 11-point
reference scale (see Appendix A, items 50-56). Although modified for the purpose
of improving comprehension amongst school age children, items still retained the
essence of the original. For example, the original item ‘Considering your life as a
whole, how does it measure up to the average for most people your own age and
sex in this area. Generally, does your life offer you far less than what is offered
the average person, more, etc.?’ was shortened to ‘How does your life compare to
the average for most people your own age?’ (0 = Far below average; 5 = About
63
average; 10 = Far above average). Reliability for the modified version of
Michalos’ scale was high (Cronbach’s  = .85) and a composite variable was
computed.
PROCEDURE
After obtaining approval from the Deakin University Human Research Ethics
Committee – health and behavioural sciences sub-committee, approval was sought
from the Department of Education and Training and The Catholic Education
Office. Approval from these organisations must be obtained prior to conducting
research in government and Catholic high-schools. Following approval from these
organisations, I was required to undergo a background police check, which is a
requirement when entering schools.
Once these authorities had given their approval, various high schools in the
Melbourne metropolitan region and Geelong were contacted by phone or e-mail.
A representative from each school was briefed on the proposed study, its aims,
obligations as a participating school and responsibilities of the researcher. Of the
21 schools contacted, three agreed to take part in the study.
Following approval by school authorities, potential research participants were
approached by me and handed parent/guardian consent forms. Students who
returned signed forms were then handed participant consent forms and at the time
of distribution, were informed as to the nature of the study. These included their
obligations as a participant (such as time commitments), the procedure for
returning the anonymous questionnaires upon completion, and their right to
withdrawal their participation at any time.
Each student who volunteered to participate then received an envelope containing
a plain language statement and a questionnaire. Participants were given time in
class to complete their questionnaire, seal it in an envelope that was provided and
return it to myself or the classroom teacher (whoever was present). In the latter
situation, questionnaires were mailed by the teacher to Deakin University. Upon
64
completion of the study, participants had the opportunity for debriefing and to
obtain a copy of the results.
RESULTS
3.2
Data Screening and Preliminary Analyses
SPSS software (version 12.0.1) was used for data screening and analysis.
To standardise data, scores have been converted to a Percentage of Scale
Maximum (%SM). For any scale that is rated 0-x, %SM is calculated through the
formula:
X  k min
k max  k min
X
kmin
kmax
=
=
=
3.2.1
Missing Data
x 100
the score or mean to be converted
the minimum score possible on the scale
the maximum score possible on the scale
SPSS frequency output revealed that the frequency of missing data for all
variables across the entire data set was less than 5%. As recommended by
Tabachnick and Fidell (2001), given that missing data appeared random, values
were replaced by regression - a sophisticated technique of data replacement that
uses linear regression analytic techniques to most accurately predict missing data.
Following the replacement of missing values, composite variables were computed
on scales with multiple items. These composite scores were used for the
remaining data screening and analysis.
65
3.2.2
Outliers
Examination of z-scores revealed univariate outliers on domain satisfaction
variables and core affect items. Comparison of mean scores on these variables
with corresponding means trimmed at the upper and lower 5% revealed that none
of these outliers significantly influenced mean scores on key variables. As a
consequence, univariate outliers were included in analyses (Pallant, 2001). No
multivariate outliers were identified with a Mahalanobis distance greater than
20.515 (critical x2 = 20.515, p< .001), a criterion recommended by Tabachnick
and Fidell (2001) for the corresponding degrees of freedom.
3.2.3
Normality and Linearity
Normality was assessed across the entire data set. Using the SPSS descriptive
statistics function, negative skews were found in the following variables:
satisfaction with health (z = -4.85), relationships (z = -6.89), safety (z = -6.8),
future (z = -3.5), the PWI (z = -4.89), primary control (z = -4.3), life satisfaction (z
= -6.77) and satisfaction with school (z = -6.18). According to Cohen and Cohen
(1983), skewness and kurtosis are acceptable within the range of -7.0 to 7.0. Thus,
no variables underwent transformation.
3.2.4
Multicollinearity and Singularity
All major independent variables, including LS, SWB, core affect, personality,
MDT,
perceived
control,
optimism
and
self-esteem
were
tested
for
multicollinearity and singularity. The highest correlations were between the affect
adjectives happy and content (r = .76) and between SWB and happy (r =.74)
Tabachnick & Fidell (2001) suggest that a researcher should think carefully
before including two variables with a bivariate correlation of greater than .70 due
to risk of inflated correlations between these variables. However, it was expected
that some affect adjectives would correlate highly due to shared core affect.
According to theory, due to the abstract nature of questioning, core affect will
have a profound influence over responses to items such as ‘How happy/content do
you generally feel’. Thus, inflated correlations between these items can be
66
attributed to shared variability in core affect. Given that the aim of this research is
to examine the affective nature of SWB, it is necessary to include both ‘content’
and ‘happy’ in analyses so that it can be precisely determined which affect terms
contribute significant variance to the prediction of LS and SWB. Based on this
reasoning, all items and composite scores were retained.
3.2.5
Sample Size
According to Tabachnick and Fidell (2001), the criterion for multiple regression
analysis is:
N  50  8m
where N= minimum number of cases and
m = number of IV’s
In the present study, the maximum number of independent variables entered in
any one regression analysis is 10. According to the rule:
N > 50 + 8 x 10
it is recommended that a minimum of 130 cases are needed for adequate statistical
power. With 146 cases, the present study meets this power requirement for all
major analyses.
3.3
Core Affective Adjectives as Predictors of SWB
Using 9 core affect adjectives as predictors, the aim of the first two multiple
regression analyses was to determine which of these contributed most strongly to
the prediction of life satisfaction (LS) and SWB. Separate analyses were
performed since people use different response strategies when making LS and
SWB judgements (e.g., Schwarz & Strack, 1999; Schwarz, Strack, Kommer, &
Wagner, 1987).
67
The following hypotheses have been made:
1.
That the mean score for LS will approximate the mean score for SWB and
that the mean SWB score will lie within the Australian adult normative
range of 73.43 to 76.43%
2.
That adjectives located on the pleasantness-unpleasantness axis of the
Circumplex Model of Affect (e.g., ‘happy’ and ‘content’) will dominate and
explain significant variance in LS and SWB
3.
That core affect will explain a significant portion of the total variance in LS
and SWB and greater unique variance in LS than in SWB
4.
That the buffer variables will explain significant variance in SWB above
core affect in a group of people experiencing homeostatic challenge
5.
That core affect is driving the relationship between SWB and related
constructs. These include perceived control, optimism, self-esteem,
extraversion, emotional stability and perceived discrepancy judgments.
To test the first hypothesis, that the mean score for LS will approximate the mean
score for SWB, a t-test was conducted. Confirming this hypothesis, the mean
score for LS (M = 75.00, SD = 18.05) did not significantly differ from the mean
score for SWB (73.90, SD = 13.95), t(145)= -1.073, p=.285 (see Table 2).
68
Table 2:
Means, standard deviations and correlations between variables (N=146)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
1.
LS
75.00
18.05
-
2.
SWB
73.90
13.95
.73
-
3.
Happy
74.86
18.13
.68
.74
-
4.
Content
73.49
17.83
.65
.68
.76
5.
Unhappy
35.88
21.72 -.53 -.49 -.59 -.45
-
6.
Discontent 38.54
23.31 -.43 -.33 -.37 -.40
.65
7.
Active
69.32
22.09
.38
.43
.39
.37 -.19 -.16
-
8.
Alert
70.48
20.25
.58
.52
.49
.53 -.27 -.22
.48
-
9.
Excited
72.47
19.43
.54
.59
.69
.65 -.30 -.22
.41
.55
10. Sleepy
58.70
25.97 -.21 -.13 -.16 -.09
.20 .15 -.19 -.16 -.14
11. Quiet
47.74
26.75 -.23 -.27 -.28 -.24
.25 .14 -.24 -.13 -.27 .38
-
-
Also, confirming the first hypothesis and consistent with SWB homeostasis theory
and Australian adult normative statistics, the mean score for SWB (73.90) was
within the normative range of 73.43 to 76.43%SM (Cummins et al., 2006). It is
also noteworthy that LS and SWB mean score standard deviations are similar to
those of the adult normative statistics. The standard deviation for SWB amongst
the adolescent sample was 13.95, compared with 12.29 for adults (a difference of
+ 1.66 percentage points); and the standard deviation for LS was 18.05 for the
adolescent sample compared with 17.38 for adults (a difference of + .67
percentage points). Thus, LS and SWB standard deviations are comparable to
adults.
In partial support of the second hypothesis, that the pleasant-unpleasant axis of the
Circumplex Model of Affect would dominate, correlations presented in Table 2
indicate that LS and SWB share the strongest relationship with adjectives located
at the ‘pleasant’ pole of Russell’s (2003) Circumplex Model of Affect. These
include the adjectives ‘happy’ (r = .65 for LS and .74 for SWB) and ‘content’ (r
= .65, .68). This finding is consistent with that of Davern (2004) who also found
that the adjectives happy and content shared the strongest relationship with LS
and SWB.
11.
-
69
Table 3 and Table 4 provide details of the relative contribution of each IV to the
prediction of LS and SWB and include the unstandardised (B) and standardised
regression coefficients (β), squared semi-partial correlations (sr2), R2 and adjusted
R2 and unique and shared variance. According to Tabachnick and Fidell (2001), in
standard multiple regression, sr (semi-partial correlation) expresses the unique
correlation between an IV and a DV when the influence of other IV’s in the model
are removed. It is, thus, a very useful measure of the importance of a predictor. In
SPSS, sr values are provided in SPSS output under the column ‘part’ and, when
squared, provide the unique contribution of an IV to the total variance of a DV.
Table 3:
Predicting LS by nine affect adjectives
Variable
1. Happy
2. Content
3. Unhappy
4. Discontent
5. Active
6. Alert
7. Excited
8. Sleepy
9. Quiet
DV: Life Satisfaction
.68
.65
-.53
-.43
.38
.58
. 54
-.21
-.23
* p<.05; ** p<.01; *** p<.001
Shared Variance = .49
B
.25*
.20*
-.12
-.07
.02
.25***
.00
.05
.00
Unique Variance = .08
β
.26
.20
-.15
-.09
.02
.28
-.00
-.06
.00
sr2
.02
.01
.01
.00
.00
.04
.00
.00
.00
R2 = .60
Adjusted R2 = .57
The R for this regression is significantly different from zero, F(9, 136) = 22.494, p
< .001. Three affects contributed significant unique variance to the prediction of
LS: alert (sr2 = .04), happy (sr2 = .02) and content (sr2 = .01). Altogether, 57% of
the variability in LS can be predicted from scores on these nine affect items,
indicating a significant influence of core affect on LS.
Table 4 provides details of the relative contribution of each affect to the prediction
of SWB.
70
Table 4:
Predicting SWB using nine affective adjectives.
Variable
DV:SWB
B
β
sr2
1. Happy
.74
.29***
.38
.04
2. Content
.68
.17*
.21
.02
3. Unhappy
-.49
-.07
-.11
.00
4. Discontent
-.33
-.01
.02
.00
5. Active
.43
.07
.11
.01
6. Alert
.52
.09
.12
.01
7. Excited
.59
.03
.04
.00
8. Sleepy
-.13
.02
.03
.00
9. Quiet
-.27
-.02
* p<.05; ** p<.01; *** p<.001 Unique Variance = .08
Shared Variance = .51
-.05
.00
R = .62
Adjusted R2 = .59
2
The R for this regression was significantly different from zero, F(9, 136) = 24.412,
p < .001. However, this time, only two affect items contributed significant, unique
variance to the prediction of SWB: happy (sr2 = .04) and content (sr2 = .02).
Consistent with the second hypothesis, affects located on the pleasant-unpleasant
axis of the circumplex (e.g., happy and content) dominated. Also, in line with the
third hypothesis that core affect will account for more variance in LS and SWB,
core affect explained 57% of the variance in LS and 59% of the variance in SWB.
However, inconsistent with the second part of the third hypothesis, results
indicated that affective adjectives accounted for the same amount of unique
variance in LS as in SWB (8.0%). To reiterate, it was hypothesised that affect
adjectives would account for greater unique variance in LS than SWB given that
LS is more abstract than the domains and core affect concerns the more abstract
component of SWB.
Additional analyses were conducted to determine whether affect adjectives would
explain more unique variance in LS than SWB when people with low SWB were
removed. The rationale for these analyses stems from the assumption that the
sample may include people considered ‘at-risk’ for depression. It is possible that
71
low scores on key affects contributed by these people may explain the finding of
no difference in the amount of unique variance explained in LS and SWB by core
affect across the entire sample. More specifically, in explanation of this, when the
homeostatic system is defeated, core affect will lose its affiliation with both LS
and SWB. However, it will lose this more in relation to LS because of its higher
degree of normal dependence on core affect.
The results of these analyses are presented in Table 5, Table 6 and
Table 7.
Table 5:
Means, standard deviations and correlations between variables (n=139 >50%SM)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
1. LS
76.83 16.02
-
2. SWB
75.80 11.15
.65
-
3. Happy
76.91 15.64
.59
.63
-
4. Content
75.04 16.35
.61
.60
.63
5. Unhappy
34.02 20.07
-.44 -.33
6. Discontent
37.67 23.20
-.43 -.29 -.33 -.36
7. Active
70.79 21.23
.29
.35 -.29
.29 -.07
-.13
-
8. Alert
71.65 19.10
.55
.52
.35
.55 -.21
-.25
.44
-
9. Excited
73.96 18.48
.45
.53
.52
.63 -.21
-.18
.36
.54
10. Sleepy
57.94 25.54
-.19 -.10
.53 -.09
.21
.13 -.19 -.17 -.09
-
11. Quiet
46.40 25.90
-.16 -.18 -.10 -.19
.19
.09 -.22 -.11 -.20
.34
11.
-
.60 -.32
.66
-
-
As expected, results presented in Table 5 indicate that the mean score for LS,
SWB and key affects has risen as a consequence of removing cases in the negative
range for SWB. Also, correlations between variables have decreased, however, it
is possible that this has occurred as a result of range restriction caused by the
splitting procedure.
10.
-
72
Table 6:
Predicting LS by nine affect adjectives (n = 139 >50%SM)
Variable
DV: Life Satisfaction
B
Β
sr2
1. Happy
.59
.22*
.21
.02
2. Content
.61
.25**
.26
.03
3. Unhappy
-.44
-.09
-.11
.01
4. Discontent
-.43
-.09
-.13
.01
5. Active
.29
.02
.03
.00
6. Alert
.55
.22***
.26
.04
7. Excited
.45
-.04
-.05
.00
8. Sleepy
-.19
-.03
-.05
.00
-.16
.01
9. Quiet
* p<.05; ** p<.01; *** p<.001 Unique Variance = .11
Shared Variance = .38
.02
.00
2
R = .52
Adjusted R2 = .49
Table 7:
Predicting SWB using nine affective adjectives (n = 139 >50%SM)
Variable
DV:SWB
B
β
sr2
1. Happy
.63
.22**
.30
.03
2. Content
.60
.15*
.22
.02
3. Unhappy
-.33
-.03
-.06
.00
4. Discontent
-.29
-.01
-.01
.00
5. Active
.35
.05
.10
.01
6. Alert
.52
.10*
.18
.02
7. Excited
.53
.02
.04
.00
8. Sleepy
-.10
.02
.04
.00
9. Quiet
-.18
-.01
* p<.05; ** p<.01; *** p<.001 Unique Variance = .08
Shared Variance = .38
-.02
.00
R = .49
Adjusted R2 = .46
2
The results of these regressions indicate that following removal of all cases in the
negative range for SWB, core affect explained more unique variance in LS than
SWB (11% versus 8%). This finding is consistent with the hypothesis that core
affect would account for more unique variance in LS than SWB because LS is
more abstract and core affect specifically concerns the more abstract component
73
of SWB. Furthermore, this finding supports the explanation of why core affect
accounted for the same amount of unique variance in LS and SWB across the
entire data set - due to the inclusion of people considered at risk for depression
The following regression analyses were carried out to determine more precisely
which affects are the strongest and most reliable predictors of LS and SWB.
3.4
Further Analyses Using Core Affect Adjectives as Predictors of LS and
SWB
From Table 3 andTable 4, it has been demonstrated that the adjectives happy,
content and alert are the strongest predictors of LS and SWB. Because high
bivariate correlations between some predictors (e.g., happy and content), can
deflate correlations between IV’s and DV’s, it was decided that further analysis of
these variables was necessary to ensure that the adjectives chosen to represent
Russell’s (2003) ‘core affect’ are the strongest and most reliable. Accordingly, to
increase confidence in the statistically significant IV’s from the standard
regressions and to increase parsimony, sequential regressions were undertaken. In
these analyses, the top four predictors from the standard multiple regressions (as
determined by the magnitude of β weights), were entered. Although only three
variables were significant when predicting LS, it was decided to include a fourth
predictor in the event that it may reach significance. In stepwise regression, the
statistical program selects which variables will enter the equation in which order,
based on a set of statistical criteria. Thus, each IV is assessed by its unique
contribution to variance in the DV (as determined by sr2 and the magnitude of β
weights) - sr2 is interpreted as the amount of variance added to R2 by each IV at
the point that it enters the equation. Thus, predictors with the greatest sr2 values
will be subject to further analysis.
The regression analysis for LS, involving the top four predictor’s ‘happy’,
‘content’, ‘alert’ and ‘unhappy’, is presented in Table 8.
74
Table 8:
Predicting life satisfaction using four affect adjectives
R2
Adj. R2
∆R2
B
SE B

sr²
Step 1
Happy
.67***
.06
.68
.46
Happy
.51***
.07
.51
.20
Alert
.30***
.06
.33
.08
Happy
.38***
.08
.38
.08
Alert
.30***
.06
.34
.09
-.18**
.06
-.21
.03
Happy
.24*
.09
.24
.02
Alert
.26***
.06
.29
.06
.46
.45
Step 2
.54
.54
.09***
Step 3
Unhappy
.57
.56
.02*
Step 4
Unhappy
-.18**
.06
-.21
.03
Content
.22*
.09
.22
.02
.59
.58
.02*
* p<.05; ** p<.01; *** p<.001
The R for the final model of this regression was significantly different from zero,
F(4, 141) = 50.665, p < .001 (LS). Altogether, 58% of the variability in LS was
predicted from these four adjectives. Interestingly, the adjective ‘unhappy’
contributed unique, significant variance to the prediction of LS (sr2 = .03) above
and beyond that explained by its bi-polar opposite ‘happy’.
The results of the step-wise multiple regression analysis, using SWB as the DV,
are presented in Table 9. The adjective ‘Unhappy’ was included alongside the top
four predictors ‘happy’ ‘content’ ‘alert’ and ‘active’ (see Table 4) for exploratory
purposes.
75
Table 9:
Predicting SWB using five affect adjectives
R2
Adj. R2
∆R2
B
SE B

Sr²
Step 1
Happy
.57***
.04
.74
.55
Happy
.34***
.06
.52
.12
Content
.23***
.07
.29
.04
Happy
.37***
.06
.49
.10
Content
.18**
.07
.30
.02
Alert
.11**
.04
.17
.02
.55
.54
Step 2
.58
.58
.03***
Step 3
.60
.59
.02*
* p<.05; ** p<.01; *** p<.001
The R for the final model of this regression was significantly different from zero,
F(3, 142) = 71.008, p < .001. Altogether, 59% of the variability in SWB was
predicted and this is the same amount of explained variance found using the nine
adjectives in Table 4. In the final model, the adjectives ‘unhappy’ and ‘active’
failed to contribute unique, significant variance to the prediction of SWB; thus,
output for these variables was not provided by SPSS output.
Interestingly, the adjective ‘content’, contributed the same amount of unique
significant variance to the prediction of LS and SWB (sr2 = .02). However, for the
prediction of LS, ‘happy’ contributed a sr2 value one fifth of that contributed to
SWB (sr2 =.02 compared with .10 respectively); while for the prediction of LS,
‘Alert’ contributed a sr2 value three times that contributed to LS (.06 versus .02
respectively). These data suggest that a different balance of core affect may be
driving LS and SWB.
In summary, for the prediction of LS and SWB, three adjectives, ‘happy’,
‘content’ and ‘alert’, were found to contribute unique, significant variance.
Interestingly, for the prediction of LS only, the adjective ‘unhappy’ contributed
76
unique variance above its supposed bi-polar opposite ‘happy’. On grounds of
simplicity, ‘unhappy’ will be excluded from further analyses and the composite
variable ‘core affect’ has been computed using the adjectives ‘happy’, ‘content’
and ‘alert’. In support of this decision, previous studies have corroborated the
finding that both LS and SWB are dominated by the adjectives ‘happy’ and
‘content’ and affects representing activated-pleasant, such as ‘alert’/‘excited’
(Davern, et al., 2007; Blore, 2008).
3.5
Predicting LS and SWB above Core Affect using the Buffer Variables
The aim of the following analyses was to examine the relative contribution of
affect and cognition to SWB. More specifically, to explore whether the cognitive
buffer variables would contribute unique variances to the prediction of LS and
SWB beyond core affect. These analyses were also conducted as part of the sixth
hypothesis which states that core affect is driving the relationship between SWB
and related constructs.
3.5.1
Predicting LS using Core Affect and the Buffer Variables
Table 10 displays means, standard deviations, simple correlations and partial
correlations between variables.
Table 10:
Means, standard deviations and correlations between variables
Variable
1. LS
Mean
75.00
SD
18.05
1.
2. Core Affect
72.95
15.95
.75
3. Optimism
66.32
18.80
2.
3.
4.
5.
.48
.65
(.00)
4. Primary Control
69.87
16.48
.44
.61
.67
(.03)
(.45)
5. Secondary Control 65.65
19.27
.46
.51
.57
.49
(.13)
(.36) (.27)
6. Self-esteem
67.10
17.13
.59
.58
.56
.54
.37
(.29)
(.30) (.29) (.11)
* (.xx) partial correlations using core affect as a co-variate are presented in brackets
6.
77
When core affect was entered as a co-variate, correlations between the buffer
variables and LS reduce significantly. In fact, the average correlation between the
buffer variables and LS decreased from .49 to .11 (a reduction of .38). Most
dramatically, the correlation between LS and optimism reduced from .48 to .00
and the correlation between primary control and SWB reduced from .44 to .03.
These results are consistent with the hypothesis that core affect is driving the
relationships between these constructs. In this light, it is particularly noteworthy
that a partial correlation of .29 between self-esteem and LS remains, indicating a
greater degree of independence than either optimism or perceived control.
The results of the hierarchical regression analysis, with core affect entered at step
1 and the buffer variables at step 2, are presented in Table 11
Table 11:
Predicting LS using core affect and the buffer variables
R2
Adj. R2
∆R2
B
SE B

sr²
.85***
.06
.75
.56
.73***
.09
.64
.19
Step 1
Core Affect
.56
.55***
Step 2
Core Affect
Optimism
-.09
.08
-.09
.00
Primary Control
-.12
.08
-.11
.00
Secondary Control
.12
.06
.13
.01
Self Esteem
.29***
.07
.28
.04
.61
.60
.05
*p<.05; ** p<.01; *** p<.001
When entered at step 2, the buffer variables (self-esteem) explain a further 5% of
the variance in LS, ∆R2 = .05, Finc(4, 140) = 4.891, p<.001. This supports Davern
(2004), who found that self-esteem contributed 4% variance in LS above core
affect in an Australian adult sample.
78
It is notable that the unique variance (sr²) contributed by core affect at Step 2 has
been markedly reduced (by .37) – indicating a high degree of shared variance
between core affect and the buffer variables.
3.4.2
Predicting SWB using Core Affect and the Buffer Variables
The following analysis examines whether the cognitive buffers contribute
significant unique variance to the prediction of SWB beyond core affect. Table 12
displays means, standard deviations, simple correlations and partial correlations
between variables.
Table 12:
Means, standard deviations and correlations between variables (n = 146)
Variable
1. SWB
Mean
73.90
SD
13.95
1.
-
2.
2. Core Affect
72.95
15.95
.76
-
3. Optimism
66.32
18.80
3.
4.
5.
.65
.65
(.32)
4. Primary Control
69.87
16.48
.59
.61
.67
(.26)
(.45)
5. Secondary Control
65.65
19.27
.52
.51
.57
.49
(.24)
(.36) (.27)
6. Self-esteem
67.10
17.13
.55
.58
.56
.54
.37
(.21)
(.30) (.29) (.11)
* (.xx) partial correlations using core affect as a co-variate presented in brackets
6.
-
As before, when core affect was entered as a covariate, correlations between SWB
and the buffer variables decrease considerably. In fact, the average correlation
between the buffer variables and SWB reduced from .58 to .26, suggesting once
again that core affect may be the driving force behind these relationships.
Interestingly, a partial correlation of .32 between SWB and optimism can be
observed, indicating some residual independence of this relationship beyond core
affect. This finding is in contrast to the partial correlation observed between LS
and optimism (.00) presented in Table 13. This suggests that optimism may play a
differential role when predicting SWB than when predicting LS.
79
The results of the regression analysis, with core affect entered at step 1 and the
buffer variables at step 2, are presented in Table 13.
Table 13:
Predicting SWB using core affect and the buffer variables
B
SE B

sr²
.66***
.05
.76
.57
Core Affect
.43***
.07
.50
.12
Optimism
.13*
.06
.17
.01
Primary Control
.07
.06
.08
.00
Secondary Control
.07
.05
.10
.01
Self Esteem
.07
.06
.08
.00
R2
Adj. R2
∆R2
Step 1
Core Affect
.57
.57
Step 2
.63
.62
.06***
* p<.05; ** p<.01; *** p<.001
Similarly to analyses involving LS, when entered at step 2, the buffer variables
explain a further 5% of the variance in SWB, ∆R2 = .05, Finc(4, 140) = 5.871,
p<.0001. However, in sharp contrast to the same analysis involving LS, only
optimism continued to make a significant contribution. Similarly to analysis
involving LS as the DV, the unique contribution (sr²) of core affect at step 2 has
reduced considerably. This time, by .45, which is greater than the .37 reduction
for the analysis using LS.
For the following analyses, SWB will be employed as the major independent
variable of interest. SWB, as measured using the PWI-SC, adopts a domain-level
representation of global life satisfaction and in addition to approximating LS,
enables exploration of domain-level satisfaction judgments. Thus, SWB is a more
comprehensive and informative measure of wellbeing.
80
3.6
Predicting SWB Using MDT
The aim of the next regression analyses is to explore how much variance in SWB
can be explained by MDT above core affect. Means, standard deviations, simple
correlation and partial correlations are presented in Table 14.
Table 14:
Means, standard deviations and correlations between variables
Variable
1. SWB
Mean
73.90
SD
13.95
1.
-
2.
2. Core Affect
72.95
15.95
.76
-
3. MDT
62.60
16.20
3.
.65
.65
(.32)
* (.xx) partial correlations using core affect as a covariate presented in brackets
A correlation of .65 between MDT and SWB indicates a strong relationship
between these constructs. However, when core affect was entered as a co-variate,
this correlation reduced considerably (from r = .65 to r = .32), indicating that
MDT may also be driven, at least in part, by core affect. The results of the
hierarchical regression analysis with core affect entered at step 1 and MDT
entered at step 2 are presented in Table 15.
Table 15:
Predicting SWB with MDT
R2
Adj. R2
∆R2
SE B

sr²
.66***
.05
.76
.57
.50***
.06
.58
.19
2.39***
.59
.28
.04
B
Step 1
Core Affect
.57
.57
Step 2
Core Affect
MDT
.62
* p<.05; ** p<.01; *** p<.001
.61
.05***
81
When entered at step 2, MDT explained a further 4% variance, ∆R2 = .05, Finc(1,
143) = 16.588, p<.0001. It is noteworthy that the B value for core affect is
considerably smaller than the B value for MDT (.50 versus 2.39). This is
interesting given that the amount of unique variance (sr²) contributed by core
affect is considerably greater than that contributed by MDT (.19 versus .04). The
explanation of this result is that B weights are unstandardised regression
coefficients that represent the factor by which the DV will increase. Although the
B value for core affect is smaller than that for MDT, standardised regression
coefficients ( weights) and sr² values inform that core affect is the greatest
unique predictor in this regression.
To examine the extent to which the buffers and MDT are explaining different
aspects of the remaining variance in SWB above core affect, another regression
analysis was performed, with the buffers entered at step 2 and MDT at step 3.
Table 16 displays means, standard deviations, simple correlations and partial
correlations between variables.
Table 16:
Means, standard deviations and correlations between variables (n = 146)
Variable
1. SWB
Mean
73.90
SD
13.95
1.
-
2.
2. Core Affect
72.94
15.95
.76
-
3. Optimism
66.32
18.80
3.
4.
5.
.65
.65 (.32)
4. Primary Control
69.87
16.48
.59
.61
.67
(.26)
(.45)
5. Secondary Control 65.65
19.27
.52
.51
.57
.49
(.24)
(.36) (.27)
6. Self-esteem
67.10
17.13
.55
.58
.56
.54
.37
(.21)
(.30) (.29) (.11)
7. MDT
62.64
16.20
.65
.65
.55
.53
.50
(.32)
(.23) (.22) (.26)
* (.xx) partial correlations using core affect as a covariate presented in brackets
6.
7.
.57
(.31)
As can be seen, all three buffer variables have a consistent, moderate relationship
with MDT (.50 to .57). However, partial correlations reveal that the relationship
-
82
between these constructs reduces considerably when the influence of core affect is
removed. The results of the regression analysis, with core affect entered at step 1,
the buffer variables at step 2 and MDT entered at step 3, are presented in Table 17.
Table 17:
Predicting SWB with core affect, the buffer variables and MDT
B
SE B

sr²
.66***
.05
.76
.57
Core Affect
.43***
.07
.50
.12
Optimism
.13*
.06
.17
.01
Primary Control
.07
.06
.08
.00
Secondary Control
.07
.05
.10
.01
Self Esteem
.07
.06
.08
.00
Core Affect
.38***
.07
.43
.08
Optimism
.12*
.06
.16
.01
Primary Control
.06
.06
.07
.00
Secondary Control
.05
.05
.07
.00
Self Esteem
.03
.06
.04
.00
1.66**
.62
.19
.02
R2
Adj. R2
∆R2
Step 1
Core Affect
.57
.57
Step 2
.63
.62
.06***
Step 3
MDT
.65
.64
.02**
* p<.05; ** p<.01; *** p<.001
When entered at step 3, MDT accounted for an additional 2.0% of the variance in
SWB, ∆R2 = .02, Finc(1, 139) = 7.718, p<.01. Thus, MDT predicts SWB
independently of both core affect and the buffer variables. Results of this
regression support the earlier observation that SWB is primarily an affective
construct, with minor independent contribution from optimism and MDT.
In summary, results of these analyses indicate that LS and SWB are primarily
affective constructs, with minor independent contribution from self-esteem,
83
optimism, and MDT in the presence of core affect. Furthermore, partial
correlations provide additional support for the hypothesis that core affect is
driving the relationship between SWB and related constructs.
3.7
Predicting SWB with Personality
The aim of the next set of analyses was to explore the relationship between
personality correlates and SWB. More specifically, to determine how much
variance in SWB can be accounted for by extraversion and emotional stability,
after removing variance attributed to core affect. Means, standard deviations,
simple correlations and partial correlations between variables are presented in
Table 18.
Table 18:
Means, standard deviations and correlations between variables
Variable
Mean
SD
1.
1. SWB
73.90
13.95
-
2. Core Affect
72.95
15.95
.76
3. Extraversion
60.45
24.71
2.
3.
4.
-
.47
.53
(.14)
4. Emotional Stability
68.07
18.57
.44
.50
.33
(.11)
(.09)
* (.xx) partial correlations using core affect as a covariate presented in brackets
-
Simple correlations reveal that a moderate relationship exists between the
personality dimensions of extraversion and emotional stability and SWB.
However, core affect, as a covariate, weakens these relationships. In fact, the
correlation between SWB and extraversion reduces from .47 to .14 and the
correlation between SWB and emotional stability reduces from .44 to .11;
indicating that the relationship between these constructs may be driven, at least in
part, by core affect. The results of the regression analysis, with core affect entered
at step 1 and personality at step 2, are presented in Table 19.
84
Table 19:
Predicting SWB using the personality dimensions of extraversion and emotional
stability
B
SE B

sr²
.66***
.05
.76
.57
Core Affect
.58***
.06
.67
.27
Extraversion
.06
.04
.10
.01
Emotional Stability
.06
.05
.08
.00
R2
Adj. R2
∆R2
Step 1
Core Affect
.57
.57***
Step 2
.58
.58
.01
* p<.05; ** p<.01; *** p<.001
When entered at step 2, neither extraversion nor emotional stability accounted for
any additional variance.
3.8
Predicting SWB Using Normative Data Divisions: Further Analyses
Homeostatic theory of SWB allows people to be categorized into relatively
homogenous groups based on level of SWB. Three main groups have been
identified: people with normal levels of SWB (SWB > 70%SM), people most
likely to be experiencing some level of homeostatic challenge (SWB between 45
and 69%SM) and people most likely to be experiencing homeostatic defeated or
depression (SWB < 45%SM).
Homeostatic theory of SWB asserts that under normal, unthreatening conditions,
set-point levels of core affect will be maintained, thus, so to will SWB.
Furthermore, because SWB is not likely to be challenged, it is predicted that the
cognitive buffers will remain un-activated and SWB will be driven by core affect.
Based on this reasoning, it is predicted that for people with SWB in the normal
range, the buffers will not contribute significant variance in SWB. When SWB
homeostasis is under threat, however, the system of cognitive buffers will be
activated in an attempt to restore and then maintain SWB within the normal range.
85
Thus, for people with SWB between 45 and 69%SM, it is predicted that the
cognitive buffers will explain unique variance in SWB above core affect (which is
operating at a diminished capacity). Finally, a loss of wellbeing, or depression,
occurs when a level of challenge has reached a threshold to which the system
cannot adapt. As a consequence, control over SWB will be taken away from
homeostasis and handed to the external, challenging agent(s). For these people, a
loss of core affect and defeat of the cognitive buffer system is predicted.
The actual composition of groups as above will, however, not be simple. As
discussed previously, on a group average basis, 70%SM is thought to represent
the line of resistance and that which SWB homeostasis is defending (Cummins et
al., 2002). If a group mean SWB falls below 70%SM, it indicates that the group
contains a higher than normal proportion of people who are at risk of depression.
Due to individual differences in SWB set-points, however, it is also likely that this
group will comprise a number of people with lower set-points (e.g., people with
SWB set-points between 55 and 65%SM) who are maintaining a level of
wellbeing that is normal to them. Given the potential for sub-groups in this
<70%SM range to contaminate findings, several data divisions have been made in
order to accommodate small sample Ns with tests of the predictions from
homeostasis.
The first of these analyses will examine people with SWB <70%SM. It is
hypothesised that when entered at step 2, the buffer variables will explain more
variance in this group than in the >70 group.
3.8.1
Predicting SWB using cases > 70%SM and <70%SM
To test this hypothesis, the sample was divided into two groups: people with SWB
scores equal to and above 70%SM (n= 100, M = 81.25, SD = 7.35) and people
with SWB below 70%SM (n= 46, M = 57.91, SD = 11.27). Splitting cases in this
manner decreases sample variance - as indicated by the reduction in the
magnitude of the standard deviations. Following this split, a sequential regression
analysis was performed on each group. It is important to note, however, that there
are only 46 cases is the group with SWB below 70%SM. According to
86
Tabachnick and Fidell (2001), this does not represent adequate statistical power
for such analyses to be reliable. As a consequence, the results are interpreted as
pilot indicators only.
Means, standard deviations, correlations and partial correlations between all major
variables for people with SWB > 70%SM and for people with SWB < 70%SM are
presented in Table 20 and Table 21.
Table 20:
Means, standard deviations and correlations between variables (SWB > 70%SM)
Variable
Mean
SD
1.
1. SWB
81.25
7.35
-
2. Core Affect
79.63
11.10
.53
3. Optimism
73.40
14.24
2.
3.
4.
5.
6.
-
.39
.45
(.20)
4. Primary Control
74.50 12.91
.31
.46
.55
(.09)
(.43)
5. Secondary Control
70.17 16.38
.29
.25
.42
.36
(.19)
(.36)
(.28)
6. Self-esteem
72.46 13.66
.25
.35
.36
.30
(.08)
(.24)
(.17)
* (.xx) partial correlations using core affect as a covariate presented in brackets
.16
(.08)
-
Table 21:
Means, standard deviations and correlations between variables (SWB < 70%SM)
Variable
Mean
SD
1.
1. SWB
57.91
11.27
-
2. Core Affect
58.41
15.26
.58
3. Optimism
50.92
18.40
2.
3.
4.
5.
6.
-
.45
.48
(.23)
4. Primary Control
59.82 18.92
.63
.53
.61
(.46)
(.48)
5. Secondary Control
55.81 21.48
.58
.58
.55
.47
(.36)
(.38)
(.24)
6. Self-esteem
55.44 18.21
.44
.51
.49
.58
.38
(.20)
(.32)
(.43)
(.12)
* (.xx) partial correlations using core affect as a covariate presented in brackets
-
87
The mean score across the buffer variables was higher for people with SWB >
70%SM ((mean for buffer variables = 72.63 (SD = 14.30) versus 55.50, (SD =
19.25 in the < 70group)). A difference of almost 5.0 percentage points between
standard deviations should also be noted. It appears that the range of scores has
increased due to selective homeostatic failure for some people. As expected, it can
be seen that correlations between the buffer variables and SWB were generally
higher for people with SWB < 70%SM (.53 vs .31). It is suspected that this occurs
for several reasons. First, there is the reduced range of values due to the splitting
procedure causing reduced correlations as a statistical artifact. Second, it is argued
that the correlations in the higher group are being driven by individual differences
in core affect alone. Finally, in the lower ranges, such individual differences
operate for some people, however, there is additional variance added by the
challenging agent causing all domains to move down together as homeostasis fails.
Thus, it is argued that correlations are being driven by a number of factors and
these will become clearer in latter analyses involving further data divisions (e.g.,
using cases between 45 and 69%SM).
Turning back to Table 21, in both samples, correlations between the buffer
variables and SWB decreased when core affect was entered into a partial
correlation matrix as a covariate. Moreover, as predicted, this effect was less
pronounced for people with SWB < 70 (average correlation reduced by .22 vs .17
for >70). It is predicted that these correlations would decrease more in the >70
group because it is most likely that the cognitive buffers are resting and core
affect is the driver of SWB.
Finally, it is noteworthy that the greatest observed correlation was found in the <
70 group - between primary control and SWB (r = .63). Even after variance
attributed to core affect was removed, a correlation of .46 between these
constructs remained. Also, a partial correlation of .36 between secondary control
and SWB was observed in this group. These correlations suggest that perceived
control may be a major buffer for people experiencing some level of homeostatic
challenge.
Regression analyses involving the split of cases > 70%SM and below 70%SM, are
presented in Table 22.
Table 22:
Predicting SWB after core affect using the buffer variables (split Cases)
SWB > 70%SM
2
R
Adj. R
2
SWB < 70%SM
∆R
2
B
SE B

sr²
.35***
.06
.53
.28
R2
Adj. R2
∆R2
B
SE B

sr²
.43***
.09
.58
.34
Step 1
Core Affect
.28
.27
.34
.32
Step 2
Core Affect
.29***
.07
.43
.13
.17
.11
.23
.03
Optimism
.74
.57
.14
.01
-.49
.92
-.08
.00
-.10
.61
-.02
.00
2.46*
.92
.41
.08
Secondary Control
.55
.43
.12
.01
1.52
.76
.29
.05
Self Esteem
.18
.50
.03
.00
.07
.88
.01
.00
Primary Control
.32
.29
.04
.53
.47
.19**
* p<.05; ** p<.01; *** p<.001
88
89
In the group with SWB > 70%SM, core affect explained 27% of the variance, F(1,
98) = 38.289, p<.0001, compared with 32% of the variance for people with SWB
< 70%SM, F (1, 44) = 22.298, p<.0001. Significance of the difference between
explained variance was assessed by first converting the correlations into z-scores
using Fishers z-score conversion table, then subtracting one z-score from the other
and then dividing by the pooled standard error for the two samples. This test
revealed no significant difference in explained variance between the two samples,
t = .382 (p > .05).
Although core affect explains more variance in the <70 group (non-significant), it
is possible that this may be due to variance contributed by people in the lower
ranges for SWB – people who’s SWB has been defeated. As expected, when the
cognitive buffers were entered at step two, they did not add significant variance in
the group with SWB > 70%SM. However, consistent with the fifth hypothesis, for
people with SWB < 70, the buffer variables, in particular primary control (sr2
= .08, p< .01), contributed a further 15% of variance, ∆R2 = .19, Finc(4, 40) =
4.021, p<.01). Interestingly, core affect fails to add significant, unique variance at
step 2. The implications of this finding will be discussed.
Another interesting result is that in contrast to findings presented in Table 17
(where optimism contributed unique, significant variance to the prediction of
SWB), optimism failed to contribute unique variance in both groups. And finally,
it is noteworthy that B weights for primary control in the > 70%SM and optimism
in the <70%SM group were negative. This, however, is not an issue of great
importance because (sr2) values for both these variables are .00. Thus, there are no
inconsistencies in the direction of regression weights and amount of explained
variance.
In summary, these analyses partially support the fifth hypothesis. In total, the
buffer variables explained a further 15.0% of variance above core affect in the
group with SWB<70%SM. However, primary control was the only significant
predictor contributing a further 8.0% of variance. However, inconsistent with
theory suggesting that core affect will explain more variance in people with
90
normal levels of SWB compared to people whose SWB is under challenge, core
affect explained more variance in the challenged group (32% versus 27%).
The following analysis examines the relation between core affect, SWB and the
buffer variables in a group most likely to be experiencing homeostatic challenge
or defeat.
3.8.2
Predicting SWB using cases Between 45 and 69%SM
For the following analysis, cases below 45%SM were removed. This was done in
order to a) observe changes in the correlations between core affect, the buffer
variables and SWB and b) observe changes in the amount of explained variance
contributed by core affect and the buffer variables to the prediction of SWB
following the removal of variance contributed by people most likely to be
depressed. As a consequence of this procedure, five cases were eliminated from
the analysis.
Means, standard deviations, correlations and partial correlations between all major
variables are presented in Table 23. Again, results must be interpreted with care
due to small sample size.
Table 23:
Means, SD’s and correlations between variables (SWB between 45 and 69%SM)
(n = 41)
Variable
Mean
SD
1.
1. SWB
61.07
6.43
-
2. Core Affect
60.65 13.72
.41
2.
3.
4.
5.
-
.09
.36
(-.06)
.34
.48
4. Primary Control
63.05 15.44
.33
(.25)
(.42)
.46
.45
.30
5. Secondary Control
58.55 20.30
.49
(.33)
(.34)
(.17)
.11
.39
.46
.23
6. Self-esteem
58.20 16.73
.34
(-.03)
(.30)
(.40) (.08)
* (.xx) partial correlations using core affect as a covariate presented in brackets
3. Optimism
6.
53.63 16.72
-
91
By removing cases below 45%SM in the <70 group, the standard deviation for
SWB reduced from 11.27 to 6.43. Furthermore, the mean score for SWB rose
3.15% from 57.91% to 61.07%. Such a reduction in variance indicates that the
group is now more homogenous in comprising people likely to be experiencing
some level of homeostatic challenge rather than also consisting of people whose
SWB is likely to have be defeated.
With the exception of perceived control, simple correlations between the buffer
variables and SWB were generally lower in the group with SWB between 45 and
69%SM when compared to the group with SWB < 70. For example, r = .11 for
self-esteem and .09 for optimism in the 45-69, compared with r = .45 and .44
respectively in the <70 group. This is as expected. Furthermore, using core affect
as a covariate, a correlation of .25 between primary control and SWB and .33
between secondary control and SWB remained. These are considerably higher
than those for people with SWB > 70%SM - primary control and SWB (r = .09)
and secondary control and SWB (r = .19). These results are consistent with
homeostatic theory which predicts a strengthened relationship between the buffer
variables and SWB when SWB set-points are challenged.
The results of this analysis, with core affect entered at step 1 and the buffer
variables at step 2, are presented in Table 24.
Table 24:
Predicting SWB using core affect and the buffer variables for individuals with
SWB between 45% and 69%SM (n = 41)
R2
Adj. R2
∆R2
B
SE B

