Abstract
Autistic people often have an atypical profile of abilities: while excelling in some structured paradigms, many report difficulties with making real-life decisions. To test whether decision-making in autism is different from in typically developing controls, we reviewed 104 studies that compared decision-making performance between autistic and comparison participants (N = 2712 autistic and N = 3189 comparison participants) between 1998 and 2022. Our searches revealed four main decision-making paradigms that are widely used in the field of decision neuroscience: perceptual discrimination, reward learning, metacognition and value-based decision-making paradigm. Our synthesis highlights that perceptual processing and reward learning were similar between autistic and comparison participants, whereas value-based decision-making and metacognitive accuracy were often different between groups. Furthermore, decision-making differences were most pronounced when the autistic participant was explicitly probed to report on an internal belief, while implicit markers of the same decision (e.g. error-related response times) were usually not different. Our findings provide evidence in favour of a metacognitive explanation of decision-making atypicalities in autism.
Lay summary
Many autistic people report difficulties with real-life decision-making. However, when doing decision-making tests in laboratory experiments, autistic people often perform as well or better than non-autistic people. We review previously published studies on autistic people’s decision-making, across different types of tests, to understand what type of decision-making is more challenging. To do this, we searched four databases of research papers. We found 104 studies that tested, in total, 2712 autistic and 3189 comparison participants on different decision-making tasks. We found that there were four categories of decision-making tests that were used in these experiments: perceptual (e.g. deciding which image has the most dots); reward learning (e.g. learning which deck of cards gives the best reward); metacognition (e.g. knowing how well you perform or what you want); and value-based (e.g. making a decision based on a choice between two outcomes that differ in value to you). Overall, these studies suggest that autistic and comparison participants tend to perform similarly well at perceptual and reward-learning decisions. However, autistic participants tended to decide differently from comparison participants on metacognition and value-based paradigms. This suggests that autistic people might differ from typically developing controls in how they evaluate their own performance and in how they make decisions based on weighing up the subjective value of two different options. We suggest these reflect more general differences in metacognition, thinking about thinking, in autism.
Introduction
Autism spectrum disorder or condition (ASC) is a neurodevelopmental condition characterised by social communication difficulties, repetitive behaviours and/or restricted interests (American Psychological Association [APA], 2013). Of the 1% of the population who are autistic, an estimated 4% (Howlin et al., 2004) to 64% (Cederlund et al., 2008) live autonomously with ordinary levels of support. Employment on either a full or part-time basis is similarly low in autistic adults (or ‘people on the autism spectrum’; Kenny et al., 2016), ranging from around 7% to 40% (Engström et al., 2003; Helles et al., 2017). While poor outcomes for autistic adults reflect a range of environmental and societal factors (e.g. lack of reasonable accommodations, stigma and victimisation), difficulties in everyday decision-making may also affect independence and employment (Luke et al., 2012). Nonetheless, autistic adults often perform comparably to typical adults on structured tests that are usually reliable indicators of cognitive functioning, such as intelligence tests (Brady et al., 2014; Dawson et al., 2007), tests of cognitive flexibility (Geurts et al., 2020; Van Eylen et al., 2011) and structured tests of theory of mind (Livingston et al., 2019). This latter point is particularly pertinent as, behaviourally, some autistic people can ‘pass’ as typical, despite underlying cognitive difficulties and mental health problems. As described by an autistic person: ‘It’s what’s going on cognitively, not behaviourally and people don’t see that. It’s frustrating because I don’t. . .get the support or understanding that I need’ (Livingston et al., 2019). This suggests that everyday functioning and decision-making are difficult for autistic people and potentially more difficult than may be expected from performance on laboratory cognitive tasks or paradigms (Geurts et al., 2020; Luke et al., 2012).
Over the last two decades, many studies have explored decision-making in autism using diverse paradigms. Bayesian inference theories explain autism as arising from an oversensitivity to new, compared to old, decision evidence (Pellicano & Burr, 2012; Van de Cruys et al., 2014; for an alternative view, see Van Boxtel & Lu, 2013), and accordingly, have led to a focus on testing perceptual decision-making (deciding on the basis of perceptual cues without feedback) and reward learning (deciding on the basis of learned reward contingencies) in autism. Strikingly, studies of perception (Greimel et al., 2013; Peiker et al., 2015; Plaisted et al., 1998; Powell et al., 2016) and learning (Brown et al., 2010; D’Cruz et al., 2016; Luman et al., 2009; Zwart et al., 2017; Zwart, Vissers, Kessels, et al., 2018; Zwart, Vissers, & Maes, 2018) have mostly found mixed results at best. Recently, the psychological study of decision-making has increasingly been investigated in the realm of ‘meta’cognition– the set of processes involved in consciously evaluating or reflecting upon first-order cognitive processes (Dunlosky & Metcalfe, 2009; Flavell, 1979; Nelson, 1990; Shea et al., 2014). First-order cognition encompasses types of cognition that are not affected by ‘thinking about thinking’, such as learning in the absence of self-evaluation or perception without awareness. There is now a plethora of work suggesting that the key distinctions between decision-making in ASC and neurotypicals can be attributed to difficulties with the metacognitive reflection upon first-order cognition, as compared to difficulties with first-order cognition itself (Carruthers, 2009; Frith & Happé, 1994; Happé & Frith, 2006). However, a direct comparison as to what extent decision-making differences in ASC arise specifically on decision-making paradigms that require metacognition, as compared to decision-making paradigms that require mostly first-order cognition, has to our knowledge never been conducted.
