The Mechanisms Underlying Incubation in Problem Solving
Ut Na Sio
Doctor of Philosophy
Department of Psychology
Lancaster University
March, 2010
i
“the perfect uselessness of knowing the answer to the wrong question”
(Ursula Kroeber Le Guin, The Left Hand of Darkness, p.57, 1969)
ii
ABSTRACT
This thesis investigated the mechanisms underlying incubation. Three unconscious
processes have been proposed to explain incubation effects: spreading activation, that is,
distributing activation to relevant memories via associative networks; selective
forgetting, that is, suppressing inappropriate strategies/concepts; and cue assimilation,
that is, re-encoding a problem in a form that facilitates the assimilation of
chance-encountered cues. The heterogeneity of settings and the insensitivity of
measurements of the incubation studies have to date made it impossible to draw any
conclusions on the role of incubation. This thesis addressed these methodological
problems, initially by carrying out a meta-analysis to synthesize past findings, and in
experimental studies by supplementing the traditional paradigm with an additional
measurement (sensitivity to relevant and misleading items), to test between the three
hypotheses.
The meta-analysis identified that incubation is susceptible to numerous procedural
moderators. Longer preparation periods gave a stronger incubation effect for visual
problems. Low-load tasks yielded a stronger incubation effect than rest during an
incubation period when solving linguistic problems. Experiments with linguistic
problems revealed that cue-assimilation occurred when solving problems at an
intermediate level of difficulty and when the incubation period was filled with low-load
tasks. This supported the cue-assimilation hypothesis and confirmed the meta-analysis
findings, and suggested an additional moderator: problem difficulty. Sensitivity to
relevant and misleading items was not influenced by incubation. The
spreading-activation and selective-forgetting hypotheses were not supported. An
experiment using visual problems, examining the impact of preparation length and
attentional allocation style (focused vs. diffuse), revealed a positive correlation between
iii
preparation length and incubation effect size. Controlling individuals’ attentional
allocation styles reduced the variance in incubation effect sizes.
Together, the meta-analysis and the experimental studies support a positive
incubation effect, and identify procedural and cognitive moderators of the effect. The
absence of knowledge-activation during the incubation period suggests that incubation
effects may rely on unconscious meta-cognitive processes.
iv
ACKNOWLEDGMENTS
I would like to take this moment to thank a number of people. They have made my years
in Lancaster University not merely a study program but an education.
First, I would like to express my gratitude to my supervisor, Prof. Tom Ormerod.
His invaluable advice and support have guided me to successfully complete this thesis.
I am also grateful to my second supervisor, Dr. Linden Ball, for being so ready and
willing to help on all occasions. I would like to thank the Psychology Department,
Lancaster University, for the financial support.
My thanks also go to my friends, spending time with them is the best incubation
period for my PhD study. Without them, I would have been defeated by Britain’s gloomy
weather and my self-doubts.
At last, I must thank my family for loving me enough to let me go.
v
DECLARATION
I declare that the thesis hereby submitted for the degree of Doctor of Philosophy at
Lancaster University is my own work and has not been previously submitted in substantially
the same form for the award of a higher degree elsewhere.
vi
PUBLISHED MANUSCRIPTS
Sio, U. N. & Ormerod, T. C. (2009). Mechanisms underlying incubation in
problem-solving: Evidence for unconscious cue assimilation. In N.A. Taatgen & H. van
Rijn (Eds.), Proceedings of the 31st Annual Conference of the Cognitive Science Society.
Amsterdam: Cognitive Science Society.
Sio, U. N. & Ormerod, T. C. (2009). Does incubation enhance problem solving? A
meta-analytic review. Psychological Bulletin, 135, 94-120.
vii
TABLE OF CONTENTS
ABSTRACT ..................................................................................................................... iii
ACKNOWLEDGMENTS ................................................................................................ v
DECLARATION.............................................................................................................. vi
PUBLISHED MANUSCRIPTS...................................................................................... vii
CHAPTER I: INTRODUCTION ...................................................................................... 7
CHAPTER II: LITERATURE REVIEW ........................................................................ 11
An Information Processing Model of General Problem Solving ................................ 11
The Nature of Insight .................................................................................................. 15
Proposed Mechanisms of Incubation ...................................................................... 23
Procedural Moderators of the Incubation Effect ..................................................... 26
Attentional Resource Allocation Styles and Incubation ......................................... 32
Methodological Problems of Past Incubation Studies ............................................ 33
CHAPTER III: THE IMPACT OF PROCEDURAL MODERATORS ON
INCUBATION EFFECTS: META-ANALYSIS............................................................. 36
Method .................................................................................................................... 37
Results ..................................................................................................................... 45
Discussion ............................................................................................................... 64
CHAPTER IV: EXPERIMENTS I, II, and III - LINGUISTIC INSIGHT PROBLEMS 73
EXPERIMENT I ......................................................................................................... 74
Method .................................................................................................................... 75
Results ..................................................................................................................... 80
Discussion ............................................................................................................... 89
EXPERIMENT II ........................................................................................................ 91
1
Method .................................................................................................................... 92
Results ..................................................................................................................... 94
Discussion ............................................................................................................. 109
EXPERIMENT III .................................................................................................... 116
Method .................................................................................................................. 117
Results ................................................................................................................... 121
Discussion ............................................................................................................. 133
DISCUSSION: EXPERIMENT I, II, and III ............................................................ 138
CHAPTER V: EXPERIMENT IV-VISUAL INSIGHT PROBLEMS .......................... 144
Method .................................................................................................................. 147
Results ................................................................................................................... 152
Discussion ............................................................................................................. 165
CHAPTER VI: GENERAL DISCUSSION .................................................................. 172
REFERENCES ............................................................................................................. 181
Appendix A ................................................................................................................... 195
Appendix B ................................................................................................................... 203
Appendix C ................................................................................................................... 227
Appendix D ................................................................................................................... 243
Appendix E ................................................................................................................... 246
Appendix F.................................................................................................................... 248
Appendix G ................................................................................................................... 249
Appendix H ................................................................................................................... 250
Appendix I .................................................................................................................... 253
2
LIST OF TABLES AND FIGURES
Table 3.1
Coding System Used In The Meta-Analysis .............................................. 40
Table 3.2
Stem-And-Leaf Display Of 117 Unbiased Effect Sizes ............................. 46
Table 3.3
The Random Variance Component, Weighted Mean, Standard Deviation,
Standard Error, And 95% Confidence Interval of the Effect Size Estimate
by Each Categorical Moderator ................................................................. 50
Table 3.4
Regression Model for Linguistic Problem Studies (N= 65) ...................... 51
Table 3.5
Regression Model for Misleading Cue Studies (N= 85) ............................ 52
Table 3.6
Regression Model on No Relevant Cue Studies (N = 82) ......................... 53
Table 3.7
Regression Model on High Cognitive-Load Incubation Task Studies (N= 75)
................................................................................................................... 54
Table 3.8
Regression Model on All Studies (N= 114) ............................................... 56
Table 3.9
Regression Model on All Studies (with Interactive Terms, N= 114) ......... 58
Table 3.10 Summary of Coefficient Differences of the Regression Model on All
Studies (with Interactive Terms, N= 114) .................................................. 60
Table 3.11
Regression Model for Studies Excluding Creative Problem (N = 100) ..... 62
Table 3.12
The Summary of Coefficient Differences of the Regression Model
Excluding Creative Problem (N= 100) ...................................................... 63
Table 4.1.1 Means and Standard Deviations of the Number of Correct Responses and
Response Latency at the First Attempt by Question Set and Condition .... 81
Table 4.1.2 Means and Standard Deviations of the Number of Correct Responses in the
Two Attempts and Improvement Score by Condition ............................... 83
Table 4.1.3 Means and Standard Deviations of the Number of Correct Responses in the
Cue and No-Cue Trials in Each Condition. ............................................... 84
3
Table 4.1.4 Means and Standard Deviations of Improvement Score on the Cue and
No-Cue Trials, and the Score Differences by Condition ........................... 85
Table 4.1.5 ANOVA Summary Table for the Effect of Condition and Cue on
Improvement Scores. ................................................................................. 85
Table 4.1.6 Means and Standard Deviations of the Lexical Decision Times for Target
Words and Neutral Words .......................................................................... 88
Table 4.1.7 ANOVA Summary Table for the Effect of Condition and Word Type on
Lexical Decision Time ............................................................................... 89
Table 4.2.1 Means and Standard Deviations of the Number of Correct Responses and
Response Time at the First Attempt in Experiment I and II ...................... 95
Table 4.2.2 Means and Standard Deviations of the Number of Correct Responses and
Response Time at the First Attempt by RAT Difficulty and Condition ..... 96
Table 4.2.3 Means and Standard Deviations of the Number of Correct Responses in the
Two Attempts by Condition and RAT Difficulty Level ............................. 98
Table 4.2.4 ANOVA Summary Table for the Effect of the RAT Difficulty and
Condition on Improvement Scores ............................................................ 99
Table 4.2.5 Means and Standard Deviations of Improvement Scores in the Presence
and Absence Of Cue, and the Score Difference by Condition and RAT
Difficulty. ................................................................................................. 101
Table 4.2.6 ANOVA Summary Table for the Effect of the RAT Difficulty, Condition,
and Cue on the Number of Correct Responses ........................................ 103
Table 4.2.7 Means and Standard Deviations of the Natural and Log-Transformed
Lexical Decision Times for Target Words and Neutral Words. ............... 107
Table 4.2.8 ANOVA Summary Table for the Effect of RAT Difficulty, Condition,and
Word Type on Log-Transformed Lexical Decision Time ........................ 108
4
Table 4.3.1 Means and Standard Deviations of the Number of Correct Responses and
Response Time at the First Attempt by Rebus Type and Condition ........ 122
Table 4.3.2 Mean and Standard Deviations of the Number of Correct Responses in the
Two Attempts by Condition and Rebus Type .......................................... 124
Table 4.3.3 ANOVA Summary Table for the Effect of Rebus Type and Condition, on
Improvement Scores ................................................................................ 125
Table 4.3.4 Mean and Standard Deviations of the Number of Correct Responses in the
Cue and No-Cue Trials in Each Condition. ............................................. 126
Table 4.3.5 Means and Standard Deviations of Improvement Score on the Cue and
No-Cue Trials, and the Cueing Effect, by Condition and Rebus Type .... 127
Table 4.3.6 ANOVA Summary Table for the Effect of Condition and Cue, and Rebus
Type on Improvement Scores. ................................................................. 128
Table 4.3.7 Means and Standard Deviations of the Original and Log-Transformed
Lexical Decision Times for Target Words and Neutral Words ................ 130
Table 4.3.8 ANOVA Summary Table for the Effect of RAT Difficulty, Condition, and
Word Type on Log-Transformed Lexical Decision Time ........................ 131
Table 4.3.9 Means and Standard Deviations of the Original and Log-Transformed
Lexical Decision Times for Misleading Hints and Neutral Words. ......... 132
Table 4.3.10 ANOVA Summary Table for the Effect of the Condition and Word Type on
Log-Transformed Lexical Decision Time................................................ 133
Table 5.1
Ethnic and Gender Distribution, and Means and Standard Deviations of the
GEFT Scores by Condition. ..................................................................... 153
Table 5.2
Solution Rate of the Visual Insight Problems (N = 32) ........................... 155
Table 5.3
Means and Standard Deviations of the Number of Insight Problem Solved
In The Pre and Post-Incubation Period in the IS and IL Conditions, and in
the Equivalent Time Period in the NI Condition. .................................... 156
5
Table 5.4
Summary of the Results of the Complete and the Backward Eliminated
Model ....................................................................................................... 158
Table 5.5
Summary of Results of the Two Hierarchical Multiple Regression Analyses
................................................................................................................. 162
Table D1
Regression Coefficients in Original and Transformed Regression Models
................................................................................................................. 245
Table E1
Regression Coefficients in Original and Transformed Regression Models
................................................................................................................. 247
Figure 3.1.
Funnel Plot of the Studies Included in the Meta-Analysis. ..................... 48
Figure 4.1.1
Task Presentation Sequence in Each Trial in Each Condition in
Experiment I. ........................................................................................... 79
Figure 5.1
Task Presentation Order in Each Trial in Experiment IV. ...................... 151
6
CHAPTER I: INTRODUCTION
Anecdotal reports of the intellectual discovery processes of individuals hailed as
geniuses (e.g., Ghiselin, 1985; Wallas, 1926; Woodworth & Schlosberg, 1954;) share a
common theme: a flash of insight pops unexpectedly into the mind of the individual
after he or she has put an unsolved problem aside for a period of time, having failed in
initial attempts to solve it. This temporary shift away from an unsolved problem that
allows a solution seemingly to emerge as if from no additional effort is termed an
‘incubation period’ (Wallas, 1926). Its importance in current thinking and practice is
illustrated by a recent search in Google Scholar for the term incubation along with either
creativity, insight, or problem which yielded 12,300 articles, with search restricted to the
years 1997 to 2009 and the subject areas of Social Sciences, Arts and Humanities. An
additional 3,300 articles were yielded when the subject areas Business Administration,
and Economics were included. Yet there are many conflicting account of incubation,
with some studies reporting a strong effect and other failing to find any effect at all. This
thesis aims to help resolve the uncertainties surrounding the phenomenon by describing
a statistical meta-analysis review of existing empirical studies of incubation and by
reporting a series of incubation experiments using novel paradigms.
One theoretical reason for studying incubation is because it is closely associated
with insightful thinking. Indeed, Wallas (1926) proposed incubation as the second of
five phases in problem-solving (the others being preparation, intimation, illumination,
and verification). Insight may be characterized as a sudden, unpredictable and
non-verbalizable solution discovery (e.g., Metcalfe & Weibe, 1987). Some researchers
see the apparently unconscious nature of solution discovery as evidence that the
processes required to achieve insight in problem-solving are qualitatively different from
those used to tackle problems that do not require insight(e.g., Jung-Beeman & Bowden,
7
2000; Wertheimer, 1985). Incubation might serve a valuable role in arbitrating between
theories of insight, in particular between special-process theories (e.g., Knoblich,
Ohlsson, Haider, & Reinhus, 1999; Ohlsson, 1992) and theories of insight as a normal
problem-solving processes (e.g., MacGregor, Ormerod, & Chronicle, 2001).
Understanding the role of incubation periods also allows us to make use of them
effectively to promote creativity in areas such as individual problem-solving, classroom
learning, and work environments. Educational researchers have tried to introduce
incubation periods in classroom activity, and positive incubation effects in fostering
students’ creativity have been reported (Lynch & Swink, 1967; Medd & Houtz, 2002;
Rae, 1997; Webster, Campbell, & Jane, 2006). However, in the absence of a
comprehensive theory or model that can explain how and why positive incubation
effects might emerge and under what conditions they are best fostered, no general
pedagogic recommendations can be made.
Two streams of mechanisms have been proposed to account for the incubation effect.
On the one hand, some researchers suggest that incubation periods provide additional time
for individuals to work on the problem covertly (Browne & Cruse, 1988; Posner, 1973).
In contrast to this conscious-work view is the proposal that performing irrelevant
incubation tasks facilitates the occurrence of some helpful unconscious problem solving
processes, such as spreading activation, selective forgetting, and cue assimilation (Bower,
Regehr, Balthazard, & Parker, 1990; Simon, 1966; Smith, 1995; Smith & Blankenship,
1991; Seifert, Meyer, Davidson, Patalano, & Yaniv, 1995; Yaniv & Meyer, 1987).
However, the diversity of the findings makes it very difficult to draw any conclusions on
the role of incubation. Some of the past studies have shown that introducing an incubation
period can improve individuals’ performance on insight problem solving (Dominowski &
Jenrick, 1972; Fulgosi & Guilford, 1968; Patrick, 1986; Smith & Blankenship, 1989,
1991). However, an incubation effect can disappear after a slight change in the
8
experimental setting (Fulgosi & Guildford, 1968; Patrick, 1986). Some researchers have
even failed to replicate the results of past studies under the same experimental setting
(Olton & Johnson, 1976). One possible reason for the conflicting findings is that there are
many moderators affecting incubation, and past studies have failed to control or adjust
them to an optimal level that fosters incubation.
Another methodological problem of past incubation studies is the insensitivity of the
measures used to assess incubation. The traditional method used in assessing incubation
in past studies was perhaps too general, in that researchers only compared individuals’
problem solving performance between incubation and control conditions. Although a
noticeable post-incubation performance improvement could support the positive impact
of incubation periods, it cannot show whether the improvement is the result of the
occurrence of a specific conscious or unconscious process. Because of these
methodological shortcomings, the nature of incubation remains unknown.
There were two objectives of this thesis. The first was to identify key moderators
affecting incubation effects. The second was to test the occurrence of three proposed
unconscious processes underlying incubation effects: spreading-activation,
selective-forgetting, and cue-assimilation. To achieve these two objectives, a
meta-analysis, which is a systematic quantitative review on past studies, was first carried
out to identify the potential procedural moderators on incubation effects. Experiments I,
II, III, and IV were then carried out to examine the impact of the potential moderators on
the incubation effect and verify the proposed mechanisms of incubation. Experiments I,
II, and III examined the occurrence of three proposed unconscious cognitive processes
and the optimal experimental settings for the occurrence of these processes. In order to
offer strong data to test the unconscious cognitive processes, incubation effects on both
problem solving performance and sensitivity to relevant and irrelevant items were
assessed in Experiment I, II, and III. Experiment IV shifted to investigate the impact of a
9
new cognitive moderator on incubation effects: attentional allocation style.
This thesis includes six chapters and is organized as follows. Chapter II is devoted
to the review of literature on insight problem-solving and incubation effects. Chapter III
presents a meta-analysis for identifying the potential procedural moderators. Chapter IV
presents the settings and findings of Experiments I, II, and III. Chapter V presents the
details of Experiment IV and the results found. Chapter VI presents a general discussion
of the findings of the meta-analysis and the experiments.
10
CHAPTER II: LITERATURE REVIEW
This chapter first presents a brief discussion of an information-processing model of
general problem solving. Following this, two major theories of insight problem solving
are described. The last section reviews research relevant to incubation effects in insight
problem solving, and discusses the potential moderators of the incubation effect as well
as methodological problems of the past studies. These moderators and methodological
issues provide the rationale for conducting the meta-analysis and the experiments
reported in the remainder of this thesis.
An Information Processing Model of General Problem Solving
Newell and Simon (1972) suggested that, although problems differ in their degree of
difficulty and contextual content, the mechanisms underlying different types of complex
problem-solving experiences are qualitatively the same and can be decomposed into a
series of basic cognitive processes. Newell and Simon put forward a general
information-processing model to explain human problem-solving behaviors.
To solve a problem, individuals have first to encode the information provided in the
task environment and construct an internal representation of the problem. The basic
elements of the internal representation are the initial and the goal states of the problem,
the moves that can transform the initial state to the goal state, and the constraints under
which moves can be applied. The problem state is the given situation of the problem,
and the goal state is the final state to that problem, which problem solvers try to achieve.
In some cases, the task environment may not provide enough information for knowing
the initial state and the goal state of the problem. Individuals may then have to make use
of their task-related knowledge in hypothesising these two problem states. Also, in order
11
increase the search efficiency, individuals usually will set up some constraints to filter
out some seemingly irrelevant or illegal moves.
Once the problem states are defined, solving a problem involves selecting and
executing a sequence of moves that can transform the initial state into the goal state in
accord with the problem constraints. The execution of each single move creates a new
problem state (current state), which will be added into the representation. Therefore, the
mental representation of a problem is dynamic rather than static during the
problem-solving process. Individuals can estimate how close they are to the goal by
calculating the difference between the current state and the goal state. There is usually
more than one possible move, and individuals do not randomly select one to execute.
Instead, individuals will apply heuristics, which are strategies that help an individual to
evaluate the effectiveness of the available moves in the search space, to select the most
effective move. The commonly-used heuristics include means-end analysis and
hill-climbing, which aim to find the moves that can maximally reduce the difference
between the current state and the goal state. This search process is halted once the goal
state is reached or an impasse is reached, in which none of the available moves can
apparently transform the current state to the goal state. In the latter case, the search
process is resumed after the internal representation is reformulated into a better form
(e.g., by releasing some unnecessary constraints or by re-encoding the initial and goal
states), or after a switch in search strategy.
In summary, there are two main stages in Newell and Simon’s (1972) model of
general problem solving. The first stage is the formation of the problem representation,
and the second stage is selecting and executing moves to make progress towards the
goal. Individuals may perform differently on the same problem because of differences in
their internal representation of the problem or because of the strategies they use in move
selection.
12
The information included in a representation may differ across individuals, and depends
on their task-related knowledge and experience. Individuals with no relevant experience
in the domain are more likely to include information which is not relevant to the
problem, and overlook subtle but relevant information in the task environment. They
may even not know the critical moves to transform an initial state to the goal state. In
other words, their representations may be mistaken or incomplete. Experienced
individuals should be more capable of selecting the relevant information to encode.
They may also generate some intermediate states as sub-goals to guide the
move-selection process. Also, additional constraints, not only based on the superficial
problem elements, but also on relevant problem solving experience, would be imposed
to filter out further irrelevant moves and narrow down the search space.
Individual differences also lie in the use of heuristics. To carry out a heuristic
search, individuals need to have an evaluation function that scores each move according
to how much it can reduce the distance between the current state and the goal/sub-goal
state. The evaluation function, again, may differ depending on an individual’s past
problem-solving experience. Individuals who have rich experience in solving similar
problems may use a strong task-specific evaluation function. However, when
encountering a new problem, individuals may only use weak task-general evaluation
functions, which may not accurately assess the goodness of each move.
Expertise research in a wide range of knowledge domains (e.g., physics, chemistry,
and mathematics) has illustrated a significant effect of past problem-solving experience
on problem representation and search strategy. Regarding problem representation,
experts tend to represent a problem according to its underlying principles, whereas
novices base their representation on superficial features of the problem (Adelson, 1984;
Chi, Feltovich, & Glaser, 1981; Heyworth, 1989; Larkin, 1983). In terms of the use of a
search strategy, novices usually work backward from the goal, in that they will apply
13
moves that can reduce the distance between goal state and the initial state. In contrast,
experts work forwards from the initial state, in that they first apply the moves that can
generate additional information to enrich the representation, even though those moves
may not make immediate progress to the goal. In summary, experts’ move-selection
process is knowledge-guided, while novices tend to use mean-end analysis, a
task-general strategy, to select moves. (Bedard & Chi, 1992; Heyworth, 1989; Kavakli &
Gero, 2002; Larkin, 1983; Lesgold et al., 1988).
The core concept of Newell and Simon’s problem-solving model is that these
aforementioned processes are explicit and occur at a conscious level, and can be
implemented on any computational system. The conceptualization of problem solving as
conscious search has become the dominant concept in human problem-solving and
artificial intelligence research. Newell and Simon (1972)’s general problem solving
model is supported by the analysis of individuals’ verbal protocols in solving different
types of problems. SOAR, a general cognitive architecture, has been developed based on
this theory to model a wide variety of intelligent behaviors, such as analogical transfer
(Gorski & Laird, 2006; Xu, Wintermute, Wang, & Laird, 2009), and decision making
(Wang & Lair, 2007).
Characterizing problem solving as conscious search implies that individuals are
consciously aware of the strategies they are using, and also of their progress towards the
goal. This assumption is contradicted by a number of phenomenological findings from
studies of a specific type of problem solving: insight problem-solving, which is
characterized as sudden, unpredictable, and nonverbalizable (Metcalfe & Weibe, 1987).
These phenomena raise a concern about whether Newell and Simon’s problem-solving
theory can be applied to insight problem-solving, a topic to which we now turn.
14
The Nature of Insight
Insight may be conceived as a sudden and unanticipated realization of the solution of a
problem (Metcalfe & Weibe, 1987). Two different perspectives, the conscious and
unconscious processing accounts, have been put forward that provide apparently
mutually exclusive accounts of the emergence of insight. The conscious account takes
the standpoint of Newell and Simon’s general problem-solving model and suggests that
insight problem-solving can be attributed to conscious search processes, even though
insight problem-solving has distinctive phenomenological features. Opposed to this is
the unconscious account, proposing that cognitive mechanisms underlying insights are
processes operating at a sub-conscious level.
From the perspective of the conscious account, insight is not viewed as a distinct
cognitive phenomenon from other kinds of problem solving. Kaplan and Simon (1990)
suggested that insight problem-solving falls within the framework of Newell and
Simon’s general problem solver model. The difficulty of insight problem-solving lies in
having inappropriate search constraints and heuristics in the move selection process that
lead to an impasse, and this impasse must be broken by a conscious switch of heuristics
or imposing appropriate constraints to confine the search to a relevant search space.
Based on this conceptualization, MacGregor, Ormerod, and Chronicle (2001)
propose what has since become to known as the Criterion for Satisfactory Progress
theory (CSPT), explaining how an inappropriate search strategy causes impasse in
insight problem solving. MacGregor et al. suggest that when insight problems do not
have a well-defined goal state, individuals will evaluate possible moves against locally
rational criteria that indicate whether progress is being made toward the goal properties.
For example, the criterion in solving the classic nine-dot problem is that each line must
cancel a number of dots given by the ratio of dots remaining to lines available. However,
15
this move-selection criterion may drive individuals away from the moves that lie on the
correct solution path. For example, when solving the nine-dot problem, individuals
usually fail to make the necessary insight to draw lines beyond the array because of the
many criterion-meeting moves available within the square shape.
When individuals fail to find moves that meet the criterion for satisfactory progress,
then they reach an impasse. As a consequence, they experience “criterion failure”, that is
a recognition that their current representation of the problem is unsatisfactory. At this
point, they will relax the requirement for maximizing progress, and will instead sample
non-maximal moves (e.g., drawing lines beyond the square when solving the nine-dot
problem), which may allow the discovery of subsequent moves that make more progress
than previous attempts. The impasse is then resolved and insight emerges. The change of
search strategy allows the exploration of the non-maximal moves that “suddenly” make
available new solution paths. This may explain the distinctive phenomenological
characteristics of insight, in which it appears to emerge suddenly and unexpectedly.
CSPT has predicted the observed behavior of individuals in several insight
problem-solving domains (Chronicle, MacGregor, & Ormerod, 2004; MacGregor,
Ormerod, & Chronicle, 2001; Ormerod, MacGregor, & Chronicle, 2002).
CSPT also suggests that the change in the search strategy occurs only if the
problem solver exhausts the initial search space and still fails to solve the problem, and
this is determined by the size of the initial search space. If the search space is small, then
individuals should realize the failure quickly, and in turn, have a stronger impulse to
abandon the inappropriate strategy. The prediction that how fast an individual has the
insight relies on the initial search space is evidenced by the findings that individuals’
problem solving performance in solving various visual insight problems is negatively
correlated with the size of the initial search space (MacGregor, Ormerod, & Chronicle,
2001; Ormerod, MacGregor, & Chronicle, 2002).
16
Unconscious processing accounts consider that the suddenness of the
emergence of insight reflects a qualitative difference between insight and non-insight
problem-solving processes. Ohlsson (1992) suggested that Newell and Simon’s problem
solving model has to be modified in order to handle insight problem-solving. He
proposed the Representational Change theory (RCT), suggesting that insight-specific
cognitive processes should be added into Newell and Simon’s model. Ohlsson (1992)
agrees that both insight and non-insight are approached at the same way initially, in that
problem solvers build a representation consisting of initial and the goal state, and search
and execute moves in transforming between the initial and the goal states. An impasse is
a state for which problem solvers fail to find a move that can transform the current
problem state to goal state. However, different from the perspective of the conscious
account, RCT(Ohlsson, 1992) suggests that reaching an impasse is mainly due to having
an incomplete or mistaken representation, rather than a wrong search strategy.
Furthermore, the problem representation is restructured via three unconscious processes:
Elaboration, Re-encoding, and Constraint relaxation. Elaboration is the process of
enriching information encoded into the initially incomplete representation. Re-encoding
is to abandon or reject the information encoded in the current representation in order to
correct the mistaken representation. Constraint relaxation is releasing unnecessary
constraints. When solving a problem, problem solvers tend to self-impose constraints on
the goal state of the problem to confine the search space. However, the constraints
sometimes are counterproductive and lead to search in an over-restricted space that does
not include the crucial problem-solving moves, and releasing inappropriate constraints
can expand the search space. These three representation-restructuring processes can be
accomplished through the unconscious spread of activation to relevant but previously
ignored memory items.
Knoblich and colleagues (Knoblich, Ohlsson, Haider, & Rhenius, 1999; Knoblich,
17
Ohlsson, & Raney, 2001) have enriched RCT by proposing that chunk decomposition is
another mechanism underlying restructuring. Chunking is the process of recoding
several stimuli into fewer but more complex and meaningful units. Consolidating
external information as different chunks can reduce working memory load during
problem solving (Chase & Simon, 1973; Ericsson & Lehmann, 1996; Miller, 1956).
Knoblich et al. (1999) examined the impact of constraints and chunks on the
likelihood of insight in the domain of arithmetic problem-solving using matchstick
algebra, where problems are presented to participants using matchsticks that form
Roman numerals. Each matchstick problem consisted of a false arithmetic statement
written in Roman numerals, an arithmetic operation, and an equals sign all constructed
out of matchsticks (e.g., IV = III + III). Individuals had to transform the statement into a
true statement by moving one stick from one position to another. In arithmetic
problem-solving, individuals will hold different levels of move-selection constraints
(from low to high), such as: numeral values cannot be changed except through
operations, operator functions cannot be modified, and arithmetic functions have to be in
the form x = F(y, z). They will also recode the task information into different chunks.
For example, XVII will be perceived as two chunks (X and VII), instead of six single
matchsticks \, /, I, I, \ and /. The chunk VII is a loose chunk relative to the chunk X,
because VII can be decomposed into different meaningful components V, I, and I, while
the components of X do not carrying any meaning. When solving arithmetic problems
requiring insight into a new use of knowledge, individuals may have to relax some
constraints and decompose some chunks. Low level constraints are local and specific,
and the relaxation of them will only cause a small portion of change in the
representation, and thus, they are easier to release. Loose chunks also have components
that are individually meaningful, and thus they are more likely to be decomposed.
Knoblich et al. suggest that insight problem-solving performance should be a function of
18
the level of the constraints that need to be relaxed and the tightness of the chunks that
need to be decomposed.
As predicted, individuals performed better in solving matchstick problems
requiring the relaxation of local constraints and decomposing loose chunks (e.g.,
problem: V–III= IV, solution: VI–II= IV) than in solving problems biased by global
constraints and tight chunks (e.g., problem: III = III + III, solution: III = III = III).
Knoblich et al. suggest that Chunk-Decomposition and Constraint-Relaxation can be
done by the suppression of activation of irrelevant constraints and chunks, and spread of
activation throughout the semantic network to activate relevant but previously ignored
memory items.
In summary, the conscious account suggests that reaching an impasse is due to the
use of wrong search strategy and it can be resolved by switching the search strategy,
which is a conscious meta-cognitive process. Unconscious accounts suggest that having
an incomplete/mistaken representation is the reason for failure to solve a problem, and
the faulty representation can be restructured into a better form by changing the
activation level of different memory items in the semantic network. Both the
unconscious and conscious accounts have received support from empirical studies.
However, it is difficult to draw a cross-study conclusion as to which account has better
predictive power. Differences in the settings of insight problems used in these studies do
not allow a direct comparison between these two theories. The insight problems used in
verifying RCT are knowledge-rich, one-move matchstick problems where only one
move is required to change the initial state to the goal state. They do not require the use
of any heuristic search to monitor progress. Therefore, matchstick problems have not
been used in testing CSPT. CPST receives support from other types of insight problem:
knowledge-lean problems. Solving a knowledge-lean problem (e.g., the nine-dot
problem) does not require the use of large amount of task-related knowledge. Therefore,
19
it is difficult to apply RCT to predicting the performance in solving knowledge-lean
insight problems.
In an attempt to fill this research gap, Jones (2003) used the “Car Park Game” as an
insight problem to test whether successful insight problem solving is the outcome of
relaxation of knowledge-based constraints or a switch in search strategy. The “Car Park
Game” is a visual problem in which individuals are asked to maneuver a taxi out of a car
park. The pathway from the taxi to the exit is obstructed by other cars. In the task, all the
cars can be moved forward or backward. Solving the problem requires individuals to
move cars away from the exit pathway so that the taxi car is able to leave the car park.
Jones suggested that individuals would hold a self-imposed constraint that the taxi
cannot be moved until the exit is clear, and therefore they would exclude the critical
move, which is moving the taxi even when the pathway is not cleared yet to enable the
movement of other cars. Participants would move the taxi only if that self-imposed
constraint was relaxed.
Jones predicted that, if RCT is correct and the self-imposed constraint causes
impasse and post-impasse restructuring is the relaxation of the unnecessary constraint
relaxation, then an impasse at the point prior to moving the taxi is expected. According
to CSPT, how quick an individual reaches an impasse depends on the effectiveness of
search within the problem space, for example, how far ahead an individual can look.
Therefore, Jones suggested that if CSPT is correct, the number of moves an individual
can look-ahead during problem solving should predict how quick he/she reaches an
impasse.
In addition to this, RCT suggests that restructuring relies on the activation of
relevant knowledge and the de-activation of irrelevant knowledge. Thus, Jones predicted
that if RCT is correct, solving a problem with a similar layout beforehand should
facilitate performance on the Car Park Game problem because the practice problem
20
would activate relevant knowledge. Jones proposed that CSPT should predict no
facilitation effect as successfully insight problem solving relies on a task-general search
strategy, but not the activation of task-relevant knowledge.
Jones analyzed participants’ eye-movements during problem solving as well as
their performance data, and suggested that both theories account for part of the
individuals’ problem solving behavior. The data indicated that participants reached an
impasse before moving the taxi, and also, solving the relevant practice problem can
enhance performance on the Car Park Game. These were in line with Jones’ predictions
based on RCT. However, the data also support predictions based on the CSPT theory,
showing that participants with a look-ahead of 3 performed far better than those with a
look-ahead of 1 or 2 steps. These findings imply the possibility that these two theories
may not be opposing each other.
However, the validity of Jones’ study is seriously in question because of some
major conceptual and methodological flaws in the study. First, the key difference
between RCT and CSPT has not been addressed in Jones’ study. RCT suggests that the
mechanisms that lead to restructuring are the unconscious processes operating within the
semantic network, while CRPT views restructuring as a conscious switch in search
strategy. However, Jones has not investigated whether restructuring occurs at a
conscious or unconscious level.
Second, it is uncertain whether individuals would really hold the constraint “the
taxi cannot be moved until the exit way is clear” while solving the Car Park Game.
Individuals usually learn from their past problem-solving experience and develop
different relevant constraints to guide the problem solving processes. However, the
“moving the car until the exit is clear” is not a necessary condition of real-life car
parking, and therefore, participants in Jones’ study may not hold this constraint when
solving the Car Park Game, implying that the Car Park Game may not be an insight
21
problem.
Third, the Car-Park-Game is a visual problem that does not require the use of a
huge amount of task-related knowledge. Hence, it is not an appropriate task to examine
whether restructuring is the outcome of the occurrence of the processes within a
semantic network. The Car Park Game is not an ideal task to test CRPT either. When
solving the Car-Park-Game, there are only two types of moves: move that can make
maximum immediate-progress (moving the car away from the pathway), or move that
make no immediate-progress at all (not moving any car away from the pathway).
Because of the narrow variation in the degree of progress among the available moves,
the problem-solving behavior may not truly reflect the impact of inappropriate
move-selective criteria on insight problem-solving.
Fourth, while CSRT does not currently contain a learning component since the
theory has not been developed with this in mind, it is a mistake to suggest that CSRT
predicts null transfer effects from training. There is extensive empirical evidence from
expert-novice research showing that past task-related experience can change the
criterion an individual uses in evaluating the moves (see the review in Chi et al., 1981).
Because of these concerns, the conclusion of Jones’ study is questionable.
Studying the role of incubation in insight problem-solving may serve a valuable
role in addressing the nature of restructuring, and testing between RCT and CSPT.
Incubation is the period of time during which a problem solver shifts attention away
from the unsolved problem to another area after a first unsuccessful attempt at solving
the problem. It has been suggested that performing incubation tasks can divert attention
away from the problem, and this would facilitate unconscious processing underlying
restructuring (Bowers, Regehr, Balthazard, & Parker, 1990; Seifert et al., 1995; Smith,
1995; Smith & Blankenship, 1991; Yaniv & Meyer, 1987). Examining the incubation
effects found with different types of insight problems may offer evidence for deciding
22
whether post-impasse restructuring is an unconscious process, and whether the processes
underlying restructuring arise at a semantic-level, a meta-cognitive level, or both.
Proposed Mechanisms of Incubation
Based on geniuses’ introspective reports of their idea discovery processes, Wallas (1926)
identified that insightful thinking process can be divided into five stages: Preparation,
Incubation, Intimation, Illumination, and Verification. Preparation is the period of time in
which the problem solver gathers information to formulate a problem representation, and
makes an initial attempt to solve it but reaches an impasse at the end. Incubation is the
period of time that the problem solver shifts attention away from the unsolved problem to
another area after the first unsuccessful attempt at solving the problem. Intimation is the
stage that the problem solver gets a feeling that the fully conscious flash of solution is
coming. Illumination is the moment that the solution to the problem suddenly appears in
the problem solver’s mind. Verification, the last stage of the creative thinking process, is
the period of time that the problem solver evaluates and modifies the solution into a better
form. Wallas’ (1926) is one of the earliest models describing the insight thinking process
systematically, and it provides a direction for researchers to study the insight process
systematically and empirically.
Several hypotheses have been proposed to account for the role of an incubation
period on insight problem solving, and similar to the debate on the nature of insight,
they can also be divided into two main kinds: conscious-work and unconscious-work.
The conscious-work hypothesis holds that incubation effects are due to issues such as
reduction of mental fatigue (Posner, 1973) or additional covert problem solving during
the incubation period (Browne & Cruse, 1988; Posner, 1973). Both sources implicate
changes in consciously controlled problem-solving activities during an incubation period.
In contrast, the unconscious-work hypothesis suggests that positive incubation effects
23
are the result of gradual and unconscious problem-solving processes that occur during
an incubation period (Bower et al., 1990; Simon, 1966; Smith, 1995; Smith &
Blankenship, 1991; Seifert et al., 1995; Yaniv & Meyer, 1987).
