Socioeconomic status and the allocation of Government resources in Australia:
How well do geographic measures perform?
Patrick Lim1, Sinan Gemici, John Rice & Tom Karmel
National Centre for Vocational Education Research
Level 11, 33 King William Street, ADELAIDE, SOUTH AUSTRALIA 5000
Ph: + 61 8 8230 8400, Fax: +61 8 8212 3436
Patrick.Lim@ncver.edu.au, Sinan.Gemici@ncver.edu.au, John.Rice@ncver.edu.au,
Tom.Karmel@ncver.edu.au
1
Corresponding Author
1
Socioeconomic status and the allocation of Government resources in Australia:
How well do geographic measures perform?
Abstract
Purpose – The aim of this paper is to compare the performance of area-based vs.
individual-level measures of socioeconomic status (SES).
Methodology - Using data from the Longitudinal Surveys of Australian Youth (LSAY),
a multidimensional measure of individual SES is created. This individual measure is used
to benchmark the relative usefulness of Socio-Economic Indexes for Areas (SEIFA), a
geographic set of measures often used in Australia to assess the SES of individuals. Both
measures are compared in terms of classification bias. The effects of using the different
SES measures on participation in post-compulsory education are examined.
Findings – SEIFA measures perform satisfactorily with regard to the aggregate
measurement of SES. However, they perform poorly when their use is aimed at
channelling resources toward disadvantaged individuals. It is at the individual level, then,
that our analysis reveals the shortcomings of area-based SES measures.
Research limitations/implications – While region based measures are relatively easy
to collect and utilise, we suggest that they hide significant SES heterogeneity within
regional districts. Hence, the misclassification resulting from the use of regional measures
to direct support for low SES groups creates a risk for resource misallocations.
Originality/value – Our finding that region-based measures are subject to significant
misclassification has important research and policy implications. Given the increasing
availability of individual-level administrative data, we suggest that such data be used as a
substitute for geographic SES measures in categorising the SES of individuals.
Keywords - Socio-economic status, young people, tertiary education, LSAY.
Paper type – Research paper
2
1. Introduction
The persistent social and economic marginalisation of individuals and groups within
society has significant detrimental direct and indirect impacts. Such marginalisation tends
to create a ‘vicious cycle’ of disadvantage, limiting access to educational opportunities,
which in turn leads to poor labour market outcomes and low earnings. Many
governments, including Australia’s, have the policy aim of breaking this cycle of
economic and social marginalisation by targeting educational assistance programs to
lower SES groups.
Policy initiatives aimed at increasing the social inclusion levels of those of low SES rely,
to a great degree, on accurate measures of SES. SES, while very real, is an abstract
notion. Determining the specific criteria by which the SES of individuals, families and
groups is established can be a complex and nuanced task. While access to financial or
other tangible resources (in the form of income or assets) is of fundamental importance,
SES is also influenced by the interaction, moderation and mediation of a variety of other
social and economic determinants.
One easy to determine, and hence frequently-used, approach is based on an individual’s
geographic location. Although geographic measures, such as the Socio-Economic
Indexes for Areas (SEIFA; Australian Bureau of Statistics [ABS] 2008), are widely used
to report the SES status of groups, these measures have been criticised for their poor
construct validity in determining SES at the individual level (Coelli 2010; Jones 2001). In
essence, the imprecision of the measure arises due to significant ‘within region’ variance
in SES within geographical areas. When assistance is targeted to individuals or groups
based on geographical measures alone, assistance may well flow to individuals of
substantial economic and social means. This results in a sub-optimal mechanism for
resource allocation.
This paper proceeds as follows. We begin with a brief discussion of variables that should
be included in an ideal measure of SES. We then introduce four area-based SES
measures that are frequently used in the Australian context. Subsequently, we use data
from the Longitudinal Surveys of Australian Youth (LSAY) to create an individual
measure of SES. We use our newly-created SES measure as a reference against which we
assess the performance of the area based measures in terms of classification bias (i.e., the
degree to which different SES measures correctly identify a respondent’s SES). We
conclude by examining the effects of using the different SES measures on participation
in post-compulsory education.
2. Approaches to Measuring SES
SES has been defined as a
short-hand expression for variables that enable the placement of persons, families,
households and aggregates such as Statistical Local Areas (SLA’s), communities and cities
into some hierarchical order, reflecting their ability to produce and consume the scarce
and valued resources of society. (Western 1998, p. 5)
Occupational status, education, income and wealth, as well as general access to social and
cultural resources all contribute to SES (Bourdieu 1973; Coleman 1987). Health, family
structure, social contacts, and migration history are additional have also been shown to
be covariates of an individual’s socioeconomic position (Pantazis 2006; Saunders 2007;
Scutella, Wilkins & Horn 2009).
3
Most young people are in some way affected directly or indirectly by parental influences.
Therefore, SES measures for children and youth should include variables based on
parental attributes. Such attributes usually include one or more dimension of parental
occupation, parental educational attainment, household income or wealth, and access to
societal and cultural resources.
