Fig. 5.

Parameter relationships to false inference in models with and without affect.

A–D – These panels show the relationship of ‘delusion score’ (from 0 to 3, derived from the proportion of false inferences, and their certainty and incorrigibility: see Methods) to various parameters of different model simulations. Each dot is one simulation, with delusion score in colour, likelihood precision on the x axes and habit resistance, mood and policy precision on the y axes, in the same format, except for plotting mood in B and absolute mood in D. Only parameters with moderate-large effects on false inferences are shown. The models in A and C did not contain affective states, and those in B and D did. A and B used the ‘consistently trustworthy’ sequence employed in Fig. 2, C and D used variants of the ‘changing trustworthiness’ sequence used in Fig. 3, Fig. 4. Note that this means that if one extreme posterior over trustworthiness is consistently applied throughout, the maximum proportion of false inferences is 100% in the ‘consistently trustworthy’ sequence, and 50% in the ‘changing trustworthiness’ sequences.

E – This bar plot illustrates the relative effect sizes (standardised regression beta weights) of the different parameters and interactions between parameters on delusion score in the model containing affect and evaluating the ‘changing trustworthiness’ sequence (Fig. 5D).