Journal Pre-proof

Autism is not associated with poor or enhanced performance on the Iowa

Gambling Task: A Meta-Analysis

Dana Zeif, Eldad Yechiam

PII: S0149-7634(19)31185-6
DOI: https://doi.org/10.1016/j.neubiorev.2020.04.016
Reference: NBR 3765

To appear in: Neuroscience and Biobehavioral Reviews

Received Date: 1 January 2020

Revised Date: 17 March 2020
Accepted Date: 11 April 2020

Please cite this article as: Zeif D, Yechiam E, Autism is not associated with poor or enhanced
performance on the Iowa Gambling Task: A Meta-Analysis, Neuroscience and Biobehavioral
Reviews (2020), doi: https://doi.org/10.1016/j.neubiorev.2020.04.016

This is a PDF file of an article that has undergone enhancements after acceptance, such as
the addition of a cover page and metadata, and formatting for readability, but it is not yet the
definitive version of record. This version will undergo additional copyediting, typesetting and
review before it is published in its final form, but we are providing this version to give early
visibility of the article. Please note that, during the production process, errors may be
discovered which could affect the content, and all legal disclaimers that apply to the journal
pertain.

© 2020 Published by Elsevier.

 1

Autism is not associated with poor or enhanced performance

on the Iowa Gambling Task: A Meta-Analysis

Dana Zeif, Eldad Yechiam

Technion – Israel Institute of Technology

of
Corresponding author: Eldad Yechiam, Max Wertheimer Minerva Center for Cognitive

ro
Studies, Faculty of Industrial Engineering and Management, Technion - Israel Institute of

Technology, Haifa 3200003, Israel. Email: yeldad@tx.technion.ac.il.

-p
This work was supported by the I-CORE program of the Planning and Budgeting

Committee and the Israel Science Foundation (1821/12). We thank the authors of the
re
original studies included in the meta-analysis.
lP

Highlights
na

 Fourteen independent studies administered the IGT to ASD and control groups.

 Individuals with ASD and controls did not differ in decision performance on
ur

the IGT

 Individuals with ASD only showed a slight performance drop in the first block
Jo

 Differences between studies were not moderated by IQ, age, gender, or study

quality

Abstract
 2

Individuals with Autism Spectrum Disorder (ASD) report difficulties in making routine

decisions. Yet there is a controversy about whether their decision performance is

impaired or enhanced compared to typically developing individuals. We focused on

studies of the Iowa Gambling Task (IGT) where contrary arguments have been made in

this regard. In a meta-analysis, we examined differences between high functioning

of
individuals with ASD and controls in decision performance (choice of long-term

advantageous options) and choice switching on the IGT. The analysis encompassed 14

ro
studies involving 433 participants with ASD and 500 controls. The results showed

-p
virtually no difference in IGT performance between groups (d = 0.04), except for a slight

disadvantage in the first block of trials for the ASD group (d = -0.16). We also found a
re
non-significant trend towards increased choice switching in the ASD group (d = -0.37)

that may be examined in future research. In sum, decision performance on the IGT is
lP

similar in individuals with ASD and controls, though their strategy may differ.

Keywords: autism, decision making, learning, performance, choice switching

na
ur

Autism Spectrum Disorder (ASD) is a heterogeneous condition characterized by

restricted interests and behaviors, as well as deficits in social communication. Individuals

Jo

with ASD also express difficulties in making everyday decisions (Luke, Ring, Redley, &

Watson, 2012; Brosnan, Chapman, & Ashwin, 2014; Gaeth, Levin, Jain, & Burke, 2016).

For example, Gaeth et al. (2016) found that high functioning adults with ASD reported

greater difficulty than controls in a variety of routine decisions such as when to go to bed,

what clothes to wear, and what food to eat. In some studies, it has been found that
 3

individuals with ASD also make poorer decisions than typically developing controls,

even in non-social contexts (e.g., Gaeth et al., 2016; Kouklari, Thompson, Monks, &

Tsermentseli, 2017). On the other hand, in other studies individuals with ASD were

found to make less biased decisions than neurotypicals (e.g., De-Martino, Harrison,

Knafo, Bird, & Dolan, 2008; South et al., 2014; Farmer, Barron-Cohen, & Skylark, 2017;

Fujino et al., 2019). Using a meta-analytic framework, the current paper evaluates the

literature on decision making performance on the Iowa Gambling task (IGT; Bechara,

of
Damasio, Damasio, & Anderson, 1994) in individuals with ASD and controls. We focus

ro
on the IGT as this decision task was most typically used in relevant investigations. Our

main research questions are: Do individuals with ASD perform more poorly in this task

-p
than controls? And are they different in terms of their exploration strategy?
re
The IGT involves repeated choices between four decks of cards, where selecting

each deck results in monetary gains and losses. Participants do not know the payoff
lP

distribution in advance and instead learn it by selecting decks and receiving payoff

feedback. In reality, two of the decks are advantageous in terms of their payoffs and two
na

are disadvantageous (as detailed in Figure 1). The disadvantageous decks yield relatively

high gains but even higher losses, leading to long term losses. By contrast, the
ur

advantageous decks yield lower gains but even lower losses, leading to long term gains.

Decks also vary with respect to the probability of negative outcomes which is either high
Jo

(50%) or low (10%). Performance in this task is typically indexed by the proportion of

selections from advantageous compared to disadvantageous decks (various

quantifications exist as detailed below).

 4

Poor IGT performance is commonly thought to reflect insensitivity to future

outcomes and increased focus on immediate outcomes (Bechara et al., 1994). However, it

can also be facilitated by imbalanced weighting of gains compared to losses and general

insensitivity to the task payoffs (Busemeyer & Stout, 2002; Yechiam, Busemeyer, Stout,

& Bechara, 2005). IGT performance is considered to be modulated by affective (or “hot”)

executive processes (Zelazo & Muller, 2002; Figner, Mackinlay, Wilkening, Murphy, &

Weber, 2006), and was found to be severely impaired in individuals with orbitofrontal

of
cortex abnormality (Bechara et al., 1994; Bechara, Tranel, & Damasio, 2000) and in

ro
chronic stimulant, opioid, and polysubstance abusers (see Kluwe-Schiavon et al., 2020).