2.1
Area-based Measures
Area-based measures of SES determine socioeconomic position by considering the
average economic profile of a given geographic location, and at the individual level
attributing this measure to all of its residents. In Australia, region-specific data is derived
from the five-yearly Census of Population and Housing. This data is aggregated into the
SEIFA, comprising four separate indexes that each represent a slightly different
approach to measuring SES.
1. Index of Relative Socioeconomic Disadvantage (IRSD): This index summarises 17
area-based variables, including low income, low education, as well as high rates of
unemployment and unskilled occupations. A low score reflects relative
socioeconomic disadvantage.
2. Index of Relative Socioeconomic Advantage and Disadvantage (IRSAD): This index
features 21 area-based variables, including low or high income, internet connection,
occupation and education. A high score reflects relative socioeconomic advantage.
3. Index of Economic Resources (IER): This index contains 15 area-based variables,
including items such as household income, housing expenditure, and wealth. A high
score indicates socioeconomic economic advantage.
4. Index of Education and Occupation (IEO): This index is composed of 9 area-based
variables, including educational attainment, enrolment in further education,
occupational information (such as skill level), and unemployment status. Higher
scores indicate socioeconomic advantage.
2.2
Parental Occupation
Parental occupation has been shown to be a strong covariate of the SES of children and
young people. Various measures exist for the measurement of parental occupation.
Internationally, the International Standard Classification of Occupations (ISCO,
International Labour Organisation 1990) is commonly used to categorise occupations. In
Australia, occupations are classified using the Australian and New Zealand Standard
Classification of Occupations (ANZSCO, ABS 2009). Both of these scales can be
converted to a continuous scale. ISCO can be converted to ISEI1 (Ganzeboom, De
Graaf, & Treima, 1992), while ANZSCO can be converted to AUSEI06 (McMillan,
Jones, & Beavis 2008). These conversions convert the categorical classifications to a
continuous scale of ‘occupational prestige’ based on the conversion of education,
earnings, social standings, and other variables to a score.
Apart from which occupational classification scheme to use, the question of which
parent’s occupation to measure has also to be decided. Three common options are
outlined below:
The ISEI is a continuous measure of occupational prestige that converts the International Standard Classification of Occupations
into a scale that ranges from 0 to 90.
1
4
1. Focus on father’s occupation only: this approach assumes that the adult male in the
household has the strongest attachment to the labour force. However, in the modern
labour market this approach probably underestimates family SES, as females
nowadays routinely make significant contributions to household income.
2. Focus on the higher-status occupation: this approach assumes that the adult with the higherstatus occupation determines the family’s overall socioeconomic position.
3. Focus on father’s occupation or, if missing or unknown, the mother’s occupation: this approach is
useful in that it helps to overcome missing values that may arise due to male
detachment from the labour force. Moreover, it is likely that young people are able to
identify the occupation of at least one of their parents or parent figures.
In this paper, we use the third option and map information on parental occupation to the
continuous ISEI scale.
2.3
Parental Education
Parental educational attainment is another important element of an accurate SES
measure for young people. In Australia, educational attainment is often classified
according to the Australian Qualifications Framework (AQF; AQF Advisory Board
2007) or the Australian Standard Classification of Education (ASCED; ABS 2001). An
alternative approach to measuring educational attainment is to focus on the length of
formal education. However, in the Australian context substantial differences exist in the
duration of qualifications, particularly for vocational education and training (VET)
courses.
Internationally, educational attainment is usually classified using the International
Standard Classification of Education (ISCED, UNESCO 1997). ISCED facilitates the
comparison of education statistics and indicators within and between countries. Table 1
provides an overview of ISCED classifications.
Table 1
ISCED Classification
ISCED Level
Qualification Level
0
Pre-primary, kindergarten, pre-school
1
Primary
2A/B
Certificate I and II (general enabling, bridging courses)
2C
Certificate I and II (basic vocational)
3A/B
Higher school certificate, university enabling courses, AQF certificate
III
3C
AQF statement of attainment
4A/B
Certificate IV
5A
Bachelor, bachelor with honours, master (research and coursework)
5B
Diploma, advanced diploma, graduate certificate, graduate diploma
6
PhD, professional doctorate
Similar to the measurement of occupation categories previously discussed, considering
parental education raises the question of which parent’s educational attainment to
measure. Three common options are outlined below:
5
1. Focus on mother’s education only: this approach is based on an argument of nurture versus
nature. Traditionally, mothers provide guidance in child rearing, and a mother who
values education is likely to instil this value in her child.
2. Focus on the higher level of education: this approach assumes that the adult with the higher
educational attainment level exerts a leading influence over the family’s overall
socioeconomic position. Problems arise when the contribution of the adult with the
lower level of educational attainment to the family’s socioeconomic position is
underestimated.
3. Focus on the mother’s education or, if missing or unknown, the father’s education: this approach
helps to alleviate the problem of missing values because it is likely that young people
are able to identify the educational attainment level of at least one of their parents or
parent figures. It also includes single-parent households.
In this paper, we use the third option and map information on parental education to the
ISCED scale.