Three different unconscious processes have been proposed to account for incubation
effects. The first involves eliciting new knowledge: over time, activation will spread
towards previously-ignored but relevant memory items. Even if relevant items do not
receive above-threshold activation, this process can still sensitize individuals to related
concepts, and thus they will be more likely to make use of external cues to solve a
problem. In addition, partially-activated concepts may combine with others to yield
fortuitous insightful ideas (Bower et al., 1990; Smith, 1995; Smith & Blankenship, 1991;
Yaniv & Meyer, 1987). The second hypothesis is selective forgetting: an incubation
period will weaken the activation of inappropriate solution concepts or/and constraints
that distract individuals during initial attempts, allowing a fresh view of the problem
(Smith, 1995; Smith & Blankenship, 1991). The third hypothesis is opportunistic
assimilation, in which an individual’s mental representation of a problem will be
re-organized into a more appropriate and stable form after initial unsuccessful attempts.
The individual is then able to capitalise upon relevant external information or to
re-arrange problem information in a manner that allows a solution to be found more
readily (Seifert et al., 1995). The first two hypotheses suggest that the incubation effect or
the emergence of the insightful solution can be attributed to autonomous processes among
semantic networks. The Opportunistic-Assimilation hypothesis does not impose any
presumption on where the restructuring occurs; it may be a switch in move-selection
strategy (MacGregor et al., 2001) or the spread of activation at a semantic level (Yaniv &
Meyer, 1987).
The conscious- and unconscious-work accounts generate different predictions
concerning the effects of activities that individuals engage in during an incubation period.
24
According to the conscious-work account, individuals benefit most from an unfilled
incubation period, as this gives them an opportunity either to relax, reduce fatigue, or to
continue working on the problem. In contrast, unconscious work accounts suggest that
unconscious problem-solving processes occur when individuals shift their attention away
from the problem to other mental activities. Thus, a certain level of involvement in other
tasks during an incubation period may facilitate post-incubation problem-solving.
A number of experimental studies have examined the role of task type presented
during an incubation period. The experimental paradigms of these incubation studies are
fairly uniform: one group of participants is interrupted with an incubation period (having
a break or performing other tasks) while solving a problem, whereas the other group
works on the problem continuously. Performance differences between these two groups
are then compared. The findings of the published studies do not give unconditional
support to either the unconscious-work or the conscious-work accounts. Patrick (1986)
found that participants who had a filled incubation period outperformed those who had an
unfilled incubation period. However, Browne and Cruse (1988) reported the opposite
pattern: participants who took a rest during an incubation period performed better than
those who had to perform tasks during an incubation period. There are also studies that
report the same level of performance by participants with filled and unfilled incubation
periods (Olton & Johnson, 1976, Smith & Blankenship, 1989). However, these studies
vary in terms of the length of incubation period, the target problems tackled, and the
nature of the interpolated tasks during the incubation period.
Because of the inconsistent findings concerning incubation, some researchers have
doubted the existence of the effect, in particular rejecting the unconscious-work
hypotheses (Brown & Cruse, 1988; Olton & Johnson, 1976; Perkins, 1995). One
explanation for conflicting findings it that there are procedural moderators other than task
type that influence the occurrence of problem-solving processes during an incubation
25
period, such as the length of the incubation period, or the nature of the problem. The
following section presents the likely key moderators of incubation, which is achieved by
reviewing the methods used in previous studies. A particular focus of this review is to
identify moderators that might discriminate conscious-work and unconscious-work
hypotheses and also the mechanisms (reduction of fatigue, additional work, activation of
new information, forgetting, restructuring) that might underlie each hypothesis.
Procedural Moderators of the Incubation Effect
Besides the incubation task moderators, other experimental parameter may also have an
impact on the incubation effect. By reviewing the methods used in previous studies, a
number of procedural moderators are identified. The following section is a discussion on
some potential moderators. Note that many moderators in past studies (e.g., number of
trials, participant characteristics) might have been included, but we focus on those we
believe fundamental to discriminating between conscious-work and unconscious-work
hypotheses, and also those relevant to the mechanisms (reduction of fatigue, additional
work; activation of new information, forgetting, restructuring) that might underlie each
hypothesis.
The interpolated task used during the incubation period
Various types of interpolated task have been used in past studies, and they can be
divided into tasks of high or low cognitive demand. Examples of high cognitive demand
tasks include mental rotation, counting backwards, and visual memory tests, whereas
reading is commonly adopted as a low cognitive demand task. High demand tasks should
fully occupy the individual’s mind, and prevent further conscious work on the unsolved
problem. Some studies report that undertaking a high cognitive demand task during an
incubation period is beneficial to the problem solving process (Kaplan, 1990; Patrick,
26
1986; Segal, 2004). Nonetheless, studies using low cognitive demand tasks that do not
require individuals to focus their conscious attention on task performance report similar
benefits (Beck, 1979; Silveira, 1972; Smith & Blankenship, 1989).
Length of the preparation period
During the preparation period, problem solvers gather information to formulate a
problem representation and make initial attempts to solve, which may lead to an impasse.
Although a problem may not be solved during the preparation period, this does not mean
that the effort the problem solver spends on the problem is fruitless. Schank (1982, 1999)
and VanLehn (1988) both suggest that failure in problem solving is important in the
human learning process. Studies by Palatano and Seifert(1994) and Seifert et al. (1995)
have found evidence of a Zeigarnik effect in insight problem solving (Zeigarnik, 1927,
1938), where individuals remembered the problems on which they got “stuck” better
than those solved immediately. Seifert et al. hypothesized that having a better memory
for failed problems might help individuals return efficiently to the problem once relevant
new information is encountered during an incubation period, thereby maximizing the
chance of solving. Evidence concerning this prediction has been obtained in an
empirical study carried by Silveira (1971), showing that problem solvers performed
better with longer preparation and incubation periods
Length of the incubation period
Longer incubation periods may allow more additional problem-solving activity, or
allow a greater degree of forgetting of misleading items or spreading of activation
memory. Thus, problem solvers may show a larger performance improvement when they
return to the problem after a long incubation period than after a short one. Some studies
report evidence supporting this contention (Silveria, 1971, Smith & Blankenship, 1989,
27
Beck, 1979, Fulgosi & Guilford, 1968). However, it is difficult to draw cross-experiment
conclusions, since there is no standard operationalization of what constitutes “long” and
“short” incubation periods. In Smith and Blankenship’s study (1989), for example, a
15-min incubation period was defined as a long incubation period, and they reported that
participants receiving this length of incubation period performed better than those
receiving a 5-min incubation period. However, in Beck’s (1979) study, a 20-min
incubation period was considered to be short, and participants’ performance in this group
did not differ from the control group. Kaplan (1989) suggested that, to judge whether the
incubation period is short or long, the length of time that problem solvers spend on initial
attempts to solve (named the “preparation period” by Wallas, 1926) should also be taken
into account. Kaplan found that a larger incubation effect was observed after increasing
the ratio of the length of preparation period to incubation period. Thus, in addition to
including incubation and preparation periods as separate moderators in the meta-analysis
reported below, a secondary analysis was also undertaken using the ratio of preparation to
incubation time as an alternative moderator.
Nature of the problem
Various different types of problem have been used in incubation studies. Some
problems, which are termed ‘creative problems’ here, require individuals to produce
multiple new ideas to meet a specific brief. For example, a verbal divergent-production
task is the Consequences test(e.g., “What would be the result if everyone suddenly lost
the ability to read and write”; Fulgosi & Guilford, 1967). Typically, there is no right or
wrong answer to these kinds of problem and performance is assessed in terms of the
numbers of solution ideas that are generated.
Other problems require individuals to discover a specific target solution that is
known in advance by the experimenter. Problems of this kind studied in the literature on
28
incubation are generally of a type described as ‘insight’ problems, in that they require
the solver to reject initial solution ideas by achieving insight into an alternative strategy
or knowledge domain. The insight problems used in the incubation studies can be
divided into visual problems, which typically require the solver to consider a
visuo-spatial array of the problem (e.g., the nine-dot problem; Scheerer, 1963), and
linguistic problems, which typically require the solver to consider linguistic information
related to the problem. The Remote Associates Task (RAT; Mednick, 1962) is one of the
most commonly used linguistic problems in incubation studies. In each RAT, three
stimulus words are presented to individuals, who then have to think of a fourth word that
can form association with each of the three words. For example, if the three stimulus
words of a RAT are “electric”, “wheel”, and “high”, the fourth word can be “chair”.
Bowden & Jung-Beeman (2003) have developed a pool of remote associate problems,
and collected the normative data regarding the resolution rate and response time for the
problems. The classification of insight problems into visual- and linguistic-based is
supported by research findings from Gilhooly and Murphy (2005) showing that solving
visual and linguistic insight problems require different types of cognitive skills.
In the remainder of this chapter, problem types are referred to as creative, visual,
and linguistic. Descriptions of the types of problem used in incubation studies are
illustrated in Appendix A. Problem type is likely to be an important determinant of
incubation, since each type creates seems to create different task demands. For instance,
the nine-dot problem appears to require the solver to restructure an initially faulty or
incomplete problem representation in searching for a representation that allows solution,
while the Consequences Task appears to require the activation of as wide a range of
different concepts as possible.
29
The presence of solution-relevant cues
Some unconscious processes proposed to explain incubation effects are purely
internal and independent of the external environment, such as the inhibition of irrelevant
memory (Smith, 1995; Smith & Blankenship, 1991) and the recombination of
partially-activated concepts (Bower et al., 1990). Others stress interactions with the
external environment, such as the proposal that spreading activation can partially
activate previously-ignored relevant memory and therefore sensitize the problem solver
to chance encounters with related stimuli (Seifert at al., 1995). A few studies have
examined effects of the presence of cues during an incubation period (Browne & Cruse,
1988; Dodds, Smith, & Ward, 2002; Dorfman, 1990; Driestadt, 1969, Olton & Johnson,
1976). Most failed to find any positive effect of cues on the incubation effect. However,
the failure reported in these studies may be due to other factors, such as the difficulty of
the unsolved problems. In order to have a fair evaluation on the impact of this moderator,
we first have to isolate the effect of other moderators on the incubation effect.
Misleading cues
Another factor that may influence the occurrence of incubation effects is the
presence of misleading cues. Smith and Blankenship (1989) carried out a series of
experiments to examine the effect of an incubation period on solving Remote Associates
Tasks (RATs), in which participants had to find a word that might accompany each of
three presented words. Smith and Blankenship presented cues (shown here in italics)
comprising misleading associates and the target word next to each of the three stimulus
words. An example of a misleading RAT is: SHIP ocean, OUTER inner, CRAWL floor.
The target solution is “space”. Performance improvements after an incubation period
were observed only when participants solved tasks containing misleading cues. They
concluded that a problem solver who is fixated on misleading information benefits more
30
from an incubation period. The misleading cues data provide critical support for
forgetting-based explanations of incubation. The presence of misleading cues is
therefore one of the potential moderators examined in this meta-analysis.
Summary
The aforementioned factors are potential moderators of incubation effects.
Understanding the link between the moderators and incubation not only helps us to find
the optimal settings to demonstrate incubation effects, but also offers insight into the
underlying mechanisms of an incubation period. There have been two reviews to date of
relations between different procedural variables and the incubation effect, and both are
qualitative in nature. Olton’s (1979) review of past incubation studies led him to
question the existence of incubation effects, given that no experimental paradigm
appeared to demonstrate incubation effect reliably. Yet, a limited number of studies were
available at that time: only 10 incubation studies were included in his review. A recent
review by Dodds, Ward and Smith (in press) with more studies included suggested that
several variables may interact to influence the effectiveness of an incubation period.
However, the qualitative nature of their review led them to conclude that findings of past
studies are too divergent and that more studies are needed to assess the impact of each
variable and to identify the optimum settings for an incubation effect.
The wide variation in experimental parameters among studies makes it difficult to
draw cross-experiment conclusions from a qualitative review. To overcome these
problems, a systematic meta-analytic review is needed. Meta-analytic review allows a
quantitative evaluation of research domains that describes the typical strength of the
effect or phenomenon, and also the relation of each moderator to the size of the effect by
using statistical analysis methods (Rosenthal, 1995). In this thesis, a meta-analysis,
reported in the next chapter, was carried out to verify the existence of a positive
31
experimental incubation effect, and more importantly, to examine the impact of the
procedural moderators on the incubation effect.
Attentional Resource Allocation Styles and Incubation
The wide divergence of procedural moderator parameters in past incubation studies may
be one of the reasons for inconsistent findings of incubation. However, there are some
conflicting findings that cannot simply be attributed to the differences in experimental
settings. Several studies have reported failure to replicate positive incubation effects
even under the same experimental settings (e.g., Olton & Johnson, 1976), implying that
there must be some internal factors influencing incubation effects.
Some studies have examined the relations between the magnitude of incubation
effects and individual differences in high-level cognitive abilities, such as intelligence or
general problem solving ability. The findings are, however, conflicting, in that both
positive (Mednick, Mednick, & Mednick, 1964; Moss, 2002) and negative (Murry &
Denny, 1969; Smith & Blankenship, 1991; Moss, 2002) correlations were found
between incubation and ability. One possible explanation is that high-level cognitive
ability is the composition of different basic cognitive and non-cognitive characteristics,
such as attentional style, memory capacity, and level of motivation. Each of these factors
may have a unique impact on incubation effects, and they may also interact with each
other and thus influence incubation effects. Therefore, the relation between high-level
cognitive ability and incubation effect size is a complex one. A more systematic
approach to examining individual differences in incubation effects would be first to
investigate the impact of each of the basic cognitive characteristic on the incubation
effect, before examining the combined effect of them.
Attentional allocation style is suggested to be one of the basic cognitive
characteristics that affects the occurrence of incubation effects (Ansburg & Hill, 2003;
32
Finke, Ward, & Smith, 1992; Martindale, 1995). Individuals have different attentional
allocation styles, some showing a tendency to allocate attention widely to different
stimuli, others being more likely to focus on a single stimulus. For those with a focused
attentional style, performing an incubation task would force them to allocate their
cognitive resources in a more diffused manner. If this led to enhanced problem-solving
performance, this result would suggest the occurrence of some unconscious problem
solving processes. For individuals who habitually allocate their attention broadly, they
are always in a mental state that facilitates the occurrence of unconscious problem
solving process, and thus, performing incubation tasks may not be that helpful to them
(Ansburg & Hill, 2003; Finke, Ward, & Smith, 1992; Martindale, 1995). This
interpretation implies that incubation periods do not benefit all the individuals to the
same degree, and the mixed findings on the past incubation studies may partly due to the
lack of control for this cognitive characteristic.
Methodological Problems of Past Incubation Studies
The literature on incubation effects and individual differences in cognitive
characteristics suggests that the likelihood of occurrence of proposed incubation
mechanisms depends on various procedural and cognitive factors. To generate strong
data to test this contention, the traditional paradigm of past incubation studies, which
only assesses performance improvement rather than specific underlying mechanisms,
has to be modified.
The experimental paradigms of most past incubation studies are rather uniform: one
group of participants has an incubation period while solving the problem, whereas the
other group is asked to solve the problem continuously. Researchers have then compared
performance differences in solving problems between these two groups of participants.
The disadvantage of simply measuring performance change is that it cannot offer direct
33
evidence to reject or support the occurrence of proposed conscious or unconscious
processes. Non-significant performance improvement does not necessarily imply the
absence of any unconscious process during the incubation period. It is possible that
unconscious processes occur during an experimental incubation period, yet, the effect
that they generate is not strong enough to cause any influence on an individual’s
problem solving performance within the current experimental conditions. However, even
if a significant positive performance improvement is observed, it still cannot verify the
occurrence of any specific unconscious processing. Performance improvement could be
the result of selective-forgetting, spreading-activation, problem-restructuring, the
combination of any of them, or none of them. Therefore, in order to offer more direct
evidence for the occurrence of these problem solving processes, a more specific
assessment of processes underlying incubation effects is needed.
To test for the presence of spreading-activation and selective-forgetting, a more
direct measurement is to assess individuals’ sensitivity to relevant and irrelevant
memory items before and after an incubation period. Yaniv and Meyer (1987), in an
attempt to confirm the occurrence of spreading-activation underlying incubation, asked
participants to solve a list of rare-word-definition tasks, and then measured their lexical
decision time for answers of the unsolved word definition tasks and matched control
words. They found that participants made quicker lexical decisions about the answers of
unsolved tasks than control words, and suggested that the residues of initial activation
persist even when individuals are not currently solving the problem, and this would
sensitize individuals to assimilate relevant external cues encountered subsequently. This
finding has offered potential bases for the incubation effect. However, one problem of
Yaniv and Meyer’s study (1987) is that the rare-word-definition task is a recall task, and
it is uncertain if similar spreading-activation would also occur in insight problem
solving. Sio and Rudowicz (2007) adapted the lexical decision tasks in assessing the role
34
of an incubation period when solving RATs. An increase in sensitivity towards relevant
memory items after an incubation period was found when solving RATs embedded with
misleading cues. However, they did not also investigate whether enhanced
problem-solving performance arose as a result of incubation. If spreading activation is
the source of incubation effects, then enhanced performance to RATs should be
accompanied by decreased lexical decision latencies.
The presentation of lexical decision tasks containing solution-relevant words could
also serve as a relevant cue to problems, and this helps test the
Opportunistic-Assimilation hypothesis. According to the Opportunistic-Assimilation
hypothesis, incubation periods help problem restructuring and increase the chance of
assimilating external cues. It implies that individuals who receive relevant cues (e.g.,
relevant lexical decision tasks) during an incubation period would outperform those who
receive cues in a non-incubation condition or those who do not receive any cues during
an incubation period. Comparing individuals’ performance improvement among
No-Incubation and incubation conditions can offer empirical evidence to test the
Opportunistic-Assimilation hypothesis.
In summary, the preceding discussion suggests the need for a quantitative review of
the past incubation studies, and provides a context for the present set of experiments. To
this end, a meta-analysis was carried out for assessing the impact of each of these
aforementioned procedural moderators, and interactions between them, on incubation
effects. Four experiments were then carried out to examine further the meta-analysis
findings, and specifically to investigate the three unconscious-processing hypotheses:
spreading-activation, selective-forgetting, and cue-assimilation. In these experiments,
besides measuring incubation effects on performance levels, lexical decision tasks were
adapted for a more direct assessment of the occurrence of the three proposed
unconscious processes.
35
CHAPTER III: THE IMPACT OF PROCEDURAL MODERATORS ON
INCUBATION EFFECTS: META-ANALYSIS
This meta-analysis addresses two questions: first, is there reliable evidence for
incubation; and second, what are the most influential moderators? To address the first
question, the effect size of incubation reported in each available study was computed.
Given that the variability among effect sizes is likely to be greater than that resulting
from subject-level sampling, a random effects model was adopted in this meta-analysis.
A heterogeneity test was carried out to verify this assumption, and then the weighted
mean under the assumption of a random effects model was computed and assessed to
determine if it was significantly larger than zero. To address the second question,
weighted least-squares linear regressions were carried out using the aforementioned
moderators as predictor variables and the incubation effect size in each study weighted
by the inverse of its variance as the criterion variable. The results of the regression
report the independent contributions of each potential moderator to the incubation effect,
while controlling for all other moderators. This approach allows us to summarize the
past studies systematically even though they vary widely in numbers and type of
experimental parameters. In addition, interactions between different moderators, such as
the nature of the interpolation task during the incubation period and the nature of the
problem, were examined.
36
Method
Literature Search
Publications that contained studies relevant to a meta-analysis of incubation were
collected through a search of the ERIC, PsycInfo, PsycArticles, and MEDLINE
databases using the keyword incubat*, intersected with one of fixation, creativ*,
divergent*, insight*, or problem. Then, references given in all the obtained articles were
systematically searched for additional relevant publications. There is a concern that
studies with statistically significant results are more likely to get published than those
without significant results, and this may lead to a biased retrieval of studies. To
ameliorate this to some extent, similar literature searches were carried out in the
ProQuest Digital Dissertations databases and using Google Scholar™ for retrieving PhD
dissertations, unpublished papers, and conference papers concerning the incubation
effect. In total, 37 relevant publications were identified and obtained. Studies meeting
the following criteria were assimilated in the analysis.
1.
The settings and difficulty of the problems were the same among all the
experimental conditions;
2.
The total length of time that participants could spend on solving the problem
consciously was the same among all the conditions;
3.
The study included a control (no incubation) group, and participants in that
group worked on the problem continuously;
4.
Participants’ problem solving performance in pre and post- incubation periods
was measured;
5.
The study reported information that allowed the computation of an effect size.
The 1st and 2nd selection criteria ensured that tasks were presented in an identical way
among different conditions, and any between-condition performance differences could
be attributed to differences in settings of the incubation period. The inclusion of criterion
37
3, a control condition (no break between the first and the second attempts at the
problem), is essential to provide a baseline against which performance in incubation
conditions can be compared. Only publications that assessed the problem solving
performance in both first and second attempts were included in the analysis (criterion 4).
Therefore, some studies (e.g., Sio & Rudowicz, 2007; Yaniv & Meyer, 1987) that did not
assess post –incubation problem solving performance were excluded. The information
required for computing effect sizes is discussed in the section “Estimation of effect
sizes”. Eight publications were excluded because the experimental studies contained
within them failed to meet one or more of the aforementioned criteria. The specific
reasons for excluding the publications are described in Appendix B, which also
describes the settings of studies included in the meta-analysis. Of the remaining 29
publications, 20 were refereed journal articles, 8 were PhD dissertations, and 1 was a
conference paper. The ratio of the refereed to other studies is 2.2:1, which is within the
suggested range of between 128:1 and 1:1 for including unpublished studies in an effort
to avoid publication bias (Thornton & Lee, 2000). Most publications included multiple
experiments, thereby allowing a reasonable sample size of independent studies (N = 117)
to be achieved.
Coding Procedure
Many of the experiments reported in the selected publications had two or more
experimental conditions, such as incubation periods of different length or different types
of task in the incubation period. For the sake of the meta-analysis, experiments with
more than one incubation condition were broken down into independent studies with
one incubation condition and one control condition. The same control group may be
included in more than one independent study, and compared with more than one
incubation condition. For example, in Goldman, Wolters, and Winogard’s (1992)
38
experiment, there were control, short-incubation-period, and long-incubation-period
conditions. The experiment was decomposed into two studies, one consisting of the
control and short-incubation-period conditions, and the other consisting of the control
and long-incubation-period conditions. To avoid inflating the degrees of freedom
available, the number of participants in the control condition was split across studies
entered into the analysis, a method advocated by Bar-Haim, Lamy, Pergamin,
Bakermans-Kranenburg, and van Ijzendoorn (2007).
There were also studies having more than one control condition. In such cases, the
control condition that had the most similar setting to the incubation condition was
chosen. For example, in Hansberry’s (1998) third experiment, participants had to solve a
list of RATs under one of three conditions: two control and one incubation. In one
control condition, the RATs were presented individually for 60 seconds. In the other
control condition, as well as in the incubation condition, each RAT was presented in two
separate 30-second blocks. Data from the latter control condition were therefore used in
computing the effect size, because this control condition and the incubation condition
had the closest settings in terms of RAT presentation.
After separating the experiments into numbers of independent studies, a standard
system was used to code each study. Background information on each independent study
(author, publication year, and the number of participants in each condition), as well as
potential moderator variables, were extracted. Table 3.1 presents the coding system used
in this meta-analysis. Appendix C presents the information extracted from each
independent study by using the coding system.
39
Table 3.1
Coding System Used in the Meta-Analysis
Variables
Coding Description
Author
Author (s) of the study
Year
Year the study was published
Total
Total number of participants
Problem Type
0 = creative problem (e.g., consequence task)
1 = visual problem ( e.g., radiation problem)
2 = linguistic problem ( e.g., remote associates task,
anagram, rebus)
Misleading Cues
0 = no misleading cues
1 = misleading cues embedded in the problem
Preparation Period
Amount of time spent on each problem before the
incubation period ( in minute)
Incubation Period
Length of the incubation period ( in minute)
Incubation Task
0 = rest
1 = low cognitive load task ( e.g., drawing a picture,
reading)
2 = high cognitive load task ( e.g., mental rotation task,
memory test)
Cues
Presence of relevant cues during the incubation period
0 = no cue
1 = yes
40
Estimation of Effect Sizes
The effect size, Cohen’s d, was computed for each study entered into the
meta-analysis. Cohen’s d in this meta-analysis comprised the difference in mean
problem-solving performance scores between the control and incubation conditions
divided by their pooled standard deviation (Hedges & Olkin, 1985). In some cases,
effect sizes had to be calculated from t and F values, frequencies or p values. If a
p-less-than value was given instead of an exact p value, the p-less-than value was treated
as an exact value, and an estimate of Cohen’s d was generated. For studies that did not
include any of the above-mentioned information but only provided statements of
non-significant differences between the control and the incubation groups, then Cohen’s
d was assumed to be zero. Among the studies that included multiple incubation
conditions, some provided a statement of non-significant performance differences
among the incubation conditions, and only reported the overall performance difference
between the control and the incubation conditions. In such cases, all incubation
conditions were assumed to generate the same magnitude of incubation effect sizes. Of
the 117 effect sizes, 88 were extracted directly from the means and standard deviations, t
value, F value, frequencies, or p value; 8 were computed from a p less than value; and
21 were estimated from statements of significance.
In some incubation studies, problem solving performance was assessed along
more than one dimension. For example, in the study carried out by Vul and Pashler
(2007), participants’ performance on RATs was measured in terms of the time spent on
solving RATs and the number of correct solutions. In such cases, a single effect size was
computed by averaging the effect size from each measure (cf. Durlak & Lipsey, 1991).
Following Hedge and Olkin’s (1985) suggestion for removing bias caused by
small sample studies, an unbiased effect size estimate was computed by multiplying the
effect size of each single study by a factor 1-3/ (4 (total N-2)-1), where total N is the
41
total number of participants of that study. Any unbiased effect size larger than 2 standard
deviations from the group mean was considered an outlier, and was recoded to the value
of the effect size found at 2 standard deviations, following a procedure for reducing the
bias caused by extreme effect sizes reported by Lipsey & Wilson (2001).
Heterogeneity Analysis
In this analysis, it was predicted that the variance in magnitude of the unbiased
effect sizes among studies was not due simply to sampling error but instead to the
difference in settings of each study (e.g., length of incubation period, nature of
incubation task, presence of cues). Therefore, analyses of the effect sizes should be
carried out under the assumption of a random effects model. To confirm the assumption
of heterogeneous distribution of effect sizes, a heterogeneity test was carried out before
running any analyses on the effect sizes. The standard measure of heterogeneity is the
Cochran’s Q test. The Q statistic is the weighted1 sum of squared differences between
the unbiased effect size estimate of each independent study and the weighted average
unbiased effect size estimate across studies. Q is distributed as a chi-square statistic with
k -1 degree of freedom, where k is the number of independent studies. If the Cochran’s
Q test for heterogeneity is statistically significant(Q is larger than the chi-square value
with k-1 degree of freedom), the assumption of the random effects model is supported.
1
The weighting was the inverse of the within-study variance of the effect estimate,
and the formula for the within-study variance was [(2 x square root of total N) + (N of
experimental x N of control x square root of the unbiased effect size)]/(2 x total N x N of
experimental x N of control), where N is the number of participants in that condition
(Cooper &Hedges, 1994). Three studies were excluded when computing the weighted
average unbiased effect size estimate and the Cochran’s Q value because they had a
within-subjects design, and thus all participants were involved in both control and
incubation conditions. Thus, the weighting formula could not apply to them.
42
Publication Bias
Prior to investigating the impact of potential moderators, a preliminary analysis
was undertaken to assess if a publication bias existed in the selection of studies despite
the inclusion of unpublished studies. A funnel plot of sample size against unbiased effect
size estimates was created. In the absence of any publication bias, it is expected that the
plot would be a funnel shape, such that the amount of scatter about the mean effect size
deceases with increasing sample size. In addition to checking the presence of publication
bias qualitatively, a weighted least-squares linear regression was carried out using the
unbiased effect size estimates as the dependent variable, and sample sized weighted by
the inverse of variance in a random effects model, which is the sum of the
between-studies variance2 (random variance component) and within-study variance of
the unbiased effect size. The regression slope (unstandardized coefficient of the
predictive variable) would be expected to approach zero if there is no publication bias
(Macaskill, Walter, & Irwing, 2001). The outcome of this analysis is reported below.
Regression Model Testing
Due to the wide variation in experiment settings among incubation studies,
observed incubation effect size differences may reflect the combined impact of different
moderators. Hence, weighted least-squares regression analyses were carried out to
reveal the true impact of each moderator on incubation effects. The regression analyses
were organized into two main sections. In the first section, the incubation studies were
first classified into different groups, in terms of the types of problem used, the cognitive
The between-studies variance was equal to [Q-(k-1)]/c, where Q is the Cochran’s
Q value and k was the number of studies. The formula for c was ((the sum of the inverse
of the within-study variance) – (the sum of the square of the inverse of the within-study
variance)/(the sum of the inverse of the within-study variance)), as suggested by Cooper
and Hedges (1994).
2
43
load of the incubation tasks, the presence of misleading cues, and the presence of
relevant cue during an incubation period. Within each sub-group, the random variance
component of the studies was computed. A larger-than-zero random variance component
implies that the variability of effect sizes within these studies is not simply due to
subject-level sampling error (Lipsey & Wilson, 2001). A weighted least-squares
regression analysis was carried out as a follow-up analysis to model the effect sizes. The
unbiased effect size estimate weighted by the inverse of variance was the outcome
variable of the regression analysis. The predictor variables included Problem Type,
Misleading Cues, Cues, Incubation Task, Preparation Period, and Incubation Period3.
The categorical variables were represented with the appropriate number of dummy
coded vectors. The categories “Creative Problem” (Problem Type), “Rest” (Incubation
Task), “No misleading cues” (Misleading Cues) and the “No Cue” (Cues) were used as
reference groups in the analysis, and their coefficients were restricted to zero.
The second section of the analyses investigated the general impact of the
moderators on the incubation effect sizes. In this section, all the incubation studies were
group together, and a weighted least-squares regression analysis was carried out to
investigate the general impact of each moderator. Again, the weighted unbiased effect
size was the outcome variable of the regression analysis. The predictor variables were
Problem Type, Misleading Cues, Cues, Incubation Tasks, Preparation Period, and
Incubation Period. Another weighted least-squares regression was carried out to examine
the interaction between the categorical variables Problem Type and Incubation Task. The
predictor variables were the appropriate number of dummy coded vectors and the
multiplicative terms of these two categorical variables, as well as the variables
Misleading Cues, Preparation Period, Incubation Period, and Cues. A more detailed
The standard deviation of the weighted mean, also known as the “standard error”,
was calculated as the square root of 1/wi, and the 95% Confidence Interval was
calculated as the weighted mean +/- 1.96*standard deviation of the weighted mean
(Hedges & Olkin,1985).
44
3
description of the selection of the dummy coded vectors and the multiplicative terms of
“Problem Type” and “Incubation Task” is presented in Appendix D.
Results
One hundred and seventeen studies were included in this meta-analysis. The total
number of participants was 3606, and the median number of participants per study was
25. An unbiased effect size estimate was computed for each independent study. Among
these studies, 85 of them report positive effect sizes. The unbiased effect size estimates
range from -.71 to 4.07, and the median was .26. The unweighted mean of the unbiased
effect size estimate was .41, with a standard deviation of .71. The upper and the lower
bound 95% confidence interval were .54 and .28. Unbiased effect sizes larger than 2
standard deviations from the mean were recoded to the value of the effect size found at 2
standard deviations. Table 3.2 gives the stem-and-leaf display showing the distribution
of the unbiased effect sizes. The unweighted mean of the adjusted unbiased effect size
estimate was .36, with standard deviation .51. The upper and the lower bound 95%
confidence interval were .26 and .45. The confidence interval does not include zero,
implying that the estimate of mean unbiased effect size is significantly larger than zero.
45
Table 3.2
Stem-and-Leaf Display of 117 Unbiased Effect Sizes
Stem
Leaf
-.7
1
-.5
9, 8, 8
-.4
0
-.3
8, 5, 3
-.2
-.1
8, 7, 4, 4, 0
-.0
3
.0
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 4, 5, 5, 6, 6, 7, 7, 9
.1
0, 0, 0, 0, 0,1, 1, 3, 4, 4, 4, 5, 7, 7
.2
1, 4, 5, 8, 8
.3
1, 4, 6, 6, 7
.4
0, 0, 1, 1, 2, 2, 4, 5, 7, 7, 8, 8, 9
.5
0, 2, 2, 6, 6, 6, 6, 6, 6, 7
.6
2, 2, 4, 5, 6, 6, 6
.7
0, 1, 2, 4, 4
.8
6
.9
9, 9
1.0
5, 5, 7, 7, 8, 9
1.1
7
1.2
1.3
1.4
1.5
46
1.6
8
1.7
0
1.8
2, 2, 2, 2
Note. Outliers were recoded.
The heterogeneity statistic, Cochran’s Q, was 173.99, which is significantly larger
than the Chi-Square critical value, df = 113, p < .001. This supports the use of a random
effects model. The variance of each unbiased effect size in random effects model was
the sum of the between-studies variance and within-study variance of the unbiased effect
size. The between-studies variance, also called the random variance component, among
these incubation studies was .0834. The mean of the weighted unbiased effect size
was .29, with .04 standard deviation3, and the 95% confidence interval was (.21, .39).
The non-zero confidence interval implies that the weighted mean is significantly larger
than zero. This answers our first question, showing the existence of a positive incubation
effect.
Figure 3.1 presents the funnel plot of sample size against estimated unbiased
effect size of each study in the meta-analysis. A weighted least-squares regression using
unbiased effect sizes weighted by the inverse of the variance as the dependent variable,
and sample size as the predictive variable, was carried out. The regression coefficient of
the predictive variable was not significantly different from zero, standardized coefficient
(β) = -.08, p = .41, suggesting the absence of publication bias. Thus no correction has
been made for publication bias.
47
Sample Size
Unbiased Effect Size
Figure 3.1.1 Funnel Plot of the Studies included in the Meta-Analysis.
Table 3.3 presents the weighted mean, standard deviation, 95% confidence interval,
and random variance component in each sub-group of each categorical moderator.
Six of the sub-groups (linguistic problems, creative problems, absence of
misleading cues, absence of relevant cues, high cognitive load task, unoccupied
incubation period) had larger-than-zero random variance components. New weightings,
under the assumption of a random effects model, were computed for each of the
sub-groups. Weighted least-squares regression analyses were carried out to find the
moderators that accounted for the effect size variability among these sub-groups. Small
numbers of studies using creative tasks and studies having unoccupied incubation
periods preclude the possibility of regression analyses with these moderators. An effect
of applying a weighting to this regression analysis is to under-estimate the original
standard error of each unstandardized coefficient. Thus, an adjusted standard error was
computed by dividing the original standard error by the square root of the mean square
48
residual, a procedure suggested by Lipsey and Wilson (2001). The corrected standard
error was used in the significance test (z-test) of each unstandardized coefficient (B).
Tables 3.4, 3.5, 3.6, and 3.7 present a summary of the regression analysis results of
each sub-group.
49
Table 3.3
The Random Variance Component, Weighted Mean, Standard Deviation, Standard Error, and 95% Confidence Interval of the Effect Size
Estimate by Each Categorical Moderator
Problem Type
Misleading Cues
Incubation Task
Cues
Linguistic
Verbal
Creativea
Yes
Noa
High Load
Low Load
Resta
Yes
No Cuea
65
35
14
29
85
76
22
16
32
82
.00281
0
.37418
0
.12004
.06478
0
.30000
0
.10409
Mean
.22
.26
.29
.35
.32
.24
.52
.46
.24
.34
Standard Deviation
.05
.08
.14
.09
.06
.05
.10
.18
.09
.06
.13,.32
.10, .41
.13, .35
.32, .72
.11, .82
.07, .41
.22, .45
Number of studies
Random Variance
Component
95% Confidence Interval
.02, .55
.17, .53 .20, .43
Mean comparison with the
t(77) = -.43 t(47) = -.19
t(112) = .30
t(90) = 1.17 t(36) = .28
t(112) = .90
reference group
Note. The lower confidence intervals of all the effects are above zero, suggesting that the mean is significant larger than zero, p < .05.
a
The reference groups in mean comparisons.
50
Table 3.4
Regression Model for Linguistic Problem Studies (N= 65)
Variable
SS
df
Mean Square
F
p
3.15*
.01
ANOVA significance test
Model
13.42
5
2.68
Residual
5.24
59
0.85
Total
63.66
64
Summary of the Regression Model
Corrected
β
B
z score
SEβ
Incubation Task
High cognitive load task
.06
.06
.21
0.27
Low cognitive load task
.54*
.43
.25
2.15
.15
.17
.12
1.24
.15
.12
.19
0.80
-.04
-.04
.13
-0.30
Misleading Cues
Ration of the preparation
period to the incubation
period
Relevant Cue
Note. Random Variance Component = .00281, R2 = .21, Another regression with the
same predicting variables, except replacing the variable “ration of length of the
preparation to the incubation period” by the variables “Incubation Period” and
“Preparation Period”, was carried out. The pattern of the results was comparable but it
had lower explanatory power, and lower significant level, R2 = .20, F(6, 58) = 2.47, p
= .03. Neither the variable “preparation period” nor “incubation period” was
significant. ANOVA = analysis of variance.
*p < .05.
51
Table 3.5
Regression Model for Misleading Cue Studies (N= 85)
Variable
SS
df
Mean Square
F
p
3.56*
.002
ANOVA significance test
Model
2.25
7
2.89
Residual
62.62
77
0.81
Total
82.87
84
Summary of the Regression Model
Corrected
β
B
z score
SEβ
Problem Type
Visual problem
-.59*
-.51
.19
-3.02
Linguistic problem
-.31
-.29
.18
-1.74
High cognitive load task
-.18
-.14
.22
-0.80
Low cognitive load task
.06
.04
.26
0.23
Length of the Incubation Period
< .001
.055
< .001
0.45
Length of the Preparation Period
.03*
.36
.01
2.16
Relevant Cue
-.03
-.02
.16
-0.20
Incubation Task
Note. Random Variance Component = .12004, R2 = .24. ANOVA = analysis of
variance.
*p < .05.