A related line of recent work has outlined what role explicit and implicit decision processes play in guiding and informing everyday decisions. The distinction between implicit and explicit metacognition rests on whether the participant is tasked with giving their response consciously (such as with verbal reports of confidence) or implicitly, when decision metrics are inferred from behaviour on a paradigm (such as how long the process of deciding takes). Several recent studies suggest that autistic decision-making difficulties on learning and perceptual paradigms are restricted to instances in which the response needs to be given explicitly, and are not seen when the response is given implicitly (Carpenter et al., 2019; Nicholson et al., 2019, 2020). This suggests that decision-making differences between autistic and comparison participants mostly arise when the decision needs to be expressed explicitly by means of metacognitive processing, but not when the decision is derived from implicit, first-order, processing.
We here sought to understand the (in)consistencies in this field by undertaking a narrative review of the decision-making literature in autism. This methodology allowed us to consider the various distinct types of decision-making paradigms and evaluate these results in light of the same question, that is, whether decision-making in autism is atypical, and if so, whether decision-making differences in autism mostly occur in decision-making challenges that require metacognitive processing. Our narrative review was influenced by recent Bayesian accounts of decision-making in ASC (Friston et al., 2013; Lawson et al., 2017; Pellicano & Burr, 2012). Building upon the finding that metacognition is particularly pertinent to decisions that require the decider to act upon subjective preferences and social decision-making, we predicted that decision-making in these domains would be more different between autistic and comparison participants, while differences would not be observed on (first-order) perceptual decision-making and reward-learning paradigms. We here built upon a distinction between decision-making paradigms that has been outlined elsewhere (Van der Plas et al., 2019). In addition, we expected that autistic people’s explicit decision-making responses would be more atypical than their implicit responses to the same decision. To test this, we reviewed published decision-making studies that compared autistic and comparison participants.
Methods
This review involved a search of the following databases: Web of Science, PsychINFO, MEDLINE, CINAHL. After piloting the search terms and making refinements, the final set of key search terms were: Autism Spectrum Disorders, Asperge*, decision-making, decision evidence, prediction error*, volatility, cognitive flexibility, reinforcement learning, implicit learning, visuo-spatial information processing, central coherence, attentional bias, prior knowledge, sensory discrimination (see Supplementary Material 1 for the full Embase search strategy). The database searches included all entries up to August 2022. We included studies that tested a diagnosed ASC sample or a general population sample with a quantitative measure of autistic traits and that involved a lab-based or online-based decision-making paradigm on a repeated number of trials (a full list of inclusion/exclusion criteria can be found in Supplementary Material 2). Finally, to aid the synthesis of the findings, we used the Quality Assessment Tool for Studies with Diverse Designs (QATSDD; Sirriyeh et al., 2012). We used the 14 items pertaining to quantitative studies to assess study quality (each item is scored 0–3, meaning scores can range from 0 to 42). An average score was created for each study by both lead authors, who then identified studies with the highest score (defined as the upper quartile of the distribution of mean quality scores); see Supplemental Tables 1a–d (highest-quality studies are denoted with ‘†’ – based on the top quartile of included studies). For the included articles, the mean study quality was 22.8 (SD = 3.24, 15.5–31).
Community involvement statement
One of the authors is an autistic person and was the main contributor to all aspects of the study (the design of the search strategy, the review process, the interpretation of the included studies, shaping the conclusions of the review and co-authoring the article). There was no other community involvement in this report.
Results
Our search strategy returned a total number of 10,133 hits; of these, 4389 were duplicates. At the screening stage, 5520 hits were excluded (because they did not contain the right task, or the right sample, or were reviews/conference proceedings/reviews, etc.). A total of 228 articles were assessed for eligibility, leading to the inclusion of 104 articles (see Supplementary Material 3).