2.4
Proxies for Income and Wealth
Household income and wealth are routinely used as indicators of SES because they
represent a direct measure of access to economic resources. Surveys such as the
Household, Income and Labour Dynamics of Australia (HILDA, Department of
Families, Housing, Community Services and Indigenous Affairs 2011) survey and the
Census of the Australian Population ask individuals to report their weekly income. In
Australia, respondents frequently perceive income-related questions as intrusive, and that
adolescents may not know their parents’ income or may be unwilling to disclose this
information, LSAY refrains from collecting income-related information directly2.
As an alternative to direct questions on income and wealth, possession-based measures
that enquire about the presence of consumer and cultural items in the household are
often used as suitable proxies in survey research (Buchmann 2002). Examples of
possession-based measures include the number of rooms, bedrooms, or bathrooms in
the home, the presence of a dishwasher, or the presence of literature or art. It is
implicitly assumed that parents who can afford certain household and cultural
possessions are assumed to provide their children with a richer set of social and
educational resources. The Longitudinal Surveys of Australian Youth (LSAY) asks a
range of questions regarding the presence of social, cultural and educational possessions
in the home; these questions are included in the individual measure of SES.
3. Method
We used data from the 2003 cohort of the Longitudinal Surveys of Australian Youth
(LSAY) to create a reference measure for SES. LSAY is a nationally-representative survey
that tracks young people in Australia from the ages of 15 to 25 as they move from school
into further study, work, and other destinations. LSAY contains a rich set of individual
background variables, which is an important prerequisite for creating an accurate SES
reference measure. Although LSAY is affected by attrition bias, we remedied this issue
by applying appropriate weights. Thus, LSAY attrition bias did not have a detrimental
2Direct
income questions are avoided to minimise attrition. If questions are too intrusive, respondents will drop out of the survey in
later years, which has a detrimental effect on overall survey quality.
6
effect on our comparisons with regard to SEIFA. Furthermore, our measure of SES was
derived using variables from the first LSAY wave from the 2003 base year, which was
not influenced by longitudinal attrition effects. We used a set of 16 SES-related
background variables as a basis for creating our SES reference measure (Table 2).
7
Table 2
LSAY Variables Used in Creating an SES Reference Measure
Variable
Type or Categories
Valid n
SE
̂
𝐩
Own desk at home
Dichotomous
10366
0.902
0.003
Own room at home
Dichotomous
10366
0.915
0.003
Own study place at home
Dichotomous
10367
0.834
0.004
Computer software at home
Dichotomous
10366
0.669
0.005
Internet at home
Dichotomous
10366
0.849
0.003
Calculator at home
Dichotomous
10366
0.971
0.002
Literature at home
Dichotomous
10365
0.363
0.005
Poetry at home
Dichotomous
10366
0.406
0.005
Art at home
Dichotomous
10366
0.556
0.005
Textbooks at home
Dichotomous
10366
0.802
0.004
Dictionary at home
Dichotomous
10366
0.973
0.002
Dishwasher at home
Dichotomous
10362
0.594
0.005
Number of books at home
0 – 10
447
0.043
0.002
11 – 25
883
0.085
0.003
26 – 100
2848
0.275
0.004
101 – 200
2347
0.226
0.004
201 – 500
2205
0.213
0.004
More than 500
1486
0.143
0.003
Parental occupation
Continuous
9417
46.635*
17.246
Parental education
None
310
0.031
0.002
ISCED 1
84
0.084
0.001
ISCED 2
2064
0.206
0.004
ISCED 3B,C
319
0.032
0.002
ISCED 3A, 4
3174
0.317
0.005
ISCED 5B
1201
0.119
0.003
ISCED5A, 6
2872
0.286
0.005
Dichotomous
10364
0.939
0.002
Own computer at home
Note: Sample sizes and proportions are unweighted. Proportions represent the per cent respondents to whom the
listed variable condition applies.
*Value represents the mean, not the proportion.
We conducted a latent-class factor analysis to create an individual reference measure of
SES. Traditional factor analysis assumes all of the variables included in the model to be
continuous. It further assumes that the emerging factors of interest follow a continuous
distribution. In our case, however, the majority of the 16 variables used in the model
were dichotomous or categorical, thereby violating basic distributional assumptions for
traditional factor analysis. We addressed this issue by using tetrachloric (dichotomous)
and polychloric (categorical) correlations in a latent class factor analysis, which was a
better fit for the particular distributional properties of our SES-related variables.
Our reference measure was designed to measure SES at the individual, rather than
aggregate, level. For the purposes of this analysis, and due to its empirical foundation, we
assumed our reference measure to achieve an unbiased classification of individual SES.
Classification bias resulting from different SES measures was determined using
standardized differences and cross-tabulations of quintiles.
8
4. Results
4.1 Factor Analysis
Eigenvalues and related statistics for the latent-class factor analysis are provided in Table
3. The interpretation of results from latent-class factor analysis is similar to that of
traditional factor analysis.