52
Table 3.6
Regression Model on No Relevant Cue Studies (N = 82)
Variable
SS
df
Mean Square
F
p
7.35*
< .001
ANOVA significance test
Model
29.34
6
4.89
Residual
49.94
75
0.67
Total
79.28
81
Summary of the Regression Model
Corrected
β
B
z score
SEβ
Problem Type
-.79**
-.66
.19
-4.10
-.22
-.21
.18
1.24
High cognitive load task
-.25
-.22
.21
-1.20
Low cognitive load task
.16
.12
.25
0.64
< .001
.02
0
Visual problem
Linguistic problem
Incubation Task
Length of the Incubation
0.13
Period
Length of the Preparation
3.01
.05**
.51
.02
.09
.07
.18
Period
Misleading Cue
0.53
Note. Random Variance Component = .10409, R2 = .37. ANOVA = analysis of
variance.
*p < .05. **p < .001.
53
Table 3.7
Regression Model on High Cognitive-Load Incubation Task Studies (N= 75)
Variable
SS
df
Mean Square
F
p
1.53
.19
ANOVA significance test
Model
7.36
6
1.47
Residual
66.19
69
0.96
Total
73.55
74
Summary of the Regression Model
Corrected
β
B
z score
SEβ
Problem Type
Visual problem
-.41*
-.34
.20
-2.01
Linguistic problem
-.28
-.30
.17
-1.58
Length of the Incubation
-0.87
-.01
-.12
.01
Period
Length of the
1.31
.02
.21
.02
Misleading cue
.08
.06
.16
0.51
Relevant cue
.09
.08
.14
0.62
Preparation Period
Note. Random Variance Component = .06478, R2 = .10. ANOVA = analysis of
variance.
*p < .05.
54
With the sub-group of studies using linguistic problems, low cognitive load tasks
generated larger incubation effects than rest alone, 
= .54, p < .05. Also, there was
an interaction between Problem Type and Incubation Task with this sub-group, such
that that low cognitive load tasks facilitated the incubation effect only when solving
linguistic problems.
With the “absence of misleading cue”, and “absence of relevant cue”
sub-groups, regression analyses reveal that, in the absence of these cues, individuals
solving visual problems had smaller incubation effects than those solving creative and
linguistic problems. There was also a positive impact of longer Preparation Periods on
the incubation effect sizes. The cognitive load of the incubation tasks did not have any
impact on the magnitude of the incubation effects in these two sub-groups. Note,
however, that the presence or absence of these effects with these specific sub-groups
does not necessarily imply the converse effects found with other sub-groups, hence a
shift to analysis of general impacts of each moderator.
In the second stage of the analysis, a weighted least-squares regression analysis
was carried out to look at the general impact of each moderator on the incubation
effect sizes. A summary of the regression analysis results is presented in Table 3.8.
The negative coefficients associated with “visual problem” and “linguistic problem”
indicate that individuals solving these two types of insight problem showed a smaller
incubation effect than individuals solving creative problems. A z-test was carried out
to compare the coefficients of “visual problem” and “linguistic problem”. The result
was not statistically significant, z score = -1.25, p > .05, suggesting the magnitude of
the incubation effect for visual and linguistic insight problems was comparable.
55
Table 3.8
Regression Model on All Studies (N = 114)
Variable
SS
df
Mean Square
F
p
4.08
< .001
ANOVA significance test
Model
25.56
8
3.20
Residual
82.22
105
0.78
Total
107.79
113
Summary of the Regression Model
Corrected
β
B
z score
SEβ
Problem Type
Visual problem
-.60**
-.52
.15
-3.61
Linguistic problem
-.31*
-.31
.14
-1.96
.16
.13
.12
1.16
.14
.10
.18
0.72
-.16
-.15
.14
-1.02
.03*
.35
.01
2.29
< .001
.04
< .001
< .001
-.01
-.01
.11
-0.09
Misleading cues
Incubation Task
Low cognitive-load
task
High cognitive-load task
Length of the Preparation
Period
Length of the Incubation
Period
Cue presented
Note. Random Variance Component = .0834, R2 = .24. ANOVA = analysis of variance.
*p < .05. **p < .001.
56
The length of preparation period was found to have a significant impact on the
magnitude of the incubation effect, β = .03, p < .05. Three bivariate correlations were
carried out to check for positive relationships between the length of preparation
period and the magnitude of the weighted incubation effect when solving the three
types of problem. There was a statistically significant positive correlation between the
weighted incubation effect size and the length of preparation period with visual
problems, r (35) = .40, p = .02, and creative problems, r (14) = .60, p = .03, but not
with linguistic problems, r (65) = -.04, p = .75.
Another weighted least-squares regression analysis was carried out to examine
the interaction between Problem Type and Incubation Task, using Misleading Cues,
Preparation Period, Incubation Period, Cues, and an appropriate number of dummy
vectors and the multiplicative terms of the variables Problem Type and Incubation
Task as predictor variables. Table 3.9 presents the results of this regression analysis.
To examine the interaction effects, the coefficient differences between their
multiplicative terms were examined to see if they were significantly larger than zero
by using z-tests. The details of equations for computing the coefficient difference can
be found in Table D1 in Appendix D. Table 3.10 presents the coefficient differences
between multiplicative terms.
57
Table 3.9
Regression Model on All Studies (with interactive terms, N= 114)
Variable
SS
df
Mean Square
F
p
4.15
< .001
ANOVA significance test
Model
35.60
12
2.97
Residual
72.18
101
.72
Total
107.78
113
Summary of the Regression Model
Problem Type x Incubation
Corrected
β
B
Task
z score
SEβ
Visual problem x Low
0.17
0.08
.35
0.50
0.90*
0.61
.35
2.60
0.72
0.43
.38
1.93
1.00*
0.97
.38
2.59
Visual problem
-1.29**
-1.10
.32
-4.00
Linguistic problem
-1.24**
-1.21
.40
-3.14
High cognitive load task
-0.91**
-0.81
.30
-3.00
Low cognitive load task
-0.12
-0.09
.33
-0.38
0.24
0.19
.15
1.54
< .001
.018
< .001
< 0.001
cognitive load task
Visual problem x High
cognitive load task
Linguistic problem x Low
cognitive load task
Linguistic problem x High
cognitive load task
Incubation Task
Misleading Cues
Length of the Incubation
58
Period
Length of the preparation
.03*
.01
.01
1.97
.03
.03
.13
.26
period
Answer was presented
Note. Random Variance Component = .08340, R2 = .33. ANOVA = analysis of
variance.
*p < .05. **p < .001.
59
Table 3.10
Summary of Coefficient Differences of the Regression Model on All Studies (with
interactive terms, N= 114)
The coefficient
The multiplicative terms
Corrected SE
z score
difference
Lij Hj – Lij Rj
0.08
.49
0.17
LijL j- LijRj
0.60
.50
1.21
LijL j- LijHj
0.52
.70
0.74
InjHj - InjRj
-0.01
.46
-0.03
InjL j- InjRj
0.05
.48
0.10
InjL j- InjHj
0.06
.66
0.09
CjHj - CjRj
-0.91*
.30
-3.00
CjL j- CjRj
-0.12
.33
-0.38
CjL j- CjH
0.79+
.45
1.77
InjHj- CjHj
-0.39
.47
-0.82
LijHj - CjHj
-0.25
.55
-0.45
LijHj- InjHj
0.14
.72
0.19
InjL j- CjLj
-1.11*
.47
-2.36
LijL j- CjLj
-0.52
.55
-0.95
LijL j- InjL j
0.60
.72
0.82
Inj Rj - Cj Rj
-1.29**
.32
-4.00
Lij Rj - Cj Rj
-1.24**
.40
-3.14
Lij Rj - Inj Rj
0.04
.51
0.08
Note. C = creative problem; Li = linguistics problem; V = visual problem; R = rest; L
= low cognitive demand task; H = high cognitive demand task; j= study j.
+
p = .08. *p < .05. **p < .001.
60
With creative problems, undertaking high cognitive load tasks was associated
with smaller incubation effects than with low cognitive load tasks or rest during the
incubation period, CjHj -CjRj = -.91, p < .05, and CjHj. - CjLj = -.79 , p = .08 . When
solving linguistic and visual problems, no differences were found for incubation
periods filled with low or high cognitive load tasks or rest. However, this regression
model has a problem in exploring the interaction between Problem Type and
Incubation Task. Among studies examining the role of an incubation period with
creative problems, 12 out of 14 studies employed high cognitive load tasks and the
remaining 2 studies employed rest during the incubation period. This unbalanced
distribution may cause bias when examining the interaction between Incubation task
and Problem Type. Hence, another regression analysis was carried out that excluded
studies using creative problems. This third regression model included the variables
Misleading Cues, Preparation Period, Cues, the dummy variables of Problem Type
(excluding Creative problem) and Incubation Task, and their multiplicative variables.
Appendix E presents a detailed description of the selection of the dummy coded
vectors and the multiplicative terms, and the equations for computing the coefficient
difference between the multiplicative terms. The results of this analysis is presented in
Table 3.11, and the coefficient differences between the multiplicative terms are
presented in Table 3.12
61
Table 3.11
Regression Model for Studies Excluding Creative Problem (N = 100)
Variable
SS
df
Mean Square
F
p
2.04*
.05
ANOVA significance test
Model
15.04
8
1.88
Residual
83.92
91
0.92
Total
98.96
99
Summary of the Regression Model
Corrected
β
B
z score
SEβ
Problem Type x
Incubation Task
Visual problem x Low
-.21
-.12
.28
-0.77
.09
.08
.24
0.38
-.15
-.15
.21
-0.68
High cognitive load task
-.01
-.01
.18
-0.05
Low cognitive load task
.45*
.41
.20
2.27
.13
.13
.11
1.10
.08
.20
.05
1.56
cognitive load task
Visual problem x High
cognitive load task
Problem Type
Visual problem
Incubation Task
Misleading Cues
Length of the preparation
period/incubation
62
Answer was presented
.01
.01
.11
0.13
Note. Random Variance Component = .00031, R2 = .15, Another regression with the
same predicting variables, except dividing the variable “ration of length of the
preparation to the incubation period” to the variables “Incubation Period” and
“Preparation Period”, was carried out. But the regression model was not significant,
F(9, 90) = 1.27, p = .10. Neither the variable “preparation period” nor “incubation
period” was significant. ANOVA= analysis of variance.
*p < .05.
Table 3.12
The Summary of Coefficient Differences of the Regression Model Excluding Creative
Problem (N= 100)
The multiplicative
The coefficient
Corrected SE
z score
terms
difference
LijHj – LijRj
-.01
.17
-0.05
LijL j- LijRj
.45*
.20
2.27
Lij Lj - LijHj
.46
.26
1.76
VjHj - VjRj
.08
.30
0.28
VjL j- VjRj
.24
.34
0.69
VjLj- VjHj
.15
.45
0.34
VjHj - LiHj
-.06
.32
-0.17
VjL j- LijLj
-.36
.35
-1.03
Vj Rj - LijRj
-.15
.21
-0.68
Note. Li = linguistics problem; V = visual problem; R = rest; L = low
cognitive demand task; H = high cognitive demand task; j= study j.
*p < .05.
63
The regression results indicate an interaction between Problem Type and
Incubation Task. When solving linguistic problems, a low cognitive-load task
generated significantly larger incubation effects than rest, LijLj -LijRj = .45, p = .05.
The difference between low and high cognitive-loads was in the same direction but
did not reach significance, and there was no significant difference between the rest
condition and high cognitive load condition. When solving visual problems, the effect
sizes among the three incubation conditions were comparable. This pattern of findings
is consistent with the previous regression analysis results. In addition, the exclusion of
creative problem studies makes the positive impact of a low cognitive-load incubation
period on linguistic problems more significant.
The model using the variables “Preparation Period” and “Incubation Period” was
found to be not significant, and the variable “Preparation Period” was not significant
in the analysis. This may be due to the decrease in the numbers of studies included in
the current regression model. Moreover, as mentioned above, a positive association
between the length of preparation period and incubation effect size was found when
solving visual and creative problems. Thus, the exclusion of creative problem studies
appears to decrease the significance of the variable “Preparation Period”, and the
significance level of the regression model using the “Preparation Period” as one of
the predictive variables.
Discussion
The meta-analysis supports the existence of incubation effects, and also identifies
some potential moderators, including the problem type, length of preparation period,
and the incubation task. Individuals solving creative problems were more likely to
benefit from an incubation period than individuals solving linguistic and visual
64
problems. Longer preparation periods gave rise to larger incubation effects. When
solving linguistic problems, a low cognitive-load task gave the strongest incubation
effects.
It is also suggested that the positive incubation effects found with creative
problems are a direct reflection of their multi-solution nature. When solving a creative
problem, individuals benefit from performing a wide search of their knowledge to
identify as many relevant connections as possible with the presented stimuli. Each
time an individual re-approaches the problem, performance is improved by extending
search to previously unexplored areas of their knowledge network. Incubation appears
to facilitate the widening of search of a knowledge network in this fashion.
Linguistic and visual problems typically have only one possible solution. In
order to solve them, individuals have to explore their memory or environment to look
for specific relevant knowledge or to adapt a specific strategy. Widening search to
new items of knowledge may not be facilitative if the solution to a problem lies
within already activated knowledge that is currently represented inappropriately.
Under this account, incubation supports knowledge activation, but it does not support
restructuring.
Another finding of the meta-analysis was the beneficial effect of an incubation
period filled with low demand tasks on solving linguistic problems. A positive effect
of a filled incubation period on problem-solving compared with rest during the
incubation period undermines the conscious-work hypothesis that incubation effects
are due to the mental fatigue reduction (Posner, 1973). There remains a possibility, of
course, that a sufficiently light load might allow additional covert problem solving
compared with a heavier task load (Browne & Cruse, 1988; Posner, 1973), but this
does not explain why a light load should be more facilitative than rest alone.
65
The positive effect of a light cognitive load may indicate competition between
controlled and automatic processes in solving linguistic problems. It has been
suggested with remote associates task performance that only strong (and in this
context, incorrect) associates are accessed when individuals focus their attention on
seeking solutions, whereas remote associates are more likely to be accessed when an
individual’s cognitive resources are allocated in a diffuse manner (Ansburg & Hill,
2001; Finke, Ward, & Smith, 1992; Martindale, 1995). During an incubation period,
low-demand tasks may occupy part of the problem solvers’ attention, preventing the
focused concentration that yields strong associates. Resting during an incubation
period may allow individuals to continue consciously working on the problem, while
performing high-demand tasks may shift attention entirely to that interpolated task,
leading to a narrow rather than diffused attentional focus. The impact of performing
high-demand tasks is analogous to the ‘verbal overshadowing’ effect reported by
Schooler and colleagues (Schooler, Ohlsson, & Brookes, 1993) in which the act of
verbalizing can impair performance by focusing individuals’ attention inappropriately
on verbalisable components of a task. The suggested role of the light-load incubation
tasks receives indirect support from recent findings which show that that visual search
can be more efficient when performed concurrently with an unrelated task than when
performed alone (Smilek, Enns, Eastwood, & Merikle, 2006). They suggest that the
dual-task condition prevents a narrow attentional focus in searching stimuli.
With visual problems, the magnitude of the incubation effects was independent
on the setting of an incubation period (filled or unfilled). Differences between visual
and linguistic insight problems may arise through a greater reliance on strategic
search rather than knowledge activation in the former than the latter. MacGregor,
Ormerod, and Chronicle (2001) proposed that in solving the nine-dot problem,
66
individuals select and execute moves that maximally reduce the distance between
current and goal states, essentially drawing lines that connect as many dots as possible.
While there remain moves available that satisfy a criterion of satisfactory progress (in
this case, the ratio of dots cancelled to lines available), individuals will persevere with
an initial representation of the problem that, in the case of the nine-dot problem, does
not include consideration of space outside the dot array. According to MacGregor et
al.’s account, individuals must experience a failure to find moves that meet a criterion
of satisfactory progress before they change the initial representation of the problem,
thereby including space outside the dot array in their attempts. An incubation period
would be helpful only if the problem solvers became aware of the necessity of a
strategy shift, but according to MacGregor et al. they are unlikely to do so unless they
encounter criterion failure as a result of reaching impasse. Seifert, Mayer, Davidson,
Palatino, and Yaniv (1995) offer an alternative account that also points to the
criticality of experiencing failure and impasse for eventual success in insight
problem-solving.
If the hypothesis that visual problems require impasse for a strategy switch to
occur is correct, a long preparation period (i.e., pre-incubation problem-solving)
should be more likely to yield benefits from subsequent incubation with visual
problems because it allows individuals to reach impasse prior to incubation. The
results of the regression analysis and the follow-up bivariate correlations are
consistent with this prediction, showing a statistically significant positive correlation
between the incubation effect size and the length of preparation period with visual
problems.
A positive correlation between length of preparation period and incubation
effect size was also found with creative problems. A long preparation period may
67
allow individuals to exhaust search in one domain, making it more likely for them to
explore a new domain in the second phase of solving. A positive correlation was not
found when solving linguistic problems, though this may simply reflect the small
variability in length of preparation period among studies using linguistic problems
(the preparation period of 82% of these studies ranged from 0.5 to 1 minute).
The meta-analysis reveals that embedding misleading cues to the problems was
not a significant predictor overall. This result is in contrast with previous reports (e.g.,
Smith et al., 1989), which suggest that incubation effects arise through forgetting of
inappropriate information. The true effect of misleading cues is underestimated in our
regression analyses, as we only examined the overall effect of misleading cues on
problem solving in general. Twenty nine independent studies included in this
meta-analysis examined the impact of misleading cues; 25 of them were examining
the impact on linguistic problems, the rest on visual problems. The weighted mean of
the effect size estimates of these studies for each problem type by the presence of
misleading cues were: Linguistic with Misleading Cue: M = .36, SD = .09; Linguistic
without Misleading Cue: M = .17, SD = .06; Visual without Misleading Cue: M = .26,
SD = .08; Visual with Misleading Cue: M = .18, SD = .39. For studies employing
linguistic problems, the presence of misleading cues induced a larger incubation
effect, but an opposite pattern of results was found in studies employing visual
problems, suggesting that the impact of misleading cues may be modality-specific. In
order to test whether the presence of misleading cues affects incubation with
linguistic problems alone, a weighted one-way analysis of variance was run to
compare the incubation effect sizes of studies using linguistic problems that included
misleading cues (25 studies) against studies using linguistic problems that did not
include misleading cues (40 studies), F(1, 65) = 3.04, MSE = 3.00, p = .08. Thus,
68
where problem materials involve linguistic stimuli, getting rid of misleading concepts
may be the key to task solution, in contrast with visual problems in which the key
may be to restructure the knowledge that is currently active. The effect of misleading
cues also offers some support for the Selective-Forgetting hypothesis, but this effect
may be task-specific.
In contrast with previous reports (e.g., Dominowski, 1972; Dreitadt, 1969;
Mednick, Mednick, & Mednick, 1964), the moderator Presence of Cues was not
found to be a significant predictor of incubation effects. It has been hypothesized that,
during an incubation period, unconscious processes such as spreading activation
sensitize individuals to solution concepts and make them more likely to utilise
externally presented cues. To examine this hypothesis, researchers have presented the
answers of unsolved problems during an incubation period, and compared
post-incubation performance with participants not receiving any cues during the
incubation period (e.g., Dodds, Smith, & Ward, 2002; Dominowski & Jenrick, 1972).
Findings of these studies are equivocal, but it does appear that problem solvers do not
always make use of solution relevant cues, even when the cue includes the solution
itself. For example, Chronicle, Ormerod, and MacGregor (2001) found that presenting
the nine-dot problem with a shaded background in the shape of the solution did not
lead to significant levels of facilitation, even when the relevance of the shading was
drawn to participants’ attention. However, because of the small number of studies that
present solution relevant cues (three with linguistic, seven with visual problems, none
with creative problems) and the wide variation in experimental parameters among
these studies, it is impossible to carry out further statistical analysis.
Overall, the meta-analysis results support the existence of incubation effects,
though there appears to be a range of effects specific to particular tasks and
69
performance conditions. When attempting creative problems that require a wide
search of knowledge, individuals benefit from an incubation period. Problems that
involve reaching some kind of insight into a unique solution do not always benefit
from incubation under all conditions. In the case of linguistic problems, such as the
RAT, there is a modest incubation effect but only where the incubation period is filled
with a low cognitive demand task. One possible explanation is that performing low
cognitive load incubation tasks allows the occurrence of some unconscious problem
solving processes, such as spreading-activation and selective forgetting. In the case of
visual problems, incubation effects arise only where there has been a sufficiently long
preparation period prior to incubation for the problem-solver to have entered a state of
impasse. Only under these conditions can an incubation period contribute to the
strategic shift needed to restructure a problem representation. Thus, the theoretical
positions of spreading activation, selective forgetting and restructuring each receive
support. However, evidence for each appears to be specific to particular problem
types.
The finding of a positive impact of an incubation period on solving creative
thinking problems supports the contention that incubation periods help the elicitation
of new ideas. Incubation is a concept central to many methodologies for encouraging
creative decision-making, especially among management science and business
communities (e.g., Rickards, 1991), and this result may be taken as supporting the
inclusion of an incubation phase in such methodologies.
It should be noted that, despite efforts to include studies from sources other than
peer-reviewed journals, the meta-analysis may be influenced by a bias in favour of
reporting significant effects at the expense of null effects. Thus, incubation effects
may be to some extent overstated in this meta-analysis, a problem common to all
70
meta-analyses. Nonetheless, the reasonably large effect sizes found with creative
problems indicate that, with this class of problem at any rate, incubation is a
potentially valuable mechanism for fostering creative thought.
Clearly, the empirical data on incubation are not straightforward. As a
consequence, the traditional narrative review approach is not amenable to drawing
strong cross-study conclusions. It is perhaps disappointing that relatively few
published studies met the necessary criteria for inclusion in the meta-analysis, since a
failure either to measure post-incubation performance or to provide effect size
information limits the conclusions that can be drawn from them. Nonetheless,
sufficient studies remain for the meta-analysis to be undertaken and to reveal some
intriguing results.
One remaining problem is the relatively narrow range of problem types that have
been explored. For instance, the majority of studies that explore incubation effects
with linguistic problems, which is the majority of studies overall, use the RAT. It is
unclear that whether the RAT can be considered an insight problem or a linguistic
completion task, suggesting it may not be representative of all linguistic
problem-solving tasks. Bowden and Jung-Beeman (2003) have found that participants
sometimes claimed that they solved RATs with insight, while sometimes reported that
they solved them without insight. Further studies should aim to explore task-specific
experimental settings for maximizing the incubation effect with a wider range of
tasks.
On the whole, the results of this meta-analysis support the existence of multiple
types of problem-specific incubation effect, and suggest that the concept of incubation
can only be understood through a close examination of the problems to which it is
applied and the conditions under which it is elicited. Therefore, two sets of experiments
71
were carried out to examine the incubation effect on linguistic and visual insight
problems separately.
The following section presents the findings of the first set of experiments
(Experiments I, II, and III) investigating the incubation effect on linguistic insight
problem. In these three experiments, the impact was examined of the presence of
misleading cues and the loading of incubation tasks on the occurrence of the three
proposed mechanisms: spreading-activation, selective-forgetting, and
cue-assimilation. To have a more specific measure of the incubation effect, both
performance improvement and the change of sensitivity to relevant and irrelevant
memory items were assessed. As a response to the problem of the narrow range of
problem types in the past studies, two types of linguistic insight problem, RATs
(Experiment I & II) and Rebuses (Experiment III), were adopted.
72
CHAPTER IV: EXPERIMENTS I, II, and III - LINGUISTIC INSIGHT
PROBLEMS
Three experiments examining linguistic insight problems were carried out in an
attempt to discriminate between three hypotheses of unconscious incubation, and to
examine the impact of incubation tasks and the presence of misleading cues on
incubation effects in more detail.
The traditional paradigm in incubation research, which only measures the
incubation effect on solution performance, offers weak data to test between the three
unconscious hypotheses. Therefore, in these three experiments, besides measuring
performance change, lexical decision tasks (LDT) were also adopted to assess
participants’ sensitivity towards relevant items and misleading cues embedded in the
problems, before and after an incubation period. Decreased lexical decision latency to
relevant items would support the spreading-activation hypothesis, while an increase in
lexical decision latency to misleading cues would support the Selective-Forgetting
hypothesis. Also, if spreading activation and selective forgetting are the sources of
incubation effects, then enhanced performance on problems should be accompanied
by a change of sensitivity to relevant items and misleading cues.
The presentation of lexical decision tasks containing solution-relevant words
could also serve as a facilitatory cue to the problem. The Opportunistic-Assimilation
hypothesis suggests that incubation periods help problem restructuring and increase
the chance of assimilating external cues. If this is the case, then facilitation by the
presentation of cues should be greater in the incubation condition than in the
No-Incubation condition. To examine the Opportunistic-Assimilation hypothesis,
73
participants’ performance on RATs in the cue-plus-incubation, the incubation-alone,
the cue-alone conditions was compared.
EXPERIMENT I
Experiment I investigated whether an incubation period can enhance performance in
solving Remote Associates Tasks (RATs), and examined two proposed unconscious
hypotheses: Opportunistic Assimilation and Spreading Activation. To verify the
Spreading-Activation hypothesis, a Lexical Decision Task (LDT) was adapted as a
supplementary measure to assess if there was any change in participants’ sensitivity to
relevant memory concepts after an incubation period. The presentation of the LDTs
could also serve as an implicit cue to the problem, and comparing the post-incubation
performance improvement in the presence and absence of cues enabled an
examination of the Opportunistic-Assimilation hypothesis.
The co-examination of these two proposed mechanisms in one experiment also
allows us to investigate how incubation can enhance cue assimilation. The
Opportunistic-Assimilation hypothesis suggests that during the incubation period, the
problem representation will be re-structured into a more appropriate form, which will
increase the likelihood of assimilating the chance-encountered cue afterwards.
However, the hypothesis does not predict the nature of the restructuring. Yaniv and
Meyer (1987) have suggested that spreading activation occurs during the incubation
period to sensitize individuals to the relevant memory items, and this will
subsequently facilitate the cue-assimilation. Yet, no studies have previously been
conducted to examine this proposal. The data of Experiment I should determine if
there is any association between cue-assimilation and spreading-activation.
74
Method
Participants
Forty-eight (37 females, 11 males) first-year Psychology students from Lancaster
University participated to gain course credits. The mean age was 19.33 with a
standard deviation of 1.48. All were native English speakers.
Materials
Remote Associates Tasks (RATs; Mednick, 1962) were used as the
problem-solving task in this study. In each RAT, three apparently unrelated words are
presented, and participants have to think of a fourth word that forms an association
with each of the three words. For example, if the three stimulus words of a RAT are
“blue”, “cake” and “cottage”, the fourth word can be “cheese”. Sixteen RATs were
selected from a pool developed by Bowden and Jung-Beeman (2003). The RATs were
randomly divided into two sets, Set A and B. The reported solution rates ranged
between 10%-31%, with a mean of 17.75% (SD = 6.32) for Set A, and 22.63% (SD =
8.18) for Set B. The mean reported solution rate of these two sets of RATs was then
compared, and the difference was not significant, t(15) = 1.33, p = .20, Coden’s d
= .69. The purpose of preparing two different sets of RATs with a similar level of
difficulty is to check that measured effects are not item-specific (cf. the language as
fixed effects fallacy: Clark, 1973). Appendix F presents the RATs given in Set A and
B.
Lexical Decision Tasks (LDTs) were adapted to examine participants’ level of
sensitivity to relevant solution concepts. A set of LDTs consisted of five items
including neutral words, pseudo-words, and the answer of the previously-presented
75
RAT (target words). All presented words were high frequency words (an occurrence
of 90 or more in a million).
Mental Rotation Tasks (MRTs) and Arithmetic Tasks (ATs) were used as
incubation tasks. Each MRT consisted of a pair of objects. Participants had to judge if
the objects in each pair were identical. Each of the MRTs created for the purpose of
this study consisted of a pair of three-dimensional objects. Each object consisted of 11
solid cubes attached face-to-face to form a rigid arm-like structure with exactly three
right-angled “elbows”. The object placed on the left hand side was the standard
stimulus and the object on the right hand side was the comparison stimulus. In half
the MRTs, the comparison stimulus was identical to the standard stimulus, whereas in
the other half the comparison stimulus was the mirror image of the standard stimulus.
The comparison stimulus was either rotated clockwise or anti-clockwise around the
x-axis, y-axis, or z-axis with an angle. The rotation angle was 60, 80, 100, 120, 140,
160, or 180 degrees relative to the orientation of the standard stimulus. The ATs
comprised three-digit additions and subtractions without regrouping.
Design
This experiment employed a 2 x 3 mixed factorial design. The within-subjects
factor was whether LDTs were presented before or after the second attempt to the
RAT. The between-subjects factor was the incubation condition with three levels: No
Incubation (NI), Incubation filled with MRTs and ATs (IN),and Rest/Unfilled
Incubation (RE) conditions. In the NI condition, after the completion of the first
attempt at a RAT, participants were immediately prompted to do an LDT set followed
by a second attempt to the same RAT(or vice versa), depending on whether it was an
LDT-before (1st, 2nd, 5th, and 8th) or LDT-after (3rd, 4th,6th, and 7th) trial. In the IN
76
condition, participants were prompted to perform MRTs and ATs for two minutes
before proceeding to complete the LDT set or the second attempt to the same RAT. In
the RE condition, participants were prompted to sit quietly and listen to soft music for
two minutes before proceeding to complete the LDT set or re-approach the same RAT.
The inclusion of the RE condition served the purpose of assessing if any measured
changes in RAT performance and lexical latency were due to performing incubation
tasks or simply due to having a two-minute rest between the two attempts.
Procedure
All the task instructions and stimuli were presented on a computer using bespoke
software. Before the start of the experiment, participants in the NI and RE conditions
were informed that the main experiment involved solving RATs and LDTs.
Participants were then given instructions on how to complete both tasks and were
reminded to finish the tasks as quickly and accurately as possible. After this, they
were prompted to solve 1 RAT and 2 LDTs for practice. Participants in the IN
condition were informed that the main experiment involved solving RATs, LDTs,
MRTs, and ATs. They were then given the task instructions and were reminded to
finish all the tasks as quickly and accurately as possible. They were then prompted to
solve 1 RAT, 1 MRT, 1 AT, and 2 LDTs for practice. After the practice phase, the main
experiment started.
During the main experiment, eight trials were presented in the same sequence for
each participant. In each trial, the three stimulus words of a RAT were presented
horizontally across the centre of the screen. Participants could respond at any time
within the 30-second presentation time by first pressing the c key, and then speaking
out their answer to a microphone connected to a tape-recorder. All the responses were
77
recorded for scoring of correct responses afterwards. Response latency was also
measured as the duration from the onset of problem presented until the participant
pressed c or until the presentation period ended. Once the participants spoke out the
answer or the 30-second period had passed, participants had to press the c key again
to proceed to the next task.
Participants in the NI condition were then prompted to do an LDT set followed
by a second attempt to the same RAT(or vice versa depending on whether it was a
LDT-before or LDT-after trial). The five items in each LDT set were presented at the
centre of the screen, one at a time. Before presenting each stimulus, a fixation cross
was presented at the centre of the screen for 0.5 second to draw participants’ attention.
Participants were prompted to press the z key if the item presented was a word or the
m key if it was a pseudo-word. The first presented item was always a neutral word or
pseudo-word, which served as a warm-up stimulus. The target word, along with
pseudo-words and neutral words, was counter-balanced across the remaining four
positions.
In the RE Condition, participants were prompted to sit quietly and listen to soft
music for two minutes before proceeding to complete the LDT set or the same RAT.
In the Incubation condition, participants were prompted to complete MRTs for one
minute and ATs for another minute (incubation period) before proceeding to complete
the LDT set or the same RAT.
In each MRT, two images were presented, one at the left and one at the
right-hand side of the screen. Participants had to press the z key if they think the
stimuli were identical, or press the m key if not. After participants responded, another
MRT would be presented. The responses and the response latencies by participants
were recorded by the computer. In each AT, participants had to write down their
78
answers on a sheet of paper, and then press the c key to proceed to next AT. In both
tasks, the response time was measured as the duration from the onset of stimulus to
the point that the participants press the c key. The number of MRTs and ATs presented
during the incubation period varied among participants, depending on how quick the
participant responded. Figure 4.1.1 presents the task presentation sequence in each
trial in the three conditions.
LDT
No Incubation
RAT
RAT
RAT
Rest
Condition:
LDT
LDT
RAT
RAT
LDT
(1st, 2nd, 5th, & 8th trials)
MUSIC
Condition:
RAT
(1st, 2nd, 5th, & 8th trials)
(3rd, 4th, 6th, & 7th trials)
Condition:
Incubation
RAT
LDT
RAT
RAT
LDT
(3rd, 4th, 6th, &7th trials)
(1st, 2nd, 5th, & 8th trials)
MRT +AT
(3rd, 4th, 6th, & 7th trials)
Figure 4.1.12 Task Presentation Sequence in Each Trial in Each Condition in
Experiment I.
79
Results
Analyses of participants’ RAT performance and lexical decision latencies were carried
out to examine the three predictions. The first prediction concerns the positive role of
incubation effect on performance level. It was predicted that participants in the IN
condition should demonstrate a larger performance improvement from the first
attempt at the RAT problem to the second attempt, compared with participants in the
other two conditions. The second prediction is based on the
Opportunistic-Assimilation hypothesis, that an incubation period can facilitate
problem restructuring and in turn increase the likelihood of cue assimilation. It was
predicted that participants in the IN condition should show a larger performance
improvement in the Cue Trials compared to the No-Cue trials, and this cueing effect
should be less noticeable in the NI and RE conditions. Shifting from the incubation
effect on performance to a qualitative change in cognitive state, the third prediction
pertains to the occurrence of spreading activation during the incubation period.
According to the Spreading-Activation hypothesis, participants in the IN condition
should have significantly shorter lexical decision latencies to target words relative to
the neutral words, and again, such differences should be less noticeable in the other
two conditions.
Background Statistics
Before running the main analyses, participants’ performance in solving the RATs in
Set A and B at the first attempt was compared to check the assumption that these two
question sets were at the same difficulty level. Table 4.1.1 presents the means and the
standard deviations of the number of correct responses and the response latency to the
RATs at the first attempt by Question Set and Condition.
80
Table 4.1.113
Means and Standard Deviations of the Number of Correct Responses and Response
Time at the First Attempt by Question Set and Condition
Number of Correct Responses
Response Time
Set A
Set B
Set A
Set B
M (SD)
M (SD)
M (SD)
M (SD)
NI
0.89 (1.26)
1.44 (1.51)
27.87 (2.64)
27.15 (1.53)
RE
0.57 (0.79)
1.25 (1.83)
28.99 (1.48)
27.42 (3.46)
IN
1.00 (1.15)
0.50 (0.53)
28.05 (2.42)
29.04 (1.25)
Condition
Note. Incorrect responses or unanswered RATs were scored as taking the maximum
permissible time (30 s).
Two 2 (Question Set) x 3 (Condition) ANOVAs, using number of correct
responses and response time as the dependent variables respectively, were carried out
to compare participants’ performance in solving RATs in set A and B. Correct
Responses: None of the effects was significant: Condition: F(2, 42) = .44, p = .51, p2
= .01; Question Set: F(1, 42) = .45, p = .64, p2 = .02; Condition x Question Set: F(2,
42) = .99, p = .38, p2 = .05. Response Time: Again, none of the effects was
significant: Condition: F(2, 42) = .38, p = .54, p2 = .01; Question Set: F(1, 42) = .77,
p = .47, p2 = .04; Condition x Question Set: F(2, 42) = 1.07, p = .35, p2 = .05. The
non-significant ANOVA results confirm the presumption that the RATs in Set A and
Set B were in similar level of difficulty. Hence, the data from both question sets were
merged in all the subsequent analyses. Also, the non-significant Condition effect
implies that there was no general problem-solving performance difference among
participants in these three conditions. This ensures that any performance difference
81
found in the subsequent analysis could not be simply attributed to differences in the
general problem solving ability.
Incubation Effects on Problem Solving Performance
Performance Improvement: If an incubation period can enhance
problem-solving performance, then participants in the IN condition should show a
larger performance improvement over the first attempt at the RATs, compared with
the other two conditions. To assess performance improvement, an improvement score
was computed for each participant, which was the proportion of RATs not solved at
the first attempt that were solved at the second attempt. In order to reveal the net
effect of an incubation period on performance improvement in the absence of any
external cue, only data in the LDT-after (3rd, 4th, 6th, & 7th) trials were included in the
analysis. One participant was excluded because of solving all the RATs presented in
the LDT-after trials at the first attempt. Table 4.1.2 presents the means and the
standard deviations of the number of correct responses in the first and the second
attempts, and the improvement score, in each condition.
82
Table 4.1.214
Mean and Standard Deviations of the Number of Correct Responses in the Two
Attempts and Improvement Score by Condition
First Attempt
Second Attempt
Improvement Score
M (SD)
M (SD)
M (SD)
0.63 (0.91)
0.92 (1.03)
.11 (.23)
NI
0.61 (0.92)
1.00 (1.32)
.14 (.26)
RE
0.73 (1.22)
0.93 (1.33)
.11 (.27)
IN
0.53 (0.52)
0.80 (0.56)
.07 (.13)
Condition
Overall
Note. The number of correct responses in the second attempt is the sum of the number
of RATs solved in the first attempt and the number of newly solved RATs in the
second attempt.
The overall improvement score, M = .11, SD = .23, was significantly larger than
zero, t(47) = 3.20, p = .002, Cohen’s d = 0.93. This is expected because extra problem
solving-time should enhance performance. The core question is whether there was
any significant difference in the improvement scores among the three conditions. A
one-way ANOVA, using Condition as the independent variable and improvement
score as the dependent variable, was carried out. The Condition effect was not
significant, F(2, 44) = .39, p = .68, ηp2 = .017, implying that the degree of
improvement was comparable among the three conditions. This does not support the
prediction that an incubation period can enhance problem-solving performance. The
non-significant incubation effect on performance may be due to the fact that the
post-incubation performance improvement, not only relies on the occurrence of
relevant internal cognitive processes during an incubation period, but also on the
external environment such as the presence of relevant external cues.
83
Opportunistic Assimilation: The Opportunistic-Assimilation account suggests
that incubation periods provide time for the problem solver to restructure the mental
representation of the unsolved problem into a better form that could increase the
chance of assimilating newly encounter-external cues into the problem. In other
words, the magnitude of the incubation effect depends on whether problem solvers
encounter relevant external cues during the incubation period. To verify this
hypothesis, the degree of RAT performance improvement in the presence and absence
of external cues (LDTs) in the three conditions was examined.
