Table 1.
Glossary of model terms and model description.
Fig 1.
Log Likelihood values of the model.
(A) Computed mean log likelihoods for each trial, coloured by dictator type. (B) Density plots of log likelihood values for each trial, coloured by dictator type. Coloured lines represent group means (C) Computed mean log likelihoods for each GPTS score quantile and clinical score cut off (Green et al., 2008). (D) Density plots of log likelihood values for each trial across each GPTS score quantile and clinical score cut off (Green et al., 2008). (E) The association between GPTS scores (minimum score = 32) and loglikelihood values. Dots = mean loglikelihood value across that score of the GPTS. Lines = 95% confidence intervals. The grey line in each plot (at -4.394) represents the loglikelihood that would be observed if the model was capturing behaviour by chance.
Fig 2.
(A) Generated Harmful Intent (HI) attributions for simulated participants at each level of paranoia at each trial within fair and unfair dictators. Dots represent the mean for each level of paranoia. Lines represent the 95% confidence interval. (B) Generated density distributions for simulated participant HI attributions (red) for each trial (1–6) within unfair and fair dictators for each level of paranoia. (C) Generated Self-Interest (SI) attributions for simulated participants at each level of paranoia at each trial within fair and unfair dictators. Dots represent the mean for each level of paranoia. Lines represent the 95% confidence interval. (D) Generated density distributions for simulated participant SI attributions (blue) for each trial (1–6) within unfair and fair dictators for each level of paranoia. (E) Smoothed linear splines for both simulated participant harmful intent and self-interest attributions by prior paranoia (minimum score = 32).
Fig 3.
Spearman rank correlations between uHI0, uSI0, uΠ, and η, and pre-existing paranoia and in-the-moment attributions.
(A) Quadratic fit for uncertainty of partner policies across the mean harmful intent attributions scored over 18 trials. (B) Quadratic fit for uncertainty of partner policies across the mean self-interest attributions scored over 18 trials. (C) Linear fit for uncertainty of partner policies across GPTS scores. (D) Quadratic fit of learning rate by mean attributions scored over 18 trials.
Fig 4.
(A) Gaussian Graphical Model of latent parameters and prior paranoid beliefs. (B) Moderated Network Model between latent parameters when moderated over prior paranoia from low to high Z-scores (-0.85–4). Red edges = negative association; green edges = positive association.
Fig 5.
Moderation effects of pre-existing paranoia on edges within the Moderated Network Model (Fig 4B).
The left panel displays the pairwise effects–the overall relationship between the parameters in the Gaussian Graphical Model of parameters—and the right panel shows the moderation effect of GPTS score on the pairwise effects–the influence of variable GPTS scores on the relationship between parameters. Both are shown with 95% confidence intervals of the bootstrapped sampling distributions. The number at the centre of the sampling distribution is the proportion of bootstrap samples in which a parameter has been estimated to be nonzero [26].
Fig 6.
Mean partner policy depending on attributes.
Okra: preference for returning a large amount to the participant. Blue / purple: preference for returning very little.
Fig 7.
Smoothed density distributions of the fitted parameters derived from the computational model.