Perceptual decision-making
On perceptual decision-making paradigms, participants are asked to discriminate features of perceptual evidence, such as the direction of moving dots, in the absence of feedback. Our search found 21 studies in this category (N = 390 ASD, mean age approximately 17.7; 1 and N = 110 comparison participants, mean age approximately 17.1; of studies reporting IQ, all participants had a minimum IQ of 70). Four studies were with samples in childhood (mean age < 13), two were with adolescents (mean age 13–17), 13 were with adults (aged 18+), one study had samples of more than one age group and one study did not report a mean age. Of these 21 studies, one (4.8%) found significant differences in perceptual sensitivity between autistic and comparison participants (where autistic participants had higher thresholds for coherence detection; Milne et al., 2002). Ten of the 21 studies (47.6%) reported no statistical difference in perceptual discrimination (i.e. on performance metrics or neural activity) between autistic and comparison participants (Greimel et al., 2013; Manning et al., 2022; Peiker et al., 2015; Pirrone et al., 2017; Plaisted et al., 1998; Sapey-Triomphe et al., 2021b; Sheppard & Altgassen, 2021). Ewbank et al. (2016) reported that neural response was unrelated to autistic traits on a repetition suppression task; Powell et al. (2016) found that autistic traits did not uniquely explain sensory thresholds. Todorova et al. (2021) reported that autistic participants were comparable to control participants on an alternate forced choice task and that both groups showed equal performance on the motion and control tasks (performance in the autism group was poorer for detecting targets, but this was driven by one autistic outlier).
The remaining 10 studies (47.6%) reported mixed results. In a visual-temporal order task, autistic participants showed a difference in visual-temporal resolution for disgust and neutral stimuli (which was not found in the comparison group; Chakrabarty et al., 2021). Two studies (9.7%) showed that people on the autism spectrum were better than comparison participants at discriminating features of simple stimuli, but their performance deteriorated when stimuli were made more complicated (Bertone et al., 2005; Zaidel et al., 2015). Similarly, on a probabilistic visual search task, autistic participants performed as well as comparison participants, except that reaction times were longer in the autism group when the target followed a distractor in a rare region of the search space (Allenmark et al., 2021). Binur et al. (2022) found that autistic participants showed an equivalent width-height bias to comparison participants, but the blur effect on perceived width was attenuated in the autism group. Two trait-based studies also fell into this category. One study, with 22 autistic participants found that repetitive behaviour scores were unrelated to performance on a switch-only or inhibition-only condition of an anti-saccade task. Yet, repetitive behaviours were associated with performance on the combined switching and inhibition condition (Cissne et al., 2022). In a sample of 222 typically developing participants, using a coherent motion task, autistic traits were unrelated to both the starting point or boundary separation parameters on a drift-diffusion model (Retzler et al., 2021).
The three remaining studies investigated perceptual discrimination differences in combination with computational modelling or neuroimaging techniques, and interpreted their findings as supporting lower global, but not local, perceptual processing in autism compared with comparison participants (Lieder et al., 2019; Stevenson et al., 2017; Van de Cruys et al., 2018).
Together, these studies suggest that autistic people perform comparably to matched control groups on decision-making paradigms in which participants need to choose between targets on the basis of perceptual evidence. Theoretically, this suggests that the sensory thresholds for visual perception are not atypical in autism, as is often proposed by Bayesian accounts of autism (e.g. Friston et al., 2003). Interesting exceptions are Bertone et al. (2005) and Zaidel et al. (2015), which showed better performance of autistic people for simple, but not for more complicated, visual stimuli. Based on this, there is no sufficient evidence to conclude autistic people perform atypically on perceptual decision-making paradigms (see Supplemental Table 1a).
Reward-based learning
We next turn to learning paradigms, in which participants make repeated decisions about choice-options which have reward contingencies that fluctuate and can be learned with feedback. This category contained 43 studies (N = 1244 ASD, mean age approximately 18.0; and N = 1373 comparison participants, mean age approximately 17.0; of those studies reporting IQ, only four studies (Reed, 2019; Kelly & Reed, 2021; Landry & Mitchell, 2021; Lung & Bertone, 2021) included participants with an IQ < 70). Of 43 learning studies, 14 included childhood samples, 9 included adolescent samples, 17 included adult samples, and 3 included samples from multiple age groups. A total of 18 studies (41.9%) did not report a statistically significant difference in learning performance between autistic and comparison participants (Barnes et al., 2008; Kruppa et al., 2019; Luman et al., 2009; Lung & Bertone, 2021; Nader et al., 2022; Solomon et al., 2010; Sapey-Triomphe et al., 2021a; Sawaya et al., 2019; Ward et al., 2022; Zwart et al., 2017; Zwart, Vissers, Kessels, et al., 2018), 2 although we note some other group differences were reported. Brown et al. (2010) found autistic participants made fewer errors on a contextual cueing task; Minassian et al. (2007) reported that autistic participants were more prone than controls to switching decks in the Iowa Gambling Task (IGT); and South et al. (2012) found autistic participants were slower at updating at the reversal learning stage.