The Opportunistic-Assimilation hypothesis predicts that participants in the IN
condition should show a significant performance improvement in the Cue Trials
compared to the No-Cue trials, and the facilitation of the presentation of the cue
should be less noticeable in the NI and RE conditions. The mean number of RATs
solved at the first and second attempts in the Cue (LDT-before:1st, 2nd, 5th , & 8th) and
No-Cue (LDT-after:3rd, 4th, 6th, & 7th) trials is presented in Table 4.1.3.
Table 4.1.315
Mean and Standard Deviations of the Number of Correct Responses in the Cue and
No-Cue trials in Each Condition.
First Attempt
Second Attempt
Cue
No Cue
Cue
No Cue
M (SD)
M (SD)
M (SD)
M (SD)
NI
.61 (0.92)
.56 (.70)
1.00 (1.08)
1.00 (1.08)
RE
.73 (1.22)
.20 (.41)
0.93 (1.33)
1.00 (1.13)
IN
.53 (0.52)
.20 (.53)
0.80 (0.56)
0.46 (0.74)
Condition
84
For each participant, the improvement score in Cue and No-Cue trials, and the
differences between these two scores (Cue – No-Cue), were calculated. One
participant was excluded because he/she solved all the RATs presented in the
LDT-after trials at the first attempt. Table 4.1.4 presents the means and the standard
deviations of the improvement scores in the presence and the absence of the cues, and
the differences between them, in each condition.
Table 4.1.416
Means and Standard Deviations of Improvement Score on the Cue and No-Cue trials,
and the Score Differences by Condition
Cue
No Cue
Cueing Effect
(Improvement Score on Cue –
No Cue)
M (SD)
M (SD)
M (SD)
NI
0.15 (0.27)
0.14 (0.26)
< .01 (.34)
RE
0.16 (0.21)
0.11 (0.27)
.05 (.30)
IN
0.07 (0.15)
0.07 0(.13)
-.01 (.12)
Condition
It was predicted that the cueing effect, indicated by the improvement score
difference between the Cue and the No-Cue trials, should be most noticeable in the IN
condition. This prediction was tested by carrying out a 2 (Cue) x 3 (Condition)
ANOVA with repeated-measures on Cue (No-Cue vs. Cue) and using improvement
score as DV. Table 4.1.5 presents the summary of the ANOVA results.
85
Table 4.1.517
ANOVA Summary Table for the Effect of Condition and Cue on Improvement Scores.
Source
df
MS
F
p
ηp2
.84
.44
.04
Between-Subjects
Condition
2
.05
Error ( between)
44
.06
Within-Subjects
Cue
1
.007
.19
.67
< .01
Cue x Condition
2
.007
.19
.83
.01
Error (within)
44
.04
Neither the main effects nor the interaction effect was significant, Cue: F(1, 44)
= .19, p = .67, p2 = .04; Condition: F(1,44) = .84, p = .44, p2 = .04; Cue x Condition:
F(2, 44) = .19, p = .82, p2 = .09. This result indicates that, in all conditions, the mean
proportion of improvement in the Cue trials and No-Cue trials did not differ
significantly. Participants in the IN condition were not more likely to assimilate the
cues to solve the problems. The Opportunistic-Assimilation hypothesis is therefore
not supported here.
The analyses of performance improvement failed to confirm that an incubation
period can enhance problem solving performance, either directly or indirectly via cue
assimilation. However, the null effect on performance change in this experiment does
not necessarily imply the absence of any unconscious problem-solving processes
during the incubation period. It may be that spreading-activation occurred during the
incubation period in this experiment to increase the activation of the relevant memory,
but the effect was too small to have any impact on the performance level. To address
86
this possibility of the occurrence of spreading-activation during the incubation period,
participants’ sensitivity to relevant items among the three conditions was compared.
Incubation Effects on Knowledge Activation
Spreading Activation: If spreading activation occurred during the incubation
period, then participants in the IN condition should be more sensitive to target words
relative to the neutral words, reflected in a quicker lexical decision times; and this
difference should be smaller in the other conditions. To test this prediction,
participants’ lexical decision times for target and neutral words in the LDT-before (1st,
2nd, 5th, and 8th) trials in the three conditions were examined. Accuracy on the LDT
was 93% or higher in all conditions. Incorrect lexical decisions and correct decisions
with extreme lexical decision times (< 50 ms or > 2500ms) were discarded (4.5% of
the data). After that, the mean and standard deviation of the lexical latencies were
calculated for each participant. Within each participant, lexical decision times longer
than 2SDs from the mean were re-coded to the value found at 2SD. The re-coded
lexical latencies for all conditions approximated a normal distribution (no skew
values exceeded +/- 1.5), therefore, natural time data were used in the analysis. Table
4.1.6 presents the means and standard deviations of the lexical decision times for
target words (answers of the unsolved RATs) and the neutral words in the LDT-before
trials by Condition.
87
Table 4.1.618
Means and Standard Deviations of the Lexical Decision Times for Target Words and
Neutral Words
Lexical Decision Time
Target Word
Neutral Word
M (SD)
M (SD)
NI
772.25 (261.06)
751.68 (200.18)
RE
688.28 (160.13)
762.23 (218.56)
IN
754.66 (208.13)
772.47 (188.50)
Condition
A2 (Word Type) x 2 (Question Set) x 3 (Condition) ANOVA with
repeated-measures on Word Type (Neutral vs. Target), using lexical decision time as
the dependent variable, was carried out to compare participants’ lexical decision
latency to target words and neutral words among the three conditions. The summary
of the results is presented in Table 4.1.7.
88
Table 4.1.719
ANOVA Summary Table for the Effect of Condition and Word Type on Lexical
Decision Time
Source
df
MS
F
p
ηp2
.19
.82
.01
Between-Subjects
Condition
2
14476.29
Error ( between)
45
74600.27
Within-Subjects
Word Type
1
13411.67
.98
.33
.02
Word Type x Condition
2
18326.10
1.34
.27
.06
Error (within)
45
23570.06
Neither the main effects nor the interaction effects was significant, Word Type:
F(1,45) = .98, p = .33, p2 = .02; Condition: F(2, 45) = .19, p = .82,p2 = .01; Word
Type x Condition: F(2, 45) = 1.34, p = .27,p2 = .06. This indicates that, in all the
conditions, participants did not respond significantly faster to the target words than
they did to the neutral words. Participants in the IN condition were not more
sensitized to the target words, and this is not in line with the Spreading-Activation
hypothesis.
Discussion
Experiment I failed to demonstrate any incubation effects, and cannot offer any
supporting evidence to the Opportunistic-Assimilation and the Spreading-Activation
hypotheses. However, the absence of an incubation effect at both behavioral
(performance) and cognitive (knowledge activation) levels may be related to the
89
difficulty of the RATs used in this experiment. The RATs used in this experiment may
have been too difficult for the participants to benefit from an incubation period,
reflected in the low number of correct responses (see Table 4.1.1). An incubation
period may not be very helpful if the problems are so difficult that the individuals do
not have the required knowledge to solve them. Also, consistent failures in solving
RATs at the first attempt may lower participants’ motivation, and in turn, affect
performance in the second attempt. Participants may even consider some of the RATs
as unsolvable after repeated failures in the first attempts, and this may suppress any
further problem-solving processes.
Therefore, to have a more appropriate evaluation of the role of an incubation
period, Experiment II was conducted to replicate Experiment I but using easier RATs.
In addition to this, Experiment II also further examined the meta-analysis findings
concerning the moderating effect of incubation-task loading on the incubation effect.
90
EXPERIMENT II
The paradigm of Experiment II was similar to that of Experiment I, except for five
modifications. First, there was an adjustment to the difficulty level of the RATs. In
Experiment II, RATs were again selected from a pool created by Bowden and
Jung-Beeman (2003), but easier RATs were chosen this time. Two sets of RATs were
prepared, one consisting of intermediate RATs and the other set consisting of easy
RATs. Second, there was no Rest condition, but instead three Incubation conditions
with incubation periods filled by high, intermediate or low cognitive load tasks. The
No-Incubation condition remained and served as a control condition. The rationale for
including these three incubation conditions is to examine the impact of the loading of
incubation tasks on incubation effects. The third modification was to the lexical
decision tasks. Although word length and frequency were controlled in Experiment I,
it is still possible that there were some uncontrolled differences between the target
words and the neutral words. To equate this ancillary lexical factor, the same items
served as both target words and neutral words across participants in this experiment.
The target words (answers for the easy RATs) presented to participants solving easy
RATs were used as neutral words for those solving intermediate RATs, and vice versa.
The fourth modification was the counterbalancing of the presentation order of the
LDT-before and LDT-after trials with each participant. In Experiment II, participants
were randomly assigned to solve the LDT before the second attempt to the RAT in the
1st, 2nd, 5th, and 8th trials or in the 3rd, 4th, 6th, and 7th trials. This counterbalancing
ensured that any measured effect was neither order-specific nor item-specific. The last
modification was the control of the number of incubation tasks presented during the
incubation period. In Experiment I, the number of MRTs and ATs presented during the
91
incubation period varied among participants, and this may affect the incubation effect
size on each participant. To control for this potential confounding factor, in
Experiment II, it was fixed that 5 MRTs (8 second each) and 8 ATs (10 second each)
were presented during the incubation period.
Method
Participants
173 (Female: 124, Male: 49) students from Lancaster University participated in
this experiment for course credits or payment, and the mean age was 20.33 (SD =
2.91). All were native English speakers.
Materials
Two sets of RATs (Easy and Intermediate) were prepared. As in Experiment I,
the RATs were selected from a pool created by Bowden and Jung-Beeman (2003).
The mean reported solution rate of the RATs in the Intermediate set was 36.16% (SD
= 8.57) with a range of 25%-45%. The mean solution rate of the RATs in the easy set
was 73.33% (SD = 5.65) with a range of 60%-80%. Appendix G presents the RATs in
the Easy and the Intermediate sets.
Three sets of MRTs and ATs (Low-, Intermediate-, and High-load) were
prepared and used as the incubation tasks. The MRTs used in Experiment I were
divided into two sets based on participants’ response time and accuracy. MRTs with
below-average accuracy rate and shorter-than-average response time were included in
the intermediate-load set. MRTs with above-average accuracy rate and
longer-than-average response time were included in the low-load sets. The mean
92
response time and accuracy rate of the selected MRTs was 5.70 (SD = .90), and
52.87% (SD = 9.59) in the intermediate-load set, and 3.17 (SD = .55) and 77.52 %
(SD = 10.33) in low-load set. The differences in response time and accuracy rate
between these two sets were significant, response time: F(1, 53)= 158.50, p < .001;
accuracy rate: F(1, 53) = 83.94, p < .001.
The MRTs included in the high-load set were created by adding one right-angled
4-cube arm to each MRT stimulus in the intermediate set. This means that the MRT
stimuli in the high-load set were 3-D objects made up of 15 solid cubes with four
right-angled arms, while the MRT stimuli in the intermediate and low set were objects
made up 11 cubes and with three right-angled arms. The increase in the figural
complexity should raise the difficulty level of the MRTs.
The ATs in the High-, Intermediate-, and Low-Load sets, were 3-digit
multiplications and divisions, 3-digit summation and subtractions with regrouping,
and 3-digit summation and subtraction without regrouping, respectively.
Lexical Decision Tasks (LDTs) were again used to examine participants’ level
of sensitivity to relevant solution concepts. Each set of LDTs consisted of five items
including neutral words, pseudo-words, and the answer to the previously-presented
RAT (target words). Neutral words included words unrelated to the stimuli and the
answers of the presented RATs in another set.
Design
This experiment employed a 2 x 2 x 4 mixed factorial design, with the
presentation of LDTs either before or after the second attempt to the RATs as the
within-subjects factor. The between-subjects factors were the RAT difficulty (easy vs.
intermediate) and the cognitive loading of the incubation tasks (no incubation, low-,
93
intermediate-, and high-load).
Procedure
Participants were randomly assigned to solve either the easy or intermediate
RATs in one of four incubation conditions. The procedure and the presentation of the
tasks in Experiment II were identical to those used in Experiment I, except that when
solving MRTs and ATs, participants could response only after the presentation time
was up ( MRT: 8 seconds, AT: 10 seconds).
Results
Similar to Experiment I, the analyses consisted of three main parts. The first part
examined the post-incubation performance improvement, the second part investigated
the Opportunistic-Assimilation hypothesis, and the last part focused on the
Spreading-Activation hypothesis. Before conducting the main analysis, participants’
RAT performance was examined to check that, first, the RATs in Experiment II were
indeed easier than those in Experiment I, and second, the easy RATs were easier than
the intermediate RATs. Participants’ RAT performance in the first attempt, in terms of
accuracy and response latency, in Experiment I and II was compared. Table 4.2.1
presents the means and the standard deviations of the number of correct responses and
response time at the first attempt in Experiment I and II.
94
Table 4.2.120
Means and Standard Deviations of the Number of Correct Responses and Response
Time at the First Attempt in Experiment I and II
RAT Performance
Correct Response
Response Time
M (SD)
M (SD)
Experiment I
0.96 (1.25)
28.04 (2.42)
Experiment II
3.56 (1.99)
22.19 (7.80)
The mean number of correct responses in Experiment II was significantly
higher than the mean in Experiment I, t(187) = 8.52, p < .001, Cohen’s d = 1.25.
Participant in Experiment II also took shorter time to solve the RATs, t(187) = 7.19, p
< .001, Cohen’s d = 1.05. These results converge to support the view that the RATs in
Experiment II were easier than those in Experiment I. Table 4.2.2 presents the means
and the standard deviations of the response time and accuracy, for easy and
intermediate RATs, in each of the incubation conditions in Experiment II.
95
Table 4.2.221
Means and Standard Deviations of the Number of Correct Responses and Response Time at the First Attempt by RAT Difficulty
and Condition
Number of Correct Responses
Response Time
Easy RAT
Intermediate RAT
Easy RAT
Intermediate RAT
M (SD)
M (SD)
M (SD)
M (SD)
No Incubation
4.41 (1.97)
2.78 (1.31)
18.59 (6.53)
23.19 (3.61)
Low Load
4.56 (1.75)
1.73 (1.80)
17.99 (4.58)
27.29 (2.50)
Intermediate Load
5.11 (1.57)
2.11 (1.24)
16.53 (5.16)
26.20 (3.08)
High Load
5.00 (1.56)
2.61 (1.69)
19.19 (5.51)
25.15 (4.14)
Condition
96
To determine if the easy RATs were easier than the intermediate RATs, two 2
(RAT Difficulty) x 4 (Condition) ANOVAs were carried out using number of correct
responses and response time as the dependent variables respectively. Correct
Response: The easy RATs were solved more often than the intermediate RATs,
indicated by the significant RAT Difficulty effect, F(1, 133) = 87.25, p < .001, p2
= .40. Non-significant effects of Condition, F(3, 134) = 1.05, p = .37, p2 = .02, and
Condition x RAT Difficulty, F(3, 133) = 1.34, p = .26, p2 = .03, suggest that in both
type of RATs, the RAT performance was comparable among the four experimental
conditions. Response Time: Consistent with the ANOVA results for correct responses,
the main effect of RAT Difficulty was significant, F(1, 133) = 90.39, p < .001, p2
= .41, while the effect of Condition and the interaction between Condition and RAT
Difficulty were not significant, Condition: F(3, 133) = .98, p = .40, p2 = .02;
Condition x RAT Difficulty: F(3,133) = 2.58, p = .06, p2 = .06. These ANOVA
results support the assumption that the intermediate RATs were more difficult than the
easy RATs.
The following sections present the results of the main analyses conducted to
investigate incubation effects on problem solving performance and sensitivity to
relevant concepts, as well as the moderating effect of the cognitive load of incubation
tasks.
Incubation Effects on Problem-Solving Performance
Performance Improvement: To determine whether an incubation period can
enhance problem solving performance, the improvement scores in the LDT-after trials
among the four conditions were compared. Twenty participants were excluded in the
97
analysis because they solved all the RATs presented in the LDT-after trials in the first
attempt. Similar to Experiment I, the improvement score was first computed for each
participant. Table 4.2.3 presents the means and the standard deviations of the number
of correct responses in the first and the second attempt, and the improvement scores
by Condition and RAT Difficulty.
Table 4.2.322
Means and Standard Deviations of the Number of Correct Responses in the Two
Attempts by Condition and RAT Difficulty
First Attempt
Second Attempt
Improvement
Score
M (SD)
M (SD)
Condition
Overall
M (SD)
Easy RAT
2.22 (0.81)
2.58 (0.80)
0.19 (0.32)
No Incubation
2.06 (1.06)
2.37 (1.02)
0.13 (0.28)
Low Load
2.21 (0.92)
2.74 (0.87)
0.29 (0.24)
Intermediate Load
2.44 (0.62)
2.78 (0.73)
0.24 (0.39)
High Load
2.17 (0.65)
2.43 (0.59)
0.12 (0.20)
Intermediate RAT
Overall
1.18 (0.93)
1.34 (1.05)
0.07 (0.21)
No Incubation
1.38 (0.86)
1.53 (0.93)
0.08 (0.23)
Low Load
0.65 (0.61)
0.82 (0.81)
0.05 (0.17)
Intermediate Load
1.05 (0.83)
1.20 (1.01)
0.07 (0.17)
High Load
1.58 (1.12)
1.74 (1.24)
1.00 (0.26)
98
The overall improvement scores on solving easy and intermediate RATs were
both significantly larger than zero, Easy RAT: M = .19, SD = .31, t(75) = 5.25, p =
< .001, Cohen’s d = 1.21; Intermediate RAT: M = .07, SD = .21, t(76) = 3.13, p = .002,
Cohen’s d = .72. A 2 (RAT Difficulty) x 4 (Condition) ANOVA, using improvement
score as the dependent variable, was carried out to check if the degree of
improvement was a function of incubation condition. The summary of the ANOVA
results is presented in Table 4.2.4.
Table 4.2.423
The ANOVA Summary Table for the Effect of RAT Difficulty and Condition on
Improvement Scores
Source
df
MS
F
p
ηp2
Between-Subjects
RAT Difficulty
1
.54
7.62
.01
.05
Condition
3
.042
0.59
.62
.01
RAT x Condition
3
.094
1.32
.27
.027
Error ( between)
145
.071
The Condition effect was not significant, F(3, 133) = .59, p = .62, p2 = .01. This
indicates that the degree of improvement was comparable among the four conditions.
This non-significant difference was found on both easy and intermediate RATs, as the
RAT x Condition effect was also not statistically significant, F(3, 133) = 1.32, p= .27,
p2 = .027. This pattern of the results does not support the positive effect of
incubation period on problem solving performance. The RAT Difficulty effect was
significant, F(1, 133)= 7.62, p = .01, p2 = .05, revealing the improvement scores of
99
participants solving easy RATs were significantly higher than those solving
intermediate RATs. This may be due to the fact that easy RATs were easier to solve,
and therefore, participants solving easy RATs were more likely to have improvement
during the second attempt.
Opportunistic Assimilation: To examine if an incubation period can facilitate
cue-assimilation and in turn enhance the subsequent problem solving process,
participants’ RAT improvement scores in the presence (LDT-before trials) and
absence of the cue (LDT-after trials) were examined. Thirty-one participants were
dropped because they solved all the RATs presented in the first attempt. Table 4.2.5
presents the means and the standard deviations of the improvement scores in the Cue
and No-Cue trials, and the differences between them, in by Condition and RAT
Difficulty.
100
Table 4.2.524
Means and Standard Deviations of Improvement Scores in the Presence and Absence
of Cue, and the Score Difference by Condition and RAT Difficulty.
Cue
No-Cue
Cueing effect
(Improvement Score
Cue – No-Cue)
M (SD)
M (SD)
Condition
M (SD)
Easy RAT
No Incubation
.44 (.40)
.14 (.28)
.31 (.42)*
Low Load
.19 (.39)
.24 (.34)
-.05 (.58)
Intermediate Load
.13 (.27)
.25 (.40)
-.12 (.49)
High Load
.38 (.41)
.14 (.22)
.23 (.40)*
Intermediate RAT
No Incubation
.30 (.34)
.08 (.23)
.23 (.29)*
Low Load
.09 (.17)
.05 (.17)
.03 (.26)
Intermediate Load
.30 (.28)
.20 (.17)
.23 (.36)*
High Load
.12 (.22)
.10 (.26)
.02 (.24)
*Significantly difference from zero, p < .05
In the No-Incubation condition, a positive cueing effect was found when solving
easy RATs and intermediate RATs (Easy RAT: .31, Intermediate RAT: .23). However,
in the Low-Load condition, the presence of cue did not facilitate the problem solving
performance (Easy RAT: -.05, Intermediate RAT: .03). In the Intermediate-Load
condition, only participants solving Intermediate RATs benefited from the presence of
101
cue (Easy RAT: -.12, Intermediate RAT: .23). Interestingly, a reverse pattern was
observed in the High-Load condition that the presence of cue could enhance the
performance only when solving easy RATs, but not intermediate RATs (Easy
RAT: .23, Intermediate RAT: .02).
Apparently, the presentation of cues could have a negative, null, or positive
effect on RAT problem-solving processes, depending on the cognitive loading of the
incubation tasks and the difficulty level of the RATs. To examine how these two
procedural factors moderate the magnitude of the cueing effect, a 2 (RAT Difficulty)
x 2 (Cue) x 4 (Condition) ANOVA with repeated-measures on Cue (Cue vs. No-Cue),
using improvement score as the dependent variable, was carried out. The ANOVA
results are presented in Table 4.2.6.
102
Table 4.2.625
ANOVA Summary Table for the Effect of RAT Difficulty, Condition, and Cue on
Number of Correct Responses
Source
df
MS
F
p
ηp2
Between-Subjects
RAT Difficulty (R)
1
.766
8.17*
.01
.057
Condition (C)
1
.105
1.12
.34
.025
RxC
3
.076
0.82
.49
.018
134
.088
Error (between)
Within-Subjects
Cue (Cu)
1
.860
11.16*
.001
.077
Cu x R
1
.023
0.30
.58
.002
Cu x C
3
.242
3.14*
.03
.066
Cu x R x C
3
.260
3.37*
.02
.070
134
.077
Error (within)
*p < .05.
There was a significant Cue effect, F(1, 134) = 11.16, p = .001, ηp2 = .077,
indicating that performing LDTs containing solution-relevant words improved
performance on the RATs more than a second attempt at the same RATs. There was
also a significant interaction between Cue and Condition, F(3, 134) = 3.14, p = .03,
ηp2 = .066. Subsequent pairwise comparisons within each condition revealed that, in
the No-Incubation condition, the mean improvement score in the Cue trials was
significantly larger than the mean score in the No-Cue trials, t(35) = 4.47, p < .001,
Cohen’s d = 6.78. However, a Cue effect was absent in the other three conditions;
103
Low-Load: t(34) = -.13, p = .90, Cohen’s d = .01; Intermediate-Load: t(36) = .96, p
= .34, Cohen’s d = .31, High-Load: t(33) = 2.02, p = .052, Cohen’s d = 1.41. The null
findings may be due to the fact that the presence of a cue had an opposite impact in
solving easy and intermediate RATs in the three incubation conditions, and they may
cancel out each other when looking at the overall cueing effect in solving RATs. This
explanation is supported by the significant Cue x RAT x Condition effect, F(3, 111) =
3.12, p = .03, ηp2 = .08.
To investigate this interaction effect further, data were split into four groups, in
terms of the experimental conditions. In each group, a 2 (RAT Difficulty) x 2 (Cue)
ANOVA with repeated-measures on Cue, using improvement score as the dependent
variable, was carried out. In the No-Incubation condition, there was a significant
effect of Cue, F(1, 34) = 20.13, p < .001,ηp2 = .372, but the Cue x RAT Difficulty
effect was not significant, F(1, 34) = .03, p = .51,ηp2 = .013. This implies that
participants solving the easy and the intermediate RATs benefited from the cue to the
same degree.
In the Low-Load Condition, neither the main effect of Cue nor the interaction
effect between Cue and RAT Difficulty was significant, Cue: F(1, 33) = .01, p = .92,
ηp2 < .001; Cue x RAT Difficulty: F(1, 33) = .31, p = .58, ,ηp2 < .001, suggesting that
the exposure to relevant cue had no impact on the subsequent problem-solving
processes.
In the Intermediate-Load Condition, the Cue effect was not significant, F(1, 33)
= .68, p = .42, ηp2 < .001. A significant Cue x RAT Difficulty effect was found in the
Intermediate-Load Condition, F(1, 33) = 6.22, p = .02, ηp2 < .151, implying that
participants solving intermediate and easy RATs benefit from the presence of cue in
different degrees. The cueing effect in intermediate RATs was significantly larger than
104
zero, M = .23, SD = .36, t(19) = 2.87, p = .01, Cohen’s d = 1.32, however, the cueing
effect in solving easy RAT was not significantly different from zero, M = -.12, SD
= .49, t(16) = -.99, p = .34, Cohen’s d = .50. This suggests that the only participants
solving the intermediate RATs could make use of the cues presented to them to solve
the RATs.
In the High-Load Condition, a significant Cue effect, F(1, 32) = 5.32, p = .03,
ηp2 = .142 and a marginally significant Cue x RAT Difficulty effect were found, F(1,
32) = 3.63, p = .07, ηp2 = .102. Different from the findings in the Intermediate-Load
Condition, the interaction effect was in the opposite direction, indicating that the
presence of cue is helpful only when solving easy RATs M = .23, SD = .40, t(14) =
2.24, p = .04, Cohen’s d = 1.32, but not intermediate RATs, M = .02, SD = .24, t(18)
= .41, p = .69, Cohen’s d = .19.
This pattern of interaction suggests that an incubation period can help us utilize
the cue, but the occurrence of the cue-assimilation process is moderated by the
loading of the incubation tasks and the problem difficulty. One of the follow-up
questions is about the mechanism underlying the cue-assimilation process. It has been
hypothesized that spreading-activation occurs during the incubation period and this
would sensitize individuals to cues encountered subsequently by chance (Yaniv &
Meyer, 1987). If the occurrence of spreading activation during an incubation period is
the basis for cue-assimilation, then a significant interaction between RAT Difficulty
and Condition on the lexical decision latency to target words is expected.
Incubation Effects on Knowledge Activation
Spreading Activation: To determine whether spreading activation occurred
during the incubation period, participants’ lexical decision times were compared for
105
the target words and the neutral words (answers for the RATs in the other sets) of the
RATs in the LDT-before trials among the four conditions were compared. Similar to
Experiment I, incorrect lexical decisions and correct decisions with extreme lexical
decision time (<50 ms or >2500ms) were discarded (1.6% of the data). Within each
participant, any lexical decision time longer than 2 SDs from the mean was recoded to
the value found at 2 SD. In this experiment, the skew value of the data in some
conditions exceed 1.5, therefore, the lexical decision times were logarithmically
transformed to diminish skew. Table 4.2.7 presents the means and standard deviations
of the natural and log-transformed lexical decision times for neutral words and target
words by Condition and RAT Difficulty.
106
Table 4.2.726
Means and Standard Deviations of the Natural and Log-Transformed Lexical
Decision Times for Target Words and Neutral Words.
Easy RAT
Target Word
Neutral Word
Target Word
Neutral Word
M (SD)
M (SD)
M (SD)
M (SD)
Condition
No Incubation
Low Load
Intermediate Load
High Load
Intermediate RAT
Original Lexical Decision Time
691.26
711.28
630.90
687.31
(183.30)
(137.25)
(129.10)
(129.46)
670.86
663.16
676.69
712.83
(292.23)
(127.20)
(164.21)
(111.06)
697.23
713.71
714.50
745.80
(160.86)
(139.80)
(165.02)
(119.40)
645.55
724.10
637.70
730.33
(114.38)
(133.25)
(129.10)
(116.47)
Log-Transformed Lexical Decision Time
No Incubation
2.82 (.11)
2.84 (.08)
2.79 (.08)
2.83 (.08)
Low Load
2.80 (.13)
2.82 (.08)
2.82 (.10)
2.85 (.07)
Intermediate Load
2.83 (.10)
2.84 (.08)
2.84 (.10)
2.87 (.07)
High Load
2.80 (.08)
2.85 (.08)
2.79 (.10)
2.86 (.07)
There was no significant difference in the log-transformed lexical decision
times to the answers of the easy and the intermediate RATs when they were presented
as neutral items in the LDTs, t(172) = .12, p = .73, Cohen’s d= .02. This ensures that
107
there is no priori difference in the baseline lexical decision time for target words and
neutral words. To investigate if spreading-activation occurred during the incubation
periods to further sensitize participants to relevant memory items, the log-transformed
data were analyzed using a 2 (Word Type) x 2 (RAT Difficulty) x 4 (Condition)
ANOVA with repeated-measures on Word Type (Target Word vs. Neutral Word).
Table 4.2.8 presents the summary of the ANOVA results.
Table 4.2.827
ANOVA Summary Table for the Effect of the RAT Difficulty, Condition, and Word Type
on Log-Transformed Lexical Decision Time
df
MS
F
p
ηp2
RAT Difficulty (R)
1
< .001
.001
.40
.006
Condition (C)
3
.011
.81
.11
.047
RxC
3
.008
.58
.98
.002
123
.013
Word Type (W)
1
.075
26.79
< .001
.180
WxR
1
.002
0.72
.40
.010
WxC
3
.006
2.03
.11
.050
WxRxC
3
< .001
0.06
.98
< .010
123
.003
Source
Between-Subjects
Error ( between)
Within-Subjects
Error (within)
In general, participants made quicker lexical decisions to the target words than
the neutral words, indicated by the significant Word Type effect, F(1, 123) = 26.79, p
108
< .001, ηp2 = .18. None of the interaction effects between Cue and other IVs.
approached significance, revealing that the lexical decision time difference between
the target words and the neutral words were comparable in all the conditions and in
both types of RATs. This pattern of results suggests that participants were, in general,
sensitized to the answer of the unsolved RATs after the initial approach to the RATs.
However, performing incubation tasks did not further sensitize participants to the
answers of the unsolved RATs. This is inconsistent with the Spreading-Activation
hypothesis. The absence of a significant Word Type x Condition x RAT Difficulty
effect also implies that the cue-assimilation process could not be attributed to the
occurrence of spread of activation within the semantic network during the incubation
period.
Discussion
Experiment II, using easier RATs as insight problems, had more fruitful results
than Experiment I. Experiment II reveals that although the incubation period did not
directly enhance performance on RATs, it facilitated participants to incorporate the
external cue into problem solving and in turn improve performance. Furthermore,
Experiment II reaffirms the meta-analysis by showing that the loading of the
incubation tasks had an impact on the post-incubation RAT performance. Experiment
II also identifies a new moderator on incubation effects: Problem Difficulty, and
reveals a significant interaction between loading and problem difficulty moderators
on the cue-assimilation process. However, Experiment II still fails to offer any
evidence in support of the Spreading-Activation hypothesis.
In the No-Incubation condition, participants solving both easy and intermediate
RATs benefited from the presentation of cues (target words given as LDTs), and the
109
magnitude of the cueing effects was comparable to the effects found in the incubation
conditions. This is different from the prediction that participants in the No-Incubation
condition should be less likely to assimilate the external cue because the problem
representation was not restructured into a better form yet. In the No-Incubation
condition, cues were presented while participants were actively solving the RAT, and
so it was very likely for the participants to aware of the association between the LDTs
and the RATs, and incorporate the cue into problem solving immediately once it was
resumed. It means that, even if participants were not having an appropriate initial
representation of the RAT, they could still utilize the cue embedded in LDTs and
solve the RAT.
However, this conscious-awareness explanation cannot account for the finding in
the incubation conditions that a positive cueing effect arose only when solving
intermediate RATs in the Intermediate-Load Condition and when solving easy RATs
in the High-Load Condition. The Opportunistic-Assimilation hypothesis, suggesting
that incubation period facilitates problem restructuring and in turn increases the
likelihood of cue-assimilation, seems to offer a better explanation for this pattern of
findings.
In the Intermediate-Load condition, performing the incubation tasks occupies
part of the participants’ attentional resource, and this would then create a diffused
attentional state. This attentional state can facilitate the problem restructuring process,
and in turn increase the likelihood of assimilation of the external cue subsequently
encountered by chance. This restructuring process would be essential especially when
solving intermediate RATs because participants’ initial problem representations may
be faulty. When solving easy RATs, the initial problem representation is likely to be
already in an appropriate form. Therefore, it may not require any restructuring, and
110
performing incubation tasks and a shift from a focused to a diffuse attentional state
would not be helpful. Indeed, the diffused attentional state may even cause the
occurrence of some unneeded unconscious processes, which may induce unnecessary
changes on the problem representation, and this may then inhibit the cue-assimilation
process.
This unconscious cue-assimilation explanation can also account for the opposite
findings in the High-Load condition and the null findings in the Low-Load condition.
In the High-Load condition, the incubation tasks were very cognitively demanding, so
that participants had to focus fully on them. This focused attentional state would not
foster the occurrence of any unconscious restructuring process. Therefore,
participants in the intermediate RAT group did not benefit from the presence of cues.
For participants solving easy RATs, a focused attentional state could inhibit the
occurrence of any redundant unconscious cognitive processes. The problem
representation, which was already at an appropriate form, would then remain
unchanged. Hence, participants could still able to incorporate the cues into problem
solving after the incubation period, and benefit from the presentation of the cues.
The null cueing effect in solving both types of RATs in the Low-Load condition
may be due to the fact the low-load incubation tasks were too easy. In Experiment II,
participants had 8 seconds to solve each MRT. According to the data from Experiment
I, participants took an average of about 3 seconds to solve an easy MRT. In the
remaining 5 seconds, participants may shift attention to think about other things, or to
think about the previously-presented but unsolved RATs. According to the
unconscious cue-assimilation explanation, shifting attention to other tasks should
increase the likelihood of cue-assimilation in solving intermediate RATs, but it should
not be helpful at all when solving easy RATs. Conversely, resuming work on the
111
unsolved RAT would facilitate the cue-assimilation process only when solving easy
RATs. Thus, it implies that whether the presence of cue has an effect on the solution
rate depended on what participants have chosen to do during the remaining time of
the incubation period. There should be large individual differences because the
experiment instructions did not give any specific guidelines as to whether participants
should think about the unsolved problem or not if they finished the MRTs and ATs
early. Thus, due to uncontrolled individual differences, it is not unexpected that the
overall cueing effects in the Low-Load condition did not approach significance.
Some may question the reliability of the findings regarding the effect of cues
because the meta-analysis reported that the presence of a cue was not a significant
moderator of the incubation effect in past incubation studies. There are several
explanations for this inconsistency. First, Experiment II reveals that the cueing effect
is moderated by the loading of the incubation tasks and the difficulty of the insight
problems. However, due to the small number of the available studies, the
meta-analysis only examined the overall effect of cue, and did not examine the
interaction between the presence of cue, loading of the incubation tasks, and problem
difficulty, on the incubation effect size. Second, there are numbers of ways to present
the cues, and the presentation of the cue in some of the past studies may not be
effective, which would certainly lessen the overall cueing effect.
In Experiment II, when solving intermediate RATs, the magnitude of the cueing
effect in the No-Incubation condition was comparable to the effect found in the
Intermediate-Load incubation condition. Some may then challenge the importance of
having an incubation period. However, in real-life problem solving situations, the
opportunity for encountering additional external cues would be low if there is no
change in the external environment. Shifting to perform other irrelevant incubation
112
tasks will cause a change in the environment, and this will increase the chance of
encountering external cues. The findings of Experiment II suggest that performing
intermediate-load incubation tasks generates the optimal attentional state for
assimilating any chance-encountered cues.
The results of Experiments II offer empirical evidence for the
Opportunistic-Assimilation hypothesis, in that incubation periods facilitate the
occurrence of unconscious processes (e.g., problem restructuring) which would
increase the likelihood of assimilating cues subsequently encountered by chance. The
findings of Experiment II enrich this hypothesis by identifying two procedural
moderators of this cue-dependent incubation effect: loading of the incubation task and
problem difficulty.
One further question concerns the nature of the unconscious processes occurring
during the incubation period. It has been suggested that a cue-assimilation effect can
be attributed to the occurrence of spreading activation during the incubation period
(Yaniv & Meyer, 1987). However, the findings reported here regarding LDT
performance challenge this view. Experiment II reveals that participants were more
sensitive to the answer even if they could not solve it at the initial attempt. This is
strong evidence for the occurrence of spreading activation during the initial attempt.
However, the incubation period did not further sensitize participants to these relevant
concepts, reflected in the non-significant difference in LDT performance between the
four conditions. This does not support the Spreading-Activation hypothesis. The
non-significant difference also implies that the cue-assimilation process cannot be
attributed to the occurrence of spreading activation during an incubation period.
On the whole, the findings of Experiment II support the unconscious role of
incubation, and offer empirical evidence for the Opportunistic-Assimilation
113
hypothesis, but offer no supporting evidence for the Spreading-Activation hypothesis.
The cue-dependent incubation effect is moderated by two factors: loading of
incubation tasks and problem difficulty. Although spreading activation did not occur
during the incubation period in solving the RATs in Experiment II, it is still too early
to jump to the conclusion that spreading activation is not a mechanism underlying an
incubation period. There is a possibility that spreading activation occurs only if an
individual’s mind is fixated on some misleading concepts, and the role of spreading
activation is to re-distribute the activation (Sio & Rudowicz, 2008). To verify this
possibility, Experiment II was replicated but using misleading insight problems.
The use of misleading insight problems also allows the examination of the
Selective-Forgetting hypothesis by comparing the lexical decision latency to
misleading concepts in incubation and no-incubation conditions. If
selective-forgetting occurs during an incubation period to desensitize individuals to
misleading items, then an increase in LDT latency to misleading items in the
Incubation condition is expected.
Another reason for replicating the experiment on other insight problems is to
examine the generalizability of the findings of Experiment II. It is unclear whether the
RAT can be considered as an insight problem or a linguistic completion task,
suggesting it may not be representative of all insight problem-solving tasks. Bowden
and Jung-Beeman (2003a) found that participants claimed that they solved RATs
sometimes with insight and sometimes without insight. This suggests the need to
replicate Experiment II with another type of linguistic insight problem.
Experiment III was carried out to address the aforementioned concerns.
