People With Autism Spectrum Conditions Make More Consistent DecisionsPsychological Science

People with autism spectrum conditions (ASC) show reduced sensitivity to contextual stimuli in many perceptual and cognitive tasks. We investigated whether this also applies to decision making by examining adult participants’ choices between pairs of consumer products that were presented with a third, less desirable “decoy” option. Participants’ preferences between the items in a given pair frequently switched when the third item in the set was changed, but this tendency was reduced among individuals with ASC, which indicated that their choices were more consistent and conventionally rational than those of control participants. A comparison of people who were drawn from the general population and who varied in their levels of autistic traits revealed a weaker version of the same effect. The reduced context sensitivity was not due to differences in noisy responding, and although the ASC group took longer to make their decisions, this did not account for the enhanced consistency of their choices. The results extend the characterization of autistic cognition as relatively context insensitive to a new domain, and have practical implications for socioeconomic behavior.

People with autism spectrum conditions (ASC) often show atypical performance on tasks that require processing of local information independently of its context (Behrmann, Thomas, & Humphreys, 2006; Frith, 1989; Happé & Frith, 2006). For example, people with ASC are better than control participants at finding figures embedded in complex shapes, their visual search is less affected by the number and similarity of distractors, and they often fail to take semantic context into account when pronouncing homographs (see Happé & Frith, 2006, for a review). Nonclinical samples scoring high on measures of autistic traits display a similar pattern of performance (e.g., M. E. Stewart, Watson, Allcock, & Yaqoob, 2009). This reduced impact of context may reflect an inability to integrate information into a coherent whole (Frith, 1989) but may also be understood solely in terms of a superior ability to process local information (Plaisted, Saksida, Alcántara, & Weisblatt, 2003).

We investigated whether the reduced context sensitivity that characterizes ASC extends to decision making. Decision making is a fundamental cognitive operation that has received relatively little attention from autism researchers (Davis & Plaisted-Grant, 2015; Luke, Clare, Ring, Redley, & Watson, 2012). Most previous studies have focused on how people with ASC represent and evaluate probabilities and rewards, often using tasks in which the decision maker must learn the payoffs and probabilities of different options by making a series of choices and receiving feedback (e.g., Mussey, Travers, Klinger, & Klinger, 2015). We took a different approach by examining whether autistic traits correlate with altered context sensitivity in a riskless choice task, in which the participant simply selects the best alternative on the basis of explicitly stated attribute values.

Conventional accounts of rational choice dictate that a person’s preference between two items be independent of the other options on offer: If one prefers salmon to steak, this should not change just because frogs’ legs are added to the menu (Luce & Raiffa, 1957). However, the choices of neurotypical adults are heavily influenced by the composition of the choice set; rather than being based on an independent assessment, the attractiveness of a given option depends on how it compares with the other values that are simultaneously present (Huber, Payne, & Puto, 1982; Simonson, 1989; Tversky, 1972). One of the most striking examples of this phenomenon is the attraction effect, which arises when people choose between two options, A and B, that “trade off” two dimensions—for example, two USB drives that respectively have lower capacity but higher longevity and higher capacity but lower longevity. When the choice set includes a third, “decoy” option that is fractionally worse than A on both dimensions, people very rarely choose the decoy, but its presence boosts the tendency to choose A rather than B—and vice versa if the decoy targets option B (Fig. 1, top panels). This kind of context-induced preference reversal occurs in many domains (e.g., Farmer, El-Deredy, Howes, & Warren, 2015), has been extensively modeled (e.g., Trueblood, Brown, & Heathcote, 2014), and is used by marketers to influence consumer behavior (Ariely, 2009).