Nine studies (20.9%) reported mixed results, showing that autistic participants learned less than comparison participants when aspects of the decision were experimentally manipulated, such as allowing for ambiguity in how feedback can be determined (when ambiguous or negative feedback is removed from the situation; Broadbent & Stokes, 2013; Solomon et al., 2015), increasing memory demand (Stoet & López, 2011), presenting feedback in a random rather than a consistent manner (Greene et al., 2019; Robic et al., 2015) or whether there was one clear rule or instruction to be followed versus when the main rule or instruction of the task was ambiguous (Kelly & Reed, 2021; Landry & Mitchell, 2021; Reed, 2019; Sapey-Triomphe et al., 2018, 2022). Importantly, all these studies found no group differences in reward learning between autistic and non-autistic participants when the paradigm had no additional memory demand (Stoet & López, 2011), when only positive feedback was given to participants (Broadbent & Stokes, 2013), when feedback was unambiguous (Solomon et al., 2015) and when there was only one clear rule to be followed (Reed, 2019; Sapey-Triomphe et al., 2018).
In a non-Western sample, Zhang et al. (2015) found mixed performance on two different gambling tasks. On the IGT, comparison participants accrued a greater net score and performed better on the fourth and fifth blocks than autistic participants. Using a dice task, Zhang and colleagues found that autistic participants were less likely to use feedback in both high- and low-risk contexts, and accrued a lower net score than comparison participants. This is similar to a study by Vella et al.(2018) in which autistic participants made more advantageous deck selections in the final block of the IGT and longer stretches of advantageous choices (while making poorer quality decisions on the Cambridge Gamble Task). A similar study with a gambling task found that cognitive flexibility (here scores on the Dimensional Change Card Sort) positively correlated with gambling task performance (Yu et al., 2021). Five studies utilised a gambling task (Chen et al., 2021; Faja et al., 2013; Johnson et al., 2006; Latinus et al., 2019; South et al., 2014). These studies reported mixed findings. Both Faja and colleagues, and Johnson and colleagues, found that autistic children made comparable choices from the safe and advantageous decks, yet their explicit knowledge of their task performance was poorer than comparison participants’ (Faja et al., 2013); Johnson et al. (2006), however, found that the performance of autistic adolescents was also less consistent (and Latinus et al. (2019) found more errors in rule-following for emotional, but not colour or identity trials). Although not assessing explicit knowledge, South and colleagues reported that autistic adolescents improved in their rate of learning compared to comparison adolescents (despite initially equivalent rates of learning and overall performance); a similar study found that autistic traits were not correlated with learning rate (but higher autistic traits were correlated with poorer performance in the volatile condition; Goris et al., 2021). Yet, Chen et al., (2021) found that high-trait individuals performed worse on the final two blocks compared to low-trait individuals (despite equivalent performance on the first four blocks). Of the five remaining studies in this category (11.6%), two studies reported more preservative errors in autistic versus comparisons adolescents (Chantiluke et al., 2015; Yeung et al., 2016), and one study reported a lower tendency to select low-frequency losses in autistic compared with comparison participants (Mussey et al., 2015). Larson et al. (2011) found that there were no differences in reaction times between autistic and comparison participants, but the autistic participants had lower peak-to-peak feedback-related negativity (FRN) activity compared to comparison participants (other electroencephalogram [EEG] activity was comparable between groups). Finally, one study by Sahuquillo-Leal et al. (2019) did not directly compare autistic and comparison participants on performance, but did find that confidence was more closely related to performance in the comparison group than in the autism group.
Notably, of the studies that did report differences in learning between autistic and comparison participants, most did not find a main group difference in implicit (behavioural) proxies of learning, such as learning-related skin conductance (Faja et al., 2013; Johnson et al., 2006), implicit behavioural adjustments by error-related P3 responses (Zwart et al., 2017) and response time slowing following errors (Doenyas et al., 2019; Nicholson et al., 2019). For example, Doenyas and colleagues (2019) showed that autistic participants were less aware of their performance accuracy than comparison participants when they gave explicit verbal reports, whereas post-error slowing (implicitly becoming more cautious about your decision-making after having made an error) did not differ between autistic and comparison participants (Doenyas et al., 2019).
Together, these studies suggest that autistic people perform comparably to matched control groups on reward-learning paradigms. In line with our prediction, we find a distinction between decision made explicitly or implicitly. Of the identified studies that did report differences in learning, seven studies report no corresponding differences in implicit (behavioural) proxies of learning (e.g. Doenyas et al., 2019). Based on this, there is not sufficient evidence to conclude autistic people decide atypically on reward-learning paradigms (see Supplemental Table 1b).