Experiment III examined the incubation effect on another type of linguistic insight
problem: Rebus. A Rebus combines verbal and visual clues to a common phrase (e.g.,
114
“you just me” = “just between you and me”; “peazzzce” = rest in peace). A correct
interpretation of the visual and verbal cues is essential in order to solve a rebus. The
validity of rebuses as insight problems was supported by a significant positive
correlation between individuals’ performance in solving rebus and self-rated insight,
and RAT performance (MacGregor & Cunningham, 2008). In Experiment III, both
neutral and misleading Rebuses were used.
Experiment III also improved the methodology of Experiment II by matching the
amount of work assigned to participants among different experimental conditions. In
Experiment II, participants in the No-Incubation condition were not required to
perform the MRTs and ATs. However, in Experiment III, participants in the
No-Incubation condition were also asked to perform the MRTs and ATs at the end of
each trial.
115
EXPERIMENT III
Experiment III had two objectives. The first one was to replicate the cue-dependent
incubation effect found in Experiment II on another insight problem: Rebus. The
second objective was to investigate if spreading activation and selective forgetting
occur during an incubation period when solving insight problems embedded with
misleading hints.
The findings of Experiment II suggest that performing intermediate-load
incubation tasks can generate the largest incubation effect in solving intermediate
difficulty problems. Therefore, in Experiment III, only Rebuses at an intermediate
level of difficulty were used because Experiment II reported a null incubation effect
when solving problems at an easy level. Also, there was only one incubation
condition: the Immediate-Load condition. The No-Incubation condition remained and
served as the control condition.
Neutral and misleading Rebuses were used in Experiment III to examine the
Spreading-Activation and Selection-Forgetting hypotheses. If spreading activation
impacts only when individuals are fixated on misleading concepts, then an increased
sensitivity to relevant memory items after an incubation period is expected only for
participants solving misleading Rebuses. If selective forgetting is one of the
mechanisms underlying incubation, then a decrease in sensitivity to misleading
concepts is expected after the incubation period, which will in turn resolve fixation on
the misleading items.
Also, it was expected that a significant cueing effect in solving Rebus would be
revealed in the Intermediate-Load condition. This prediction arises because,
according to the findings of Experiment II, performing incubation tasks can facilitate
116
unconscious problem restructuring, and in turn, increase the likeliness of assimilating
external cues.
Method
Participants
69 (51 females, 18 males) students from Lancaster University participated for
course credits or payment. The mean age was 20.99 years with a standard deviation of
4.82. All participants were native English speakers.
Materials
Two sets of Rebus, neutral and misleading, were prepared and used as the
problem-solving tasks in this study. Eight Rebuses selected from the pool developed
by MacGregor and Cunningham (2008), with a solution rate range from 25-50%,
were included in the neutral set. Each Rebus was presented inside a rectangular border.
These 8 Rebuses were also used as the stimuli in the misleading set. In the misleading
set, a misleading hint was presented below the rectangular border of each of these
eight Rebuses. The misleading hint is a word that can form an association with one of
the clues of the presented Rebus but is not part of the answer. It was expected that the
misleading hints would direct the participants to interpret the clue in an inappropriate
way and search in the wrong direction. To avoid participants realizing that the
presented hints were always irrelevant, two filler Rebuses, with a relevant hint(a word
which is part of the answer) presented below the rectangular border, were added to the
misleading set. The relevant hints in the filler Rebus should also lure the participants
to attempt to use the misleading hints to solve the misleading Rebuses. To match the
117
number of stimuli in the neutral and misleading sets, these two filler Rebuses were
also included in the neutral set, but the hints below the rectangular box were removed.
In total, there were 10 Rebuses in each set, but the filler Rebuses were not
included in the analyses. The Rebuses presented in the neutral and misleading sets are
presented in Appendix H.
Lexical Decision Tasks (LDTs) were again adapted to examine participants’
level of sensitivity to the relevant solution concepts and the misleading hints. A set of
LDTs consisted of five items including neutral words, pseudo-words, and target words
(a word which is part of the answer). If the previously-presented Rebus was a
misleading one, the misleading hint was also included in the LDTs. All presented
words were high frequency words (an occurrence of 90 or more in a million). To
ensure that there was no baseline lexical decision time difference between the neutral,
target, and misleading words, 15 participants were recruited from the Lancaster
University to perform a set of lexical decision tasks including the presentation of
these three types of words and the pseudo-words used in Experiment III. The
presentation of the lexical decision tasks was the same as the one in Experiment II.
The log-transformed LDT data were analyzed using one-way ANOVA with
repeated-measures on Word Type: neutral, relevant, and misleading hint. The main
effect of Word Type was not significant, F(2, 28) = .24, p = .82, ηp2 = .031, implying
that there is no prior difference in the lexical decision latencies for these three types of
words. Hence, any difference in lexical decision latencies in the Experiment III
cannot be attributed to this factor.
Mental Rotation Tasks (MRTs) and Arithmetic Tasks (ATs) presented in
Experiment III were the intermediate-load MRTs and ATs in Experiment II.
118
Design
This experiment employed a 2 x 2 x 2 mixed factorial design. The
within-subjects factor was whether the LDTs were presented before or after the
second attempt at the Rebus. The two between-subjects factors were the Incubation
conditions: No Incubation (NI) and Incubation (IN), and the types of the Rebuses:
Neutral and Misleading.
Similar to Experiment II, the presentation order of the LDT-before and
LDT-after trials was counterbalanced across participants. Participants were randomly
assigned to receive the LDT-before trials either in the 1st, 2nd, 4th, 7th, and 9th trials or
in the 3rd,5th, 6th, 8th, and 10th trials. The two filler Rebuses were always presented at
the first and fifth trials.
The main experiment consisted of 10 trials. In every trial, regardless of
condition, participants were asked to solve 1 Rebus (either neutral or misleading,
depending on the type of Rebus they were assigned to solve), 5 LDTs, 5 MRTs, and 8
ATs. The presentation order of these four types of tasks varied among participants,
depending on which incubation condition they were assigned. Each trial in the
non-incubation condition began with the presentation of a Rebus, followed by the
presentation of a set of LDTs and then the re-presentation of the same Rebus (or vice
versa, depending on whether it was a LDT-before or LDT-after trial). After that, 5
MRTs and 8 ATs were presented. In the incubation condition, after the first
presentation of the Rebus, 5 MRTs and 8 ATs were presented; then the LDTs were
presented followed by the presentation of the same Rebus (or vice versa, depending
on whether it was a LDT-before or LDT-after trial).
119
Procedure
Participants were randomly assigned to solve the neutral or the misleading Rebus
set, in either the No-Incubation or Incubation condition. All the task instructions and
stimuli were presented on a computer using bespoke software.
Participants were first informed that the main experiment involved solving
Rebuses, MRTs, ATs, and LDTs. They were then given instructions on how to
complete the tasks and were reminded to finish the tasks as quickly and accurately as
possible. After this, they were prompted to solve one Rebus, one MRT, one AT, and
two LDTs for practice. For participants assigned to solve the misleading set, the
practice Rebus presented to them was a Rebus in which a relevant hint appearing
below the rectangular frame. Prior to the main experiment, participants were
reminded that the hints presented below the rectangular frame could be relevant or
misleading.
Regardless of condition, at the beginning of each experimental trial, a Rebus was
presented for 30 seconds. After the time was up or the participants answered the
Rebus, participants in the No-Incubation condition were then prompted to do an LDT
set followed by a second attempt to the same Rebus (or vice versa depending on
whether it was a LDT-before or LDT-after trial). After that, participants were told to
solve MRTs and ATs for two minutes. Participants in the Incubation condition were
first prompted to solve the MRTs and ATs, before performing the LDTs and
re-approaching the problems. If participants solved the Rebus at the first attempt, they
were instructed to give the same answer again in the second attempt to the same
Rebus.
Each Rebus was presented in the middle of the screen. Participants were told
that they could respond at any time within the presentation time (30 seconds) by first
120
pressing the C key to pause the timer, and then reporting their answer verbally to the
experimenter for immediate judgment. If their answer was correct, then they were
prompted to do the next task; otherwise, they had to continue to solve the Rebus until
the time was up or until a solution was reached. The number of correctly solved
problems and the response latencies were recorded. The response latency was
measured as the duration from the onset of problem presented until the participant got
the right answer or until the presentation period ended. The Rebuses in the misleading
set were presented in the same way except that the hint for each Rebus was presented
for 1 second prior to the presentation of the Rebus, and the hint also appeared below
the rectangular frame during the presentation of the Rebus. The presentation of the
LDTs, MRTs and ATs were the same as in Experiment II.
Results
Before running the main analyses to verify the predictions, response time and
accuracy for the neutral and misleading Rebuses at the first attempt were compared to
verify that the misleading hints did indeed mislead the participants and impair their
performance. Table 4.3.1 presents the means and the standard deviations of number of
correct responses and response time to the neutral and misleading Rebuses at the first
attempt in each incubation condition.
121
Table 4.3.128
Means and Standard Deviations of the Number of Correct Responses and Response
Time at the First Attempt by Rebus Type and Condition
Number of Correct Responses
Response Time
Neutral
Misleading
Neutral
Misleading
Rebus
Rebus
Rebus
Rebus
M (SD)
M (SD)
M (SD)
M (SD)
No Incubation
3.82 (1.23)
2.82 (2.10)
20.24 (4.15)
23.00 (5.87)
Incubation
4.67 (1.33)
2.44 (1.46)
16.91 (3.85)
25.19 (4.05)
Condition
Note. The two filler rebuses were excluded, and only 8 rebuses were included in the
analysis.
Two 2 (Rebus Type) x 2 (Condition) between-subjects ANOVAs was carried out,
using number of correct responses and response time as the dependent variables. The
factor Rebus Type had a significant effect on both number of correct responses, F(1,
66) = 18.56, p < .001,ηp2 = .220, and response time, F(1, 66) = 25.90, p < .001,ηp2
= .282. This supports the assumption that the misleading hints would induce a
negative effect on participants’ performance with Rebus problems. The effect of
Condition was not significant: number of correct responses: F(1, 66) = .28, p = .60,
ηp2 = .004, response time; F(1, 66) = .39, p = .54,ηp2 = .006, suggesting that there
were no performance differences between participants in these conditions.
There was a significant Rebus Type x Condition interaction in response latencies,
F(1, 66) = 6.49, p = .01,ηp2 = .09. Two independent t-tests were carried out as
follow-up analyses to compare the response time to neutral and misleading Rebuses in
each condition. Participants in the Incubation condition took significantly longer to
122
solve the misleading Rebuses than the neutral Rebuses, p < .001. Participants in the
No-Incubation condition also took longer to solve the misleading RATs, but the
difference did not approach significance. The null effect in the No-Incubation
condition may be due to the fact that response time was a less sensitive and reliable
performance indicator than number of correct responses. However, the overall results
support the view that the presence of the misleading hints could reduce participants’
performance on Rebuses.
Similar to the Experiment I and II, the number of correct responses at the first
and the second attempt was used to examine the incubation effect on problem solving
performance, and the data on lexical decision tasks were analyzed to assess if there
were any changes in sensitivity to misleading and relevant items. Four separate
analyses were carried out to address four questions: 1) if performing
intermediate-load incubation tasks could enhance performance on Rebuses, 2) if an
incubation period filled with intermediate-load tasks could facilitate the
cue-assimilation process, 3) if spreading activation occurred only when solving
misleading Rebuses, and 4) if selective forgetting occurred during the incubation
period to suppress the activation of the misleading concepts.
Incubation Effect on Problem-Solving Performance
Performance Improvement: To assess the incubation effect on performance,
an improvement score, which was the proportion of Rebuses not solved at the first
attempt that were solved at the second attempt, was calculated for each participant.
These numbers were then averaged across participants solving the same type of
Rebuses in the same condition. Similar to Experiments I and II, only data on Rebuses
presented in the LDT after trials were examined. The data on the filler Rebuses were
123
not included in the analysis. Table 4.3.2 presents the means and standard deviations of
the number of correct responses in each attempt, and the improvement score, in the
LDT-after trials, by Condition and Rebus type.
Table 4.3.229
Mean and Standard Deviations of the Number of Correct Responses in the Two
Attempts by Condition and Rebus Type
First Attempt
Second Attempt
Improvement Score
M (SD)
M (SD)
M (SD)
Condition
Overall
Neutral Rebus
2.14 (0.71)
2.41 (.82)
.14 (.28)
No Incubation
2.06 (0.75)
2.35 (.86)
.14 (.28)
Incubation
2.24 (0.75)
2.47 (.80)
.14 (.28)
Misleading Rebus
Overall
1.26 (1.47)
1.61 (.95)
.11 (.18)
No Incubation
1.43 (1.03)
1.81 (.98)
.13 (.20)
Incubation
1.11 (1.02)
1.44 (.92)
.09 (.16)
The overall improvement score was significantly larger than zero, Neutral
Rebuses: t(33) = 2.86, p < .001, Cohen’s d = 1.00; Misleading Rebuses: t(33) = 3.55,
p < .001, Cohen’s d = 1.24. If the incubation period can improve the performance on
Rebuses, then the improvement scores of participants in the Incubation condition
should be larger than the scores of those in the No-Incubation condition. A 2
(Condition) x 2 (Rebus Type) ANOVA on improvement scores was conducted to
124
verify this prediction. The summary of the results is presented in Table 4.3.3.
Table 4.3.330
ANOVA Summary Table for the Effect of Rebus Type and Condition on Improvement
Scores
Source
df
MS
F
p
ηp2
Between-Subjects
Rebus Type
1
.014
.24
.62
.004
Condition
1
.004
.08
.78
.001
Rebus Type x Condition
1
.004
.08
.78
.001
Error ( between)
64
.056
Neither the main effects nor the interaction effects were significant, implying
that participants in the Incubation condition did not demonstrate a larger performance
improvement in solving neutral and misleading Rebuses than participants in the
No-Incubation condition. This does not support the hypothesis that an incubation
period can enhance problem-solving performance.
Opportunistic Assimilation: Another objective of Experiment III was to
replicate the findings of Experiment II concerning the impact of an incubation period
on cue assimilation. In Experiment II, performing irrelevant intermediate-load
incubation tasks did not inhibit the cue-assimilation process, but instead facilitated the
unconscious cue-assimilation process. If this cue-dependent incubation effect is not
item-specific, then participants in the Incubation condition in this experiment should
be able to assimilate the external cues to solve the Rebuses in the second attempt. In
other words, it was predicted that, in the Incubation condition, the degree of
125
performance improvement in the LDT-before trial should be significantly larger than
in the LDT-before trials than in the LDT-after trials. A similar pattern of findings was
expected in the No-Incubation condition, but the mechanism underlying this
cueing-effect would be conscious, rather than unconscious. To verify this prediction,
the difference between the improvement score in LDT-before (Cue) and LDT-after
(No-Cue) trials in each condition was compared. The means and the standard
deviations of the number of correct responses to Rebuses in the Cue and No-Cue
Trials by Condition and Rebus Type are presented in Table 4.3.4. Table 4.3.5 presents
the mean improvement score and its standard deviation in Cue and No-Cue trials by
Rebus Type and Condition.
Table 4.3.431
Mean and Standard Deviations of the Number of Correct Responses in the Cue and
No-Cue Trials in Each Condition.
First Attempt
Second Attempt
Cue
No Cue
Cue
No Cue
M (SD)
M (SD)
M (SD)
M (SD)
Condition
Neutral Rebus
No Incubation
1.76 (0 .97)
2.06 (0.75)
2.18 (0.95)
2.35 (0.86)
Incubation
2.33 (1.14)
2.33 (0.84)
2.72 (1.13)
2.56 (0.86)
Misleading Rebus
No Incubation
1.24 (1.25)
1.59 (1.17)
1.94 (1.20)
1.94 (1.09)
Incubation
1.33 (1.08)
1.11 (1.02)
1.67 (1.03)
1.44 (0.92)
126
Table 4.3.532
Means and Standard Deviations of Improvement Score on the Cue and No-Cue trials,
and the Cueing Effect, by Condition and Rebus Type
Cue
No-Cue
Cueing Effect
(Improvement Score Cue –
No-Cue)
Condition
M (SD)
M (SD)
M (SD)
Neutral Rebus
No Incubation
.19 (.32)
.14 (.28)
.07 (.47)
Incubation
.27 (.37)
.14 (.29)
.11 (.51)
Misleading Rebus
No Incubation
.24 (.25)
.13 (.20)
.14 (.31)
Incubation
.11 (.16)
.09 (.16)
.01 (.20)
A 2 (Rebus Type) x 2 (Cue) x 2 (Condition) ANOVA with repeated-measures
on Cue (Cue vs. No-Cue) was carried out on improvement scores to compare the
degree of improvement in different incubation condition and in solving different types
of Rebuses. The ANOVA results are presented in Table 4.3.6.
127
Table 4.3.633
ANOVA Summary Table for the Effect of Condition and Cue and Rebus Type on
Improvement Scores.
Source
df
MS
F
p
ηp2
Between-Subjects
Condition (Co)
1
.02
0.34
.56
.006
Rebus Type (R)
1
.02
0.38
.54
.006
Co x R
1
.11
1.86
.18
.030
Error ( between)
59
.06
Within-Subjects
Cue (Cu)
1
.224
2.97
.09
.050
Cu x Co
1
.018
0.24
.63
.004
Cu x R
1
.001
0.01
.91
< .001
Cu x Co x R
1
.053
0.70
.41
< .012
Error (within)
59
.075
Neither the main effects nor the interaction effects approached significance,
implying that, in both conditions and both types of Rebuses, the rate of improvement
in the presence and the absence of a cue did not significantly differ. Experiment III
fails to reveal any the facilitatory effect of incubation period on cue assimilation.
However, the null finding in this experiment does not necessarily refute the existence
of a positive role for an incubation period on cue assimilation. The absence of any
cueing effect in this experiment may be due to the ineffectiveness of the cue used,
indicated by the non-significant main effect of Cue (p = .09) on the improvement
score. A more detail discussion on this view is presented in the next section.
128
Incubation Effects on Knowledge Activation
Data from the lexical decision tasks were analysed to determine whether
spreading activation and selective forgetting occur during the incubation period. Prior
to these analyses, the data were first screened to remove outliers and incorrect
responses. Accuracy on the lexical decision tasks was 93% or higher in all groups.
Incorrect lexical decisions and correct decisions with extreme lexical decision time (<
50 ms or >2500ms) were discarded (1.2% of the data). After that, the mean and
standard deviation of the lexical decision latencies were calculated for each
participant. Within each participant, lexical decision times longer than 2SDs from the
mean were recoded to the value found at 2 SDs. The recoded lexical latencies in some
conditions had skew values exceeding 1.5. To diminish skew, the lexical decision
times were logarithmically transformed.
Spreading Activation: Participants’ lexical decision time to target words and
neutral words in the LDT-before trials were examined to test the hypothesis that being
misled is the criterion for the occurrence of spreading activation during an incubation
period. The means and standard deviations of the natural and log-transformed lexical
decision time for neutral words and target words in the LDT-before trials by
Conditions and Rebus Type are presented in Table 4.3.7.
129
Table 4.3.734
Means and Standard Deviations of the Original and Log-Transformed Lexical
Decision Times for Target Words and Neutral Words
Neutral Rebus
Target Word
M (SD)
M (SD)
Condition
No Incubation
Neutral Word
Misleading Rebus
Target Word
Neutral Word
M (SD)
M (SD)
Original Lexical Decision Time
544.39 (83.04)
Intermediate Load 546.42 (123.84)
625.27 (66.76)
561.58 (95.65) 634.64 (92.72)
645.48 (92.52) 567.81 (103.54) 643.34 (82.04)
Log-Transformed Lexical Decision Time
No Incubation
2.73 (.07)
2.79 (.05)
2.74 (.07)
2.80 (.06)
Intermediate Load
2.73 (.09)
2.81 (.06)
2.74 (.07)
2.81 (.06)
It was predicted that participants who were assigned to solve misleading Rebuses
in the Incubation condition would make quicker lexical decisions to target words
relative to the control words. This LDT difference should be less noticeable with
participants solving neutral Rebuses in the same condition and on those solving the
misleading Rebuses in the No-Incubation condition. The log-transformed data were
analyzed using a 2 (Word Type) x 2 (Rebus Type) x 2 (Condition) ANOVA with
repeated-measures on Word Type (Target Word vs. Neutral Word). Table 4.3.8
presents the summary of the ANOVA results.
130
Table 4.3.835
ANOVA Summary Table for the Effect of RAT Difficulty, Condition, and Word Type on
Log-Transformed Lexical Decision Time
df
MS
F
p
ηp2
Rebus Type (R)
1
.003
.52
.47
.009
Condition (C)
1
< .001
< .01
.96
< .001
RxC
1
< .001
.05
.82
.001
Error ( between)
61
.006
Word Type (W)
1
.054
34.80
< .001
.363
WxR
1
< .001
0.20
.66
.003
WxC
1
< .001
0.09
.77
< .001
WxRxC
1
< .001
0.11
.74
.002
Error (within)
61
.002
Source
Between-Subjects
Within-Subjects
Participants generally made quicker lexical decisions to the target words than the
neutral words, indicated by the significant Word Type effect, F(1, 61) = 34.80, p
< .001,ηp2 = .363. The three way interaction between Condition x Rebus Type x Word
was not significant, F(1, 61) = .04, p = .84, ηp2 = .001, implying that participants who
solved misleading Rebuses in the Incubation condition were not more sensitized to
relevant concepts, as compared with other participants. This does not support the
suggestion that spreading activation occurs during an incubation period only when
the problem solver is misdirected.
131
Selective Forgetting: Lexical decision times for misleading hints and neutral
words were analysed to test the Selective-Forgetting hypothesis that an incubation
period helps desensitize individuals to misleading but over-activated memory items.
Only data from participants who were assigned to solve the misleading Rebuses were
included in the analysis (N = 35). The means and standard deviations of the natural
and log-transformed recoded lexical decision time for neutral words and misleading
hints in the LDT-before trials by Conditions are presented in Table 4.3.9.
Table 4.3.936
Means and Standard Deviations of the Original and Log-Transformed Lexical
Decision Times for Misleading Hints and Neutral Words.
Log-Transformed Lexical Decision
Natural Lexical Decision Time
Time
Misleading Hints Neutral Words
Condition
Misleading Hints
Neutral Words
M (SD)
M (SD)
M (SD)
M (SD)
No Incubation
588.27 (111.70)
604.38 (88.89)
2.76 (.07)
2.77 (.06)
Incubation
595.17 (148.72)
612.10 (92.52)
2.77 (.09)
2.78 (.05)
As a corollary to the Selective-Forgetting hypothesis, it was predicted that
participants in the Incubation condition should have longer lexical decision times to
misleading hints relative to neutral words, as compared with participants in the
No-Incubation condition. A 2 (Word Type) x 2 (Condition) mixed-subjects ANOVAs
with repeated-measures on Word Type (Misleading Hint vs. Neutral Word), using
lexical decision time as the dependent variable, was carried out to verify the
prediction. The summary of the ANOVA results is in Table 4.3.10.
132
Table 4.3.10371
ANOVA Summary Table for the Effect of the Condition and Word Type on
Log-Transformed Lexical Decision Time
df
MS
F
p
ηp2
Condition (C)
1
< .001
0.05
.83
.001
Error ( between)
32
.007
Word Type (W)
1
.005
2.12
.16
.060
WxC
1
< .001
0.02
.83
.001
Error (within)
32
.002
Source
Between-Subjects
Within-Subjects
The main effect of Word Type was not significant, F(1, 32) = 2.12. p = .16, ηp2
= .06, indicating that participants’ sensitivity to neutral words and misleading hint was
comparable. The non-significant interaction between Word Type and Condition
implies that a similar non-significant difference was found in both Incubation and
No-Incubation conditions. This is not in line with the prediction that participants in
the Incubation condition would be less sensitive to misleading hints.
Discussion
Experiment III failed to replicate the findings of Experiment II on cue assimilation.
Experiment III reported a non-significant cueing effect in both the Incubation and
No-Incubation conditions. Similar to Experiment II, Experiment III reported a
non-significant post-incubation change in participants’ sensitivity to target words.
This does not support the possibility that spreading activation occurs during
133
incubation period only when problem solvers are misled. Experiment III also
examined the Selective-Forgetting hypothesis, and the findings did not support the
view that a selective-forgetting mechanism occurs during an incubation period to
suppress further the activation level of the misleading concepts.
As in Experiment II, the cues for Rebuses were presented as the items in LDTs.
However, a significant cueing effect was not found in Experiment III. This raises the
question about the validity and the generalizability of the findings of Experiment II.
Yet, the failure in replicating the significant cueing effect in Experiment III may
simply due to the ineffectiveness of the cues used in Experiment III. Unlike the cues
presented in Experiment II, which were the answer of the unsolved RAT, the cues in
Experiment III were just part of the answer, e.g., the word “high” was used as the cue
to the Rebus with the answer “Search high and low”. Also, solving a Rebus requires
the correct interpretation of both verbal AND visual clues (e.g., “you just me” = “just
between you and me”). Therefore, simply presenting one single cue may have limited
effect on the solution rate. This view is in line with Kershaw and Ohlsson’s (2004)
account that difficulty of many insight problems is determined by multiple factors,
and presenting one single cue can only remove a single factor and this is not
effectively enough to help the individual solve the problem.
The LDT performance data in Experiment III and II were more consistent with
each other. Experiment III and II both revealed that the reaction times for target words
were significantly shorter than the neutral words after the first approach to the
unsolved problems, suggesting that activation spreads towards the answers of the
problem during the initial attempt. However, this spreading-activation process did not
continue during the incubation period further to sensitize participants to the answers,
reflected in the null Condition effect on lexical decision latencies to target words in
134
both experiments. Also, the non-significant Condition effect on participants solving
misleading Rebuses eliminates the possibility that spreading activation occurs during
an incubation period only if an individual is misled. This pattern of findings does not
support the Spreading-Activation hypothesis.
Another key finding of Experiment III is the absence of a selective-forgetting
mechanism during the incubation period. Experiment III revealed that participants in
the Incubation condition were not less sensitive to the misleading hints, as compared
with participants in the No-Incubation group. Also, in both conditions, the sensitivity
to misleading hints and neutral words was comparable. There are two possible
explanations for the non-significant differences.
One explanation is that participants were not paying any attention to the
misleading hints, and therefore, the misleading hints could not induce a misleading
effect. This would explain the absence of a selective-forgetting mechanism during the
incubation period. According to this account, participants’ Rebus performance should
not be impaired by the presence of misleading hints. However, the number of correct
responses to misleading Rebuses was significantly lower than the number of correct
responses to neutral Rebuses. Also, each misleading hint was presented alone for one
second prior to the presentation of the corresponding Rebus and during the
presentation of that Rebus, and it would be very difficult for participants to ignore the
misleading hint. Furthermore, if participants did not process the misleading hints at all,
then participants’ sensitivity to the misleading hints should be the same, independent
of whether the Rebuses were solved or not solved after the first attempt. However, the
results of a repeated-measures ANOVA on log-transformed lexical decision latencies,
using Word Type (Neutral Words, Misleading Hints of the solved Rebuses, and
Misleading Hint of the unsolved Rebuses) as the within-subjects factor contradict this
135
prediction. The ANOVA reported a significant Word Type effect was found, F(1, 48)
= 5.85, p = .02,ηp2 = .196. Subsequent pairwise comparisons indicated that the lexical
decision time to misleading hints of the solved Rebuses was significantly shorter that
the lexical decision time to the misleading hints of the unsolved Rebuses, p = .044
and to the neutral words, p < .001, and there was no lexical decision latency
difference between neutral words and misleading hints of the unsolved Rebuses.
Therefore, the possibility that participants did not process the misleading hints is
rejected.
Although participants in this experiment were misled by the misleading hints,
they were able to suppress the activation of the misleading hints, indicated by the
findings that participants were not sensitized to the misleading hints of the unsolved
Rebuses. However, the finding that participants were sensitized to misleading hints of
the solved Rebuses suggests that the suppression inhibition may not persist until the
Rebus is solved. It may be that participants restructure the initial problem
representation when they realize the irrelevance of the misleading hints and the
inappropriateness of the initial problem representation. Once the representation is
successfully restructured into a better form, the suppression mechanism may be
terminated and the misleading hint would then be treated as part of the information
presented in the task environment. For those who did not manage to restructure the
problem, the suppression mechanism would persist. This may be the reason of the
findings that participants’ sensitivity to the misleading hint of the Rebus at the end of
the first attempt was the function of whether the Rebus was solved or unsolved.
However, the introduction of the incubation period did not further strengthen the
suppression mechanism, reflected in the null Condition x Word Type effect on the
lexical decision times. This pattern of findings suggested that selective forgetting
136
occurs during the initial approach but not during the incubation period. This does not
support the Selective-Forgetting hypothesis. The findings of Experiment III suggest
that selective forgetting is not the mechanism underlying incubation period to
overcome the fixation effect.
137
DISCUSSION: EXPERIMENT I, II, and III
Experiment I, II, and III adapted a multi-dependent variable approach to examine the
three unconscious-work hypotheses concerning incubation. The combined findings of
these three experiments provides most support for the Opportunistic-Assimilation
hypothesis, in which performing incubation tasks can facilitate the assimilation of
chance-encountered cues, and that this cue-dependent incubation effect is moderated
by the loading of the incubation tasks. This is consistent with the findings of the
meta-analysis. The present experiments also further extend the meta-analysis findings
by showing that problem difficulty is another potential moderator of the incubation
effect. Experiment I, II, and III, however, do not offer evidence to support the
Spreading-Activation and Selective-Forgetting hypotheses.
The present experiments reveal that cue assimilation was more likely to occur
when solving problems at an intermediate level of difficulty and when the incubation
period was filled with intermediate-load tasks. Performing intermediate-load
incubation tasks should occupy a part of participants’ cognitive resource and help
generate a diffused attentional state. This is in line with the meta-analysis findings
that incubation effects are most significant when the incubation periods are filled with
irrelevant incubation tasks that can occupy part but not all of an individual’s cognitive
resource and prevent a focused attentional state. These consistent findings support the
contention that the processes underlying the incubation period are unconscious in
nature. The significant interaction between problem difficulty and loading of the
incubation task implies that problem difficulty is another moderator of the incubation
effect.
The non-significant changes in lexical decision latency suggest that the
138
incubation effects found in the experiments may not be the outcome of the occurrence
of knowledge-activation processes during the incubation period. The findings of
Experiment I, II, and III revealed that participants were sensitized to target words and
they could also inhibit the activation of the misleading hints even if they did not solve
the problems. However, introducing the incubation period did not further increase the
activation level of the target words or strengthen the inhibition mechanism. The
absence of spreading activation during the incubation period in Experiment III also
undermines the possibility that spreading activation occurs during the incubation
period when individuals are fixated on misleading concepts.
Some may challenge the present findings as they are not fully consistent with
the findings of some previous studies that also used LDTs to assess the incubation
effect. Sio and Rudowicz (2008) created a list of misleading RATs where the stimuli
of these RATs activate irrelevant knowledge in the domain of GO, and these RATs
misled GO experts to believe that the misleading RATs were GO-relevant. They asked
the Chinese GO experts to solve these misleading RATs and neutral RATs, either
continuously (control group), with filled incubation period, or with two minutes rest.
They have found that the GO experts in the filled-incubation condition made quicker
lexical decisions to unsolved misleading RATs, as compared with those in the control
condition.
The disparity between Sio and Rudowicz ’s (2008) and Experiment III’s
findings may be due to the difference in the nature of the misleading effect, as well as
the cognitive characteristics of the participants in these two studies. In Sio and
Rudowicz’s (2008) study, GO experts were asked to solve a list of RATs where the
stimuli activate irrelevant GO-related knowledge. The misleading effect was the result
of the automatic activation of the domain-specific knowledge. However, in
139
Experiment III, the misleading effect was induced by presenting an external object
prior and during the presentation of Rebus. This misleading effect may not be as
strong as the internally-generated fixation effect in Sio and Rudowicz’s (2008) study.
Moreover, unlike the participants in Experiment III, the participants in Sio and
Rudowicz’s study were GO experts and their GO-related knowledge was well
connected and easily activated. When an individual is solving an insight problem
containing a stimulus in his/her domain of expertise, domain-specific knowledge will
be activated automatically, implying that the activation in a semantic network will be
unevenly distributed. Having a well-structured task-specific knowledge network and
being strongly fixated on irrelevant task-specific knowledge may be the necessary
conditions for the occurrence of spreading activation.
The study of Zhong, Dijksterhuis, and Galinsky (2008) also reported a
decreased lexical decision times for answers of the unsolved RATs after performing
irrelevant incubation tasks. However, a direct comparison between the findings of the
present experiments and their study is inappropriate because of methodological
differences between them. In Experiment I, II, and III, a set of LDTs was presented
after the initial attempt to one RAT. In Zhong et al.’s (2008) study, participants had to
solve a list of RATs and then a list of LDTs containing the answers of the
previously-presented RATs. This means that participants in their study were having
more than one unsolved RAT in the mind while performing the LDTs. Having many
unsolved problems may magnify the feeling of reaching an impasse, and increase the
perceived need for restructuring, and this may facilitate the occurrence of
spreading-activation. Whether these kinds of meta-cognitive effect arise at a
conscious or unconscious level remains an open question.
The findings of Experiment III offer evidence to reject the Selective-Forgetting
140
hypothesis. Experiment III revealed that selective forgetting occurred during the
initial approach to suppress the activation level of the misleading hints, yet this
inhibition mechanism was not further strengthened during the incubation period. This
does not support the Selective-Forgetting hypothesis. One may argue that the
misleading effect generated in Experiment III was a relatively weak one and
participants could overcome it easily. An incubation period would help suppress the
activation of the misleading memory items only if individuals are strongly fixated on
them. However, the study of Sio and Rudowicz (2008), using RATs that can generate
a strong fixation effect also reported the absence of any activation-suppression
mechanism during the incubation period.
If selective forgetting is not the mechanism underlying an incubation period,
then an alternative explanation is needed to account for the findings of the study of
Smith and Blankenship (1989), which have been considered as strong evidence of the
Selective-Forgetting hypothesis. Smith and Blankenship (1889) found that
participants in the incubation condition were less able to recall the misleading cues
and demonstrated a larger degree of performance improvement on the insight
problems, as compared with those who had to solve the insight problems continuously.
Smith and Blankenship (1989) concluded that forgetting of misleading cues was the
mechanism underlying performance improvement. However, forgetting of misleading
cues may simply reflect the natural decay of participants’ memory on the
problem-related information after the incubation period. The incubation periods in
Smith and Blankenship study were 5 and 15 minutes long, which are much longer that
the incubation period in Experiment III and in the study of Sio and Rudowicz (2008).
Therefore, it is likely that participants in Smith and Blankenship’s (1989) study had
forgotten part of the problem-related information after the incubation period. Also,
141
they did not compare participants’ performance on recalling other problem-related
information. Therefore, it is difficult to conclude whether the forgetting of the
misleading cues is the result of selective or general forgetting. There is a possibility
that the performance improvement was the outcome of the occurrence of other
mechanisms during the incubation period. The co-occurrence of forgetting the
misleading cue and performance improvement created an illusion that selective
forgetting occurred during the incubation period to overcome a fixation effect.
The absence of evidence for spreading activation and selective forgetting in the
present experiments suggests that any restructuring processes that occur during the
incubation period are not the knowledge-activation processes proposed by the
unconscious account. It may be the case the incubation effects on cue assimilation
reported in the present experiments are the outcome of the occurrence of
meta-cognitive processes (e.g., search strategic shift). In order to verify this
possibility, incubation experiments should be carried out on knowledge-lean visual
insight problems, where their solutions require changes in search strategy. .A
significant incubation effect in solving visual insight problem would be an evidence
of the occurrence of meta-cognitive processes during an incubation period.
The meta-analysis presented in the previous chapter has also suggested that
incubation effects, and by implication the mechanisms by which individual attempts
to solve problems, differed according to whether were linguistic or visual. Therefore,
in order to fully understand the incubation effect, incubation experiments using visual
insight problems are needed.
Apart from examining the incubation effect on different classes of problems,
there is another research gap that needs to be filled in order fully to understand the
nature of incubation. The meta-analysis and the Experiments I, II, and III have
142
identified some key procedural factors moderating the incubation effect, and this
offers a potential explanation to the mixed findings of past studies. However, not all
the inconsistencies of past findings can be attributed to the heterogeneous
experimental settings. For example, Olton and Johnson (1976) failed to obtain
evidence of incubation, even following the experimental settings of previous
incubation studies that reported positive incubation effect(e.g., Dreistadt, 1969) . This
suggests the possibility of the existence of additional cognitive moderators of the
incubation effect. One of the potential cognitive moderators may be attentional
allocation styles. The findings of Experiment II and the meta-analysis results suggest
that incubation tasks help create a diffuse attentional state that facilitates the
occurrence of some relevant unconscious processes. According to this, individuals
who typically have a focused attentional state should benefit more from the
incubation period than those who have already have a tendency to allocate their
attention broadly. Experiment IV, an incubation study of visual insight problems, was
carried out to address these questions.
143
CHAPTER V: EXPERIMENT IV-VISUAL INSIGHT PROBLEMS
The meta-analysis supports the existence of multiple types of problem-specific
incubation effects. Experiment I, II, III investigated incubation effects on linguistic
insight problems. Experiment IV was carried out to examine incubation effects on
visual insight problems. Experiment IV also investigated the impact of the length of
preparation period and attentional allocation styles on incubation in solving visual
insight problems.
The regression models presented in Chapter II have significant but limited
predictive power, in that they can only account for 33% of the variance in the
incubation effect sizes. This implies the some key moderators are missing from the
model. The findings of Experiments I, II, and III have suggested that problem
difficulty is one of these missing moderators. However, the moderators of incubation
are not necessarily procedural in nature. The inconsistency of the findings between
studies sharing the same or similar paradigm suggests the existence of non-procedural
moderators. One of the potential non-procedural moderators is individuals’ attentional
allocation styles.
The findings of Experiments I, II, and III suggest that performing incubation
tasks helps divert individuals’ attention away from the problem, and this facilitates
relevant unconscious processes. According to this hypothesis, individuals who have a
focused attentional state should benefit more from an incubation period than those
who always allocate their attention broadly. As a consequence, even under the same
procedural settings, individuals may still benefit differentially from an incubation
period. However, most of the pervious incubation studies have overlooked the impact
of this cognitive moderator. The lack of control of this cognitive characteristic may be
144
one of the reasons for the diversity of past findings. To fill this research gap,
Experiment IV examined the link between attentional allocation styles and incubation
effects.
