Fig. 2: Reduced ΔES weighting replicated in transdiagnostic dimensional patient–control sample.
From: Indecision and recency-weighted evidence integration in non-clinical and clinical settings
a, Participants’ goal was to determine which of two possible stimuli was more abundant. They were presented with a sequence of draws until they decided to declare for a stimulus. Each draw consisted of five stimuli at a time in varying proportions, allowing for the sequences to be constructed so as to reduce the collinearity between the number of draws, current and previous evidence, which in turn enables us to study their specific representations in the brain. Here again, evidence strength (ES) at draw d is quantified as the cumulative difference in evidence for the 2 gems with respect to the gem that is more abundant at draw d (for example, here 4 blue gems minus 1 yellow gem constitutes an evidence strength of 3 in the first draw, and 8 yellow gems minus 7 blue gems yields an evidence strength of 1 in the third draw). Cumulative evidence strength in the previous draw ESd−1 is defined as the lagged ES by one draw, and evidence strength update ΔES is quantified as the signed difference between the cumulative ES at draw d−1 and draw d. The maximum number of draws varied between 4 and 8 in the short-horizon condition and between 10 and 14 in the long-horizon condition. The horizon condition was cued by the colour of the frame (different colours were presented in the task). Participants received 2 points for a correct response, lost 2 points for an incorrect response, and lost 1 point for non-decisions. Points were later translated into bonus payments. b, To pool across clinical and non-clinical groups, we reduced the dimensionality of 7 questionnaires using an exploratory factor analysis (N = 105). This revealed a 3-factor solution, where the second factor mapped primarily onto items in questionnaires assessing OCD (OCI-R and Padua Inventory-Washington State University Revision (PI-WSUR)). We used this OC factor to quantify individual differences across all participants, plotted here by categorical group, that is, for participants from the general population with low and high obsessive–compulsive scores (‘low C’ and ‘high C’, respectively), healthy controls (‘Control’), participants with generalized anxiety disorder (‘GAD’) and participants with OCD (‘OCD’). Boxplots show the median (centre line) and IQR; whiskers extend to the largest value within 1.5× IQR. c, The number of draws to a decision varied across trials and participants in the short- and long-horizon condition (N = 105). Note that the number of draws to a decision per se was not associated with the transdiagnostic OC factor (ρs = −0.117, P = 0.236), yet this is not surprising given the highly constraining horizon manipulation. d, The probability to make a decision was predicted in a GLMM from experimental factors as well as variability in the OC factor (N = 105, data are presented as beta coefficients ± s.e., scale adjusted to show main effects of experimental factors on the left and effects associated with individual differences in OC on the right y axis). Experimental factors (left y axis) comprised the number of draws, the cumulative evidence strength from the start of the game until the previous draw ESd−1, ΔES and the horizon condition (controlling for interactions of the horizon condition with all other experimental factors). More draws, higher previous and current evidence strength, and a short-horizon condition all increase the probability to make a decision (all P < 0.001). Individual differences in the OC factor (right y axis) are quantified as the main effect of the OC factor score and its interaction with the experimental factors. The GLMM shows no significant effect of the OC factor per se (P = 0.318), or differences in weighting the current number of draws (P = 0.155), the horizon condition (P = 0.320), or the previous evidence (P = 0.272) in making a decision. Conversely, it shows that those with higher OC factor scores weight ΔES less in making a decision (P = 0.013, highlighted by shaded area). e, Illustration of the association between ΔES weighting and the OC factor (using individual GLMs per participant; N = 92 after outlier exclusions were applied to all beta coefficients in the GLM, defined as any participant with any beta values 1.5× IQR above the third quartile or below the first quartile). f, OCD patients with stronger obsession symptom severity had more attenuated weighting of ΔES (ρs = −0.475, P = 0.012, N = 27). All tests are two-sided and not corrected for multiple comparisons.