If the tendency of people with ASC to prioritize local information and to be relatively insensitive to the other elements of a stimulus array extends to decision making, then they should be less influenced by decoy options and make fewer context-induced preference reversals. Correspondingly, we hypothesized that adults with ASC would make more consistent choices—indicative of a more rational, independent valuation of alternatives—than would neurotypical adults. This possibility is important because choice consistency is regarded as normative in conventional economic theory, so reduced context sensitivity would provide a new demonstration that autism is not in all respects a “disability” (Baron-Cohen, 2000). More important, context effects on choice speak to the nature and basis of autistic cognition. Many studies of altered context sensitivity among people with ASC focus on perceptual tasks, such as pitch discrimination, visual search, and motion-coherence detection, with corresponding theoretical frameworks that emphasize “low-level” processes, such as enhanced perceptual discrimination or altered magnocellular sensitivity (see Happé & Frith, 2006, for a review). Altered preferences in a choice task involving verbally described consumer products would suggest the need for a broader characterization and integrated theorizing across levels and domains of processing (Davis & Plaisted-Grant, 2015; Pellicano & Burr, 2012). Finally, the possibility of reduced contextual influence has practical implications for the economic and social functioning of people with ASC: Attraction-effect decoys influence many real-world decisions (e.g., Doyle, O’Connor, Reynolds, & Bottomley, 1999), and reduced decoy sensitivity among people with ASC would affect their financial, consumer, political, and relationship choices.

