Model-based analysis of EEG data. a, Model comparison results comparing blockwise surprise Ib, to the traditional categorical model, KL surprise, and surprise without forgetting Ie. We report the log Bayes factor (or likelihood ratio), so that positive values > 3 indicate evidence (20:1 odds) in favor of Ib (indicated by dotted line). b, Posterior density of GLM parameters weighting the explanatory variable containing blockwise surprise at the group level. Error bar indicates SEM. c, Posterior densities showing influence of surprise on single-trial P300 estimate for each participant. Error bars indicate STD.