Examining the impact of attentional allocation styles on incubation effect also
offers additional data to determine the moderating role of preparation periods on
incubation. It has been suggested that a long preparation period allows individuals to
explore the entire initial problem space to reach an impasse, which is a necessity for
the occurrence of processes underlying restructuring, and these processes can be
further facilitated by a post-impasse incubation period (Seifert et al., 1995). In line
with this, the meta-analysis presented in Chapter III revealed that when solving visual
insight problem, long preparation periods gave rise to larger incubation effects.
If exhausting the problem space and reaching an impasse is crucial for
benefiting from an incubation period, then individuals’ attentional allocation styles,
which affect how efficient they are in executing focused search, should influence the
incubation effect size. Also, the impact should be more profound when the
preparation period is limited. When the preparation period is short, individuals who
tend to allocate their attention broadly may not have enough time to exhaust the
search space. Thus, they may be less likely to benefit from the incubation period.
However, when the preparation period is long, most of the individuals (either with
focused or diffuse attentional allocation style) would have enough time to reach an
impasse, and therefore, have the same opportunity to benefit from the incubation
period. Hence, if the proposed reach-of-impasse explanation is correct, then an
interaction between preparation period length and individuals’ attentional allocation
styles on the incubation effect size is predicted.
The core difficulty in conducting an experiment to examine the impact of
145
attentional allocation style on incubation is the recruitment of participants with
similar background (e.g., age, gender, intelligence) but significantly different
attentional styles (focused vs. diffuse). The findings of studies carried out by Nisbett
and his colleagues on cultural differences in cognitive processing offer a solution to
this problem. Nisbett(2003) suggested that Asians and Westerners have different
attentional allocation styles in problem solving, and have demonstrated this difference
under laboratory settings. Ji, Peng and Nisbett(2000) found that Americans, compared
with East Asians, were more field-independent, in that their perception of an object
was less likely to be influenced by the background in which it appeared. Chua,
Boland and Nisbett(2005) went further to examine if the field-independence
difference was due to the ways Westerns and Asians allocate their attention in
perception. They studied the eye movements of Chinese and American graduate
students when perceiving scenes with a single foreground object on a complex
background. American students were found to fixate more on the focal object, while
Chinese students made more saccades to the background. These patterns of results
suggest that Asians tends to allocate their attention broadly in scene perception while
Westerns would focus narrowly on the focal object. Asian students were also more
likely to bind the foreground object with the background, and they were also less
likely to recognize correctly old foreground objects when presented in a new
background. American students were less influenced by the backgrounds of the scenes,
and they outperformed Asian students in recognizing the focal objects against either
new or old background (Chua, Boland, & Nisbett, 2005). A similar pattern of
performance differences was observed when Masuda and Nisbett(2006) asked Asian
and Western students to view a sequence of scene pictures while changes in focal
object and background were introduced during the sequence of presentations. They
146
found that Asian students were more likely to notice changes in contextual
information (e.g., change of the background color) while Western students were more
likely to notice changes in the foreground object.
In summary, the findings of studies carried out by Nisbett and his colleagues
converge to support the view that Asians have a tendency to allocate their attention
broadly, while Westerners are usually narrowly focused in problem solving. If it is the
case that an individual’s attentional allocation style is a moderator of the incubation
effect, then an East-West difference in the degree to which they benefit from different
types of incubation period is predicted.
In Experiment IV, University students from Britain and Hong Kong were
recruited because of the known differences in attentional allocation styles between
Asians and Westerners. They were asked to solve a list of visual insight problems in a
non-incubation condition, an incubation condition with a long preparation period, or
an incubation condition with a short preparation period. The study predicted a
significant moderating effect of the length of preparation period and attentional
allocation style on incubation, and a significant interaction between these two factors.
Method
Participants
Fifty University students from Hong Kong Universities (Female: 26, Male: 24)
and 59 University students from Lancaster University (Female: 30, Male: 29) were
recruited for this study. The mean age of the HK participants was 22.63 with a
standard deviation of 3.74. The mean age of the UK participants was 19.19 with a
standard deviation of 1.12. The reason for recruiting students from these two cultural
147
backgrounds is due to reported differences in attentional allocation styles between
Asians and Westerners (Nisbett, 2003).
Materials
Eight visual insight problems were used as the problem-solving tasks in this
study, and they were clustered into four different sets (two in each set), in terms of the
nature of restructuring required. The English and Chinese4 descriptions of the
problems are presented in Appendix I. These problems have been used as insight
problems in many insight studies (Ash & Wiley, 2006; Gilhooly & Murphy, 2005;
Ohlsson, 1992), excepting Problem 7 (Set 4), which is the modified version of the
eight-coin insight problem used only by Ormerod, MacGregor, & Chronicle (2002).
All these problems are knowledge-lean, in that their solutions require a shift of search
strategy. Also, these visual problems are relatively language-free, which is very
important for this study with participants speaking different languages.
The intermediate-load Mental Rotation Tasks (MRTs) and Arithmetic Tasks
(ATs) presented in Experiment III were used as the incubation tasks in this
experiment.
Witkin’s Group Embedded Figure Test(GEFT, Witkin, Oltman, Raskin, &
Karp, 1971) was used to determine participants’ attentional allocation style. GEFT is
originally designed as a measure of individuals’ field-dependence: the degree of
perception of an object is influenced by the background or environment in which it
resides. The test required individuals to trace out a simple shape embedded in a larger
complex figure. It consists of a practice section containing seven simple items and
4
The Chinese descriptions were translated and confirmed by a bilingual (English-Chinese) psychology
postgraduate in the University of Hong Kong.
148
two test sections, each of which contains nine more difficult items. The completion
time limit for the second and the third sections was 5 minutes each. The test scores
ranged from 0 to 18 depending on the number of shapes traced correctly.
Field-independent individuals should be more effective in ignoring the surrounding
field and disembedding the simple shapes from the complex figures, and in turn, score
higher on the GEFT.
There is a strong relationship between an individual’s degree of
field-dependence and their attentional allocation style. Field-dependent individuals
tend to allocate attention broadly and are more capable of noticing the global aspects
of the to-be-processed information, while field-independent individuals tend to focus
on partial aspect and better at focused and control processes (Bennink, 1982; Clark &
Roof, 1988; Cochran & Davis, 1987; Goode, Goddard, & Pascual-Leone, 2002;
Miyake, Witzki, & Emerson, 2001; Tsakaniko, 2006).
In this study, GEFT was used to measure participants’ attentional allocation
styles. It was expected that participants with more focused attentional allocation styles
should have high scores on the GEFT.
Procedure
This experiment entailed a between-subjects design with incubation conditions
(NI: No Incubation; IL: incubation with long preparation length; and IS: incubation
with short preparation length) varying randomly between participants. Participants
were tested individually and were randomly assigned to one of the three experimental
conditions. In all conditions, participants were informed that they had first to solve a
list of visual insight problems, MRTs and ATs, which were presented on a computer
using bespoke software, and then proceed to finish the GEFT.
149
Before the start of the experiment, participants were first given task instructions
and were exposed to one practice visual non-insight matchstick problem for 180
seconds or until the problem had been solved, and one practice MRT(8 seconds) and
one practice AT(10 seconds).
The main experiment consisted of 8 trials. In each trial, regardless of condition,
participants were exposed to the same number of tasks: 1 visual non-insightful
matchstick problem for 180 seconds in total or until the problem had been solved, and
5 MRTs (8 seconds each) and 8 ATs (10 seconds each). The presentation order and
the relative timing of these tasks differed between conditions.
In the NI condition, an insight problem was presented for 180 seconds or until it
was solved, then participants performed 5 MRTs and 8 ATs. After that, they proceeded
to the next trial involving another insight problem, followed by MRTs and ATs. In the
IS condition, the insight problem was first presented for 90 seconds or until it was
solved, participants were then prompted to solve the 5 MRTs and 8 ATs (incubation
period). Following the interruption, participants proceeded to the next trial if they had
solved the insight problem before the interruption. If not, the insight problem was
presented for another 90 seconds (or until a solution was reached). Participants in the
IL condition were interrupted to perform the MRTs and ATs while solving the insight
problem, but the maximum presentation time of the insight problem at the pre- and
post- incubation stage was 120 seconds and 60 seconds, respectively. Figure 5.1
illustrates the sequence of the various tasks in each condition. Eight trials were
included in each condition, and the presentation order was varied over participants,
and insight problems in the same set were not presented in consecutive trials.
During the presentation of the insight problems, participants could respond at
any time by first pressing the C key to pause the timer, then writing down the
150
response on the sheet provided and showing it to the experimenter for immediate
judgment. If their answer was correct, they were prompted to do the MRTs and ATs;
otherwise, they were asked to continue to solve the problem until the time was up or
until a solution was reached. The presentation of the MRTs and ATs was the same as
in Experiment II. The number of visual insight problems that were solved and the
responses time were recorded. In the analyses, responses time was dropped as a
performance indicator because a large variance in participants’ responding styles was
observed, which may affect the reliability of this measure. Conservative participants
pressed C and presented their answers to the researcher only if they were certain
about the accuracy of their answers, while some participants would press C and
present their answer even they were not sure whether the answers were correct or not.
After finishing the computerized tasks, participants completed a paper-based
GEFT. The test was administered following the standard procedures outlined in the
GEFT manual.
IS
90s on insight
problem
IL
120s on insight problem
NI
180s on insight problem
90s on insight
problem
120s on MRTs + ATs
120s on MRTs + ATs
60s on insight
problem
120s on MRTs + ATs
Figure 5.13. Task Presentation Order in Each Trial in Experiment IV.
151
Results
To assess the incubation effect, the number of insight problems solved in the
post-incubation period in the IS and the IL conditions was compared with
performance in the NI condition in the equivalent time period5. If incubation periods
facilitate problem solving processes, then participants in the IS and the IL conditions
should solve more insight problems in the post-incubation period as compared with
those in the NI condition in the equivalent time period. Furthermore, such
performance differences between the IL and the NI conditions should be larger than
the difference between the IS and the NI conditions if the preparation length is a
positive moderator of incubation.
Apart from this, a positive correlation between the post-incubation improvement
and individuals’ GEFT scores is expected if individuals who tend to allocate attention
narrowly show more likely to benefit from an incubation period. Also, the strength of
this correlation in the IS condition should be stronger than in the IL condition if the
proposed link between preparation length and attentional allocation styles on the
incubation effect is correct.
Background Statistics
Participants’ gender and ethnic distribution, and the mean and the standard
deviation of their GEFT scores (a higher score implies a more focused attentional
allocation style) in each condition are presented in Table 5.1.
5
Performance during the 90th- 180th second in the NI condition was compared with the post-incubation
performance in the IS condition. Performance during the 120 th- 180th second in the NI condition was
compared with the post-incubation performance in the IL condition.
152
Table 5.138
Ethnic and Gender Distribution, and Means and Standard Deviations of the GEFT
Scores by Condition.
HK
Gender
UK
GEFT
Gender
GEFT
scores
scores
Male
Female
M (SD)
Male
Female
M (SD)
NI
7
7
16.00 (2.15)
9
9
12.28 (4.11)
IS
10
7
15.82 (2.35)
10
10
13.75 (5.17)
IL
9
9
15.72 (2.93)
10
10
14.55 (4.08)
Condition
A 2 (Ethnicity) x 3 (Condition) ANOVA using GEFT scores as the dependent
variable was carried out to check if there was a HK-UK difference on GEFT scores
within each condition. There was a significant Ethnicity effect on GEFT scores, F(1,
101) = 10.21, p = .002, η2 = .092, indicating that HK participants had significantly
higher GEFT scores than the UK participants. This implies that HK participants are
more field-independent. This is opposite to the findings reported by Nisbett(2003). An
individual’s GEFT performance has been reported to be positively correlated with
his/her academic achievement(e.g., Copeland, 1983; Luk, 2002; Murphy, Casey, Day,
&Young, 1997), and therefore, the opposite difference may be due to the possibility
that the HK participants outperformed the UK participants in academic matters. This
sampling bias, however, does not invalidate this study. The core reason for recruiting
participants from HK and UK was to maximize the variance in attentional allocation
styles among the participants. The significant Ethnicity effect confirms that there was
153
a wide variance in attentional allocation styles (focused vs. diffuse) between
participants, albeit in the opposite direction to that expected initially. This pattern of
differences was found in all the three conditions, as neither the Condition effect (p
= .54) nor the Ethnicity x Condition (p = .37) effect was significant.
Table 5.2 presents the solution rate of each visual problem. In order to
calculate the solution rate in an un-interrupted problem solving setting, only data from
participants in the No-Incubation condition (N = 32) were used. Insight problems in
Set 4 were excluded in all the following analyses because of the extreme within-set
and between-set differences in solution rate.
154
Table 5.239
Solution Rate for the Visual Insight Problems (N = 32)
Solution Rate
Set 1:
Insight Problem 1
88%
Insight Problem 2
72%
Set 2:
Insight Problem 3
59%
Insight Problem 4
31%
Set 3:
Insight Problem 5
78%
Insight Problem 6
78%
Set 4:
Insight Problem 7
90%
Insight Problem 8
12%
155
Table 5.3 presents the mean and the standard deviations for the number of insight
problem solved during the pre- and post-incubation period in the IS and the IL
conditions, and during the equivalent time period in the NI condition. GEFT scores
were positively correlated with the number of insight problems solved at the
pre-incubation stage. This is consistent with the past findings (e.g., Frank & Noble,
1985; Witkin et al., 1971) that higher GEFT scores predicts better insight problem
solving performance.
Table 5.340
Means and Standard deviations of the Number of Insight Problem Solved in the Pre
and Post-Incubation Period in the IS and IL Conditions, and in the Equivalent Time
Period in the NI Condition.
Number of Insight Problem Solved
IS
NI
IL
NI
M (SD)
M (SD)
M (SD)
M (SD)
Pre-Incubation
2.95 (1.45)
3.00 (1.63)
3.00 (1.26)
3.50 (1.24)
Post-Incubation
0.70 (0.78)
1.06 (1.01)
1.00 (0.89)
0.56 (0.56)
0.29+
0.47**
0.39*
0.46**
Time Period
Correlation
between no. of
problems solved
in pre-incubation
period and GEFT
scores
+
p < .06. *p < .05. **p < .001.
156
In the main analyses, regressions were carried out to examine the incubation
effect in the IS and the IL conditions separately. To avoid over-fitting the regression
models with irrelevant variables, two backward stepwise multiple regressions were
first carried out to examine if any of the proposed predictor variables was redundant:
one regression for the IS vs. NI comparison and one for the IL vs. NI comparison,.
In both backward regression analyses, the outcome variable was the number of
insight problems solved during the post-incubation period or the equivalent time
period in the NI condition. The predictor variables included Initial Performance (the
number of problem solved during the pre-incubation period or the equivalent time
period in the NI condition), Gender (female vs. male), Ethnicity (UK vs. HK),
Condition (No Incubation vs. Incubation), GEFT (GEFT scores, ranging from 0-18).
The categorical variables Gender, Ethnicity, and Condition were dummy-coded. The
categories (with the associated predictor variables following in parentheses) of female
(Gender), HK (Ethnicity), and No Incubation (Condition) were used as reference
groups in the analyses, and their coefficients were restricted to zero.
Table 5.4 presents a summary of the results of the complete and the backward
eliminated regression models. In both the IS vs. NI and the IL vs. NI regression
analysis, Ethnicity was found to be a non-significant predictor variable and was
removed in the backward eliminated model. Therefore, Ethnicity was not included in
the main regression analyses. The tolerance statistics of each of the remaining
predictors are all well above 0.2, suggesting that there was no collinearity within the
data.
157
Table 5.441
Summary of the Results of the Complete and the Backward Eliminated Model
IS vs. NI (N = 69)
B
SE B
β
Tolerance
IL vs. NI (N = 72)
F
Adjusted
B
SE B
β
R2
statistics
Tolerance
F
Adjusted
R2
statistics
Complete
4.90*
.223
.219+
.997
.004
.870
.084
-.239+
.737
.020
.027
.098
.733
.334
.189
.213+
.884
.342
.176
.219+
.997
-.153
.082
-.240+
.752
Gender
.506
.199
.281*
.932
.342
Ethnicity
-.232
.220
-.129
.775
.007
Initial Performance -.318
.075
-.129*
.727
-.152
GEFT
.047
.027
.210*
.802
Condition
-.399
.194
-.221+
.986
.478
.198
.265*
.948
.070
-.491*
.818
.177
.190
2.63*
.104
3.33*
.118
Backward Eliminated
Gender
Initial Performance -.292
5.83*
.221
158
+
GEFT
.054
.026
.239*
.848
.020
.026
.097
.781
Condition
-.399
.194
-.221*
.986
.334
.187+
.213+
.884
p < .10. *p < .05.
159
Moderating Effect of Preparation Length and Attentional Allocation Style
Two hierarchical regressions were carried out, one for the IS vs. NI comparison and one
for the IL vs. NI comparison, to examine the impact on post-incubation performance of the
preparation length and attentional allocation styles, as well as the interaction between them.
In both regression analyses, the outcome variable was the number of insight problems solved
during the post-incubation period or the equivalent time period in the NI condition. The
predictor variables included Initial Performance, Gender, Condition, GEFT, and the
interaction variable GEFT x Condition, which was the multiplicative term of the variable
GEFT and Condition.
The predictor variables were separated into four blocks. The variables in the first block
were the control variables, which included Gender, Ethnicity, and Initial Performance. To
examine the impact of the incubation period on insight problem solving performance,
Condition was added in the second block. GEFT and GEFT x Condition were placed in the
third and the forth block of each model respectively to examine if participants’ attentional
allocation style could moderate the incubation effect and if the moderating effect was more
significant in the IS condition.
A positive coefficient of Condition was expected if the incubation period enhances
performance. The coefficient in the IL vs. NI model should be larger than in the IS vs. NI
model if the preparation length is the positive moderator of the incubation effect. A positive
coefficient of GEFT is also expected because of the reported positive correlation between
GEFT scores and insight problem solving performance (e.g., Frank & Noble, 1985; Witkin et
al., 1971). The magnitude of the coefficient of GEFT x Condition could indicate if there is
any link between attentional allocation style and incubation. If the hypothesis that individuals
with a focused attentional allocation style benefit more from the incubation period is correct,
then a significant and positive coefficient of this interaction variable is expected. Also,
160
according to the predicted interaction between the preparation length and attentional
allocation styles on the incubation effect, the coefficient of the variable GEFT x Condition in
the IS vs. NI model should be larger than the one in the IL vs. NI model. Table 5.5 presents
the summary of the results of these two hierarchical regression models as each additional
block of variable was entered.
161
Table 5.542
Summary of Results of the Two Hierarchical Multiple Regression Analyses
IS vs. NI (N = 69)
Block
1
Variable
B
SE B

F
IL vs. NI (N = 72)
Adjusted
Sig.
R2
F-change
.155
.001
B
SE B

.347
.179
.222+
-.159
.073
-.250*
.348
.175
.223*
-.399*
-.124
.073
-.195+
.198
-.179+
.370
.181
.236*
.198
.265*
.342
.176
.219+
Gender
.458
.205
.255*
Initial
-.236
.068
-.397*
Gender
.445
.202
.247*
Initial
-.237
.067
Condition
-.355
Gender
.478
7.24*
F
Adjusted
Sig.
R2
F-change
4.11*
.082
.021
4.27*
.123
.044
3.33*
.118
.448
Performance
2
6.06*
.182
.078
Performance
3
5.83**
.221
.044
162
-.292
.070
-.491*
-.153
.082
-.240+
Condition
-.399
.194
-.221*
.334
.187
.213+
GEFT
.054
.026
.239*
.020
.026
.097
Gender
.460
.201
.256*
.375
.178
.240*
Initial
-.285
.072
-.480**
-.155
.082
-.243+
Condition
-.817
.742
-.453
1.082
.707
.690
GEFT
.036
.041
.159
.048
.037
.234
GEFT x
.029
.050
.259
-.052
.048
-.529
Initial
Performance
4
4.68*
.213
.576
Performance
Condition
+
p < .08. *p < .05. **p < .001.
163
2.91*
.120
.277
In the IS vs. NI model, the negative coefficients of Condition, 2-Block:  = -.179, p
= .07, 3-Block:  = -.221, p = .04, indicated that having an incubation period preceded by a
short preparation period impaired performance. Yet, a positive incubation effect was found
with a long preparation period, as reflected by the positive coefficient of Condition in the IL
vs. NI model, 2-Block:  = .236, p = .04. This supports the moderating effect of the length
of preparation period on incubation.
The predicted interaction between GEFT and Condition is not found, IS vs. NI:  =
-.259, p = .56, IL vs. NI:  = .-53, p = .28, indicating that participants with different GEFT
score did not benefit differentially from an incubation period. This does not support the
hypothesis that individuals with more focused attentional allocation style benefit more from
the incubation period.
However, there was an unexpected interaction between preparation length and GEFT
on the outcome variable. In the IS vs. NI model, the coefficient of GEFT was .24, p = 0.04,
indicating that participants with focused attentional allocation styles solved more insight
problems during the post-incubation period or during the equivalent time period in the NI
condition. The inclusion of GEFT in the regression model also amplified the negative
incubation effect (without GEFT:  = -.197, p = .08; with GEFT:  = -.221, p = .04). This
indicated that controlling for individual differences in attention allocation styles improved
the stability of the negative incubation effect. In the IL vs. NI model, the impact of GEFT
scored was attenuated,  = .097, p = .45, and adding the variable GEFT to the model
decreased the explanatory power of it, ∆ Adjusted R2 = -.005. This discrepancy suggests that
the impact of attentional allocation style on incubation varies depending on the preparation
length.
164
The number of problems solved at the pre-incubation period was a strong negative
predictor in both the IS vs. NI and the IL vs. NI models. This is not surprising because the
more problems participants solved at the beginning, the fewer problems remained unsolved
during the post-incubation period or the equivalent time period in the NI condition, and this
certainly limited the maximum number of problems individuals could solve during that time
period. There was also a robust Gender effect, such that male participants outperformed
female participants in the post-incubation problem solving stage. The effect was significant
even after controlling for the pre-incubation performance and individuals’ attentional
allocation styles.
Discussion
This experiment demonstrated a positive incubation effect on insight problem-solving and a
significant moderating effect of preparation length on incubation, and more importantly,
revealed that there is an interaction between individuals’ attentional allocation styles and
preparation length on the incubation effect.
The difference in the regression coefficient of Condition between the IS vs. NI and
the IL vs. NI model suggests that individuals benefit from an incubation period only if it is
preceded by a long preparation period. This is consistent with the meta-analysis reported in
Chapter 3, and supports the suggestion that a long preparation period allows individuals to
explore the entire initial faulty problem space and reach an impasse, which would maximize
the likelihood of the occurrence of the restructuring processes during the incubation period
(Seifert et al., 1995). The negative impact of Condition in the IS model suggests that
preparation length not only can amplify, but also reverse the incubation effect. In the IS
165
condition, the preparation period was too short for participants to exhaust the initial problem
space and reach an impasse. An “early” incubation period may interrupt the initial conscious
search, delays the onset of the impasse, and inhibits the occurrence of restructuring.
Therefore, continued effort will indeed be more effective because individuals can have more
time to search through the problem space and reach an impasse, and move on to the
post-impasse restructuring stage.
One may argue that the positive coefficient of Condition in the IL vs. NI model does
not necessarily imply a positive role of incubation period in visual insight problem-solving.
It could be that participants in the IL condition had inferior pre-incubation performance, as
compared with those in the NI condition. As a consequence, participants in the IL condition
had a better chance to show improvement in the post-incubation problem solving period. If
an initial performance difference existed between participants in these two conditions, then
a significant association between the predictor variables Condition and Initial Performance
in the IL vs. NI model would be expected. A multiple regression was carried out using the
Initial Performance as the outcome variable, and Gender and Condition as the predictor
variables, was carried out. The regression model did not reach significance, F(1, 69) = 1.56,
p = .22, and none of the predictor variables had a significant impact on the outcome variable,
Gender:  = .067, p = .58, Condition:  = -.197, p = .10. Therefore, this suggestion can be
rejected, and the findings of this experiment support the conclusion that preparation length
is a positive moderator of the incubation effect with visual insight problems.
Another key finding is the differential impact of GEFT in the IS vs. NI and the IL vs.
NI models. The IS vs. NI comparison revealed that participants with focused attentional
allocation style (indicated by the GEFT scores) solved more insight problems during the
post-incubation period or during the equivalent time period in the NI condition. One may
166
suggest that this finding simply reflects the positive correlation between GEFT scores and
insight problem solving performance, which has been reported in past studies (e.g., Frank &
Noble, 1985; Witkin et al., 1971). However, this explanation fails to account for the
non-significant contribution of GEFT in the IL vs. NI model. Therefore, a more
sophisticated account is required, and it is suggested that it could again be related to the
importance of reaching an impasse during insight problem-solving.
As mentioned previously, the preparation period in the IS condition may be too short
for participants to exhaust the initial problem space to reach an impasse. When participants
in the IS condition re-approached the problem in the post-incubation period, they resumed
the interrupted conscious search. Participants with a more focused attentional state should
be more effective in navigating the problem space, implying that they would be more likely
to reach an impasse during the post-incubation period and restructure the problem
representation. Therefore, they would still outperform those participants who have a
diffused attentional allocation style. The positive impact of having a focused attentional
state would then overcome the general negative effect of the “early” incubation period.
Accordingly, controlling individual differences in this cognitive style should make the
negative effect of incubation period more significant. In line with this prediction, the
regression coefficient of “Incubation” was changed from non-significant to significantly
negative after adding the variable “GEFT” into the IS model.
This reach-impasse explanation also account for the non-significant contribution of
GEFT in the IL vs. NI model. Participants in the IL condition had a longer preparation
period to explore the initial problem space. This increased the chance of encountering an
impasse at pre-incubation stage, especially for those with a diffused attentional allocation
style. Therefore, participants with different attentional allocation style would still have the
167
same opportunity to benefit from the incubation period. It means that individuals’ attentional
allocation style should no longer be a strong predictor on the post-incubation performance
in the IL condition. This explains the non-significant impact of “GEFT” in the IL model.
The combined findings of the IS vs. NI and the IL vs. NI model support the importance of
reaching an impasse in order to benefit from an incubation period.
Inconsistent with the hypothesis, participants with different attentional allocation styles
benefited from (IL condition) or were impaired by (IS condition) the incubation period in
the same degree, as indicated by the non-significant coefficient of the interaction variable
GEFT x Condition in both models. It may be the case that the link between attentional
allocation style and the magnitude of the incubation effect is significant only when external
cues are presented during the incubation period. Individuals who are narrowly focused are
less capable of noticing the external cues, as compared with those allocating their attention
broadly (Ansburg & Hill, 2003). Performing incubation tasks should help these individual
become less focused, and increase the chance of cue-assimilation.
There is also an unexpected finding that male participants demonstrated a larger
improvement than female participants in the post-incubation insight problem-solving stage.
This effect remained significant even after controlling for the pre-incubation performance
and GEFT scores. This observed association cannot be simply attributed to the male-female
differences in general visual insight problem-solving ability. One possible explanation is
that male participants benefited more than the female participants from the incubation
period. To examine this possibility, the interaction variable Gender x Condition was added
into the 4-block regression models presented in Table 5.5. However, this additional variable
was not a significant predictor in the IS vs. NI,  = -.185, p = .33, and the IL vs. NI model, 
= -.204, p = .32, suggesting that male and female participants benefitted from the incubation
168
period to a similar degree. An alternative account of this gender effect may be related to
their differential tendency to switch spatial/visual search strategies at the post-impasse stage.
Male participants might be more likely than female participants to switch to other
spatial/visual search strategies after reaching an impasse, and therefore, they are more likely
to overcome the impasse and solve the problems in the post-impasse period. This
explanation is not without ground. There are numerous studies revealing that males tend to
outperform female on a diverse set of spatial tests (see review in Linn & Petersen, 1985),
and flexible spatial switching may be one of the mechanisms of their superior performance
on this aspect.
The findings of this experiment not only further the current understanding of the
nature of incubation, but also offer us a new perspective on the nature of processes
underlying restructuring. The current debate on the nature of post-impasse restructuring is
dominated by two opposite accounts: conscious- and unconscious-work. Researchers in the
conscious-work stream consider insight problem solving as a conscious and controlled
strategic search of the problem space, with impasse being due to the use of wrong search
heuristics and with the resolution of impasse dependent on modifying the initial search
heuristic (Kaplan & Simon, 1990; MacGregor, Ormerod, & Chronicle, 2001). Alternatively,
the unconscious-work account proposes that impasse is caused by the inappropriate use of
task-specific information and knowledge, and it can be overcame by inhibiting the
activation of misleading items and spreading-activation to relevant items in long-term
memory.
The findings of this study offer support to both of these two polarized accounts. The
positive incubation effect in the IL condition suggests that the post-impasse restructuring
processes should function at a sub-conscious level. Otherwise, the restructuring processes
169
and incubation tasks would compete for the same limited cognitive resource, and
participants’ insight problem solving performance should be impaired. Researchers
supporting the conscious-work account on incubation may argue that the incubation period
provides extra time for participants to work on the problem covertly. However, this
explanation is contradicted by the finding of a negative incubation effect in the IS condition.
The findings of this experiment support the suggestion that incubation tasks occupy part of
the problem solver’s attention, and the resulting diffused attentional state facilitates the
occurrence of unconscious post-impasse restructuring processes (Ansburg & Hill, 2003;
Finke, Ward, & Smith, 1992; Martindale, 1995).
The processes underlying restructuring, however, may not necessarily be the
memory-activation processes proposed by the unconscious-work account. The insight
problems adapted in this experiment were knowledge-lean problems (except the matchstick
arithmetic problems), in that their solution does not call upon access (or suppression) of the
task-specific knowledge. The restructuring processes of these insight tasks are more likely
to be at the meta-cognitive level (e.g., strategic switching). This account is consistent with
provides the strategy-based restructuring proposed by the conscious-work account.
In sum, Experiment IV supports the positive role of incubation, and suggests that
preparation length and individuals’ attentional allocation styles determine how likely it is
that individuals encounter an impasse before the incubation period, which in turn affect
whether they benefit from an incubation period. The significant moderating effect of
individuals’ attentional allocation styles also implies that, even under identical experimental
settings, it is not guaranteed that we can always replicate the same pattern of incubation
effect. In order to examine the true effect of incubation periods, future studies should
control for the individual differences in this cognitive style. Our results also provide
170
empirical evidence for unconscious restructuring, and suggest that restructuring may rely on
both cognitive and meta-cognitive processes. An integration of unconscious- and
conscious-work accounts is needed in order to build a complete model of insight.
171
CHAPTER VI: GENERAL DISCUSSION
A meta-analysis and four experiments were carried out in an attempt to identify the potential
procedural and cognitive moderators of incubation, and to examine the three dominant
unconscious-work hypotheses of the incubation effect: spreading activation, selective
forgetting, and cue assimilation. The findings converge to support the existence of
unconscious processes during incubation, and suggest that the magnitude of the incubation
effect varies depending on the loading of the incubation tasks, the type and the difficulty of
the problem, preparation length, and an individual’s attentional allocation style.
The meta-analysis, synthesizing results of 117 studies (extracted from 29 publications),
revealed that the incubation effect is highly susceptible to task-specific procedural
moderators. In linguistic insight problems, light-load incubation tasks yielded a stronger
incubation effect than did rest or heavy load tasks. Light-load tasks may help divert an
individual’s attention to other areas, and this facilitates helpful unconscious processes. In
visual insight problems, the length of the preparation period was the only significant
moderator. Four experiments were then carried out to examine the incubation effect on
linguistics and visual insight problems.
Three experiments were conducted on linguistic problems (Experiment I, II, & III) with
the objectives being to verify the three unconscious-work hypotheses, and to examine the
impact on incubation of the loading of the incubation tasks. These three experiments
compared the incubation effects when solving linguistic problems at different difficulty levels
(easy, intermediate, and difficult) and when having incubation period filled with tasks of
varying cognitive loadings (low, intermediate, and high) were compared. To examine the
incubation effect on problem-solving performance and sensitivity to relevant and irrelevant
172
items, the traditional paradigm that only measures performance improvement was
supplemented with lexical decision tasks to assess the change of sensitivity to relevant and
irrelevant concepts. Experiments I, II, and III found that cue assimilation occurred when
solving problems at an intermediate level of difficulty and when the incubation period was
filled with tasks at intermediate-load. This supports the cue-assimilation hypothesis and the
meta-analysis findings, and also suggests an additional moderator: problem difficulty. Also,
participants were sensitized to the relevant items and could ignore the misleading cues, even
though they could not solve the problems at the first attempt. However, the sensitivity to these
two item types remains the same after the incubation period, suggesting the absence of any
activation redistribution. Therefore, the Spreading-Activation and Selective-Forgetting
hypotheses are not supported.
Experiment IV was carried out to examine the impact of the preparation length and
attentional allocation style on incubation, and the interaction between them. Again, the
findings of the experiment were in line with the meta-analysis results, revealing that
increasing the preparation time can change the incubation effect size from negative to
positive. Also, controlling individuals’ attentional allocation styles helps reduce the variance
in incubation effect size among participants. The findings also confirm the interaction
between preparation length and attentional allocation styles of individuals on incubation. It is
suggested that individuals with focused attentional allocation styles are more likely to exhaust
the initial search space and reach an impasse, and reaching an impasse is a necessary
condition for the occurrence of restructuring, which can be further facilitated by performing
incubation tasks.
The combined findings of the meta-analysis and the experimental studies have offered
an explanation of the inconsistencies between past findings, and further our current
173
understanding of the incubation period. The impact of the procedural and cognitive
moderators on incubation suggests that the variance of past findings can be attributed to
differences in experimental parameters and the lack of control of cognitive moderators.
However, attentional allocation style is not the only cognitive moderator of the incubation
effect. It has been reported that individuals’ working memory span is an indicator of their
ability to control attention (Kane, Bleckley, Conway, & Engle, 2001), and this affects how
fast they are in exhausting the initial problem space and reaching an impasse. Therefore, an
individual’s working memory should also affect they can benefit from an incubation period.
Another potential moderator is the characteristics of an individual’s knowledge network. As
mentioned earlier, the structure of an individual’s knowledge network (e.g., the richness of
the connectivity and the strength of the inter-connections), may affect the occurrence of
spreading activation during the incubation period, and in turn moderate the incubation effect.
However, none of these proposed cognitive factors have been examined in past incubation
studies, and so further research adopting an individual-differences approach should be
carried out in order to fill this gap.
The novel paradigm used in these experiments has also generated strong data to
examine the underlying mechanism of incubation. It has long been suggested that there are
gradual and unconscious cognitive processes underlying incubation. However, the
heterogeneity of settings, insensitivity of measurements, and diversity of the findings make
it very difficult to draw any conclusions on the role of incubation. The
multi-dependent-variable paradigm used in these experiments offered strong data to support
the cue-assimilation hypothesis and confirm the absence of spreading-activation and
selective-forgetting during the experimental incubation periods. However, Sio and
Rudowicz (2007), using a similar paradigm, reported evidence for the Spreading-Activation
174
hypothesis. They revealed, after a short incubation period, GO experts became more
sensitive to the answer of the RATs which contain irrelevant GO information. These mixed
findings point to the possibility that spreading-activation occurs only if the fixation effect is
very strong or/and only if individuals’ task-related knowledge is well connected (rich and
strong inter-node connections) and easily activated. Therefore, instead of completely
rejecting the Spreading-Activation hypothesis, further studies have to be carried out to
examine the conditions for the occurrence of spreading-activation. Both the present
experiments and those of Sio and Rudowicz (2007) reported the absence of selective
forgetting during the incubation period. These studies offer evidence to reject the
Selective-Forgetting hypothesis.
Although spreading activation and selective forgetting did not occur during the
incubation period, participants’ sensitivity towards the target words and misleading hints
remained unchanged after the incubation period. There was no dissipation on the activation
of the target words and any weakening of the inhibition of the misleading hints even when
participants had shifted attention to the incubation tasks. This finding is analogous to the
replication of the Zeigarnik effect (1927, 1938), in that individuals remember unsolved
problems better. The present experiments show that the representation of an unsolved
problem is maintains at a heightened level. The persistence of such activation may provide a
basis for some other unconscious problem-solving processes to occur during the incubation
period. The non-significant change of participants’ sensitivity to relevant and irrelevant
items after the incubation period in the linguistic experiments suggests that these
unconscious processes may not be semantically-based. The positive incubation effect on
visual insight problems, in which solving them requires strategic switching/releasing
unnecessary search constraints, suggests the possibility of the occurrence of meta-cognitive
175
processes during an incubation period.
These findings not only offer a new perspective on the mechanism of incubation
period, but also question the commonly held assumption in human cognitive research that
meta-cognitive processes require conscious awareness (Brown, Bransford, Ferrara, &
Campione, 1983; Flavell, 1979; Flavell & Wellman, 1977). Recently, a few studies have
supported the possibility of unconscious meta-cognition, showing that individuals can
change their problem-solving strategy to optimize performance without conscious
awareness (Newton & Roberts, 2005; Reder & Schunn, 1996; Siegler & Stern, 1998).
However, the tasks examined in these studies are relatively simple, such as arithmetic
computation (Reder & Schunn, 1996; Siegler & Stern, 1998) and spatial tasks (Netwon &
Roberts, 2005). The simplicity of the tasks used in these studies has limited the implications
of the findings. Examining the strategic aspect of insight problem-solving and the impact of
an incubation period should bring further data to the debate on unconscious meta-cognition.
The findings of this thesis also challenge the current perspective of insight
problem-solving. The current debate on the nature of insight is dominated by two accounts:
The Representational Change Theory (RCT, Knoblich, Ohlsson, Haider, & Rhenius, 2001;
Ohlsson, 1992;) and the Criterion for Satisfactory Progress theory (CSPT, Macgregor,
Ormerod, & Chronicle, 2001). RCT proposes that, unlike non-insight problem solving,
insight emerges after an individual restructures the problem representation, which is the
outcome of the occurrence of unconscious processes. CSPT, however, suggests insight
problem-solving relies on a conscious change in search strategy, and such strategic shifts
occur only if an individual reaches an impasse, and this is no different from the non-insight
problem processes. Studies have been carried out to test between these the two theories.