Previous studies of individuals with ASD yielded inconsistent findings with

-p
respect to IGT performance. Some studies found that individuals with ASD made fewer
re
advantageous selections than typically developing controls (e.g., Zhang et al., 2015;

Kouklari, Thompson, Monks, & Tsermentseli, 2017). Others reported that individuals
lP

with ASD made more advantageous selections (e.g., South et al., 2014), while still others

recorded no difference between groups (e.g., Johnson et al., 2006; Yechiam, Arshavski,
na

Shamay-Tsoory, Yaniv, & Aharon, 2010). Recently, Kouklari et al. (2017) argued that

the studies showing better or equal performance of individuals with ASD were
ur

underpowered. This suggests that a meta-analysis is useful for integrating the various

findings.
Jo

Our main goal in the present meta-analysis is to examine whether individuals with

ASD differ in decision performance from controls. In addition, we also examine group

differences in exploration strategy on the IGT in light of previous findings showing that

ASD is associated with higher switching between choice options (Johnson et al., 2006;
 5

Minassian, Paulus, Lincoln & Perry, 2006; Pellicano et al., 2011). Specifically, Johnson

et al. (2006) examined whether young adults with ASD were more likely to switch

between decks on the IGT compared to typically developing controls. Choice switching

was quantified by calculating the average and maximal run of consecutive selections

from the same deck.1 The results indicated that both average and maximal runs were

considerably smaller for individuals with ASD compared to controls (similar findings

were also observed by Mussey, Travers, Grofer, Klinger, & Klinger, 2014). Also, shorter

of
runs of consecutive choices were correlated with more severe autistic syndromes in the

ro
ASD group. The reasons for the increased choice switching in ASD are yet unclear and

possible explanations range from an impairment in prototype (Johnson et al., 2006) and

-p
implicit learning (Mussey et al., 2014), to an increased drive for exploration and
re
discovery (Yechiam et al., 2010; Levin, Gaeth, Levin, & Burke, 2019). In the present

meta-analysis, we examine whether this phenomenon is robust across different studies.

lP
na
ur
Jo

1
High choice switching implies short runs of same deck choices (formally, the average run equals the
number of trials divided by the number of switches + 1).
 6

Method
The review was conducted in compliance with the recommendation checklist of the

PRISMA guidelines for systematic review (McInnes et al., 2018). To find relevant

articles, we searched Google Scholar, Medline, and EMBASE using the search terms

[(“autism spectrum” or “Asperger” or “Pervasive Developmental Disorder”) AND “Iowa

Gambling Task”]. No constraints were made on the year or language of publications (last

search March 2020). In addition, we screened reference lists of included papers to

of
identify additional articles, and scanned all conference proceedings of the International

ro
Society for Autism Research (INSAR) and the International Meeting for Autism

Research (IMFAR). Titles and abstracts were reviewed for all articles returned in this

-p
search. Abstracts in foreign language were translated (with Google Translate). Both
re
published and unpublished studies were included.

The following eligibility criteria were used: With respect to study type, we
lP

included all studies that administered the IGT to individuals diagnosed with ASD and to

typically developing control persons. All sub-diagnoses were included (Autistic Disorder,
na

Asperger Syndrome, Pervasive Developmental Disorder Not Otherwise Specified), in

line with the current DSM-V manual which no longer distinguishes these sub-types
ur

(APA, 2013). With respect to outcome measures, we included all IGT variants that use

four choice alternatives with probabilities as indicated in Figure 1 and with payoffs as in
Jo

Figure 1 or their linear transformations. In addition to the original IGT, this includes the

Hungry Donkey Task (HDT; Crone & van der Molen, 2004), which uses linearly

transformed IGT outcomes and requires participants to choose from four doors to earn

apples in order to feed a “hungry donkey”. We did not include versions of the IGT that

only have two options (e.g., Faja, Murias, Beauchaine, & Dawson, 2013; Gonzalez-
 7

Gadea et al., 2016) since their payoff structure is essentially different. In most studies the

task involved 100 trials, and in studies with additional trials we analyzed the first 100

trials for standardization.

With respect to types of participants, we included all age groups and all studies of

high functioning individuals as evidenced by mean IQ scores of 85 and above; or in the

absence of an IQ score a mean education level of 13 years and higher denoting tertiary

education. We also did not impose any restrictions on the dependent measures (i.e., did

of
not remove extreme values). Consequentiality, in two studies (Yechiam et al., 2010:

ro
Vella et al., 2018) we incorporated all observations including ones that were removed

from the studies original analysis. No publication date or publication status restrictions

were imposed.
-p
re
Methods of the analysis and inclusion criteria were specified in advance and

documented. Eligibility assessment was performed independently in an unblinded

lP

standardized manner by the two authors. Disagreements were resolved by consensus. We

used a data extraction protocol based on the Cochrane Consumers and Communication
na

Review Group’s data extraction template. One author extracted data from included

studies and the other author checked the extracted data. If the study did not include
ur

averages and standard deviations of IGT indices, we requested data from the

corresponding author. In cases where we obtained no response and relevant indices were
Jo

available in graphs we extracted relevant information using an image analysis software,

ImageJ® (V. 1.48).

Information was extracted from each included study with regards to: (1): The task

type (IGT or HDT), (2) IGT performance and choice switching indices, (3) participants’
 8

demographics including mean age, gender, and IQ in each group, (4) the type of IQ test,

and (5) sample size. Additionally, each article was reviewed to examine whether

information about how the participants were diagnosed was disclosed, and the criteria for

diagnosis (e.g., DSM-IV).