Results Section: Choose Top of page Abstract Method Results << Discussion References ASC study: context effects among people with ASC Responses to each product pair were placed into one of four categories. Consistent choices were those in which the decision maker chose the same item on both presentations of a particular product pair. Attraction-effect preference reversals were cases in which the person’s selection switched when the decoy changed (the person chose A when the decoy targeted A and chose B when it targeted B). Non-attraction preference reversals were preference reversals in which the person chose the nontarget options on both presentations (e.g., the person chose A when B was the target and chose B when A was the target). Decoy selections were cases in which the person chose the decoy on one or both presentations of a given product pair. As the left panel of Figure 2 shows, the ASC group made more consistent choices than did the control group, t(203.9) = 5.15, p < .001, d = 0.60, 95% confidence interval (CI) = [0.34, 0.85], and showed fewer preference reversals—attraction-effect reversals: t(206.4) = 2.27, p = .024, d = 0.26, 95% CI = [0.01, 0.51]; non-attraction reversals: t(194.6) = 2.12, p = .035, d = 0.25, 95% CI = [0.00, 0.50]. They also made fewer decoy selections, t(294.9) = 5.46, p < .001, d = 0.53, 95% CI = [0.28, 0.78]. We had a surprisingly high proportion of females in our ASC sample, but the choices of male and female ASC participants differed very little (see Table S10 in the Supplemental Material), which indicates that the sample’s gender composition does not affect the representativeness of the results for the ASC population. Download Open in new tab Download in PowerPoint Our primary analysis tested these effects more rigorously with a series of mixed-effects logistic regressions that examined the effects of group on choice, controlling for age, gender, and cognitive ability (regressions were conducted using lme4 for R; Bates, Mächler, Bolker, & Walker, 2015). The fixed-effects predictors were participant group (control = 0, ASC = 1), ICAR score, gender (female = 0, male = 1), and age. In all analyses, the predictors were standardized prior to each regression, and we included random intercepts for participant and product pair and by-product random slopes for the effects of group, age, gender, and ICAR score, thereby allowing the effects of these variables to differ across product pairs; random effects were uncorrelated (Barr, Levy, Scheepers, & Tily, 2013; dropping the random slopes did not impair model adequacy and led to virtually identical results). All confidence intervals (CIs) are 95% Wald CIs; the regression coefficients are presented graphically in Figure S1 of the Supplemental Material. Our first regression contrasted consistent choices (coded 1) with preference reversals (coded 0). Participants in the ASC group made more consistent choices than did control participants, b = 0.200, 95% CI = [0.059, 0.341], p = .005, demonstrating reduced context sensitivity and a more rational decision-making style. There was little effect of age, b = 0.093, 95% CI = [−0.046, 0.231], p = .190; gender, b = 0.078, 95% CI = [−0.062, 0.217], p = .274; or cognitive ability, b = 0.096, 95% CI = [−0.062, 0.253], p = .233. We also contrasted decoy selections (coded 1) against all other choice types (coded 0). Decoy selections are rare and represent a form of noisy responding/inattention (the decoy is manifestly worse than the target option). This kind of error was negatively related to general cognitive functioning, b = −0.679, 95% CI = [−0.944, −0.414], p < .001, and was less prevalent among ASC participants than control participants, b = −0.371, 95% CI = [−0.640, −0.102], p = .007, a result consistent with the higher attention-check failure rate in the control group. Decoy selections did not depend on age, b = 0.007, 95% CI = [−0.234, 0.248], p = .954, or gender, b = 0.036, 95% CI = [−0.214, 0.283], p = .784. Finally, we contrasted attraction-effect and non-attraction preference reversals. Non-attraction-effect choices were very rare and, like decoy choices, they likely reflect noisy responding; they were more common among people with lower cognitive ability, b = 0.689, 95% CI = [0.357, 1.021], p < .001, but did not differ between the ASC and control groups, b = −0.037, 95% CI = [−0.381, 0.307], p = .834, and were unrelated to age, b = 0.014, 95% CI = [−0.295, 0.324], p = .929, or gender, b = −0.229, 95% CI = [−0.528, 0.069], p = .132. AQ study: context effects among high- and low-AQ groups in the general population The choice proportions for low-AQ and high-AQ participants are plotted in the right panel of Figure 2 and show an attenuated version of the pattern found in the ASC study: High-AQ participants made more consistent choices than did low-AQ participants, t(366.6) = 3.00, p = .003, d = 0.31, 95% CI = [0.11, 0.52], and showed fewer attraction-effect preference reversals, t(366.9) = 2.16, p = .031, d = 0.22, 95% CI = [0.02, 0.43], although both groups exhibited a clear context effect. The high-AQ group also made slightly fewer decoy selections, t(355.3) = 1.40, p = .163, d = 0.15, 95% CI = [−0.06, 0.35], and non-attraction preference reversals, t(328.8) = 1.80, p = .072, d = 0.19, 95% CI = [−0.02, 0.40]. When applying the regression analyses to this study, we included study version and its interactions with all other fixed-effects variables to examine the consistency of the findings across participant samples and stimulus sets. In the same way as for other variables, version was standardized prior to each regression, and the interaction terms were computed by multiplying the standardized predictors. (No random effects