Table 1. Comparisons of fits of self-esteem models .

Bayesian Information Criterion (BIC) measures are summed across all participants. Lower BIC values indicate a more parsimonious model fit. Mean squared error over self-esteem ratings indicates goodness of fit. k is the number of fitted parameters. Models 1–4 were fit to both choice behavior and self-esteem ratings and BIC measures comprise both the summed log likelihood of the model prediction over choice behavior and the summed log density of the model prediction over the self-esteem ratings. Models 5–6 were fit solely to the self-esteem ratings in order to allow for a fair comparison with a model without expectations (model 6), which by definition would not provide a good fit for the behavioral choice data. See Materials and Methods for details on the computational models.

Model	k	Mean r2	Median r2	BIC	BIC-BICmodel1
1: Learning and positive bias	9	0.31	0.27	−633	0
2: Learning, but no bias	8	0.29	0.25	−378	255
3: Correct initial beliefs about approval	7	0.25	0.22	409	1042
4: Separate term for expectations	10	0.34	0.32	−502	131

Model	k	Mean r2	Median r2	BIC	BIC-BICmodel1
5: Free initial beliefs about approval	5	0.32	0.31	−5671	0
6: Outcome valence only	7	0.23	0.18	−5581	90