The main decision performance index was the percentage of advantageous

selections across trials. It should be noted that most alternative indices of decision

performance in the IGT (e.g., decks A + B, decks C + D, decks C + D  A  B) produce

of
identical effect sizes because they either include the addition of a constant (C + D =

ro
100%  A + B) or multiply the performance index by two (C + D  A  B = C +D –

[100%  C – D] = 2C + 2 D – 100%) which has no effect on effect sizes because both the

-p
mean distance and standard deviation are multiplied by the same number.
re
In addition, we also conducted a secondary analysis of IGT performance in each

block of 20 trials to examine effects of learning. In the absence of data on variances in

lP

different blocks we approximated the variance per block from the overall task variance.

To reduce the number of statistical comparisons and the chance of inflating type-I error
na

rates, differences between groups were tested on the first block of trials which denotes

the beginning of the learning process and is highly affected by the extent of exploration
ur

(Yechiam et al., 2001), and in the last block which maximally incorporates effects of

learning.
Jo

In our examination of choice switching, the dependent variable was mean run size

or an alternative index of choice switching. In the absence of information about mean run

size, we used the probability of choice switching, or the rate of lose-shift decisions minus

the rate of gain-stay decisions, in this order.

 9

We examined several moderating variables in order to account for possible

differences between study results. First, we tested the effect of differences in IQ, age, and

gender between the ASD and control group to evaluate the effect of imbalanced groups

(as suggested by Kouklari et al., 2017). Additionally, we examined the effect of these

indices across the two groups. It was of particular interest to examine the effect of age

and IQ as the papers that spearheaded the argument that ASD individuals show higher

performance typically focused on older adolescents with high IQ (South et al., 2008,

of
2014) while the studies demonstrating reverse findings focused on younger individuals

ro
with somewhat lower IQ (e.g., Kouklari et al., 2018a). We also planned to examine the

moderating effect of task type in the event that different variants of the IGT were used by

relevant studies.
-p
re
In addition, in light of reviewer suggestions we examined whether study results

were moderated by the quality of the study methods and data collection process. First, we
lP

examined the difference between studies for which we had raw data at the individual

participant level versus the group level. Secondly, we examined effect sizes in
na

incentivized studies and the remaining studies.2 Thirdly, we also rated the quality of each

study, using the Cochrane Collaboration’s risk of bias assessment checklist (Higgins,
ur

2011; see Supplementary section).

The meta-analysis was based on all relevant studies reported in the included
Jo

articles. In order to examine differences between ASD and control groups, we calculated

the standardized mean differences (SEM = Cohen’s d) for all relevant measures. After

2
In cocaine users, for example, it was found that significant performance deficits emerged in non-
incentivized IGT versions (e.g., Stout, Busemeyer, Lin, Grant, & Bonson, 2004, Verdejo-Garcia et al.,
2007) but not in incentivized versions (see Hulka et al., 2014).
 10

calculating the SEM for each single study, aggregated effect sizes were calculated based

on a random effect model with generic inverse variance weighting (Hedges & Olkin,

2014). For the analysis of binary moderators we used Z-tests, while for the analysis of

continuous moderators we used meta-regressions (Hedges & Olkin, 2014). To reduce the

number of statistical tests, we conducted two meta-regressions, one for the differences

between IQ, age, and gender between groups, and the other for their average across the

two groups.

of
To inspect whether publication biases exist in the data, we plotted the effect size

ro
of each study by its inverse standard error. The asymmetry of these funnel plots indicates

whether there is a publication bias in favor of studies with positive or negative effect

-p
sizes (Light & Pillemer, 1984; Egger, Smith, Schneider, & Minder, 1997). We used
re
Egger’s regression test (Egger et al., 1997) to examine the significance of the bias and a

trim-and-fill procedure (Duval & Tweedie, 2000) to correct for asymmetries in the
lP

number of studies with extreme effect sizes. All analyses were conducted using Wilson’s

Macros (Wilson, 2005) and replicated with JAMOVI, an open software using the R
na

package “Metafor()” (Viechtbauer, 2010)

ur

Results

On the basis of our search and inclusion criteria, we identified 17 studies published
Jo

between 2006 and 2018. Figure 2 provides a complete flow diagram of the search process

(as prescribed in Moher et al., 2009). Three studies of Koukari et al. (Kouklari et al.,

2017; Kouklari, Tsermentseli, & Monks, 2018a; 2018b) seemed to include IGT data from

the same group of ASD individuals with small differences in the size of the sample due to
 11

the crosstabulation with other tasks and questionnaires. In order to keep the independence

assumption of the meta-analysis, we used the study with the largest number of

participants (Kouklari et al., 2018a). An additional study could not be used due to lack of

data (averages and standard deviations) (Isaacson, Crone, & Solomon, 2005).3 The

remaining 14 analyzed studies included 433 participants with ASD and 500 controls.

Decision performance data was available for all of these studies. Block by block IGT data

was available for 12 of the studies,4 while choice switching data was available for only

of
six studies.

ro
A summary of the included studies appears in Table 1. As can be seen, the studies

focused primarily on male participants which is consistent with the male-dominance in

-p
autism (e.g., Loomes, Hull, & Mandy, 2017). One study included only males while the
re
rest had a male majority. Participants’ ages varied, stressing the need to examine the

moderating effect of age. Most studies focused on pre-adolescents and adolescents

lP

(11/14), and some on young adults (3/14); there were no studies of older adults. All but

one study reported the participants’ IQ. In Torralva et al. (2013) IQ levels were not
na

reported but most ASD and control participants had college level education, confirming

their high functioning status. All studies reported the diagnosis criteria used for allocating
ur

participants to the ASD condition. Finally, most studies administered the IGT except for

two which used the HDT (South et al., 2008; Gilbert, Zhou, Donehey, Buirkle, & Faja,
Jo

2017). Descriptive data on study quality appears in the Supplementary section.