Sr²
.19**
.07
.41
.17
.11
.08
.24
.04
-.12
.07
-.30
.06
Step 1
Core Affect
.17
.14
Step 2
Core Affect
Optimism
Primary Control
.14*
.07
.34
.08
Secondary Control
.13*
.05
.40
.11
.06
-.10
.01
Self Esteem
-.04
.36
* p<.05; ** p<.01; *** p<.001
.27
.20*
92
When entered at step 1, core affect explained 14% of the variance, F(1, 39) =
7.713, p = .01. This is considerably less than the 32% of variance explained for
people with SWB <70 and the 27% of variance explained for people with
SWB >70. When entered at step 2, secondary control (sr2 = .11, p< .05) and
primary control (sr2 = .08, p< .05) added a further 13% of variance, ∆R2 = .13, Finc
(4, 35) = 2.684, p < .05. This additional variance is comparable to the 15% of
variance contributed by primary control and, to a lesser extent, secondary control,
in the same analysis involving people with SWB <70. Most importantly, core
affect fails to add significant, unique variance at Step 2 with almost all of the
significant variance explained by the buffer variables.
In summary, these findings are consistent with SWB homeostasis in that, for
samples under homeostatic challenge, the influence of core affect over SWB
appears to shift to the buffers. It seems that individual differences in buffer
strength are driving SWB amongst the challenged group. Furthermore, as people’s
SWB is increasingly challenged, the buffer variables (perceived control), will be
activated in attempt to counter threats to wellbeing.
3.9
Predicting SWB using MDT: Further Analyses Using the Adolescent
Sample
For exploratory purposes, multiple regression analyses were carried out to test
whether MDT would contribute unique variance in SWB in people with SWB
between 45 and 69%SM and >70%SM. To reiterate, MDT represents a cognitive
approach to SWB measurement and is based on the premise that SWB results
from perceived gaps or discrepancies between needs, wants, desires etc. (Michalos,
1985). As stated in the previous section, under normal conditions, when SWB is
intact, SWB will be driven by core affect. However, when challenged, thoughts
pertaining to particular events and experiences will exert greater influence over
SWB. Given that MDT was found to independently predict SWB across the entire
data set and based on the premise that MDT represents a cognitive approach to
SWB measurement, it is expected that MDT will predict SWB, above core affect,
for people most likely to be experiencing some level of homeostatic challenge.
93
Table 25 displays means, standard deviations, simple correlations and partial
correlations between variables for people with SWB > 70%SM and people with
SWB between 45 and 69%SM.
Table 25:
Means, SD’s and correlations between variables
SWB > 70%SM SWB
Between 45 and 69%SM
Variable
Mean
SD
1.
2.
1. SWB
81.25
7.35
-
2. Core Affect
79.63
11.10 .53
-
3. MDT
68.90
12.80 .32
(.17)
.36
3.
-
Mean
SD
61.07
6.43
1.
2.
3.
.42
-
-
60.65 13.72
.40
51.30 12.00
.19
(-.06)
-
* (.xx) partial correlations using core affect as a covariate presented in brackets
For people with SWB > 70%SM, the correlation between MDT and SWB
revealed a weak relationship between these constructs (r = .32). Consistent with
theory suggesting that core affect will operate to a diminished extent when SWB
set-points are challenged, this same correlation was a low .19 for people with
SWB between 45 and 69%SM. When variance attributed to core affect was
removed, these correlations reduced to .17 (SWB > 70%SM) and -.06 (SWB
between 45 and 69%SM).
The results of regression analyses, with core affect entered at step 1 and MDT at
step 2 are presented in Table 26.
Table 26:
Predicting SWB After Core Affect Using MDT
SWB > 70%SM
2
R
SWB between 45 and 69%SM
2
Adj. R
∆R
2
B
SE B

sr²
.35***
.06
.53
.28
R2
Adj. R2
∆R2
B
SE B

sr²
.19
.07
.41
.17
Step 1
Core Affect
.28
.27
.17
.14
Step 2
Core Affect
.32***
.06
.48
.19
.20
.08
.43
.15
MDT
.87
.52
.15
.02
.34
.88
.06
.00
.30
.29
.02
.17
.13
.01
* p<.05; ** p<.01; *** p<.001
94
95
For both groups, MDT entered at Step 2 did not contribute additional significant
variance. This is in contrast to findings using the complete sample where MDT
was found to add a further 4% variance at Step 2. Results indicate that
discrepancy judgments may not comprise the cognitive component of the
remaining variance in SWB in a group of adolescents with normal levels of
wellbeing and in those experiencing homeostatic challenge.
3.10 Predicting SWB Using Normative Data Divisions: Conclusions
The most important results from these analyses have come from the separation of
SWB into three groups representing various levels of SWB. This has revealed that
different factors may be driving SWB depending on a person’s level of SWB. As
homeostatic theory of SWB would suggest, the split of cases >70%SM and
<70%SM revealed that correlations between the buffer variables and SWB were
higher for people with SWB < 70%SM (.53 vs .31). Furthermore, in all samples, it
was found that correlations between the buffer variables and SWB decreased
when core affect was entered into a partial correlation matrix as a covariate. This
effect was more pronounced in the group of people with normal levels of SWB.
Finally, consistent with the fifth hypothesis and with homeostatic theory of SWB,
the split of cases revealed that the buffer variables contributed unique variance
above core affect in a group of people experiencing homeostatic challenge.
3.11 Comparative Analysis Against an Adult Data Set
To explore whether these findings generalise to other populations, parallel
analyses were conducted on a comparative adult data set with a larger N.
Method
Participants
The adult data set consisted of a sample of participants from the Australian Unity
Longitudinal Wellbeing Project. In total, 577 surveys were mailed to participants
and 387 were returned, representing a response rate of 67%. Of the 67% of
individuals who responded, 45% were male (174) and 55% were female (213).
96
The age of participants ranged from 19 to 85 years, with a mean age of 56.5 (SD =
14.72 years).
Measures
The same set of measures as used for the adolescent sample was used in this
survey. However core affect was represented by the adjectives content, active and
happy. It is notable that ‘active’ was used in these analyses to represent core affect
in place of ‘alert’. The reason is that ‘active’ rather than ‘alert’ contributed unique
variance to the prediction of SWB in this data set.
Data Screening and Preliminary Analysis
Data were screened and cleaned using the same method as for the adolescent data
(see 3.1).
Results
Similarly to previous analyses involving the adolescent sample, after exploring
the relative affective/cognitive component of SWB across the entire data set, cases
were split into two groups – those with SWB > 70%SM and those with SWB
between 45%SM and 69%SM (people who, on an average basis, are most likely
to be experiencing some degree of homeostatic challenge). The results of these
analyses are presented in the following Tables. Table 27 displays means, standard
deviations, simple correlations and partial correlations between variables using all
cases.
97
Table 27:
Means, SD’s and correlations between variables: comparative analysis (n=387)
Variable
Mean
SD
1.
2.
3.
1. SWB
72.45
14.95
-
2. Content
73.40
17.90
.76
-
3. Active
66.10
20.20
.47
.42
-
4. Happy
74.40
17.70
.74
.87
.49
5. Optimism
62.02
16.27
4.
5.
6.
7.
-
.48
.50
.33
.53
(.12)
6. Primary Control
61.56
13.68
.27
.30
.26
.30
.42
(.03)
(.31)
7. Secondary Control 65.56
15.23
.29
.38
.24
.39
.55
.52
(-.05)
(.43) (.45)
8. Self-esteem
76.17
17.26
.55
.60
.34
.63
.63
.37
(.11)
(.43) (.23)
* (.xx) partial correlations using core affect as a covariate presented in brackets
.43
(.25)
Simple correlations between the buffer variables and SWB were generally lower
in the adult sample than in the adolescent sample (average correlation of .40
compared to .58 for the adolescent sample). Interestingly this difference is largely
attributed to a difference in correlations between SWB and primary control (.32
lower in the adult sample) and between SWB and secondary control (.23 lower in
the adult sample). This suggests a stronger relationship between perceived control
and SWB in the adolescent sample.
When core affect was entered as a covariate in a correlation matrix, the average
correlation reduced by .35, compared with .32 for the adolescents (see Table 12).
Moreover, the partial correlation between optimism and SWB was greater in the
adolescent sample than in the adult sample (.32 versus .12 respectively). This
finding indicates that in the adolescent sample, optimism shows greater
independence from core affect.
Table 28 displays the results of the regression analysis, with core affect entered at
step 1 and the buffer variables at step 2.
8.
-
98
Table 28:
Predicting SWB using core affect and the buffer variables: comparative analysis
B
SE B

sr²
Content
4.14***
.54
.50
.06
Active
1.02***
.27
.14
.01
Happy
2.04***
.57
.24
.01
Content
3.99***
.54
.48
.05
Active
.95***
.27
.13
.01
1.69**
.58
.20
.01
Optimism
.09*
.04
.10
.01
Primary Control
.03
.04
.03
.00
Secondary Control
.09*
.04
.09
.01
Self Esteem
.02
.01
.06
.00
R2
Adj. R2
∆R2
Step 1
.62
.62
Step 2
Happy
.63
.64
.02
* p<.05; ** p<.01; *** p<.001
When entered at step 2, optimism and secondary control each provided an
additional 1% of variance ∆R2 = .02, Finc(4, 379) = 3.091, p < .05. This is a very
similar result to that found across the adolescent sample (see Table 13).
To explore the relative contribution of core affect and the buffers for people with
SWB > 70%SM and people with SWB between 45 and 69%SM, further
regressions
were
conducted
similar
to
those
presented
earlier.
99
Table 29 and Table 30 display means, standard deviations, simple correlations and
partial correlations between variables for people with SWB > 70%SM and for
people with SWB between 45 and 69%SM.
100
Table 29:
Means, SD’s and correlations between variables (SWB > 70%SM) (n=261)
Variable
Mean
SD
1.
2.
3.
4.
1. SWB
80.90
7.02
-
2. Content
80.40
11.40
.58
-
3. Active
71.20
16.40
.26
.21
-
4. Happy
80.10
11.40
.57
.84
.25
5. Optimism
66.05
14.33
5.
6.
7.
.30
.32
.18
.33
(.12)
6. Primary Control 63.54
13.11 .20
.23
.23
.26 .37
(.04)
(.30)
7. Secondary Control 67.51
15.05 .35
.43
.24
.41 .54
.51
(.11)
(.45) (.44)
8. Self-esteem
80.83
13.35 .42
.41
.28
.50 .48
.31
(.18)
(.38) (.19)
* (.xx) partial correlations using core affect as a covariate presented in brackets
-
8.
-
.38
(.21)
-
Table 30:
Means, SD’s and correlations between variables (SWB Between 45 and 69%SM)
(n=261)
Variable
Mean
SD
1.
2.
3.
1. SWB
59.31
6.58
-
2. Content
64.00
16.20
.43
-
3. Active
58.50
20.90
.22
.18
-
4. Happy
65.70
16.50
.36
.67
.42
5. Optimism
56.87
14.20
4.
5.
6.
7.
8.
-
.11
.33
.21
.48
(-.09)
6. Primary Control
58.30 13.82 .09
.18
.09
.20
.36
(-.01)
(.30)
7. Secondary Control
61.80 15.04 .12
.25
.08
.34
.60
(.00)
(.53)
8. Self-esteem
70.48 17.94 .16
.43
.10
.46
.57
(-.05)
(.45)
* (.xx) partial correlations using core affect as a covariate presented in brackets
.47
(.43)
.28
.45
(.21) (.35)
-
101
Simple correlations presented in Table 29 (SWB > 70%SM) revealed that the
relationship between the buffer variables and SWB was lower in the adult sample
when compared to the >70%SM adolescent sample (average correlation of .32
compared with .44). Also, similarly to that found using the adolescent sample,
correlations between the buffer variables and SWB decreased considerably when
the influence of core affect was removed - an average reduction of .21 compared
with .30 for the adolescent sample). These data indicate that for people with
SWB > 70%SM, core affect has a greater influence over the adolescent responses.
For people with SWB between 45 and 69%SM (Table 30), the main difference
between the two samples is that the correlation between primary/secondary
control and SWB are significantly greater in the adolescent sample (primary
control =.34, secondary control = .46) than in the adult sample (primary control
= .09, secondary control = .12). Furthermore, after removing variance attributed to
core affect, correlations between these constructs and SWB remained
considerably greater in the adolescent sample (primary control = .25, secondary
control = .33) than in the adult sample (primary control = -.01, secondary control
= .00), indicating a greater degree of independence.
Table 31 displays results of the regression analyses involving the comparative
adult data set for people with SWB > 70%SM and for people with SWB between
45 and 69%SM. In both regressions, core affect is entered at step 1 and the buffer
variables at step 2.
Table 31:
Predicting SWB using core affect and the buffer variables for adults with SWB between 45 and 69%SM
SWB > 70%SM (N = 261)
SWB Between 45 and 69%SM (N = 103)
B
SE B

sr²
Content
2.26***
.56
.37
Active
.53*
.22
Happy
1.38*
.57
2
R
Adj. R
2
∆R
2
B
SE B

sr²
.04
1.44**
.50
.36
.07
.13
.01
.39
.32
.12
.01
.22
.01
.29
.53
.07
.00
R2
Adj. R2
∆R2
Step 1
.37
.37
.20
.18
Step 2
Content
2.15***
.57
.35
.04
1.46**
.51
.36
.07
Active
.40
.22
.09
.01
.40
.32
.13
.01
Happy
.92
.59
.15
.01
.44
.58
.11
.00
Optimism
.02
.03
.03
.00
-.05
.06
-.11
.01
-.01
.03
-.02
.00
.00
.05
.00
.00
Secondary Control
.03
.03
.06
.00
.03
.05
.06
.00
Self Esteem
.03*
.01
.14
.01
-.00
.02
-.02
.00
Primary Control
.40
.38
.02*
.21
.15
.01
* p<.05; ** p<.01; *** p<.001
102
103
What is interesting here is that in contrast to prediction, when entered at step 2,
the buffers did not add additional, significant variance in the group with SWB
between 45 and 69%SM. This finding is also inconsistent with findings using the
sample of adolescents. To reiterate, in the adolescent sample (Table 24), the buffer
variables (e.g., primary control and secondary control) added a further 13% of
variance. Furthermore, the overall amount of explained variance was considerably
lower in the adult sample than in the adolescent sample (15% versus 27%). Unlike
analyses involving the adolescent data set, the buffer variables do not appear to
add any significant variance for adults most likely to be experiencing some level
of homeostatic challenge.
3.12 Predicting SWB using MDT: Further Analyses Using the Comparative
Adult Data Set
To explore whether the relationship between SWB and MDT followed the
adolescent data, regressions were conducted, similarly to those presented in Table
26. Table 32 displays means, standard deviations, simple correlations and partial
correlations between variables for people with SWB > 70%SM and people with
SWB between 45 and 69%SM.
Table 32:
Means, standard deviations and correlations between variables
SWB > 70%SM
SWB Between 45 and 69%SM
Variable
Mean
SD
1.
2.
3.
1. SWB
80.90
70.20
-
2. Content
80.40
11.40
.58
-
3. Active
71.20
16.40
.26
.21
4. Happy
81.00
11.40
.57
.84 .25
5. MDT
66.40
12.20
4.
Mean
SD
1.
59.31
65.80
-
64.00
16.20 .43
-
58.50
20.90 .22
.18
-
65.70
16.50 .36
.67
.42
-
.35
.42 .07 .37 52.00 12.90 .38 .42 -.01
(.15)
(.26)
* (.xx) partial correlations using core affect as a covariate presented in brackets
.32
-
2.
3.
4.
104
When core affect was entered into a correlation matrix as a covariate, the
correlation between SWB and MDT reduced from .35 to .15 for people with
SWB > 70%SM. This finding is similar to that found across the adolescent data
(Table 25), where the correlation reduced from .32 to .17. For adults with SWB
between 45 and 69%SM, this same correlation reduced from .38 to .26; whereas
in the adolescent sample, it reduced from .19 to -.06. It appears that discrepancy
judgments may be an independent predictor in only the adult sample.
The results of regression analyses, with core affect entered at step 1 and the buffer
variables at step 2 are presented in Table 33.
Table 33:
Predicting SWB after core affect using MDT
SWB > 70%SM
SWB between 45 and 69%SM
B
SE B

sr²
Content
2.26***
.56
.37
Active
.53*
.22
Happy
1.38*
.57
2
R
2
Adj. R
∆R
2
R2
Adj. R2
∆R2

sr²
.50
.36
.07
.39
.32
.12
.01
.29
.53
.07
.00
B
SE B
.04
1.44**
.13
.02
.22
.01
Step 1
.37
.37
.20
.18
Step 2
Content
.20***
.06
.32
.02
.11*
.05
.27
.04
Active
.06*
.02
.13
.02
.05
.03
.16
.02
Happy
.14*
.06
.22
.01
.01
.05
.03
.00
MDT
.08*
.03
.13
.01
1.34**
.49
.26
.06
.39
.38
.01*
.26
.23
.06
* p<.05; ** p<.01; *** p<.001
105
106
In the group with SWB > 70%SM, MDT, when entered at step 2, explained an
additional 1% of variance, ∆R2 = .01, Finc(1, 256) = 5.937, p <.05. In the group
with SWB between 45 and 69%SM, MDT, when entered at step 2, explained an
additional 5% of the variance, ∆R2 = .06, Finc(1, 98) = 7.464, p <.01. However, the
total amount of explained variance is still very low (23%).
In contrast to findings with the adolescent sample, results of these analyses
indicate that MDT independently predicts SWB, above core, for adults with SWB
between 45 and 69%SM. This finding is of particular interest given that for 45 69%SM, MDT mean scores and standard deviations amongst the adolescent
sample (M = 51.30, SD = 12.90) and adult sample (M = 52.00, SD = 12.90) were
remarkably similar – although partial correlations between SWB and MDT were
greater in the adult sample than in the sample of adolescent (r = .26 versus r = .06).
3.12.1 Comparative Analysis Against an Adult Data: Summary of Results
The most interesting finding from these analyses is that the buffer variables, when
entered at step 2 of a stepwise multiple regression analysis, did not add additional,
significant variance in the 45 - 69 group for adults. However, in the adolescent
sample, the buffer variables (e.g., primary control and secondary control)
contributed a further 13% of variance. The results of these regression analyses are
consistent with correlations presented in Table 23 and Table 30. More specifically,
simple correlations between primary control and SWB and between secondary
control and SWB were significantly greater in the adolescent sample (see Table
23) (primary control =.34, secondary control = .46) when compared to the adult
sample (see Table 30) (primary control = .09, secondary control = .12). Thus,
findings for adults are inconsistent with theory.
Another major difference between the adult and adolescent data was that for the
45 - 69%SM group, MDT only contributed unique, significant variance above
core affect in the adult sample.
107
3.13 Exploratory Analyses: Satisfaction with School as a Unique Construct
Given that most adolescents spend much of their time at school, it would seem
intuitive that satisfaction with school may meet the criterion of a PWI domain for
this group. The aim of the next multiple regression analyses were explore whether
school satisfaction predicts LS above all seven domains included in the PWI-SC,
thus, qualifying as a new domain. To determine this, a stepwise multiple
regression analysis was carried out with the seven domains entered at step 1 and
school satisfaction entered at step 2.
Means, standard deviations, simple correlations and partial correlations between
variables are presented in Table 34.
Table 34:
Means, standard deviations and correlations between variables
Variable
Mean
SD
1.
1. Life Sat.
75.00
18.05
-
2. Std Living
77.05
18.28
2.
3.
4.
5.
6.
.47
(.15)
3. Health
74.04 20.36 .66 .34
(.42) (.07)
4. Achieving
69.73 17.85 .47 .31
.52
(.21) (.08) (.34)
5. Relationships
80.09 18.57 .57 .40
.57
.46
(.18) (.08) (.32) (.24)
6. Safety
81.08 18.14 .58 .35
.52
.39
.58
(.20) (.04) (.24) (.13) (.30)
7. Neighbourhood
65.56 24.20 .42 .38
.40
.21
.33
.48
(.13) (.20) (.20) (.01) (.07) (.30)
8. Future
69.73 20.41 .46 .43
.49
.44
.45
.46
(.07) (.20) (.24) (.22) (.13) (.14)
9. School Sat.
78.97 19.29 .53 .53
.43
.42
.56
.53
(.11) (.26) (.10) (.17) (.22) (.21)
* (.xx) partial correlations using core affect as a covariate presented in brackets
7.
9.
.32
(.10)
.32 .45
(.05) (.13)
The mean score for school satisfaction was third highest at 78.97 (SD = 19.29).
Additionally, school satisfaction correlated fourth highest with life satisfaction
at .53 (.11). Results of the regression analysis, with all seven domains entered at
step 1 and school satisfaction entered at step 2, are presented in Table 35.
8.
-
108
Table 35:
Predicting LS using 7 PWI domains and satisfaction with school
R2
Adj. R2
∆R2
B
SE B