However, it may be a mistake to consider the RCT and CSPT as two opposite and mutually
176
exclusive accounts. The findings of the linguistic experiments suggest that unconscious
cognitive processes occur during the incubation period, and in turn, facilitate the
post-incubation problem solving performance. This supports RCT in that insight problemsolving processes are unconscious. The findings of the visual experiments, however, support
CSPT. CSPT suggests that insight problem solving relies on a conscious change in search
strategy, and such strategic shifts occur only if an individual reaches an impasse. In the
visual experiments, participants benefit from the incubation period only if they have reached
an impasse already, and introducing the incubation period too early induces a negative effect
on problem-solving performance. Instead of perceiving the RCT and CSPT as two
competing accounts, it is possible that these two theories capture different aspects of the
insight problem-solving process. An integration of these two theories is needed in order to
explain fully the insight problem-solving process.
The present findings also suggest that it is a misconception to regard insight and
non-insight problem-solving processes as involving two distinctive types of processes.
Instead, it is suggested that insight problem-solving can be divided into two stages. In the
first stage, individuals carry out a focused search of the initial problem space to find out and
execute the moves to reach a solution, and this process is no different from the non-insight
problem-solving processes. However, the settings of the insight problem usually mislead
individuals to build an inappropriate initial problem representation, which does not include
all important moves for solving the problem. Therefore, instead of reaching a solution,
individuals usually reach an impasse. Restructuring the problem representation will then
occur to overcome the impasse, and this process is facilitated by a diffuse attentional state. A
recent study of Ash and Wiley (2006) also offers some empirical support to this proposal.
They asked participants to solve a list of visual insight problems with different, initially
177
faulty search spaces. They found that increased individual ability to control attention
predicted a better performance in solving insight problems with a large search space.
However, such a correlation was not observed when solving problems with a small search
space. They suggest that when solving problem with a large search space, participants with
better control of attention are quicker in exploring the initial search space and better at
maintaining different failed solutions in memory. Therefore, it is quicker for such
participants to realize that the initial search space is incorrect, and so they started the
restructuring process earlier. This is consistent with the proposal that the pre-impasse insight
problem stage involves a controlled and focused search through the problem space. When
solving problems with a small faulty search space, most participants would reach an
impasse quickly. Ash and Wiley suggest that performance on the problems depends solely
on how successful the restructuring process is, and the absence of any link between the
ability to control attention and success in solving problems with a small initial search space
implies that the restructuring process is not necessarily a conscious search process.
In sum, this PhD thesis suggests that successful insight problem-solving requires a shift
of attentional state from focused to diffuse after reaching an impasse. However, no studies
have yet been done that measure the attentional state at pre- and post-impasse insight
problem-solving stages, and correlate them with the occurrence of insight-related cognitive
processes and the insight problem solving performance. To measure attentional state and the
cognitive mechanisms during insight problem-solving, advanced techniques such as EEG
have to be adapted. EEG data can identify neural activities associated with different
cognitive processes, such as focused search, generating constraints, and encoding and
processing semantic information (Klimesch, 1999; Sauseng, Klimesch, Schabus, &
Doppelmayr, 2005, Klimesch, Doppelmayr, Pachinger, & Ripper, 1997). Comparing the
178
EEG data at the pre-and post-impasse stage should offer strong evidence to test the
two-stage model. There are few attempts to measure the neural activity during insight
problem-solving. However, most of them only record the neuroimaging data either at
pre-problem presentation (Jung-Beeman et al., 2004; Kounios et al., 2006) or during the
pre-solution period (Kounios et al., 2006). For those assessing the actual insight
problem-solving processes, they have not taken the importance of reaching an impasse into
account. For example, Luo, Niki, and Knoblich (2006) examined brain activation levels at
different time intervals. However, the validity of this approach depends on whether
individuals reach the same problem state at similar times.
This PhD thesis, along with the study of Ash and Wiley (2006), suggest that there are
individual differences in the onset of impasse, depending on how quickly an individual
executes the controlled search and exhausts the initial problem space. Therefore, in order to
capture accurately the nature of pre-impasse and post-impasse insight problem solving
processes, one must know the onset of impasse for each individual. Examining an
individual’s eye-movement allows us to time-lock the onset of impasse without interrupting
the problem solving process (Knoblich, Ohlsson, & Raney, 2001). One direction for future
research would be combining EEG and eye-tracking, or other kinds of advanced techniques
to study insight.
Overall, this PhD thesis has furthered the current theoretical understanding of
incubation and insight. The meta-analysis and the experimental studies reveal that the
incubation effect is moderated by numerous procedural and cognitive moderators, and the
inconsistency of past studies is very likely due to the lack of control for these factors. The
positive effect of performing irrelevant incubation tasks on insight problem solving supports
the conceptualization of unconscious restructuring. Furthermore, the findings of the present
179
experiments suggest that unconscious restructuring occurs only if an individual has
exhausted the search space and reached an impasse. This implies that successful insight
problem solving is the joint effort between pre-impasse conscious search and post-impasse
unconscious processes. This offers support to both the RCT and CSPT theories, and also
suggests the necessity of integrating these two theories. In a broader sense, the positive
incubation effect with knowledge-lean problems raises a question about whether
meta-cognition requires awareness. In order to have a more complete picture of how insight
emerges and how it can be facilitated, there are few questions that we need to address,
including the nature of cognitive processes in pre- and post-impasse insight problem-solving
(conscious vs. unconscious), and the nature of the restructuring (semantic, strategic, or both).
To answer these questions, further studies have to combine different advanced techniques to
capture neural, cognitive, and behavioral data on insight problem solving.
180
REFERENCES
(References marked with an asterisk indicate studies included in the meta-analysis)
Adelson, B. (1984). When novices surpass experts: The difficulty of a task may increase
with expertise. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 10 (3), 483-495
Anderson, R. J. (1994). Representations and requirements: The value of ethnography in
system design. Human-Computer Interaction, 9 (2), 151-182.
Ansburg, P. I., & Hill, K. (2003) Creative and analytic thinkers differ in their use of
attentional resources. Personality and Individual Differences. 34, 1141-1152.
Ash, I. K., & Wiley J. (2006). The nature of restructuring in insight: An
individual-differences approach. Psychonomic Bulletin & Review, 13 (1), 66-73.
Bar-Haim, Y., Lamy, D., Pergamin, L., Bakermans-Kranenburg, M. J., & van Ijzendoorn M.
J. (2007). Threat-related attentional bias in anxious and nonanxious individuals: A
Meta-analytic study. Psychological Bulletin, 133, 1-24.
*Beck, J. (1979). The effect of variations in incubation activity and incubation time on
creative response production. Unpublished doctoral dissertation, New York
University.
Bedard, J. & Chi, M. T. H. (1992). Expertise. Current Directions in Psychological Science,
1 (4), 135-139.
*Bennett, S. (1975). Effect of incubation on the associative process of creativity.
Unpublished doctoral dissertation. University of Notre Dame.
Bennink, C. D. (1982). Individual differences in cognitive style, working memory, and
181
semantic integration. Journal of Research in Personality, 16, 267-280.
*Both, L., Needham, D., & Wood, E. (2004) Examining tasks that facilitate the experience
of incubation while problem solving. Alberta Journal of Educational Research, 50 (1),
57-67.
Bowden, E.M., & Jung-Beeman, M. (2003). Aha! Insight experience correlates with
solution activation in the right hemisphere. Psychonomic Bulletin & Review, 10,
730-737.
Bowden, E. M., & June-Beeman, M. (2003). Normative data for 144 compound remote
associate problems. Behavior Research Methods, Instruments, & Computers, 35 (5),
634-639.
Bowers, K. S., Regehr, G., Balthazard, C. G., & Parker, K. (1990). Intuition in the context
of discovery. Cognitive Psychology, 22, 72-110.
*Brockett, C. (1985). Neuropsychological and cognitive components of creativity and
incubation. Unpublished doctoral dissertation, Virginia Commonwealth University.
Brown, A. L., Bransford, J. D., Ferrara, R. A., & Campione, J. C. (1983). Learning,
remembering, and understanding. In J. H. Flavell, & E. M. Markman ( Eds.),
Handbook of child psychology (4th ed.): Cognitive development, 3, 77-166. New York:
Wiley.
*Browne, B. A., & Cruse, D. F. (1988). The incubation effect: Illusion or Illumination.
Human Performance, 1 (3), 177-185.
Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55-81.
Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of
physics problems by experts and novices. Cognitive Science, 5, 121-152.
*Christensen, B.T., & Schunn, C. D. (2005). Spontaneous access and analogical incubation
182
effects. Creativity Research Journal, 17 (2&3), 207-220.
Chronicle, E.P., MacGregor, J.N., & Ormerod, T. C. (2004). What makes an insight problem?
The roles of heuristics, goal conception and solution recoding in knowledge-lean
problems Journal of Experimental Psychology: Learning, Memory & Cognition, 30,
14-27.
Chronicle, E. P., Ormerod, T. C., & MacGregor, J. N. (2001). When insight just won't come:
The failure of visual cues in the nine-dot problem. Quarterly Journal of Experimental
Psychology, 54A, 903-919.
Chua, H. F., Boland, J. E., & Nisbett, R. E. (2005). Cultural variation in eye movements
during scene perception. Proceedings of the National Academy of Sciences, 102,
12629-12633.
Clark, H. H. (1973). The language-as-fixed-effect fallacy: A Critique of Language statistics
in psychology research. Journal of Verbal Learning and Verbal Behavior, 12,
335-359.
Clark, H. T., & Roof, K. D. (1988). Field dependence and strategy use. Perceptual and
Motor Skills, 66, 303-307.
Cochran, K. F., & Davis, J. K. (1987). Individual differences in inference processes. Journal
of Research in Personality, 21, 197-210.
Cooper, H., & Hedges, L. V. (1994). The Handbook of Research Synthesis. New York:
Russell Sage Foundation.
Copeland, B.D., 1983. The relationship of cognitive style to academic achievement of
university art appreciation students. College Student Journal, 17 ( 2), 157–162.
Dodds, R. A., Smith, S. M., & Ward, T. B. (2002). The use of environmental clues during
incubation. Creativity Research Journal, 14 (3/4), 287-305
183
*Dodds, R. A., Smith, S. M., & Ward, T. B. (2002). The use of environmental clues during
incubation. Creativity Research Journal, 14 (3/4), 287-305
Dodds, R. A., Ward, T. B., & Smith, S. M. (in press). Incubation in problem solving and
creativity. In M. A. Runco (Ed.), Creativity Research Handbook, Vol. 3. Cresskill, NJ:
Hampton Press.
*Dominowski, R. L., & Jenrick, R. (1972). Effect of hints and interpolated activity on
solution of an insight problem. Psychonomic Science, 26 (6), 335-338.
*Dorfman, J, Shames, V. A., & Kihlstrom, J. F. (1996). Intuition, incubation, and insight:
implicit cognition in problem solving. In G. Underwood (Ed.), Implicit Cognition (pp.
257–287). Oxford: Oxford University Press.
*Dreistadt, R. (1969). The use of analogies and incubation in obtaining insights in creative
problem solving. The Journal of Psychology, 71, 159-175.
Durlak, J.A., & Lipsey, M.W. A practitioner's guide to meta-analysis. American Journal of
Community Psychology, 19, 291-332.
Ericsson, K. A., & Lehmann, A. C. (1996). Expert and exceptional performance: Evidence
of maximal adaptation to task. Annual Review of Psychology, 47, 273-305.
Finke, R.A., Ward, T.B., & Smith, S.M. (1992). Creative Cognition: Theory, Research, and
Applications, Cambridge, MA: MIT Press.
Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of
cognitive-development inquiry. American Psychologist, 34, 906-911.
Flavell, J. H., & Wellman, H. M. (1977). Metamemory. In R. V. Kail, Jr. & J. W. Hagen
(Eds.), Perspectives on the development of memory and cognition, 3-33. Hillsdale, NJ:
Lawrence Erlbaum Associates, Inc.
Frank, B. M., & Noble, J. P. (1985). Field independence-dependence and cognitive
184
restricting. Journal of Personality and Social Psychology, 47 (5), 1129-1135.
*Frankel, A. (1990). The effect of incubation and eidetic imagery training on tasks of
problem solving. Unpublished doctoral dissertation, Fordham University.
*Fulgosi, A., & Guilford, J. P. (1968). Short-term incubation in divergent production.
American Journal of Psychology, 81, 241-246.
*Fulgosi, A., & Guilford, J. (1972). A further investigation of short-term incubation. Acta
Instituti Psychologici, 64-73, 67-70.
*Gall, M., & Mendelsohn, G.A. (1967). Effects of facilitating techniques and subject
experimenter interaction on creative problem solving. Journal of Personality and
Social Psychology, 5, 211-216.
Geary, D. C., Hoard, M. K., Byrd-Craven, J., & DeSoto, M. C. (2004) Strategy choices in
simple and complex addition: Contributions of working memory and counting
knowledge for children with mathematical disability. Journal of Experimental Child
Psychology, 88, 121–151.
Ghiselin, B. (1985). The Creative Process. A Symposium. Berkeley: University of California
Press.
Gilhooly, K. J., & Murphy, P. (2005). Differentiating insight from non-insight problem.
Thinking and Reasoning, 11, 279-302
*Goldman, W., Wolters, N., & Winograd, E. (1992). A demonstration of incubation in
anagram problem solving. Bulletin of the Psychonomic Society, 30, 36-38.
Goode, P. E., Goddard, P. H., & Pascual-Leone, J. (2002). Event-related potentials index
cognitive style differences during a serial-order recall task. International Journal of
Psychophysiology, 43, 123-140.
Gorski, N.A., & Laird, J.E. (2006) Experiments in Transfer Across Multiple Learning
185
Mechanisms. Proceedings of the ICML-06 Workshop on Structural Knowledge
Transfer for Machine Learning. Pittsburgh, PA.
*Hansberry, M. T. (1999). Fixation and incubation effects in problem solving. Unpulished
doctoral dissertation. University of Hampshire
Heyworth, R. M. (1989). Expert-novice differences in the solving of a basic problem in
chemistry. Chinese University Education. Journal, 17 (1), 59-71
Hedges L. V., & Olkin, I. (1985). Statistical Methods for Meta-Analysis. Orlando: Academic
Press.
*Henley, R. J. (1999). Priming Incubated Problems. Unpublished doctoral dissertation.
University of Sussex.
*Jamieson, B. A. (1999). Incubation and aging: The nature of processing underlying insight.
Unpublished doctoral dissertation University of Georgia.
Ji, L., Peng, K., & Nisbett, R. E. (2000). Culture, control, and perception of relationships in
the environment. Journal of Personality and Social Psychology, 78, 943-955.
Jones, G. (2003). Testing two cognitive theories of insight. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 29 (5), 1017-1027.
Jung-Beeman, M., & Bowden, E. (2000). The right hemisphere maintains solution-related
activation for yet-to-be solved insight problems . Memory & Cognition, 28,
1231-1241.
Kane, M. J., Bleckley, M. K., Conway, A. R. A., & Engle, R.W. (2001). A
controlled-attention view of working-memory capacity. Journal of Experimental
Psychology: General,130 (2), 169-183.
*Kaplan, C. (1990). Hatching a theory of incubation: Does putting a problem aside really
help? If so, why? Unpublished doctoral dissertation, Carnegie-Mellon University.
186
Kavakli, M. & Gero, J. S. (2002): Structure of Concurrent Cognitive actions: A Case Study
on Novice & Expert Designers. Design Studies, 23 (1), 25-40
Kershaw, T.C., & Ohlsson, S. (2004). Multiple causes of difficulty in insight: The case of
the nine-dot problem. Journal of Experimental Psychology: Learning, Memory, &
Cognition, 30 (1), 3-13.
Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive and memory
performance: a review and analysis. Brain Research Review, 29, 169-195.
Klimesch, W., Doppelmayr, M., Pachinger, T., & Ripper, B. (1997). Brain oscillations and
human memory performance: EEG correlates in the upper alpha and theta bands.
Neuroscience Letters, 238, 9-12
Knoblich, G., Ohlsson, S., Haider, H., & Rhenius, D. (1999). Constraint relaxation and
chunk decomposition in insight problem solving. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 25, 1534–1555.
Knoblich, G., Ohlsson, S., & Raney, G. E. (2001). An eye movement study of insight
problem solving. Memory & Cognition, 29, 1000-1009.
Kounios, J., Frymiare, J.L., Bowden, E.M., Fleck, J.I., Subramaniam, K., Parrish, T.B., &
Jung-Beeman, M. (2006). The prepared mind: Neural activity prior to problem
presentation predicts solution by sudden insight. Psychological Science, 17, 882-890.
Larkin, J. H. (1983). The role of problem representation in physics. In A.L. Stevens, & D.
Gentner (Eds.), Mental models (pp. 75-99). Hillsdale, NJ: Lawrence Erlbaum
Associate.
Lesgold, A., Rubinson, H., Feltovich, P., Glaser, R., Klopfer, D., & Wang, Y. (1988).
Expertise in a complex skill: diagnosing X-ray pictures. In M. T. H. Chi, R. Glaser, &
M. Farr (Eds.), The nature of expertise. Hillsdale, NJ: Erlbam.
187
Lipsey, M. W., & Wilson, D. B. (2001). Practical Meta-analysis. Thousand Oaks: Sage.
Luk, S. C. (2002). The relationship between cognitive style and academic achievement.
British Journal of Educational Technology, 29 (2), 137-147.
Luo, J., NiKi, K. & Knoblich, G. (2006). Perceptual contributions to problem solving:
Chunk decomposition of Chinese characters. Brain Research Bulletin, 70 (4-6),
430-443.
Lynch, M. D., & Swink, E. (1967). Some effects of priming, incubation and creative
aptitude on journalism performance. Journal of Communication, 17 (4), 372-382.
Macaskill, P., Walter, S. D., & Irwing, L. (2001). A comparison of methods to detect
publication bias in meta-analysis. Statistics in Medicine, 20, 641-654.
MacGregor, J. N., Ormerod, T. C., & Chronicle, E. P. (2001). Information-processing and
insight: A process model of performance on the nine-dot and related problems.
Journal of Experimental Psychology: Learning, Memory and Cognition, 27, 176-201.
Martindale, C. 1995. Creativity and connectionism. In S.M. Smith, T. B. Ward, & R. A.
Finke (Eds.) The Creative Cognition Approach (pp. 249-268). Cambridge, MA: MIT
Press.
Masuda, T., & Nisbett, R. E. (2006). Culture and change blindness. Cognitive Science, 30,
381-399
*Medd, E., & Houtz, J, C. (2002). Educational Research Quarterly, 26 (2), 13-16.
Metcalfe, J., & Weibe, D. (1987). Intuition in insight and non-insight problem-solving.
Memory and Cognition, 15, 238-246.
*Mednick, M., Mednick, S., & Mednick, E. (1964). Incubation of creative performance and
specific associative priming. Journal of Abnormal and Social Psychology, 69, 84-88.
Mednick, S. A. (1962). The associative basis of the creative process. Psychological Review,
188
69, 220-232.
Mill, G. A. (1956). The Magical number seven, plus or minus two: some limits on our
capacity for processing information. Psychological Review, 63, 81-97.
Miyake, A., Witzki, A., & Emerson, M. (2001). Field dependence-independence from a
working memory perspective: A dual-task investigation of the Hidden Figures Test.
Memory, 9, 455-457.
*Moss, S. A. (2002). The Impact of Environmental Clues in Problem Solving and
Incubation: The Moderating Effect of Ability. Creativity Research Journal, 14 (2),
207.
Murphy, H. J., Casey, B., Day, D. A., & Young, J. D. (1997). Scores on the Group
Embedded Figures Test by undergraduates in information management. Perceptual
and Motor Skills, 84, 1135-1138.
*Murray, H. G., & Denny, J. P. (1969). Interaction of ability level and interpolated activity
(opportunity for incubation) in human problem solving. Psychological Reports, 24 (1),
271-276.
Newell, A. & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice
Hall
Newton, E.J., & Roberts, M.J. (2005). The window of opportunity: A model for strategy
discovery. In M.J. Roberts and E.J. Newton (Eds.), Methods of Thought: Individual
Differences in Reasoning Strategies (pp. 129-158). Hove, United Kingdom:
Psychology Press.
Nisbett, R. E. (2003). The geography of thought: How Asians and Westerners think
differently and why. New York: The Free Press.
Ohlsson, S. (1992). Information-processing explanations of insight and related phenomena.
189
In M. T. Keane & K. J. Gihooly (Eds.), Advances in the psychology of thinking (pp.
1-44). Hemel Hempstead : Harvester Wheatsheaf
Olton, R. M. (1979). Experimental studies of incubation: Searching for the elusive. Journal
of Creative Behavior, 13, 9-22.
*Olton, R. M., & Johnson D. M. (1976). Mechanism of incubation in creative problem
solving. American Journal of Psychology, 89 (4), 617-630.
Ormerod, T. C., MacGregor, J. N. & Chronicle, E. P. (2002). Dynamics and Constraints in
Insight Problem Solving. Journal of Experimental Psychology Learning, Memory, and
Cognition, 28 (4), 791-799.
Patalano, A. L., & Seifert. C. M. (1994). Memory for impasses during problem solving.
Memory and Cognition, 22 (2), 234-242.
*Patrick, A. S. (1986). The role of ability in creative "incubation". Personal and Individual
Differences, 7 (2), 169-174.
*Penney, C. G., Godsell, A. Scott, A., & Balsom, R. (2004). Problem variables that promote
incubation effects. Journal of Creative Behavior, 38 (1), 35-55.
Perkins, D. N. (1995). Insight in minds and genes. In R. J. Sternberg & J. E. Davidson
(Eds.), The nature of insight. (pp. 495-533). Cambridge, MA: The MIT Press.
*Peterson, C. (1974). Incubation effects in anagram solution. Bulletin of the Psychonomic
Society, 3, 29-30.
Posner, M.I. (1973). Cognition: An Introduction. Glenview, IL: Scott, Foresman.
Rae, C. M. (1997). The creative power of doing nothing. Writer, 110 (7), 13-15.
Reder, L. M., & Schunn, C. D. (1996). Metacognition does not imply awareness: Strategy
choice is governed by implicit learning and memory. In L. M. Reder (Ed.), Implicit
memory and metacognition (pp. 45–78). Hillsdale, NJ: Erlbaum.
190
Rickards, T. (1991). Innovation and creativity: woods, trees and pathways. R & D
Management, 21 (2), 97-108.
Rosenthal, R. (1995). Writing meta-analytic reviews. Psychological Bulletin, 118, 183-192.
Sauseng, P., Klimesch, W., Schabus, M., & Doppelmayr, M. (2005). Fronto-parietal
coherence in theta and upper alpha reflect central executive functions of working
memory. International Journal of Psychophysiology, 57, 97-103.
Schooler, J. W., Ohlsson, S., & Brooks, K. (1993). Thoughts beyond words: When language
overshadows insight. Journal of Experimental Psychology: General, 122, 166–183.
Schank, R. C. (1982). Dynamic Memory: A Theory of Reminding and Learning in
Computers and People. Cambridge: Cambridge University Press.
Schank, R. C. (1999). Dynamic Memory Revisited. Cambridge: Cambridge University Press.
Scheerer, M. (1963). Problem solving. Scientific American, 208 (4), 118-128
*Seabrook, R., & Dienes, Z. (2003). Incubation in problem solving as a context effect.
Proceedings of the 25th meeting of the Cognitive Science Society, Boston, July 31-Aug
2 , 2003. Mahwah, NJ: Lawrence Erlbaum Associates.
*Segal, E. (2004). Incubation in insight problem solving. Creativity Research Journal, 16
(1), 141-149.
Seifert, M. C., Meyer, D. E., Davidson, N., Patalano, A. L., & Yaniv, I. (1995).
Demystification of cognitive insight: Opportunistic assimilation and the
prepared-mind perspective. In R. J. Sternberg & J. E. Davidson (Eds.). The Nature of
Insight(pp. 65-124). Cambridge, MA: MIT Press.
Siegler, R. S., & Stern, E. (1998). Conscious and unconscious strategy discoveries: A
microgenetic analysis. Journal of Experimental Psychology: General, 127, 377-397.
*Silveira, J. (1971). Incubation: The effect of interruption timing and length on problem
191
solution and quality of problem processing. Unpublished doctoral dissertation,
University of Oregon.
Simon, H. A. (1966). Scientific discovery and the psychology of problem solving. In R. G.
Colodny (Ed.), Mind and Cosmos: Essays in Contemporary Science and Philosophy
(pp. 22-40). University of Pittsburgh Press.
*Sio, U. N. & Rudowicz, E. (2007). The Role of an incubation period in creative problem
solving. Creativity Research Journal, 19 (2-3), 307-318.
Smilek, D., Enns, J. T., Eastwood, J. D., & Merikle, P. A. (2006). Relax! Cognitive style
influences visual search. Visual Cognition, 14, 543-564.
Smith, S. M. (1995). Fixation, incubation, and insight in memory and creative thinking. In S.
M. Smith, T. B. Ward, & R. A. Finke (Eds.). The Creative Cognition Approach (pp.
135-146). Cambridge, MA: MIT Press.
*Smith, S. M., & Blankenship, S. E. (1989). Incubation effects. Bulletin of the Psychonomic
Society, 27 (4), 311-314.
*Smith, S. M., & Blankenship, S. E. (1991). Incubation and the persistence of fixation in
problem solving. American Journal of Psychology, 104, 61-87.
*Snyder, A., Mitchell, D. J., Ellwood, S., Yates, A., & Palllier, G. (2004). Nonconscious idea
generation. Psychological Reports, (94), 1325-1330
Thornton, A., & Lee, P. (2000). Publication bias in meta-analysis – its causes and
consequences. Journal of Clinical Epidemiology, 53 (2), 207-216.
*Torrance-Perk, J. (1997). The incubation effect: Implication for underlying mechanism.
Unpulished doctoral dissertation. University of Waterloo.
Tsakanikos, E. (2006). Associative learning and perceptual style: Are associated events
perceived analytically or as a whole? Personality and Individual Differences, 40,
192
579-586.
VanLehn, K. (1988). Towards a theory of impasse-driven learning. In H. Mandl & A.
Lesgold (Eds), Learning: Issues for Intelligence Tutor System (pp. 19-41). New York:
Springer-Verlag.
*Vul, E., & Pashler, H. (2007).Incubation is helpful only when people are misled. Memory,
and Cognition, 35 (4), 701-710.
Wallas, G. (1926). The Art of Thought. London: Jonathan Cape.
Wang, Y. & Laird, J. E. (2007). The importance of action history in decision making in
reinforcement learning. In proceedings of the 8th International Conference on
Cognitive Modeling. Ann Arbor, Michigan, USA.
Webster, A., Campbell, C., & Jane, B. (2006). Enhancing the creative process for learning
in primary technology education. International Journal of Technology & Design
Education, 16 (3), 221-235.
Wertheimer, M. (1985). A Gestalt perspective on computer simulations of cognitive
processes. Computers in Human Behavior, 1, 19–33.
Witkin, M.A., Oltman, P., Raskin, E., Karp, S. (1971). A Manual for the Embedded Figures
Test. Palo Alto, CA: Consulting Psychologists Press
Woodworth, R. S., & Schlosberg, H. (1954). Experimental Psychology. New Delhi: Oxford
& IBH.
Xu, J., Wintermute, S., Wang, Y. J., & Laird, J. (2009). Transferring Learned Search
Heuristics (Report No. CCA-TR-2009-04). Ann Arbor, MI: Center for Cognitive
Architecture, University of Michigan.
*Yaniv, I., & Meyer, D. E, (1987). Activation and meta-cognition of inaccessible stored
information: Potential bases for incubation effects in problem solving. Journal of
193
Experimental Psychology: Learning, Memory, and Cognition, 13 (2), 187-205.
Zeigarnik, B. (1927). Untersuchungen zur Handlungs- und Affektpsychologie: III. Das
Behalten erledigter und unerledigter Handlungen [Investigations on the psychology of
action and affection: III The memory of completed and uncompleted actions].
Psychologische Forschung, 9, 1-85.
Zeigarnik, B. (1938). On finished and unfinished tasks. In W. D. Ellis (Ed.), A sourcebook of
Gestalt psychology (pp.300-314). New York: Humanities press.
Zhong, C. B., Dijsterhuis, A., & Galinsky, A. D. (2008). The Merits of Unconscious
Thought in Creativity. Psychological Science, 19, 912-918.
194
Appendix A
Insight Problems Used in Incubation Studies
Task Type
Description
Solution/Sample Task
Participant has to list all the uses he/she can think of
/
Creative Problems
Brick Task
for a brick.
Consequence Task
Participant has to list out as many consequences of an
Sample Task:
event as he/she can foresee
What would be the results if everyone
suddenly lost the ability to read and
write?
Creative Writing
--
Sample Task:
Write about three concrete objects: a
195
Koosch ball, a wooden type of propeller,
and a triangular frisbee
Visual Insight Problems
Candle Problem
Participant has to support a candle on wall by using
Solution:
some matches and a box of tacks.
Use a tack to attach the box to the wall,
then drip some wax onto the box and
mount the candle on the box
Farm Problem
Participant has to divide an L-shaped farm into four
parts which have the same size and shape.
196
Solution:
Hat Rack Problem
Participant has to construct a stable hatrack by using
Solution:
two boards and a Camp.
Wedge the two boards between the
ceiling and the floor, holding them in
place with the clamp, and with the clamp
also serving as a hook.
Insightful mathematic
Participant has to compute separately the area of the
Solution
problem
square and that of the parallelogram shown below:
Restructure the given shape into partially
overlapping triangle ABG and ECD. The
sum of their area is 2 x ab/2 = ab
197
Necklace problem
Necklace Problem
Solution:
Participant is given four pieces of chain, and each chain Open all 3 links of one chain, and join the
is made up of three links, he/she has to join all the
other three chains together.
pieces by only opening and closing three links.
Radiation Problem
A patient has an inoperable tumor in the middle of the
Solution:
body, and there is a ray at a strong intensity that can
Direct multiple low-intensity rays
destroy the tumor, but the ray also harms the healthy
simultaneously toward the tumor from
tissue that it travels through. At low intensities, the ray
different directions
will spare the healthy tissue, but will not destroy the
tumor. Participant has to think out a way to use the ray
to destroy the tumor without damaging healthy tissue.
Saugstad’s “ball Problem”
Participant has to transfer steel balls from a drinking
Solution:
glass to a cylinder from a distance of 8 ft by using the
Bend the nail into a hook then attach it to
198
following objects: nail, a pair of pliers, a length of
the string. Use it to drag back the frame
string, a pulley, elastic bands, and newspaper. The glass
and remove the balls. Transfer the balls
sits on a moveable frame.
into the cylinder by using a tube
constructed of newspapers and elastic
band.
Tree Problem
Participant has to plant 10 trees in five rows with four
Solution:
trees in each row
(The trees are represented by the dots)
199
Linguistic Insight Problems
Anagram
Participant has to rearrange the scrambled letters to
Sample Task:
form a new word
The scrambled letters are “t s l t i n e”,
and one possible solution is “silent”
RAT
Three words are presented to the participant; he/she has Sample Task:
to think of a word that can form associations with each
The three stimulus words are “ blue,
of them.
cake, cottage”, and one possible solution
is the word “cheese”
Rebus
Participant has to figure out the phrase portrayed by the
Sample task:
pictogram
The pictogram
200
The answer is “first aid”
Riddle
--
Sample Task
A wine bottle is half-filled and corked.
How can you drink all of the wine
without removing the cork from the
bottle?
The answer is: The cork can be opened
by pushing it in.
Word Associates Task
Six words are presented to the participant, he/she has to Sample Task:
think out a word that can form an association with each
The six stimulus words are “school, chair,
of them.
jump, noon, heels, wire”, and one
possible solution is the word “high”
Word Fragment
Participant has to complete a word that has various
Sample task:
Completion Task
letters missing
The stimulus is “OC_ _N”, and one
201
possible answer is “OCEAN”
202
Appendix B
Settings of Incubation Studies
Misleading
cues
Cues during
Preparation Incubatio
Year
Author
Problem
embedded in
Included in
Incubation Task incubation
period
Other Factors
Incubation Effect
n period
Meta-Analysis
some
period
problems
Mednick, M. T.,
Analogy task
Mednick, S. A.,
1964
RAT
Mednick, E.
Yes, in "Analogy
Not
No
1 min
No, no control
vs. Analogy
Yes
--
task +Cues"
specified
group
task + Cues
(Experiment 1)
203
condition
Cues are either
relevant,
Mednick, M. T.,
Yes, high ability
Analogy task
Mednick, S. A.,
1964
Not
RAT
No
1 min
Mednick, E.
No, no
incorrect, or
group in "Analogy information for
vs. Analogy
Yes
irrelevant;
specified
task + Cues
task + Cues"
estimating effect
condition
size
Participants'
(Experiment 2)
problem solving
ability
Non-verbal task
Gall,M., &
Yes, estimated
vs. Free
1967 Mendelsohn, G.
RAT
No
2 min
0, 25 min
Yes
--
No
from the
associates task
A.
p-less-than value
+ Cues
Yes, estimated
from the
Number-Series
Fulgosi, A.,
Consequence
1968
0, 10, 20
No
Guilford, J. P.
Task
2 min
Task ( 10 and
min
No
Yes, in 20 min
statement of
condition
significance, and
--
20 min)
the p-less-than
value
204
In control
Farm Problem: Yes,
condition, half
Farm
in "Guess card
Guess Card vs.
of the
Problem &
1969 Dreistadt, R
+Cues" condition.
No
5 min
0, 8 min
Guess Card +
Yes
participants
Tree Planting
Yes
Tree Planting
cues
received cues
Problem
Problem: No
during the whole
incubation effect.
experiments.
Multiple choice
logical
Murry, H. G. &
1969
Participants'
traced complex No
problem solving
Yes, in low ability
No
Denny J. P.
syllogisms, and
Saugstad's
5 min
0, 5 min
ball problem
Yes
group.
sequences of
ability
numbers and
letters
No, no
Number-Series
Fulgosi, A.,
Consequence
1972
0, 30 min,
No
Guilford, J. P.
Task
2 min
Task (30 min
60 min
No
Yes, in 30-min
information for
condition
estimating effect
--
and 60 min)
size
205
Free associate
task vs.
One group of
Anagram vs.
participants
Yes, estimated
Dominowski, R.,
Hat Rack
1972 & Jenrick,R.
from the
No
5 min
0, 10 min Free associate
Yes
received cues
No
Problem
statement of
( Experiment 1)
task +Cues vs.
during the
Anagram +
whole experiment
significance
Cues
Free associate
One group of
task vs.
participants
Anagramvs.
received cues
Yes, estimated
Dominowski, R.,
Hat Rack
1972 & Jenrick,R.
from the
No
3 min
0, 3 min
Free associate
No
during the whole No
Problem
( Experiment 2)
statement of
task +Cues vs.
experiment .Parti
Anagram +
cipants' problem
Cues
solving ability
significance
206
Read book (30
Yes, in long
min) vs. Read
Silveira, J. M.
Necklace
1972
3 min vs. 13 0,
30,
No
(Experiment 1 )
Problem
preparation (13
book + Free
min
No
--
210 min
Yes
min) and incubation
Activity (210
(210 min)group
min)
Silveira, J. M.
Necklace
1972
Read book +
No
(Experiment 2)
Problem
Silveira, J. M.
Consequence
1972
1974 Peterson, C.
210 min
--
Yes
Yes
No
--
No
Yes
Task Difficulty
Yes
Yes
--
No
Read book +
2 min
210
min
Task
Anagram
No
Free Activity
No
(Experiment 3)
13 min
Free Activity
No
20 sec
3.6 mins Other Anagram No
Listen music
vs. Solve
1975 Bennett, S. M.
RAT
No
1 min
10 min
No, no control
No
mathematics
problems
207
group
Rest vs. Stroop
test +
backward
Counting vs.
Review the
Olton, R, M., &
Farm
Johnson, D. M.
Problem
1976
0, 15
No
10 min
problem vs.
Yes
--
No
Yes
min
Have lecture on
creative
problem
solving vs.
Listen Music
Relax (20 min
Verbal
Yes, the longer the
and 30 min) vs.
Divergent
1979 Beck, J.
0, 20, 30
No
12 min
Thinking
incubation the
Writing Essay
No
--
min
Yes
stronger the
(20
min and
Task
benefits
30 min)
Brick and
1985 Brockett, C. A.
No
10 min
0, 20 min Questionnaires No
RAT
208
--
Yes
Yes
Yes, estimated
Conversation
1986 Patrick, A. S.
RAT
No
2 min
0, 5 min
vs. Mental
Yes, in
No
--
rotation task
"Mental
from the
rotation task"
p-less-than value,
Condition
and the statement
of significance
No, did not
Rare-word
Yes, participants
Yaniv, I, &
1987
Definition
Not
Not
specified
Specified
No
Meyer, D. E.
Questionnaire
No
--
Task
measure the
were more sensitive post-incubation
to relevant concepts problem solving
performance
Listen music,
Graph drawing,
Browne, B. A . &
Farm
1988 Donna F. Cruse
Memorizing a
No
Problem
20 min
0, 5 min
No
portion of an
(Experiment 1)
oceanorgaphy
text
209
--
No
Yes
Listen music,
Graph drawing,
Browne, B. A . &
Farm
1988 Donna F. Cruse
20 min or 25
No
Problem
Memorize a
0, 5 min
min
No
--
No
Yes
portion of an
(Experiment 2)
oceanorgaphy
text
Yes, the
performance of 15
Rest( 5 min and
Yes,
min incubation > 5
15 min) vs.
min incubation and
Smith, S. M. &
Useful cues were
Misleading
1989 Blankenship, S.
Rebus
0, 5, 15
Music
0.5 min
or Useful
control group.
No
min
presented in the
Perception Task
E. (Experiment 1)
Yes
Larger forgetting
second attempt
cues
( 5 min and 15
effect in 15 min
min)
incubation group.
No main effect of
the filler task
210
Yes, but no
difference between
Yes,
Rest( 5 min and
the 5 min and the
15 min) vs.
15 min conditions.
Smith, S. M. &
Misleading
1989 Blankenship, S.
Rebus
0, 5, 15
Music
0.5 min
or Useful
Larger forgetting on
No
min
--
Yes
Perception Task
misleading cues
( 5 min and 15
in the 15 min
min)
group. No main
E. (Experiment 2)
cues
effect of the filler
task
211
Rebus (10 min)
vs. Music
perception
task+rebus (15
Yes,
Yes, but no
Smith, S. M. &
min) vs.