3
This study involved only 10 individuals with ASD.
4
Mussey et al. (2015) used 25 trials blocks, and we decided to include their data given the proximity to 20
trials and the fact that different blocks had closely similar effect sizes.
 12

Decision performance:

Figure 3 shows the effect sizes of the difference between groups in decision performance

(i.e., overall rate of advantageous selections). In all, three studies indicated a significant

performance advantage to the ASD group, two studies indicated significantly poorer

performance for this group, while the remaining nine studies indicated no difference.

Across studies, the overall effect size (Cohen’s d) was 0.04 (CI 95% [-.18, .26], Z = 0.37, p

= .71), which is very close to zero.

of
Only a modest portion of the variance was attributable to heterogeneity across

ro
studies (I2 = 60.45%), with the remaining variance (about 40%) being expected due to

sampling error. Nevertheless, we conducted planned tests of moderation to further

-p
understand the difference between study results. An analysis using meta-regression
re
indicated that the variance in effect sizes was not explained by differences between

groups in mean age (B = 0.01, CI 95% [-0.02, 0.05]), mean IQ (B = 0.006, CI 95% [-0.01,
lP

0.02), or gender ratio (B = -0.70, CI 95% [-1.73, 0.32]), and was therefore not driven by

imbalanced groups. The variance in effect sizes was also not explained by the average of
na

these indices in each study across groups (age: B = -0.004, CI 95% [-0.19, 0.19]; IQ: B =

0.02, CI 95% [-0.01, 0.06]; gender: B = -0.10, CI 95% [-.6.89, 6.68]).

ur

The effect size was also not affected by the studied task (Z = 0.05, p = .96).

Specifically, the effect size for the IGT was 0.05 (CI 95% [-.21, .30]), while for the HDT
Jo

it was 0.03 (CI 95% [-.32, .39]). Effect sizes also did not differ based on whether raw

data at the individual level was available to us (Z = 0.40, p = 0.69), whether the study was

incentivized (Z = 1.80, p = .07), or the rated quality of the study (Z = 0.61, p = .54).5

5
The study quality level had only two levels (see Supplementary section) and was therefore analyzed as a
binary moderator.
 13

We next examined whether differences between groups were affected by learning.

Figure 4 presents the effect size of the difference in each block of trials (the forest plot

for each of these blocks is available in the Supplementary section). Across studies, in the

first block of trials the ASD group exhibited significantly poorer performance (d = -0.16,

CI 95% [-.32, -.0003], Z = 2.00, p = .046) though the effect size of the decrement was

small. This difference disappeared in the second block, with the ASD group showing

slightly better performance than the control group from that point on. In the last block of

of
trials there was no significant difference in performance between groups (d = 0.20, CI 95%

ro
[-.10, .49], Z = 1.30, p = .19).

Choice switching:
-p
re
The main results for choice switching are shown in Figure 5. There was a significant

effect in two studies (a marginal effect in one) while the remaining studies indicated no
lP

significant effect. Across studies the effect size was estimated as d = -0.37, which

indicates a small to medium decrease in run sizes (and an increase in choice switching) in
na

the ASD group. However, this difference was not significant (CI 95% [-.82, .08]; Z = 1.61,

p = .11). A moderate portion of the variance was attributable to heterogeneity across

ur

studies (I2 = 73.01%). Meta-regressions conducted as above did not yield any significant

moderators (all p’s > .15).

Jo

Publication bias:

Figure 6 shows the inversed standard error of the effect size as a function of its

magnitude. As indicated in the figure, in the analysis of decision performance no

 14

asymmetry was visually apparent and no significant publication bias was detected (Z =

1.03; p = .91; Egger et al. 1997). An examination of absolute effect sizes showed similar

findings. For choice switching an asymmetry was visually apparent and a significant

publication bias was detected (Z = 3.59; p = .01; Egger et al. 1997). However, given the

small number of studies with choice switching data, this latter analysis needs to be

cautiously addressed. In both indices the rate of extreme positive and negative effect sizes

was identical, and therefore we did not conduct a trim-and-fill procedure.

of
ro
General discussion

Perhaps the clearest finding in the current meta-analysis is that decision performance of

-p
individuals with ASD in the Iowa Gambling Task did not differ from that of typically
re
developing individuals. The effect size of the difference was near zero (d = 0.04) and it

was not influenced by gender ratio, mean age, or mean IQ. Moreover, the variance in
lP

effect sizes was only partially affected by heterogeneity across different studies, and 40%

of it was attributable to sampling error. This suggests that both the argument that
na

individuals with ASD show elevated decision capacities on the IGT (e.g., South et al.,

2014) or exhibit poor decision making in this task (e.g., Kouklari et al., 2018a) should be
ur

subject to additional scrutiny as currently the evidence across studies does not support

either claim.
Jo

The second finding we observed in individuals with ASD is a small but

significant decrement in decision performance in the very first block of the task, which

was followed by a non-significant trend of higher performance in subsequent blocks. This

pattern of reduced initial performance followed by somewhat higher performance is

 15

characteristic in individuals employing relatively explorative decision strategies (see e.g.,

Yechiam, Erev, & Gopher, 2001; Gray & Lindstedt, 2017). Correspondingly, our

analysis of choice switching showed a non-significant trend of increased choice

switching in individuals with ASD (d = .37), which may be examined in future research.

