Extended Data Fig. 6: Control analyses ruling out maximum play surprise as an alternative explanation for game surprise findings.

(a) Maximum unsigned play surprise within a game was greater for the positive and negative games chosen by participants than a null distribution of all games (left) [positive vs. null Mann-Whitney U = 6.9*10^5, p = 0.002, r = 0.20, 95% confidence interval: (−0.03,−0.007); negative vs. null Mann-Whitney U = 5.8*10^5, p < 0.001, r = 0.30, 95% confidence interval: (−0.05,−0.02)]. We therefore selected subsets of the positive and negative games that did not differ from the null distribution (right) (p > 0.10, enforced). (b) Using these modified distributions, negative games still had greater unsigned full-game (left) and within-game surprise (right) values than the null distribution (99% and 100% of the samples, respectively). Shown are averages across resamples. (c, d) We plotted the t-statistics of subsets of the game data that did not differ significantly in maximum play surprise. (c) t-statistics for positive (top) and negative (bottom) full-game surprise. (d) t-statistics for positive (top) and negative (bottom) within-game surprise. Both negative distributions remained significant, whereas both positive ones were not significant.