Sr²
Step 1
Std Living
.18**
.07
.18
.02
Health
.32***
.07
.36
.07
Achieving
.07
.07
.07
.00
Relationships
.12
.08
.12
.01
Safety
.20*
.08
.20
.02
Neighbourhood
.03
.05
.05
.00
Future
.01
.06
.01
.01
.57
.55
Unique Variance = .13
Shared Variance = .42
Step 2
Std Living
.15*
.07
.15
.01
Health
.32***
.07
.36
.07
Achieving
.06
.07
.06
.00
Relationships
.10
.08
.10
.00
Safety
.17*
.08
.17
.01
Neighbourhood
.04
.05
.05
.00
Future
.00
.06
.00
.00
School Satisfaction
.11
.07
.11
.01
.58
.56
.01
Unique Variance = .10
Shared Variance = .46
* p<.05; ** p<.01; *** p<.001
While it can be seen that school satisfaction accounts for no additional significant
variance when entered at step 2, it is interesting that only three domains do
contribute unique variance to the prediction of life satisfaction. These are health
(sr² = .07, p< .001), standard of living (sr² = .02, p< .01) and safety (sr² = .02,
p< .05). These results will be discussed.
109
CHAPTER 4: STUDY 1 DISCUSSION
The major aim of this study was to replicate the findings of Davern, et al., (2007),
that SWB is primarily an affective construct, with an independent contribution
from cognition. Additionally, this study sought to test a number of hypotheses
based on Cummins’ homeostatic theory of SWB as follows: 1. That the mean
score for LS will approximate the mean score for SWB and that the mean score
for SWB will lie within the Australian adult normative range; 2. That adjectives
located at the pleasantness-pole of the circumplex (core affect) will explain
significant variance in LS and SWB; 3. That core affect will explain greater
unique variance in LS than in SWB; 4. That the buffer variables will explain
significant variance in SWB above core affect in a group of people experiencing
homeostatic challenge and 5. That core affect is driving the relationship between
SWB and related constructs as perceived control, optimism, self-esteem,
extraversion, emotional stability and perceived discrepancy judgments. The
results will now be discussed in relation to each hypothesis.
4.1
Hypothesis one: That the mean score for LS will approximate the mean
score for SWB and that the mean score for SWB will lie within the
Australian adult normative range
According to SWB homeostasis theory, although the mean score for LS should
approximate the mean score for SWB, these two scores are not expected to be the
same. This is because the domain-based satisfactions that comprise the measure of
SWB represent the ‘first-level deconstruction’ of satisfaction with ‘life as a
whole’ (International Wellbeing Group, 2006). Together, the seven domains
represent broad, semi-abstract areas of life that describe the overall experience of
LS. Because the domains are more specific than LS, they deviate further from the
general positive mood state that is core affect. Thus, although LS and SWB scores
should not substantially differ from one another, they are not expected to be
identical.
A simple t-test confirmed this hypothesis. Consistent with Australian adult
normative statistics, the mean score for SWB (73.90%) was just within the
110
Australian adult normative range of 73.43 to 76.43%SM (Cummins et al., 2006).
Additionally, the mean score for LS (75.00) was also within the Australian adult
normative range of 75.20 to 79.10%SM (Cummins et al., 2006). It is also
noteworthy that the standard deviation for SWB is 13.95, compared with 12.29 for
adults (a difference of + 1.66 percentage points). This higher variance is
consistent with a mean score for SWB at the lower end of the adult normal range.
Homeostasis theory predicts that as a group mean falls towards the bottom of the
normal range, it indicates that the group contains a higher than normal proportion
of people who are at risk of depression and that some of these people will be
exhibiting a loss of homeostatic control. Consequently, their SWB will be below
their set-point-range and these low scores will cause the sample variance to
increase.
It is important to note that this relatively low mean and high SD has implications
for a number of the hypotheses being tested. This is because the pattern of results
will differ depending on whether the homeostatic system is resting or under
threat/defeat. Under conditions of low challenge, SWB will be maintained within
its set-point range and core affect is believed to be driving SWB. However, when
aversive circumstances defeat the homeostatic system, SWB will fall below
equilibrium levels. As a consequence, core affect no longer reflects SWB. Instead,
the experience of SWB will reflect the loss of positive affect as control over SWB
is assumed by cognitions and affects associated with the challenging agent. Thus,
because this sample has a mean SWB at the lower end of the normal range, the
above considerations will be relevant to subsequent discussion of the hypothesis
testing.
In summary, the first hypothesis was supported. Consistent with SWB
homeostasis, the mean score for LS does not significantly differ from the mean
score for SWB. However, while the mean score for SWB is within the normative
range, the fact that it lies at the bottom of the range has implications for some of
the hypotheses tested by the present study.
111
4.2
Hypotheses two and three: That adjectives located on the pleasantnessunpleasantness axis of the Circumplex Model of Affect will dominate
and explain significant variance in LS and SWB and that core affect
will explain greater unique variance in LS than in SWB
Consistent with hypothesis two, the adjectives happy (pleasant), content (pleasant)
and alert (activated) were found to be independent, significant predictors of both
LS and SWB. These three adjectives explained 59% of the variance in SWB and
57% of the variance in LS respectively, suggesting a strong affective component.
These findings are generally consistent with those of Davern, et al., (2007). These
authors also found that adjectives happy and content explained significant
variance in SWB, however, in their study, ‘excited’ (rather than ‘alert’) was found
to make a unique contribution. Together, these three adjectives explained 64% of
the variance in LS. According to the circumplex, excited represents pleasantactivated and is closer to the pleasant pole than alert. In the present study, excited
produced very low regression weights and did not contribute unique variance to
the prediction of either LS or SWB. Nonetheless, both alert and excited represent
activated-positive states, thus, the basic contribution of core affect to SWB has
been confirmed.
This relative amount of explained variance, however, is the reverse of the
theoretical ordering, with core affect explaining slightly more variance in SWB.
According to homeostasis theory, the domains contain more cognition, thus, core
affect should explain more variance in LS than SWB due to the differing levels of
abstraction. (Cummins et al., 2003). The results, however, showed that core affect
explained the same amount of unique variance in LS as in SWB (8.0%).
One explanation for these findings calls on the previous observation that the
sample has a relatively low mean, indicating the possibility of some students
being at risk of depression. These people may be responsible for the failure of
core affect to explain more of the variance in LS than SWB because of the
following: core affect is predicted to explain more variance in LS because it is
more abstract. However, this is dependent on a normally functioning homeostatic
system.
As
homeostasis
becomes
dysfunctional
under
challenge,
the
112
determination of both LS and SWB progressively shifts from core affect to the
challenging agent. In this process, core affect explains less unique variance in
both constructs. Moreover, the reduction in explained variance is somewhat
greater in LS because it is more abstract and therefore rather more determined by
core affect in the first place. So, the presumed sequence of change is as follows:
1.
When the system is functioning normally, core affect will explain more
variance in LS than in SWB.
2.
As the system is challenged and homeostasis loses its grip, core affect will
start to lose its affiliation with both LS and SWB. However, it will lose this
more in relation to LS because of its higher degree of normal dependence
on core affect.
Thus, a three stage sequence is envisaged. In normal conditions, the contribution
of unique variance from core affect should be LS > SWB. As homeostasis is
placed under challenge this should change to LS = SWB. Then, in conditions of
rampant homeostatic failure, the contribution of core affect to both constructs will
continue to fall to insignificance as the measure of both LS and SWB increasingly
represents affect generated by the challenging agent.
In a direct, empirical test of this explanation, all cases in the negative range for
SWB (<50%SM) were removed and regressions were conducted again. 50%SM
was used as a cut-off point because scores below this are outside the normal range
and represent a sub-group of people most likely to be experiencing severe
homeostatic defeat. Thus, only cases in the positive range for SWB were included
in the analyses. Consistent with theory, the results subsequently indicated that
core affect now explained 52% of the total variance in LS and 49% of the total
variance in SWB. Additionally in this truncated sample, core affect explained 3%
more unique variance in LS than SWB (11% versus 8% unique variance).
It is also notable that the mean score for each of the key affects (happy, content
and alert) were also slightly raised in this >50%SM sample, indicating that lower
ratings on these affects in the original analyses may have contributed to lack of
113
support for the hypothesis. It is therefore concluded that the variance contributed
by people with abnormally low SWB may have been responsible for the failure of
core affect to explain more of the total variance and unique variance in LS than
SWB across the entire data set. When the sample was restricted to people whose
SWB was in the normal range, the hypothesis was supported.
4.3
Hypothesis four: That the buffer variables will explain significant
variance in SWB above core affect in a group of people experiencing
homeostatic challenge
This hypothesis is based on homeostatic theory which states that the buffers
(positive cognitive biases of perceived control, optimism and self-esteem) will be
activated in an attempt to stabilise and maintain SWB within the normal range
during times of challenge (Cummins & Nistico, 2002). This was tested by the
separation of SWB into groups representing two levels of challenge as people
with SWB >70 and people with SWB <70. According to homeostasis theory, 70%
is thought to represent the homeostatic line of resistance (Cummins et al., (2003)
which the system is defending. Thus, on an average level basis, people with SWB
below this point are considered to be most likely in the challenged range for SWB.
Splitting the cases in this manner revealed that, as predicted, correlations between
the buffer variables and SWB were higher for people with SWB < 70%SM than
for people with SWB > 70 (.53 vs .31 respectively). Furthermore, when core
affect was entered into a partial correlation matrix as a co-variate, the reduction in
the correlation between SWB and the buffers was more pronounced in the group
of people with normal levels of SWB. These results support a number of
theoretical predictions as follows:
1.
The fact that the buffers correlated more strongly with SWB in the
<70%SM group is consistent with the idea that when SWB set-points are
challenged, the buffers will play a greater role in maintaining SWB stability.
2.
The fact that core affect, as co-variate, reduced the correlations between the
buffer variables and SWB more strongly in the >70 group, is consistent
114
with its greater influence with normal levels of SWB. When core affect was
entered into a step-wise multiple regression analysis at step 1 and the
buffers at step 2, the buffers added a further 15% (<.01) of variance in the
group with SWB < 70%SM compared with 2% (ns) of the variance for the
SWB > 70 group. Again, consistent with homeostatic theory, the buffers
became more activated and explained significantly more variance in the
challenged group than in the non-challenged group.
It is notable, however, that among the buffers, primary control was the only
significant, unique predictor. It is interesting that neither self-esteem nor optimism
contributed variance since they are also regarded as buffers in the homeostatic
model (see Cummins & Nistico, 2002). The reason for this may be that selfesteem and optimism share a greater amount of variance with core affect. In fact,
in the presence of core affect, the partial correlation between optimism and SWB
reduced from .45 to .23 and that between self-esteem and SWB reduced from .44
to .20. In comparison, the partial correlation between primary control and SWB
remained moderate at .46, indicating a greater degree of independence from core
affect. Thus, these results suggest that the relationship between optimism and
SWB and between self-esteem and SWB is strongly influenced by the relationship
these constructs share with core affect.
The following analysis was conducted using a second split of cases. The aim was
to further explore the relationship between the buffer variables and SWB using a
restricted sample of challenged individuals only. However, this time, the sample
was split using people with SWB between 45 and 69%SM. It was predicted that
when entered at step 2, the buffer variables will explain significant variance in
SWB above core affect.
4.4
Exploratory analysis: Predicting SWB using cases between 45 and
69%SM
Using the adolescent sample, the results indicate that correlations between the
buffer variables and SWB are generally lower when compared to the 0-70 group.
However, this has almost certainly occurred due to range restriction. In
115
accounting for the SWB variance, when core affect is entered at step one it
explains 14% of variance. This is considerably less than the 32% explained in the
0-70 group and the 27% explained for people with SWB >70. The higher
explained variance for the 0-70 group may also be due to the extended range of
values, however, the >70 group has an approximately equal range of values (7099) to the 45-69 group. The difference in explained variance between the 45-69
and 70-99 groups (27% vs 14% respectively) is consistent with homeostasis
theory. More specifically, that core affect will explain more variance in
populations with normal levels of SWB because core affect is the main driver of
SWB under conditions of low challenge (Davern, et al., 2007).
It is also notable that, in the challenged 45-69 group, core affect failed to add
significant, unique variance at step 2, with all of the significant variance explained
by primary and secondary control. This contrasts with the 0-70 group, where only
primary control made a unique contribution at step 2. Thus it appears that, when
cases in the defeated range for SWB are removed, secondary control becomes
activated and plays a supportive role to primary control in maintaining SWB
during times of challenge.
The aim of the next analysis was to further explore the relationship between
multiple discrepancies thoery and SWB using the challenged individuals only.
MDT represents a cognitive contribution to SWB measurement and theory
suggests that there should be greater relative contribution by MDT in challenged
samples. The reason for this is that when set-points are challenged, core affect
loses its grip over SWB and SWB becomes increasingly cognitive (Davern, et al.,
2007). Consistent with this theory, it was predicted that MDT would continue to
predict SWB above core affect in the 45-69 group. However, contrary to this
prediction, MDT did not contribute unique variance above core affect either in
this or in the > 70 group. The implication of this finding is that it appears that the
same cognitive processes that are driving SWB in challenged adult samples, for
example, perceived strong discrepancies between desired and experienced
conditions, are not operating to the same extent in the present sample of
adolescents.
116
4.5
Hypothesis five: That core affect is driving the relationship between
SWB and related variables.
The final hypothesis was that core affect would be driving the relationship
between SWB and the related constructs of perceived control, optimism, selfesteem, extraversion, neuroticism and MDT. To investigate this hypothesis,
partial correlations when controlling for core affect were examined and in support
of the hypothesis, the simple correlations between SWB and all related variables
reduced dramatically. In a number of instances, this reduction was greater than r
= .30. These findings suggest that a large body of research which supports a
relationship between SWB and related constructs should be re-examined in the
presence of appropriate affective controls. For example, research on self-esteem
(e.g., Cummins & Nistico, 2002; Diener & Diener, 1995); optimism (e.g.,
Cummins & Nistico, 2002; Lucas, Diener, & Suh, 1996; Scheier & Carver, 1985);
and perceived control (Cummins & Nistico, 2002; Myers & Diener, 1995;
Peterson, 1999).
Interestingly, the correlations between LS and the buffer variables underwent a
greater reduction than those correlations between SWB and the buffer variables
(an average reduction of .38 compared with .32 respectively). Again, this is
consistent with homeostatic theory which states that core affect more closely
involves LS because it is more abstract (Davern, et al., 2007).
In conclusion, the results support the hypothesis that SWB is primarily an
affective construct, with adjectives happy (pleasant), content (pleasant) and alert
(activated) explaining 59% of the variance in SWB and 57% of the variance in LS
respectively. Furthermore, consistent with Davern, et al., (2007), MDT and the
buffer variables (e.g., optimism) contributed an additional 7% variance above core
affect, suggesting that both affect and cognition are unique predictors of SWB. It
was also found that in challenged populations, primary control was the only buffer
variable to contribute unique, significant variance in SWB. In contrast to
homeostasis theory, neither self-esteem nor optimism contributed variance and
this is a particularly notable finding given that they are also regarded as buffers in
the homeostatic model (see Cummins & Nistico, 2002). Finally, partial
117
correlations suggest that core affect may be driving the relationship between SWB
and related constructs. In support of this, the relationship between SWB and
personality, the buffer variables and MDT reduced considerably in the presence of
core affect.
4.6
Exploratory analyses: Personality and MDT as predictors of SWB
A secondary aim of this study was to explore the relationship between constructs
believed to have an important influence over SWB. These include personality and
MDT. Considering personality first, there is a large body of research which
suggests a link between extraversion, neuroticism and SWB. Numerous authors
either argue for, or assume, personality as being one of the strongest and most
consistent single predictors of SWB (e.g., Brickman & Campbell, 1971; DeNeeve
& Cooper, 1998; Emmons & Diener, 1985; Vitterso, 2001; Vitterso, 2002) and
that it plays a crucial role in determining individual SWB set-points (Headey &
Wearing, 1989, 1992). Inconsistent with these views, hierarchical multiple
regression analysis demonstrated that extraversion and neuroticism (reverse coded
as emotional stability) accounted for no additional variance above core affect.
Furthermore, when core affect was entered into a partial correlation matrix as a
covariate, correlations between extraversion and SWB and between neuroticism
and SWB reduced dramatically (see Table 18). These findings are consistent with
Davern, et al., (2007) who also found that personality is not the driving force
behind SWB in the presence of suitable affective and cognitive variables. In fact,
similar to their conclusions, the data suggest that the relationship between
personality and SWB is dependent on the shared variance from core affect. Thus,
if core affect is driving both personality and SWB, then individual differences in
set-point for core affect are causing personality and SWB to correlate. This result,
if confirmed, suggests that previous research supporting strong correlations
between personality and SWB should be revisited.
In terms of MDT (Michalos, 1985), once again there is a strong tradition of
thought that SWB includes a cognitive component (e.g., Campbell, Converse, &
Rodgers, 1976; Davern, et al., 2007; Diener & Diener, 1996; Steel & Ones, 2002;
Veenhoven, 1994). Moreover, since cognitive processing must be necessary for
118
people to make judgments of wellbeing about past, present and future aspirations
and goals and needs in relation to their current condition, it seems reasonable to
assume that this component can be investigated through MDT. Using the entire
sample, hierarchical regression analyses, with core affect entered at step 1 and
MDT at step 2, revealed that MDT explained an additional 4.0% variance. In
support of this result, Davern, et al., (2007) argue that MDT plays a strong but
subsidiary role in explaining variance in SWB. Thus, while SWB is highly
affective in nature, it has now been empirically demonstrated in two different
samples that MDT makes an important independent contribution; indicating that
discrepancy judgments may well comprise the cognitive component of SWB.
To explore whether MDT and the buffer variables are explaining the same or a
different portion of the remaining variance in SWB above core affect, an
additional hierarchical regression analyses was conducted with the buffer
variables entered at step 2 and MDT at step 3. It was found that MDT added a
further 2.0% variance above that explained by core affect and the buffer variables,
indicating that although there is a degree of independence between these
constructs, the buffer variables and MDT share some variance. Altogether, 64% of
the variance was explained by core affect (57%), the buffer variables and MDT (a
further 7% variance combined). This result again supports Davern, et al., (2007)
who found that SWB is primarily driven by affects, with cognition playing a
subsidiary role. Furthermore, findings are in agreement with research suggesting
that SWB is a construct involving both affective and cognitive processes (e.g.,
Diener, 1984; Diener & Diener, 1996; Veenhoven, 1994).
In order to explore whether these findings generalise to adult data, parallel
analyses were conducted on an adult data set with a larger sample.
4.7
Comparative analyses against an adult data Set
A further aim of this study was to compare data from adolescents and adults in
order to observe similarities and differences between the two age groups.
According to the data, there is a great deal of consistency in the amount of
explained variance in SWB accounted for by core affect between the adult data of
119
Davern, et al., (2007) (64%), the present adolescent data (57%) and the
comparative adult data set (62%). However, the simple correlations between the
buffer variables and SWB were generally higher in the adolescent sample.
Interestingly, this difference is largely attributed to the correlations with primary
control (.32 higher) and secondary control (.23 higher). This suggests that
perceived control shares a stronger relationship with SWB in the adolescent
sample. Thus, suggesting the buffer variables may be operating more
independently of core affect in the adolescent sample.
To explore whether the buffer variables contribute additional variance in SWB
above core affect across the entire adult sample, a regression analysis was
conducted. When entered at step 2, the cognitive buffer variables (optimism and
secondary control) added a further 2% (p< .05) of variance above core affect. This
is 3% less than that contributed by optimism within the adolescent data. Thus,
whilst in both samples, core affect is accounting for a majority of the variance in
SWB, a greater portion of the remaining variance is explained by the buffer
variables (optimism) in the adolescent sample. These results suggest that
optimism is explaining a greater proportion of the remaining variance in SWB
above and beyond that explained by core affect in the adolescent sample than is
optimism and secondary control in the adult sample.
A further aim of these comparative analyses was to explore the relative
contribution of core affect and the buffers for adolescents and adults with SWB >
70%SM and between 45 and 69%SM. It was found that within the >70 group,
core affect, at step 1, explained 10% more variance in the adult sample than in the
adolescent sample (37% versus 27%).
In the ‘challenged’ 45-69 group, core affect again explained 4% more variance in
the adult sample than in the adolescent sample at step 1 (18% versus 14%
respectively). However, this difference was not significant and may be accounted
for by measurement error. Thus, in challenged groups, core affect appears to be
operating similarly in adults and adolescents. One major difference between the
two samples, however, is that primary control (sr² = .08) and secondary control
(sr² = .11), when entered at step 2, accounted for additional variance in the
120
adolescent sample, compared with no additional variance contributed by any of
the buffer variables in the adult sample. Thus, it can be concluded that primary
control and secondary control appear more activated in the challenged adolescent
sample than in the challenged adult sample and contribute significantly more
variance in SWB.
Finally, the influence of discrepancy judgements was weak throughout. While
MDT explained an additional 5% variance above core affect for adults, it
explained no additional variance within the adolescent sample. According to
Michalos (1985), a close fit between needs, wants and desires etc. is necessary to
produce feelings of happiness, contentment and satisfaction - pleasant affects
associated with life satisfaction. It appears that after controlling for pleasant affect,
the strength of perceived discrepancies had no additional influence on SWB in the
adolescent sample. Thus, although it is argued that cognitive processes dominate
assessment of discrepancies, evidence supports MDT as largely a measure of
pleasant affect.
4.8
Exploratory analyses: School satisfaction as a unique construct
The aim of this analysis was to examine whether satisfaction with school might be
included as a new domain in the PWI-SC. A new domain must predict unique,
significant variance in global life satisfaction, above the other seven life domains.
To explore whether satisfaction with school met this criterion, it was entered at
step 2 of a hierarchical regression analysis, following all seven domains entered at
step 1. Despite correlating fourth highest with life satisfaction (r = .53), with a
standardised regression weight of  =.11 (fifth highest), satisfaction with school
contributed a sr² value of .01 (ns). Thus, satisfaction with school did not meet the
criteria for consideration as a new domain. It is interesting to note, however, that
four existing domains also failed to meet this criterion. In fact, the only domains
to independently predict life satisfaction were satisfaction with health, standard of
living and safety. This finding is in contrast to research conducted by Cummins et
al., (2003) on adult populations, where all domains, except safety, are consistent
unique predictors of LS.
121
As further evidence that the domains are operating differently between the PWISC and PWI-A, the domain of safety contributed 2% unique variance to the
prediction of LS (p<.05). According to research conducted on the PWI-A
(International Wellbeing Group, 2005), the domain of safety rarely makes a
unique contribution in Australia, but is retained because it does so in other
countries. Finally, the domain of ‘health’ topped the list of domains that
contributed the most unique variance (sr² = .07) in the PWI-SC. Adult data
(Cummins et al., 2006) informs that this domain often ranks 3rd or 4th in adult
populations (N = approximately 30,000) in terms of both mean score (75.12%)
and unique variance contributed to LS (1%). Further, this domain correlates with
LS at between .35 (Survey 9) and .41 (Survey 1) respectively.
There are a number of potential explanations for these differences. The first is that
while adolescents have been shown to experience levels of positive affect similar
to adults (e.g., Davern, et al., 2007) they may have only a small number of
concerns in comparison to those of adults (Cummins et al., 2006). The finding that
standard of living, health and safety were the only domains to predict unique
variance in LS is consistent with this notion. Looking first at standard of living, it
seems intuitive that adolescents would have a preoccupation with material
possessions, such as whether or not they have the latest game console or whether
their new shoes are better than those of their friends. Keeping up to date with the
latest trends and a lifestyle of relative comfort would certainly be favored by an
adolescent keen on making an impression on his/her friends. Thus, it is not
surprising that standard of living would be related to overall life satisfaction in
adolescents.
Turning now to the domain of safety and safety is an issue that has been at the
centre of school and government education programs in Australia for many years.
A number of initiatives have been implemented to educate children and
adolescents and to increase their awareness on safety and related issues. Examples
of some of these programs include Quit Victoria (www.quit.org.au) – a joint
venture between the Cancer Council Victoria, The Department of Human Services,
the National Heart Foundation and Vic Health, who are dedicated to educating
adolescents and adults about the dangers of smoking and tobacco. Kid Power
122
(www.kidpower.org/School-age.html) is another organisation comitted to
teaching children to learn and practice different ways of being powerful and safe.
Given the saliency of this domain amongst adolescents, we would expect this
domain to contribute uniquely to LS.
Finally, it would be reasonable to assume that satisfaction with health would be
closely related to adolescents’ overall life satisfaction. Although perceived health
concerns amongst adolescents will likely differ from those of their parents and
grandparents, issues relating to health and ill health are still important. For
example, broken bones, sprained ankles and the common cold are all relatively
minor ailments, however, these have the potential to prevent adolescents from
engaging in activities that are fun and exciting. Thus, for the majority of
adolescents who are in good health, satisfaction with this domain is likely to be
closely related to LS.
A second explanation for these findings requires consideration of two factors.
Firstly, it is important to highlight that the amount of explained variance
contributed by standard of living, health and safety (R2 = .55) is marginally
greater than that observed using all seven domains in adult populations (R2 =.48 .52; International Wellbeing Group, 2006). Secondly, the amount of unique
variance contributed by the significant domains in the present study was also
comparable to that frequebtly observed in the Australian Unity Wellbeing Index
(approximately 13% versus 13-17%). Given that the amount of explained variance
and unique contributions from all 7 domains to LS is relatively comparative
between the two age groups, a major statistical difference between the present
study and analyses conducted using the Australian Unity Wellbeing Index data
must be brought to attention - that of sample size. In the present study, analyses
were conducted with 146 cases. According to Tabachnick and Fidell (2001), using
the formula:
N> (8/f2) + (m - 1) where f2 = .01, .15 and .35 for small, medium and large effect
sizes respectively
123
the sample size of the present study was adequate and statistical analyses did not
lack power. It could be argued that regressions carried out on approximately 2000
cases in the Australian Unity Wellbeing Index may be responsible for the majority
of domains contributing unique variance in LS. According to Tabachnick and
Fidell, it is possible to have too many cases and as the number of cases becomes
quite large, any multiple correlation or regression coefficient may reach
significance, even one that predicts negligible variance. Thus, in the present study,
the failure of four domains to predict unique variance in LS, despite the remaining
three contributing the same or more variance than all six or seven in the
Australian Unity Wellbeing Index, may be an issue of sample size. These findings
will be examined further in study 2.
4.9
Summary
Collectively, these results generally support the notion that SWB is primarily an
affective construct with minor independent contribution from cognition (Davern,
et al., 2007). This view is confirmed by the finding that 57% of the variance in LS
and 59% of the variance in SWB was explained by three key affects (happy,
content and alert). MDT and the buffer variables (optimism) contributed an
additional 7% variance above core affect, suggesting that both affect and
cognition are unique predictors of SWB. Thus, affects happy, content and alert
were used to represent the measure of core affect, as initially conceived by Russell
(2003).
Further analysis provided support for the idea of homeostasis. For example,
according to homeostasis theory, core affect should explain more unique variance
in LS than the PWI because LS is more abstract (Cummins & Nistico, 2002;
Davern, et al., 2007). While using the whole sample it was found that core affect
explained the same amount of unique variance in LS and SWB; when the sample
was restricted to people scoring >50 points, core affect explained more variance
in the predicted direction. This finding confirms homeostasis theory and
demonstrates the importance of understanding the sample characteristics when
engaging in such investigations.
124
Using a split of cases below 70%SM and between 45 and 69%SM, the hypothesis
that the cognitive buffer variables will contribute additional variance above core
affect in populations experiencing some level of homeostatic challenge was
partially supported. When entered at step 2 of a hierarchical regression analysis,
primary control contributed unique variance in the <70 group. In the 45-69 group,
primary control and secondary control were significant, unique predictors. These
findings cast some doubt on particular aspects of homeostasis theory. More
specifically, this theory predicts that all the buffer variables will be activated in an
attempt to maintain and stabilise SWB within the set-point range. Examination of
simple and partial correlations indicated that the relationship between the nonsignificant buffers of self-esteem and optimism with SWB may be dependent on
shared variance from core affect. Thus, primary control and secondary control are
relatively more independent of core affect and this may explain their significant
contribution. This leads to the next conclusion – that core affect appears to be
driving the relationship between SWB and constructs normally considered
independent of one-another in the literature.
When variance attributed to core affect was removed from such relationships,
their shared variance dramatically reduced, indicating that the relationship
between these constructs and SWB may be driven by shared variance from core
affect. Of particular interest here is the dramatic decrease in the relationships
between extraversion and SWB and between emotional stability and SWB. A
major implication of these findings is that a large body of research which
presumes an independent link between personality and SWB should be revisited
with adjustments made for relevant affective variables.
It is also notable that in the adolescent sample, the cognitive component of SWB,
measured using MDT, failed to contribute unique variance above core affect in
the 45 to 69 group. The implication of this finding is that, contrary to previous
research (e.g., Davern, et al., 2007), the cognitive component of SWB may not
comprise discrepancy judgments as in MDT. Alternatively, the data suggest that
this cognitive component may be explained elsewhere, for example, in thoughts
pertaining to perceptions of control. According to these results, in the presence of
core affect, primary control and secondary control share the strongest unique
125
relationship with SWB. Thus, control beliefs appear to be driving a portion of the
remaining variance in SWB above core affect. This seems intuitive; for example,
when a person’s wellbeing is challenged, belief that circumstances are within
his/her immediate control may have a powerful influence over his/her wellbeing.
Finally, it was found that school satisfaction did not contribute unique variance
beyond the seven domains and so cannot be considered as an additional domain.
Interestingly, however, neither did satisfaction with achieving in life, relationships,
future security or satisfaction with community connectedness. Moreover,
satisfaction with health made a more substantial contribution in the adolescent
sample than in the adult sample. Taken together, these findings have important
implications regarding the validity of the PWI-SC, which is based on the premise
that all seven domains represent the first level deconstruction of life as a whole
and should contribute uniquely to LS.
4.10 Conclusions
These results generally confirmed the predictions derived from homeostasis
theory. The results also provided overwhelming support for the hypothesis that
SWB is primarily an affective construct. According to the data, core affect
explained a significant portion of variance in both LS and SWB. Furthermore,
core affect appears to be driving the relationship between SWB and related
constructs, suggesting that previously reported correlations between SWB and
constructs such as self-esteem, optimism, perceived control, MDT and personality,
may be driven by shared variance with core affect. Notably, however, in
challenged populations, it was found that primary control (in the <70 group) and
primary control and secondary control (in the 45-69 group) were the only buffer
variables to remain independently linked to SWB in the adolescent sample. Thus,
maintaining a sense of control during times of challenge may be critical and
representative of a strategic use of cognition by adolescents to restructure their
experiences so as to prevent, or at least minimize, the impact of challenging and
unavoidable events. For example, blaming others when things go wrong and
believing that ones circumstances will improve with intervention are cognitive
strategies that may be used by adolescents to reinforce a sense of wellbeing.
126
Finally, satisfaction with school did not meet criteria as a new unique construct
and thus, did not qualify as a new domain to be included in future revisions of the
PWI-SC.
127
CHAPTER 5: STUDY 2
INTRODUCTION
The first study determined that three affects, which represent the pleasantunpleasant axis and activation-deactivation axis of the Circumplex Model of
Affect, may be considered to represent a measure of core affect as initially
conceived by Russell (2003). It was found that, the three affect terms - happy,
content and alert - explained 57% of the variance in LS and 59% of the variance
in SWB respectively. These results support the notion that SWB is primarily an
affective construct, with minor independent contribution from cognition (e.g.,
from MDT and optimism). Importantly, these results also cast doubt on a large
body of research which suggests that personality is the main driver of SWB.
Hierarchical multiple regression analyses indicated that, in the presence of core
affect, extraversion and emotional stability did not contribute additional variance
to the prediction of SWB.
The aim of this second study is to replicate study 1 with a new sample of highschool students. More specifically, Structural Equation Modeling (SEM) will be
used in addition to multiple regression analyses, to examine a) the structure of
affect and b) the structure of SWB. Moreover, SEM and associated model fit
statistics will be used to determine which theoretical model provides the best fit to
the data. This analysis will determine whether a purely Affective model for SWB
is superior to a Personality-driven model, a Cognitive model, or a model for SWB
that comprises both an affective and a cognitive component.
As in study 1, a further aim is to explore whether satisfaction with school will
contribute additional variance in LS above all 7 domains, thus, qualifying as a
new domain in the PWI-SC. Numerous items reflecting various aspects of
satisfaction with school have also been included in the questionnaire so as to
determine what aspects of school life contribute uniquely to the prediction of this
construct.
128
METHODOLOGY
Participants
The 205 participants were attending various high-schools in the Melbourne
metropolitan region and country Victoria. This sample comprised 53 males
(25.9%) and 152 females (74.1%), representing 2 students in year 7 (1.0%), 5
students in year 8 (2.4%), 6 students in year 9 (2.9%), 43 students in year 10
(21.0%), 49 students in year 11 (23.9%) and 100 students in year 12 (48.8%).
Participants’ ages ranged from 13 – 20 years, with a mean age of 16.7 years.
Questionnaire
A 67-item paper and pencil questionnaire titled The Young Australian Wellbeing
Index (see Appendix B) was self-completed by each student, under conditions of
information privacy, in their regular classroom. The questionnaire is similar to
that used in study one; however, 9 additional items assessing satisfaction with
various aspects of school life were also included (see Appendix B, items 14-22).
5.1
Major Dependent Variable and Other Variables
The same measures of LS, SWB, core affect, primary control, secondary control,
self-esteem, optimism, MDT, personality and school satisfaction were used as for
study one. Chronbach’s  for the scaled variables are as follows: SWB (.85); selfesteem (.82); optimism (.89); primary control (= .79); secondary control (74);
extraversion (= .74); and emotional stability (.54)
5.2
Other Measures: School Index
A further 9 items adapted from the Piers-Harris Children’s Self-Concept Scale
(Piers, 1986) were included in this study. These items were designed to tap levels
of satisfaction with various aspects of school life and were measured using an 11point end defined scale. Examples include ‘How satisfied are you with your
abilities at school’ and ‘How satisfied are you with your teachers at school’ (0
=Very Dissatisfied; 5 = Neutral and 10=Very Satisfied) (see Appendix B, items
14-22). Cronbach’s alpha for these 9 items was high at .87.
129
5.3
Procedure
After obtaining approval from the Deakin University Human Research Ethics
Committee – health and behavioural sciences sub-committee to conduct the study,
approval was again sought from the Department of Education and Training and
the Catholic Education Office. Once these authorities had given their approval,
various high schools in the Melbourne metropolitan region and country Victoria
were contacted by phone or e-mail. As in study 1, a representative from each
school was briefed on the proposed study, its aims, obligations as a participating
school and responsibilities of the researcher. Of the 17 schools contacted, four
agreed to take part in the study.
As this study is a replication of study 1, a more detailed description of the
procedure can be found in the Method section of study 1.
5.4
Data Analytic Strategy
In addition to multiple regression analysis, data will be analysed using SEM in
Amos (Version 6.0). SEM is similar to multiple regression, however, it takes into
account the modeling of interactions, nonlinearities, correlated independent
variables, measurement error, correlated error terms, multiple latent independent
variables each measured by multiple indicators, and one or more latent dependent
variables, also each with multiple indicators (Garson, 1998). Researchers are
increasingly turning to SEM as an alternative to path analysis, factor analysis,
time series analysis, and analysis of co-variance (which are all seen as special
cases in SEM) (Kline, 1998). The benefits of using SEM over these alternatives
include: more flexible assumptions; the ability to test models rather than
coefficients alone; the ability to model mediating variables; the ability to test
models with multiple dependants; the attractiveness of SEM’s graphical modeling
interface; and the ability to compare and pit different theoretical models against
one another to determine the best model fit (Garson, 1998). In the present study,
SEM will be used to determine which theoretical model fits the data best.
130
Assessing Goodness of Fit: Which Statistics to Interpret
When assessing how well a structural equation model fits the data, it is important
to highlight the inappropriate reliance on the chi-square statistic Kline (1998).
Chi-square is a measure of the discrepancy between the saturated (best-fitting)
model and the model being tested. Accompanying the chi-square statistic is a p
value that indicates significance of difference between a model and the data - if
p>.05, then model and data are not significantly different from one another,
indicating a close fit. Although the present sample of 205 cases meets the
requirement for adequate sample size (to be discussed in the results section),
according to Kline (1998) chi-square is overly sensitive to sample size and as a
result, large sample sizes may reach significance even when there is only a small
discrepancy between data and model. One technique that reduces sensitivity of the
chi-square statistic is to divide chi-square by degrees of freedom (χ2/df). Ideally,
χ2/df should not be greater than 3.0, indicating model parsimony. Once calculated,
χ2/df can be interpreted alongside absolute and incremental fit indices. Kline
further adds that the fit statistics chosen for interpretation will vary from
researcher to researcher, however, the author encourages the presentation of a
minimal set that includes: (χ2/df); an index describing overall proportion of
explained variance (e.g., Squared Multiple Correlation; SMC); an index based on
standardised residuals (e.g., Root Mean Squared Error of Approximation;
RMSEA); and an index detailing the discrepancy between the saturated (bestfitting) and independence model (e.g., Normed Fit Index; NFI). A RMSEA value
of .05 or less would indicate a close fit of the model in relation to the degrees of
freedom. Furthermore, the value for NFI should approximate 1.0, indicating little
or no difference between the best-fitting model and independence model; however,
an NFI >.90 is acceptable. These criteria will be adopted in the analyses to follow.
Finally, Akaike Information Criterion (AIC) will be used to assess the complexity
of the specified models. The AIC is a modification of standard goodness-of-fit
statistic (chi-square) that includes a penalty for complexity (Kline, 1998).
According to Kline, AIC is analogous to indexes of model fit; and the AIC for a
model is computed using the following formula: AIC = χ2 – 2df. Thus, complex
models (with fewer degrees of freedom) undergo larger reductions in their χ2
131
values. Lower AIC values are preferred as these represent more parsimonious
models.
RESULTS
5.5
Data Screening and Preliminary Analyses
SPSS software (version 14.0) was used for data screening and analysis. As in
study 1, all scores have been converted to a Percentage of Scale Maximum
(%SM).
5.5.1
Missing Data
The frequency of missing data for all variables across the entire data set was less
than 5%. As before, values were replaced by regression.
5.5.2
Outliers
Examination of z-scores and scatter plots revealed univariate outliers on domain
satisfaction variables (with the exception of satisfaction with health) and core
affect items (with the exception of ‘alert’ and ‘quiet’). However, comparison of
mean scores on these variables with corresponding means trimmed at the upper
and lower 5% revealed that none of these outliers significantly influenced mean
scores on key variables (Pallant, 2001). As a consequence, univariate outliers
were included in analyses. No multivariate outliers were identified with a
Mahalanobis distance greater than 20.515 (critical x2 = 20.515, p < .001), a
criterion recommended by Tabachnick and Fidell (2001) for the corresponding
degrees of freedom.
5.5.3
Normality and Linearity
Normality was assessed across the entire data set. Using the SPSS descriptive
dtatistics function, negative skews were found in the following variables: LS (z= 5.85), standard of living (z = -3.80), health (z = -3.50), achieving in life (z = 4.20), relationships (z = -6.30), safety (z = -6.98), community connectiveness (z =
-5.71), future security (z= -4.01) the PWI (z = -4.71), happy (z = -5.10), content (z
132
= -3.90), unhappy (z = -3.20), excited (z= -3.20), sleepy (z = -3.01) and
satisfaction with school (z = -3.40). According to Cohen and Cohen (1983),
skewness and kurtosis are acceptable within the range of -7.0 to 7.0. Thus, no
variables underwent transformation.
5.5.4
Multicollinearity and Singularity
All major independent variables, including LS, SWB, core affect, personality,
MDT, primary control, secondary control, optimism and self-esteem were tested
for multicollinearity and singularity. The highest observed correlations were
between the variables ‘happy’ and ‘content’ (r = .86), ‘LS’ and ‘happy’ (r = .82),
‘unhappy’ and ‘discontent’ (r = .80), ‘LS’ and ‘content’ (r = .74), ‘LS’ and ‘SWB’
(r = .73), ‘SWB’ and ‘content’ (r = .71) and ‘SWB’ and ‘happy’ (r = .70). For the
reasons discussed in study 1, all items and composite scores were retained.
5.5.5
Sample Size
According to Tabachnick and Fidell (2001), the criterion for multiple regression
analysis is:
N > 50 + 8m
where N= minimum number of cases and
m = number of IV’s
In the present study, the maximum number of independent variables entered in
any one regression analysis is 10. According to the rule:
N > 50 + 8 x 10
it is recommended that a minimum of 130 cases are needed for adequate statistical
power. With 205 cases, the present study meets this power requirement for all
major multiple regression analyses. As SEM relies on tests which are sensitive to
sample size as well as to the magnitude of differences in covariance matrices, it
has been suggested that a sample size of at least 100 cases, preferably 200, is
necessary (Loehlin, 1992). The present study meets this requirement.
133
5.6
Core Affective Adjectives as Predictors of SWB
Using 9 core affect adjectives as predictors, the aim of the first two multiple
regression analyses were to repeat the analysis undertaken in study 1, by
determining which of these contributed most strongly to the prediction of LS and
SWB.
Similar, modified hypotheses are to be tested as for study 1. These are:
1.
That the mean score for LS will approximate the mean score for SWB and
that the mean SWB score will lie within the Australian adult normative
range of 73.43 to 76.43%
2.
That adjectives happy, content and alert will dominate and explain
significant variance in SWB
3.
That core affect will explain greater unique variance in LS than in SWB
when the system is functioning normally
4.
That in circumstances of homeostatic challenge, but not in homeostatic rest,
primary control and secondary control will explain variance in SWB
beyond core affect
5.
That core affect is driving the relationship between SWB and related
constructs. These include perceived control, optimism, self-esteem,
extraversion, emotional stability and MDT
6.
That in confirmation of study 1, only the three domains of health, safety
and relationships will contribute unique variance to LS
To test the first hypothesis, that the mean score for LS will approximate the mean
score for SWB, a t-test was conducted. Confirming this hypothesis, the mean
score for LS (M = 72.15, SD = 20.27) did not significantly differ from the mean
score for SWB (73.61.61, SD = 14.18), t(204)= 1.515, p=.131. Table 36 displays
means, standard deviations and correlations between variables.
Table 36:
Means, standard deviations and correlations between variables (N=205)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
1.
LS
72.15
20.27
-
2.
SWB
73.61
14.18
.73
-
3.
Happy
69.61
19.70
.82
.70
-
4.
Content
68.63
19.10
.74
.71
.86
-
5.
Unhappy
38.20
21.94
-.66
-.52
-.69
-.64
-
6.
Discontent
38.39
21.44
-.56
-.47
-.53
-.52
.80
-
7.
Active
62.78
21.16
.42
.47
.47
.50
-.35
-.25
-
8.
Alert
64.68
18.24
.40
.51
.44
.49
-.30
-.23
.64
-
9.
Excited
68.78
19.10
.47
.50
.59
.53
-.36
-.23
.46
.50
-
10. Sleepy
61.86
24.52
-.16
-.18
-.13
-.13
.16
.10
-.24
-.26
-.09
-
11. Quiet
47.61
24.75
-.27
-.19
-.28
-.22
.37
.33
-.21
-.17
-.32
.23
11.
-
134
135
Confirming the second part of the first hypothesis and consistent with both SWB
homeostasis and with Australian adult normative statistics, the mean score for
SWB (73.61) was just within the adult normative range of 73.43 to 76.43%SM
(Cummins et al., 2006). Furthermore, the mean score for SWB in this sample was
not significantly different from that found in study 1 (73.90, SD = 13.95), t(349)
= .19, p>.05. However, the mean score for LS (72.15) was below the Australian
adult normative range of 75.20%SM to 79.10%SM (Cummins et al., 2006). It is
also noteworthy that while the standard deviation for SWB is slightly greater than
study 1 (14.18 vs. 13.95) the standard deviation for LS is notably larger (20.27 vs.
18.05, a difference of + 2.22 percentage points). Taken together, these results
indicate that the mean score for SWB is at the lower end of the normal range and
the variance has increased due to a greater than normal proportion of people with
low wellbeing. In fact, the number of people with very low SWB (<50) in this
sample is 14 (6.8%) - compared with 7 people (4.7%) in study 1.
To compare study 1 and study 2 mean scores and standard deviations for LS,
SWB and affects, Table 37 has been constructed.
Table 37:
Comparing study 1 and study 2 mean scores and standard deviations for LS, SWB
and affects
Variable
Mean (study one) Mean (study two) SD (study one) SD (study two) Significance
(N=146)
(N=205)
(N=146)
(N=205)
of Difference
1. LS
75.00
72.15
18.05
20.27
p > .05
2. SWB
73.90
73.61
13.95
14.18
p > .05
3. Happy
74.86
69.61
18.13
19.70
p < .01
4. Content
73.49
68.63
17.83
19.10
p < .05
5. Unhappy
35.88
38.20
21.72
21.94
p > .05
6. Discontent
38.54
38.39
23.31
21.44
p > .05
7. Active
69.32
62.78
22.09
21.16
p < .01
8. Alert
70.48
64.68
20.25
18.24
p < .01
9. Excited
72.47
68.78
19.43
19.10
p < .05
10. Sleepy
58.70
61.86
25.97
24.52
p > .05
11. Quiet
47.74
47.61
26.75
24.75
p > .05
136
While the mean scores for LS and SWB do not differ between the two samples,
scores on key affects are generally lower in study 2. The mean score for happy is
lower by 5.25 percentage points, t(349) = 2.58, p <.01; content by 4.86 percentage
points, t(349) = 2.44, p <.05; and alert by 5.8 percentage points, t(349) = 2.75, p
<.01. Also, the mean score for active is lower by 6.54 percentage points, t(349) =
2.78, p <.01; and excited by 3.69 percentage points, t(349) = 1.77, p <.05.
In order to compare correlations between LS, SWB and affects between studies 1
and 2, Table 38 has been constructed.
Table 38:
A comparison of means, standard deviations and correlations between variables
Variable
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
1.
LS
-
.73
.82
.74
-.66
-.56
.42
.40
.47
-.16
-.27
2.
SWB
.73
-
.70
.71
-.52
-.47
.47
.51
.50
-.18
-.19
3.
Happy
.68
.74
-
.86
-.69
-.53
.47
.44
.59
-.13
-.28
4.
Content
.65
.68
.76
-
-.64
-.52
.50
.49
.53
-.13
-.22
5.
Unhappy
-.53
-.49
-.59
-.45
-
.80
-.35
-.30
-.36
.16
.37
6.
Discontent
-.43
-.33
-.37
-.40
.65
-
-.25
-.23
-.23
.10
.33
7.
Active
.38
.43
.39
.37
-.19
-.16
-
.64
.46
-.24
-.21
8.
Alert
.58
.52
.49
.53
-.27
-.22
.48
-
.50
-.26
-.17
9.
Excited
.54
.59
.69
.65
-.30
-.22
.41
.55
-
-.09
-.32
10. Sleepy
-.21
-.13
-.16
-.09
.20
.15
-.19
-.16
-.14
-
.23
11. Quiet
-.23
-.27
-.28
-.24
.25
.14
-.24
-.13
-.27
.38
-
N=146 (study 1); N = 205 (study 2)
As found in study 1, LS shares the strongest relationship with adjectives located at
the ‘Pleasant’ pole of Russell’s (2003) Circumplex Model of Affect. These
include ‘happy’ (r = .82) and ‘content’ (r = .74). SWB also shares the strongest
relationship with ‘happy’ (r = .70 SWB) and ‘content’ (r = .71). Significant
differences in correlations between LS and affects for studies 1 and 2 can be
observed for happy, z = 3.00, p<.01 and alert, z = 2.19, p<.05. When compared to
study 1, the correlation between LS and happy increased by .14 whilst the
correlation between LS and alert reduced by .18. Finally, no significant
137
differences in correlations between SWB and affects were observed in the two
samples.
To more thoroughly test the second hypothesis that affective adjectives will
account for a significant portion of total variance in LS and SWB and the third
hypothesis that affective adjectives would account for greater unique variance in
LS than SWB, additional multiple regression analyses were conducted. The means,
standard deviations and regression coefficients are presented in Tables 39 and 40.
Table 39 provides details of the relative contribution of each affect to the
prediction of LS.
Table 39:
Predicting LS by nine affect adjectives
1. Happy
DV: Life
Satisfaction
.82
2. Content
Variable
B

sr2
.64***
.62
.08
.74
.08
.08
.00
3. Unhappy
-.66
-.06
-.06
.00
4. Discontent
-.56
-.13
-.13
.01
5. Active
.42
.02
.02
.00
6. Alert
.40
.02
.02
.00
7. Excited
.47
-.02
-.01
.00
8. Sleepy
-.16
-.03
-.04
.00
9. Quiet
-.27
* p<.05; ** p<.01; *** p<.001
.00
.00
Unique Variance = .09
Shared Variance = .59
.00
2
R = .70
Adjusted R2 = .68
The R for this regression is significantly different from zero, F(9, 195) = 49.824, p
< .001; and only one affect contributed significant unique variance to the
prediction of LS: happy (sr2 = .08). Altogether, 68% of the variability in LS was
predicted, indicating a significant influence of core affect on LS. This contrasts
with study 1 where the three affects that contributed significant unique variance to
the prediction of LS were alert (sr2 = .04), happy (sr2 = .02) and content (sr2 = .01).
138
Table 40 provides details of the relative contribution of each affect to the
prediction of SWB.
Table 40:
Predicting SWB using nine affective adjectives.
Variable
DV:SWB
B
β
sr2
1.
Happy
.70
.21**
.29
.02
2.
Content
.71
.23***
.30
.02
3.
Unhappy
-.52
.07
.11
.00
4.
Discontent
-.47
-.13*
-.20
.01
5.
Active
.47
.02
.04
.00
6.
Alert
.51
.12*
.15
.01
7.
Excited
.50
.06
.09
.00
8.
Sleepy
-.18
-.04
-.07
.00
9.
Quiet
-.19
.03
.06
* p<.05; ** p<.01; *** p<.001
Unique Variance = .06
Shared Variance = .53
.00
2
R = .59
Adjusted R2 = .57
The R for this regression was significantly different from zero, F(9, 195) = 31.030,
p < .001. However, this time, four affect items contribute significant, unique
variance to the prediction of SWB: content (sr2 = .02), happy (sr2 = .02), alert (sr2
= .01) and discontent (sr2 = .01). Altogether, 59% of the variability in LS can be
predicted from scores on these four affect items, indicating a significant influence
of core affect on SWB. This result is again different from this same analysis
performed in study 1 where only happy and content contributed unique variance.
In partial support of the second hypothesis, affects located on the pleasantunpleasant axis of the circumplex dominate the explained variance for both LS
and SWB. Moreover, in line with the third hypothesis that core affect will account
for more variance in LS than SWB, core affect explained 68% of the variance in
LS and 57% of the variance in SWB. Finally, consistent with the second part of
139
the third hypothesis, the affective adjectives account for greater unique variance in
LS (9%) than SWB (6%).
As in study 1, additional analyses were conducted to determine whether the
affective adjectives would explain more unique variance in LS than SWB when
people with low SWB were removed. To reiterate the logic of this procedure,
when the homeostatic system is defeated, core affect will lose its affiliation with
both LS and SWB. However, it will lose this more in relation to LS because of its
higher degree of normal dependence on core affect. Thus, by removing cases in
the negative range for SWB (<50), it is expected that core affect will account for a
greater percentage of unique variance in LS than in SWB.
Table 41 Displays means, standard deviations and correlations between variables
for the group with SWB >50%SM.
Table 41:
Means, standard deviations and correlations between variables (n=191 >50%SM)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
1.
LS
74.50
18.20
-
2.
SWB
75.93
11.49
.67
-
3.
Happy
71.78
17.53
.77
.66
-
4.
Content
70.84
17.45
.67
.64
.84
5.
Unhappy
36.39
20.55
-.61 -.46 -.65 -.59
-
6.
Discontent
36.65
20.60
-.49 -.37 -.46 -.43
.78
7.
Active
64.08
20.44
.37
.42
.44
.46 -.30 -.17
-
8.
Alert
66.18
17.34
.35
.46
.40
.44 -.27 -.17
.64
-
9.
Excited
70.26
18.05
.41
.43
.53
.46 -.32 -.15
.44
.45
10. Sleepy
62.05
24.25
-.27 -.30 -.25 -.23
.23
.15 -.31 -.31 -.18
-
11. Quiet
46.39
24.04
-.26 -.12 -.27 -.20
.37
.33 -.20 -.16 -.34
.20
11.
-
-
-
As expected, the results presented in Table 41 indicate that in the >50% group, the
mean score for LS, SWB and key affects has risen as a consequence of removing
-
140
cases in the negative range for SWB. However, similar to the results for the whole
sample (see Table 40), the mean scores on key affects following this procedure
are generally lower than in study 1. The mean score for ‘happy’ is lower than in
study 1 by 5.13 percentage points, t(349) = 2.88, p <.01; ‘content’ is 4.20
percentage points lower, t(349) = 2.31, p <.05 and ‘alert’ is 5.47 percentage points
lower than in study 1, t(349) = 2.74, p <.01. Also, the mean score for ‘active’ is
lower by 6.71 percentage points, t(349) = 2.96, p <.01.
Finally, a comparison of correlations between LS, SWB and affects between
studies 1 and 2 for people with >50%SM indicates increased correlations between
‘LS’ and ‘happy’ (by r = .18, z = 3.14, p < .01) and ‘LS’ and ‘unhappy’ (by r
= .17, z = 2.17, p<.05). There is also a decrease in the correlation between ‘LS’
and ‘alert’ (by r = .20, z = 2.32, p<.05). No significant differences in correlations
from study 1 to study 2 were observed between SWB and any of the affects.
These results will be discussed.
The results of the regression analyses are presented in the Tables 42 and 43.
Table 42:
Predicting LS by nine affect adjectives (n = 191 >50%SM)
Variable
DV: Life
Satisfaction
.77
B
.63***
β
.60
sr2
.08
1.
Happy
2.
Content
.67
.06
.05
.00
3.
Unhappy
-.61
-.05
-.06
.00
4.
Discontent
-.49
-.11
-.13
.01
5.
Active
.37
.00
.00
.00
6.
Alert
.35
.03
.02
.00
7.
Excited
.41
.00
.00
.00
8.
Sleepy
-.27
-.05
-.07
.00
-.01
.00
R = .62
Adjusted R2 = .60
9. Quiet
-.26
* p<.05; ** p<.01; *** p<.001
-.01
Unique Variance = .09
Shared Variance = .51
2
141
Table 43:
Predicting SWB using nine affective adjectives (n = 191 >50%SM)
Variable
DV:SWB
B
β
sr2
1. Happy
.66
.23**
.35
.03
2. Content
.64
.13*
.19
.01
3. Unhappy
-.46
.04
.07
.00
4. Discontent
-.37
-.09
-.16
.01
5. Active
.42
.01
.02
.00
6. Alert
.46
.10*
.16
.01
7. Excited
.43
.07
.10
.01
8. Sleepy
-.30
-.06*
-.12
.01
9. Quiet
-.12
.06*
.12
.01
* p<.05; ** p<.01; *** p<.001
Unique Variance = .09
Shared Variance = .41
R2 = .52
Adjusted R2 = .50
Following the removal of all cases in the negative range for SWB, core affect
explains the same amount of unique variance in LS and SWB (9%). Further, this
is the same amount of unique variance explained using the whole sample; and 2%
less unique variance than the 11% found in study 1. This finding is also
inconsistent with the third hypothesis which states that core affect will account for
more unique variance in LS than SWB because LS is more abstract and core
affect specifically concerns the abstract component of SWB. These results will be
discussed.
Finally, in contrast to study 1, in the regression involving SWB as the DV,
adjectives ‘sleepy’ and ‘quiet’ made a significant unique contribution; despite the
mean score for these variables being no different from those found in study 1.
As in study 1, additional regression analyses were carried out to determine more
precisely which affects are the strongest and most reliable predictors of LS and
SWB.
142
5.6.1
Further Analyses Using Core Affect Adjectives as Predictors of LS and
SWB
These additional analyses were conducted using the whole sample to ensure that
the predictors chosen to represent core affect are the strongest and most reliable.
In these analyses, the top four predictors from the standard multiple regressions
(as determined by the magnitude of β weights), were subject to hierarchical
multiple regression analyses. For the prediction of LS, these included happy (β
= .64), discontent (β = -.13), content (β = .08) and unhappy (β = -.06) (see Table
39) and for the prediction of SWB, these included content (β = .23), happy (β
= .21), discontent (β = -.13) and alert (β = .12) (see Table 40).
The hierarchical regression analysis for LS, involving the top four predictors
happy, discontent, content and unhappy, is presented in Table 44.
Table 44:
Predicting life satisfaction using four affect adjectives
R2
Adj. R2
∆R2
B
SE B

sr²
Step 1
Happy
.67
.84***
.04
.82
.66
.74***
.05
.72
.38
-.17***
.04
-.18
.02
.66
Step 2
Happy
Discontent
.69
.69
.02***
* p<.05; ** p<.01; *** p<.001
The R for the final model of this regression was significantly different from zero,
F(2, 202) = 225.251, p < .001. Altogether, 69% of the variability in LS was
predicted from scores on the adjectives happy and discontent. In the final model,
the adjectives ‘content’ and ‘unhappy’ failed to contribute unique, significant
variance thus, output for this variable was not provided by SPSS. This result will
be discussed.
143
The regression analysis for SWB, involving the top four predictor’s content,
happy, discontent and alert, is presented in Table 45.
Table 45:
Predicting SWB using four affect adjectives
B
SE B

Sr²
.53***
.04
.71
.50
Content
.45***
.04
.61
.28
Alert
.16***
.04
.21
.03
Content
.25***
.07
.34
.03
Alert
.16***
.04
.20
.03
Happy
.23***
.07
.32
.03
R2
Adj. R2
∆R2
Step 1
Content
.51
.51
Step 2
.54
.54
.03***
Step 3
.57
.57
.03***
* p<.05; ** p<.01; *** p<.001
The R for the final model of this regression was significantly different from zero,
F(3, 201) = 88.494, p < .001. Altogether, 57% of the variability in SWB was
predicted and in the final model, the adjective ‘discontent’ failed to contribute
unique, significant variance; thus, output for this variable was not provided by
SPSS.
Consistent with data from study 1, content, alert and happy were significant
independent predictors of SWB. However, in the final regression for this same
analysis in study 1, happy was the greatest single predictor (sr2 = .10), followed
by content (sr2 = .02) and alert (sr2 = .02). This contrasts with the current
regression, where all three adjectives contribute the same amount of unique
variance (sr2 = .03).
144
In summary, the mean scores for LS and SWB are within the normal ranges and
do not significantly differ from their levels in study 1. For the prediction of LS,
‘happy’ and ‘discontent’ made a unique contribution, together explaining 69% of
variance. This is different from study 1 where adjectives ‘happy’, ‘unhappy’,
‘content’ and ‘alert’ explained 59% variance. However, consistent with study 1,
for the prediction of SWB, adjectives ‘happy’, ‘content’ and ‘alert’ explained 57%
variance - 2% less variance (ns) than in study 1.
In terms of the unique contribution of these adjectives to LS, ‘happy’ and
‘discontent’ both contributed unique, significant variance. Interestingly, the
correlation between LS and ‘happy’ was significantly greater in study 2 when
compared to study 1. It is possible that ‘happy’ consumed a majority of variance
that may otherwise be attributed to ‘content’ and ‘alert’. In terms of the prediction
of SWB, consistent with study 1, ‘content’, ‘alert’ and ‘happy’ together explained
57% of the variability in SWB. Finally, as predicted, across the entire data set,
core affect was found to explain more unique variance in LS than SWB (9%
versus 6%). However, in contrast to theory, when cases in the negative range for
SWB were removed (SWB < 50), core affect now explained the same amount of
unique variance in both measures (9%) This is 3% less unique variance than the
11% contributed by core affect to the prediction of LS in this same analysis in
study 1. Interestingly, across the entire data set, mean scores on key affects
‘happy’, ‘content’ and ‘alert’ all significantly reduced from study 1, as did mean
scores on adjectives ‘excited’ and ‘active’. Furthermore, in the group with SWB >
50%SM, mean scores for ‘happy’ and ‘content’ were also significantly lower
when compared to study 1. It appears that lower scores on these key affects may
be responsible for the failure of core affect to explain more unique variance in LS
than SWB.
5.7
Predicting LS and SWB above Core Affect using the Buffer Variables
The aim of the following analyses is to examine the relative contribution of core
affect and the buffer variables to SWB. This analysis is also conducted as part of a
series of analyses examining the fifth hypothesis which states that core affect is
driving the relationship between SWB and related constructs.
145
5.7.1
Predicting LS using Core Affect and the Buffer Variables
Table 46 displays means, standard deviations, simple correlations and partial
correlations between variables.
Table 46:
Means, standard deviations and correlations between variables (N=205)
Variable
1. LS
Mean
72.15
SD
20.27
1.
-
2.
2. Core Affect
67.64
16.29
.77
-
3. Optimism
62.30
19.60
3.
4.
5.
.56
.56
(.24)
4. Primary Control
65.79
16.54
.41
.53
.46
(.01)
(.23)
5. Secondary Control
65.19
17.72
.40
.50
.56
.50
(.03)
(.39) (.33)
6. Self-esteem
63.72
18.10
.63
.62
.57
.36
.47
(.30)
(.34) (.06) (.23)
* (.xx) partial correlations using core affect as a covariate are presented in brackets
When core affect is entered as a covariate, the correlations between the buffer
variables and LS reduce in a fashion similar to that found in study 1. In fact, the
average correlation between the buffer variables and LS decreases from r = .50
to .15 (an average reduction of .35). In comparison, this same average reduction
was .38 in study 1 (from r = .49 to .11). These data are again consistent with the
hypothesis that core affect is driving the relationship between SWB and related
variables. The most striking difference between the two samples is that the partial
correlation between LS and optimism is greater in the present sample (.24) when
compared to study 1 (.00).
The result of the hierarchical regression analysis examining the influence of the
buffer variables on LS above core affect is presented in Table 47.
6.
-
146
Table 47:
Predicting LS using core affect and the buffer variables
B
SE B

sr²
.95***
.06
.77
.58
Core Affect
.74***
.08
.59
.17
Optimism
.15*
.06
.15
.01
R2
Adj. R2
∆R2
Step 1
Core Affect
.59
.58***
Step 2
Primary Control
-.01
.07
-.01
.00
Secondary Control
-.08
.06
-.07
.00
.07
.22
.03
Self Esteem
.24***
.64
.63
.05***
*p<.05; ** p<.01; *** p<.001
When entered at step 2, the buffer variables of self-esteem and optimism explain a
further 5% of the variance in LS, ∆R2 = .05, Finc(4, 199) = 6.823, p<.001. This
supports Davern et al’s., (2007) contention that LS is primarily an affective
construct with very minor independent contribution from self-esteem (sr² =.03,
p<.001) and optimism (sr² =.01, p<.05) in the presence of core affect. Consistent
with study 1, the unique variance (sr²) contributed by core affect at step 2
markedly reduced (a reduction of .41 compared with .37 for study 1), indicating a
high degree of shared variance between core affect and the buffer variables.
However, what is most interesting here is that optimism made a significant
contribution. In study 1, only self-esteem made a significant contribution above
core affect.
5.7.2
Predicting SWB using Core Affect and the Buffer Variables
The following analysis examines whether the cognitive buffers will contribute
significant unique variance to the prediction of SWB beyond core affect. Table 48
displays means, standard deviations, simple correlations and partial correlations
between variables.
147
Table 48:
Means, standard deviations and correlations between variables (n = 205)
Variable
1. SWB
2.
Core Affect
3.
Optimism
Mean
SD
73.61
14.18
67.64
16.29
1.
-
2.
.75
-
3.
4.
5.
.56
.56
(.24)
4. Primary Control
.36
.53
.46
65.79
16.54
(.06)
(.23)
5. Secondary Control
.43
.50
.56
.50
65.19
17.72
(.11)
(.39) (.33)
6. Self-esteem
.66
.62
.57
.36
.47
63.72
18.10
(.37)
(.34) (.06) (.23)
* (.xx) partial correlations using core affect as a covariate presented in brackets
62.31
6.
19.63
-
Moderate correlations between SWB and the buffer variables can be seen in Table
48, with the largest correlations between SWB and self-esteem (r = .66) and
between SWB and optimism (r = .56). However, when variance attributed to core
affect is removed, correlations between SWB and the buffer variables decrease
considerably. In fact, this average correlation reduces from r =.50 to .20. The
largest partial correlation is between SWB and self-esteem (r = .37). This is in
contrast to study 1 where the greatest observed partial correlation was between
SWB and optimism (r = .32). Interestingly, in study 1, the partial correlation
between SWB and self-esteem was the lowest of all the buffer variables (r = .21).
The result of the regression analysis, with core affect entered at step 1 and the
buffer variables at step 2, is presented in Table 49.
148
Table 49:
Predicting SWB using core Affect and the buffer variables
B
SE B