Metacognition
Metacognition is usually experimentally measured as the ability to monitor behaviour (e.g. choice accuracy) with subjective assessments (e.g. self-confidence), with ‘better’ metacognitive ability being when confidence is more tightly coupled to actual performance. Our search revealed 10 studies that compared metacognitive ability between autistic and comparison participants (N = 340, mean age approximately 18.8; and N = 539, mean age approximately 18.3, respectively, for ASC and comparison samples; of those reporting IQ all studies included participants with an IQ > 70). One study included a childhood sample, four included adolescent samples, five included adult samples, the mean age was not reported for one sample, and two included samples from more than one age group. Three out of all 13 metacognition studies (23.1%) measured metacognitive ability as an association between objective accuracy and self-reported confidence, and two of these found lower metacognitive ability in autistic participants than non-autistic participants (Doenyas et al., 2019; Grainger et al., 2016b). One study (7.7%) tested metacognition as the association between subjective estimations of being in control of actions and the extent to which actions are actually under experimental control and found that autistic participants considered themselves more ‘in control’ of their decisions than they objectively were, compared to non-autistic participants (Zalla et al., 2015). Two studies (15.4%) investigated metacognition as the ability to estimate emotional states, and showed a reduced ability to reflect upon and judge internal bodily and emotional states in autism (Hall et al., 2007; McMahon et al., 2016). One study used a Fisher task (participants are asked to infer which lake a fisher had caught a fish in based on (a) the fisherman’s preference for lake size and (b) the proportion of red and black fish in each lake), and found (in a sample of both autistic and comparison participants) that autistic traits were not associated with circular inference (Angeletos Chrysaitis et al., 2021).
Three papers in this category (23.1%) tested both explicit and implicit metacognitive ability, as an association between objective accuracy and self-reported confidence or behavioural indicator of confidence, respectively. Two of these articles found that autistic people had reduced explicit, but not implicit, metacognitive ability (Carpenter et al., 2019; Nicholson et al., 2019). One article found that overall calibration (agreement between correct responses and confidence) was equivalent between the autistic and comparison groups; but autistic participants showed lower initial confidence and tended to display less under- and over-confidence than comparison participants (Brewer et al., 2022). Finally, three articles (23.1%) found additional support for intact implicit uncertainty monitoring in autism. These studies found no group differences in how much evidence is sampled by autistic and comparison participants prior to deciding (an implicit proxy of decision cautiousness; Brosnan et al., 2014; Jänsch & Hare, 2014), or response to initial uncertainty, estimated volatility and belief updating (although cluster analysis did show that, in the high volatility cluster, autistic participants showed greater volatility than comparison participants; Kreis et al., 2021).
Together, these studies suggest that autistic people have lower metacognitive ability in evaluating and reflecting upon decisions, especially when the decision also involves subjective values, such as emotional states (Hall et al., 2007; McMahon et al., 2016). Importantly, both studies that report comparable metacognitive ability between autism and comparison groups did not match for first-order performance (Carpenter et al., 2019), which is a known artefact of measuring metacognitive ability (Fleming & Lau, 2014). Based on these studies, we conclude that some autistic people may have altered metacognitive evaluations compared with non-autistic people (see See Supplemental Table 1c).
Value-based decision-making
Value-based decisions are subjective decisions that have no clear right or wrong answer. Instead, a decider is asked to rely on introspection to select the option that best matches their personal preferences. Our search revealed 27 studies of value-based decision-making in autistic and comparison participants (N = 738, mean age approximately 19.0 and N = 1547, mean age approximately 18.5, respectively, for ASC and comparison samples; of those reporting IQ all participants had an IQ > 70). Three studies included a childhood sample, 5 included an adolescent sample, 17 included an adult sample, 1 study did not report the age of the participants and 1 included a sample from multiple age groups. To discuss the variety of value-based decision-making paradigms in an informative manner, we split the paradigms into three subtypes described below (see Supplemental Table 1d).
Discounting tasks
Value-based decision-making can be tested with discounting tasks. On these paradigms, participants make a number of decisions between varying numbers of monetary rewards and varying amounts of an aversive aspect of the decision, such as how long participants need to wait before the reward will be received (temporal discounting tasks), how risky the decision is (risk-taking tasks) or how much effort needs to be exerted to obtain the outcome (effort discounting tasks). We reviewed seven discounting tasks (25.9% of all value-based studies), of which three investigated temporal discounting (Carlisi et al., 2017a; Demurie et al., 2012; Murphy et al., 2017). Two of these studies found that autistic participants discounted (i.e. gave lower value to) rewards more for longer waiting delays than comparison participants (Carlisi et al., 2017a; Murphy et al., 2017) and found that this effect was reflected in increased ventromedial prefrontal cortex (vMPFC) activity during delayed versus immediate choices (Carlisi et al., 2017a; Murphy et al., 2017). The two risk-taking studies showed that autistic participants were more risk-averse than comparison participants (Gosling & Moutier, 2018; Wu et al., 2018). However, Gosling and Moutier (2018) found autistic participants were risk-seeking when it was rational to do so; and Wu et al. (2018) found that autistic participants made more consistent choices for equal-value framing effects than comparison participants. These two studies suggest autistic people can overcome risk-aversion when it is beneficial to do so (as in both cases, risky choices were equivalently, or less preferred by autistic people when there was nothing to gain). The two remaining effort-discounting tasks found that autistic participants were less sensitive to reward for themselves (Damiano et al., 2012) and others (Mosner et al., 2017), which was reflected in higher effort exertion for the same outcome compared with comparison participants.