Table 3
Initial Latent-class Factor Analysis
Factor
Eigenvalue
Difference
Prop. Explained
Cumulative
1
6.28
NA
39.25
39.25
2
1.74
4.54
10.89
50.14
3
1.48
0.26
9.26
59.39
4
1.15
0.34
7.16
66.55
5
0.81
0.33
5.08
71.63
6
0.72
0.09
4.52
76.15
7
0.71
0.02
4.41
80.56
8
0.65
0.06
4.06
84.61
9
0.54
0.11
3.37
87.98
10
0.48
0.06
2.97
90.95
11
0.42
0.06
2.61
93.56
12
0.35
0.07
2.18
95.74
13
0.21
0.08
1.32
97.06
14
0.29
0.06
1.81
98.87
15
0.17
0.04
1.07
99.94
16
0.01
0.16
0.06
100.00
Four factors had Eigenvalues in excess of 1. Collectively, these four factors explained
over 65% of the total variance in the model. The four factors included income and
wealth, study resources, computing resources, and cultural resources (Table 4).
Table 4
Loadings for the Four-factor Model (Varimax Rotation)
LSAY
Variable
Study
Resources
Income/Wealth
Computing
Resources
Cultural
Resources
Desk
0.628
0.167
0.243
0.259
Own room
0.524
0.312
-0.026
-0.052
Study place
0.758
0.100
0.097
0.257
Software
0.413
-0.043
0.547
0.377
Internet
0.251
0.390
0.641
0.062
Calculator
0.675
-0.033
0.321
0.206
Literature
0.148
0.319
0.136
0.787
Poetry
0.134
0.172
0.040
0.880
Art
0.231
0.149
0.116
0.619
Textbooks
0.384
0.023
0.214
0.559
Dictionary
0.667
0.012
0.353
0.382
Dishwasher
0.237
0.422
0.272
0.089
Parental occupation
0.060
0.507
0.101
0.165
9
No. of books
0.109
0.305
0.098
0.453
Parental education
0.005
0.503
0.068
0.227
Computer in home
0.198
0.279
1.109
0.171
Note: The highest loading for each variable across all four factors is bolded.
Table 4 demonstrates that the variable capturing computer availability in the home
produced a loading in excess of 1, indicating an estimation anomaly. A likely cause for
this anomaly was possible collinearity between the presence of a computer in the home,
the availability of software, and internet access. To obviate further estimation problems,
we eliminated the ‘computer in the home’ variable from the model. Results from the
latent-class factor analysis for the modified model are provided in Table 5 and Figure 1.
Table 5
Loadings for the Three-factor Model
LSAY Variable
Educational
Resources
Income/
Wealth
Cultural
Resources
Desk
0.664
0.219
0.229
Own room
0.396
0.291
-0.032
Study place
0.685
0.130
0.231
Software
0.608
0.091
0.324
Internet
0.464
0.493
0.063
Calculator
0.776
0.052
0.141
Literature
0.210
0.302
0.791
Poetry
0.176
0.128
0.883
Art
0.287
0.146
0.608
Textbooks
0.484
0.057
0.524
Dictionary
0.788
0.091
0.323
Dishwasher
0.299
0.482
0.083
Parental occupation
0.047
0.515
0.188
No. of books
0.132
0.293
0.463
-0.014
0.495
0.256
Parental education
Note: The highest loading for each variable across all three factors is bolded.
10
Figure 1
Scree Plot of Eigenvalues
Eigenvalues
7.00
6.00
5.00
4.00
3.00
2.00
1.00
0.00
0
2
4
6
8
10
12
14
16
The modified model identified three factors with Eigenvalues greater than 1 as
underlying traits of SES, including educational resources, income and wealth, and cultural
resources. However, the Eigenvalue difference between the first factor and the other two
factors was substantial. Given the disproportionately large amount of variance explained
by the first factor, we isolated the first factor to generate one composite measure of SES
consisting of a variety of home resources and parental background dimensions (Table 6).
Table 6
Loadings for the Single-factor Model
LSAY Variable
Composite SES Factor
Desk
0.628
Own room
0.340
Study place
0.618
Software
0.598
Internet
0.499
Calculator
0.640
Literature
0.844
Poetry
0.803
Art
0.653
Textbooks
0.659
Dictionary
0.782
Dishwasher
0.426
No. of books
0.523
Parental occupation
0.359
Parental education
0.380
We used the single-factor model as our SES reference measure for the remainder of this
paper. Our decision was based on practical considerations related to the interpretability
and useability of subsequent analyses. Given that our primary interest centred on the
evaluation of various different measures of SES, the composite SES measure from the
11
single-factor model provided a less complex reference for comparison. For the
remainder of this paper, we refer to our SES reference measure as SES-C (SESComposite).3
4.2
Comparing SEIFA and SES-C
Given sustained criticisms surrounding the accuracy of area-based SES measures, this
section compares SEIFA to SES-C in terms of classification bias. To carry out
meaningful comparisons, we assumed our SES-C measure to be an unbiased
classification variable against which to assess the extent of misclassification resulting
from SEIFA.
Despite the availability of various SEIFA measures, our investigation exclusively
considered the SEIFA Postal Area (SEIFA-POA) indexes. This limitation was necessary
because information contained in the LSAY dataset is limited to respondents’ residential
postcode. SEIFA-POA consists of four indexes, including education and occupation,
economic resources, relative advantage and disadvantage, and relative disadvantage.
