Table 1.
Characteristics of the participants.
Fig 1.
Subjects have to guess whether the majority of the (initially) hidden cards (left panel) are yellow or blue. They can gather information by selecting cards which then reveal their colour. The subjects are free to open as many cards as they want (middle panels) until they feel certain enough and declare for one colour (right panel). In the ‘fixed’ condition, there is no explicit cost associated with sampling. Correct choices are rewarded with 100 credit points, incorrect ones are punished with the same amount. In the ‘decreasing’ condition, sampling is associated with lower potential wins. For every opened card, subjects will win 10 points less; starting from 250 with no cards opened. False decisions, again, result in a loss of a fixed 100 points.
Fig 2.
Performance differences in information gathering.
Adolescents with OCD won significantly more points than controls across the entire task. This was mainly driven by an increase in points won during the decreasing condition (A). OCD patients opened significantly more boxes (B), but did not differ in performance accuracy (C; accuracy defined as the proportion of choices of the colour that was in the majority of currently opened cards). (D) Total draws to decision correlated significantly with self-reported general indecisiveness in OCD patients, linking task behavior to real-world perceived difficulties. *** p< = .001, ** p < .01, * p < .05, t p < .10.
Fig 3.
Choice probabilities (policy π) of the winning model.
For each stage of a game (x axis), the model predicts the average probability of choosing yellow colour (Py), blue colour (Pb), or continuing to sample (pink, PND). For the fixed condition (top), OCD patients (right) are more likely to continue sampling in conditions where the majority is not clearly apparent. In the decreasing condition (bottom), the probability of not deciding is lower than in the fixed condition in both groups (less pinkish colours). It is also noteworthy that OCD adolescents flexibly adapt their strategy in the decreasing condition, indicating against a general inflexibility in OCD.
Fig 4.
Increased decision thresholds in patients with OCD.
The OCD group showed increased decision thresholds in both conditions. The model’s inherent decision threshold showed a distinction between the OCD group and controls (A). When using the model to simulate data, we again found a marked increase in decision threshold (as indicated by the mean evidence difference at choice at each stage) for OCD patients (B). Note the growing evidence difference for the first few draws increases because the magnitude of the difference in evidence is limited by the actual samples (e.g., the difference after 2 draws can be no more than 2). The increasing errorbars in the decreasing condition at later stages shows that even simulated agents rarely sample beyond the 15th step in this condition.