An obvious limitation of the current meta-analysis is that we focused on

performance in the IGT and not in other decision tasks. The IGT assesses the specific

decision capacity of selecting options that result in better long term outcomes over others

of
which result in better short term outcomes (Bechara et al., 1994), which is important in

ro
many contexts, such as drug abuse, gambling, drinking, and impulsive behavior (see e.g.,

Kluwe-Schiavon et al., 2020; Mullan, Wong, Allom, & Pack, 2011; Oldershow et al.,

-p
2009). However, the IGT was designed to mimic an environment involving high
re
uncertainty about decision outcomes and might not predict behavior in settings where

decision outcomes are based on clear rules. Especially, in settings with clear rules and no
lP

uncertainty (such as riskless decisions and decisions under risk) some studies found

reduced decision bias in individuals in ASD (e.g., De Martino et al., 2008; Farmer et al.,
na

2017). The current findings might not be generalizable to these contexts. Alternatively,

though, it might be that in settings involving lower uncertainty as well, differences

ur

between individuals with ASD and typically developing persons are not as large as

previously considered.
Jo

Another limitation of the current meta-analysis is that it focused on high

functioning individuals with ASD, and its conclusions are therefore limited to this sub-

population. As far as we are aware, very few studies examined decision making in low-

functioning individuals with ASD (see e.g., Chuan, 2015; Lambrechts et al., 2019).
 16

Chuan (2015) administered the HDT to individuals with ASD whose mental age was

about five and to an equivalent control group, and reported no significant differences in

the rate of disadvantegous selections. Nevertheless, this should be examined further.

In closing, it has been previously demonstrated that individuals with ASD have

difficulties in various emotional domains that affect decision performance (e.g., Johnson

et al., 2006; De Martino et al., 2008; Luke et al., 2012) as well as in relevant cognitive

domains including working memory (Alloway, Rajendran, & Archibald, 2009; Geurts, de

of
Vries, & van den Bergh, 2014), planning (Hill, 2004; Geurts et al., 2014), and response

ro
inhibition (Hill, 2004; Van Eylen et al., 2015). On the other hand, individuals with ASD

have compensatory advantages in germane domains such as logical consistency (De

-p
Martino et al., 2008) and low level discrimination (Mottron et al., 2006). It seems that in
re
learning to select long-term advantageous options on the IGT, these different strengths

and weaknesses balance one another.

lP

References
na

Alloway, T.P., Rajendran, G., & Archibald, L.M.D. (2009). Working memory in children

with developmental disorders. Journal of Learning Disabilities, 42, 372-382.

ur

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental

disorders (5th ed.). Washington, DC: APA.

Jo

Bechara, A., Damasio, A.R., Damasio, H., & Anderson, S. (1994). Insensitivity to future

consequences following damage to human prefrontal cortex. Cognition, 50, 7-15.

 17

Bechara, A., Tranel, D., & Damasio, H. (2000). Characterization of the decision-making

deficit of patients with ventromedial prefrontal cortex lesions. Brain, 123, 2189-

2202.

Brady, D.I. (2011). Executive functions and emotional intelligence in Asperger’s

Syndrome: A neuropsychological framework. Unpublished doctoral dissertation.

University of Calgary, Alberta.

Brosnan, M., Chapman, E., & Ashwin, C. (2014). Adolescents with autism spectrum

of
disorder show a circumspect reasoning bias rather than ‘jumping- to- conclusions’.

ro
Journal of Autism and Developmental Disorders, 44, 513-520.

Busemeyer, J.R., & Stout, J.C. (2002). A contribution of cognitive decision models to

-p
clinical assessment: decomposing performance on the Bechara gambling task.
re
Psychological Assessment, 14, 253-262.

Carlisi, C.O., Norman, L., Murphy, C.M., Christakou, A., Chantiluke, K., Giampietro, V.,
lP

Simmons, A., Brammer, M., Murphy, D.G., Mataix-Cols, D., & Rubia, K. (2017).

Shared and disorder-specific neurocomputational mechanisms of decision-making

na

in autism spectrum disorder and obsessive-compulsive disorder. Cerebral Cortex,

27, 5804-5816.
ur

Chuan, M.J. (2015). Cognición temporal en personas adultas con autismo: Un análisis

experimental. Unpublished doctoral dissertation. Complutense University of

Jo

Madrid.

Crone, E., & Van der Molen, M. (2004) Developmental changes in real life decision

making: Performance on a gambling task previously shown to depend on the

ventromedial prefrontal cortex. Developmental Neuropsychology, 25, 251-279.

 18

De Martino, B., Harrison, N.A., Knafo, S., Bird, G., & Dolan, R. J. (2008). Explaining

enhanced logical consistency during decision making in autism. Journal of

Neuroscience, 28, 10746-10750.

Duval, S., & Tweedie, R. (2000). Trim and fill: a simple funnel‐ plot–based method of

testing and adjusting for publication bias in meta‐ analysis. Biometrics, 56, 455-

463.

Egger, M., Smith, G.D., Schneider, M., & Minder, C. (1997). Bias in meta-analysis

of
detected by a simple, graphical test. British Medical Journal, 315, 629-634.

ro
Faja, S., Murias, M., Beauchaine, T.P., & Dawson, G. (2013). Reward-based decision

making and electrodermal responding by young children with autism spectrum

-p
disorders during a gambling task. Autism Research, 6, 494-505,
re
Farmer, G.D., Baron-Cohen, S., & Skylark, W.J. (2017). People with autism spectrum

conditions make more consistent decisions. Psychological Science, 28, 1067-1076.

lP

Figner, B., Mackinlay, R.J., Wilkening, F., & Weber, E.U. (2009). Hot and cold

cognition in risky decision making: accounting for age and gender differences in
na

risk taking. Journal of Experimental Psychology: Learning, Memory, and

Cognition, 35, 709-730.

ur

Flynn, J.R. (2007). What is Intelligence? Beyond the Flynn Effect. Cambridge, UK:

Cambridge University Press.

Jo

Fujino, J., Tei1, S., Itahashi, T., Aoki, Y., Ohta, H., Kanai, C., Kubota, M., Hashimoto,

R., Nakamura, M., Kato, N., & Takahashi, H. (2019). Sunk cost effect in

individuals with autism spectrum disorder. Journal of Autism and Developmental

Disorders, 49, 1-10.