sr²
.65***
.04
.75
.56
Core Affect
.49***
.05
.56
.15
Optimism
.09*
.04
.13
.01
-.08
.05
-.09
.00
Secondary Control
.01
.05
.01
.00
Self Esteem
.21***
.05
.27
.04
R2
Adj. R2
∆R2
Step 1
Core Affect
.56
.56
Step 2
Primary Control
.64
.63
.07***
* p<.05; ** p<.01; *** p<.001
When entered at step 2, the buffer variables explain a further 7% of variance in
SWB, ∆R2 = .07, Finc(4, 199) = 9.763, p<.0001. Similar to the prediction of LS,
self-esteem (sr² = .04, p<.001) and optimism (sr² = .01, p<.05) make significant
unique contributions. The finding that self-esteem contributes unique variance is
in contrast to study 1 where optimism was the only buffer variable to make a
significant unique contribution. Finally, it is important to note that at step 2,
primary control contributed negative standardised and unstandardised regression
coefficients. However, these are non-significant. These results will be discussed.
For reasons outlined in study 1, that SWB is a more comprehensive and
informative measure of wellbeing than LS, SWB will be employed as the major
independent variable of interest for the following analyses.
5.8
Predicting SWB Using MDT
As in study 1, the aim of the next regression analyses is to explore how much
variance in SWB can be explained by MDT above core affect. Means, standard
deviations, simple correlation and partial correlations are presented in Table 50.
149
Table 50:
Means, standard deviations and correlations between variables (N=205)
Variable
1. SWB
2.
Core Affect
3.
MDT
Mean
SD
73.61
14.18
67.64
16.29
1.
2.
3.
.75
-
.52
.59
(.14)
* (.xx) partial correlations using core affect as a covariate presented in brackets
61.19
15.23
As shown in Table 50, the correlation between MDT and SWB is moderate at .52.
Interestingly, this same correlation in study 1 (r = .65) is .13 greater than that
observed in the present study. However, this difference is not significant.
Furthermore, as found in study 1, when core affect was entered as a covariate, this
correlation reduced considerably (from r = .52 to r = .14, p<.01). This reduction is
greater than that observed in study 1 (e.g., from r = .65 to r = .32, p<.01),
indicating that core affect may be driving the relationship between these
constructs to a greater extent than in study 1. A significant difference in partial
correlations (r = .14 vs .32, z = -1.75, p<.05) supports this.
The results of the hierarchical regression analysis with core affect entered at step
1 and MDT entered at step 2 are presented in
Table 51.
Table 51:
Predicting SWB with MDT
150
SE B

sr²
.65***
.04
.75
.56
Core Affect
.59***
.05
.68
.30
MDT
.11*
.05
.12
.01
R2
Adj. R2
∆R2
B
Step 1
Core Affect
.56
.56
Step 2
.57
.57
.01*
* p<.05; ** p<.01; *** p<.001
When entered at step 2, MDT explained a further 1% variance, ∆R2 = .05, Finc(1,
202) = 4.107, p<.05. This is 3% less variance than the 4% variance contributed by
MDT in this same analysis in study 1. Interestingly, standardised and
unstandardised regression coefficients for MDT are considerably lower when
compared to study 1. Further, and unique variance (sr²) contributed by core affect
at step 2 is 11% greater than that observed in study 1. Taken together, these
results suggest that in the present sample, core affect appears the more dominant
influence over SWB, with MDT playing a significant, but very subsidiary role.
To investigate the influence of individual discrepancy judgements on SWB, an
additional analysis was conducted using all 7 items comprising MDT. Means,
standard deviations, simple correlation and partial correlations are presented in
Table 52.
Table 52:
Means, standard deviations and correlations between variables (N=205)
Variable
1. SWB
Mean
73.61
SD
14.18
1.
-
2.
2. Core Affect
67.64
16.29
.75
-
3. Self-wants
64.71
22.23
.65
4. Self-other
62.29
20.77
5. Self-deserves
58.19
19.48
.60
(.22)
.53
(.25)
.26
(.03)
.52
.33
3.
4.
5.
.70
(.56)
.46
.64
(.35) (.58)
-
6.
7.
8.
9.
151
6. Self-needs
55.17
7. Self-progress
65.66
8. Self-future
68.12
9. Self-best
54.20
19.49
.43
(.12)
24.10 .19
(.00)
21.11 .35
(.17)
24.75 .22
(-.09)
.48
.25
.33
.37
.52
(.31)
.38
(.29)
.26
(.06)
.43
(.27)
.59
(.46)
.41
(.33)
.28
(.13)
.38
(.24)
.59
(.52)
.33
.23
(.27) (.13)
.19
.36
.22
(.09) (.24) (.15)
.38
.35
.64
.25
(.29) (.22) (.60) (.15)
* (.xx) partial correlations using core affect as a covariate presented in brackets
This shows low to moderate correlations between discrepancy judgments and
SWB, with SWB sharing the greatest relationship with the variables: ‘Is your life
as good as you want it to be’ (self-wants, r = .60) and ‘how does your life
compare to the average for most people your own age’ (self-other, r = .53).
However, when core affect was entered as a covariate, these correlations reduced
considerably. The highest mean scores were observed on the variables: ‘this time
next year, will your life be better or worse than it is now’ (self-future, M = 68.12;
SD = 21.11) and ‘compared to one year ago, is your life better or worse’ (selfprogress, M = 65.66, SD = 24.10). Relatively low discrepancies on these variables
reflect higher levels of perceived satisfaction with these particular aspects of life.
The results of the hierarchical regression analysis with core affect entered at step
1 and MDT entered at step 2 are presented in Table 53.
Table 53:
Predicting SWB with MDT
B
SE B

sr²
.65***
.04
.75
.56
Core Affect
.51***
.05
.59
.18
Self wants
.09
.05
.14
.01
Self-others
.14**
.05
.21
.01
.05
-.12
.01
R2
Adj. R2
∆R2
Step 1
Core Affect
.56
.56
Step 2
Self-deserves
-.08
-
152
Self-needs
.02
.05
.03
.00
Self-progress
.00
.04
.00
.00
Self-future
.07*
.03
.11
.01
-.07*
.04
-.13
.01
Self-best
.63
.61
.06***
Unique variance = .23
Shared Variance = .38
* p<.05; ** p<.01; *** p<.001
When entered at step 2, three discrepancy judgments explained a further 6%
variance, ∆R2 = .06, Finc(7, 196) = 4.824, p<.001; and this is 5% more variance
than that explained by the composite variable ‘MDT’ in the above analysis. The
three contributing discrepancy judgments were: ‘how does your life compare to
the average for most people your own age’ (sr² = .01); ‘this time next year, will
your life be better or worse than it is now’ (sr² = .01); and ‘how does your life
right now compare to the best you have had’ (sr² = .01). Finally, the unique
variance contributed by core affect at step 2 is 11% greater than that observed in
study 1. Taken together, these results suggest discrepancy judgments play a
significant, but very subsidiary role in explaining SWB. The results also indicate
that when all 7 discrepancy judgments are included, these explain considerably
more variance than when these variables are combined into a composite.
A further regression examines the extent to which all of the buffers and MDT are
explaining different aspects of the remaining variance in SWB above core affect.
Here, the buffers are entered at step 2 and MDT at step 3. Table 54 displays
means, standard deviations, simple correlations and partial correlations between
variables.
Table 54:
Means, standard deviations and correlations between variables (N = 205)
Variable
Mean
SD
1.
1.
SWB
73.61
14.18
-
2.
Core Affect
67.64
16.29
.75
3.
Optimism
62.31
19.63
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
-
.50
(.33)
.36
(.06)
.45
(.17)
.37
(.14)
.30
(.15)
.29
(.05)
.25
.14)
.23
(.08)
.30
.14)
.47
(.23)
.45
(.19)
.44
(.25)
.24
(.09)
.25
(.02)
.28
(.18)
.27
(.13)
.26
(.09)
.64
(.39)
.62
(.44)
.29
(.12)
.38
(.13)
.27
(.15)
.31
(.14)
.29
.08)
.70
(.56)
.46
(.35)
.52
(.31)
.38
(.29)
.26
(.06)
.43
(.27)
.64
(.58)
.59
(.46)
.41
(.33)
.28
(.13)
.38
(.24)
.59
(.52)
.33
(.27)
.19
(.09)
.38
(.29)
.23
(.13)
.36
(.24)
.35
(.22)
.22
(.15)
.64
(.60)
.25
(.15)
153
.56
.56
(.24)
4. Primary Control
65.79
16.54
.36
.53
.46
(-.06)
(.23)
5. Secondary Control
65.19
17.72
.43
.50
.56
(.11)
(.39)
6.
Self-esteem
63.72
18.10
.66
.62
.57
(.37)
(.34)
7.
Self-wants
64.71
22.23
.60
.65
.52
(.22)
(.24)
8.
Self-other
62.29
20.77
.53
.52
.51
(.25)
(.30)
9.
Self-deserves
58.19
19.48
.26
.33
.26
(.03)
(.10)
10. Self-needs
55.17
19.49
.43
.48
.38
(.12)
(.15)
11. Self-progress
65.66
24.10
.19
.25
.24
(.00)
(.12)
12. Self-future
68.12
21.11
.35
.33
.36
(.17)
(.23)
13. Self-best
54.20
24.75
.22
.37
.30
(-.09)
(.13)
* (.xx) partial correlations using core affect as a covariate presented in brackets
154
As can be seen, a low to moderate relationship exists between the buffer variables
and discrepancy judgments (.23 to .64). However, as found in study 1, partial
correlations reveal that the relationship between these constructs reduces
considerably when the influence of core affect is removed. The results of the
regression analysis, with core affect entered at step 1, the buffer variables at step 2
and MDT entered at step 3, are presented in Table 55.
Table 55:
Predicting SWB with core affect, the buffer variables and MDT
R2
Adj. R2
∆R2
B
SE B

sr²
Step 1
Core Affect
.65***
.04
.75
.56
Core Affect
.49***
.05
.56
.15
Optimism
.09*
.04
.13
.01
Primary Control
.-08
.05
-.09
.00
Secondary Control
.01
.05
.01
.00
Self Esteem
.21***
.05
.27
.04
Core Affect
.46***
.06
.53
.34
Optimism
.07
.04
.10
.00
-.07
.05
-.08
.00
Secondary Control
.00
.05
.00
.00
Self-esteem
.16**
.05
.20
.02
Self-wants
.06
.05
.09
.00
Self-other
.07
.05
.10
.00
-.05
.04
-.07
.00
Self-needs
.03
.04
.04
.00
Self-progress
.00
.03
.00
.00
Self-future
.05
.03
.08
.00
-.07*
.03
-.12
.01
.56
.56
Step 2
.64
.63
.07***
Step 3
Primary Control
Self-deserves
Self-best
.66
* p<.05; ** p<.01; *** p<.001
.64
.01
155
When entered at step 3, MDT accounted for no additional significant variance,
despite the item ‘how does your life right now compare to the best you have had’
(self-best) contributing a significant regression coefficient. This is in contrast to
findings from study 1 where MDT explained a further 2.0% variance above core
affect and the buffer variables. Thus, in the present sample, the remaining portion
of variance in SWB can be explained by self-esteem and optimism, with no
independent contribution from MDT.
5.9
Predicting SWB with Personality
In study 1, neither extraversion nor emotional stability accounted for additional
variance in SWB above core affect. The aim of the next regression analyses was
to repeat this analysis with the new sample.
Means, standard deviations, simple correlations and partial correlations between
variables are presented in Table 56.
Table 56:
Means, standard deviations and correlations between variables
Variable
Mean
SD
1.
1.
SWB
73.61
14.18
-
2.
Core Affect
67.64
16.29
.75
3.
Extraversion
61.96
22.58
2.
3.
4.
-
.41
.50
(.06)
4. Stability
62.00
19.51
.48
.47
.26
(.23)
(.03)
* (.xx) partial correlations using core affect as a covariate presented in brackets
-
Examination of simple correlations reveals that a moderate relationship exists
between extraversion and SWB and between emotional stability and SWB.
However, as found in study 1, core affect as a covariate weakens these
relationships. The correlation between SWB and extraversion reduces from .41
to .06 and the correlation between SWB and emotional stability reduces from .48
156
to .23, indicating that the relationship between these constructs may be driven, at
least in part, by core affect. Interestingly, a correlation of .23 between SWB and
emotional stability remains, indicating a small, yet significant degree of
independence. The results of the regression analysis, with core affect entered at
step 1 and personality at step 2, are presented in Table 57.
Table 57:
Predicting SWB using the personality dimensions of extraversion and emotional
stability
B
SE B

sr²
.65***
.04
.75
.56
Core Affect
.57***
.05
.65
.27
Extraversion
.03
.03
.04
.00
Stability
.12**
.04
.17
.02
R2
Adj. R2
∆R2
Step 1
Core Affect
.56
.56
Step 2
.58
.58
.02**
* p<.05; ** p<.01; *** p<.001
When entered at step 2, emotional stability accounts for an additional 2% variance,
∆R2 = .02, Finc(1, 201) = 5.803, p<.01. This finding contrasts results from study 1
where neither extraversion nor stability made a significant unique contribution at
step 2. This result will be discussed.
In summary, consistent with findings from study 1, these analyses indicate that LS
and SWB are primarily affective constructs, with minor independent contribution
from self-esteem. Emotional stability also made a unique contribution. Finally,
within study 1 and study 2, both self-esteem and optimism made significant
unique contributions to either LS or SWB, but in different ways. In study 1, selfesteem was the only buffer variable to contribute unique variance in LS above
core affect; whilst optimism alone contributed uniquely to the prediction of SWB.
157
Other contrasts between the two sets of results are also evident. While in study 1,
MDT made an independent contribution, in study 2 it did not. Similarly, in study
1, emotional stability explained a further 2% variance in SWB above core affect
whereas in study 2 it did not.
Overall, these results support the hypothesis that core affect is driving the
relationship between SWB and related constructs. However, what has also been
observed is a varying pattern of significant, but very subsidiary contributions from
other variables. Whether these contributions represent random effects or result
from systematic error variance remains undetermined.
5.10 Predicting SWB Using Normative Data Divisions: Further Analyses
As in study 1, cases were split according to level of SWB and separate analyses
were performed on these groups. To test the fourth hypothesis that that in
circumstances of homeostatic challenge, but not in homeostatic rest, primary
control and secondary control will explain variance in SWB beyond core affect,
regression analyses were conducted with core affect entered at step 1 and the
cognitive buffer variables at step 2.
5.10.1 Predicting SWB using cases between 45 and 69
To test the fourth hypothesis that primary control and secondary control will
explain significant variance in SWB above core affect in a group of people
experiencing homeostatic challenge, the sample was divided into two groups:
people with SWB scores equal to and above 70%SM (n= 133, X = 82.16, SD =
6.89) and people with SWB scores between 45 and 69%SM (n= 64, X = 60.36,
SD = 6.76). As discussed in study 1, splitting cases in this manner decreases
sample variance - as indicated by the reduction in the magnitude of the standard
deviations. Results will be interpreted with care as there are only 64 cases in the
45-69 group.
Means, standard deviations, correlations and partial correlations between all major
variables for people with SWB in the greater than 70 group and in the 45-69
group are presented in Table 58 and Table 59.
158
Table 58:
Means, standard deviations and correlations between variables (SWB > 70%SM)
(n = 133)
Variable
1. SWB
Mean
82.16
SD
6.89
1.
-
2.
2. Core Affect
75.19
11.43
.60
-
3. Optimism
69.00
17.51
3.
4.
5.
6.
.34
.41
(.13)
4. Primary Control
69.75 16.02 .11
.34
.33
(-.13)
(.23)
5. Secondary Control
70.20 16.02 .27
.33
.44
.32
(.10)
(.36)
(.24)
6. Self-esteem
71.85 14.56 .42
.45
.39
.25
.39
(.20)
(.25)
(.11)
(.28)
* (.xx) partial correlations using core affect as a covariate presented in brackets
-
Table 59:
Means, standard deviations and correlations between variables (SWB Between 45
& 69%SM) (n=64)
Variable
1. SWB
Mean
SD
60.36
6.76 -
2. Core Affect
55.36
13.34
50.78
17.37
3. Optimism
4. Primary Control
5. Secondary Control
6. Self-esteem
59.32
55.26
49.13
1.
.34
.40
(.33)
.28
14.89
(.11)
.40
15.96
(.31)
.30
14.35
(.20)
2.
3.
4.
5.
6.
.28
.56
.41
.34
.39
(.30)
.52
(.46)
.49
(.43)
.66
(.57)
.22
(.04)
.26
(.15)
-
* (.xx) partial correlations using core affect as a covariate presented in brackets
As expected and similarly to that found in study 1, the mean score across the
buffer variables was higher for people with SWB > 70%SM (mean for buffer
variables = 70.20, SD = 16.02 versus 53.62, SD = 15.64 in the < 70group). Also,
159
as expected, it can be seen that correlations between the buffer variables and SWB
were generally higher for people with SWB < 70%SM (.35 vs .29).
Interestingly, in the 45-69 group, the mean score for core affect is 5.29 percentage
points lower (p<.05) when compared to study 1 and mean scores across the buffer
variables are also generally lower when compared to study 1. In fact, the mean
score for self-esteem is 9.07 points lower in the present sample (p<.01). Also, as
found in study 1, correlations between the buffer variables and SWB generally
decreased when core affect was entered into a partial correlation matrix as a
covariate. Moreover, as predicted, this effect was less pronounced for people in
the 45-69 group (average correlation reduced by .05 vs .11 for >70 group). It was
expected that these correlations would decrease more in the >70 group because it
is most likely that the cognitive buffers are resting and core affect is the driver of
SWB.
Regression analyses involving the split of cases > 70%SM and between 45 and
69%SM, are presented in Table 60.
Table 60:
Predicting SWB after core affect using the buffer variables (split cases)
SWB > 70%SM
2
R
2
Adj. R
SWB between 45 & 69%SM
∆R
2
B
SE B

sr²
.36***
.04
.53
.35
R2
Adj. R2
∆R2
B
SE B

sr²
.17***
.06
.34
.12
Step 1
Core Affect
.36
.35
.12
.10
Step 2
Core Affect
.32***
.05
.53
.25
.11
.07
.21
.04
Optimism
.04
.03
.09
.01
.08
.06
.21
.03
-.07*
.03
-.16
.04
-.05
.08
-.11
.01
Secondary Control
.02
.03
.05
.00
.11
.07
.26
.04
Self Esteem
.08*
.04
.16
.03
.04
.06
.08
.01
Primary Control
.41
* p<.05; ** p<.01; *** p<.001
.39
.05*
.25
.19
.13*
160
161
In the group with SWB > 70%SM, core affect explained 35% of the variance, F (1,
132) = 73.287, p<.0001 at step 1. This is 8% more variance than the 27%
contributed by core affect in study 1. In the 45 – 69 group, core affect explained
10% of variance, F(1, 62) = 8.121, p<.01. This is 4% less variance than that
contributed by core affect in study 1. Consistent with SWB homeostasis, core
affect explained more variance at step 1 in the > 70 group, although this effect
was not observed in study 1. Interestingly, and in contrast to the fifth hypothesis,
although the overall amount of explained variance from the buffers in the 45-69
group was significant, ∆R2 = .13, Finc(4, 58) = 3.926, p<.05, no variable
contributed unique significant variance. However, it is likely that this is due to
sample size. Nonetheless, this finding is also in contrast to findings from study 1
where primary control and secondary control were significant unique contributors.
In the > 70 group, primary control and self-esteem contributed unique variance at
step 2, suggesting that in this group, primary control plays a subsidiary role to
core affect. Finally, it is particularly noteworthy that for both groups, the
correlation between primary control and SWB was negative, as were the
regression coefficients for this variable. It is suspect that this may be occurring
because core affect is acting as a suppressor variable.
In summary, the results of these analyses are not consistent with hypothesis four.
In total, the buffer variables explained a further 9% variance (adjusted) in the 4569 group and, although this was significant, no buffer variable contributed
significant unique variance. However, it is likely that this non-significance is due
to small sample size resulting from the splitting procedure. To determine whether
or not this may be the case, data from studies 1 and 2 were combined and analyses
conducted again.
To reiterate, according to Tabachnick and Fidell (2001), the criterion for multiple
regression analysis is:
N > 50 + 8m
where
N= minimum number of cases and
m = number of IV’s
162
With 5 five IV’s in this particular analyses, a minimum sample size of 90 cases is
required. Thus, with a sample sizes of 41 (study 1) and 64 (study 2), there is a
need to combine the data.
Means, standard deviations, correlations and partial correlations for variables in
the combined analyses are presented in Table 61.
Table 61:
Means, standard deviations and correlations between variables (SWB Between 45
& 69%SM) (n=105)
Variable
1. SWB
2. Core Affect
3. Optimism
4. Primary Control
5. Secondary Control
6. Self-esteem
Mean
SD
60.64
6.61 -
57.43
13.68
51.90
17.10
60.78
56.54
52.67
1.
.37
.29
(.19)
.31
15.14
(.16)
.43
17.76
(.31)
.23
15.88
(.10)
2.
3.
4.
5.
6.
.32
.48
.45
.37
.43
(.34)
.49
(.41)
.45
(.37)
.50
(.36)
.34
(.20)
.26
(.12)
-
* (.xx) partial correlations using core affect as a covariate presented in brackets
As shown in Table 61, the greatest observed partial correlation is between
secondary control and SWB (r = .31). Thus, secondary control shares the greatest
relationship with SWB, independent of core affect.
Regression analyses involving combined data from studies 1 and 2 for cases
between 45 and 69%SM, are presented in Table 62.
163
Table 62:
Predicting SWB using core affect and the buffer variables (n=105)
B
SE B

sr²
.18***
.04
.37
.13
Core Affect
.09
.05
.19
.03
Optimism
.02
.04
.05
.00
Primary Control
.02
.05
.04
.00
Secondary Control
.11*
.04
.29
.05
Self Esteem
.02
.04
.04
.00
R2
Adj. R2
∆R2
Step 1
Core Affect
.14
.13
Step 2
.23
.19
.09***
In the 45 – 69 group, core affect explained 13% of variance, F(1, 103) = 16.158,
p<.01, which is 3% less variance than that contributed by core affect in study 1
and 1% more variance contributed in study 2. In partial support of the fourth
hypothesis, secondary control contributed unique variance at step 2 (sr² = .05, p
= .001). However, the most striking result from this analysis is that at step 2, core
affect no longer significantly predicts SWB. This finding is consistent with SWB
homeostasis theory, which states that core affect will lose its affiliation with SWB
when challenged. Further, when this occurs, cognition will assume greater control.
These results will be discussed.
5.11 Model Testing: Evaluating Model Fit
Regression analyses have so far demonstrated core affects’ dominance over
personality and MDT in explaining significant variance in SWB. The purpose of
the following structural equation models is to explore which theoretical model fits
the data best. Using AMOS and maximum likelihood estimation, the models
being examined are the same as Davern, et al., (2007). These include:
1.
An Affective-Cognitive model
164
2.
A Personality-driven model
3.
A Multiple Discrepancies model
An example of a simplified Affective-Cognitive model of SWB to be tested is
shown in Figure 9.
Personality
Core affect
SWB
MDT
Figure 9:
A simplified Affective-Cognitive model of SWB
As shown in Figure 9, core affect is proposed to have a direct influence on SWB,
personality and MDT. Direct paths are also specified between personality and
SWB and between MDT and SWB to determine the influence of these constructs
on SWB in the presence of core affect.
The results of the first structural equation model are presented in the following
section.
5.11.1 An Affective-Cognitive Model
Similar to Davern, et al., (2007), the following analysis examines how well an
Affective-Cognitive model of SWB fit the data. In the final model, co-variances
were fitted between error terms and these are theoretically justified. For example,
error terms were fitted between interrelated SWB domains with moderate to high
correlations. These included: standard of living and future (r = .48), standard of
165
living and health (r = .49), health and future (r = .50), health and achieving (r =
.55), achieving and future (r = .60), relationships and community (r = .52) and
community and future (r = .50). Co-variances were also fitted between interrelated
MDT error terms. These included: self-wants and self-deserves (r = .46), selfwants and self-others (r = .70), self-wants and self-needs (r = .52), self-others and
self-deserves (r = .64), self-others and self-needs (r = .59), self-deserves and selfneeds (r = .59) and self-progress and self-best (r = .64). It is important to note that
while it is clear that these constructs are related, they are not multi-collinear. For
example, the greatest correlation between interrelated domain satisfactions and
discrepancy judgments was observed between self-wants and self-others (rself-wants
↔ self-others
=0.70), whilst latent constructs core affect and SWB correlated at .88.
According to both Predhazur (1997) and Tabachnik and Fidell (2007),
multicollinearity occurs when two variables share a zero-order correlation greater
than 0.9. Thus, no variables in the model are collinear.
Although the χ2/df for the Affective-Cognitive model is less than 3.00 (χ2/df =
1.824), indicating an acceptable level of model fit, the value for χ2 is significant
(χ2 = 242.61, p = .000) and this suggests that the model's covariance structure is
significantly different from the observed covariance matrix. The other fit Indices
for this model are presented in 63.
Table 63:
Absolute and relative fit Indices for the Affective-Cognitive model of SWB
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
242.61
133
1.824
.000
356.61
.88
.064
.80
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root Mean
Square Error of Approximation; SMC = Squared Multiple Correlation.
An SMC of .80 indicates that 80% of the variance in SWB is explained by the
Affective-Cognitive model; an NFI of .88 is just below acceptability and indicates
a need to respecify the model; and an RMSEA of .064 is greater than .05,
suggesting that the model is not a particularly close fit in relation to the degrees of
166
freedom. However, an AIC value of 356.61 for the model is less than that of the
saturated model, suggesting that this is a relatively parsimonious model.
Despite explaining 80% of the variance in SWB, results suggest that the
Affective-Cognitive model does not fit the data particularly well. A path model
specified according to the affective-cognitive model is presented in Figure 10.
ES
E
.27
.25
.31
stand.
SOL living
.33
.12*
health
.52***
.03
.89
.54
.50***
.80
happy
.46
.82
.69***
Core Affect
content
achieving
SWB
rel/ships
.40
.26
safety
alert
.78***
.61
.16
.28
community
MDT
.52
fut security
.84
self-wants
.56
others
.24
deserves
.45
needs
.17
.21
progress
future
* p<.05; ** p<.01; *** p<.001
Note: E = extraversion; ES = emotional stability; MDT = multiple discrepancies theory;
SWB = subjective wellbeing
Figure 10: Affective-Cognitive model of SWB
As can be seen in Figure 10, all pathways leading from core affect are significant.
Core affect is a powerful predictor of SWB ( =. 69), MDT ( =. 78), emotional
stability ( =. 50) and extraversion ( =. 52). Interestingly, in the presence of core
.22
best
167
affect, neither MDT nor extraversion predicts SWB. However, emotional stability
contributes unique, significant variance ( =. 12). Table 64 provides a summary of
standardised regression coefficients and associated significance values for the
variables in this model.
Table 64:
Analysis of an Affective-Cognitive model of SWB (N=205)
Core Affect  SWB

.69
.000
Core Affect  MDT
.78
.000
Core Affect  Emotional Stability
.50
.000
Core Affect  Extraversion
.52
.000
MDT  SWB
.16
.123
Emotional Stability  SWB
.12
.041
Extraversion  SWB
.03
.680
Pathways
p
The results presented in Table 64 emphasise core affect as the most dominant
influence over SWB. Not only does core affect contribute a standardized
regression coefficient over four times that contributed by MDT, it significantly
predicts MDT and personality. This suggests that these constructs are highly
affective in nature. Consistent with this assumption, MDT has a non-significant
effect on SWB in the presence of core affect, suggesting that the relationship
MDT shares with SWB may be driven by the relationship MDT shares with core
affect.
Overall, according to model fit statistics, the Affective-Cognitive model does not
fit the data particularly well and can be improved.
5.11.2 A Personality Model for SWB
This next SEM examines how well a Personality-driven model of SWB fit the
data. The χ2/df is 2.53, indicating an acceptable level of model fit. However, the
value for χ2 is significant (χ2 = 334.29, p = .000) and this suggests that the models
168
covariance structure is significantly different from the observed covariance matrix.
The other fit indices for this model are presented in Table 65.
Table 65:
Absolute and relative fit iIndices for the Personality-Driven model of SWB
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
334.29
132
2.53
.000
450.29
.84
.09
.78
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
An SMC of .78 indicates that 78% of the variance in SWB is explained by the
Personality-driven model of SWB; an NFI of .84 is below acceptability and
indicates a need to respecify the model; and an RMSEA of .09 is greater than .05,
suggesting that the model is not a close fit in relation to the degrees of freedom.
Finally, an AIC value of 450.29 for the model is greater than that of the saturated
model, suggesting that the personality driven model lacks parsimony. The
comparison between these models is presented later.
The path model specified according to the Personality-driven model is presented
in Figure 11.
169
happy
content
.84
alert
.85
.25
.27
stand.
SOL living
.35
.44***
Core Affect
.29
.72***
E
health
.39***
.49
.00
ES
achieve
.78
.42
SWB
.10
rel/ships
.36
.31***
safety
.47***
.31
.25
.25*
community
MDT
.47
fut security
.68
self-wants
.67
others
.30
deserves
.37
needs
.20
.16
progress
future
.23
best
* p<.05; ** p<.01; *** p<.001
Note: E = extraversion; ES = emotional stability; MDT = multiple discrepancies theory;
SWB = subjective wellbeing.
Figure 11: SEM of a Personality-driven model of SWB
As can be seen in Figure 11, the pathways between emotional stability and SWB
( =. 10) and between extraversion and SWB ( =. 00) are not significant.
However, both emotional stability and extraversion predict core affect ( = .39
and .44 respectively) and MDT ( = .47 and .31 respectively). More importantly,
however, the path between core affect and SWB ( = .72) remains strong and
significant. Finally, a significant path between MDT and SWB ( = .25) suggests
a unique contribution from MDT. Taken together, these results indicate that core
affect is the most dominant influence over SWB, with MDT again playing a
significant, yet subsidiary role. Furthermore, when core affect and MDT mediate
170
the relationship between personality and SWB, the influence of personality on
SWB in negligible.
Table 66 provides a summary of standardised regression coefficients and
associated significance values for variables in this model.
Table 66:
Analysis of a Personality-driven model of SWB (N=205)
Emotional StabilitySWB

.10
.143
ExtraversionSWB
.00
.974
Emotional Stability Core Affect
.39
.000
ExtraversionCore Affect
.44
.000
Emotional Stability MDT
.47
.000
ExtraversionMDT
.31
.000
MDTSWB
.25
.014
Core AffectSWB
.72
.000
Pathways
p
Results presented in Table 66 emphasise core affect as the most dominant
influence over SWB in this model. In fact, core affect contributed a standardised
regression coefficient over seven times that contributed by personality and over 2
and a half times that contributed by MDT, further suggesting its dominance over
these constructs.
Overall, despite explaining 78% variance, according to model fit statistics, the
Personality-driven model is a poor fit to the data.
5.11.3 The Multiple Discrepancies Theory Model
Similar to the previous analyses, co-variances were included in this model. The
χ2/df for this model was 1.971 and indicates an acceptable level of model fit.
However, the value for the chi-square statistic is significant (χ2 = 262.11, p = .000)
and this suggests that the model's covariance structure is significantly different
171
from the observed covariance matrix. The other fit indices for this model are
presented in Table 67.
Table 67:
Absolute and relative fit indices for MDT model of SWB
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
262.11
133
1.971
.000
376.11
.87
.07
.80
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
Fit indices presented in Table 67 suggest an average fit of the model to the data.
An SMC of .80 indicates that 80% of the variance in SWB is explained by the
model specified according to MDT; an NFI of .87 is below acceptability and
indicates a need to respecify the model; and an RMSEA of .07 is above .05,
suggesting that the model is not a particularly close fit in relation to the degrees of
freedom. However, an AIC value of 376.11 for the model is less than that of the
saturated or best fitting model, suggesting that this is a relatively parsimonious
model. A path model specified according to MDT is presented in Figure 12.
172
ES
E
.21
.29
.31
stand.
SOL living
.03
.33
.12*
self-wants
.75
others
.60
health
.54
.45***
.54***
deserves
needs
progress
.80
.24
.39
achieve
.46
.20
MDT
SWB
rel/ships
.40
.18
.19
safety
.82***
.67
future
.65***
.28
.23
community
best
Core Affect
.51
fut security
.88
excited
.83
happy
.26
content
* p<.05; ** p<.01; *** p<.001
Note: E = extraversion; ES = emotional stability; MDT = multiple discrepancies theory;
SWB = subjective wellbeing
Figure 12: Multiple discrepancies theory model of SWB
As can be seen in Figure 12, not all pathways leading from MDT are significant.
While MDT predicts core affect ( =. 82), extraversion ( =. 45), and emotional
stability ( =. 54), MDT fails to significantly predict SWB when the relationship
between these variables is mediated by core affect and personality. Consistent
with the Affective-Cognitive model and the Personality-driven model of SWB,
the path between core affect and SWB remains strong and significant ( =. 65).
From these results, it is apparent yet again that core affect is the most dominant
influence on SWB, with emotional stability ( =. 12) playing a significant, yet
subsidiary role in this model.
173
Table 68 provides a summary of standardised regression coefficients and
associated significance values for variables in this model.
Table 68:
Analysis of a multiple discrepancies model of SWB (N=205)
MDT  SWB

.20
.121
MDT  Core Affect
.82
.000
MDT  Extraversion
.45
.000
MDT  Emotional Stability
.54
.000
Extraversion  SWB
.03
.560
Emotional Stability  SWB
.12
.049
Core Affect  SWB
.65
.000
Pathways
p
Results presented in Table 68 indicate that core affect contributed a standardized
regression coefficient over three times greater than that contributed by MDT. In
contrast, MDT did not predict SWB. Thus, data again emphasise core affect as the
most dominant influence over SWB.
In summary, despite explaining 80% variance, model fit statistics indicate that the
Personality-driven model of SWB does not provide a good fit to the data.
5.12 A Comparison of the SEM models
A summary of the fit statistics for each model are presented in Table 69. The
models have been ranked from the best fitting model to the worst fitting model.
Table 69:
Summary of absolute and relative fit indices for all models
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
Affect-Cog
242.61
133
1.824
.000
356.61
.88
.06
.80
MDT
262.11
133
1.971
.000
376.11
.87
.07
.80
Personality
334.29
132
2.53
.000
450.29
.84
.09
.78
Model
174
Statistically, fit statistics indicate that the Affective-Cognitive model shown in
Figure 10 is the preferred description of how these variables are interacting with
one another. This result is consistent with Davern, et al., (2007). For example,
although all the models had significant values for chi-square, that for the
Affective-Cognitive model was the lowest – indicating best model fit. The
Affective-Cognitive model also ranked first on all other fit statistics. Importantly,
the Affective-Cognitive model had the lowest AIC value, indicating that this is the
most parsimonious of all the models.
Turning now to the specific models of SWB, it was consistently demonstrated that
core affect was the strongest single predictor in all three models, contributing
standardised regression coefficients of between two-and-a-half and seven times
greater than any other construct. In further support of the dominance of core affect,
regression coefficients between personality and SWB (in the Personality-driven
model) and between MDT and SWB (in the MDT model) were non-significant
when core affect mediated these relationships. Thus, the relationship between
these variables and SWB appears dependent upon shared variance with core affect.
In summary, the Cognitive-Affective model shown in Figure 10 is the statistically
preferred description of how the variables relate to one another, explaining 80%
of the variance in SWB and this finding in consistent with Davern et al., (2007).
5.13 Exploratory Analyses: School Satisfaction as a Unique Construct
The aim of this exploratory analysis was to determine whether school satisfaction
would meet the criteria to be considered a unique construct. In study 1, a stepwise multiple regression analysis was conducted to determine whether school
satisfaction predicts LS above the seven domains that comprise the measure of
SWB, thus, qualifying as a new domain. In study 1, satisfaction with school failed
to meet this criterion. The following analysis repeats this investigation using the
new data.
Means, standard deviations, simple correlations and partial correlations between
variables are presented in Table 70.
175
Table 70:
Means, standard deviations and correlations between variables
Variable
1. Life Sat.
Mean
72.15
SD
20.27
2. Std Living
70.73
20.46
1.
-
2.
3.
4.
5.
6.
7.
8.
.54
(.33)
3. Health
70.29
21.80 .53 .49
(.23) (.34)
4. Achieving
66.63
20.22 .66 .45 .55
(.36) (.24) (.34)
5. Relationships
78.44
17.92 .49 .36 .39 .54
(.10) (.13) (.14) (.29)
6. Safety
78.88
20.80 .52 .31 .33 .41 .45
(.18) (.07) (.05) (.10) (.20)
7. Neighbourhood 81.22
16.45 .37 .30 .33 .43 .52 .42
(.07) (.13) (.13) (.23) (.37) (.24)
8. Future
69.07
19.72 .56 .48 .50 .60 .46 .44 .50
(17) (.28) (.26) (.35) (.17) (.14) (.33)
9. School Sat.
67.56
21.07 .44 .30 .21 .43 .41 .38 .31 .37
(.12) (.10) (.06) (.18) (.18) (.16) (.13) (.10)
* (.xx) partial correlations using core affect as a covariate presented in brackets
The mean score for school satisfaction is the second-lowest of the domain scores
at 67.56 (SD = 21.07) and is 11.41 points lower (p<.01) than this same score in
study 1. Results of the regression analysis, with all seven domains entered at step
1 and school satisfaction entered at step 2, are presented in Table 71.
9.
-
176
Table 71:
Predicting LS using 7 PWI domains and satisfaction with school
B
SE B

Sr²
Std Living
.20***
.06
.20
.03
Health
.10
.06
.10
.01
Achieving
.33***
.07
.33
.05
Relationships
.09
.07
.08
.00
Safety
.23***
.06
.23
.04
-.07
.07
-.06
.00
.11
.07
.11
.01
R2
Adj. R2
∆R2
Step 1
Neighborhood
Future
.58
.56
Unique Variance = .14
Shared Variance = .42
Step 2
Std Living
.19***
.06
.19
.03
Health
.11*
.06
.12
.01
Achieving
.31***
.07
.30
.04
Relationships
.07
.07
.06
.00
Safety
.21***
.06
.22
.03
-.07
.07
-.06
.00
Future
.10
.07
.10
.00
School Satisfaction
.10
.05
.10
.01
Neighborhood
.59
.57
.01
Unique Variance = .12
Shared Variance = .45
* p<.05; ** p<.01; *** p<.001
When entered at step 2, satisfaction with school did not account for any additional
significant variance. However, it is noteworthy that the 1% unique variance
contributed by this variable was approaching significance (p=.059). Interestingly,
at step 1, only three domains contributed significant unique variance. These were
achieving in life (sr² = .05, p< .001), safety (sr² = .04, p< .001) and standard of
living (sr² = .03, p< .001). This finding is somewhat consistent with those from
study 1 where three domains, health (sr² = .07, p< .001), standard of living (sr²
= .02, p< .001) and safety (sr² = .02, p< .001) were significant unique predictors.
177
Thus, while unique contributions from standard of living and safety are significant
across both studies, in the present sample, satisfaction with achieving in life is
now a unique predictor in the place of satisfaction with health (as found in study
1).
5.14 Satisfaction with School as a Unique Construct: Exploratory Analysis
Using Combined Data
The aim of the following analysis was to repeat the above analysis using
combined data from studies 1 and 2. Although the criteria for the regressions
undertaken in the previous two studies were met (N ≥ 50 + 8m, Tabachnick &
Fidell, 2001), it was decided to run this analysis again with greater statistical
power. Altogether, 351 cases were used in the following stepwise regression
analysis.
Means, standard deviations, simple correlations and partial correlations between
variables are presented in Table 72.
Table 72:
Means, standard deviations and correlations between variables (N=351)
Variable
Mean SD
1.
1. Life Sat.
73.33 19.41
-
2. Std Living
73.36 19.80
3. Health
71.85 21.27
4. Achieving
67.92 19.30
.52
(.24)
.58
(.31)
.59
(.31)
.52
(.14)
.55
(.21)
.33
(.12)
.52
(.15)
.47
(.09)
5. Relationships 79.12 18.18
6. Safety
79.80 19.74
7. Neighborhood 74.71 21.45
8. Future
69.34 19.98
9. School Sat.
72.31 21.09
2.
3.
4.
5.
6.
7.
8.
9.
.44
(.24)
.41
(.18)
.38
(.12)
.33
(.06)
.25
(.10)
.45
(.23)
.41
(.18)
.54
(.34)
.47
(.21)
.40
(.13)
.30
(.15)
.50
(.26)
.30
(.00)
.51
(.26)
.41
(.13)
.26
(.10)
.53
(.31)
.43
(.17)
.50
(.25)
.37
.39
(.22) (.25)
.46
.45
.37
(.17) (.17) (.22)
.46
.43
.18
.39
(.19) (.17) (-.01) (.09)
-
178
Results of the regression analysis, with all seven domains entered at step 1 and
school satisfaction entered at step 2, are presented in Table 73.
Table 73:
Predicting LS using 7 PWI domains and satisfaction with school
R2
Adj. R2
∆R2
B
SE B

