Fig. 2: Stories, not sentences, elicit a lasting influence on spontaneous thought.

A Histograms of participant responses, across three separate experiments, to the question: “To what extent did the text linger in your mind after reading it?”. Participants provided their rating on a 7-pt scale: 7 = very much, 1 = not at all. Black dashed line represents the mean rating per condition. n = 80 participants per condition, per experiment. B Histograms of how accurately a document classifier could discriminate between pre- and post-story free association across three separate datasets. Classifiers were trained within-condition per dataset (n = 80 participants), using a leave-one-participant-out cross-validation procedure with 500 bootstraps. Solid lines represent the mean classification accuracy. Null distributions were estimated by randomly shuffling the labels of the held-out participant’s word chains (pre, post) and recalculating classification accuracy over 500 permutations. Likelihood of achieving mean classification from the null distribution was calculated using a one-sided permutation test [ns p > .05; * p = <.05; ** p = <0.01; Note 1: all ps are uncorrected with respect to multiple comparisons; Note 2: minimum p value estimate for this analysis is p < 0.002]. The exact accuracies and p values for each dataset and condition are as follows: Carver-Replication – Intact [74% accuracy, p < 0.002], Sentence-scrambled [55% accuracy, p = 0.130]; Carver-Rewrite – Intact [60% accuracy, p = 0.036], Sentence-scrambled [62% accuracy, p = 0.004]; July – Intact [81% accuracy, p < 0.002], Sentence-scrambled [51% accuracy, p = 0.420]. Source data for all panels are provided as a Source Data file.