involving version or its interactions were computed because all participants and all product pairs arose only for one version.) None of the effects were modulated by version, and excluding version and its interactions made no difference to the results. (The full set of regression coefficients for these terms is shown in Fig. S7 in the Supplemental Material.) High-AQ participants were more likely to make consistent choices than were low-AQ individuals, b = 0.159, 95% CI = [0.011, 0.307], p = .035. In addition, consistent choice was positively related to age, b = 0.171, 95% CI = [0.037, 0.305], p = .013, and more common among males than females, b = 0.243, 95% CI = [0.110, 0.377], p < .001; it was also weakly related to cognitive ability, b = 0.148, 95% CI = [−0.006, 0.302], p = .060. The tendency to choose the decoy did not differ between the low-AQ and high-AQ groups, b = −0.033, 95% CI = [–0.241, 0.176], p = .758, but, as in the ASC study, it was negatively related to general cognitive functioning, b = −0.424, 95% CI = [−0.623,−0.224], p < .001; it was also slightly lower in males than in females, b = −0.201, 95% CI = [−0.398, −0.005], p = .044, but was independent of age, b = −0.075, 95% CI = [−0.287, 0.138], p = .490. Finally, non-attraction-effect choices were again very rare and more prevalent among people with lower cognitive ability, b = 0.538, 95% CI = [0.205, 0.870], p = .002. The proportion of non-attraction reversals was not affected by AQ, b = 0.013, 95% CI = [−0.271, 0.297], p = .928, or by gender, b = −0.220, 95% CI = [−0.508, 0.068], p = .134, or by age, b = −0.121, 95% CI = [−0.429, 0.186], p = .439. The demographics questions asked participants whether they had ever been diagnosed with an ASC (response options: “yes,” “no,” “prefer not to say”). Seven of the 194 (3.6%) participants in the high-AQ group answered “yes,” whereas none of the 176 in the low-AQ group did so. Excluding these participants made little difference to the estimated coefficients, but the confidence intervals increased, including the confidence intervals for the effect of participant group on the tendency to make consistent choices versus preference reversals, b = 0.155, 95% CI = [0.006, 0.304], p = .041. A further 13 high-AQ and 2 low-AQ participants selected “prefer not to say”; excluding these participants led to a weaker effect, b = 0.123, 95% CI = [–0.026, 0.272], p = .107. Thus, independently of the effects of age, gender, and general cognitive performance, people scoring high on the AQ made more consistent choices than did those with low AQ scores. The effect was weaker than in our ASC study, may reflect the presence of people with ASC in the high-AQ sample, and could have been driven by different mechanisms (Gregory & Plaisted-Grant, 2016), but it provides converging evidence for an association between autistic traits and a reduction in context sensitivity during choice. Additional analyses We probed three possible contributors to the enhanced rationality of the ASC group’s choices and the high-AQ group’s choices: a drive for greater internal consistency, a reduction in noisy responding, and a slower, more deliberative decision-making style. Full results for these additional analyses, including regression coefficients for demographic control variables, are provided in the Supplemental Material (Figs. S2–S6 and S10–S14). First choices One possible reason for a reduction in context-induced preference reversals is that people remember their own past choices and strive to be consistent. If the ASC group had better memory or a stronger drive for consistency than the control group, then a change in the attributes of the decoy between successive presentations of a product pair would have less effect on their choices, as we found, but this would apply only to within-subject preference reversals: The memory and consistency mechanisms would reduce the ASC participants’ tendency to switch their preference when the context changed, but they would be just as susceptible as control participants to the effects of the decoy when they first encountered a given pair of products. However, analyzing responses to the first occurrence of each product category revealed the same pattern as our main analysis: Decoy selection was less common among ASC participants than among control participants, b = −0.410, 95% CI = [−0.742, −0.078], p = .016, and, more important, people with ASC were less likely than control participants to choose the target item rather than the competitor, b = −0.131, 95% CI = [−0.228, −0.034], p = .008. The ASC group were therefore less influenced than control participants by the decoy even when they had never seen the competing options before. This pattern indicates a reduced influence of local context rather than an effect driven by memory or need for consistency. Notably, applying the same analysis to the data from our AQ study revealed no effect of AQ group on the tendency to choose the decoy, b = −0.056, 95% CI = [−0.328, 0.217], p = .689, or on the tendency to choose the target rather than the competitor, b = 0.001, 95% CI = [−0.103, 0.105], p = .980. Thus, consistent with our primary analysis, this analysis indicates that the difference in context sensitivity between people with ASC and control participants was more pronounced than the difference in context sensitivity between members of the general population with low and high AQ scores: The only effect of AQ group was on the tendency to make within-subject preference reversals, whereas, in comparison with the control group, the ASC group both made fewer preference reversals and showed a reduced tendency to select the target when a given stimulus pair was encountered for the first time. Noisy