Visualization, parameter recovery, and behavioral validation of the Hybrid model.

<p><b>(a)</b> To confirm that we can identify each parameter independently from other parameters within the Hybrid model, we conducted a parameter recovery analysis. To do so, we simulated artificial data and estimated the model on that data to see if the ‘true parameters’ used to simulate data could be identified. Correlations between true parameters and estimates indicate parameter recovery for that specific parameter (see <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1012602#sec006" target="_blank">Materials and Methods</a> for parameter recovery procedure). All parameters included in the model showed positive correlation between the true and estimated parameters, indicating that the Hybrid model could be recovered: inverse temperature (log transform; Pearson’s r = 0.81, P = 1.86e–11), risk preference (Pearson’s r = 0.86, <i>P</i> = 7.32e–14), OCU<sub>normalized</sub> (Pearson’s r = 0.76, <i>P</i> = 2.75e–09), ω<sub>follow</sub> (Pearson’s r = 0.88, <i>P</i> = 4.38e–15), ω<sub>oppose</sub> (Pearson’s r = 0.90, <i>P</i> = 1.46e–16), and ω<sub>utility</sub> (Pearson’s r = 0.73, <i>P</i> = 2.44e–08). <b>(b-d)</b> To confirm that the ω<sub>follow</sub> and ω<sub>oppose</sub> parameters are capturing an individual’s behavior, we compared these parameters to a model-agnostic measure of an individual’s tendency to conform to or oppose the choices of others. In healthy adults, we previously showed individuals who were more risk averse (or seeking) were more likely to conform when others chose the safe (or risky) choices but less likely to conform when others chose the risky (safe) choices. Moreover, in these individuals utilizing the OCU-based model, the tendency to conform to others’ safe choices was negatively correlated with the tendency to conform to others’ risky choices. Individuals following this pattern will fall close to the y = 1 − x line in <b>b</b>. Conversely, individuals who tend to use the OCU-free weights to a greater extent in the decision-making process will fall further away from this line, since they are more likely to follow or oppose others regardless of the type of information. Thus, to validate our model, we test if the orthogonal distance from the y = 1 − x line (the solid lines in <b>b</b> are shown as examples) correlates with the extent to which individuals use the OCU-free weights. As expected, there is a significant correlation between this model-agnostic measure of conformity tendency and both ω<sub>follow</sub> (<b>c</b>; r = 0.72, <i>P</i> = 4.07e−08, robust correlation) and ω<sub>oppose</sub> (<b>d</b>, r = −0.77, <i>P</i> = 1.33e−09, robust correlation) across all individuals, which shows that the model parameters accurately capture individuals’ behavioral choices. Each point represents an individual participant, and lines are the regressions between the indicated parameters.</p> <p>(PDF)</p>