Misleading
1989 Blankenship, S.
Rebus
Yes, estimated
0, 10, 15
0.5 min
or Useful
difference among
Mathematics
No
--
min
E. (Experiment 3)
from the
different incubation
Task +Rebus
p-less-than value
cues
conditions
(15 min) vs.
Rest+ Rebus
(15min) vs.
Rebus (15 min)
Yes, estimated
Yes,
Read story ( 5
Misleading
min ) vs.
Smith, S. M. &
1989 Blankenship, S.
from the
Rebus
0.5 min
0, 5 min
Yes, in "Read
No
or Useful
Mathematic
cues
tasks ( 5 min)
E. (Experiment 4)
--
p-less-than value
story" condition
and the statement
of significance
212
Other word
Yes, in both
Dorfman, J.
Word Puzzle
(Experiment 3)
Task
1990
No
0, 5, 15
problem vs.
min
other word
0.49 min
No
--
incubation
Yes
conditions
problem + Cues
Number series
Yes, estimated
task , and cues
Dorfman J
Word Puzzle
1990
0, 3, 8, 13
No
(Experiment 4)
0.49 min
Task
from the
presented in
No, not significant
delay
statement of
sub-sequence
significance
phase
Kaplan, C. A.
Consequence
1990
Psychometric
No
(Experiment 1)
Task
Kaplan, C. A.
Consequence
1990
Task
Kaplan, C. A.
Consequence
(Experiment 3)
Task
0, 30 min
No
--
Yes
Yes
No
--
Yes
Yes,
test battery
Division task
No
(Experiment 2)
2 min
2 min
0, 32 min
+lecturre
Yes, estimated
1990
No
2 min
0, 30 min Lecture
No
--
No
from statement of
significance
213
Yes , solve a list of
problem in
randomized order
Division
Kaplan, C. A.
Consequence
1990
( problem +
No
(Experiment 4)
2 min
40 min
problem and
No
--
Task
Yes
incubation task) t =
insight problem
1.82, p < 0.05 (10,
10, number of
participants)
Yes, when solving
problems with
Smith, S. M. &
Participants'
Read science
1991 Blankenship, S. E RAT
Yes
0.5 min
0, 5 min
misleading cues and
No
problem solving
fiction
Yes
greater effect for
(Experiment 1).
ability
low ability
participants
Smith, S. M. &
Participants'
Read
1991 Blankenship, S.
RAT
Yes
1 min
0, 5 min
Yes, when solving
science
No
problem solving problem with
fiction
E. (Experiment 2)
ability
214
misleading cues
Yes
Yes, when solving
Yes, estimated
problem with
Smith, S. M. &
Free association
Participants'
0, 0.5, 2
1991 Blankenship, S.
RAT
Yes
0.5 min
tasks ( 0.5 min No
problem solving
min
E. (Experiment 5)
from p-less-than
misleading cues,
value and
smaller effect for
and 2 min)
ability
statement of
low ability
significance
participants
General
knowledge
questionnaire
Goldman, W. P.,
(20 min) vs.
Wolters, N. C.
1992
0, 20,
Anagram
No
40 min
W., & Winograd,
Yes, in "1440 min"
General
No
--
1440 min
Yes,
condition.
knowledge
E.
questionnaire +
free activity
(1440 min)
Houtz, J. C. &
Life-relevant
Franke;, A. D.
problem
1992
No
10 min
10 min
Anagram task
215
No
--
Yes
Yes
Word
Fragment
Rest +lexical
Completion,
decision task
Torrance-Perks,
Candle
1997
Yes,
25 min
0, 10 min vs. Rest +
J. (Experiment 1) Problem,
lexical decsion
Radiation
task + cues
Problem,
RAT
216
Yes
--
No
Yes
Candle
Problem:
Memory test vs.
Candle
Memory test +
Problem,
Yes,
Radiation
Misleading
Problem,
cues
cues; Radiation
Torrance-Perks,
1997
1 min
0, 8 min
Problem: Read Yes
J. (Experiment 2)
RAT
story vs. Read
story +Cues;
RAT:
Analogy vs.
Analogy + cues
217
--
No
Yes
Two-String
Problem :
Analogy +
Two-String
Cues; Hatrack
Problem,
Yes,
Hatrack
Misleading
Torrance-Perks,
1997
Problem: Paired
3 min
0, 8 min
J. (Experiment 3)
Yes
--
Yes
Yes
No
--
No
Yes
problem solving No
Yes
associated task
Problem,
cues
+ Cues ;
Radiation
Radiation
Problem: Read
Story + Cues
Hansberry, M. T.
1998
Riddles
Yes,
1 min
0, 15 min RAT
(Experiment 2)
Participants'
Hansberry, M. T.
1998
RAT
Yes,
0.5 min
0, 10 min RAT
No
(Experiment 3)
ability
Yes, estimated
Henley, R. J.
1999
0, 1440
Anagram
(Experiment 3.2)
No
0.25 min
Free Activity
No
--
No
from statement of
min
significance
218
Yes, estimated
Henley, R. J.
1999
0, 1440
Anagram
No
0.93 min
(Experiment 4)
Free Activity
No
--
No
from statement of
min
significance
Jamieson, B. A.
1999
Mathematics
RAT
Yes
0.33 min
0, 5 min
(Experiment 1)
Jamieson, B. A.
1999
No
--
No
Yes
No
--
No
Yes
vs. Insight
Nature of the
Yes, only when
problem +
cues ( answer,
answer or unrelated
relevant
word was presented
test vs. Insight
information or
during the
problem +
related word)
incubation period
problem
Mathematics
RAT
Yes
0.33 min
0, 5 min
(Experiment 2)
problem
Insight problem
+ Drawing task
Dodds , R.,
Smith, S. M., &
2002
No, no control
RAT
No
10 min
15 min
Make a word
Yes
Ward, T. B.
group
(Experiment 1)
Make a word
test + Cues
219
Insight problem
+ Drawing task
vs. Insight
Nature of the
Yes, in informed
problem +
cues ( answer,
group and only
relevant
when answer was
vs.
information or
presented during
Insight problem
related word)
incubation
Dodds , R.,
Smith, S. M., &
2002
RAT
No
10 min
0, 15 min Make a word
Yes
Ward, T. B.
test
(Experiment 2)
+ Make a word
test + Cues
220
Yes
Half of the
participants were Yes, if the problem
told that they
solvers were
Unrelated
would turn back
working related
writing task vs.
to the initial
writing task during
Yes, estimated
from the
Medd., E &
Creative
Houtz, J.
Writing
2002
Yes,
10 min
0, 10 min
No
p-less-than value
Related Writing
assignment later, incubation period
task
and prompted to and knew that they
and statement of
significance
think about it
had to get back to
during the
the problem later
incubation period
221
Yes. High ability
Verbal
Tasks were
participants: solve
described as
more solvable
solvable or
RATs after
unsolvable.
incubation. Low
Participants’
ability participants:
problem
solve
solving ability
unsolvable RATs
No, no control
2002 Moss, S. A.
RAT
No
0.5 min
0, 15 min reasoning task
Yes
group
+Cues
more
after incubation
Nature of the
Seabrook, R., &
2003
Word
Anagram
No
0.25 min
7 min
Dienes, Z
Yes, estimated
Yes, in
Yes
cues ( Irrelevant
generation tasks
from p-less-than
Relevant-Cue group
or Revelant)
value
Searching a
Both, L.,
specific letter
2004 Needham, D., & Anagram
No
1.67 min
0, 6 min
Yes, in Incubation
No
(A to Z) and
Wood, E.
Questionnaire
222
--
Yes
group
Searching a
specific letter
Both, L.,
(A to Z) ,
2004 Needham, D., & Anagram
Yes
1.67
0, 6 min
Yes, in Incubation
Yes
--
Questionnaire
Yes
group
Wood, E
and backward
digit task
Word
Penny, C. G.,
completion task
Godsell, A.,
+ Cues (15
Yes, in 15 min
15 , 180
2004 Scott, A., &
Anagram
No
5.75 min
condition, filler
min, 180 min)
min
No
No, no control
-tasks did not make group
Balsom, R.
vs. Inactivity
(Experiment 1)
( 15
a different
min)
223
min, 180
Word
Penny, C. G.,
completion task
Yes, in 15 min
Godsell, A.,
15, 1440 + Cues (15 min,
2004 Scott, A., &
Anagram
No
5.75 min
condition, filler
No
min
No, no control
--
1440 min) vs.
tasks did not make group
Inactivity (15
a different
Balsom, R.
(Experiment 2)
min, 1440 min)
Word
Yes, but no
completion task
Penny, C. G.,
difference between
+Cues (30 min,
Godsell, A.,
2004
0, 30, 120
Anagram
No
5.75 min
Scott, A., &
the short and long
120 min) vs.
No
--
min
Yes
incubation period.
Inactivity
Balsom, R.
Filler tasks did not
( 30min, 120
make a different
min)
Read paper ( 4
Yes, in word puzzle
Insightful
min, 12 min)
0, 4 , 12
2004 Segal, E.
Mathematic
No
20 min
condition, and
vs. Word
min
Puzzle
No
--
Yes
larger effect in 4
puzzle ( 4 min,
min condition
12 min)
224
Snyder, A.,
Mitchell, J.,
Divergent
Ellwood, S., &
thinking task
2004
No
5 min
5 min
Conversation
No
--
Unknown
No
Yates A.
No, preparation
Nature of the
Christensen, B.
Analogous or
Insight
2005 T. & Schunn, C.
Not
specified
specified
No
Puzzle
D.
Not
and incubation
cues ( analogous Yes, in analogous
distracter rating Yes
periods
cue or distracter
task
cue condition
unspecified and
cue)
no control group
225
The RATs were
Yes, when solving
either
problem with
Chess-Related,
Mental
misleading cues,
Neutral, or
Rotation Task
No, did not
experts were more
Chess-Unrelated
Sio, U. N &
2007
+ Mathematics
RAT
Yes
1 min
0, 2 min
Rudowicz, E
measure the
sensitive to relevant
No
containing
Task vs.
post-incubation
concepts after an
misleading chess
Listening
problem solving
incubation period
cues. The
Music
performance
filled with mental
participants were
rotation and
either experts or
mathematics task
novices in chess
Vul, E. &
2007
Anagram
No
12 min
0, 5 min
Video game
No
Task difficulty
RAT
Yes
12
0, 5 min
Video game
No
Task difficulty
No
Yes
Pashler,H.
Vul, E. &
2007
Yes, in the fixation
min
Pashler,H.
Yes
group
226
Appendix C
Information Extracted from Each Independent Study Included in the Meta-Analysis
Misleading
ID
Year
Author
Total N
Nature of the Problem
Unbiased
Preparation
Incubation
Incubation Task
Cues
Cues
Effect Size
2 (Non-Verbal
1
1967 Gall, M. & Mendelsohn, G.
60
2 (RAT)
0
2
25
0
-.58
1
-.58
0
0
0
.52
0
.34
1
.99
Task)
2 (Associations
2
1967 Gall, M. & Mendelsohn, G.
60
2 (RAT)
0
2
25
Training)
Fulgosi, A., & Guilford, J.
3
1967
2 (Number series
50
1 (Consequence Task)
0
2
10
P.
task)
Fulgosi, A., & Guilford, J.
4
1967
2 (Number series
49
1 (Consequence Task)
0
2
20
P.
task)
1 (Guess playing
5
1969
Dreistadt R. (Study 1)
20
1 (Farm Problem)
0
5
8
card)
1 (Guess playing
6
1969
Dreistadt R. (Study 2)
20
1 (Farm Problem)
0
5
8
card)
227
1 (Guess playing
7
1969
Dreistadt R. (Study 3)
20
1 (Tree Problem)
0
5
8
0
.21
1
.48
0
.62
0
-.59
0
.11
card)
1 (Guess playing
8
1969
Dreistadt R. (Study 4)
20
1 (Tree Problem)
0
5
8
card)
2 (Multiple choice
logical syllogisms,
1 (Saugstad's ball
Murry, H. G. & Denny J. P.
9
1969
and traced
36
problem, low ability
0
5
5
(Study 1)
sequences of
group)
numbers and
letters)
2 (Multiple choice
logical syllogisms,
1 (Saugstad's ball
Murry, H. G. & Denny J. P.
10
1969
and traced
36
problem, high ability
0
5
5
(Study 2)
sequences of
group)
numbers and
letters)
11
1971
Silveria, J. M. (Study 1)
18
1 (Necklace Problem)
0
228
3
30
1 (Read book)
1 (Read book for
.06
12
1971
Silveria, J. M. (Study 2)
18
1 (Necklace Problem)
0
3
210
30 mins and free
0
activity 3 hrs)
13
1971
Silveria, J. M. (Study 3)
18
1 (Necklace Problem)
0
13
30
1 (Read book)
0
.17
0
.42
0
.44
0
0
0
0
1 (Read book for
14
1971
Silveria, J. M. (Study 4)
18
1 (Necklace Problem)
0
13
210
30 mins and free
activity 3 hrs)
1 (Read book for
15
1971
Silveria, J. M. (Study 5)
32
1 (Necklace Problem)
0
13
210
30 mins and free
activity 3 hrs)
Dominowski, R. L., &
16
1972
2 (Free
27
1 (Hat Rack Problem_
5
10
Jenrick, R.
Association)
Dominowski, R. L., &
17
1972
2 (Free
30
1 (Hat Rack Problem)
3
3
Jenrick, R.
18
1974
19
1976
Peterson, C.
Association)
24
1 (Anagram)
21
1 (Farm Problem)
.33
1.8
Anagram
1
.65
10
15
0 (Rest)
0
.10
Olton, R. M., & Johnson,
0
D. M. (Study 1)
229
2 (Stroop test +
Olton, R. M., & Johnson,
20
1976
21
1 (Farm Problem)
0
10
15
Counting
0
.11
1
.10
D. M. (Study 1)
backward)
Olton, R. M., & Johnson,
21
1976
2 ( Review the
21
1 (Farm Problem)
0
10
15
D. M (Study 1)
problem)
Olton, R. M., & Johnson,
22
1976
21
1 (Farm Problem)
0
10
15
2 ( Have lecture)
1
0
20
1 (Farm Problem)
0
10
15
0 (Listen music)
0
-.10
20
1 (Farm Problem)
0
10
15
0 (Rest)
2
.10
20
1 (Farm Problem)
0
10
15
2 (Having lecture)
2
-.03
0
12
20
0 (Relax)
0
2.19
0
12
30
2 (write essay)
0
1.07
D. M. (Study 1)
Olton, R. M., & Johnson,
23
1976
D. M. (Study 1)
Olton, R. M., & Johnson,
24
1976
D. M. (Study 1)
Olton, R. M., & Johnson,
25
1976
D. M. (Study 1)
0 (verbal divergent
26
1979
Beck, J. (Study 1)
60
thinking task)
0 (verbal divergent
27
1979
Beck, J. (Study 2)
60
thinking task)
230
0 (verbal divergent
28
1979
Beck, J. (Study 3)
60
0
12
20
0 (Relax)
0
4.07
0
12
30
2 (Write essay)
0
4.04
0
10
20
2 (Questionnaire)
0
.41
0
.33
20
2 (Questionnaire)
0
.36
0
2
5
2 (Conversation)
0
0
0
2
5
0
.66
0
.47
thinking task)
0 (verbal divergent
29
1979
Beck, J. (Study 4)
60
thinking task)
30
1985
Brockett, C. A. (Study 1)
30
31
1985
Brockett, C. A. (Study 2)
30
0 (Brick)
2 ( Remote Associates
Task)
2 (Remote Associate
32
1986
Patrick, A. S.
30
Tasks)
2 (Remote Associate
33
1986
Patrick, A. S.
30
2 (Mental
Tasks)
Rotation Task)
Browne, B. A . & Cruse, D.
34
1988
0 (Listening
60
1 (Farm Problem)
0
20
5
F. (Study 1)
music)
Browne, B. A . & Cruse, D.
35
1988
53
1 (Farm Problem)
0
20
5
1 (Graph drawing)
2
.24
55
1 (Farm Problem)
0
20
5
2 (Memory test)
0
-.17
F. (Study 1)
Browne, B. A . & Donna F.
36
1988
Cruse (Study 3)
231
Smith, S. M. & Steven, B.
37
1989
26
2 (Rebus)
1
0.5
5
26
2 (Rebus)
1
0.5
5
0 (Rest)
0
.45
0
.48
0
.49
0
.50
1
.05
1
.05
1
.41
1
.41
1
.56
(Study 1)
Smith, S. M. & Steven, B.
38
1989
2 (Music
(Study 1)
Perception Task)
Smith, S. M. & Steven, B.
39
1989
26
2 (Rebus)
1
0.5
15
26
2 (Rebus)
1
0.5
15
0 (Rest)
(Study 1)
Smith, S. M. & Steven, B.
40
1989
2 (Music
(Study 1)
Perception Task)
Smith, S. M. & Steven, B.
41
1989
25
2 (Rebus)
1
0.5
5
25
2 (Rebus)
1
0.5
5
0 (Rest)
(Study 2)
Smith, S. M. & Steven, B.
42
1989
2 (Music
(Study 2)
Perception Task)
Smith, S. M. & Steven, B.
43
1989
25
2 (Rebus)
1
0.5
15
25
2 (Rebus)
1
0.5
15
0 (Rest)
(Study 2)
Smith, S. M. & Steven, B.
44
1989
2 (Music
(Study 2)
Perception Task)
Smith, S. M. & Steven, B.
45
1989
29
2 (Rebus)
1
(Study 3)
232
0.5
10
2 (Rebus)
Smith, S. M. & Steven, B.
46
1989
29
2 (Rebus)
1
0.5
15
0 (Rebus +Rest)
1
.56
29
2 (Rebus)
1
0.5
15
2 (Rebus+Maths)
1
.56
29
2 (Rebus)
1
0.5
15
0 (Rebus+Music)
1
.56
29
2 (Rebus)
1
0.5
15
2 (Rebus)
1
.56
49
2 (Rebus)
1
0.5
5
1 (Read Story)
0
0
49
2 (Rebus)
1
0.5
5
2 (Maths Task)
0
.49
0
.49
15
0
0
0
1.07
0
0
(Study 3)
Smith, S. M. & Steven, B.
47
1989
(Study 3)
Smith, S. M. & Steven, B.
48
1989
(Study 3)
Smith, S. M. & Steven, B.
49
1989
(Study 3)
Smith, S. M. & Steven, B.
50
1989
(Study 4)
Smith, S. M. & Steven, B.
51
1989
(Study 4)
2 ( Word associate
52
1990
Dorfman, J. (Study 1)
15
2 (Number Series
task)
Task)
2 ( Word associate
53
1990
Dorfman, J. (Study 1)
15
2 (Number Series
0
.49
5
task)
Task)
2 ( Word associate
54
1990
Dorfman, J. (Study 2)
27
2 (Number Series
0
task)
.49
3
Task)
233
2 ( Word associate
55
1990
Dorfman, J. (Study 2)
27
2 (Number Series
0
.49
8
task)
2 ( Word associate
56
1990
Dorfman, J. (Study 2)
26
0
0
0
0
0
.06
0
.71
0
0
0
1.08
0
1.70
0
.64
Task)
2 (Number Series
0
.49
13
task)
Task)
2 (Psychometric
57
1993
Kaplan, C. A. (Study 1)
278
0 (Consequence test)
0
2
30
test battery)
2 (Division
58
1993
Kaplan, C. A. (Study 2)
64
0 (Consequence test)
0
2
2
problem)
59
1993
Kaplan, C. A. (Study 3)
36
0 (Consequence test)
0
2
30
60
1993
Kaplan, C. A. (Study 3)
20
0 (Consequence test)
0
4.57
28.08
2 (Lecture)
2 (Division and
insight problem)
Smith, S. M. &
2 (Remote Associates
61
1991 Blankenship, S. E. (Study
18
1 (Read science
1
.5
5
Test )
fiction)
1)
Smith, S. M. &
2 (Remote Associates
62
1991 Blankenship, S. E. (Study
21
1 (Read science
0
Test )
.5
5
fiction)
1)
234
Smith, S. M. &
63
1991
2 (Remote Associates
30
Blankenship, S. E. (Study2
1 (Read science
0
1
5
Test )
0
.37
0
.74
0
0
0
0
0
.99
0
.99
fiction)
Smith, S. M. &
2 (Remote Associates
64
1991 Blankenship, S. E. (Study
30
1 (Read science
1
1
5
Test )
fiction)
2)
Smith, S. M. &
2 ( Free
2 (Remote Associates
65
1991 Blankenship, S. E. (Study
16
0
.5
2
Associations
Test )
5)
Task)
Smith, S. M. &
2 ( Free
2 (Remote Associates
66
1991 Blankenship, S. E. (Study
18
0
.5
.5
Associations
Test )
5)
Task)
Smith, S. M. &
2 ( Free
2 (Remote Associates
67
1991 Blankenship, S. E. (Study
17
1
.5
.5
Associations
Test )
5)
Task)
Smith, S. M. &
2 ( Free
2 (Remote Associates
68
1991 Blankenship, S. E. (Study
18
1
.5
2
Associations
Test )
5)
Task)
235
Goldman, W. P.,
69
Wolters,
1992 N. C. W., & Winograd, E.
2 ( General
36
2 (Anagram)
0
.25
20
(Study 1)
Knowledge
0
.13
0
.66
0
.03
0
-.38
1
-.41
0
.14
1
.07
Questionnaire)
2 ( General
Goldman, W. P.,
70
Wolters,
1992 N. C. W., & Winograd, E.
Knowledge
36
2 (Anagram)
0
.25
1140
(Study 1)
Questionnaire +
Free Associations
Task)
71
1992
Houtz, J.C. & Frankel
72
1997 Torrance-Perks, J. (Study 1)
105
1 (Creative Writing)
0
10
10
0
0.5
10
2 (Word fragment
15
0 (Rest+Lexical
completion)
decision test)
2 (Word fragment
73
1997 Torrance-Perks, J. (Study 2)
15
0 (Rest+Lexical
0
0.5
10
completion task)
decision test)
2 (Word fragment
74
1997 Torrance-Perks, J. (Study 3)
15
0 (Rest+Lexical
1
0.5
10
completion)
decision test)
2 (Word fragment
75
1997 Torrance-Perks, J. (Study 4)
15
1 (Anagram)
0 (Rest+Lexical
1
completion)
0.5
10
decision test)
236
2 (Memory test:
cues presented as
76
1997 Torrance-Perks, J. (Study 5)
8
1 (Candle Problem)
0
1
8
2
.17
0
-.71
1
.41
0
.00
2
.15
0
.14
one of the
stimulus)
77
1997 Torrance-Perks, J. (Study 6)
8
1 (Candle Problem)
0
1
8
2 (Memory test )
1 (Read story :
78
1997 Torrance-Perks, J. (Study 7)
8
1 (Radiation Problem)
0
5
8
analogy to the
problem)
1 (Read unrelated
79
1997 Torrance-Perks, J. (Study 8)
8
1 (Radiation Problem)
0
5
8
story)
2 (Analogy : have
2 (Remote Associates
80
1997 Torrance-Perks, J. (Study 9)
8
the same solution
0
1
8
Tasks)
as the unsolved
RAT)
Torrance-Perks, J. (Study
81
1997
2 (Remote Associates
7
10)
2 (Neutral
0
Tasks)
1
8
analogy)
237
2 (Memory
Torrance-Perks, J. (Study
82
1997
test :cue presented
7
1 (Candle Problem)
1
1
8
11)
2
.56
0
-.33
1
.52
0
.00
2
.17
0
.28
a one of the
stimulus)
Torrance-Perks, J. (Study
83
1997
7
1 (Candle Problem)
1
1
8
2 (Memory test )
12)
1 (Read
Torrance-Perks, J. (Study
84
1997
7
1 (Radiation Problem)
1
5
8
story :analogy to
13)
the problem)
Torrance-Perks, J. (Study
85
1997
1 (Read unrelated
7
1 (Radiation Problem)
1
5
8
14)
story)
2 (Analogy: have
Torrance-Perks, J. (Study
86
1997
2 (Remote Associates
7
15)
the same solution
1
1
8
Tasks)
as the unsolved
RAT)
Torrance-Perks, J. (Study
87
1997
2 (Remote Associates
7
16)
2 (Neutral
1
Task)
1
8
analogy)
238
2 (Remote
88
1998 Hansberry, M. T. (Study 1)
32
2 (Riddle)
0
1
15
0
.66
0
.04
0
.28
associates task)
2 (Remote Associates
89
1998 Hansberry, M. T. (Study 2)
20
2 (Remote
0
.5
10
Task)
associates task)
2 (Remote Associates
90
1998 Hansberry, M. T. (Study 2)
20
2 (Remote
1
.5
10
Task)
associates task)
2 ( Remote Associates
91
1999 Jamieson, B. A. (Study 1)
52
0
.33
5
2 (Maths problem)
0
.00
0
.33
5
2 (Maths problem)
0
.10
Task)
2 ( Remote Associates
92
1999 Jamieson, B. A. (Study 2)
52
Task)
93
1999 Henley, R. J. (Study 3.2)*
48
2 ( Anagram)
0
.25
1440
1 (Free activity)
0
0
94
1999
26
2 ( Anagram)
0
.93
1440
1 (Free activity)
0
0
2
.31
1
-.14
Henley, R. J. (Study 4)*
2 (Insight Problem
Dodds , R., Smith, S. M., &
95
2002
2 (Remote Associates
45
Ward, T. B. (Study 1)
0
10
.5
+ Make a word
Test )
task)
2 (Insight Problem
Dodds , R., Smith, S. M., &
96
2002
2 (Remote Associates
45
Ward, T. B. (Study 2)
0
10
.5
+ Make a word
Test )
task)
239
2 (Insight Problem
Dodds , R., Smith, S. M., &
97
2002
2 (Remote Associates
42
Ward, T. B. (Study 3)
0
10
.5
+ Make a word
0
.10
0
.00
Test )
task)
Dodds , R., Smith, S. M., &
98
2002
2 (Remote Associates
70
Ward, T. B. (Study 4)
99
2 (Insight Problem
0
10
.5
Test )
+ Drawing test)
2002
Medd, E., Houtz, J. C.
15
1 ( Creative Writing)
0
10
10
2 (Writing)
1
1.05
100 2002
Medd, E., Houtz, J. C.
15
1 (Creative Writing)
0
10
10
2 (Writing)
0
0
60
2 (Anagram)
0
6
7
1
.74
2 ( Word
101 2003 Seabrook, R., & Dienes, Z*
generation)
2 (Search letter
Both, L., Needham, D., &
102 2004
.48
98
2 (Anagram)
0
1.67
6
and Answer
0
Wood, E.
questionnaire)
2 (Search letter
Both, L., Needham, D., &
103 2004
82
2 (Anagram)
0
1.67
6
and Answer
0
.09
1
.74
Wood, E.
questionnaire)
Penney, C., Godsell, A.,
104 2004
9
2 (Anagram)
0
Scott, A., & Balsom, R.
240
5.75
30
1 ( Free Activity)
Penney, C., Godsell, A.,
105 2004
9
2 (Anagram)
0
5.75
120
1 ( Free Activity)
1
1.05
9
2 (Anagram)
0
5.75
30
1 ( Free Activity)
0
.62
9
2 (Anagram)
0
5.75
120
1 ( Free Activity)
0
.70
0
20
4
1 (Read paper)
0
1.17
0
20
12
1 (Read paper)
0
1.09
0
20
4
2 (Word puzzle)
0
.90
0
20
12
2 (Word puzzle)
0
.57
0
1
5
2 (Video game)
0
.86
1
1
5
2 (Video game)
0
-.14
Scott, A., & Balsom, R
Penney, C., Godsell, A.,
106 2004
Scott, A., & Balsom, R
Penney, C., Godsell, A.,
107 2004
Scott, A., & Balsom, R
1 ( Insightful
108 2004
Segal, E. (Study 1)
20
mathematic problem)
1 ( Insightful
109 2004
Segal, E. (Study 2)
21
mathematic problem)
1 ( Insightful
110 2004
Segal, E. (Study 3)
20
mathematic problem)
1 ( Insightful
111 2004
Segal, E. (Study 4)
23
mathematic problem)
Vul, E. & Pashler,H. (Study
112 2007
2 (Remote Associates
25
1)
Test )
Vul, E. & Pashler,H. (Study
113 2007
2 (Remote Associates
25
2)
Test )
241
Vul, E. & Pashler,H. (Study
114 2007
14
2 (Anagram)
0
1
5
2 (Video game)
0
-.17
14
2 (Anagram)
0
1
5
2 (Video game)
0
-.35
14
2 (Anagram)
0
1
5
2 (Video game)
0
.36
14
2 (Anagram)
0
1
5
2 (Video game)
0
.25
3)
Vul, E. & Pashler,H. (Study
115 2007
4)
Vul, E. & Pashler,H. (Study
116 2007
5)
Vul, E. & Pashler,H. (Study
117 2007
6)
*Study with a within-subjects design: Individual was in both incubation and control conditions, the weighting formula would not apply to it,
hence, it would be excluded from the regression analyses.
242
Appendix D
Description of the selection of the dummy coded vectors and the multiplicative terms
in the regression model on all studies (with interaction terms, N = 114)
The regression model describing the interaction effect of Nature of Problem and
Incubation Task on the weighted unbiased effect size estimate, included the
multiplicative terms between Nature of Problem (In, Li, Di) and Incubation Task (H, L,
I) as predictors. The regression model is specified as follows:
Yj = ajXj + b1LijHj + b2LijLj + b3Lij Rj + b4VjHj + b5VjLj + b6VjRj + b7CjHj
+ b8 Cj Lj + b9 CjRj + kj
where Yj is the weighted unbiased effect size estimation of the study j, and Xj is a
vector of other categorical and explanatory variables (Misleading Cues, Cue, and
Preparation Period, Incubation Period) of that study. aj is the corresponding vector of
coefficients. b1j to b9j are the coefficients of the multiplicative terms, and kj is the
error term. The dummies I j and Di are eliminated through substituting equations Li
j+In j+Di j
= 1 and I j +L j +Hj = 1 into the model. The transformed regression model is
as follows:
Yj = ajXj + (b3-b9)Lij+ (b6-b9)Vj + (b7-b9)Hj + (b8-b9)Lj +
(b1-b7-b3+b9)LijHj + (b2-b8-b3+b9) LijLj + (b4-b7-b6+b9) InjHj +
(b5-b8-b6+b9)VjLj.
Regression analysis was carried out to find out the coefficient of the each variable in
the transformed model. To examine the interaction effect of Nature of the Problem and
Incubation Task, the difference between coefficients’ multiplicative terms in the
243
original model was compared. For example, the coefficient of the compound dummies
VjRj and CjRj was compared to check if an incubation period filled with low cognitive
tasks would improve the performance on Visual problems more than on divergent
thinking tasks. A comparison between the original and the transformed model
indicated that the coefficient of each variable in the transformed model was actually
the combination of the coefficients in the original model. The coefficients’ difference
between the compound dummies in the original model could be found by
re-interpreting the coefficient in the transformed model. Table D1 represents a list of
coefficient differences between compound dummies in the original model and the
equivalent combination of the coefficients in the transformed model
244
Table D143
Regression Coefficients in Original and Transformed Regression Models
The coefficient difference between
The combination of coefficients
compound dummies in the original
in the regression analysis
regression model
LijHj – Lij Rj
LijHj + Hj
LijL j- Lij Rj
LijLj + Lj
VjHj- Vj Rj
VjHj + Hj
VjLj - Vj Rj
VjLj + Lj
CjHj - Cj Rj
Hj
CjLj - Cj Rj
Lj
VjHj - CjHj
VjHj + Vj
LijHj - CjHj
LijHj + Lij
VjL j- CjLj
VjLj +Vj
LijLj- CjLj
LijL j+ Lij
Vj Rj - Cj Rj
Vj
Lij Rj - Cj Rj
Lij
245
Appendix E
Description of the selection of the dummy coded vectors and the multiplicative terms
in the regression model for studies excluding creative problem (N= 100)
The third regression model describing the interaction effect of Nature of
Problem and Incubation Task on the weighted unbiased effect size estimate, included
the multiplicative terms between Nature of Problem (V, Li) and Incubation Task (H, L,
R) as predictors. The regression model is specified as follows:
Yj = ajXj + b1LijHj + b2LijLj + b3Lij Rj + b4VjHj + b5VjLj + b6Vj Rj + kj
where Yj is the weighted unbiased effect size estimation of the study j, and Xj is a
vector of other categorical and explanatory variables (Misleading Cues, Cue, and
Preparation Period, Incubation Period) of that study. aj is the corresponding vector of
coefficients. b1j to b9j are the coefficients of the multiplicative terms, and kj is the
error term. The dummy Li and I j were eliminated through substituting equations Li
j+V j
= 1 and R j +L j +Hj = 1 into the model. The transformed regression model is as
follows:
Yj = ajXj + (b6-b3)Vj + (b1-b3)Hj + (b2-b3)Lj + (b3-b1-b6+b4) VjHj +
(b5-b2-b6+b3)VjLj.
Regression analysis was carried out to find out the coefficient of the each
variable in the transformed model. To examine the interaction effect of Nature of the
Problem and Incubation Task, the difference between coefficients’ multiplicative
terms in the original model was compared. For example, the coefficient of the
compound dummies Vj Rj and Lij Rj was compared to check if an incubation period
246
filled with low cognitive tasks would improve the performance on Visual problems
more than on linguistic insight tasks. A comparison between the original and the
transformed model indicated that the coefficient of each variable in the transformed
model was actually the combination of the coefficients in the original model. The
coefficients’ difference between the compound dummies in the original model could
be found by re-interpreting the coefficient in the transformed model. Table E1
represents a list of coefficient differences between compound dummies in the original
model and the equivalent combination of the coefficients in the transformed model.
Table E144
Regression Coefficients in Original and Transformed Regression Models
The coefficient difference between
The combination of coefficients
compound dummies in the original
in the regression analysis
regression model
LijHj - Lij Rj
Hj
LijL j- Lij Rj
Lj
LijL j- LijHj
Lj - Hj
V jHj- Vj Rj
VjHj + Hj
VjLj - Vj Rj
VjLj + Lj
VjLj - VjHj
VjLj + Lj -VjHj - Hj
LijHj - VjHj
VjHj + Vj
LijLj - VjLj
VjLj + Vj
Lij Rj - Inj Rj
Vj
247
Appendix F
RATs used in Experiment I
Answer
Solution Rate*
Over, Plant, Horse
power
10%
Wise, Work, Tower
clock
13%
Tooth, Potato, Heart
sweet
28%
Spoon, Cloth, Card
Table
26%
Stick, Maker, Point
Match
21%
Home, Arm, Room
rest
21%
battle
18%
fair
10%
Fence, Card, Master
Post
13%
Pea, Shell, Chest
Nut
23%
Rope, Truck, Line
tow
21%
Office, Mail, Hat
Fall
41%
Reading, Service, Stick
lip
10%
Cut, Cream, War
cold
31%
Rain, Test, Stomach
acid
31%
Off, Military, First
base
31%
Set A
Cry, Front, Ship
Way, Ground, Weather
Set B
*Solution rates reported in Bowden and Jung-Beeman’s study (2003)
248
Appendix G
RATs used in Experiment II
Stimulus
Answer
Solution Rate*
Soap
74%
Cottage, Swiss, Cake
Cheese
64%
Glass, Rush, Happy
Hour
72%
Safety, Cushion, Point
Pin
66%
Polish, Finger, Down
Nail
76%
Fur, Rack, Tail
Coat
79%
River, Note, Account
Bank
79%
Note, Chain, Master
Key
26%
Spoon, Cloth, Card
Table
26%
Rain, Test, Stomach
Acid
31%
Guy, Rain, Down
Fall
41%
Age, Mile, Sand
Stone
44%
Chamber, Mask, Natural
Gas
44%
Health, Taker, Less
Care
44%
Easy RATs
Opera, Hand, Dish
Intermediate RATs
*Soluion rates reported in Bowden and Jung-Beeman’s study (2003)
249
Appendix H
Rebuses Used in Experiment III
Rebus
Problem
Answer
Relevant
Misleading
Item
Cue
1
NA
Add insult to injury
add
(filler item)
Just between you
2
between
both
small
centre
fall
tear
and me
It is a small world
3
after all
4
Waterfall
250
NA
5
Long john
long
(filler task)
6
Search high and low
high
Gap
7
Splitting headache
split
She
thick
pool
too
mirror
Blood is thicker than
8
water
9
Too big to ignore
251
Through thick and
10
thin
thin
252
repeat
Appendix I
Visual Insight Problems Used in Experiment IV
Set
1
Problems
Solutions
Problem 1
English Description:
Move part of the operator
Move one stick to make the following statement
into a correct one.
Chinese Description: 移動1根火柴使以下算式變得正確
1
Restructuring
Problem 2
English Description: Move one stick to make the following statement
into a correct one
Chinese Description: 移動1根火柴使以下算式變得正確
253
2
Problem 3
Re-conceiving the
English Description: There are 10 coins on the table, please move 3
orientation of the shape
coins to make the triangle point downwards
Chinese Description: 枱上有10個硬幣, 請移動3個硬幣 使三角形
指向下方
2
Problem 4
English Description: Add 2 more square enclosures that would put
each circle in a pen by itself
Chinese Description: 加上2個正方形,使每個圓形也有自已的空間
254
3
Problem 5
Shapes in different sizes
English Description: Remove 2 sticks to leave two squares with no
loose end.
Chinese Description: 移歨2根火柴使剩下2個正方形, 火柴需互相
連結, 沒有缺口
3
Problem 6
English Description: There are 9 sticks on the table arranged, please
move 3 sticks to make 5 equilateral triangles.
255
Chinese Description: 枱上有9根火柴, 請移動3根火柴以變出5個等
邊三角形
4
Problem 7
English Description:
Re-arrange into a 3-D object
There are 4 coins on the table, please move 1
coin so that each coin touches exactly 3 other coins
Chinese Description:枱上有4個硬幣,請移動1個硬幣, 使每個硬幣
能觸到另外的3個硬幣
256
4
Problem 8
English Description: There are 6 sticks on the table, please use them to
make 4 equal-sided triangles
Chinese Description:枱上有6根火柴, 請用它們砌出4個等邊三角形
257
258