responding Next, we tested whether the enhanced consistency of the ASC group and the high-AQ group was driven by a reduction in random responding rather than an altered sensitivity to contextual stimuli (Pettibone, 2012). This analysis and the next were not part of our preregistered analysis strategy but seemed like useful explorations. We computed the proportion of decoy selections across the 20 test trials for each participant as an index of noisy responding (Pettibone, 2012) and reran the regression analysis that contrasted consistent with inconsistent choices with this noisy-responding measure included as a predictor. The results were virtually identical to those of the original analyses: The ASC individuals in the ASC study made more consistent choices (fewer preference reversals) than did the control participants, b = 0.188, 95% CI = [0.047, 0.329], p = .010, and the high-AQ group in the AQ study made more consistent choices than did the low-AQ group, b = 0.157, 95% CI = [0.010, 0.304], p = .037. The index of noisy responding only weakly predicted the tendency to make consistent choices rather than preference reversals (ASC study: b = −0.130, 95% CI = [−0.277, 0.017], p = .082; AQ study: b = −0.124, 95% CI = [−0.263, 0.014], p = .078). The reduced attraction effect among the ASC group and the high-AQ group is therefore unlikely to have been due to a change in random or inattentive choice. Response times Finally, we examined whether the greater consistency of the ASC group was due to slower, more deliberative de-cision making. We computed each participant’s mean re-sponse time (RT) across the 20 test trials and log-transformed these means to normalize the data and reduce the influence of extreme values. We examined the relationship between this measure and participant group with linear regression (including the same additional predictors as in our main analyses) and then assessed the association between RTs and choice behavior by rerunning our primary analysis with the RT variable added as a predictor. Participants with ASC had longer response latencies than control participants (ASC group: geometric mean = 24.3 s, 95% CI = [21.9, 27.1]; control group: geometric mean = 17.8 s, 95% CI = [16.3, 19.5]; b = 0.088, 95% CI = [0.014, 0.163], p = .020). However, the effects of ASC on choice behavior were not altered by including RT as a predictor: ASC participants remained more likely than control participants to make consistent choices, b = 0.189, 95% CI = [0.048, 0.331], p = .009, and were less likely than control participants to make decoy selections, b = −0.249, 95% CI = [−0.495, −0.001], p = .049. In addition, although participants with shorter RTs were more likely to make decoy selections and non-attraction preference reversals (decoy selections: b = −0.604, 95% CI = [−0.811, −0.396], p < .001; non-attraction preference reversals: b = 0.513, 95% CI = [0.172, 0.853], p = .003), there was no meaningful association between response latency and the tendency to make consistent choices rather than preference reversals, b = 0.079, 95% CI = [−0.059, 0.216], p = .263. Similar results emerged in the AQ study. The response latencies of the low-AQ and high-AQ groups were very similar to one another (low-AQ group: geometric mean = 17.7 s, 95% CI = [16.2, 19.3]; high-AQ group: geometric mean = 16.8 s, 95% CI = [15.7, 18.0]; b = −0.031, 95% CI = [−0.085, 0.024], p = .269), and although participants with shorter RTs were more likely to make decoy selections, b = −0.303, 95% CI = [−0.521, −0.083], p = .007, there was no association between response latency and the tendency to make consistent choices, b = −0.001, 95% CI = [−0.137, 0.135], p = .990, and controlling for response latency made very little difference to the effect of AQ group on choice consistency, b = 0.161, 95% CI = [0.013, 0.309], p = .033. In short, participants who rushed their decisions were more likely to make random responses, but there is no indication that the reduced context sensitivity of people with high AQ scores or with ASC was a consequence of their taking longer over their choices. Their increased decision time is consistent with research showing that people with ASC are reluctant to make decisions at all, and do not simply take a more deliberative strategy than neurotypical individuals (Luke et al., 2012).

Timothy J. Pleskac served as action editor for this article. Declaration of Conflicting Interests

The authors declared that they had no conflicts of interest with respect to their authorship or the publication of this article. Funding

This research was funded by the Wellcome Trust (Grant RG76641) and the Isaac Newton Trust (Grant RG70368). Supplemental Material

Additional supporting information can be found at http://journals.sagepub.com/doi/suppl/10.1177/0956797617694867 Open Practices

The design and analysis plans for the autism-spectrum-conditions study were preregistered at the Open Science Framework, and all data and materials for both studies have been made publicly available at the Open Science Framework; these files can be accessed at https://osf.io/7b9au. The data are also publicly available via Apollo, the University of Cambridge institutional repository, and can be accessed at https://doi.org/10.17863/CAM.7153; the materials are available from the same repository and can be accessed at https://doi.org/10.17863/CAM.7281. The complete Open Practices Disclosure for this article can be found at http://journals.sagepub.com/doi/suppl/10.1177/0956797617694867. This article has received badges for Open Data, Open Materials, and Preregistration. More information about the Open Practices badges can be found at http://www.psychologicalscience.org/publications/badges.