Sr²
Step 1
Std Living
.19***
.04
.19
.03
Health
.18***
.04
.20
.02
Achieving
.23***
.05
.22
.03
Relationships
.10*
.05
.10
.005
Safety
.22***
.04
.23
.03
Neighborhood
.01
.04
.02
.00
Future
.06
.05
.07
.00
.56
.55
Unique = 11.5%
Shared = 43.5%
Step 2
Std Living
.17****
.04
.17
.02
Health
.19****
.05
.21
.03
Achieving
.21****
.05
.21
.02
Relationships
.08
.05
.07
.00
Safety
.20***
.05
.21
.03
Neighborhood
.02
.04
.03
.00
Future
.06
.05
.06
.00
School Satisfaction
.10*
.04
.11
.01
.57
.56
.01*
Unique = 11.0%
Shared = 45.0%
* p<.05; ** p<.01; *** p<.001
The R for step 1 of this regression is significantly different from zero, F(7, 343) =
62.195, p < .001. This time, five domains contributed significant unique variance
to the prediction of LS: standard of living (sr2 = .03), health (sr2 = .02), achieving
in life (sr2 = .03), relationships (sr2 = .005) and safety (sr2 = .03). Altogether, 55%
of the variability in LS can be predicted from scores on these five domains. When
entered at step 2, satisfaction with school accounted for an additional 1.0% of the
179
variance, ∆R2 = .01, Finc(1, 342) = 5.892, p<.05. Thus, satisfaction with school
now meets the criteria of a unique construct and can be considered as a domain
representing the first level deconstruction of satisfaction with life as a whole.
Interestingly, at step 2, satisfaction with relationships becomes a non-significant
predictor; indicating that school satisfaction is a more important predictor in this
regression than satisfaction with relationships. These results will be discussed.
5.15 Additional Exploratory Analyses: Predicting School Satisfaction
In anticipation of the possibility that satisfaction with school would meet the
criterion for a new domain, several items were included in the questionnaire to
allow an exploration of the domain in terms of its own deconstruction. These 9
items were adapted from the Piers-Harris Children’s Self-Concept Scale (Piers,
1986).
Means, standard deviations, simple correlations and partial correlations between
variables are presented in Table 74.
Table 74:
Means, standard deviations and correlations between variables (N=205)
Variable
Mean SD
1. School Sat. 67.56 21.07
2. Behavior
74.98 19.37
3. Abilities
68.63 21.05
4. Appearance 70.93 19.67
5. Popularity
71.41 21.18
6. Safety
81.56 20.33
7. Travel
79.46 21.52
8. Teachers
71.02 20.59
9. Classmates. 78.34 19.56
10. Friends
84.78 18.72
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
.58
(.45)
.54
(.40)
.40
(.21)
.39
(.20)
.51
(.37)
.41
(.27)
.57
(.49)
.41
(27)
.41
(.29)
.62
(.51)
.45
(.29)
.39
(.19)
.48
(.33)
.33
(.16)
.54
(.46)
.48
(.36)
.44
(.33)
.48
(.31)
.40
(.21)
.36
(.17)
.31
(.14)
.48
(.39)
.25
(.07)
.22
(.06)
.50
(.33)
.49
(.34)
.27
(.09)
.39
(.27)
.39
(.25)
.29
(.15)
.60
(.47)
.39
(.23)
.30
(.16)
.60
(.50)
.55
(.46)
.52
(.41)
.39 .32
(.28) (.21)
.57 .33 .43
(.47) (.20) (.34)
.51 .37 .31 .75
(.41) (.27) (.21) (.71)
* (.xx) partial correlations using core affect as a covariate presented in brackets
-
180
The correlations presented in Table 74 indicate a moderate relationship between
most items and school satisfaction. Furthermore, when variance attributed to core
affect is removed, these correlations reduce substantially, indicating a degree of
dependence. Interestingly, moderate partial correlations between satisfaction with
school and satisfaction with teachers at school (r = .49), satisfaction with school
and behavior at school (r = .45) and satisfaction with school and satisfaction with
abilities at school (r = .40) remain. Thus, the relationship between school
satisfaction and these domains is not dependant upon their shared relationship
with core affect.
Results of the standard multiple regression analysis is presented in Table 75.
Table 75:
Predicting satisfaction with school
DV: Satisfaction
with School
.58
B
1.81*

