Adults with High Autistic Traits Are Reluctant to Trade Accuracy for Monetary Reward: A Probabilistic Reasoning Experiment
Information processing has been considered as a cause of dysfunctions in autism spectrum disorders (ASD; e.g., van de Cruys et al., 2014). Different empirical studies, however, provided seemingly contradictory findings on whether and how autistic people differ in probabilistic reasoning (Brosnan, Chapman, & Ashwin, 2014; Jänsch & Hare, 2014). What were confounded in previous studies are the abilities to judge probabilistic information and to use the information for decision-making.
To investigate how people with various autistic traits are driven by accuracy (the goal of judgment) and monetary reward (the goal of decision-making) to different extents by introducing monetary reward into a probabilistic reasoning task. In certain experimental conditions, accuracy and reward formed the opposing forces: pushing expected accuracy over a point would reduce expected reward.
We recruited forty-eight 18- to 28-year-old college students and measured their autistic traits with Autism-Spectrum Quotient (AQ) on a 4-point scale (0, 1, 2, and 3). Participants played a computerized Bead Task, in each trial of which they saw two bead jars. One of the jars had a blue-to-pink beads ratio of 80%:20% or 60%:40%, and the other had the opposite ratio. Participants were told that one jar had been randomly chosen and their task was to judge which jar was chosen. For each correct judgment, participants could win up to 10 game points. Before judging, they could draw 0 to 20 beads sequentially from the chosen jar and stop sampling at any time. This information gathering could incur a cost: 0, 0.1, or 0.4 points for each bead drawn. There were 288 trials with three cost conditions blocked and two ratio conditions randomly mixed in each block.
We divided the participants evenly into three groups based on AQ scores (IQ-matched) and performed an ANOVA (Group × Cost × Ratio) on the jump-to-conclusion rate (the percentage of the trials with no more than two beads drawn). Jump-to-conclusion was not an optimal choice when the sampling cost was low, but could be more rewarding when the sampling cost was high. We found a three-way interaction (Figure 1): in the 80%:20% ratio condition where accuracy and reward were rarely in conflict, all the groups adequately adapted to the change of the cost. However, in the 60%:40% ratio condition, the high-AQ group was less likely than the other two groups to trade accuracy for expected reward when the cost climbed.
The distribution of the reaction time was bimodal. Consistent with our findings in the number of beads drawn, the reaction time gradually shifts from the shorter to the longer in the low- and middle-AQ groups with the drawing cost rising, while the shift in the high-AQ group seemed somewhat cost-resistant (Figure 2).
In sum, we found that autistic traits influence how people use probabilistic information for decision-making. People with high autistic traits are highly driven by the accuracy of judgment even though pursuing the accuracy means costing their expected reward, while people with low autistic traits are more adaptively driven by expected reward.