 19

Gaeth, G.J., Levin , I.P., Jain, G., & Burke, E.V. (2016). Toward understanding everyday

decision making by adults across the autism spectrum. Judgment and Decision

Making, 11, 537-546.

Geurts, H.M., de Vries, M., & van den Bergh, S.F. (2014). Executive functioning theory

and autism. In S. Goldstein, & J. Naglier (Eds.), Handbook of executive functioning

(pp.121-141). New York: Springer.

Gilbert, R., Zhou, M., Donehey, J., Buirkle, J., & Faja, S. (2017). Frontal asymmetry and

of
reward-based decision making in children with high functioning autism spectrum

ro
disorder. Paper presented at the International Society for Autism Research

(INSAR) meeting, May 2017, San Francisco, CA.

-p
Gonzalez-Gadea, M.L., Baez, S., Torralva, T., Castellanos, F. X., Rattazzi, A., Bein, V.,
re
Rogg, K., Manes, F., & Ibanez, A. (2013). Cognitive variability in adults with

ADHD and AS: Disentangling the roles of executive functions and social
lP

cognition. Research in Developmental Disabilities, 34, 817-830.

Gonzalez-Gadea M.L., Sigman M., Rattazzi A., Lavin C., Rivera-Rei A., Marino J.,
na

Manes, F., & Ibanez, A. (2016). Neural markers of social and monetary rewards in

children with Attention-Deficit/Hyperactivity Disorder and Autism Spectrum

ur

disorder. Scientific Reports, 6, 1-11.

Gray, W.D., & Lindstedt, J.K. (2017). Plateaus, dips, and leaps: Where to look for
Jo

inventions and discoveries during skilled performance. Cognitive Science, 41,

1838-1870.

Hedges, L.V., & Olkin, I. (2014). Statistical Methods for Meta-Analysis. New York, NY:

Academic press.
 20

Higgins, J.P.T., Altman, D.G., Gøtzsche, P.C., Jüni, P., Moher, D., Oxman, A.D.,

Savovic, J., Schulz, K.F., Weeks, L., & Sterne, J.A. (2011). The Cochrane

Collaboration's tool for assessing risk of bias in randomised trials. British Medical

Journal, 343, d5928.

Hill, E.L. (2004). Evaluating the theory of executive dysfunction of autism.

Developmental Review, 24, 189-233.

Hulka, L.M., Eisenegger, C., Preller K.H., Vonmoos, M., Jenni, D., Bendrick, K.,

of
Baumgartner, M.R., Seifritz, E., & Quednow B.B. (2014). Altered social and non-

ro
social decision-making in recreational and dependent cocaine users. Psychological

Medicine, 44, 1015-1028.

-p
Isaacson, K., Crone, E., & Solomon, M. (2005). Decision making in children with high
re
functioning autism. Paper presented at the International Meeting for Autism

Research (IMFAR), May 2005, Boston, MA.

lP

Johnson, S.A., Yechiam E., Murphy, R.M., Queller, S., & Stout, J.C. (2006).

Motivational processes and autonomic responsivity in Asperger’s disorder:

na

Evidence from the Iowa gambling task. Journal of the International

Neuropsychological Society, 12, 668-676.

ur

Kluwe-Schiavon, B., Viola, T.W., Sanvicente-Vieira, B., Lumertz, F.S., Salum, G.A.,

Grassi-Oliveira, R., & Quednow, B.B. (2020). Substance related disorders are
Jo

associated with impaired valuation of delayed gratification and feedback

processing: A multilevel meta-analysis and meta-regression. Neuroscience &

Biobehavioral Reviews, 108, 295-307.

 21

Kouklari, E.C., Thompson, T., Monks, C.P., & Tsermentseli, S. (2017). Hot and cool

executive function and its relation to theory of mind in children with and without

autism spectrum disorder. Journal Of Cognition And Development, 18, 399-418.

Kouklari, E-C., Tsermentseli, S., & Monks, C.P. (2018a). Hot and cool executive

function in children and adolescents with autism spectrum disorder: Cross-sectional

developmental trajectories. Child Neuropsychology, 24, 1088-1114.

Kouklari, E-C., Tsermentseli, S., & Monks, C.P. (2018b). Developmental trends of hot

of
and cool executive function in school-aged children with and without autism

ro
spectrum disorder: links with theory of mind. Development and Psychopathology,

31, 541-556.

-p
Lambrechts, A., Cook, J., Ludvig, E., Alonso, E., Anns, S., Taylor, M., & Gaigg, S.
re
(2019). Reward devaluation in autistic children and adolescents with complex

needs: a feasibility study. Submitted for publication.

lP

Levin, I.P., Gaeth, G.J., Levin, A.M., & Burke, E.V. (2019). Decision- making processes

of high- functioning adults on the autism spectrum. In K. Morsanyi, & R.M.J.

na

Byrne (Eds.), Thinking and reasoning in autism (pp. 39-58). London: Routledge.

Light, R.J., & Pillemer, D.B. (1984). Summing up. The science of reviewing research.
ur

Cambridge, MA: Harvard University Press.

Loomes, R., Hull, L., & Mandy, W.P.L. (2017). What is the male-to-female ratio in
Jo

Autism Spectrum Disorder? A systematic review and meta-analysis. Journal of the

American Academy of Child and Adolescent Psychiatry, 56, 466-474.

Luke, L., Ring, H., Redley, M., & Watson, P. (2012). Decision-making difficulties

experienced by adults with autism spectrum conditions. Autism, 16, 612-621.

 22

McInnes, M.D.F., Moher, D., Thombs, B.D., McGrath, T.A., Bossuyt, P.M., the

PRISMA-DTA Group, & Willis, B.H. (2018). Preferred reporting items for a

systematic review and meta-analysis of diagnostic test accuracy studies: The

PRISMA-DTA statement. Journal of the American Medical Association, 319, 388-

396.