.17
sr2
.01
2. Abilities
.54
2.06**
.21
.02
3. Appearance
.40
-.07
-.01
.00
4. Popularity
.39
-.16
-.02
.00
5. Safety
.51
1.95*
.19
.02
6. Travel
.41
.83
.09
.00
7. Teachers
.57
2.77***
.27
.04
8. Classmates
.41
-.70
-.07
.00
9. Friends
.41
1.60
.14
Variable
1. Behavior
* p<.05; ** p<.01; *** p<.001
Unique Variance = .10
Shared Variance = .39
.01
2
R = .52
Adjusted R2 = .49
The R for this regression is significantly different from zero, F(9, 195) = 23.152, p
< .001. Altogether, 49% of the variability in satisfaction with school was
predicted. Only four of the nine domains contribute significant unique variance.
These are satisfaction with teachers at school (sr² = .04, p< .001), satisfaction with
abilities at school (sr² = .02, p< .01), satisfaction with safety at school (sr² = .02,
p< .05), and satisfaction with behavior at school (sr² = .01, p< .05).
181
5.16 Study 2 Results: Summary
In summary, the results from study 2 are generally consistent with those from
study 1. In particular, the mean score for SWB was within the Australian adult
normative range, if at the lower end. The adjectives happy, content and alert
explained 57% of variance in SWB, comparable to the 59% of variance in study 1.
Inconsistent with hypothesis three, using the entire data set, core affect explained
3% more unique variance in LS than in SWB (9% versus 6%). However, when
cases below 50 were removed, core affect explained 1% more unique variance in
LS than in SWB (9% vs 8%), which is in the predicted direction.
Inconsistent with the fourth hypothesis, none of the buffer variables explained
variance in SWB beyond core affect. However, using combined data from studies
1 and 2, secondary control became a significant predictor. Consistent with the
fifth hypothesis, core affect is clearly driving the relationship between SWB and
related constructs. This was demonstrated using both multiple regression analysis
and SEM.
In terms of the possible models that were considered, structural equation
modelling indicated that the Affective-Cognitive model of SWB provided the
statistically preferred description of how the variables relate with one-another.
In partial support of the sixth hypothesis, satisfaction with ‘standard of living’ and
satisfaction with ‘safety’ significantly predicted LS, however, satisfaction with
‘health’ did not. Finally, satisfaction with school met the criteria for consideration
as a new domain when analyses were conducted using combined data from studies
1 and 2. Thus, there is evidence that satisfaction with school may be a unique
construct.
182
CHAPTER 6: STUDY 2 DISCUSSION
The aim of study 2 was to replicate the major findings from study 1. These are: 1.
That the mean score for LS will approximate the mean score for SWB and that the
mean score for SWB will lie within the Australian adult normative range; 2. That
adjectives happy, content and alert will dominate the explained variance in SWB;
3. That core affect will explain greater unique variance in LS than in SWB when
the system is functioning normally; 4. That in circumstances of homeostatic
challenge, but not in homeostatic rest, primary control and secondary control will
explain variance in SWB beyond core affect; 5. That core affect is driving the
relationship between SWB and related variables; 6. That in confirmation of study
1, only the three domains of health, safety and relationships will contribute unique
variance to LS.
6.1
Hypothesis one: That the mean score for LS will approximate the mean
score for SWB and that the mean score for SWB will lie within the
Australian adult normative range
As discussed in study 1, domain-based satisfactions that comprise the measure of
SWB represent the ‘first-level deconstruction’ of satisfaction with ‘life as a
whole’ (International Wellbeing Group, 2006). Thus, although the mean score for
LS should approximate the mean score for SWB, due to the differing levels of
abstraction between these measures, the two scores are not expected to be the
same. Consistent with the first hypothesis, a t-test confirmed that the mean score
for LS (72.15 points) was not statistically different from the mean score for SWB
(73.61 points). Furthermore, consistent with study 1 and with Australian adult
normative statistics, the mean score for SWB was within the Australian adult
normative range of 73.43 to 76.43 points (Cummins et al., 2006). However, the
mean score for LS was below the Australian adult normative range for this
variable, which is 75.20 to 79.10 points (Cummins et al., 2006). It is also
noteworthy that whilst the standard deviation for SWB is comparable to study 1
(14.18 vs. 13.95), the standard deviation for LS is somewhat larger in study 2
(18.05 vs 20.27) - a difference of + 2.22 percentage points.
183
The implication of these findings is that the present sample of adolescents most
likely comprises a greater than normal proportion of people with low wellbeing.
To reiterate, according to homeostasis theory, mean population scores below
70.00 points represent a defeat of the homeostatic system for a significant
proportion of the population (Cummins, 2002). The inability of the system to
maintain mean life satisfaction above the line of resistance causes the distribution
to collapse and become more platykurtic. When this occurs, the standard deviation
for the population will increase and this is precisely what has been observed in the
present data set.
There are also implications for the comparison of results between the two studies.
Compared to study 1, the proportion of people with low SWB (<50) has risen by
2.1% from 4.7% (7 cases) to 6.8% (14 cases). Thus, it is likely that this sample
may contain a higher than normal proportion of people who are at risk of
depression (Cummins, et al., unpublished) and who are exhibiting a loss of
homeostatic control. As discussed previously, this has important implications for a
number of hypotheses being tested as the pattern of results will differ depending
on whether the homeostatic system is resting or under threat/defeat.
In summary, results from studies 1 and 2 confirm that the mean SWB score for
adolescents approximates adult norms. However, since the mean score lies at the
lower end of the normal range, the sample may comprise of a greater proportion
of people in homeostatic defeat and at risk for depression.
6.2
Hypotheses two and three: That adjectives happy, content and alert will
explain significant variance in SWB and that core affect will explain
greater unique variance in LS than in SWB when the system is
functioning normally
As for study 1, it was hypothesised that adjectives located at the pleasantunpleasant axis of the Circumplex Model of Affect will dominate in explaining
significant variance in SWB. This hypothesis was based on findings from Davern,
et al., (2007) that SWB is primarily a measure of pleasant affect, with some
activation. The results of study 2 corroborate these findings. More specifically, the
184
adjectives happy (pleasant), content (pleasant) and alert (activation) explained
57% of variance in SWB and this is comparable to the 59% variance explained by
these variables in study 1. Thus, it has now been empirically demonstrated in two
independent samples of adolescents that SWB is primarily a measure of pleasant
affect with some activation. Interestingly, however, the adjectives that contributed
unique variance in LS differed between the two studies. Whereas in study 1,
happy, alert, content and unhappy all contributed unique, significant variance,
happy explained 66% of variance, with discontent adding a further 2% variance
(p<.001) in study 2. Thus, for the prediction of LS, happy was the only common
predictor in the two samples.
The explanation of this discrepancy may be found in the sample. In the present
study, the correlation between LS and happy (r = .82) was significantly greater
(p<.01) than this same correlation in study 1 (r = .68). Consistent with these
correlations, regression analyses indicated that the unique contribution of happy to
LS in the present study (sr² = .08) was considerably greater than that in study 1
(sr² = .02). Thus, it appears that happy is explaining a majority of the unique
variance and since the present sample is at the lower end of the normal range, is
likely to consist of a greater than normal proportion of people with low SWB.
Because of this, discontent becomes a unique predictor, explaining the remaining
portion of variance in negative affect (i.e., residual negative variance).
Turning now to hypothesis three, as in study 1, it was hypothesised that core
affect would explain more unique variance in LS than in SWB due to its greater
level of abstraction (Cummins et al., 2003), with responses to the domains
containing more cognition. In study 1, it was found that this effect was dependent
on a normally functioning homeostatic system. That is, while across the entire
data set, core affect explained the same amount of unique variance in LS as SWB
(8.0%), when cases in the defeated range for SWB were removed (<50), core
affect explained 3% more unique variance in LS than SWB (11% vs. 8%).
Curiously, however, the present sample conformed more closely to theoretical
expectations than the first sample. This is unexpected given that the second
sample contains a higher proportion of people with low SWB. In the second
185
sample, across the entire data set, core affect explains 3% more unique variance in
LS than in SWB (9% versus 6%), thus confirming the hypothesis. However, when
cases below 50 were removed, core affect explained the same amount of unique
variance in LS (9%) but 2% more in SWB (8%). Thus, when the cases below 50
were removed, the hypothesis was just supported, with core affect explaining 9%
of the variance in LS and 8% of the variance in SWB.
These results are consistent with homeostasis theory in that, when the system is
functioning normally, core affect will explain more variance in LS than in SWB.
However, as homeostasis becomes dysfunctional under challenge, the
determination of both LS and SWB progressively shifts from core affect to the
challenging agent. In this process, core affect explains less unique variance in
both constructs. Moreover, the reduction in explained variance is somewhat
greater in LS (although this effect was not observed in the second sample)
because it is more abstract and therefore rather more determined by core affect in
the first place.
While these results are generally confirmatory of the hypothesis, the fact that the
first whole sample failed to show the differential LS vs SWB differences in
explained variance is inconsistent with theory. However, if the respondents
scoring <50 are exerting a strong effect on the whole sample, this result would be
expected. A further anomaly within the results is in the second sample, such that
when cases below 50 were removed, core affect explained the same amount of
unique variance in LS. This finding contrasts theory which posits that core affect
will explain more unique variance when the system is functioning normally.
In summary, it has been demonstrated, in two independent samples, that for the
prediction of SWB, adjectives happy, content and alert have proven significant,
unique predictors. This finding is consistent with Davern, et al., (2007) and
confirms that SWB is largely a measure of pleasant affect, with some activation.
Furthermore, for the prediction of LS, happy and discontent were unique
predictors. This finding contrasts the same analysis in study 1 where happy, alert,
content and unhappy all contributed unique, significant variance. The implication
of this finding is that in the present sample, happy is more highly correlated with
186
LS than in study 1 and as a consequence, is consuming a majority of unique
variance most likely attributed to unhappy when these variables are less related.
Finally, inconsistent with theory, core affect did not explain additional unique
variance in LS when cases below 50 were removed. Although this result is an
anomaly, it is unlikely that the magnitudes of the differences observed are of
sufficient magnitude or consistency to pose a significant challenge to theory.
However, the result warrants further investigation.
6.3
Hypothesis four: That in circumstances of homeostatic challenge, but
not in homeostatic rest, primary control and secondary control will
explain variance in SWB beyond core affect
To test this hypothesis, the sample was split into two groups (> 70 points and 4569 points). A step-wise multiple regression analysis was then performed on each
group. As expected, correlations between the buffer variables and SWB were
somewhat, but not significantly greater in the 45-69 group and this difference was
in the expected direction. This trend is consistent with the idea that when SWB
set-points are challenged, the buffers will play a greater role in maintaining SWB
stability.
Within the > 70 group, core affect explained 35% of the variance and this is
significantly greater (p<.01) than the 10% variance explained by core affect in the
45-69 group. This is consistent with homeostatic theory which predicts that core
affect will explain more variance in SWB in groups that are maintaining their
SWB within the normal range because core affect is the main driver of SWB
under conditions of low challenge (Davern, et al., 2007).
In other respects, these results provide only partial support of the fourth
hypothesis. While the overall amount of explained variance contributed by the
buffers in the 45-69 group was significant at step 2, no single buffer variable
contributed unique significant variance. This differs from the same analyses in
study 1, where primary control and secondary each contributed unique significant
variance above core affect.
187
To determine whether the failure of the buffer variables to contribute unique
variance above core affect in the challenged sample may be due to sample
characteristics or measurement error, data for challenged individuals across both
studies were combined. This analysis revealed that, in the presence of the buffer
variables, core affect no longer significantly predicted SWB, with all the
significant variance explained by secondary control. This is consistent with theory.
In challenged populations, core affect will lose its affiliation with SWB as
cognition assumes control.
It is interesting that secondary control is the only buffer variable to reliably
predict SWB. By using secondary control, individuals enhance their personal gain
by accommodating to existing realities (Weisz, et al., 1984). According to the
discrimination model (Thompson et al., 1998), the most adaptive control beliefs
are context dependant, with primary control considered most adaptive when a
situation is controllable and secondary control considered most adaptive when a
situation is resistant to change. Typically, the school environment is one that is
dominated by rules and regulations. Students are expected to arrive at a certain
time in the morning, attend particular classes, have lunch at pre-determined times
in allocated spaces and leave at a specified time. For most students, exercising
primary control is difficult in an institution where failing to obey the rules will
often result in penalty. Thus, because the school environment does not allow
much primary control, embracing secondary control strategies may be a more
constructive approach to the maintenance of wellbeing.
It is also notable that the buffers of self-esteem and optimism made no
contribution to SWB. One explanation is that these constructs are largely driven
by core affect. The partial correlations presented in Table 59 support this
argument. If this is confirmed by future studies it will require a reconceptualisation of self-esteem and optimism as SWB buffers.
Finally, there is also an anomaly that concerns the nature of the relationship
between the buffer variables and SWB. According to theory, all such relationships
should be positive because they are driven by core affect. Zero-order correlations
confirmed this. However, according to the data from study 2, in both the
188
unchallenged >70 and challenged 45-69 groups, primary control contributed
negative standardised and unstandardised regression coefficients when entered
after core affect.
One explanation is that core affect may be acting as a suppressor variable.
According to Cohen, Cohen, West and Aiken (2003), shared variance between
one or more variables introduced into a model may suppress or control some
variance in another explanatory or independent variable. Thus, introducing core
affect into the model could cause a reversal of the valence of the relationship
between primary control and SWB through negative suppression. The implication
is that the residual variance in primary control, not accounted for by core affect,
may be contributing variance in SWB that is opposite in valence to what would
normally be observed in the uncontrolled relationship.
The nature of this residual variance is uncertain. However, exercising primary
control implies some energy expenditure on the part of the individual attempting
to alter their world. This may produce stress and anxiety that, in turn, reflects the
residual negative variance in primary control. This result is somewhat consistent
with the discrimination model (Thompson et al., 1998) previously discussed.
More specifically, the negative variance contributed by primary control may
reflect the inappropriate use of this type of control in a situation where actual
control is low. For example, in a low control situation (such as in a classroom
setting) attempts to exercise primary control may have resulted in failure or
disappointment when certain outcomes were not met.
In summary, these results generally support theory. As predicted, in both samples,
there is evidence to support the hypothesis that core affect will have greater
influence over SWB when the homeostatic system is functioning normally. The
results concerning the buffer variables, however, are not so consistent with theory.
For example, while it was found in study 1 that both primary and secondary
control contributed unique significant variance in the challenged group, this did
not occur in study 2. Moreover, using the combined data from both studies, only
secondary control made a unique contribution. The implication of this finding is
189
that secondary control is the only buffer variable to reliably predict SWB and
appears to be the buffer most intimately linked to SWB.
As previously mentioned, it was also found that self-esteem, primary control and
optimism made no unique contribution. The implication of this finding is that core
affect is consuming a majority of variance otherwise attributed to the buffers in
the uncontrolled relationship. Another important finding was that results differed
when the samples were combined, highlighting the dangers of drawing major
conclusions on the basis of the results of a single study. Finally, the finding that
primary control contributed negative regression coefficients after controlling for
core affect is probably due to its action as a suppressor variable, thereby causing a
reversal of the valence of the normally positive relationship between primary
control and SWB.
6.4
Hypothesis five: That core affect is driving the relationship between
SWB and related variables.
There is now published evidence that core affect may be driving the relationship
between SWB and related constructs (Davern, et al., 2007). Study 1 provided
overwhelming support for this notion. Partial correlations indicated that in the
presence of core affect, the relationship between SWB and the buffer variables,
MDT and personality, reduced considerably, suggesting that the previously
reported strength of correlations between these constructs and SWB should be
revised in the presence of suitable affective controls. The results of the second
study support those of study 1. Again, when variances attributed to core affect
were removed, all correlations between SWB and related constructs underwent
considerable reductions. Thus, there is further evidence in the second sample of a
need to reinterpret previous findings within the literature.
6.5
Exploratory analyses: Personality and MDT as predictors of SWB
As in study 1, a secondary aim of the second study was to explore the relationship
between SWB, personality and MDT. Considering personality first and in contrast
to study 1, emotional stability contributed a further 2% (p<.01) variance above
core affect. Although this finding confirms a weak link between personality and
190
SWB, it does not corroborate previous assertions (e.g., Brickman & Campbell,
1971; DeNeeve & Cooper, 1998; Emmons & Diener, 1985; Vitterso, 2001;
Vitterso & Nilsen, 2002) that personality is one of the strongest and most
consistent single predictors of SWB (at least not in the presence of core affect) or
that it plays a crucial role in determining individual SWB set-points (Headey &
Wearing, 1989, 1992). Rather, the data suggest that emotional stability plays a
significant, yet very subsidiary role in explaining variance in SWB above core
affect. This claim can be made on the basis that almost all of the variance in SWB
is explained by core affect. Further, previously reported correlations demonstrate
that when core affect is controlled, the relationship between personality and SWB
decreases
considerably.
Thus, previously reported
correlations
between
personality and SWB are dependent upon the relationship personality shares with
core affect. In other words, when the affective content is removed from
personality, the residual relationship is, at best, very weak. The implication of this
is that previously reported correlations between personality and SWB should be
revisited in the presence of suitable affective controls.
In terms of MDT, there is a strong tradition of thought that SWB comprises a
cognitive component (e.g., Campbell, Converse, & Rodgers, 1976; Diener &
Diener, 1996; Steel & Ones, 2002; Veenhoven, 1994) and this is best measured
through the most comprehensive of the ‘gap’ approach to SWB – as from MDT
(Michalos, 1985). As in study 1, the assumption that SWB is largely a measure of
cognition was tested. In this second sample, MDT explained an additional 1%
unique variance (p<.05) above core affect - rather less than it contributed in study
1. This is consistent with the view of Davern, et al., (2007) that SWB is largely a
measure of core affect with a minor independent contribution from MDT. The
implication of this finding is similar to that for personality – that the previously
reported correlations between MDT and SWB are dependent upon the relationship
MDT shares with core affect. Thus, MDT is largely a measure of core affect and,
with respect to personality, previously reported correlations between MDT and
SWB should be revisited in the presence of suitable affective controls.
An additional analysis involving MDT determined the influence of the individual
discrepancy items on SWB above core affect. The reason for this analysis was to
191
more precisely determine the nature of the remaining cognitive component of
SWB above core affect. It was found that three discrepancy judgements (‘selfothers’, ‘self-future’ and ‘self-best’) together explained a further 6% variance 5% more variance than that explained by the composite variable ‘MDT’. A reason
that these 3 three discrepancies are a more powerful predictor of SWB than the
composite MDT is that the inclusion of four additional items that are not
intimately tied to SWB to form the composite MDT has deflated the overall
correlation between the composite MDT and SWB. Consequently, the composite
MDT variable will appear less related. This has important implications for the
way in which MDT should be used in future investigations of this type.
In terms of the individual discrepancies, the ‘self-other’ discrepancy contributed
the strongest regression coefficients. Thus, it appears that the comparisons
adolescents make with their peers significantly influences SWB. This finding is
somewhat consistent with a study by Michalos (1991), who using a global sample
of over 9,000 students aged 17-25, found that social comparison discrepancies had
the second greatest impact on LS. The ‘self-future’ discrepancy also made a
unique contribution in the second study, (sr² = .01) and this not surprising given
that this item is worded similarly to the item that measures optimism. To reiterate,
across the whole sample, optimism also made a unique contribution above core
affect (sr² = .01). Given the similarity between the two measures, it was expected
that the effect of the self-future discrepancy on SWB would disappear in the
presence of a suitable measure of optimism. This prediction was confirmed.
Another interesting result is that a negative relationship was observed between the
‘self-best’ discrepancy and SWB, indicating that introducing core affect into the
model as a covariate is again causing negative suppression. The implication is that
residual variance in the ‘self-best’ discrepancy, not accounted for by core affect,
may be contributing variance in SWB that is opposite in valence to what would
normally be observed in the uncontrolled relationship. This residual negative
variance may stem from a specific thought about an aspect or situation in a
person’s life that has not improved since the previous year. For example, although
perceptions of life overall may be better than they were a year ago, specific
aspects, such as the status of a relationship, may not have improved. Thus,
192
negative variance associated with such a thought may be responsible for the
observed reversal of valance in the ‘self-best’ discrepancy item.
Finally, when all 7 discrepancy judgments were entered above core affect and the
buffer variables, only one discrepancy judgment (self-best) contributed unique
variance (sr² = .01). This confirms that MDT appears largely a measure of core
affect, whilst a considerable amount of the remaining variance is shared with the
buffer variables (e.g., with optimism). This implies that, above core affect and the
buffer variables, discrepancy judgments capture only a small portion of the
remaining cognitive variance.
In summary, the results are again consistent with the proposition that SWB is
primarily an affective construct, with some subsidiary contribution from MDT.
Thus, although the results are in agreement with research suggesting that SWB is
a construct involving both affective and cognitive processes (e.g., Diener et al.,
2003; Veenhoven, 1994), the proportional contribution is very different from that
envisaged by these authors.
6.6
Model testing using SEM in AMOS
The aim of the structural equation models was to determine which theoretical
model fits the data best. Models being tested were the same as Davern, et al.,
(2007). These were a Cognitive-Affective model of SWB, a Personality-driven
model of SWB and a Multiple Discrepancies model for SWB. Consistent with
Davern (2007), according to model fit statistics, the Affective-Cognitive model
was the best fitting model. It is particularly interesting to note that the model fit
statistics indicated a poor fit for the Personality-driven model. This challenges a
large body of research which shows a strong link between extraversion,
neuroticism and SWB (e.g., Brickman & Campbell, 1971; DeNeeve & Cooper,
1998; Emmons & Diener, 1985; Headey & Wearing, 1989, 1992; Vitterso, 2001;
Vitterso & Nilsen, 2002). These current results indicate that the vast majority of
these relationships are caused by the common element of core affect. A similar
conclusion is made in respect of MDT. Taken together, these results
overwhelmingly support core affect as the major component of SWB and again
193
suggest that the strength of the relationship between personality and SWB and
between MDT and SWB should be revisited in the presence of suitable affective
controls.
6.7
Hypothesis six: That in confirmation of study 1, only the three domains
of health, safety and relationships will contribute unique variance to LS.
In adult data from Australia, six domains usually make a unique contribution
when regressed against LS. One domain which has not frequently been found to
make a contribution is safety (Cummins, et al., 2003). One of the unexpected
results from study 1 was that only three domains contributed significant unique
variance in LS. These were satisfaction with health, safety and standard of living.
In study 2, three domains made a contribution - standard of living and safety, as in
study 1, and achieving in life instead of health. In attempting to explain this
finding, it is notable that 50% of participants are year 12 students and this study
took place toward the end of these students time at school. For many of these
students, a sense of fulfilment and achievement nearing the completion of their
school days would be a significant source of satisfaction and thus, an important
contributor to their overall level of LS. Similarly, in adult samples, satisfaction
with achieving in life consistently ranks amongst the top three domains in terms
of its unique contribution toward LS. Thus, adolescents, like adults, have a need
for achievement satisfaction.
In relation to the other two domains, these results provide additional support for
the conclusions from study. These are that like adults, adolescents will have a
preoccupation with material possessions and keeping up to date with the latest
trends such that satisfaction with standard of living will be highly related to LS;
and that given the saliency of safety and related issues to high school students,
satisfaction with safety will be related to LS. It is important to note that
satisfaction with standard of living is a consistent and strong predictor of LS in
adult samples and that the unique contribution by this domain is not just restricted
to adolescent samples. Similarly to adolescents, adults also have a need for
material goods and services that make life easier, more comfortable and enjoyable.
194
In attempting to explain why satisfaction with health did not make a unique
contribution in this sample, it may be that students were so preoccupied with
homework and exams that a need for achievement satisfaction dominated their list
of concerns. Interestingly, the domain of health consistently contributes unique
variance in adult samples.
Finally, an anomaly within the two samples is that in adult populations,
satisfaction with relationships consistently ranks in the top few domains in terms
of its unique contribution to LS (e.g., Cummins et al., 2006). However, along with
satisfaction with future security and community connectiveness, this domain did
not provide any additional explanatory variance for adolescents. This is an
interesting result given that adolescents, like adults, require mutual and supportive
relationships.
An explanation may be found in the wording of the question. For example, the
PWI-A asks ‘How satisfied are you with your personal relationships?’, whilst the
PWI-SC asks ‘How satisfied are you with getting on with the people you know?’.
It is possible that the PWI-A form of the question is more abstract and will be
influenced more heavily by core affect. As a consequence, LS and the domain of
relationships will be closely related due to shared cored affect. On the other hand,
the PWI-SC alternative more specifically directs attention away from
‘relationships generally’ to ‘people you know’. The latter may evoke a mental
search for people whom the responder does not feel particularly close to, for
example, peers not considered friends, teachers, coaches or the neighbours. Thus,
this response will deviate from core affect, involve greater cognition and as a
consequence, will be less related to LS. If these findings are confirmed, they
would indicate that this domain is not equivalent between the adult and school
versions.
In summary, these findings offer some support for stability in adolescent domainbased satisfaction judgements. First, as would be expected and similarly to adults,
adolescents have a need for a standard of living that meets their requirements for a
comfortable life. Unlike adults, however, adolescents consider safety as an
important component of their overall wellbeing. This may be so because
195
adolescents see themselves as a somewhat more vulnerable group. In terms of
satisfaction with health, whilst this was a unique predictor amongst adolescents in
study 1, it was not in study 2. In explanation of this, it is proposed that at times,
adolescents may consider other needs, such as that for achievement, as more
important. For example, during exam time, health concerns may be overlooked
and achievement satisfaction will dominate their list of concerns and will thus be
more closely related to LS. Finally, unlike the adult samples, it was found that
satisfaction with relationships did not predict LS. However, this may be so
because of the wording of the question. Adolescents, perhaps even more so than
adults, need mutual and supportive relationships.
6.8
Exploratory analyses: Satisfaction with school as a unique construct
A secondary aim was to explore whether satisfaction with school would meet the
criterion for a new domain on the PWI-SC by predicting unique, significant
variance in global life satisfaction above the other seven life domains. Consistent
with study 1, it did not achieve this criterion. However, it is noteworthy that the
1% unique variance contributed by this variable was approaching significance (p
= .059). To determine whether the results may be an artefact of sample size, the
analysis was repeated using the combined data from studies 1 and 2. The rationale
for this procedure was to increase the statistical power of the analysis and
decrease the chance of committing Type II error (Howell, 2004).
When the 351 cases were combined, the results were substantially different from
those using the smaller samples. At step 1, five domains made a significant
contribution. The exceptions were community connectedness and future security.
The second difference is that when entered at step 2, school satisfaction
contributes a further 1% variance (p<.05) above all 7 domains, thus, qualifying as
a unique construct. Thus, school satisfaction has fulfilled the major criterion for a
domain and should be considered for inclusion in a future revision of the PWI-SC.
It is also notable that, in the presence of school satisfaction, satisfaction with
relationships no longer predicted LS. As discussed, this may be so due to the
wording of the question.
196
It is also important to note that when data from studies 1 and 2 were combined,
despite more domains becoming significant contributors of unique variance, the
overall amount of explained variance did not increase. The implication is that
increasing the number of cases in the analysis has had an effect on the number of
significant predictors due to increased power of the analysis (the ability to detect
an effect when an effect is present). However, these additional significant
domains have not resulted in further explanatory variance.
6.9
Exploratory analyses: Predicting satisfaction with school
In the event that satisfaction with school did meet criteria to be considered a
unique construct, additional items had been included in the questionnaire.
Adapted from the Piers-Harris Children’s Self-Concept Scale (Piers, 1986), these
items were designed to explore the conceptual nature of school satisfaction in
terms of those aspects of school life that best represent the construct satisfaction
with school. It was found that 49% of variance in school satisfaction was
explained by four of the items. These are: satisfaction with teachers at school,
satisfaction with abilities at school, satisfaction with safety at school and
satisfaction with behaviour at school.
Whilst the face validity of these items is high, since they all have obvious
relevance to the school experience, previous research also supports a link between
these constructs and adolescent wellbeing. For example, the finding that
satisfaction with teachers was a unique predictor of school satisfaction is
somewhat consistent with a study by Baker (1999). Using a small sample of 3rd
through 5th grade students in the USA, it was found that perceptions of a caring,
supportive relationship with a teacher and a positive classroom environment were
related to school satisfaction by as early as the 3rd grade. Furthermore, in partial
support of the influence of teachers on LS, a study by Suk-Un and Moon (2006),
using a sample of 299 ‘high-ability’ students, sought to investigate the
psychological wellbeing and school life satisfaction of students in Korea.
According to the results of their open ended questions, students appreciated the
expertise of their teachers and also reported satisfactory relationships with their
teachers. In fact, 47.1% of students were most satisfied with the relationship they
197
had with their teachers and other students. Interestingly, the next most highly
rated category was satisfaction with ‘freedom in school’, with 30.4% of students
reporting satisfaction with this aspect of school life. Thus, there is some evidence
that students value their teachers and that satisfaction with teachers may be related
to students overall level of school satisfaction. This is intuitive, given the
multifaceted role that teachers play at school and the high level of interaction they
have with their students.
With respect to satisfaction with abilities at school, there is some evidence that
perceived academic competence is related to global life satisfaction. For example,
a study by Huebner, Gilman and Laughlin (1999), using a sample of American
elementary school students (n = 183) and middle school students (n = 290), found
a positive relationship between perceived academic competence and LS (r = .36
and .37 respectively). In this study, LS was measured using the Students Life
Satisfaction Scale (SLSS; Huebner, 1991) – a 7-item global self report scale. It is
important to note, however, that it is likely that these relationships may be driven
to a large extent by core affect. If theory is correct and core affect is influencing
all satisfaction judgments, then these correlations may simply reflect shared
variances attributed to core affect. In a similar vein, using a sample of older
adolescents (698 middle and high school students), Suldo and Huebner (2004)
report a moderate correlation (r = .45) between scores on the Multi-dimensional
Students Life Satisfaction Scale (MSLSS; Huebner, 1994) and academic selfefficacy. However, the issue that core affect may be the driving force behind these
relationships must again be raised. Nonetheless, there is some evidence that
perceived abilities at school are related to the student experience.
In terms of satisfaction with behaviour at school, a review of the literature
revealed little work has examined the direct effect of this domain on school
satisfaction. However, an intuitive explanation for this connection is that students
gain a sense of personal control over aspects of their school life when they behave
in certain ways. For example, when students engage with peers, teachers and
principals in a meaningful and positive way, they are likely to elicit positive
feedback in return. Ongoing constructive interactions of this kind tend to reinforce
further positive behaviours and as a consequence, a happy, productive school
198
environment is likely to ensure. Thus, satisfaction with behaviour would be
related to the overall school experience.
Finally, the inclusion of satisfaction with safety as a predictor corroborates data
from studies 1 and 2 where it was found that satisfaction with safety predicted LS
in both samples. As discussed, safety is an issue that has been at the centre of
school and government education programs for many years. For example, Quit
Victoria (www.quit.org.au) and Kid Power (www.kidpower.org/School-age.html)
are two organisations devoted to teaching children and adolescents to be safe.
Given the saliency of this domain amongst adolescents and the need for
adolescents to feel safe and secure, it is to be expected that this domain would
contribute uniquely to students’ overall sense of satisfaction with school.
In summary, the results of these exploratory analyses suggest that by increasing
the sample size, the number of domains that reach significance and predict LS will
increase. However, these additional variables will not necessarily have a
significant impact on the total amount of explanatory variance. Thus, there is a
need to conduct further research on student samples to identify those domains that
are of greatest importance using larger samples. Furthermore, when data were
combined, school satisfaction met the criterion as an independent construct. As
such, school satisfaction should be considered as a new domain on a future
revision of the PWI-SC. Finally, four variables were found to explain almost 50%
of the variance in school satisfaction and there is some evidence within the
literature that satisfaction with teachers, perceived abilities, safety and behaviour
at school are related to student wellbeing. These findings raise awareness of
factors that are associated with and influence the adolescent school experience.
6.10 Summary and Conclusions
The collective results of this second study are generally consistent with those of
study 1. Both studies found that the mean score for SWB was within the
Australian adult normative range, albeit at the lower end. A normal but low mean
SWB score with evidence of increased variance suggests that the sample
comprises a greater than normal proportion of people with low wellbeing. Both
199
studies also support the notion that SWB is primarily an affective construct with a
minor independent contribution from cognition (Davern, et al., 2007). The
implication of this finding is that, in contrast to some research (e.g., Michalos,
1985), SWB is a construct driven by affect and not cognition. Thus, results
suggest a reinterpretation of previous literature using suitable affective controls.
The analyses provided general support for homeostasis theory. Particularly,
homeostasis theory predicts that in normal populations, core affect will explain
more unique variance in LS than the PWI because LS is more abstract (Cummins
& Nistico, 2002; Davern, et al., 2007). Although this was not confirmed for the
entire sample, it was confirmed when cases below 50 were removed – providing
support for theory in relation to when the homeostatic system is functioning
normally. However, the finding that when cases below 50 were removed, core
affect explained the same amount of unique variance in LS is an anomaly within
the results.
Also in support of SWB homeostasis theory is that core affect explained more
variance in the unchallenged >70 group than in the challenged 45-69 group.
Moreover, the finding that secondary control contributed unique variance in the
45-69 group is also consistent with theory. It is important to note, however, that
this result was obtained using combined data from studies 1 and 2. Thus, the
failure to obtain significant results using the separate samples is most likely due to
inadequate statistical power caused by small sample size. Finally, SWB
homeostasis theory predicts that when challenged, core affect will tend to lose its
affiliation with SWB. Consistent with this, in the presence of the buffer variables,
core affect no longer significantly predicted SWB in the 45-69 group. Thus, when
challenged, control over SWB moved from core affect to secondary control.
Support was found for the argument that core affect may be acting as a suppressor
variable and thereby causing a reversal of the valence of the normally positive
relationship between SWB and some constructs. It was found that in both the
unchallenged >70 and challenged 45-69 groups, primary control contributed
negative regression coefficients. In a similar vein, in the presence of core affect, a
negative relationship was observed between the ‘self-want’ discrepancy and SWB.
200
The implication of these findings is that residual variance in these constructs not
accounted for by core affect may be contributing negative variance. As such, the
use of core affect as a covariate provides a very different view of the unique
linkage between these variables.
In relation to the issue of shared variance, the results of study 2 corroborate those
of study 1 where it was found that core affect is driving the relationship between
SWB and constructs normally considered independent of one-another. These
include extraversion, emotional stability, self-esteem, perceived control, optimism
and MDT. When variance attributed to core affect was removed from such
relationships, their shared variance dramatically reduced, indicating that the
relationship between these constructs and SWB may be driven by shared variance
from core affect. As in study 1, the more interesting reduction was the dramatic
decrease in the relationships between extraversion and SWB and between
emotional stability and SWB. A major implication of these findings is that a large
body of research which presumes an independent link between personality and
SWB should be revisited with adjustments made for relevant affective variables.
Another major finding from study 2 was that according to model fit statistics, the
Affective-Cognitive model was the best fitting model. This result overwhelmingly
supports core affect as the major component of SWB and again suggests that the
strength of the relationship between personality and SWB and between MDT and
SWB should be revisited in the presence of suitable affective controls.
A further aim of study 2 was to explore the relationship between domainsatisfactions and LS. Using the combined data-set, all domains except satisfaction
with relationships, community connectiveness and satisfaction with future
security were significant unique predictors. An implication of this finding is that
increasing the number of cases in the analysis has had an effect on the number of
significant predictors due to increased power of the analysis. However, these
additional significant domains have not resulted in further explanatory variance.
Also, using the combined data set, satisfaction with school predicted LS above all
7 domains. Thus, there is evidence that satisfaction with school should be
considered as an eighth domain on a future revision of the PWI-SC. Finally, four
201
domains (satisfaction with teachers at school, abilities at school, safety at school
and behaviour at school) were found to explain almost half of the variance in
satisfaction with school. Research generally supports a link between these
domains and satisfaction with school.
Thus, the results of study 2 are generally consistent with those of study 1, with the
major finding being that contrary to previous research, it is core affect not
personality which is the main driver of SWB. The aim of study 3 is to demonstrate
the importance of affect to past, present and future research.
202
CHAPTER 7: A RE-ANALYSES OF PUBLISHED RESEARCH
INVESTIGATING THE IMPORTANCE OF AFFECT TO MEASURES OF
SUBJECTIVE WELLBEING
INTRODUCTION
Studies 1 and 2 confirmed the hypothesis that SWB is primarily an affective
construct. In support of this, regression analyses indicated that three affects happy, content and alert - explained 59% (study 1) and 57% of variance (study 2)
respectively. Furthermore, according to SEM, the Affective-Cognitive model of
SWB provided the best fit to the data and this is consistent with Davern, et al.,
(2007). These results support the hypothesis that core affect is driving the
relationship between SWB and related constructs. Most particularly, in the
presence of core affect, the relationships between personality, MDT, the buffer
variables and SWB reduced considerably. This suggests that previously reported
correlations between these constructs and SWB should be revisited in the
presence of suitable affective controls.
The aim of this chapter is to investigate past studies in this light. A series of reanalyses of past literature will be conducted using SPSS for Windows (12.0; SPSS,
Inc., Chicago, Il) and AMOS (7.0; Smallwaters Corp, Chicago, Il). The published
analyses will be remodelled from the alternative theoretical perspective using each
respective author’s measure of affect which represents the closest approximation
to core affect.
General Method
A literature search was undertaken for studies that contained the following
information:
1.
A measure of LS or SWB, personality and a measure of affect
2.
The correlations between these variables
3.
The means and standard deviations for all measured variables
203
4.
The sample size
This set of information allows the results to be re-analysed using SEM in AMOS.
Further, this technique allows theoretical models to be compared against oneother to determine which is best.
Relevant studies were identified as follows:
1.
Headey, B., & Wearing, A. (1989). Personality, life Events, and subjective
wellbeing: Toward a dynamic equilibrium model. Journal of Personality
and Social Psychology, 57(4), 731-739.
2.
Vitterso, J. (2001). Personality traits and subjective wellbeing: emotional
stability, not extraversion, is probably the more important predictor.
Personality and Individual Differences, 31, 903-914.
3
Vitterso, J., & Nilsen, F. (2002). The conceptual and relational structure of
subjective wellbeing, neuroticism, and extraversion: Once again,
neuroticism is the more important predictor. Social Indicators Research,
57, 89-118.
4.
Libran, E.C. (2006). Personality dimensions and subjective wellbeing. The
Spanish Journal of Psychology, 9(1), 38-44.
5.
Zheng, X., Sang, D., & Lin, Q. (2004). Personality, cognitive and social
orientations and subjective wellbeing among Chinese students. Australian
Psychologist, 39(2), 166-171.
Conversion of Data to Percentage of Scale Maximum (%SM)
All LS and SWB scores provided in each study were converted into a standard
form called ‘percentage of scale maximum’ (%SM). As discussed in previous
chapters, this process allows comparisons to be made with other data that have
been derived from different response scales. It is important to note, however, that
not all mean scores and standard deviations have been converted because, in some
204
of the studies, authors did not specify certain details, for example, the number of
points on a particular scale.
Data were converted to the standard 0 – 100 %SM through the use of the formula
below:
X  k min
k max  k min
X
kmin
kmax
=
=
=
x 100
the score or mean to be converted
the minimum score possible on the scale
the maximum score possible on the scale
205
REANALYSIS ONE: HEADEY AND WEARING (1989)
This paper by Headey and Wearing was chosen for the first re-analysis because
the work of these authors is highly regarded and commonly cited. Headey and
Wearing (1989, 1992) were the first to suggest that individuals may have a ‘setpoint’ level of SWB. Inspired by work suggesting that a majority of people in
Western industrialised nations have levels of wellbeing that are generally high and
positive, the aim of Headey and Wearing’s research was to detail sources of
psychological wellbeing and to understand how people cope with change.
According to these authors, prior to the late 1980’s, there were several general
theories of SWB and they critique each one of them. The first is that the traits of
extraversion and neuroticism are the major determinants of SWB. Although they
do not dispute personality as playing an integral role, their major criticism of this
approach is that a ‘pure’ personality model accounts for only a moderate amount
of variance in SWB (e.g., Diener, 1984) and excludes other important
determinants, such as demographic variables, social determinants and life events.
They also note that the personality model implies that SWB will be highly stable
over time since extraversion and neuroticism are stable traits. However, they cite a
number of studies that report only moderate levels of stability in individual SWB
scores (e.g., Abbey & Andrews, 1985; Atkinson, 1982; Campbell, Converse, &
Rodgers, 1976).
The second theory they discuss is whether major life experiences are the most
likely causes of change in SWB. According to adaptation level theorists Brickman,
Coates and Janoff-Bulman (1978), adaptation to these experiences can explain
stability in SWB. In support of their theory, these authors present data
demonstrating how both major favourable events (e.g., winning the lottery) and
unfavourable events (e.g., becoming a quadriplegic or paraplegic) had little longterm effect on SWB.
In their critique of this study, Headey and Wearing point to the small sample sizes
(22 lottery winners and 29 accident victims) as the reason this comparison did not
reach statistical significance. Moreover, while the accident victims did score
206
significantly lower than controls, this difference was reported as being
surprisingly small. For these reasons, Headey and Wearing argue that Brickman,
et al’s., results are easily over interpreted. They point out that while adaptation is
one of several important processes that may reduce the impact of some major life
events, the work of Brickman, et al., is often cited as showing that adaptation is so
rapid and complete that events have only a minor impact on SWB (e.g., Argyle,
1987; Costa & McCrae, 1980; Diener, 1984). In the view of Headey and Wearing,
this clearly is not true.
With these criticisms in mind, Headey and Wearing sought to offer a more
integrated account that describes the relationship between person characteristics,
events and SWB. The main thrust of their work centres on age and personality as
playing an integral role in determining individual SWB set-points and maintaining
stable levels of wellbeing. According to their Dynamic Equilibrium model,
although personality characteristics provide the stable set-point for wellbeing,
unusual circumstances have the potential to alter individual set-points above or
below equilibrium. However, these deviations are usually short lived because
dispositional influences on SWB, such as genetics and stable personality factors,
play an important equilibrating function that ensures under most circumstances,
SWB reverts back to baseline levels. Thus, dynamic equilibrium theory postulates
that personality and age will have a profound influence over their measure of LS,
however, favourable and unfavourable events will predict LS above and beyond
that explained by these variables.
To test the validity of the Dynamic Equilibrium model, Headey and Weary
conducted a step-wise multiple regression analysis with age and personality
entered at step 1 and the recent experience of favourable and unfavourable events
entered at step 2. Consistent with their model, they found that at step 1, both
extraversion (β = .18, p < .05) and neuroticism (β = -.20, p < .05) significantly
predict SWB-HW. However, in contrary to prediction, age did not. Furthermore,
as hypothesised, when entered at step 2, favourable events (β = .11, p < .05) and
unfavourable events (β = -.28, p < .05) significantly predict SWB-HW. Thus,
recent events predict SWB-HW above and beyond that explained by personality.
207
Altogether, the variables comprising Headey and Wearing’s Dynamic Equilibrium
model explained 17% variance in SWB-HW.
Interestingly, although affect was not integrated into their model, Headey and
Wearing acknowledge that affect may comprise an important component of SWB.
They state “one could question whether the items [comprising the LS index] tap
cognitive life satisfaction or whether they also include a substantial affective
element” and “we accept that the LS index may no longer be the most pure
measure of life satisfaction available” (p. 732). Thus, despite regarding affect as a
potentially important contributor, these authors did not conduct any analyses
examining the direct effect of affect on SWB, let alone include affect alongside
variables in their model.
The aim of this re-analysis is to a) re-construct Headey and Wearing’s model
according to paths they have specified in order to test the utility of their model b)
determine how well Headey and Wearing’s Dynamic Equilibrium model fits their
data and c) explore the influence of PA as a predictor when included alongside
variables in their model. It is hypothesised that in the presence of PA, variables
comprising Headey and Wearing’s dynamic equilibrium model will make a weak,
subsidiary contribution to the prediction of LS, with a majority of the variance
explained by affect. To test this hypothesis, a step-wise multiple regression
analysis will be conducted with variables comprising the Dynamic Equilibrium
model entered at step 1 and PA entered as a predictor at step 2.
METHODOLOGY
Participants
At the beginning of this study, participants were 942 members of the Victorian
Quality of Life Panel study interviewed on four occasions (1981, 1983, 1985 &
1987). Of those, 509 of the participants were female (54%) and 433 were male
(46%) and their ages ranged from 18-65 years. By 1987, 649 participants
remained.
208
7.1
Major Dependent Variable and Other Variables
Subjective Wellbeing Headey & Wearing (SWB-HW)
Life satisfaction was measured using a scale comprising 6 items rated on the 9point Delighted-Terrible scale (Andrews & Withey, 1976; Headey, Holmstrom, &
Wearing, 1985). Two of the items asked participants ‘How do you feel about your
life as a whole’ and these were placed near the beginning and near the end of the
interview (r = .67). The other four items were: How do you feel about ‘the sense
of purpose and meaning in your life’?; ‘what you are accomplishing in life?’,
‘how exciting your life is’ and ‘the extent to which you are succeeding and getting
ahead in life’?’ According to these authors, the six items inter-correlated over .4
and were averaged to form LS. Furthermore, the LS index had a Chronbach’s
alpha of .92 (1987 survey).
Affect
Affect was measured using Bradburn’s (1969) PA/NA scale. Each scale is based
on five yes-no items asking about affects experienced during the past few weeks.
Questions reflecting positive feelings include: Did you feel…
1.
particularly excited or interested in something?
2.
proud because someone complimented you on something you had done?
3.
pleased about having accomplished something?
4.
on top of the world?
5.
that things were going your way?
PA scores were generated by summing yes ratings for each scale (0-5). The scale
produced Chronbach’s alpha of .64.
Personality
The Eysenck Personality Inventory, form B (EPI; Eysenck & Eysenck, 1964),
which includes measures of extraversion and neuroticism, was used to measure
personality. Furthermore, based on research conducted by Costa and McCrae
209
(1985) and Norman (1963), who regard openness to experience as a third major
dimension, a measure of this construct was also included in their 1987 survey.
Thus, extraversion, neuroticism and openness to experience comprise the measure
of personality.
Life Events
Life events were measured using a modified version of the List of Recent
Experiences (LRE; Henderson, Bryne, & Duncan-Jones, 1981). Modification to
the original scale involved including items that measure the occurrence of recent
positive events as well as recent negative events. Headey and Wearing note that
the LRE measures continuing experiences, such as ‘serious problems/arguments
with your children’, as well as discrete events, such as ‘you made new friends’.
RESULTS
Two models will be investigated in this re-analysis. One model examines the
utility of Headey and Wearing’s Dynamic Equilibrium model and the second
model incorporates affect as driving the relationships between all constructs in
Headey and Wearing’s model.
7.2
Headey and Wearing’s Dynamic Equilibrium Model
The first analysis examines a) how much variance in LS can be explained by
Headey and Wearing’s Dynamic Equilibrium model and b) fit statistics for this
model. Means, standard deviations and correlations between variables are
presented in Table 76. These values are taken from the Appendicis (p. 739) in the
original publication.
210
Table 76:
Means, standard deviations and correlations between variables (N = 649)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
1. SWB-HW*
72.50%
10.66%
-
2. PA*
66.80%
27.80%
.31
-
3. Age
37.20
13.27
.03
-.13
-
4. EXT
14.13
3.25
.20
.18
-.22
-
5. NEU
11.19
4.83
-.23
.00
-.11
-.12
-
6. O
3.76
.51
.08
.23
-.13
.08
.01
-
7. Fav Event
3.16
2.21
.09
.31
-.44
.20
.02
.17
-
8. Adv Event
2.42
2.07
-.29
.07
-.16
-.01
.18
.12
.16
* scores converted to %SM statistic
Note: SWB-HW = subjective wellbeing Headey and Wearing; PA = positive affect; EXT =
extraversion; NEU = neuroticism; O = openness to experience; Fav Event = recent
favourable events experienced; Adv Event = recent adverse events experienced.
As shown in Table 76, SWB-HW shares the strongest relationship with NA (r = .38) and PA (r = .31). Furthermore, the mean score for SWB-HW is 72.50, which
is below the Australian adult normative range on the PWI which is 73.43 to 76.43.
However, the scale to measure LS contains items other than those measuring
satisfaction. For example, one item on this scale asks ‘how exciting your life is’.
This item, at least in part, is a measure of affect.
In terms of meeting the assumptions for multiple regression analysis, according to
the formula prescribed by Tabachnick and Fidell (2001), the minimum number of
cases required is 122. With 649 participants in this study, sample size
requirements have been met. In regards to the issue of normality and singularity,
as shown in Table 76, the greatest observed correlation between any two variables
is -.44. Thus, this assumption has not been violated.
A re-analysis of Headey Wearing’s model will now be conducted in order to
confirm their analysis. To achieve this aim, paths specified according to their
model must first be drawn in AMOS. It is important to note at this time that a path
coefficient is a standardised regression coefficient (β) showing the direct effect of
an independent variable on a dependent variable in the path model (Kline, 1998).
-
211
A path model specified according to Headey and Wearing’s Dynamic Equilibrium
model of SWB is presented in Figure 13.
O
NEU
EXT
.09*
.10**
.17***
.16***
.11**
.10**
-.16***
.17
.19
.05
ADV
-.28***
SWB-HW
.12**
FAV
.07
-.41***
-.13***
AGE
* p<.05; ** p<.01; *** p<.001
Note: NEU = neuroticism; O = openness to experience; EXT = extraversion; ADV =
recent adverse events experienced; FAV = recent favourable events experienced; SWBHW = subjective wellbeing Headey and Wearing
Figure 13: SEM of Headey and Wearing’s Dynamic Equilibrium Model
In the model presented in Figure 13 above, direct paths are drawn between all
other variables and SWB-HW because these variables are believed to have a
direct influence on LS. According to Headey and Wearing (p. 736), levels of
favourable and unfavourable events are predictable on the basis of stable person
characteristics (e.g., age and personality). Thus, direct paths are drawn between
age and events and between personality and events. However, they also state (p.
735) that they had no strong prior expectations about which of these person
characteristics would influence favourable and unfavourable event scores.
However, they linked extraversion and openness to experience with favourable
events and neuroticism and openness to experience with unfavourable events.
Thus, paths specifying these relationships are included in the model.
212
Looking at the fit statistics for this re-constructed model, the χ2 is 74.58 (p = .000)
which suggests that this model does not provide a good fit. A χ2/df value of 7.458
and fit indices presented in Table 77 below, also suggest a generally poor fit of the
model to the data. An SMC of .17 indicates that 17% of the variance in SWB-HW
is explained, which is the same figure they reported; an NFI of .81 indicates that
this model is not a close fit in relation to the best fitting model; and an RMSEA
of .10 is above .05, suggesting that the model does not provide a close fit in
relation to the degrees of freedom. Finally, an AIC value of 112.58 for the model
is significantly greater than that of the saturated or best fitting model (AIC =
56.00), suggesting that this model lacks parsimony. Absolute and relative fit
Indices for this model are presented in Table 77.
Table 77:
Absolute and relative fit indices for Headey and Wearing’s Dynamic Equilibrium Model
χ2
Df
χ2/df
P
AIC
NFI
RMSEA
SMC
74.58
10
7.46
.000
112.58
.81
.10
.17
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root Mean
Square Error of Approximation; SMC = Squared Multiple Correlation.
Although fit statistics suggest a generally poor fit of their model to the data, the
path model presented in Figure 13 above offers some support for Headey and
Wearing’s theory. For example, although age did not significantly influence
SWB-HW, as dynamic equilibrium theory predicts, neuroticism, openness to
experience and extraversion all contributed unique variance, as did the recent
experience of both favourable and unfavourable events. Together, these variables
explained 17% variance. Furthermore, consistent with their predictions, both
neuroticism and openness to experience significantly predicted adverse events;
whilst extraversion and openness to experience significantly predicted favourable
events. Together, personality and age explained 19% variance in favourable
events and 5% variance in unfavourable events. Finally, with respect to mediation
and the direct effects of both favourable and unfavourable events on SWB-HW
were significant. The implication of this is that events only partially mediate the
213
relationship between personality, age and SWB-HW. Thus, there is an
independent relationship between events and SWB-HW.
In summary, the re-constructed model specified according to Headey and
Wearing’s dynamic equilibrium theory presented in Figure 13 explains 17% of
variance in SWB-HW, which is the same as originally reported. Importantly,