Framing studies
Eleven articles (40.7% of all value-based studies) investigated how value-based decisions change when choice-options are presented with decision frames, for example, distinct emphases or with a third choice-irrelevant option. These articles consistently found that framing effects were less pronounced in autistic than in comparison participants (Bellamy et al., 2021; De Martino et al., 2008; Farmer et al., 2017; Fujino et al., 2017, 2019, 2020; Hooper et al., 2019; Klapwijk et al., 2017; Panasiti et al., 2016; Shah et al., 2016). However, Fujino et al. (2017) found, in an economic decision-making paradigm with ambiguous and risky contexts, that autistic participants were less sensitive to context change in the ambiguous condition, compared to comparison participants (with no differences in the risky condition, similar to Gosling & Moutier, 2018). Importantly, Shah et al. suggest that these effects are related to individual differences in alexithymia – the ability to identify and describe one’s own emotions – in comparison, but not in autistic, participants (Shah et al., 2016). This latter finding is consistent with Damiano and colleagues (2012), but also suggests that restricted access to emotional information does not impair decision-making in autistic participants as much as in comparison participants. A final framing study showed that self-referential beliefs in autism are less affected by negative arousal (Kuzmanovic et al., 2019).
Social decision-making
Social decision-making paradigms study decisions made in (artificial) social settings. Our search revealed nine social decision-making paradigms (33.3% of all value-based studies). One study used a measure of autistic traits and found that social decisions made by those with higher autistic traits are not disadvantageous, but better suited to situations in which co-operation is less advantageous than working alone (Craig et al., 2017). The remaining studies compared autistic participants with comparison participants. Chambon et al. (2017) showed that autistic participants’ social decisions are less affected by prior expectations than those of comparison participants (Chambon et al., 2017). Woodcock et al. (2020) employed an ultimatum game, on which participants are asked to make or accept versus reject equitable (i.e. half-half) or unequitable (i.e. lower than half) monetary offers from an endowment that needs to be shared with a play partner. They showed that autistic participants made less equitable offers than comparison participants, which was explained by lower perspective-taking ability in autistic participants. In addition, they found that autistic participants reported more negative frustration after having received an unfair offer than comparison participants. Intriguingly, while such frustration following a negative offer predicted greater rejection of unfair offers in the comparison group, this was not the case in the autism group (Woodcock et al., 2020). This is similar to a study by Mantas et al., (2022) who reported that – in a prisoner’s dilemma task – autistic participants were less able to predict their opponents’ choices; however, there were no differences between groups (autistic and comparison) for total score earned over the course of the game or co-operation levels.
Van Hoorn et al. (2017) showed that autistic and comparison participants responded equivalently to social influence. Specifically, both groups donated more tokens to the group when exposed to prosocial information, and conversely, donated fewer tokens when exposed to antisocial information. A study by Komeda et al. (2016) found that autistic participants were more likely to rate a vignette character as good (when presented with bad characteristics, good behaviours and the event had a good outcome) compared to comparison participants; yet, when the judgements were about the pleasantness of a vignette character, autistic participants were less likely to take the outcome of the event into consideration than comparison participants. The two final articles highlight the impact of perspective-taking on the social decisions made by autistic participants (Balsters et al., 2017; Large et al., 2019). These studies leveraged advice-taking paradigms, in which an independently made decision can be updated on the basis of social advice. They found that autistic participants with greater perspective-taking difficulties took less advice from others than comparison participants (Large et al., 2019).
Together, these studies suggest that autistic people choose differently from non-autistic people on decisions that involve reflecting upon subjective values. Theoretically, subjective decision-making requires a decider to act upon their internal preferences and priorities, which is a metacognitive process (De Martino et al., 2013). Interestingly, across the distinct value-based decision-making paradigms investigated here (discounting, framing and social decision-making tasks), more complex decisions are increasingly more dissimilarly made by autistic and comparison participants, suggesting value complexity is driving any reported decision-making atypicalities between autistic and non-autistic participants. Based on this, we conclude autistic people decide differently from non-autistic people on value-based decision-making paradigms (see Supplemental Table 1d).