Minassian, A., Paulus, M., Lincoln, A., & Perry, W. (2006). Adults with autism show

increased sensitivity to outcomes at low error rates during decision-making.

of
Journal of Autism and Developmental Disorders, 37, 1279-1288.

ro
Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., & The PRISMA Group (2009).

Preferred reporting items for systematic reviews and meta-analyses: The PRISMA

statement. PLOS Medicine, 6, e1000097.

-p
re
Mottron, L., Dawson, M., Soulières, I., Hubert, B., & Burack, J. (2006). Enhanced

perceptual functioning in autism: An update, and eight principles of autistic

lP

perception. Journal of Autism and Developmental Disorders, 36, 27-43.

Mullan, B., Wong, C., Allom, V., & Pack, S.L. (2011). The role of executive function in
na

bridging the intention-behaviour gap for binge-drinking in university students.

Addictive Behaviors, 36, 1023-1026.

ur

Mussey, J.L., Travers, B.G., Klinger, L.G., & Klinger, M.R. (2015). Decision‐ Making

Skills in ASD: Performance on the Iowa Gambling Task. Autism Research, 8, 105-
Jo

114.

Oldershaw, A., Grima, E., Jollant, F., Richards, C., Simic, M., Taylor, L., & Schmidt, U.

(2009). Decision making and problem solving in adolescents who deliberately self-

harm. Psychological Medicine, 39, 95-104.

 23

Pellicano, E., Smith, A.D., Cristino, F., Hood, B.M., Briscoe, J., & Gilchrist, I.D. (2011).

Children with autism are neither systematic nor optimal foragers. Proceedings of

the National Academy of Sciences, 108, 421-426.

Sanjurjo, N.S., Montanes, P., Sierra Matamoros, F.A., & Burin, D. (2015). Estimating

intelligence in Spanish: Regression equations with the word accentuation test and

demographic variables in Latin America. Applied Neuropsychology: Adult, 22,

252-261.

of
Sawa, T., Kodaira, M., Oiji, A., Sasayama, D., Iwadare, Y., Ushijima, H., Usami, M.,

ro
Watanabe, K., & Saito, K. (2013). Dysfunction of orbitofrontal and dorsolateral

prefrontal cortices in children and adolescents with high-functioning pervasive

-p
developmental disorders. Annals of General Psychiatry, 12, 31.
re
South, M., Chamberlain, P.D., Wigham, S., Newton, T., Le Couteur, A., McConachie,

H., Gray, L., Freeston, M., Parr, J., Kirwan C.B., & Rodgers, J. (2014). Enhanced
lP

decision making and risk avoidance in high-functioning autism spectrum

disorder. Neuropsychology, 28, 222-228.

na

South, M., Ozonoff, S., Suchy, Y., Kesner, R.P., McMahon, W.M., & Lainhart, J.E.

(2008). Intact emotion facilitation for nonsocial stimuli in autism: Is amygdala

ur

impairment in autism specific for social information? Journal of the International

Neuropsychological Society, 14, 42-54.

Jo

Stout, J.C., Busemeyer, J.R., Lin, A., Grant, S.J., & Bonson, K.R. (2004). Cognitive

modeling analysis of decision-making processes in cocaine abusers. Psychonomic

Bulletin and Review, 11, 742-747.

 24

Torralva, T., Gleichgerrcht, E., Roca, M., Ibanez, A., Marenco, V., Rattazzi, A., &

Manes, F. (2013). Impaired theory of mind but intact decision-making in Asperger

syndrome: Implications for the relationship between these cognitive domains.

Psychiatry Research, 205, 282-284.

Van Eylen, L., Boets, B., Steyaert, J., Wagemans, J., & Noens, I. (2015). Executive

functioning in autism spectrum disorders: Influence of task and sample

characteristics and relation to symptom severity. European Child and Adolescent

of
Psychiatry, 24, 1399-1417.

ro
Vella, L., Ring, H.A., Aitken, M.R.F., Watson, P.C., Presland, A., & Clare, I.C.H.

(2018). Understanding self-reported difficulties in decision-making by people with

-p
autism spectrum disorders. Autism, 22, 549-559.
re
Verdejo-Garcia, A., Benbrook, A., Funderburk, F., David, P., Cadet, J.L., & Bolla, K.I.

(2007). The differential relationship between cocaine use and marijuana use on
lP

decision-making performance over repeat testing with the Iowa Gambling Task.

Drug and Alcohol Dependence, 90, 2-11.

na

Viechtbauer, W. (2010). Conducting meta-analyses in R with the metaphor package.

Journal of Statistical Software, 36, 1-48.

ur

Wilson, D.B. (2005). Meta-analysis macros for SAS, SPSS, and Stata. Retrieved from

http://mason.gmu.edu/~dwilsonb/ma.html
Jo

Yechiam, E., Arshavski, O., Shamay-Tsoory, S.G., Yaniv, S., & Aharon, J. (2010).

Adapted to explore: Reinforcement learning in Autistic Spectrum Conditions.

Brain and Cognition, 72, 317-324.

 25

Yechiam, E., Busemeyer, J.R, Stout, J.C., & Bechara, A. (2005). Using cognitive models

to map relations between neuropsychological disorders and human decision making

deficits. Psychological Science, 16, 973-978.

Yechiam, E., Erev, I., & Gopher, D. (2001). On the potential value and limitations of

emphasis change and other exploration-enhancing training methods. Journal of

Experimental Psychology: Applied, 7, 277-285.

Zelazo, P.D., & Muller, U. (2002). Executive function in typical and atypical

of
development. In U. Goswami (Ed.), Handbook of childhood cognitive development

ro
(pp. 445-469). Oxford: Blackwell.

Zhang, L., Tang, J., Dong, Y., Ji, Y., Tao, R., Liang, Z., Chen, J., Wu, Y., & Wang, K.