Correlations between our SES-C reference measure and each of the four SEIFA-POA
indexes were in close proximity to each other (Table 7), prompting us to choose the
most-highly correlated Index of Education and Occupation (SEIFA-EO) for comparison
purposes.
Table 7
Correlation of SEIFA-POA with SES-C
Variable
Correlation with SES-C
SEIFA-POA Index of Education and Occupation
0.30
SEIFA-POA Index of Economic Resources
0.26
SEIFA-POA Index of Relative Advantage
0.29
SEIFA-POA Index of Relative Disadvantage
0.29
The low correlations between SES-C and any of the SEIFA indexes indicated that the
classification of individuals into SES groups was highly sensitive to the particular SES
measure chosen.
4.2
Individual-level Comparison
In determining the performance of SEIFA-EO with that of SES-C we used three
different approaches. In the first approach, the distributions of SEIFA-EO and SES-C
were both standardized to a Gaussian with m = 1,000 and sd = 100. Each individual in
the dataset thus had a standardised score for SEIFA-EO and SES-C. Standardised scores
were then subtracted from one another (see distribution of differences depicted in Figure
2. While the majority of the differences fell between -1.5 (-150) and +1.5 (150) standard
deviations, substantial differences were observed well beyond this range.
Our SES measure overlaps to some extent with the Index of Economic, Social, and Cultural Status (ESCS), a measure of SES
developed by the Organisation for Economic Cooperation and Development (OECD 2005) for use in the Program for International
Student Assessment (PISA). Similar to our measure, ESCS is derived from family background variables, including parental
occupation, parental education, and home possessions. Despite the high correlation between our SES-C measure and ESCS (r =
0.75), important differences exist. ESCS scores are obtained as component scores for the first principal component from factor
analysis, whereby 0 is the score of an average OECD student and 1 the standard deviation across equally weighted OECD countries.
The need for multi-country adjustment renders ESCS less reliable when considering only the Australian context. This loss in reliability
is reflected in the considerably lower reliability coefficient for ESCS compared with our SES-C reference measure (standardised
Cronbach’s alpha for ESCS for Australia = 0.61; standardized Cronbach’s alpha for SES-C = 0.74).
3
12
Figure 2
SEIFA-EO and SES-C Distribution of Differences
In the second comparison, SES-C scores were plotted against SEIFA-EO scores (Figure
3). The low correlation between the two measures was clearly reflected in the shape of
the plot. Assuming that a score of 900 or below (i.e., one standard deviation or more
below the SES mean score) represents low-SES, we examined the classification of
individuals classified as low-SES by either of the two measures (Quadrant 2). Quadrants
1 and 3 depict the number of misclassified individuals when using SEIFA-EO.
Specifically, Quadrant 1 shows actual high-SES individuals who were misclassified as
low-SES using SEIFA-EO, whereas Quadrant 3 contains actual low-SES individuals who
were misclassified as high-SES using SEIFA-EO.
Figure 3
SEIFA-EO and SES-C Classification Plot
Q1
Q4
Q2
Q3
13
Cross-tabulation of quintiles was a third approach to determining classification bias
between SEIFA and SES-C. Table 8 presents the cross-tabulation of SES-C and SEIFAEO quintiles.
Table 8
SEIFA-EO and SES-C Quintiles
SEIFA
SES-C
1
2
3
4
5
Total
1 = lowest
5.53
5.19
4.09
3.38
1.80
20
2
4.57
4.51
4.34
4.04
2.52
20
3
4.20
4.24
4.29
4.03
3.30
20
4
3.54
3.72
4.04
4.03
4.65
20
5 = highest
2.03
2.42
3.59
4.30
7.64
20
Total
20
20
20
20
20
100
The diagonal sum of Table 8 yielded a correct classification rate of only 26.3%. A further
35% of individuals were slightly misclassified (first-order off-diagonals). Almost 40% of
individuals in this sample were, however, severely misclassified (second-order and above
off-diagonals). Our results were very similar to those reported by Coelli (2010) who
compared SEIFA and income levels using the Household, Income and Labour
Dynamics in Australia (HILDA) dataset.
4.3
Aggregate-level Comparison
Cross-tabulations at the individual level illustrated the strength of classification bias
caused by using SEIFA indexes. However, these analyses did not lend themselves to an
examination of classification at the aggregate level. To assess aggregate-level
misclassification, we conducted two logistic regression analyses. The first examined
whether an individual participated in a Bachelor or higher degree by age 19. The second
examined whether an individual participated in Vocational Education and Training
(VET) by age 19. We note that these two regressions were not mutually exclusive and
that in individual may have indeed participated in both. Both regression analyses were
undertaken twice for each outcome, using our SES-C reference measure in the first run,
and SEIFA-EO in the second4.
Our focus on participation in post-school education as an outcome measure was
motivated by the considerable policy interest in enhancing higher education access for
low-SES youth. Table 9 provides the aggregate-level predicted probabilities of
participation in higher education and VET by age 19 that resulted form the regression
analyses. Predicted probabilities were categorised into low, medium and high SES (details
for the regression models are provided in the Appendix).
