Fig. 6: Overview of the main analyses for experiment 4.

In this experiment participants learned about an out-group (i.e., fashion models) instead of the regular in-group (i.e., students). Additionally, three models were added to our original set of five models to capture stereotypic inclinations (STE). These stereotypic models have a darker colour in the figures and are indicated by STE in the model names. a As hypothesized, the best fitting model was Model 5-STE [Fine Granularity & Stereotypic RP] (n = 29). This model uses the expected stereotypical self-ratings from models as a reference point and fine granularity for generalization. b Simulated data for the best performing model (n = 29). Contrary to participants’ data the best performing model was model 1 [No Learning]. This indicates that participants used too complex a strategy for learning about the fashion models (see Supplementary Fig. 8 for details). c Both plots display the average absolute PEs over time ± SEM. Top) Participants’ data show a decrease in the PEs over time (ρ:−0.307, least squares line (red)), an indication of learning over time. Bottom) Simulated data from the best fitting model (Model 5-STE) show a large decrease in PEs over time (ρ:−0.788). d Of the three regressors (representing: 1 RW learning, 2 Coarse granularity, 3 Fine granularity), only the third regressor was significant (one-sided t-test), regressor 1: t(28) = 2.0906, p = 0.9771, regressor 2: t(28) = 4.3546, p = 0.9999, regressor 3: t(28) = −2.6794, p = 0.0061, indicating participants (n = 29) used fine granular representations during learning. Individual data points represent participants’ parameter estimates. Boxplots summarize these parameter estimates (median (middle line), 25th, and 75th percentile (box), most extreme points not considered outliers (whiskers), outliers (1.5 times interquartile range) indicated with + signs). Due to the high correlations between regressors one should be careful when drawing conclusions based on these regressors. [One-sided t-test; * indicates p < 0.05, ** indicates p < 0.001, no correction for multiple comparisons]. CG coarse granularity, FG fine granularity, RP reference point, # number of, PEs prediction errors, SEM standard error of the mean, LSLine least squares line.