however, the reconstructed model fit the data quite poorly. No attempt was made
to improve model fit by correlating error terms, since there was no statistical
justification for doing so. Also, re-estimating error variance in the DV had no
influence on model fit.
7.3
Re-analysis Incorporating Affect
The purpose of the next structural equation model is to explore the relationships
between variables in Headey and Wearing’s model in the presence of PA. The aim
of this analysis is to demonstrate that in the presence of PA, unique contributions
from personality, recent events and age will weaken due to the dominance of
affect over SWB-HW. Furthermore, because affect is believed to be driving life
satisfaction, it is hypothesised that the amount of explained variance in this model
will increase over the previous model.
It is important to note that the model specified in Figure 14 below is similar to that
presented above in Figure 13. However, direct paths between PA and SWB-HW,
PA and extraversion and PA and favorable events have been included because
study 1 and study 2 results indicated that PA has a direct influence over these
constructs. The relationship between PA and openness to experience is not as
clearly defined, so no paths between these constructs have been included in the
model. A path model incorporating PA alongside variables comprising Headey
and Wearing’s Dynamic Equilibrium model is presented in Figure 14 below.
214
.12
NEU
.03
O
EXT
.18***
.04
.07*
.17***
.10**
.06
.13***
-.16***
.23
.26
.05
ADV
-.29***
.04
SWB-HW
FAV
-.27***
.30***
.25***
.27***
-.40***
.06
-.13***
PA
AGE
* p<.05; ** p<.01; *** p<.001
Note: NEU = neuroticism; O = openness to experience; EXT = extraversion; ADV =
recent adverse events experienced; FAV = recent favourable events experienced; SWBHW = subjective wellbeing Headey and Wearing; PA = positive affect; NA = negative
affect
Figure 14: An Affectively-driven model of SWB-HW
Turning now to the model in Figure 14, the χ2 is 112.115 (p = .000) suggesting
that this model is not a good fit to the data. A χ2/df value of 8.008 and fit statistics
presented in Table 78 below also reflect a poor level of model fit. This is
reinforced by an NFI of .80 and an RMSEA value of .10. Further, an AIC value of
265.12 suggests that this model lacks parsimony. While an SMC of .26 indicates
that only 26% of the variance in SWB-HW is explained by this model, notably,
however, this is 9% more variance than that explained by the original. Absolute
and relative fit Indices for this model are presented in Table 78.
215
Table 78:
Absolute and relative fit Indices for an Affectively-driven model of SWB-HW
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
112.115
14
8.008
.000
256.12
.80
.10
.26
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root Mean
Square Error of Approximation; SMC = Squared Multiple Correlation.
Fit indices presented in Table 78 suggest a generally poor fit of the model to the
data. However, as hypothesised, in the revised model, PA significantly influences
SWB-HW (β = .30, p<.001). In terms of personality, while neuroticism remains a
consistent predictor of SWB-HW, in the presence of PA, the strength of the
association between neuroticism and SWB-HW has weakened slightly (from β
=.10, p<.01 to β = .07, p<.05). Further, openness to experience is no longer a
unique predictor. Interestingly, the relationship between adverse events and SWBHW remains strong in the presence of PA (β = -.29, p<.001). This suggests that
adverse events may be explaining residual, negative variance in SWB-HW
contributed by people experiencing some kind of hardship and who may be
experiencing a level of SWB below their set-point.
Collectively, although fit statistics for the Affectively-driven model suggest a
poor fit to the data, results nonetheless indicate that PA is the main driver of
SWB-HW, with residual, negative variance explained by adverse events. Not only
did PA contribute a greater standardardised regression co-efficient than
personality, events and age, including affect into Headey and Wearing’s model
had a significant impact on the total amount of explanatory variance. More
specifically, PA contributed a further 9% variance in SWB-HW above variables
comprising their Dynamic Equilibrium model.
A final analysis was conducted to determine how well a model comprising only
those variables that contributed uniquely to the prediction of SWB-HW in the
previous model fit the data. These were PA, extraversion, neuroticism and adverse
events. This final model is presented in Figure 15 below.
216
.03
EXT
NEU
-.17***
.13***
.26
.18***
.18***
SWB-HW
-.29***
.32***
.11
PA
ADV
* p<.05; ** p<.01; *** p<.001
Note: NEU = neuroticism; EXT = extraversion; ADV = recent adverse events
experienced; SWB-HW = subjective wellbeing Headey and Wearing; PA = positive affect;
NA = negative affect
Figure 15: A simplified, Affectively-driven model of SWB-HW
The χ2 for the model in Figure 15 above is not significant (χ2 =15.24, p > .05),
suggesting this model is a good fit to the data. Furthermore, the χ2/df value of 3.0
indicates that this is a parsimonious model. Fit statistics presented in Table 79,
however, are mixed. For example, an NFI of .96 indicates that this model is a
close fit in relation to the best fitting model, however, an RMSEA of .06 is outside
the range of acceptability (<.05) and an AIC value of 35.24 is greater than that of
the saturated, or best fitting model (AIC = 30.00). Finally, an SMC of .26
indicates that 26% of the variance in SWB-HW is explained by this model and
this is the same amount of variance explained as the model in Figure 14 using all
of the variables. Absolute and relative fit Indices for this model are presented in
Table 79.
Table 79:
Absolute and relative fit indices for an Affectively-driven model of SWB-HW
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
15.24
5
3.0
.009
35.24
.96
.06
.26
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root Mean
Square Error of Approximation; SMC = Squared Multiple Correlation.
217
Collectively, fit indices presented in Table 79 suggest that this model provides a
reasonable fit to the data. The strength of association between all variables in this
model are relatively similar to those presented in Figure 14, however, the
elimination of all non-significant predictors from the previous model has
improved model fit; without reducing the total amount of explanatory variance.
7.4
Summary
In summary, Headey and Wearing proposed that stability in SWB-HW can be
explained by their Dynamic Equilibrium model. More specifically, according to
their theory, personality and age predispose people to experience moderately
stable levels of favourable and unfavourable events and moderately stable levels
of SWB. A re-analysis of their data was conducted and it was found that PA not
personality, age or recent events, was the strongest predictor of SWB-HW. In fact,
PA explained a further 9% variance in SWB-HW above variables comprising the
Dynamic Equilibrium model. Results of this re-analysis suggest a re-interpretation
of Headey and Wearing’s data in the presence of suitable affective controls.
218
REANALYSIS TWO: VITTERSO (2001)
In his introduction, Virtersso cites a vast number of research papers that regard
extraversion as one of the strongest predictors of SWB (e.g., Argyle & Lu, 1990;
Costa & McCrae, 1980; Diener & Larson, 1993). However, he questions
extraversion as the cardinal SWB trait and proposes that emotional stability may
be the more important predictor. In support of this view he cites DeNeeve and
Cooper (1998), who conducted a meta-analysis of the literature and found that
when personality factors were grouped according to the Five Factor Model,
emotional stability (the positive pole of neuroticism) was the strongest predictor
of life satisfaction and happiness. Further, Vitterso argues that closer examination
of much of the literature that argues in favour of extraversion as the more
important factor reveals a number of ambiguities in its relationship with
subjective wellbeing. These are a) that the effect size of prediction between
extraversion to subjective is often quite small when compared to that of emotional
stability and b) that when emotional stability is controlled, the unique contribution
of extraversion to subjective wellbeing often attenuates or disappears altogether
due to shared variance between these constructs. Thus, Vitterso concludes that the
evidence supports emotional stability as being the more important predictor of
subjective wellbeing. At its core, this is a curious argument since extraversion and
neuroticism are supposed to be independent traits – the basis of the 5-factor model.
Based on his review, a primary aim of Vitterso’s study was to explore the
relationship between SWB and personality traits and to determine the actual
strength of zero-order correlations between extraversion and SWB. More
specifically, he hypothesised that:
1.
The association between emotional stability and indicators of SWB are
substantially stronger than their association with extraversion.
2.
The association between extraversion and SWB will be substantially
reduced when emotional stability is controlled for in a multiple regression
analysis.
219
Vitterso performed a correlational analysis on questionnaire data he collected at
two different time points. These clearly support the first hypothesis that there is a
stronger association between emotional stability and measures of SWB than
between extraversion and SWB. The zero-order correlations between emotional
stability and the overall subjective wellbeing variable ((subjective wellbeing =
satisfaction with life + (positive affect – negative affect)) were .59 (using data
collected at time point one (T1) and .65 (using data collected at time point two
(T2). These same correlations between extraversion and subjective wellbeing
were .30 (T1) and .24 (T2). When the average correlation across both time points
is squared to give the shared variance, emotional stability explains 38% variance
in subjective wellbeing compared with 7% variance explained by extraversion.
To test the second hypothesis that the association between extraversion and
subjective wellbeing will be substantially reduced when emotional stability is
controlled, Vitterso conducted a series of multiple regression analyses using
subjective wellbeing measured by the Satisfaction With Life Scale (SWLS; Pavot
& Diener, 1993). He found that controlling for extraversion, emotional stability
explained 34% of the average variance in satisfaction with life (SWL) measured
across both time points. This compared with 1% additional variance contributed
by extraversion above emotional stability. Thus, his second hypothesis, that the
association between extraversion and subjective wellbeing will be substantially
reduced when emotional stability is controlled, was supported.
Interestingly, Vitterso seems to ignore the high correlations between his measure
of positive affect (PA) and subjective wellbeing. For example, in his T1 data, the
zero-order correlation between positive affect and subjective wellbeing was .45
and .47 using his T2 data respectively. These same correlations between
emotional stability and subjective wellbeing were .45 (T1) and .47 (T2)
respectively. The implication of this is that Vitterso may have missed an
important additional variable by not including PA as a predictor variable.
The following re-analyses reconstruct Vitterso’s model in AMOS to see how well
his model fits his data. This model will then be expanded by the inclusion of PA.
220
It is hypothesised that in the presence of PA, both extraversion and emotional
stability will make weak, subsidiary contributions to subjective wellbeing.
Analyses will be conducted on questionnaire data Vitterso collected at two
different time points. Throughout this re-analysis, data sets will be referred to as
T1 (data collected at time point one) and T2 (data collected at time point two).
Step 1 of each analysis will re-construct Vitterso’s model where extraversion and
emotional stability will be regressed onto subjective wellbeing. Step 2 of each
analysis will include PA as a predictor SWL.
It is important to note that the re-analysis will be conducted using the overall
subjective wellbeing variable (SWL) as the DV. The reason for this is that the
Vitterso’s measure of subjective wellbeing was computed according to the
tripartite concept of subjective wellbeing ((subjective wellbeing = satisfaction
with life + (positive affect – negative affect)), with the variables transformed into
z-scores before the subjective wellbeing variable was computed. The decision to
use the SWL as the DV is forced because since affect is a component of the DV, it
cannot be used as the IV.
METHODOLOGY
Participants
Data from 264 Norwegian high school students were collected at T1 (August,
1995). 225 participants remained at followed-up in May 1996 (T2). The mean age
of participants at T1 was 19 years (SD = 1.1 years), with an age range of 16-25
years.
7.5
Major Dependent Variable and Other Variables
Overall Life Satisfaction (LS)
Overall life satisfaction was assessed using the Satisfaction With Life Scale
(SWLS; Pavot & Diener, 1993) which consists of five items rated on a 7-point
scale ranging from 1 (‘strongly disagree’) to 7 (‘strongly agree’). An example of
some of these items include: ‘In most ways my life is close to my ideal’ and ‘I am
221
satisfied with my life’. Cronbach’s alphas for the scale were .83 for both T1 and
T2 data. The autocorrelation from T1 to T2 was .62 (p<.001)
Personality
Personality was measured using the Norwegian Big Five Inventory (Engvik,
1993), however, only items measuring extraversion and emotional stability were
analysed in the present investigation. The emotional stability inventory consists of
12 adjectives rated on a 7-point bi-polar scale (e.g., worried-not worried,
contented-discontented and confident-hesitant). Cronbach’s alphas for the 12
items were .80 and .83 at T1 and T2, respectively; and the autocorrelation from
time one to time two was .73 (p<.001). The extraversion inventory also consists of
12 adjectives rated on a 7-point polar bi-polar scale (e.g., not social-social,
outgoing/contented-discontented and confident-hesitant). Cronbach’s alphas for
the 12 items were .80 and .83 at T1 and T2 respectively
Positive Affect
Positive affect was measured with questions about how often, in terms of the
percentage of their time, participants felt the following affective states: joy,
satisfaction and happiness. Responses to these items ranged from 0 (0% of waking
time) to 10 (100% of waking time). Cronbach’s alphas for this scale was .81 (T1)
and .83 (T2). The autocorrelation for this scale from T1 to T2 was .52 (p<.001).
RESULTS
To repeat Vitterso’s model, extraversion and emotional stability will be entered at
step 1 of a step-wise multiple regression analysis. Means, standard deviations and
correlations between variables are presented in Table 80. These values are taken
from Table 1 (p. 908) in the original publication.
222
Table 80:
Means, standard deviations and correlations between variables
T1: N=264
T2: N=225
Variable
Mean
SD
1.
2.
3.
1. ES*
55.00
14.77
-
2. EXT*
62.33
11.05
.32
-
3. SWL*
64.10
13.67
.39
.22
-
4. PA*
60.00
17.90
.49
.42
.45
4.
Mean
SD
1.
60.00 11.35
-
2.
3.
-
65.17 16.67 .23
-
68.23 12.80 .47
.14
-
63.80 16.70 .50
.36
.47
* scores converted to %SM statistic
Note: ES = emotional stability; EXT = extraversion; SWL = satisfaction with life; PA =
positive affect
As shown in Table 80, in the T1 data, SWL shares the strongest relationship with
PA (r = .45), followed by emotional stability (r = .39). In the T2 data, however,
the SWL shares the strongest relationship with PA and emotional stability (r
= .47).
With respect to the assumptions of multiple regression analysis, the sample size is
adequate since the minimum number of cases needed for acceptable statistical
power is 90. Sample sizes also meet the requirements for SEM. Moreover, since
the largest correlation between any two variables are .49 (T1 data) and -.56 (T2
data), there is no risk of collinearity.
The re-analysis of Vitterso’s regression analyses is re-constructed in AMOS to
determine the strength of association between extraversion, emotional stability
and SWL. Figure 16 and Figure 17 below show the before and after addition of
PA for T1 data while Figures 18 and 19 show the same for T2 data.
7.6
4.
Re-constructing Vitterso’s Model (T1)
Vitterso’s model of satisfaction with life is presented in Figure 16 below.
-
223
EXT
.11
.14
SWL
ES
.36***
* p<.05; ** p<.01; *** p<.001
Note: EXT = extraversion; ES = emotional stability; SWL = satisfaction with life
Figure 16: Predicting SWL with personality (Step 1; T1 data)
In the model presented in Figure 16 above, direct paths are drawn between
extraversion, emotional stability and SWL because these variables are believed to
have a direct influence on subjective wellbeing (Vitterso, 2001). The fit statistics
show that χ2 is 28.412 (p = .000) and suggests that this model does not provide a
good fit to the data. A χ2/df value of 28.412 further suggests a generally poor fit of
the model to the data.
Although these fit statistics suggest a generally poor fit, the path coefficients offer
some support for Vitterso’s theory. That is, only emotional stability significantly
influences SWL.
7.7
Re-analysis Incorporating Affect (T1 data)
Step 2 of this analysis, which includes PA as driving the relationship between
variables in Vitterso’s model, is presented in Figure 17 below.
224
.18
.42***
EXT
.24
.01
PA
SWL
.34***
.24
.22***
.49***
ES
* p<.05; ** p<.01; *** p<.001
Note: PA = positive affect; NA = negative affect; EXT = extraversion; ES = emotional
stability; SWL = satisfaction with life
Figure 17: Predicting SWL with PA above personality (Step 2; T1 data)
In the model presented in Figure 17, direct paths are drawn between PA and SWL,
extraversion and emotional stability, because PA is believed to have a direct
influence on these variables. Looking at the fit statistics for the affective model of
SWL and the χ2 was 5.538 (p = .019) and suggests that this model does not
provide a good fit to the data. Furthermore, a χ2/df of 5.538 is greater than 3.0,
suggesting a poor fit in relation to the degrees of freedom.
A comparison of fit statistics for both models is presented in the section below.
7.8
A Comparison of the fit Statistics for the Personality and AffectivelyDriven Models of SWL (T1 data)
To compare model fit statistics for the Personality and Affectively-driven models
of SWL, Table 81 has been constructed.
225
Table 81:
A comparison of model fit statistics (T1 data)
χ2
df
χ2/df
P
AIC
NFI
RMSEA SMC
Vitterso’s
28.412
1
28.412
.000 28.412
.62
.32
.14
Model
Affective
1
.019 23.54
.97
.13
.24
5.538
5.538
Model
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
Fit statistics presented in Table 81 above clearly indicate that while neither model
provides a good fit to the data, the alternative Affective-model of SWL provides a
far better description of how the variables relate to one another. While the χ2 for
both models are greater than 3.0 and significant, the χ2 value for the affective
model is considerably less than that of the Personality-driven model. Furthermore,
the AIC value for the Affective-model is smaller than that of the Personalitydriven model, suggesting that this is a more parsimonious model; and whilst both
RMSEA values are greater than .05, suggesting that neither model provide a close
fit in relation to the degrees of freedom, the RMSEA for the Affective-model is
considerably smaller. In support that the affective model is a better fitting model,
the NFI value for this model was .97, indicating that this model is an excellent
relative fit in relation to the best fitting model. Finally, the Affective-model
explained 10% more variance (SMC) than the Personality-driven model.
7.9
Summary of Model Comparisons (T1)
While the fit statistics suggest that both models do not fit the data particularly well,
the alternative Affective-model model presented in Figure 17 offers a better
description of how the variables relate to one-another and suggests affect may be
driving the relationship between SWL and related variables. In this model, PA is
the strongest predictor of SWL and explains 24% and 18% of variance
respectively in emotional stability and extraversion. Moreover, including PA into
the model increases the total explained variance by 10% and in the presence of PA,
the relationship between emotional stability and SWL reduced significantly (from
226
 = .36, p<.001 to  = .22; z = .175, p<.05). Thus, PA is the strongest predictor of
SWB and appears to be driving a considerable portion of variance between
personality and SWL.
7.10 Re-constructing Vitterso’s Model (T2 Data)
The next analyses will replicate those above using Vitterso’s T2 data. Figure 18
below shows extraversion and emotional stability regressed against SWL (a reconstruction of Vitterso’s model).
EXT
.03
.22
SWL
ES
.46***
* p<.05; ** p<.01; *** p<.001
Note: EXT = extraversion; ES = emotional stability; SWL = satisfaction with life
Figure 18: Predicting SWL with personality (Step 1; T2 data)
The fit statistics show that χ2 is 12.175 (p = .000) and suggests that this model
does not provide a good fit to the data. Furthermore, a χ2/df value of 6.008 further
suggests a generally poor fit. As before, the path coefficients offer support for
Vitterso’s theory.
7.11 Re-analysis Incorporating Affect (T2 data)
Step 2 of this regression, with PA entered as a predictor, is presented in Figure 19
below.
227
.13
.36***
EXT
.33
-.05
PA
SWL
.35***
.25
.33***
ES
.50***
* p<.05; ** p<.01; *** p<.001
Note: PA = positive affect; NA = negative affect; EXT = extraversion; ES = emotional
stability; SWL = satisfaction with life
Figure 19: Predicting SWL with PA above personality (Step 2; T2 data)
Looking at the fit statistics for the Affective-model of SWL and the χ2 was 3.800
(p = .150) this suggests that the model provides an excellent fit to the data. A χ2/df
value of 1.900 is less than 3.0 and further reflects an excellent level of model fit
and model parsimony.
A comparison of fit statistics for both models is presented in the following section.
7.12 A Comparison of the fit Statistics for the Personality and Affectively
Driven Models of SWL (T2 data)
To compare model fit statistics for the Personality and Affectively-driven Models
of SWL, Table 82 has been constructed.
Table 82:
A comparison of model fit statistics (T2 data)
Vitterso’s
Model
Affective
Model
χ2
Df
χ2/df
P
AIC
NFI
12.175
2
6.088
.002
22.175
.82
.22
.22
3.800
2
1.900
.150
19.800
.98
.06
.33
RMSEA SMC
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
228
Fit statistics presented in Table 82 above clearly highlight the alternative
Affective model’s dominance over Vitterso’s Personality-driven model of SWL.
First, the χ2 and χ2/df values for the Affective-model suggest excellent model fit,
whilst those for the Personality-driven model suggest a poor level of model fit.
Furthermore, the AIC value for the Affective-driven model of SWL is below that
of the saturated or best fitting model as provided is AMOS output (AIC = 20.00),
suggesting model parsimony. However, that of the Personality-driven model of
SWL is above that of the saturated or best fitting model (AIC = 12.00), suggesting
that this model lacks parsimony. Also, although the RMSEA for the Affectivemodel is greater than .05, it is considerably lower than that of the Personalitydriven model, suggesting that the Affective-model provides a closer fit in relation
to the degrees of freedom. Finally, the Affective-model explained 11% more
variance than the Personality-driven model.
7.13 Summary of Model Comparisons (T2)
Fit statistics presented in Table 82 provide overwhelming support that the
alternative Affective-model of SWL presented in Figure 19 is a better fitting
model than the Personality-driven model proposed by Vitterso (see Figure 18). PA
proved a stronger predictor of SWL than either emotional stability or extraversion
and was found to explain 25% and 13% of variance in these constructs
respectively. Furthermore, including PA into the model increased the total amount
of explained variance by 11% and in the presence of PA, the relationship between
emotional stability and SWL reduced considerably, although not significantly
(from  = .36, p<.001 to  = .22; z = .163, p=.052). Thus, according to the data,
PA is the strongest predictor of SWB and appears to be driving a considerable
portion of the variance between personality and SWL.
7.14 Summary
A primary aim of these re-analyses were to reconstruct Vitterso’s model in AMOS
to see how well his model fit his data and then compare this model against a
proposed Affective-model of SWL. It was hypothesised that in the presence of PA,
both extraversion and emotional stability will make weak, subsidiary
contributions to SWL. While, emotional stability proved a stronger predictor of
229
SWL than did extraversion, the proposed Affective-model of SWL provided a
better fit to the data than Vitterso’s Personality-driven model of SWL. As
hypothesised, in the presence of PA, unique contributions from extraversion and
emotional stability reduced considerably. Collectively, these results suggest that
PA is a stronger predictor of SWL than is personality and that the alternative
Affective-model of SWL offers a better description of how the variables relate to
one-another than the Personality-driven model.
230
REANALYSIS THREE: VITTERSO AND NILSEN (2002)
Following on from his 2001 paper, Vitterso and Nilsen (2002), sought to
investigate the conceptual structure of subjective wellbeing and to compare the
effect sizes of neuroticism and extraversion as predictors of subjective wellbeing.
In their introduction, these authors focus specifically on extraversion and
neuroticism as the personality traits that are most closely related to subjective
wellbeing. They also point out that, within the literature, extraversion is regarded
as the dominant trait in this regard. However, in a similar vein to the previous
paper, they question this finding and argue that a closer inspection of the literature
reveals neuroticism to be the more important predictor (e.g., DeNeve & Cooper,
1998). In an attempt to shed light on this issue, a major aim of Vitterso and
Nilsen’s study was to corroborate previous findings that neuroticism is the more
important predictor of subjective wellbeing. In the remainder of this text,
subjective wellbeing according to Vitterso and Nilsen’s conceptualisation will be
referred to as subjective wellbeing - Vitterso and Nilsen (SWB-VN).
To test their hypotheses that neuroticism is the more important personality trait
and to explore the associations between extraversion, neuroticism and life
satisfaction, they used SEM. Analyses were conducted with ESQ 5.7 for Windows
(Bentler, 1995). According to their data and consistent with their prediction that
neuroticism will share a more intimate and stronger association with SWB-VN, it
was found that neuroticism explained eight times as much variance as
extraversion (24% vs 3% variance). However, since these constructs are related (r
= .50), these authors demonstrate through model comparisons the importance of
including both extraversion and neuroticism as independent variables in their
model. It is important to note that they found a simple factor structure comprising
neuroticism and extraversion was not supported by model fit statistics and that
considerable modification of the model was needed to provide even a mediocre
goodness-of-fit for the trait model. These authors further concede that a model
consisting of personality and affect provided the best fit to the data. However,
affect does not feature in any of their models as having a direct influence on
subjective wellbeing. Rather, in their model, affect is regarded as an outcome and
not a cause.
231
In summary, Vitterson and Nilsen conclude that their analyses support the notion
of an overall subjective wellbeing construct that comprises the three nested
dimensions of life satisfaction, PA and NA. These authors further argue that
neuroticism, not extraversion, is the more important personality dimension and
that which is driving subjective wellbeing.
The aim of the following re-analysis is to reconstruct Vitterso and Nilsen’s model
to explore how well their model fits their data and to offer an alternative model
based on the notion of core affect. The alternative, Affectively-driven model of
life satisfaction will then be compared to the reconstructed version to determine
which theoretical model provides the best description of how the variables relate
to one another. In this re-analysis, where Vitterso and Nilsen constructed
subjective wellbeing according to the tripartite concept ((subjective wellbeing =
satisfaction with life + (positive affect – negative affect)) as their dependent
variable, the present re-analysis will employ subjective wellbeing as measured
using the Satisfaction With Life Scale (SWLS; Pavot & Diener, 1993) as the
dependent variable. As discussed in the previous re-analysis (Vitterso, 2001),
Vitterso and Nilsen’s measure of SWB cannot be used as a dependent variable
because PA cannot comprise part of the dependent variable as well as be used as
an independent variable. Thus, results of the re-construction analysis are likely to
vary somewhat from the original.
METHODOLOGY
Participants
Participants were 461 individuals representative of the adult population in
northern Norway. The mean age of this sample was 47.8 years (SD = 15.8 years),
with an age range of 19-88 years. 45.3% of participants were female and 54.8%
male. Data were collected from participants by mail in the summer of 2000.
232
7.15 Major Dependent Variable and Other Variables
Satisfaction with Life (SWL)
Satisfaction with life was assessed using the Satisfaction With Life Scale (SWLS;
Pavot & Diener, 1993). Cronbach’s alpha for this scale was .87.
Personality
Neuroticism was measured using the revised NEO personality inventory (NEOPI-R; Costa & McCrae, 1992), with 6 facets, each one comprised of 8 items for a
total of 48 items. These facets and associated Cronbach’s alphas included: 1.
anxiety ( = .76); 2. angry/hostility ( = .68); 3. depression ( = .78); 4. selfconsciousnesses ( = .68); 5. impulsiveness ( = .62); and 6. vulnerability (
= .76).
Extraversion was measured using the NEO-PI-R, which also consists of 6 facets
comprising 8 items for a total of 48 items. These facets and associated Cronbach’s
alphas included: 1. warmth ( = .74); 2. gregariousness ( = .70); 3. assertiveness
( = .73); 4. activity ( = .64); 5. excitement-seeking ( = .60); and 6. positive
emotions ( = .79).
Similarly to the procedure undertaken by Vitterso and Nilsen, composite variables
for extraversion and neuroticism were created for use in the analyses.
Positive Affect
Positive affect was measured by asking participants to indicate how well each of
the following affect terms – ‘happy’, ‘glad’, ‘pleased’ and ‘joyful’ - described the
situation in their own lives. The precise wording of the question is not provided.
Answers were marked on a 7-point rating scale with end-points labeled ‘does not
fit’ (1) and ‘fits very well’ (7). Cronbach’s alpha for this scale was .85.
It is important to note that whilst this measure of PA includes the adjectives happy
and pleased (adjectives representing the pleasant axis of the Circumplex Model of
Affect), this scale is not a measure of core affect per se. For example, this scale
233
asks respondents to describe how well the adjective ‘happy’ describes a situation
in their own lives. In doing so, the respondent’s attention has been directed away
from trait affect to a specific circumstance. Thus, this item is a measure of a
person’s affective experience toward a particular event or situation (state affect)
and is not a true measure of trait mood.
Negative Affect
Negative affect was measured by asking participants to indicate how well each of
the following affect terms – ‘sad’, ‘anxious’, ‘angry’ and ‘worried’ - described the
situation in their own lives. Answers were marked on a 7-point rating scale with
end-points labelled ‘does not fit’ (1) and ‘fits very well’ (7). Cronbach’s alpha for
this scale was .79.
RESULTS
7.16 Vitterso and Nilsen’s Model
The first analysis is a re-construction of Vitterso and Nilsen’s model. The aim of
this re-analysis is to explore model fit statistics for their Personality-driven model
of subjective wellbeing. Means, standard deviations and correlations between
variables are presented in Table 83.
234
Table 83:
Means, standard deviations and correlations between variables (N = 457)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
1.
SWLS*
64.00
14.28
-
2.
PA*
66.67
13.67
.68
-
3.
NA*
35.00
13.54
-.27
-.21
-
4.
Anxiety (N)
9.80
5.70
-.45
-.46
.55
-
5.
Angry/hostility (N)
11.10
4.90
-.23
-.21
.40
.44
-
6.
Depression (N)
13.40
5.90
-.43
-.40
.56
.68
.48
-
7.
Self-consciousness (N)
13.10
5.10
-.28
-.26
.39
.55
.41
.62
-
8.
Impulsiveness (N)
15.10
4.80
-.14
.04
.20
.25
.35
.23
.24
-
9.
Vulnerability (N)
10.60
4.80
-.40
-.35
.41
.62
.42
.61
.57
.36
-
10. Warmth (E)
23.50
4.40
.27
.33
-.20
-.22
-.31
-.22
-.26
-.08
-.24
-
11. Gregariousness (E)
17.80
5.40
.14
.24
-.21
-.22
-.12
-.20
-.23
.17
-.08
.50
-
12. Assertiveness (E)
15.70
5.50
.17
.19
-.23
-.38
-.08
-.41
-.54
.04
-.39
.14
.23
-
13. Activity (E)
20.00
4.50
.18
.27
-.15
-.28
-.04
-.27
-.28
.04
-.31
.34
.19
.46
-
14. Excitement-seeking (E) 14.30
15. Positive-emotions (E)
20.20
5.10
.03
.22
-.01
-.12
.13
-.05
-.05
.30
-.11
.11
.20
.24
.27
-
5.50
.30
.44
-.21
-.24
-.10
-.31
-.26
.27
-.21
.43
.43
.35
.41
.29
15.
-
* scores converted to %SM statistic
Note: SWL = satisfaction with life; PA = positive affect; NA = negative affect; (N ) = neuroticism sub-component; (E) = extraversion sub-component
234
235
In Table 83, it can be seen that the SWL shares the strongest relationship with PA
(r = .68). With respect to the assumptions of structural equation modelling, the
sample size of 475 cases meets this requirement. Furthermore, it is clear from
Table 83 that none of the variables are multi-collinear.
The re-analysis will be conducted using SEM in AMOS to determine how well
Vitterso and Nilsen’s Personality-driven model of satisfaction with life fits their
data. It is important to note that in this model, co-variances were fitted between
error terms for each of the 6 personality facets comprising each measure of
extraversion and neuroticism and these are theoretically justified. For example,
error terms were fitted between interrelated extraversion and neuroticism domains
with a moderate correlation or greater (correlations between .41 and .68, see Table
83). These included: anxiety and angry hostility; anxiety and depression; anxiety
and self-consciousness; anxiety and vulnerability; angry hostility and depression;
angry hostility and self-consciousness; angry hostility and vulnerability;
depression and self-consciousness; depression and vulnerability; depression and
assertiveness; warmth and gregariousness; warmth and positive emotions;
gregariousness and positive emotions; assertiveness and activity; and activity and
positive emotions.
A simplified version of a model similar to Vitterso and Nilsen’s is presented in
Figure 20 below.
236
N
E
-.23**
.65***
.64***
.95***
SWLS
.89
.48
.41
NA
PA
p<.05; ** p<.01; *** p<.001
Note: N = neuroticism; E = extraversion; SWL = satisfaction with life scale; NA =
negative affect; PA = positive affect
Figure 20: A simplified version of Vitterso and Nilsen’s Personality-driven
model of satisfaction with life
In the model presented in Figure 20 above (constructed according to the
theoretical viewpoint of Vitterso and Nilsen) direct paths are drawn between
neuroticism, extraversion and satisfaction with life, because these variables are
believed to have a direct influence on satisfaction with life (p. 96). Furthermore,
direct paths are drawn between neuroticism and NA and between extraversion and
PA, because Vitterso and Nilsen argue that the overall subjective wellbeing
concept comprises the presence of PA and absence of NA (p. 94). Paths linking
PA, NA and SWL are not present because Vitterso and Nilsen consider PA and
NA as outcome variables and variables that have no direct influence on subjective
wellbeing. In their view, personality is driving affect, thus, paths specifying these
relationships have been included in the model.
Looking at the fit statistics for Vitterso and Nilsen’s model and the χ2 was 642.89
(p = .000) and suggests that this model does not provide a good fit to the data. A
χ2/df value of 7.746 further suggests a generally poor fit of the model to the data.
Interestingly, in contrast to Vitterso and Nilsen’s prediction, extraversion, not
neuroticism, is the stronger predictor of satisfaction with life.
237
7.17 Re-analysis Incorporating Affect
A second, Affectively-driven model of satisfaction with life is presented in Figure
21 below.
.05
N
-.22
-.07
.71***
PA
.47
SWL
.56***
E
-.09
.32
p<.05; ** p<.01; *** p<.001
Note: N = neuroticism; E = extraversion; SWL = satisfaction with life; PA = positive
affect
Figure 21: An Affectively-driven model of satisfaction with life
In the model presented in Figure 21 above, direct paths are drawn between PA
and satisfaction with life and between PA and extraversion and PA and
neuriticism, because PA has been found to be driving these constructs (see results
of studies 1 and 2). Direct paths between extraversion, neuroticism and
satisfaction with life are also depicted so as to investigate the influence of
personality on satisfaction with life in the presence of PA. It is important to note
that NA has been omitted from this model because NA does not reflect core affect
according to the manner in which core affect has been conceptualised in this
thesis. Furthermore, neuroticism has been retained because of its significant
influence in the previous model (Figure 20) and because it is a primary aim to
determine the influence of this construct on subjective wellbeing in the presence
of PA. Looking at the fit statistics for the Affective-model of satisfaction with life
and the χ2 was 451.350 (p = .000) and suggests that this model does not provide a
good fit to the data. A χ2/df value of 7.922 further suggests a generally poor fit of
the model to the data.
238
7.18 A Comparison of fit Statistics for the Personality and Affectively-driven
Models of SWL
To compare model fit statistics for the Personality and Affectively-driven models
of SWL, Table 84 has been constructed.
Table 84:
A comparison of model fit statistics
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
Personality681.349
72
9.463
.000
777.349
.76
.14
.48
driven Model
Affective
451.530
57
7.922
.000
547.530
.83
.12
.47
Model
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
As mentioned, the χ2 and χ2/df values for both these models indicate poor levels of
model fit. The AIC values for both models are also greater than those of the
saturated or best fitting models (AIC = 240.00 for the Personality-driven model
and AIC = 210.00 for the Affective model respectively), indicating that both
models lack parsimony. The NFI values for both models are also less than .90 and
suggest that these models are not a close fit in relation to the best fitting models.
RMSEA values of .14 for the Personality-driven model and .12 for the Affectivemodel suggest that these models do not provide a close fit in relation to the
degrees of freedom. Finally, SMC’s of .48 and .47 indicate that 48% and 47% of
variance in SWL is explained by the Personality-driven model and Affectivemodel respectively.
According to fit statistics, both of these models are comparatively poor fitting
models. However, in the model presented in Figure 21 above, it is clear that PA (β
= .71, p<.001) is the strongest predictor of satisfaction with life. In fact, in the
presence of PA, extraversion and neutoticism no longer predict satisfaction with
life, suggesting that the previously observed relationship between extraversion,
239
neutoticism and satisfaction with life appears driven by variance extraversion and
neutoticism share with PA.
7.19 Summary
The aim of Vitterso and Nilsen’s study was to corroborate a growing body of
research (in Vitterso and Nilsen’s view) which suggests that neutoticism, not
extraversion, is the more important predictor of subjective wellbeing. Consistent
with their prediction, neuroticism was found to explain eight times as much
variance as extraversion. However, these authors acknowledge that a simple factor
structure comprising neutoticism and extraversion was not supported by model fit
statistics and that a model consisting of personality and affect provided the best fit
to their data. The current re-analysis aimed to construct a similar model using
subjective wellbeing (as measured by the Satisfaction With Life Scale) as the
dependent variable. An alternative, Affect-driven model was also constructed. It
was found that both models were relatively poor fits to the data. However,
standardised regression-coefficients revealed that PA, not extraversion or
neutoticism, was the greatest single predictor of satisfaction with life. Further, in
the presence of PA, the influence of extraversion and neutoticism on satisfaction
with life became non-significant The implication is that PA, not personality, is the
main driver of satisfaction with life and that previously reported correlations
between personality and this construct should be revisited in the presence of
suitable affective controls.
240
REANALYSIS FOUR: LIBRAN (2006)
The fourth article chosen for re-analysis is the paper by Libran (2006), published
in the Spanish Journal of Psychology. In his introduction, Libran cites a number
of studies in support of subjective wellbeing as a multi-dimensional construct that
comprises both cognitive and emotional elements (e.g., Andrews & Withey, 1976;
Diener, 1984; Pavot, Fugita & Diener, 1997). Libran further argues that since
subjective wellbeing is considered a stable trait, certain personality dimensions
must therefore be related to the experience of happiness. In support of this view,
Libran cites a number of studies proposing a link between personality and
subjective wellbeing (e.g., Costa & McCrae, 1980; Hayes & Joseph, 2003).
Libran also describes studies examining the specific relationship between the
personality dimensions of extraversion and neuroticism and subjective wellbeing
and cites a number of research papers that argue in favour of extraversion as the
more important correlate (e.g., Headey & Wearing, 1989; Hotard, McFatter,
McWhirter, & Steggal, 1989; Lu, 1995). According to Libran, therefore, the
relationship between extraversion and subjective wellbeing is based on the
premise that extraverts are happier because they seem to have better social skills,
are more assertive and are more cooperative (p. 39). However, Libran also
challenges the view that extraversion is the dominant subjective wellbeing trait
and like Vitterso and Nilsen, cites a number of studies (e.g., David, Green, Martin,
& Suls, 1997; DeNeve & Cooper, 1998; Ryan & Frederick, 1997) that suggest
neuroticism (reverse coded as emotional stability) is the more important predictor.
The aim of Libran’s empirical study was to determine the strength of the
relationship between subjective wellbeing and its components. The analysis
involved zero-order correlations and a number of step-wise multiple regression
analyses.
According to the results of Libran’s step-wise multiple regression analyses (p. 41),
neuroticism had a greater influence (β = -.62, p<.000) on subjective wellbeing
((subjective wellbeing = satisfaction with life + (positive affect – negative affect))
and satisfaction with life (SWL, β = -.40, p<.000) than did extraversion (β = .28,
p<.000 and .21, p = .000 respectively). Altogether, neuroticism explained 44%
241
variance in subjective wellbeing, with extraversion contributing a further 8%
variance at step 2. For the prediction of SWL, neuroticism explained 19%
variance, with a further 4% variance contributed by extraversion. Thus, Libran
concludes that neuroticism, not extraversion, is the strongest predictor of
subjective wellbeing and SWL.
While Libran acknowledges the importance of affect to subjective wellbeing (p.
39) and cites a number of studies that support affect as comprising an important
and independent component of subjective wellbeing (e.g., Bradburn, 1969; Costa
& McCrae, 1980), he fails to include affect in his analyses as having a direct
influence on subjective wellbeing. Rather, Libran treats affect as an outcome
variable (i.e., as a product of personality). For example, using stepwise multiple
regression analysis, Libran reports that extraversion explains 25% of variance in
PA (β = .46, p<.000), whilst neuroticism contributes a further 5% (β = -.25,
p<.000). For the prediction NA, Libran reports that neuroticism explains 45%
variance (β = .68, p<.000), with extraversion contributing a further 1% variance (β
= .09, p<.025). Thus, Libran does not examine the direct influence of affect on
subjective wellbeing and satisfaction with life, instead treating these constructs as
outcome variables determined by personality.
A primary aim of this re-analysis is to determine the strength of the relationship
between neuroticism, extraversion and satisfaction with life in the presence of
affect. The second aim is to determine which theoretical model fits the data best the Personality-driven model or the Affectively-driven model. It is hypothesised
that in the presence of affect, the association between personality and satisfaction
with life will reduce considerably and that affect, not personality, is driving
satisfaction with life.
To test these hypotheses, analyses will be conducted using SEM in AMOS. The
first model explores how well Libran’s Personality-driven model of satisfaction
with life fits the data. A second, Affectively-driven model of satisfaction with life
is also explored. It is important to note that similarly to previous re-analyses, the
present re-analyses will be conducted using Pavot & Diener’s (1993) Satisfaction
with Life Scale (SWLS) as the DV. Libran’s alternative measure of subjective
242
was computed according to the tripartite concept of subjective wellbeing. As
previously discussed, there are statistical implications surrounding the use of this
variable.
METHODOLOGY
Participants
Participants were 368 students from various faculties of the University of Rovira i
Virgili of Tarragona (faculties of law, psychology pedagogy, economics, etc.).
214 participants were female and 154 male, with a mean age of 24.2 years (SD=
4.76). Participation in this study was voluntary and anonymous.
7.20 Major Dependent Variable and Other Variables
Overall Life Satisfaction (SWL)
Satisfaction with life was assessed using a translated version (Atienza, Pons,
Balaguer, & Garcia-Merita, 2000) of the Satisfaction With Life Scale (SWLS;
Pavot & Diener, 1993) which consists of five items on a 7-point scale ranging
from 1 (‘strongly disagree’) to 7 (‘strongly agree’). An example of one of these
items includes: ‘In most ways my life is close to my idea’ and ‘I am satisfied with
my life’. The author found that Cronbach’s alpha for this scale was .82.
Personality
Personality was measured using the Spanish version (Aguilar, Tous, & Andres,
1990) of the Eysenck Personality Questionnaire-Revised (EPQ-R; Eysenck,
Eysenck & Barrett, 1985). Altogether, 23 items comprised the measure of
extraversion and 24 items the measure of neuroticism. The scales consisted of a
Yes/No response format. According to Libran, Cronbach’s alpha for the
extraversion scale was .83; whilst that for neuroticism was .82.
Positive and Negative Affect
Positive and Negative affect were measured using the Spanish version (Sandin,
Chorot, Lostao, Joiner, Santed, & Valiente, 1999) of the Positive and Negative
243
Affect Schedule (PANAS; Watson, Clark, & Tellegen, 1988). This inventory
consisted of 20 items – 10 that describe PA and 10 that describe NA. Each group
of descriptors is added separately, providing scores in both scales. Participants
were required to respond on a 5-point Likert type scale, ranging from 1 (nothing
or almost nothing) to 5 (very much) expressing the degree to which they generally
experience the particular feeling or emotion described by the item. Example of
items from the PA scale include happy, active and proud; whilst examples of
items on the NA scale include guilty, nervous and fearful. According to Libran,
Cronbach’s alpha for the PA scale was .87, while that for the NA scale was .85.
It is important to note that this measure of affect taps only those affect terms that
represent activated states. For example, the 10 PA and NA descriptors represent
high poles of PA and NA and not those from either the low PA or NA poles
(Carrol, Yik, Russell, & Feldman-Barrett, 1999). Thus, the PANAS is not a
complete, comprehensive measure of the full spectrum of known affective states.
As a consequence, the relationship between affect and subjective wellbeing may
be considerably weaker than that found between core affect and subjective
wellbeing.
RESULTS
7.21 A Personality-driven Model of Satisfaction With Life
The first analysis will examine how well a Personality-driven model of SWL fits
the data. Means, standard deviations and correlations between variables are
presented in Table 85 below. Correlations between variables are taken from Table
1 (p. 41) in the original publication. Mean scores and standard deviations were
provided by the author through direct correspondence. Where possible, these were
converted into the percentage of scale maximum (%SM) statistic.
244
Table 85:
Means, standard deviations and correlations between variables (N=368)
Variable
1. SWLS*
2. PA*
3. NA*
4. EXT
5. NEU
Mean
44.00
55.75
25.25
15.60
10.80
SD
8.00
10.70
7.32
4.20
4.80
1.
.42
-.35
.28
-.43
2.
3.
4.
5.
6.
.66
-.63
.38
-.66
-.01
.50
-.32
-.02
.67
-.16
-
* scores converted to %SM statistic
Note: SWL = satisfaction with life; PA = positive affect; NA = negative affect; EXT =
extraversion; NEU = neuroticism
In Table 85, it can be seen that SWL shares the strongest relationship with PA (r
= .66), and neuroticism (r = -.66). With respect to the assumptions of multiple
regression, the sample size is adequate since the minimum number of cases
needed for acceptable statistical power is 90. The sample size also meets the
criteria for SEM. Moreover, since the largest correlation between any two
variables used in the analyses is .66, there is no risk of collinearity.
A re-analysis of a Personality-driven model of SWL will be conducted using SEM
in AMOS to determine how well Libran’s model fits his data. A model
constructed according to Libran’s theoretical standpoint is presented in Figure 22
below.
.47***
.28
PA
E
.22
-.25***
.23***
SWL
-.42***
.09*
.47
.68***
N
NA
* p<.05; ** p<.01; *** p<.001
Note: E = extraversion; N = neuroticism; SWL = satisfaction with life; NA = negative
affect; PA = positive affect
Figure 22: A Personality-driven model of satisfaction with life
245
In the model presented in Figure 22, direct paths are drawn between neuroticism,
extraversion and satisfaction with life, because these variables are believed to
have a direct influence on satisfaction with life (p. 39). Furthermore, direct paths
are drawn between neuroticism and NA and between extraversion and PA because
Libran argues that personality predicts affect (p. 39).
Consistent with Libran, neuroticism and extraversion were both significant
predictors of satisfaction with life, PA and NA. Together, these variables
explained 22% variance in satisfaction with life, 28% variance in PA and 47%
variance in NA. These are similar to variances reported by Libran (p. 41).
Looking at the fit statistics for the Personality-driven model and the χ2 was 78.932
(p = .000) and suggests that this model does not provide a good fit to the data. A
χ2/df value of 15.786 further suggests a generally poor fit of the model to the data.
7.22 Re-analysis Incorporating Affect
A second, Affectively-driven model of SWL is presented in Figure 23 below.
.10
N
-.33***
-.32***
.27***
PA
.50***
E
.28
.55***
SWL
.09
.25
p<.05; ** p<.01; *** p<.001
Note: N = neuroticism; E = extraversion; SWL = satisfaction with life; PA = positive
affect
Figure 23: An Affectively-driven model of satisfaction with life
In the model presented in Figure 23, direct paths are drawn between PA and
satisfaction with life, PA and extraversion and between PA and neuroticism,
246
because PA is argued to be driving these constructs. Direct paths between
extraversion, neuroticism and SWL are also depicted so as to investigate the
influence of personality on satisfaction with life in the presence of PA. Similarly
to previous critiques, NA has been omitted from the model because this construct
does not reflect core affect as core affect is defined in this thesis.
In the model, neuroticism and PA significantly predict satisfaction with life.
Together, these variables explain 28% variance and this is 6% more variance than
that explained by the Personality-driven model of SWL. PA was also significantly
related to extraversion – explaining 25% variance in this construct. Looking at
the fit statistics for the Affective-model of satisfaction with life and the χ2
was .089 (p = .956) and suggests that this model provides an excellent fit to the
data. A χ2/df value of .045 further suggests an excellent fit of the model to the
data.
7.23 A Comparison of Model fit Statistics for the Personality and
Affectively-driven Models of SWL
To compare model fit statistics for the Personality-driven and Affectively-driven
Models of SWL, Table 86 has been constructed.
Table 86:
A comparison of model fit statistics
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
Personality78.932
5
15.786 .000
98.932
.85
.20
.22
driven Model
Affective
.089
2
.045
.956
16.089 1.000
.000
.28
Model
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
The Affective-model was the only model to have an AIC value lower than that of
the saturated or best fitting model provided in AMOS output (AIC = 30.00 for the
Personality-driven model and AIC = 20.00 for the Affective model respectively),
247
indicating that the Affective model is more parsimonious. The NFI value for the
Affective model was greater than .90, suggesting that this model provides a close
fit in relation to the best fitting model. In contrast, the NFI value for the
Personality-driven model was less than .90, indicating that this model does not
provide a close fit in relation to the best fitting model. Similarly, the RMSEA
value for the Affective model was less than .05, suggesting a close fit in relation
to the degrees of freedom. On the other hand, this ame value for the Personalitydriven model was greater than .05 and this suggests that these models do not
provide a close fit in relation to the degrees of freedom. Finally, SMC’s of .22
and .28 indicate that 22% and 28% of variance in satisfaction with life can be
explained by the Personality-driven model and Affectively-driven models
respectively. It is also notable that whilst the Affective-model explains 6% more
variance, the χ2 and χ2/df values are also considerably lower than those of the
Affectively-driven model, suggesting that comparatively, this model provides a
much better fit.
It is also interesting to note that in the Affectively-driven model, although
standardised regression coefficients between extraversion and SWL and between
neuroticism and SWL have reduced in the presence of PA (from β = .23 to β =.09,
z = 1.84, p<.05 for extraversion and from β = -.42 to β = -.33; z = -1.29, p>.05 for
neuroticism), neuroticism is the strongest predictor of SWL.
According to Pavot and Diener (1993), the mean SWL score in Libran’s sample
was ‘slightly dissatisfied’. To determine whether this result may be a direct
consequence of a low mean SWL score, the analysis below is conducted using NA
as a predictor. To reiterate, according to homeostasis theory, when set-points are
challenged, PA will loose its affiliation with subjective wellbeing. When this
occurs, subjective wellbeing will be under increasing control from those forces
that threaten homeostasis. For example, negative affect associated with the
particular challenge.
An alternative, Affective-model of satisfaction with life that incorporates NA as a
predictor is presented in Figure 24 below.
248
.55
N
-.31***
-.14*
PA
.30
.32***
.50***
.67***
SWL
-.25***
NA
p<.05; ** p<.01; *** p<.001
Note: N = neuroticism; E = extraversion; SWL = satisfaction with life; PA = positive
affect
Figure 24: An alternate, Affectively-driven model of satisfaction with life (b)
The model presented in Figure 24 is similar to that presented in Figure 23 except
that direct paths are drawn between NA and SWL and between NA and
neuroticism because it is argued that NA may be driving these constructs in this
sample. Also, extraversion has been eliminated from this model because of its
non-significant influence on satisfaction with life in the previous model. Looking
at the standardised regression coefficients and PA, NA and neuroticism all have a
significant influence over satisfaction with life. Together, these variables explain
30% of variance and this is 8% more variance than that explained by the
Personality-driven model presented in Figure 22 and 2% more variance than that
explained by the former Affective-model (Figure 23).
According to model fit statistics, both the χ2 of .070 (p = .966) and the χ2/df value
of .035 suggest an excellent fit of the model to the data.
To compare model fit statistics for all three models of satisfaction with life, Table
87 has been constructed.
249
Table 87:
A comparison of model fit statistics for all three models
Personalitydriven Model
Affective
Model (a)
Affective
Model (b)
χ2
df
χ2/df
P
AIC
NFI
RMSEA
SMC
78.932
5
15.786
.000
98.932
.85
.20
.22
.089
2
.045
.956
16.089
1.000
.000
.28
.070
2
.035
.966
16.070
1.000
.000
.30
As shown in Table 87, fit statistics reveal that the Affectively-driven models
provide a much better fit and explain more variance in satisfaction with life than
the Personality-driven model. In terms of model fit, there is little separating the
two Affectively-driven models of satisfaction with life, with both models
providing an excellent fit to the data. However, the second Affective-model (to be
referred to as Affective model b) explains 2% more variance than the first
Affective model (to be referred to as Affective-model a). Furthermore, referring to
Affective-model (b), the influence of neuroticism on satisfaction with life reduced
considerably when compared to the Personality-driven model presented in Figure
22 (from β = -.42 to β = -.14; z = -4.05, p<.001. Furthermore, PA and NA were
found to predict 55% of the variance in neuroticism. According to these data,
there is sufficient evidence to suggest that a) the relationship between neuroticism
and SWL appears mainly dependent on the relationship neuroticism shares with
NA and b) NA and PA are driving a significant portion of variance in neuroticism.
Finally, when the variance NA and PA share with neuroticism was removed, PA
is the strongest predictor of SWL and this is consistent with subjective wellbeing
homeostasis theory.
7.24 Summary
In summary, Libran argues that personality is an important determinant of
subjective wellbeing, PA and NA and contends that within the literature, there is
much confusion as to which personality trait shares the strongest relationship with
subjective wellbeing. To shed light on this issue, Libran conducted a study
250
examining the effect of extraversion and neuroticism on subjective wellbeing.
Consistent with his predictions, neuroticism was the best predictor of SWL,
explaining 19% variance, with extraversion contributing a further 4% variance.
Furthermore, according to Libran’s analyses, personality explained 30% and 46%
of variance in PA and NA respectively. The re-analysis of Libran’s data using
SEM in AMOS confirmed these results. However, it also found that PA and NA,
not personality, were the underlying drivers of SWL. The model comprising PA
and NA as predictors (see Figure 24) explained 8% more variance than Libran’s
Personality-driven model. Finally, in contrary to Libran’s claim that personality is
an important determinant of SWL, in the presence of PA and NA, neuroticism
made a significant, but very subsidiary contribution, while extraversion had no
effect. The implication of this re-analysis is that Libran’s findings should be reinterpreted in the presence of suitable affective controls.
251
REANALYSIS FIVE: ZHENG, SANG AND LIN (2004)
Zheng, Sang and Lin (2004) conceptualise subjective wellbeing as a construct that
includes life satisfaction, positive emotions and the absence of negative emotions.
They also cite a number of papers that support the link between extraversion,
neuroticism and subjective wellbeing (e.g., Argyle & Lu, 1990; Diener, 1984;
Myers & Diener; 1995). However, Zheng et al., argue that the relationship
between psychoticism and subjective wellbeing has been virtually neglected and
this may be because psychoticism is characterised by ego-centrism (not caring
about others welfare) and is not conductive to the formation of a good relationship
(an important resource of subjective wellbeing according to Myers & Diener,
1995). Thus, an aim of their study was to explore the relationship between
psychoticism and subjective wellbeing.
Zheng et al., also consider the cognitive and social contributions to subjective
wellbeing (see Larsen, Diener, & Croponzano, 1987; Scheier & Carver, 1993;
Seidlitz & Diener, 1993). They describe findings from cross-cultural studies (e.g.,
Diener, Diener, & Diener, 1995) showing that poor nations score low on
individualism and show “average subjective wellbeing scores” (p. 167). On the
other hand, people in wealthier countries score higher on both individualism and
subjective wellbeing. Zheng et al., therefore propose that cognition and social
orientations will make an additional contribution to subjective wellbeing above
personality and that these variables may mediate the relationship between
personality and subjective wellbeing (p. 167).
To test their predictions, Zheng et al., conducted a step-wise multiple regression
analysis with personality entered at step 1 and their cognitive and social variables
(e.g., individualism, collectivism and psychotocism) at step 2. Their sample
comprised 201 participants at the South China Normal University. They found
that neuroticism (β = -.18, p < .05) and extraversion (β = .17, p < .05) together
explained 12% variance in satisfaction with life (Pavot & Diener, 1993). When
entered at step 2, only the ‘present-future orientation’ variable (β = 17, p < .05)
contributed unique variance (a further 5% unique variance) above extraversion
and neuroticism. These results provide some support for their theory and are
252
regarded as evidence that personality can predict subjective wellbeing reliably in
non-western cultures.
The authors conducted an additional path analysis to determine whether a)
personality correlates with subjective wellbeing because of the influence of
cognitive and social orientations or b) personality, cognitive and social
orientations have an independent influence on subjective wellbeing. The results of
these analyses, however, were uninformative. For example, Zheng et al., report
that “cognitive and social orientations are also effective predictors of subjective
wellbeing” and that “they make an additional contribution to the variance of life
satisfaction” (p. 170), however, they provide no empirical evidence through path
analysis to support this claim.
It is notable that they ignore the zero-order correlation of .38 between positive
affect and satisfaction with life, which is the strongest relationship of any two
measured variables.
The following re-analyses will reconstruct the final step of Zheng et al’s., stepwise multiple regression analysis using all their significant predictors. This
reconstructed model, which will be called the Personality-Cognitive and Social
Orientation model of satisfaction with life (P-C-SO model), will then be compared
to an Affectively-driven model of satisfaction with life to determine which
theoretical model explains more variance and is a better fit to the data. It is
hypothesised that a) in the presence of PA, the association between personality,
present-future orientation and satisfaction with life will reduce considerably and b)
the Affective-model of satisfaction with life will fit the data better.
METHODOLOGY
Participants
201 students (90 male and 111 female) from the South China Normal University
in mainland China participated in this study. The age range of these participants
was 17 to 24 years (M = 20.30 years).
253
7.25 Major Dependent Variable and Other Variables
Overall Life Satisfaction (LS)
Life satisfaction was measured using the Satisfaction With Life Scale (SWLS;
Pavot & Diener, 1993) which consists of five items rated on a 7-point scale
ranging from 1 (‘strongly disagree’) to 7 (‘strongly agree’). In the present study,
Cronbach’s alpha for this scale was .76. It is important to note that for
comparability with item scores on the PA and NA scales (see below), data were
transformed by the authors from a 1-7 scale to a 1-9-point scale using the formula
[ item scores on 1-7 scale x 9/7].
Personality
The personality traits of extraversion and neuroticism were measured using the
Revised Eysenck Personality Questionnaire Short Scale for Chinese (EPQ-RSC;
Qian, Wu, Zhu, & Zhang, 2000). This measure yielded four 12-item sub-scale
scores for each dimension. Participants were required to indicate if the statements
are consistent with them by stating ‘yes’ or ‘no’.
Positive Affect
Positive affect was measured using the Positive Affect Scale (Diener, Suh, &
Oishi, 2000). This is a 6-item self-report scale designed to assess positive
emotional experiences. Participants are required to indicate how often they felt six