The regression models also included covariates that controlled for Indigenous status, regionality, family structure and academic
achievement.
4
14
Table 9
Probability of Higher Education Participation by Age 19
SES Measure
Low SES
Predicted
Probability
Medium SES
Predicted
Probability
High SES
Predicted
Probability
Higher Education Participation
SES-C
0.230
0.374
0.570
SEIFA
0.267
0.383
0.542
SES-C
0.314
0.243
0.170
SEIFA
0.301
0.241
0.173
VET Participation
Despite the considerable strength of classification bias resulting from the use of SEIFA
and SEIFA composites at the individual level, both measures performed reasonably well
at the aggregate level. However, SEIFA over-stated participation probabilities for the low
and medium SES quintiles, and under-stated them for the high SES quintile for higher
education. For VET participation, the difference between SES-C and SEIFA was
smaller, with either measure of SES providing reasonably accurate aggregate VET
participation rates. Regardless of the differences between SEIFA and SES-C, it was
evident that low-SES individuals are much more likely to participate in VET than in
higher education, even after controlling for academic ability.
5. Conclusion
Given sustained policy interest in improving the social inclusion of disadvantaged youth,
we examined two fundamentally different approaches to measuring SES. The first
approach was based on the average economic profile of a given geographic location (i.e.,
SEIFA). The second approach was based on a mixture of possessions and individual
background variables. Our investigation was motivated by the frequent use of SEIFA in
directing economic resources to disadvantaged youth. Doing so assumes that SEIFA can
properly identify youth from low-SES backgrounds. Results from our analysis question
the adequacy of this assumption.
SEIFA measures perform satisfactorily with regard to the aggregate measurement of
SES, as evidenced by our model for overall participation in higher education. Yet, the
foremost concern of social inclusion policies lies with the ability to channel resources
toward disadvantaged individuals rather than aggregate population averages. It is at the
individual level, then, that our analysis reveals the shortcomings of area-based measures
as SES classifiers.
SEIFA and other region-based measures are often used by governments to assess the
SES of individuals and groups as they are readily derivable and do not raise concerns
regarding privacy and intrusiveness. For example, the Australian Government has
allocated $168.5 million to increase enrolment in higher education among low-SES
youth. The allocation of these funds is primarily based on SEIFA. Given the extent to
which SEIFA misclassifies the SES of young individuals, the efficiency of policies that
seek to increase access to higher education and other critical resources for disadvantaged
stakeholders may suffer.
Results from our investigation prompt us to discourage the use of SEIFA in the context
of social inclusion. Instead of employing geographic measures, recently-developed social
inclusion frameworks for Australia (Saunders 2007; Scutella 2009) and the European
15
Union (Atkinson 2004) collectively emphasise the importance of individual SES
parameters, including education, access to services, as well as economic, social, and
political participation.
A potential solution to this measurement conundrum may lie in the use of extensive
administrative and other individual-level data held by, or available to, governments. All
developed economies have various forms of social support systems in place that are
criteria-based and allow, to a greater or lesser degree, some insight into the ‘true’ SES of
households and individuals. Given the increasing availability of individual-level
administrative data, we suggest that such data be used as a substitute for geographic SES
measures in categorising the SES of individuals. Achieving a better understanding of the
SES of individuals and groups is the first step to effectively improving the SES of
marginalised groups within the community.
6. Implications for Practice
Our analysis suggests that SES is complex and multifaceted, and clearly, as such, will be
challenging to measure. Any measurement of SES is somewhat subjective, as indeed is
the measure proposed in this paper. The measure that we have adopted takes a
possession-based approach to cultural and educational resources in addition to parental
occupation and education.
An implication of our work is that governments should utilise a measure of SES that is
focused on an assessment of individual household attributes and resources.
Governments do have some of this information available, and indeed in Australia
undertake household-level assessments of financial resources to determine eligibility for a
variety of welfare benefits (Mendes, 2009). This ‘administrative data’ includes a
significant amount of detail allowing governments to identify low SES households. For
example, where households contain recipients of Australia’s Disability Support Pension,
details of the nature and severity of the illness are held, along with information relating to
(a) an independent determination of the claimant’s capacity to undertake work, (b) their
financial resources, and (c) disposable income (Cai, Vu and Wilkins, 2008).
More commonly, for recipients of unemployment benefits, Australia’s government
collects detailed information regarding address, age, assets and income, whether the
claimant has children and, if so, whether the claimant pays or receives payments for these
children. When claimants are young people (under 25) the government also, in some
cases, collects information on parental assets and income.
In different jurisdictions, the use of this data is controlled in different ways. Few
governments, however, preclude the matching of anonymised data for group or
population level comparisons (for example, at the institutional level) (Jutte, Roos and
Brownell, 2011). Indeed, in Australia, part of the allocation of low SES participation
funding for universities is based on an assessment of the proportion of an institution’s
students whose parents are in receipt of various welfare benefits.
Given increasing access to relevant data for measuring individual SES, governments and
other stakeholders may benefit from using such data to devise more accurate measures
of SES. In the absence of such measures, institutions or individuals who might be
beneficiaries of targeted government policies may miss out due to misclassification.