-p
(2015). Similarities and differences in decision-making impairments between
re
Autism Spectrum Disorder and Schizophrenia. Frontiers in Behavioral

Neuroscience, 9, 259.
lP
na
ur
Jo
 26

Table 1: Details of the included studies. Sample size for the ASD and control groups

(NASD/C) followed by the percentage of males and mean age and IQ in each group, the IQ

test used, and the ASD diagnosis criteria.

Study NASD/C %MaleASD/C AgeASD/C IQASD/C IQ test ASD Diagnosis

Johnson et al., 2006 15/14 73/71 16.1/15.9 113.9/114.2 WASI DSM-IV

South et al., 2008 32/30 97/95 19.7/19.2 107.7/112.7 WASI DSM-IV

of
Yechiam et al., 2010 15/28 93/93 15.6/15.6 100.5/101.5 WASI-SiS1 ICD-10

Brady, 2011 29/33 76/76 18.8/18.9 112.8/110.4 WASI DSM-IV-TR

ro
Gonzalez-Gadea et al., 2013 23/21 65/52 33.0/38.2 93.4/93.1 WAT2 DSM-IV

Sawa et al., 2013 19/19 89/89 13.2/13.1 96.0/97.3 WASI DSM-IV-TR

Torralva et al., 2013

South et al., 2014

25/25

48/56
72/60

94/75
-p
33.9/36.4

13.2/12.6
-/-

109.8/113.8
-

WASI
DSM-IV

DSM-IV
re
Mussey et al., 2015 15/18 -/- 18.8/19.0 103.4/94.8 KBIT DSM-IV

Zhang et al., 2015 37/80 84/84 18.9/19.2 103.2/108.1 RPMT DSM-IV-TR

lP

Carlisi et al., 2017 24/20 100/100 14.6/15.1 113.1/119.7 WASI ICD-10

Gilbert et al., 2017 34/28 91/86 9.0/9.7 108.0/112.1 WASI DSM-V

na

Kouklari et al., 2018a 79/91 82/66 11.3/10.8 95.9/99.8 WASI DSM-IV

Vella et al., 2018 38/37 66/68 34.1/34.0 116.4/114.2 WASI-V DSM-IV

ur

Notes: WASI = Wechsler Abbreviated Scale of Intelligence (full scale IQ); WASI-SiS = WASI
Similarities Subscale; WASI-V = WASI Verbal IQ; WAT = Word Accentuation test; RPMT = Raven
Jo

Progressive Matrices Test; KBIT = Kaufman Brief Intelligence Test. 1 – WASI-SiS scores were
converted to IQ using the norms in Flynn (2007). 2 - WAT scores were converted to IQ using the
norms in Sanjurjo, Montanes, Matamoros, and Burin (2015).
 27

Figure 1: Left – An illustration of the Iowa Gambling Task. Right – the available

alternatives in the task and their outcomes: The top two decks (A, B) are disadvantageous

and the bottom two (C, D) are advantageous. Note that as in the example on the left,

gains and losses can occur simultaneously.

Win $100 every card

A 0.5 probability of losing $250

of
Win $100 every card
B 0.1 probability of losing $1,250

ro
C Win $50 every card
0.5 probability of losing $50

-p
D
Win $50 every card
0.1 probability of losing $250
re
lP
na
ur
Jo
 28

Figure 2: Flow diagram of the literature search.

Articles identified from Articles identified

database searching from other sources
(n = 1101) (n = 3)

(n = 904)
Articles after duplicates removed Excluded (n = 1033)
(n = 1078) Review or theory paper: 408
Irrelevant disorder: 284
Healthy population only: 188

of
Brain-lesions study: 66
No IGT: 49
Articles screened on the basis of Animal research: 15
title and abstract No control group: 10

ro
(n = 1078) Re-analysis: 8
Abstract not available: 5

-p Excluded (n = 29)
re
Full text article assessed for Did not use the IGT: 19
eligibility No ASD group: 4
(n = 46) Re-analysis: 4
IGT for children:2
lP

Full text articles corresponding Excluded (n = 3)

na

to all inclusion criteria Non independent data:2

(n = 17) No data: 1
ur

Studies included in the analysis

(n = 14)
Jo
 29

Figure 3: Forest plot of performance differences between the ASD and control group

across all studies. Effect sizes falling to the right of zero indicate higher IGT performance

for the ASD group. Error bars represent 95% confidence intervals (CI) around the effect

sizes. The right hand side presents the effect sizes and CI in brackets for each study.

Below is the random effect (RE) model estimate.

of
ro
-p
re
lP
na
ur
Jo
 30

Figure 4: Random effect model estimates for performance differences in each of five

blocks of trials. Positive scores denote higher IGT performance for the ASD group, while

negative scores denote higher performance of the control group. Error bars represent 95%

confidence intervals.

0.5

of
0.4

0.3

ro
Cohen’s d (ASD – control)

0.2

0.1

-p
0

-0.1
re
-0.2

-0.3
lP

-0.4
Block of trial
-0.5
1 2 3 4 5
na
ur
Jo
 31

Figure 5: Forest plot for run size differences between the ASD and control groups. Effect

sizes falling to the left of zero indicate shorter run sizes (and greater choice switching) for

the ASD group. Error bars represent 95% confidence intervals (CI) around the effect

sizes. The right hand side presents the effect sizes and CI in brackets for each study.

Below is the random effect (RE) model estimate.

of
ro
-p
re
lP
na
ur
Jo
 32

Figure 6: Funnel plots displaying the inverse standard error of the effect size (Cohen’s d),

plotted as a function of the magnitude of the effect size in each study. The dotted lines

denote the predicted 95% confidence intervals.

Decision performance Choice switching

Inverse standard error

Inverse standard error

of
ro
Effect size

-p Effect size
re
lP
na
ur
Jo