positive emotions (these six emotions were not specified) in the last week using a
9-point scale (1 = not at all, 9 = all the time). The final PA scale scores were
obtained by summing the scores of each item and dividing by 6. In the present
study, Cronbach’s alpha for this scale was .81.
Negative Affect
Negative affect was measured using the Negative Affect Scale (Diener, et al.,
2000). This is an 8-item self-report scale designed to assess negative emotional
experiences. Participants are required to indicate how often they felt eight
negative emotions (these eight emotions were not specified) in the last week using
254
a 9-point scale (1 = not at all, 9 = all the time). ). The final NA scale scores were
obtained by summing the scores of each item and dividing by 8. In the present
study, Cronbach’s alpha for this scale was .81.
Cognitive and Social Orientations
Cognitive and social orientations were measures using the Cognitive and Social
Orientations Scale (Diener et al., 2000). This is a 13-item self-report scale
developed to assess five kinds of cognitive and social orientations. Some of these
include individualism (e.g., ‘it is important for me that I do my job better than
others’), collectivism (e.g., ‘it is important for me to maintain harmony within my
group’) and present and future orientations on enjoying life (e.g., ‘I would rather
enjoy the present and not worry about the future’). Participants were required to
indicate how strongly they agreed or disagreed with each statement on a 9-point
scale ranging from 1 (strongly disagree) to 9 (strongly agree).
RESULTS
7.26 Zheng et al’s., Model of Subjective Wellbeing
The first analysis examines how well Zheng et al’s., model fits their data. A
second, Affectively-driven model of satisfaction with life will then be investigated
in order to determine which of these two models provides the best description of
how the variables relate to one-another. Means, standard deviations and
correlations between variables are presented in Table 88. These values are taken
from Tables 1 and 2 (p. 169) in the original publication. All variables that will be
used in the following re-analysis have been converted to the %SM statistic.
255
Table 88:
Means, standard deviations and correlations between variables (N = 201)
Variable
Mean
SD
1.
2.
3.
4.
5.
6.
7.
8.
1. SWL*
52.75
17.63
-
2. PA*
51.75
19.14
.38
3. NA*
39.63
15.88 -.26 -.29
4. IND
5.79
1.06
-.10
.31
-.04
5. COL
6.90
1.19
.11
.29
-.01 -.08
6. PFO*
49.00
26.95
.14
.04
-.04
.04
.03
-
7. FAT
2.38
1.82
-.17 -.32
.15
.01
-.12
.09
8. SEC
7.86
1.52
.06
.10
-.06 -.08
.03
-.03 -.38
-
9. EXT
7.66
2.66
.29
.37
-.22
.21
.07
-.07 -.21
.19
10. NEU
5.42
3.13
-.28 -.37
.57
.03
-.04 -.03
9.
10.
-
11. P
3.13
1.68 -.09 -.09 -.02 -.08 -.22
* scores converted to %SM statistic
.03
-
.05
-.02 -.45
-
.00
.01
.05
-.05
Note: SWL = satisfaction with life; PA = positive affect; NA = negative affect; IND =
individualism; COL = collectivism; PFO = present-future orientation; FAT = fatalist
orientation; SEC = self-created orientation; EXT = extraversion; NEU = neuroticism; P =
psychoticism
As shown in Table 88, satisfaction with life shares the strongest relationship with
positive affect (r = .38).
11.
In terms of meeting the assumptions for multiple
regression analysis, according to the formula prescribed by Tabachnick and Fidell
(2001), the minimum number of cases required is 82. With 201 participants in this
study, the sample size requirement for SEM has also been met. In regards to the
issue of normality and singularity and as shown in Table 88, the greatest observed
correlation between any two variables is .38. Thus, this assumption has not been
violated. Zheng et al’s., model of satisfaction with life is presented in Figure 25.
-
256
EXT
.26***
.13
NEU
PFO
-.20**
SWL
.15*
p<.05; ** p<.01; *** p<.001
Note: EXT = extraversion; NEU = neuroticism; PFO = present-future orientation SWL =
satisfaction with life scale
Figure 25: Predicting satisfaction with life with personality and present-future
orientation
In the model presented in Figure 25, direct paths are drawn between extraversion,
neuroticism, present-future orientation and satisfaction with life, because these
variables were unique predictors of satisfaction with life in Zheng et al’s., stepwise multiple regression analysis (see Table III in original paper). Consistent with
Zheng et al., extraversion, neuroticism and present-future orientation all
contributed unique variance in satisfaction with life. Together, these variables
explained 13% variance and this is 4% less variance than the 17% variance
reported by these authors.
Looking at the fit statistics for this model and the χ2 was 11.554 (p = .009) and
suggests that this model does not provide a good fit to the data. Furthermore, a
χ2/df of 3.851 is greater than 3.0, suggesting that this model is not a good fit in
relation to the degrees of freedom.
An alternate, Affectively-driven model of satisfaction with life will now be
investigated.
257
7.27 Re-analysis Incorporating Affect
An affectively-driven model of satisfaction with life is presented in Figure 26
below.
.14
.37***
EXT
.15*
PA
.29***
NEU
-.11
SWL
.17
PF
.14*
p<.05; ** p<.01; *** p<.001
Note: EXT = extraversion; NEU = neuroticism; PF = present-future orientation; PA =
positive affect; SWL = satisfaction with life
Figure 26: An affectively-driven model of satisfaction with Life (a)
The model presented in Figure 26 is similar to that presented in Figure 25,
however, direct paths have been drawn between PA and satisfaction with life and
between PA and extraversion, because PA is believed to be driving these
constructs. According to the results and consistent with SWB homeostasis theory,
PA is the strongest predictor of satisfaction with life in this model. Moreover, in
the presence of PA, standardised regression coefficients between extraversion,
neuroticism, present-future orientation and satisfaction with life reduced
considerably. In fact, in the presence of PA, neuroticism has no significant
influence over satisfaction with life.
Looking at the fit statistics for this model and the χ2 was 60.231 (p = .000) and
suggests that this model does not provide a good fit to the data. Furthermore, a
χ2/df of 12.046 is greater than 3.0, suggesting that this model is not a good fit in
relation to the degrees of freedom
258
To explore whether model fit would improve if neuroticism was removed, a third
analysis was conducted using PA, extraversion and present-future orientation as
predictors. It was decided to remove neuroticism from the analysis because this
variable contributed no unique variance in the former model (see Figure 26). A
path model specified according to this model is presented in Figure 27.
.14
.37***
PA
EXT
.19**
SWL
.19
PF
.14*
* p<.05; ** p<.01; *** p<.001
Note: EXT = extraversion; NEU = neuroticism; PF = present-future orientation; PA =
positive affect; SWL = satisfaction with life scale
Figure 27: An Affectively-driven model of satisfaction with life (b)
As expected, in this model, PA remains as the strongest single predictor of
satisfaction with life. Moreover, extraversion and present-future orientation also
contribute unique variance. Looking at the fit statistics for this model and the χ2
was 1.996 (p = .369) and suggests that this model provides an excellent fit to the
data. Furthermore, a χ2/df of .998 is less than 3.0, suggesting that this model is an
excellent fit in relation to the degrees of freedom.
7.28 A Comparison of the Fit Statistics for all Three Models of Satisfaction
with Life
To compare model fit statistics for Zheng et al’s., model and the two alternative
Affectively-driven models of satisfaction with life, Table 89 has been constructed.
259
Table 89:
A comparison of model fit statistics for all three models
χ2
Df
χ2/df
P
AIC
NFI
RMSEA SMC
Zheng et
11.554
3
3.851
.009 25.554
.73
.12
.13
al’s., Model
Affective
60.231
5
12.046
.000 80.231
.55
.24
.17
Model (a)
Affective
1.996
2
.998
.369 17.996
.97
.000
.19
Model (b)
Note: NFI = Normed Fit Index; AIC = Akaike Information Criterion; RMSEA = Root
Mean Square Error of Approximation; SMC = Squared Multiple Correlation.
Fit statistics presented in Table 89 clearly indicate that the second Affective
model (to be referred to as Affective model b) provided the best description of
how the variables relate to one-another, whilst also explaining the most variance.
For example, while the χ2 for the other two models are greater than 3.0 and
significant, the χ2 value for the Affective-model (b) is below 3.0 and significant,
indicating excellent model fit. Furthermore, the AIC value for this model is the
only AIC value less than that of the saturated, or best fitting model (as provided
by AMOS output), suggesting that this is the most parsimonious model. Also, the
RMSEA value for Affective model (b) is the only RMSEA value less than .05,
suggesting that this model provides the closest fit in relation to the degrees of
freedom. In further support that Affective model (b) provides the greatest
description of how the variables relate to one-another, the NFI value for this
model was .97, indicating that this model is an excellent fit in relation to the best
fitting model. Finally, Affective model (b) explained 6% more variance than
Zheng et al’s, model and 2% more variance than the former Affective model (to
be referred to as Affectivemodel a) that incorporated neuroticism. Thus, evidence
clearly supports that Affective model (b) is the more superior model. However, it
is important to note that the overall amount of explained variance in Affectivemodel (b) is still considerably low. One reason for this may be that the group
mean SWL score is low (see Pavot & Deiner, 1993) and this may suggest that a
higher than normal proportion of people in this group may be at risk of depression.
As discussed previously, these people may be responsible for the failure of affect
260
to explain more of the variance in subjective wellbeing. To reiterate, as
homeostasis becomes dysfunctional under challenge, the determination of
subjective wellbeing progressively shifts from affect to the challenging agent. In
this process, affect will explain less variance subjective wellbeing.
7.29 Summary
The aim of Zheng et al’s., study was to test their hypothesis that cognitive and
social orientations make an additional contribution to subjective wellbeing above
personality. They did this through the use of step-wise multiple regression
analysis. However, there was only minor support for their hypothesis in that
present-future orientation was the only variable to predict satisfaction with life
above extraversion and neuroticism. The aim of the re-analysis was to explore the
validity of Zheng et al’s model using SEM in AMOS. It was hypothesised that a)
in the presence of PA, the association between personality, present-future
orientation and satisfaction with life will reduce considerably and b) that an
Affective model of satisfaction with life would fit the data better. Both of these
hypotheses were supported and it was found that an Affectively-driven model of
satisfaction with life (Affective model b) provided the best fit to the data and
explained the most variance. The implication of these findings is that affect, not
personality or any of the cognitive and social variables, is the strongest predictor
of satisfaction with life. Further, the evidence suggests that affect appears to be
driving a considerable portion of the variance between personality and satisfaction
with life.
261
CHAPTER 8: STUDY 3 DISCUSSION
Studies 1 and 2 determined that SWB is a construct primarily driven by core
affect and not personality. In light of these results, the aim of study 3 was to
demonstrate the importance of affect to research in subjective wellbeing studies
generally. This involved a series of re-analyses of past studies using SEM in
AMOS. Two general hypotheses were tested. These were:
1.
That in the presence of affect, the relationship between personality and
subjective wellbeing would reduce considerably
2.
That Affectively-driven models of subjective wellbeing would explain more
variance and would provide a better fit to the respective data sets than would
Personality-driven models of subjective wellbeing.
Collectively, the results of each of the five papers chosen for re-analysis provide
general support for the hypotheses being tested. For example, a re-analysis of a
paper by Heading and Wearing (1989) revealed that when positive affect was
included alongside variables in their model as a predictor, the overall amount of
explained variance increased by 9%. Further, in the presence of PA, unique
contributions from personality, favorable events and age weakened. Also, as
hypothesised, the Affectively-driven model of subjective wellbeing provided the
best fit to the data. These findings are consistent with those of studies 1 and 2 and
support a growing body of research suggesting that core affect is the main
determinant of subjective wellbeing and may be driving the relationship between
subjective wellbeing and related constructs (e.g., Davern, et al., 2007). Results of
this re-analysis further suggest a re-interpretation of Headey and Wearing’s data
in the presence of suitable affective controls.
Two papers by Vitterso (Vitterso, 2001; Vitterso & Nilsen, 2002) were also reanalysed as part of study 3 and similar results were found. More specifically,
Affectively-driven models of satisfaction with life provided better fits to the data
than the Personality-driven models originally constructed by the authors.
Furthermore, in both of these re-analyses, as hypothesised, in the presence of PA,
262
unique contributions from extraversion and emotional stability (Vitterso, 2001)
and between extraversion and neuroticism (Vitterso & Nilsen, 2002) reduced
considerably. In fact, in the Vitterso and Nilsen (2002) re-analysis, unique
contributions from extraversion and neuroticism were eliminated when variance
attributed to affect was controlled. Collectively, these results suggest that in both
data sets, PA is a stronger predictor of satisfaction with life than personality and
that the Affective-models of satisfaction with life offer better descriptions of how
the variables relate to one-another than do the Personality-driven models. Thus,
the results of these re-analyses provide further support that subjective wellbeing is
a construct driven by affect and not personality as is commonly agreed within the
literature (e.g., Brickman & Campbell, 1971; DeNeeve & Cooper, 1998; Emmons
& Diener, 1985; Headey & Wearing, 1989, 1992; Vitterso, 2001; Vitterso &
Nilsen, 2002).
The final two papers chosen for re-analysis were Libran (2006) and Zheng, Sang,
and Quin (2004). What is interesting about these two studies is that both studies
used the Satisfaction with Life Scale (Pavot & Diener, 1993) as a measure of
subjective wellbeing and the mean score for subjective wellbeing in both these
samples are considered low. For example, according to Pavot and Diener, a score
of 20 (approximately 57.00 percentage points) represents the neutral point of the
scale – the point at which the respondent is about equally satisfied as dissatisfied.
Mean satisfaction with life scores of 44 points (Libran, 2006) and 52.75 points
(Zheng, et al., 2004) indicate that people in both these samples were on average,
‘slightly dissatisfied’ with their lives. According to homeostasis theory, as
homeostasis becomes dysfunctional under threat, core affect may no longer reflect
subjective wellbeing. Instead, the experience of subjective wellbeing will reflect
the loss of positive affect as control over subjective wellbeing is assumed by
cognitions and affects associated with the challenging agent. Homeostasis theory
predicts that when this occurs, subjective wellbeing becomes less predictable and
a majority of variance remains to be explained. Consistent with homeostasis
theory, in Libran’s sample, PA and NA contributed a majority of the 30%
explained variance in satisfaction with life. Furthermore, using Zheng et al’s., data,
PA contributed a majority of the 19% variance explained. As predicted, in these
challenged populations, a considerable amount of variance is unexplained and it is
263
suspected that this is so because positive affect has lost some of its affiliation with
subjective wellbeing.
Collectively, the results of these re-analyses clearly support affect as the
underlying driver of subjective wellbeing. In all five of the re-analyses,
Affectively-driven models of subjective wellbeing provided better fits to the data
and explained comparatively more variances than the Personality-driven models
of subjective wellbeing. Further, results of all five re-analyses demonstrated that
in the presence of affect, the relationship between SWB and related variables
reduced considerably from those variances reported in the respective original
publications.
The finding that affect was a more consistent and dominant predictor of subjective
wellbeing is strengthened by the fact that all of these studies used relatively
inferior measures of affect. For example, Libran (2006) used the PANAS (Watson,
Clark, & Tellegen, 1988) to measure positive and negative affect. Of great
importance to this discussion is knowledge that the PANAS only measures high
activated affective states. More specifically, affects representing low PA or NA
poles are not represented in this scale. The implication of this is that the PANAS
very poorly represents core affect as defined in the thesis. For example, the
adjective ‘happy’, which studies 1 and 2 determined as one of the strongest
predictors of SWB and an essential component of core affect, does not feature in
this scale. Thus, the PANAS is a relatively poor measure of affect and will not
accurately reflect the affective component of subjective wellbeing. As a
consequence, the PANAS and measures of subjective wellbeing will appear less
related. This is precisely what has been observed.
In summary, the results of study 3 highlight the importance of treating affect as a
causal variable and not just as an outcome variable. Furthermore, it is clear from
these results that researchers should employ contemporary measures of affect
which represent the essential elements of core affect. In so doing, these measures
will adequately reflect the underlying and intimate relationship that exists between
affect and subjective wellbeing.
264
CHAPTER 9: SUMMARY AND OVERVIEW
Theoretical Overview and Hypotheses
This thesis concerns the conceptual structure of subjective wellbeing and has
attempted to determine the relative affective/cognitive component of this
construct. The investigation comprised three linked studies and the overall finding
is that subjective wellbeing is a construct primarily driven by core affect, with
minor independent contribution from cognition. The basis of this work is as
follows.
Within the literature, it is generally agreed that subjective wellbeing is a broad
construct that comprises global judgments of life satisfaction, domain-based
satisfactions and people’s emotional responses (Diener, Suh, Lucas, & Smith,
1999). As such, subjective wellbeing comprises both a cognitive and affective
component (e.g., Diener & Diener, 1996; Steel & Ones, 2002). Despite this
agreement, however, the relationship between the component parts of cognition,
affect and SWB is not well understood. There is also a lack of consensus within
the literature regarding what is driving subjective wellbeing. Some researchers
(e.g., Headey & Wearing, 1989, 1992; Vitterso, 2001; Vitterso & Nilsen, 2002),
argue that individual differences in subjective wellbeing can be explained by
individual differences in personality. However, Davern, et al., (2007) have
provided evidence that subjective wellbeing is essentially driven by core affect,
with cognitive discrepancies playing an important but subsidiary role.
Core
affect,
as
initially defined
by Russell
(2003),
refers
to
that
“neurophysiological state consciously accessible as the simplest raw (nonreflective) feelings evident in moods and emotion” (Russell, 2003; p. 148). In this
seminal paper, Davern et al., found that six affect terms contributed significant
unique variance to the prediction of satisfaction with life as a whole (LS). Further,
five of these affects (representing the pleasant-unpleasant axis and activationdeactivation axis of the Circumplex Model of Affect) were found to explain 64%
of the variance in SWB measured using the PWI. From these studies, Davern et
al., placed their findings into the context of ‘core affect’ and proposed a small
265
group of unique affect predictors as a measure of this construct. Finally, using
structural equation modeling, these authors explored the relative strength of core
affect (content, happy and excited) in three separate models of subjective
wellbeing that incorporated cognition (MDT; Michalos, 1985) and all five factors
of personality (NEO Personality Inventory; Costa & McCrae, 1992). Based on
their findings, Davern et al., concluded that the Affective-Cognitive model
provided the best fit to the data, explaining 90 percent of the variance in SWB. In
this model, core affect is proposed to have a direct influence on SWB, personality
and MDT. Further, direct paths were also specified between personality and SWB
and between MDT and SWB to account for any residual variance in SWB that
may be explained by these variables (See Figure 28 below).
A primary aim of this thesis was to replicate these findings. It was hypothesised
that adjectives located at the pleasantness-pole of the Circumplex Model of Affect
would explain significant variance in LS and SWB. Further, it was also
hypothesised that an Affective-Cognitive model for SWB would offer a better
description of how the variables relate to one another than would a Personalitydriven model for SWB.
Another major aim of this thesis was to test a number of theoretical predictions
based on SWB homeostasis theory. Analogous to the homeostatic maintenance of
body temperature, this theory posits that SWB is actively controlled and
maintained (Cummins & Nistico, 2002). According to homeostasis theory, under
normal, unthreatening conditions, SWB is controlled by the homeostatic system.
As a consequence, SWB will be steadily maintained within the normal range and
core affect will be driving SWB. However, conditions of prolonged and aversive
threat can challenge personal beliefs, thus threatening SWB set-points. When
external circumstances defeat the homeostatic SWB set-point, the homeostatic
system will forfeit control to the external agent, resulting in a drop in SWB. Under
these circumstances, the contribution of core affect to SWB will decrease as the
measure of SWB increasingly represents affect generated by the challenging agent.
Further, homeostasis theory predicts that the buffer variables (perceived control,
self-esteem and optimism), will be activated in challenged populations in an
attempt to maintain and stabilise SWB within the set-point range. Based on this
266
theory, it was hypothesised that the buffer variables would explain significant
variance in SWB above core affect in a group of people experiencing homeostatic
challenge.
Homeostasis theory also predicts that core affect should explain more variance in
LS than in SWB due to the differing levels of abstraction and responses to the
domains containing more cognition. Further, homeostasis theory posits that, when
the system is functioning normally, core affect will explain more variance in LS
than in SWB. However, as homeostasis becomes dysfunctional under challenge,
the determination of both LS and SWB progressively shifts from core affect to the
challenging agent. Thus, the amount of variance explained in LS and SWB will
vary depending on whether the system is functioning normally or under challenge.
Another important issue to be addressed concerns the nature of the relationship
between SWB and related variables. These include self-esteem, optimism,
perceived control, perceived discrepancy judgements (as measured using MDT)
and personality. Within the literature, these constructs are regarded as independent
entities and generally share a moderate correlation with SWB. However, in this
thesis, the relationship between these constructs and SWB was revisited in the
presence of core affect. It was argued that core affect is the cohesive force that is a
considerable portion of variance in each of these variables and is the common
element causing them to correlate with each other. In light of this theory, it was
hypothesised that core affect is driving the relationship between SWB and these
related variables.
Finally, there has been no previous attempt to determine the normative
characteristics of SWB for adolescents. Thus, another aim of this thesis was to
address this important issue and determine those domains that are most important
to high-school students.
Methodology
To test the hypotheses, data were collected from Australian high-school students
in the Melbourne metropolitan region, country Victoria and Geelong. Altogether,
267
146 students (in study 1) and 205 students (in study 2) participated. Participants in
both studies represented students in years 7-12. Their mean age was 14.4 years (in
study 1) and 16.7 years (in study 2). As part of these studies, students were
required to complete a paper and pencil questionnaire measuring SWB, core affect,
self-esteem, perceived control, optimism, personality and perceived multiple
discrepancies (MDT).
Results and Implications
Normative Data
Consistent with Australian adult normative statistics, studies 1 and 2 determined
that the mean scores for SWB (73.90% in study 1) and (73.61% in study 2) were
just within the Australian adult normative range of 73.43 to 76.43%SM (Cummins
et al., 2006). However, these mean scores were at the lower end of the normal
range and homeostasis theory predicts that, as a group mean falls towards the
bottom of the normal range, it indicates that the group contains a higher than
normal proportion of people who are at risk of depression. This is important
information since it indicates that some of these people will be exhibiting a loss of
homeostatic control and consequently their SWB will be below their set-pointrange. This is relevant for a number of the hypotheses being tested as the pattern
of results will differ depending on whether the homeostatic system is resting or
under threat/defeat.
Core Affect
Consistent with Davern et al., who found that SWB is primarily driven by affects
representing the pleasantness-unpleasantness axes and activation-deactivation
axes of the Circumplex Model of Affect, study 1 and study 2 determined that
three affects (happy, content and alert) explained 59% and 57% of the variability
in SWB respectively. These three affects and their combination are considered to
represent core affect. Thus, results corroborate previous research suggesting that
SWB is highly affective in nature and that SWB is comprised mainly of pleasant
affect, with some activation.
268
Homeostasis Theory
Homeostasis theory asserts that, since domains contain more cognition than LS,
core affect should explain more variance in LS than SWB due to the differing
levels of abstraction. In confirmation of this, core affect explained more unique
variance in global life satisfaction (LS) than the PWI. However, this effect was
only observed when the sample was restricted to people scoring in the operational
range for homeostasis of >50 points on a 0-100 point scale. This demonstrates that,
as homeostasis becomes dysfunctional under challenge, the determination of both
LS and SWB progressively shifts from core affect to the challenging agent. In this
process, core affect explains less unique variance in both constructs. Moreover,
the reduction in explained variance is somewhat greater in LS because it is more
abstract and therefore rather more determined by core affect in the first place.
A second major prediction from homeostasis theory is that the buffer variables
(perceived control, self-esteem and optimism), will be activated in challenged
samples in an attempt to maintain and stabilise SWB within the set-point range. In
contrast to the hypothesis, however, this was not observed. When data from
studies 1 and 2 were combined, secondary control was the only buffer variable to
contribute additional unique variance above core affect in the homeostatically
challenged situation. The implication of this finding is that secondary control may
be relatively more independent of core affect than self-esteem and optimism.
The Relationship Between SWB and Related Variables
Another important issue addressed in studies 1 and 2 is the nature of the
relationship between SWB and related variables including self-esteem, optimism,
perceived control, perceived discrepancy judgments (as measured using MDT)
and personality. Further testing supported the notion that core affect also appears
to be driving the relationship between SWB and these constructs. For example,
when variance shared with core affect was controlled, the strength of relationship
between these variables and SWB reduced considerably. The implication is that
core affect, not personality, is the major cohesive force underpinning these
relationships. Further, these findings suggest that a large body of research which
supports a relationship between SWB and related constructs should be re-
269
examined in the presence of suitable affective controls. A model specifying the
relationships described is specified in Figure 28.
Personality
Buffer Variables
SWB
Core Affect
MDT
Figure 28: An Affective-model for SWB demonstrating the proposed influence of
core affect on SWB and related variables
In Figure 28, a direct path is drawn between core affect and SWB because core
affect is argued to have a direct influence over SWB. Direct paths are also
specified between core affect and personality, the buffer variables and MDT,
because core affect is believed to be driving these constructs. Finally, paths
between personality, the buffer variables, MDT and SWB have also been
specified to represent residual variance in SWB contributed by these variables that
is not accountable by core affect.
Collectively, the major implication of these results is that core affect is the driving
force behind SWB and may also be responsible for set-point stability in SWB
ratings in Cummins’ homeostatic theory of SWB. Furthermore, the data suggest
that core affect may also be driving the relationship between SWB and related
variables. Additionally, since core affect is driving both personality and SWB,
individual differences in set-point levels of core affect may be causing personality
and SWB to correlate.
270
Model Testing Using Structural Equation Modelling in AMOS
Another major aim of studies 1 and 2 was to explore Davern et al’s., finding that
an Affective-Cognitive model for SWB provided the best fit to the data.
Consistent with their findings, structural equation modeling revealed that the
Affective-Cognitive model was the best fitting model, explaining 80% of the
variance. Also, in contrast to a large body of research which shows a strong link
between extraversion, neuroticism and SWB (e.g., Brickman & Campbell, 1971;
DeNeeve & Cooper, 1998; Emmons & Diener, 1985; Headey & Wearing, 1989,
1992; Vitterso, 2001; Vitterso, 2002), model fit statistics indicated a relatively
poor fit for the Personality-driven model of SWB. The implication of these
findings is that core affect, not personality, is the main driver of SWB and that the
vast majority of these relationships are caused by the common element of core
affect.
School Satisfaction as a Unique Construct
Given that high-school students spend a majority of their week at school (at least
during the school term), it seemed intuitive to include an item measuring
satisfaction with this domain. Further, previous research supports an association
between aspects of school life and subjective wellbeing (e.g., Baker, 1999;
Huebner, Gilman, & Laughlin, 1999; Suk-Un & Moon, 2006; Suldo & Huebner,
2004). Thus, another aim of this thesis was to explore whether a single item
measuring satisfaction with school would fulfill the criteria for a new domain
within the PWI-SC. To be included as a new domain, any new domain must
contribute independently to the prediction of LS.
Combined data from both studies demonstrated that satisfaction with school
explained unique variance in global life satisfaction (LS) above all 7 existing
domains - qualifying satisfaction with school as a unique construct. This is
evidence that satisfaction with school should be considered as an eighth domain
on a future revision of the PWI-SC. Using the combined data set, all domains
except satisfaction with relationships, community connectiveness and satisfaction
with future security were significant unique predictors. Interestingly, an anomaly
within the two samples is that in adult populations, satisfaction with relationships
271
consistently ranks in the top few domains in terms of its unique contribution to LS,
whereas in the adolescent sample, it did not contribute any unique variance.
However, a comparison of this item in the adult and adolescent form of the
questionnaires revealed that differential wording of the items may account for this
difference.
In summary, satisfaction with school explained unique variance in LS above all 7
existing domains. This qualifies satisfaction with school as a unique construct.
Further, from these data, clear adult/adolescent differences in domain satisfactions
have been identified. The implication is that parents, teachers and the wider
community need to be aware of the unique and changing needs of adolescents so
that appropriate support, understanding and assistance can be offered to those in
need.
Re-analyses of Past Research Papers
In light of the results from studies 1 and 2, the aim of study 3 was to demonstrate
the importance of affect to research in subjective wellbeing studies generally. This
involved a series of re-analyses of past studies using SEM to replicate the original
results and to examine change when core affect was introduced into the model.
These re-analyses clearly support affect as the underlying driver of subjective
wellbeing. In all five re-analyses, Affectively-driven models of subjective
wellbeing provided better fits to the data and explained comparatively more
variances than Personality-driven models of subjective wellbeing. Further, results
of all five re-analyses demonstrated that in the presence of affect, the relationship
between SWB and related variables reduced considerably from that reported in
the original publications. This finding is striking given that the majority of these
studies used inferior measures of affect. For example, Libran (2006) used the
PANAS (Watson, Clark, & Tellegen, 1988) to measure positive and negative
affect. Of great importance to this discussion is the fact that the PANAS only
measures those affect terms that represent high activated states. As a consequence,
affect and subjective wellbeing will appear less related. It is suggested that had
272
these studies used a measure of core affect similar to that used in studies 1 and 2,
the predictive power of affect would be considerably greater.
Collectively, the results of study 3 highlight a) the importance of treating affect as
a causal variable and not just as an outcome variable and b) a need to employ core
affect as a co-variate when investigating the relationship between other variables
related to SWB.
Directions for Future Research
This thesis offers a number of insights and directions for future research into
subjective wellbeing. First, the results highlight a need to understand the sample
characteristics. For example, the mean scores for SWB in studies 1 and 2 were
both at the lower end of the Australian adult normative range. This means that the
present sample most likely comprises a greater than average number of people
who may be experiencing homeostatic defeat and who may be at risk for
depression. The implication of this has been demonstrated, such that the
normative and defeated sub-samples behave differently and in accordance with
homeostatic theory.
A second goal for future research, especially those studies basing their hypothesis
on SWB homeostasis theory, is to gain a more comprehensive understanding of
how the buffer variables operate. According to theory, all three buffer variables
(perceived control, optimism and self-esteem) should be activated during
challenging times in an attempt to stabilise and maintain SWB within the normal
set-point range. However, the combined data revealed that only secondary control
made a unique contribution in a group of these challenged individuals. It is
important that a future study investigate this issue further so that necessary
revisions to theory are considered.
Finally, more research is needed to quantify the normative characteristics of SWB
in adolescents. In this thesis, time and sampling constraints restricted the number
and demographic variability of the participants. If a future study can recruit a
more diverse and representative sample of high-school students, the relationship
273
between age, gender, year level at school and SWB can be more precisely
determined. This will also assist in the identification of ‘at-risk’ groups and enable
appropriate resources to be allocated to high-school students in need of greater
support, assistance and understanding.
Conclusion
In summary, the results presented in this thesis generally conform to the
homeostatic model. They are also consistent with those of Davern et al., in
suggesting that SWB is a construct driven by core affect. Further, this research
suggests that core affect is the cohesive force that causes significant relationships
between SWB and related variables. Through the process of natural selection, it is
possible that humans have evolved to experience a level of core affect that is
stable and positive. Not only does core affect provide the motivation for living, it
also represents the SWB set-point that homeostasis is defending.
274
REFERENCES
Abbey, A., & Andrews, F.M. (1985). Modelling the psychological determinants
of life quality. Social Indicators Research, 16, 1-16.
Abramson, L.Y., Seligman, M.E.P., & Teasdale, J.D. (1978). Learned
helplessness in humans: Critique and reformulation. Journal of Abnormal
Psychology, 87, 49-74.
Affleck, G., Tennen, H., & Gershman, K. (1985). Cognitive adaptations to highrisk infants: The search for mastery, meaning, and protection from future
harm. American Journal of Mental Deficiency, 89, 653-656.
Aguilar, A., Tous, J.M., & Andres, A. (1990). Adaptacion y estudio psicometrico
del EPQ-R. Anuario de Psicologia, 46, 101-118.
AMOS 7.0. (2006). Smallwaters Corp: Chicago, Il.
Andrews, F. M., & Robinson, J.P. (1991). Measures of subjective-wellbeing, In
Measures of personality and social psychological attitudes. (pp. 61-114).
Robinson, J.P., Phillip, R., Wrightsman, L.S. San Diego, CA. Academic Press.
Andrews, F. M., & Withey, S. B. (1976). Social indicators of wellbeing:
Americans' perceptions of life quality. New York: Plenum Press.
Argyle, M. (1987). The psychology of happiness. London: Metheun.
Argyle, M., & Lu, L. (1990). The happiness of extraverts. Personality and
individual differences, 11, 1011-1017.
Aspinwall, L.G., & Taylor, S. E. (1992). Modeling cognitive adaptation: A
longitudinal investigation of the impact of individual differences and coping
on college adjustment and performance. Journal of Personality and Social
Psychology, 63(6), 989-1003.
Atienza, F.L., Pons, D., Balaguer, I., & Garcia Merita, M.L. (2000). Propiedades
psicometricas de la Escala de Satisfaccion con la Vida en adolescentes.
Psicothema, 12, 331-336.
Atkinson, T. (1982). The stability and validity of QOL measures. Social
Indicators Research, 10, 113-133.
275
Averill, J.R. (1973). Personal control over aversive stimuli and its relationship to
stress. Psychological Bulletin, 80(4), 286-303.
Baker, J. (1999). Teacher-student interaction in urban at-risk classrooms:
Differential behavior, relationship quality, and student satisfaction with
school. The Elementary School Journal, 100(1), 57-70.
Blascovich, J., & Tomaka, J. (1991). Measures of self-esteem. In, J.P. Robinson,
P.R. Shaver, L.S. Wrightsman (Eds.). Measures of personality and social
psychological attitudes. San Diego, CA: Academic Press.
Block, J., & Robbins, R.W. (1993). A longitudinal study of consistency and
change in self-esteem from early adolescence to early adulthood. Child
Development, 64, 909-923.
Blore, J.D. (2008). Subjective wellbeing: An assessment of competing theories.
Thesis. Deakin University, Melbourne.
Bradburn, N.M. (1969). The structure of psychological wellbeing. Chicago:
Aldine.
Brickman, P., & Campbell, D.T. (1971). Hedonic relativism and planning the
good society. In M.H Appley (Ed.). Adaptation-level theory (pp. 287-305).
New York: Academic Press.
Brickman, P., Coates, D., & Janoff-Bulman, R. (1978). Lottery winners and
accident victims: Is happiness relative? Journal of Personality and Social
Psychology, 36(8), 917-927.
Browne, M. W. (1992). Circumplex models for correlational matrices.
Psychometrika, 57(4), 469-497.
Campbell, A. (1981). The sense of wellbeing in America: Recent patterns and
trends. New York, NY: McGraw-Hill.
Campbell, A., Converse, P. E., & Rodgers, W. L. (1976). The quality of American
life: Perceptions, evaluations and satisfactions. New York: Russell Sage
Foundation.
Carroll, J. M., Yik, M. S. M., Russell, J. A., & Feldman-Barrett, L. (1999). On the
psychometric principles of affect. Review of General Psychology, 3(1), 14-22.
276
Chang, E.C., Maydeu-Olivares, A., & D'Zurilla, T. (1997). Optimism and
pessimism as partially independent constructs: Relationship to positive and
negative affectivity and psychological wellbeing. Personality and Individual
Differences, 23(3), 433-440.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation for the
behavioural sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
Cohen, J., Cohen, P., West, S.G., & Aiken, L.S. (2003). Applied multiple
regression /correlation for the behavioural sciences (3nd ed.). New York:
Academic Press.
Costa, P. T., Jr., & McCrae, R. R. (1980). Influence of extraversion and
neuroticism on subjective wellbeing. Journal of Personality and Social
Psychology, 38, 668-678.
Costa, P. T., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEOPI-R) and NEO Five-Factor Inventory (NEO-FFI) professional manual.
Odessa, Florida: Psychological Assessment Resources.
Cropanzano, R., Weiss, H.M., Hale, J.M.S., & Reb, J. (2003). The structure of
affect: reconsidering the relationship between negative and positive affectivity.
Journal of Management, 29(6), 831-857.
Cummins, R. A. (1995). On the trail of the gold standard for subjective wellbeing.
Social Indicators Research, 35, 179-200.
Cummins, R. A. (1996). The domains of life satisfaction: an attempt to order
chaos. Social Indicators Research, 38, 303-328.
Cummins, R. A. (1998). The second approximation to an international standard
for life satisfaction. Social Indicators Research, 43, 307-334.
Cummins, R. A. (2000a). Normative life satisfaction: measurement issues and a
homeostatic model. In B. Zumbo (Ed.), Methodological developments and
issues in quality of life research. Amsterdam: Kluwar.
Cummins, R. A. (2000b). Personal income and subjective wellbeing: a review.
Journal of Happiness Studies, 1, 133-158.
277
Cummins, R. A. (2000c). Objective and subjective quality of life: An interactive
model. Social Indicators Research, 52, 55-72.
Cummins, R. A. (2003). Normative life satisfaction: measurement issues and a
homeostatic model. Social Indicators Research, 64, 225-256.
Cummins, R. A., Davern, M., Okerstrom, E., Lo, S. K., & Eckersley, R., (2005)
Australian Unity Wellbeing Index: Special report on city and country living.
Retrieved February 20, 2006, from Deakin University, Australian Centre on
Quality of Life Website:
http://www.deakin.edu.au/research/acqol/index_wellbeing/Survey_12.1.pdf
Cummins, R. A., Eckersley, R., Pallant, J., Van Vugt, J., & Misajon, R. (2003).
Developing a national index of subjective wellbeing: The Australian Unity
Wellbeing Index. Social Indicators Research, 64, 159-190.
Cummins, R. A., Gullone, E., & Lau, A. L. D. (2002). A model of subjective
wellbeing homeostasis:
The role of personality. In E. Gullone, R. A.
Cummins, R.A., & A. L. D. Lau (Eds.), The universality of subjective
wellbeing indicators.
Social Indicators Research Book Series (2002).
Dordrecht: Kluwer.
Cummins, R.A., Lau, A.D.L. (2005). Personal Wellbeing Index - School Children
(PWI-SC). (3rd Ed.). Melbourne: Deakin University.
Cummins, R.A., Lau, A.D.L., & Davern, M. (unpublished). Homeostatic
mechanisms and quality of life. In K.C. Land, (Ed.), Handbook of social
indicators
and
quality-of-life
studies.
Volume
I:
Theoretical
and
Methodological Foundations. New York: Springer.
Cummins, R.A., Lau, A.D.L., Mellor, D., & Stokes, M.A. (in press). Encouraging
governments to enhance the happiness of their nation: Step 1: Understanding
subjective wellbeing. Social Indicators Research.
Cummins, R. A., & Nistico, H. (2002). Maintaining life satisfaction: The role of
positive bias. Journal of Happiness Studies, 3, 37-69.
278
Cummins, R.A., Woerner, J., Tomyn, A., Gibson, Knapp, T. (2005). The
Australian Unity Wellbeing Index: Survey 14. The Wellbeing of Australians –
Personal relationships. Melbourne: Australian Centre on Quality of Life,
School of Psychology, Deakin University.
Cummins, R.A., Woerner, J., Tomyn, A., Gibson, A., & Knapp, T. (2006). The
Australian Unity Wellbeing Index: Survey 16. The wellbeing of Australians –
mortgage payments and home ownership. Melbourne: Australian Centre on
Quality of Life, School of Psychology, Deakin University.
Davern, M. (2004). Subjective wellbeing as an affective construct. Thesis. Deakin
University, Melbourne.
Davern, M., Cummins, R.A., & Stokes, M. (2007). Subjective wellbeing as an
affective-cognitive construct. Journal of Happiness Studies, 8(4) (in press).
David, J.P., Green, P.J., Martin, R., & Suls, J. (1997). Differential roles of
neuroticism, extraversion, and event desirability for mood in daily life. An
integrative model of top-down and bottom-up influences. Journal of
Personality and Social Psychology, 73, 149-159.
Dember, W.N., Brooks, J. (1989). A new instrument for measuring optimism and
pessimism: Testetest reliability and relations with happiness and religious
commitment. Bulletin of the Psychonomic Society, 27(4), 365-366.
DeNeve, K. M. (1999). Happy as an extraverted clam? The role of personality for
subjective wellbeing. Current Dissertations in Psychological Science, 8(5),
141-144.
DeNeve, K. M., & Cooper, H. (1998). The happy personality: A meta-analysis of
137 personality traits and subjective wellbeing. Psychological Bulletin, 124,
197-229.
Diener, E. (1984). Subjective wellbeing. Psychological Bulletin, 95, 542-575.
Diener, E. (2000). Subjective wellbeing: The science of happiness and a proposal
for a national index. American Psychologist, 55(1), 34-43.
Diener, E., & Diener, M. (1995). Cross-cultural correlates of life satisfaction and
self-esteem. Journal of Personality and Social Psychology, 68(4), 653-663.
279
Diener, E., & Diener, M. (1996). Most people are happy. Psychological Science,
7(3), 181-185.
Diener, E., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The Satisfaction
With Life Scale. Journal of Personality Assessment, 49(1), 71-75.
Diener, E., & Larsen, R. J. (1993). The experience of emotional wellbeing. In M.
Lewis, & J.M. Haviland, Handbook of emotions. New York: Guildford.
Diener, E., Lucas, R. E., Oishi, S., & Suh, E. M. (2002). Looking up and looking
down: weighting good and bad information in life satisfaction judgements.
Journal of Personality and Social Psychology, 28(4), 437-445.
Diener, E., Napa Scollon, C., & Lucas, R. E. (2004). The evolving concept of
subjective wellbeing: the multifaceted nature of happiness. Advances in Cell
Aging and Gerontology, 15, 187-219.
Diener, E., Napa Scollon, C., Oishi, S., Dzokoto, V., & Suh, E.M. (2000).
Positivity and the construction of life satisfaction judgements: Global
happiness is not the sum of its parts. Journal of Happiness Studies, 1(2), 159176.
Diener, E., Suh, E. M., Lucas, R. E., & Smith, H. L. (1999). Subjective wellbeing:
three decades of progress. Psychological Bulletin, 125, 276-302.
Diener, E., Suh, E. M., & Oishi, S. (2000). International College Survey 2001.
Unpublished questionnaires, University of Illinois at Urbana-Champaign.
Dimberg, U. (1982). Facial reactions to facial expressions. Psychophysiology, 33,
S34.
Eid, M., & Diener, E. (2004). Global judgements of subjective wellbeing:
Situational variability and long-term stability. Social Indicators Research,
65(3), 245-277.
Emmons, R.A., & Diener, E. (1985). Personality correlates of subjective
wellbeing. Personality and Social Psychology Bulletin, 11(1), 89-97.
Engvik, H. (1993). "Big Five". Tidsskrift for Norsk Psykologforening, 30(9), 884896.
280
Eysenck, H.J., & Eysenck, S.B.G. (1964). Manual of the Eysenck Personality
Inventory. London: Stoughton.
Eysenck, S.B.G., Eysenck, H.J., & Barrett, P. (1985). A revised version of the
Psychotocism Scale. Personality and Individual Differences, 6, 21-29.
Fabrigar, L. R., Visser, P. S., & Browne, M. W. (1997). Conceptual and
methodological issues in testing the circumplex structure of data in
personality and social psychology. Personality and Social Psychology Review,
1, 184-203.
Feldman, L. A. (1995a). Valence focus and arousal focus: Individual differences
in the structure of affective experience. Journal of Personality and Social
Psychology, 69(1), 153-166.
Feldman, L. A. (1995b). Variations in the circumplex structure of mood.
Personality and Social Psychology Bulletin, 21(8), 806-817.
Folkman, S. (1984). Personal control and stress and coping processes: A
theoretical analysis. Journal of Personality and Social Psychology, 46(4),
839-852.
Gable, S.L., Reis, H.T., & Elliott, A.J. (2000). Behavioural activation and
inhibition in everyday life. Journal of Personality and Social Psychology, 78,
1135-1149.
Garson, G.D. (1998, 2008). Structural equation modelling. Retrieved March 31,
2007 from: http://www2.chass.ncsu.edu/garson/pa765/structur.htm
Gosling, S.D., Rentfrow, P.J., Swann, W.B. Jr., (2003). A very brief measure of
the Big-Five personality domains. Journal of Research in Personality, 37(6),
504-528.
Guttman, L.A. (1954). A new approach to factor analysis: The radex. In P.R.
Lazarsfeld (Ed.). Mathematical thinking in the social sciences. Glencoe, Ill:
Free Press.
Hanestad, B.R., & Albrektsen, G. (1992). The stability of quality of life
experience in people with Type 1 diabetes over a period of a year. Journal of
Advances Nursing, 17, 777-784.
281
Hayes, N., & Joseph, S. (2003). Big 5 correlates of three measures of subjective
wellbeing. Personality and Individual Differences, 34, 723-727.
Headey, B., Holmstrom, E., & Wearing, A. (1985). Models of wellbeing and illbeing. Social Indicators research, 17, 211-234.
Headey, B., & Wearing, A. (1989). Personality, life events, and subjective
wellbeing: toward a dynamic equilibrium model. Journal of Personality and
Social Psychology, 57(4), 731-739.
Headey, B., & Wearing, A. (1992). Understanding happiness: A theory of
subjective wellbeing. Longman Cheshire, Melbourne.
Helson, H. (1965). Adaptation Level Theory: An experimental and systematic
approach to behaviour. New York, NY: Harper & Row.
Henderson, A.S., Bryne, D.G., & Duncan-Jones, P. (1981). Neurosis and the
social environment. New York: Academic Press.
Holloway, J. (2003). Neighbourhood subjective wellbeing and the role of
perceived control in maintaining wellbeing. Unpublished honours thesis,
Deakin University, Burwood, Australia.
Hottard, S.R., McFatter, R., McWhirter, R., & Stegall, M. (1989). Interactive
effects of extraversion, neuroticism, and social relationship on subjective
wellbeing. Journal of Personality and Social Psychology, 57, 321-331.
Howell, D.C. (2004). Fundamental statistics for the behavioural sciences (5th ed.).
CA: Wadsworth
Huebner, E. S. (1991). Initial development of the Student’s Life Satisfaction Scale,
School Psychology International, 12, 231–40.
Huebner, E. S. (1994). Preliminary development and validation of a
multidimensional
life
satisfaction
scale
for
children.
Psychological
Assessment, 6, 149–58.
Huebner, E.S., Gilman, R.., & Laughlin, J.E. (1999). A multimethod investigation
of the multidimensionality of children’s wellbeing reports: Discriminant
validity of life satisfaction and self-esteem, Social Indicators Research, 46, 1–
22.
282
Huelsman, T. J., Furr, R.M., & Nemanick, R. C. (2003). Measurement of
dispositional affect: Construct validity and convergence with a circumplex
model of affect. Educational and Psychological Measurement, 63(4), 655-673.
Huelsman, T. J., Nemanick, R. C., & Munz, D. C. (1998). Scales to measure four
dimensions of dispositional mood: positive energy, tiredness, negative
activation, and relaxation. Educational and Psychological Measurement,
58(5), 804-819.
Hunskaar, S., & Vinsnes, A. (1991). The quality of life in women with urinary
incontinence as measured by the Sickness Impact Profile. Journal of the
American Geriatrics Society, 39(4), 378-382.
International Wellbeing Group. (2005). Personal Wellbeing Index - Adult (PWIA).
International Wellbeing Group. (2006). Personal Wellbeing Index - Adult (PWIA).
Kid Power. Retrieved July 20, 2006 from: www.kidpower.org/School-age.html
Kline, R.B. (1998). Principles and practice of structural equation modelling. New
York: Guilford Press.
Landua, D. (1992). An attempt to classify satisfaction changes: Methodological
and content aspects of a longitudinal problem. Social Indicators Research,
26(3), 221-241.
Larsen, R. J., & Diener, E. (1992). Promises and problems with the circumplex
model of emotion. In M. S. Clark (Ed.), Review of Personality and Social
Psychology: Emotion (Vol. 13, pp. 25-59). Newbury Park, CA: Sage.
Larsen, R. J., Diener, E., & Cropanzano, R.S. (1987). Cognitive operations
associated with individual differences in affect intensity. Journal of
Personality and Social Psychology, 53, 767-774.
Lazarus, R.S. (1966). Psychological stress and the coping process. New York,
NY, US: McGraw-Hill.
Lazarus, R.S. (1981). The costs and benefits of denial. In S. Breznits (Ed.), Denial
and Stress. Ney York: International Universities Press.
283
Libran, E.C. (2006). Personality dimensions and subjective wellbeing. The
Spanish Journal of Psychology, 9(1), 38-44.
Lingoes, J.C., (1973). Guttman-Lingoes nonmetric program series. Ann Arbor,
Mich: Mathesis.
Loehlin, J. C. (1987, 1992). Latent variable models: An introduction to factor,
path, and structural analysis. (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
Lovibond, P. F., & Lovibond, S. H. (1995). The structure of negative emotional
states: Comparison of the Depression Anxiety Stress Scales (DASS) with the
Beck Depression and Anxiety Inventories. Behaviour Research and Therapy,
33(3), 335-343.
Lu, L.A. (1995). The relationship between subjective wellbeing and psychosocial
variables in Taiwan. Journal of Social Psychology, 135, 351-357.
Lucas, R.E., Clark, A.E., Georgellis, Y., & Diener, E. (2003). Reexaming
adaptation and the set point model of happiness: Reactions to changes in
martial status. Journal of Personality and Social Psychology, 84(3), 527-539.
Lucas, R. E., Diener, E., & Suh, E. M. (1996). Discriminant validity of wellbeing
measures. Journal of Personality and Social Psychology, 71(3), 616-628.
Lykken, D.T., & Tellegen, A. (1996). Happiness is a stochastic phenomenon.
Psychological Science, 7(3), 186-189.
Maslow, A. H. (1965). Motivation and personality (2nd ed.). New York: Harper
and Row.
McDougal, W. (1908). An introduction into social psychology (Metheum,
London).
Mehrabian, A., & Russell, J. A. (1974). An approach to environmental psychology.
Cambridge, Mass: MIT Press.
Michalos, A. C. (1985). Multiple discrepancies theory (MDT). Social Indicators
Research, 16, 347-413.
Michalos, A. C. (1991). Global report on student wellbeing, Vol. II: Family,
friends, living partner, and self-esteem. New York, NY, US: Springer-Verlag
Publishing.
284
Murphy, S.T., & Zajonc, R.B. (1987). Affect and awareness: Comparisons of
subliminal and supraliminal affective priming. Paper presented at the 95th
Annual Convention of the American Psychological Association, New York,
NY.
Murphy, S.T., & Zajonc, R.B. (1993). Affect, cognition and awareness: Affective
priming with suboptimal and optimal stimulus. Journal of Personality and
Social Psychology, 64, 723-739.
Myers, D.G. (2000). The funds, friends, and faith of happy people. American
Psychologist, 55, 56-67.
Myers, D.G., & Diener, E. (1995). Who is happy? Psychological Science, 6(1),
10-19.
Ng, Y.K. (1997). A case for happiness, cardinalsim, and interpersonal
comparability. The Economic Journal, 107, 1848-1858.
Norman, W.T. (1963). Towards an adequate taxonomy of personality attributes.
Journal of Abnormal and Social Pychology, 66, 574-583.
Oatley, K., & Johnson-Laird, P.N. (1987). Towards a cognitive theory of
emotions. Cognition and Emotion, 1, 29-50.
O'Brien, W.H., VanEgeren, L., & Mumby, P.B. (1995). Predicting health
behaviors using measures of optimism and perceived risk. Health Values:
The Journal of Health Behavior, Education & Promotion, 19(1), 21-28.
Olason, D.T., & Roger, D. Optimism, pessimism and 'fighting spirit': A new
approach to assessing expectancy and adaptation. Personality and Individual
Differences, 31(5), 755-768.
O'Malley, P.M., & Bachman, J.G. (1979). Self-esteem and education: Sex and
cohort comparisons among high school seniors. Journal of Personality and
Social Psychology, 37(7), 1153-1159.
Pallant, J. (2001), SPSS Survival Manual. Sydney: Allen & Unwin.
Pavot, W., & Diener, E. (1993). Review of the satisfaction with life scale.
Psychological Assessment, 5(2), 164-172.
285
Pavot, W., Fujita, F., & Diener, E. (1997). The relation between self-aspect
congruence, personality and subjective wellbeing. Personality and Individual
Differences, 22, 183-191.
Penland, E.A., Masten, W.G., Zelhart, P., Fournet, G.P., & Callahan, T.A. (2000).
Possible selves, depression and coping skills in university students.
Personality and Individual Differences, 29(5), 963-969.
Peterson, C. (1999). Personal control and wellbeing. In: Wellbeing: The
foundations of hedonic psychology. Kahneman, D., Diener, E., & Schwarz, N.
New York, NY, US: Russell Sage Foundation.
Peterson, C. (2000). The future of optimism. American Psychologist, 55(1), 44-55.
Peterson, C., Maier, S.F., & Seligman, M.E.P. (1993). Learned helplessness: A
theory for the age of personal control. New York: Oxford University Press.
Piers, E.V. (1986). The Piers-Harris Children’s Self-Concept Scale, Revised
Manual. Los Angeles, CA: Western Psychological Services.
Pyszczynski, T., Holt, K., & Greenberg, J. (1987). Depression, self-focused
attention, and expectancies for positive and negative future life events for self
and others. Journal of Personality and Social Psychology, 52(5), 994-1001.
Qian, M., Wu, G., Zhu, Y., & Zhang, X. (2000). Development of the revised
Eysenck personality questionnaire short scale for Chinese (EPQ-RSC). Acta
Psychologica Sinica, 32, 317-323.
Quit Victoria. Retrieved July 20, 2006 from: www.quit.org.au
Remington, N. A., Fabrigar, L. R., & Visser, P. S. (2000). Re-examining the
circumplex model of affect. Journal of Personality and Social Psychology,
79(2), 286-300.
Remondet, J.H., & Hansson, R.O. (1991). Job-related threats to control among
older employees. Journal of Social Issues, 47(4), 129-141.
Rosenberg, M. (1965). Society and the adolescent self-image. New Jersey:
Princeton University Press.
Rosenberg, M. (1989). Society and the adolescent self-image (rev. ed.).
286
Rothbaum, F., Weisz, J.R., & Snyder, S.S. (1982). Changing the world and
changing the self: A two-process model of perceived control. Journal of
Personality and Social Psychology, 42, 5-27.
Rotter, J.B. (1966). Generalised expectancies for internal versus external control
of reinforcement. Psychological Monographs, 80, 1-28.
Runciman, W. G. (1966). Relative deprivation and social justice. Berkeley:
University of California Press.
Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and
Social Psychology, 39(6), 1161-1178.
Russell, J. A. (2003). Core affect and the psychological construction of emotion.
Psychological Review, 1, 145.
Ryan, R.M., & Frederick, C. (1997). On energy, personality, and health:
Subjective vitality as a dynamic reflection of wellbeing. Journal of
Personality, 65, 529-565.
Sandin, B., Chorot, P., Lostao, L., Joiner, T., Santed, M., & Valiente, R. (1999).
Escalas PANAS de afecto positivo y negativo: Validacion factorial y
convergencia transcultural. Psicothema, 11, 37-55.
Scheier, M. F., & Carver, C. S. (1985). Optimism, coping, and health: Assessment
and implications of generalized outcome expectancies. Health Psychology,
4(3), 219-247.
Scheier, M. F., & Carver, C. S. (1993). On the power of positive thinking: The
benefits of being optimistic. Current Directions in Psychological Science, 2,
26-30.
Scheier, M. F., Carver, C. S., & Bridges, M. W. (1994). Distinguishing optimism
from neuroticism (and trait anxiety, self-mastery, and self-esteem): A
reevaluation of the Life Orientation Test. Journal of Personality and Social
Psychology, 67(6), 1063-1078.
Schimmack, U., Diener, E., & Oishi, S. (2002). Life-satisfaction is a momentary
judgement and a stable personality characteristic: Use of chronically
accessible and stable sources. Journal of Personality, 70(3), 345-384.
287
Schlosberg, H. (1952). The description of facial expressions in terms of two
dimensions. Journal of Experimental Psychology, 44, 229-237.
Schwarz, N., & Strack, F. (1999). Reports of subjective wellbeing: Judgemental
processes and their methodological implications, In D. Kahneman, E. Diener,
& N. Schwarz. (Eds.). Wellbeing: The foundations of hedonic psychology.
New York, NY: Russell Sage Foundation.
Schwarz, N., & Strack, F., Kommer, D., & Wagner, D. (1987). Soccer, rooms,
and the quality of your life: Mood effects on judgements of satisfaction with
life in general and with specific domains. European Journal of Social
Psychology, 17, 69-79.
Seidlitz, L., & Diener, E. (1993). Memory for positive versus negative life events:
Theories for the differences between happy and unhappy persons. Journal of
Personality and Social Psychology, 64, 654-664.
SPSS for Windows: Release 12.0. (2003). Chicago: SPSS Inc.
Steel, P., & Ones, D.S. (2002). Personality and Happiness: A national level
analysis. Journal of Personality and Social Psychology, 83(3), 767-781.
Suh, E., Diener, E., & Fugita, F. (1996). Events and subjective wellbeing: Only
recent events matter. Journal of Personality and Social Psychology, 70(5),
1091-1101.
Suk-Un, Jin., & Moon, Sidney, M. (2006). A study of wellbeing and school
satisfaction among academically talented students attending a science high
school in Korea. Gifted Child Quarterly, 50(2), 169-184.
Suldo, S. M., & Huebner, E. S. (2004). The role of self-efficacy in adolescents
subjective wellbeing: poster session presented at the International Positive
Psychology Summit, Washington, DC, 1 October 2004.
Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics (4th ed.).
MA: Allyn & Bacon.
Thompson, S.C. (1981). Will it hurt less if I cannot control it? A complex answer
to a simple question. Psychological Bulletin, 90, 89-101.
288
Thompson, S.C., & Spacapan, S. (1991). Perceptions of control in vulnerable
populations. Journal of Social Issues, 47(4), 1-21.
Thompson, S.C., Thomas, C., Rickabaraugh, C.A., Tantamjarik, P., Otsuki, T.,
Pan, D., Garica, B.F., & Sinar, E. (1998). Primary and secondary control over
age-related changes in physical appearance. Journal of Personality, 66(4),
583-605.
Veenhoven, R. (1991). 'Is happiness relative'? Social Indicators Research, 24(1),
1-34.
Veenhoven, R. (1994). 'Is happiness a trait'? Test of the theory that a better society
does not make people any happier'. Social Indicators Research, 32, 101-160.
Vitterso, J. (2001). Personality traits and subjective wellbeing: Emotional stability,
not extraversion, is probably the important predictor. Personality and
Individual Differences, 31(6), 903-914.
Vitterso, J., & Nilsen, F. (2002). The conceptual and relational structure of
subjective wellbeing, neuroticism, and extraversion: once again, neuroticism
is the important predictor of happiness. Social Indicators Research, 57, 89118.
Watson, D., Clark, L., & Tellegen, A. (1988). Cross-cultural convergence in the
structure of mood: A Japanese replication and comparison with U.S. finding.
Journal of Personality and Social Psychology, 47, 127-144.
Watson, D., & Tellegen, A. (1985). Toward a consensual structure of mood.
Psychological Bulletin, 98(2), 219-235.
Weisz, J. R., Rothbaum, F.M., & Blackburn, T.C. (1984). Standing out and
standing in: The psychology of control in America and Japan. American
Psychologist, 39(9), 955-969.
Wenholt, R. (1975). Map Motivatieleer. Department of Social Sciences. Erasmus
University Rotterdam.
Woodworth, R. S., & Schlosberg, H. (1938). Experimental Psychology (3rd Ed.).
London: Methuen & Co.
289
Wundt, W. (1924). An introduction to psychology (R. Pinter, Trans.). London:
Allen & Unwin. (Original work published 1912).
Yik, M. S. M., Russell, J. A., & Feldman Barrett, L. (1999). Structure of selfreported current affect: integration and beyond. Journal of Personality and
Social Psychology, 77(3), 600-619.
Zautra, A., & Reich, J. (1980). Positive life events and reports of wellbeing: Some
useful distinctions. American Journal of Community Psychology, 8(6), 657670.
Zheng, X., Sang, D., & Lin, Q. (2004). Personality, cognitive and social
orientations and subjective wellbeing among Chinese students. Australian
Psychologist, 39(2), 166-171.
290
APPENDIX A:
YOUNG AUSTRALIAN WELLBEING INDEX
(STUDY 1 QUESTIONNAIRE)
291
292
293
APPENDIX B:
YOUNG AUSTRALIAN WELLBEING INDEX
(STUDY 2 QUESTIONNAIRE)
294
295
296
THE END