Likewise, situations in which government resources are misdirected can be avoided. We
thus argue for better use of administrative data to foster the development of more
16
accurate SES measures for individual classification. This would require negotiating issues
of privacy and data-security.
7. References
ABS (Australian Bureau of Statistics) 2001, Australian standard classification of education
(ASCED), Commonwealth of Australia, Canberra.
—— 2008, An introduction to socio-economic indexes for areas (SEIFA), Commonwealth of
Australia, Canberra.
——2009, ANZSCO – Australian and New Zealand Standard Classifications of Occupations,
Canberra.
Atkinson, AB, Marlier, E & Nolan, B 2004, ‘Indicators and targets for social inclusion in
the EU’, Journal of Common Market Studies, vol. 42, pp. 47-75.
Bourdieu, P 1973, ‘Cultural reproduction and social reproduction’, in Papers in the sociology
of education, ed R Browen, Tavistock, London, pp. 71-112.
Bradley, RH & Corwyn, RF 2002, ‘Socioeconomic status and child development’, Annual
Review of Psychology, vol. 53, pp. 371-399.
Buchmann, C 2002, ‘Measuring family background in international studies of education:
conceptual issues and methodological challenges’, in Methodological advances in crossnational surveys of educational achievement, eds AC Porter and A Gamoran, National
Academy Press, Washington, pp. 150-197.
Cai, L, Vu, H and Wilkins, R 2008, The Extent and Nature of Exits from the Disability
Support Pension, Australian Bulletin of Labour, vol. 34, pp. 1-27.
CHSE (Centre for the Study of Higher Education) 2008, Participation and equity: a review of
the participation in higher education of people from low socio‐economic backgrounds and Indigenous
people, University of Melbourne, Melbourne.
Coelli, MB 2010, The forgotten second quartile: parental income and youth post-secondary education
enrolment in Australia, University of Melbourne, Melbourne.
Coleman, JS 1987, ‘Families and schools’, Educational Researcher, vol. 16, pp. 32-38.
Cresswell, J 2004, Immigrant status and home language background: implications for Australian
student performance in PISA 2000, Australian Council for Educational Research,
Camberwell.
Crul, M & Vermeulen, H 2003, ‘The second generation in Europe’, International Migration
Review, vol. 37, pp. 965-986.
Department of Families, Housing, Community Services and Indigenous Affairs 2011,
Household, Income and Labour Dynamics of Australia, viewed January 2011,
http://www.fahcsia.gov.au/sa/families/progserv/Pages/ldi-hilda.aspx.
Ganzeboom, HB G, De Graaf, PM & Treima, DJ 1992, ‘A standard international socioeconomic index of occupational status’, Social Science Research, vol.21, pp.1-56.
Goldthorpe, JH 1983, ‘Women and class Analysis: in defence of the conventional view’,
Sociology, vol. 17, pp. 465-485.
Gordon, D, Adelman, L, Ashworth, K, Bradshaw, J, Levitas, R, Middleton, S, Pantazis,
C, Patsios, D, Payne, S, Townsend, P & Williams, J 2000, Poverty and social exclusion in
Britain, Joseph Rowntree Foundation, York.
Haveman, R & Wolfe, B 1995, ‘The determinants of children's attainments: a review of
methods and findings’, Journal of Economic Literature, vol. 33, pp. 1829-1878.
International Labour Organization 1990, International standard classifications of occupations
(ISCO-88), Geneva, viewed January 2011,
http://www.ilo.org/public/english/bureau/stat/isco/index.htm.
17
Jerrim, J & Micklewright, J 2009, Children’s education and parent’s socioeconomic
status: distinguishing the impact of mothers and fathers, University of Southampton,
Southampton.
Jones, RG 2001, Identifying higher education students from low socio-economic status
backgrounds and regional and remote areas, Department of Education, Science and
Training, Canberra.
Jutte, D, Roos, L and Brownell, M 2011, Administrative Record Linkage as a Tool for
Public Health Research, Annual Review of Public Health, vol. 32, pp. 91-108.
Korupp, S, Ganzeboom, H & Van Der Lippe, T 2002, ‘Do mothers matter? A
comparison of models of the influence of mothers’ and fathers’ educational and
occupational status on children’s educational attainment’, Quality and Quantity, vol. 36,
pp. 17–42.
Marks, GN, McMillan, J, Jones, FL & Ainley, J 2000, The measurement of socioeconomic status
for the reporting of nationally comparable outcomes of schooling, Australian Council for
Educational Research, Camberwell.
McMillan, J, Jones, F & Beavis, A 2008, Introducing the Australian socioeconomic index 2006,
Australian Demographic and Social Research Institute, Canberra.
Mendes, P 2009, Retrenching or renovating the Australian welfare state: the paradox of
the Howard government's neo-liberalism. International Journal of Social Welfare, vol. 18,
pp. 102–110.
Pantazis, C, Gordon, D & Levitas, R (eds) 2006, Poverty and social exclusion in Britain: the
Millennium survey, The Policy Press, Bristol.
Saunders, P, Naidoo, Y & Griffiths, M 2007, Towards new indicators of disadvantage:
deprivation and social exclusion in Australia, Social Policy Research Centre, Sydney.