Discussion
We conducted a narrative literature review to test whether decision-making in intellectually able autistic individuals is atypical, and if so, what type of decision-making is most challenging. We found that differences in decision-making between autistic and comparison participants were most pronounced when the decision concerned metacognitive evaluations (e.g. Carpenter et al., 2019; Doenyas et al., 2019; Grainger et al., 2016a, 2016b; Nicholson et al., 2019) and/or value-based preferences (e.g. Carlisi et al., 2017a; Gosling & Moutier, 2018; Murphy et al., 2017). Perhaps surprisingly – given the amount of theoretical and empirical work on altered perception and reward learning in autism – there was no consistent evidence in favour of altered decision-making in perception or reward-learning paradigms in autism compared with comparison groups (e.g. Barnes et al., 2008; Brown et al., 2010; Carlisi et al., 2017b).
Another key feature of our results is that decision-making difficulties in autism were more pronounced at, or limited to, the explicit (verbal) level, such as the way in which choice preferences are reflected upon or evaluated, but much less evident in the (implicit) cognitive process of deciding itself. More specifically, of the nine studies that did report differences in explicit (verbal) aspects of learning and decision-making, seven studies did not find any group differences in implicit decision-making measures (Chantiluke et al., 2015; D’Cruz et al., 2016; Johnson et al., 2006; Larson et al., 2011; McPartland et al., 2012; Nicholson et al., 2019; Solomon et al., 2015).
How do these results fit in with existing theories of decision-making in autism? Bayesian or predictive coding accounts of decision evidence integration in autism suggest that decision-making difficulties in autism can be explained by an imbalance between expectations and new sensory evidence (Pellicano & Burr, 2012; Van de Cruys et al., 2014). Specifically, it has been proposed that autistic people rely more on factual (new) evidence and less on what can be expected based on similar (prior) experiences. This atypical balancing of expectations and sensory evidence can explain some of the differences in value-based decision-making between autistic and comparison participants reported here. These include, decreased loss-aversion and attraction effect on how subjective choices are evaluated (De Martino et al., 2008; Farmer et al., 2017), a tendency towards risk-seeking when the environment demands it (Gosling & Moutier, 2018) and how (little) alternative cues that usually elicit strong expectations (e.g. facial traits and emotions) impact decision-making in autism (Damiano et al., 2012; Hooper et al., 2019; Klapwijk et al., 2017). In these cases, the priors are default options that can be re-evaluated on the basis of changed circumstances or contexts. These six studies all unequivocally suggest that autistic participants update their priors less in response to temporary circumstances than do comparison participants. However, importantly, Shah et al. point out that a reduced impact of these prior expectations on value-based decision-making in autism can be explained by an atypical interpretation of those expectations, and not necessarily an absence of those expectations (Shah et al., 2016). This point potentially highlights the metacognitive difficulties that drive these differences. This view was confirmed in several of our reviewed studies; for example, autistic (versus comparison) participants were less aware of their objective accuracy when they rated their confidence verbally, but despite this reduced insight, still behaviourally adapted their decisions to account for ongoing errors (Carpenter et al., 2019; Doenyas et al., 2019; Kuzmanovic et al., 2019; Nicholson et al., 2019). The reduced impact of changing circumstances here is restricted to participants’ awareness that the circumstance has changed, but is still expressed in their behaviour. Together, these studies suggest that decision-making difficulties in autism mostly arise at a metacognitive level, as a difficulty with understanding and reasoning why one decided one way or another, but that the decision-making process itself may not necessarily be different in autistic versus non-autistic participants.
We believe our results have implications for research on decision-making in autism. Autism has traditionally been associated with difficulties with theory of mind or perspective-taking. Later accounts have suggested that ‘reading one’s own mind’ may require the same cognitive processes as tracking others’ beliefs and desires, as tested in theory of mind paradigms (Williams & Happé, 2009, 2010). Yet, only a handful of empirical studies have broadened this view into suggesting that self-evaluative metacognition is also affected in autism (Balsters et al., 2017; Carpenter et al., 2019, 2020; Doenyas et al., 2019; Grainger et al., 2016b). Our work suggests that exactly this aspect of cognition may explain most of the reported decision-making differences between autistic and comparison participants. This interpretation could also explain why, in general, fewer differences in decision-making were found in these common lab-based paradigms than would be predicted based on the difficulties people on the autism spectrum describe in everyday decision-making. Everyday instances of (social) decision-making are ambiguous and convoluted, and often provide unreliable feedback or information (Jänsch & Hare, 2014; Robic et al., 2015); they thus rely to a greater extent on metacognition than laboratory paradigms. Our results suggest that it is this feature of everyday decision-making that autistic people struggle most with. Future studies should develop laboratory decision-making paradigms that are more predictive of everyday decision-making challenges (Burgess, 2006; Nick et al., 2003). Based on our results, such paradigms would advance the development of decision-making supports for any areas of everyday decision-making in which autistic people seek assistance or scaffolding (Luke et al., 2012).