Scutella, R, Wilkins, R & Horn, M 2009, Measuring poverty and social exclusion in Australia: a
proposed multidimensional framework for identifying socio-economic disadvantage, University of
Melbourne, Melbourne.
Western, J, McMillan, J & Durrington, D 1998, Differential access to higher education: the
measurement of socioeconomic status, rurality, and isolation, Commonwealth of Australia,
Canberra.
18
7. Appendix
This appendix presents the regression co-efficients for the regression of participation in
VET and higher education against SES-C, SEIFA and a range of control variables.
Table A1 Logistic Regression Co-efficients - VET Participation for SES-C.
Parameter
df
Estimate
Intercept
1
-0.094
0.605
0.876
0.000
0.000
0.050
-0.785
0.067
<0.001
SES-C
Males
1
Females
Non-Indigenous
P-value
Reference Category
1
0.028
Indigenous
Non-Metropolitan
SE
0.202
0.890
Reference Category
1
-0.293
Metropolitan
0.065
<0.001
Reference Category
Government
1
-0.741
0.107
<0.001
Catholic
1
-0.657
0.117
<0.001
Independent
Reference Category
Single-Family
1
-1.250
0.458
0.006
Nuclear-Family
1
-1.262
0.455
0.006
Mixed Family
1
-1.394
0.465
0.003
Other Family structure
1
-0.660
0.503
0.190
Unknown Family structure
Reference Category
Maths Achievement score
1
0.003
0.001
<0.001
Reading Achievement score
1
0.003
0.001
<0.001
Science Achievement score
1
-0.000
0.001
0.980
Note. The probability modelled is the probability of NOT undertaking VET study by age
19
19
Table A2 Logistic Regression Co-efficients - VET Participation for SEIFA.
Parameter
df
Estimate
Intercept
1
-5.260
1.354
<0.001
0.006
0.001
<0.001
-0.786
0.067
<0.001
SEIFA
Males
1
Females
Non-Indigenous
P-value
Reference Category
1
0.019
Indigenous
Non-Metropolitan
SE
0.202
0.925
Reference Category
1
-0.191
Metropolitan
0.069
0.005
Reference Category
Government
1
-0.659
0.110
<0.001
Catholic
1
-0.595
0.119
<0.001
Independent
Reference Category
Single-Family
1
-1.260
0.458
0.006
Nuclear-Family
1
-1.246
0.455
0.006
Mixed Family
1
-1.387
0.465
0.003
Other Family structure
1
-0.671
0.503
0.182
Unknown Family structure
Reference Category
Maths Achievement score
1
0.003
0.001
<0.001
Reading Achievement score
1
0.003
0.001
<0.001
Science Achievement score
1
7.4E-06
0.001
0.991
Note. The probability modelled is the probability of NOT undertaking VET study by age
19
20
Table A3 Logistic Regression Co-efficients – HE participation for SES-C.
Parameter
df
Intercept
1
SES-C
Males
1
Estimate
0.587
<0.001
0.002
0.000
<0.001
-0.630
0.065
<0.001
Reference Category
1
0.389
Indigenous
Non- Metropolitan
P-value
-8.523
Females
Non-Indigenous
SE
0.271
0.152
Reference Category
1
-.034
Metropolitan
0.068
<0.001
Reference Category
Government
1
-0.747
0.082
<0.001
Catholic
1
-0.309
0.093
0.001
Independent
Reference Category
Single-Family
1
-0.286
0.357
0.424
Nuclear-Family
1
0.163
0.351
0.643
Mixed Family
1
-0.623
0.370
0.092
Other Family structure
1
-0.010
0.405
0.980
Unknown Family structure
Reference Category
Maths Achievement score
1
0.007
0.001
<0.001
Reading Achievement score
1
0.004
0.001
<0.001
Science Achievement score
1
0.001
0.001
0.066
Note. The probability modelled is the probability of undertaking Bachelor or higher by age
19
21
Table A4 Logistic regression co-efficients – HE participation for SEIFA.
Parameter
df
Estimate
SE
P-value
Intercept
1
-14.778
1.248
<0.0001
SEIFA
1
0.008
0.001
<0.0001
Males
1
-0.640
0.065
<0.0001
Females
Non-Indigenous
Reference Category
1
0.3944
Indigenous
Non- Metropolitan
0.272
0.147
Reference Category
1
-0.205
Metropolitan
0.073
<0.0001
Reference Category
Government
1
-0.688
0.085
<0.0001
Catholic
1
-0.269
0.0947
0.004
Independent
Reference Category
Single-Family
1
-0.333
0.358
0.352
Nuclear-Family
1
0.180
0.352
0.609
Mixed Family
1
-0.638
0.370
0.085
Other Family structure
1
-0.030
0.406
0.006
Unknown Family structure
Reference Category
Maths Achievement score
1
0.007
0.0006
<0.0001
Reading Achievement score
1
0.004
0.0007
<0.0001
Science Achievement score
1
0.002
0.0007
0.024
Note that probability modelled is the probability of undertaking Bachelor or higher by age
19
22