In the introduction to this article, we discussed findings that autistic people often struggle with independence (in terms of living status and employment), and we suggested that decision-making may be associated with these difficulties. While it is plausible that cognitive factors (such as metacognition) may be related to these real-world outcomes, it is also vital to consider other plausibly related factors. For example, autistic people often report experiencing stigma (Perry et al., 2020), and a recent study found that autistic people were more likely to be perceived as deceptive or dishonest compared to neurotypical people (Lim et al., 2021). Moreover, autistic people face barriers to accessing physical healthcare (Mason et al., 2019), and face unmet needs in a range of areas (Lai & Weiss, 2017), including employment (i.e. stigma, not understanding ASC or communication difficulties; Black et al., 2020). Thus, intra-personal factors like metacognitive ability can only ever be part of the explanation for the real-world outcomes of autistic people.
It is important to note the limitations of the studies included in this review. The most significant limitation was that of the representativeness of the samples. Samples were often largely – or exclusively – male, and the vast majority of included studies set a minimum IQ. This is problematic given recent estimates of the male-to-female ratio in autism as 3:1 (Loomes et al., 2017), and a significant proportion of autistic people have a co-occurring intellectual disability (estimated to be around 55%, Charman et al., 2011). This suggests that the results of some empirical studies reviewed here may not generalise to the wider autistic population. In addition, there was no evidence of user involvement in any of the reviewed studies. Research indicates that researchers and autistic people (or their relatives) have different views on what ought to be researched (e.g. applied over basic science, and research into mental health and well-being; Frazier et al., 2018; Pellicano et al., 2014); moreover, there is growing evidence that involving autistic people in the design of studies is feasible and helps to better illuminate the phenomena of interest (e.g., challenging the assumption that autistic people are not socially interested; Jaswal & Akhtar, 2018). On the basis of our results, we would recommend the field to increase autistic individuals’ co-design and test more representative autistic populations (Pellicano & den Houting, 2022) to understand if the described findings are replicated with autistic people with intellectual disability.
Strengths and limitations
It is also important to note the limitations of the present review. First, this review provides a general overview of trends and findings from studies of decision-making in autism. We aimed to be as inclusive as possible in our design and reviewed experimental results from N = 2712 autistic and N = 3189 comparison participants on approximately 60 different cognitive paradigms. This approach allowed us to make broad conclusions about autistic decision-making performance on average. However, ours is not a typical review in the sense that describing the results of each article in detail would be prohibitively lengthy. Instead, we decided to highlight those results that are informative to the main theoretical framework presented. By doing so, some intriguing findings and details from some of the studies have inevitably been left out. We therefore strongly recommend readers to refer to the Tables and the original publications for greater detail.
Second, we did not pre-register this review, although the main authors did pre-register the hypothesis of altered metacognitive, but not first-order, decision-making in autistic compared with non-autistic participants for a related empirical research project (https://osf.io/u6ecx/). A future review of this area should pre-register the review protocol to better align with the Open Science movement.
Third, while our search strategy led to a large number of ‘hits’ and identified many relevant articles, the terms used did not bring up every article using some of the paradigms and tasks above. We did not, for example, include every perceptual judgement study with autistic participants, only those whose title or abstract contained at least one of our search terms related to decision-making. The interested reader is therefore directed to more specific reviews to gain a full picture of the literature on any particular task or paradigm (e.g. Zeif & Yechiam, 2020).
We end by cautioning that, just because autistic people in most of the reviewed studies generally scored well on structured laboratory paradigms of decision-making, this does not mean that they are not faced with enormous challenges in managing daily choices and living activities (as reflected in some first-person accounts). Metacognition is involved in navigating daily life and is crucial for assessing how to meet one’s needs and preferences (be they social, financial, vocational, etc.). Our review suggests that autistic people may face difficulties with navigating daily life decisions at this metacognitive level, even though implicit decision-making is often unaffected. This suggests that future studies may nuance autistic people’s support to take into account any challenges with metacognition. Our results suggest that a need for decision-making support in this population may be subtle and often go unnoticed, but that this need may, nonetheless, be far-reaching.
Supplemental Material
sj-docx-1-aut-10.1177_13623613221148010 – Supplemental material for Decision-making in autism: A narrative review
Supplemental material, sj-docx-1-aut-10.1177_13623613221148010 for Decision-making in autism: A narrative review by Elisa van der Plas, David Mason and Francesca Happé in Autism
Supplemental Material
sj-docx-2-aut-10.1177_13623613221148010 – Supplemental material for Decision-making in autism: A narrative review
Supplemental material, sj-docx-2-aut-10.1177_13623613221148010 for Decision-making in autism: A narrative review by Elisa van der Plas, David Mason and Francesca Happé in Autism
Footnotes
Author contributions
All authors conceptualised the project. E.v.d.P. and D.M. collected the data; and E.v.d.P., D.M., and F.H. wrote the article.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: EvdP and FH were supported by the Mental Health and Justice Project, which is funded by a grant from the Wellcome Trust (203376/2/16/Z). DM was not supported by any specific funding.”
Supplemental material
Supplemental material for this article is available online.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
