Fig. 2: Retrieval accuracy across days as a function of semantic relatedness and learning manipulation.
A On Day 1, a two-tailed t-test revealed that pairs that underwent testing during Round 3 showed a significant effect of relatedness (blue bars; t(79) = 9.979, p < 0.001, d = 1.12, 95% CI = [0.19 0.28]). Restudied pairs from Day 1 are shown (maroon bars) but were not analyzed since the Day 1 restudy score merely reflects the ability to re-type the target word that was displayed on the screen, rather than memory strength. B On Day 2 (our critical measure of learning outcomes), a RM-ANOVA revealed that there was better cued recall of target words from related pairs than unrelated pairs (F(1,79) = 362.23, p < 0.001, ηG2 = 0.462), and better recall for tested pairs (blue bars) than restudied pairs (maroon bars) (F(1,79) = 25.24, p < 0.001, ηG2 = 0.450), but no statistically significant interaction. C Splitting tested word pairs based on whether or not they were successfully recalled on Day 1 reveals a significant semantic relatedness by learning manipulation interaction (F(2,154) = 23.53, p < 0.001, ηG2 = 0.045), showing a larger testing effect (i.e. an advantage for tested pairs that were correctly recalled at Day 1 over restudied pairs) for unrelated pairs than for related pairs. Solid blue bars reflect tested pairs that were recalled correctly at Day 1, striped bars reflect tested pairs that were not recalled correctly at Day 1. D The interaction was driven by a larger testing effect (i.e. an advantage for tested pairs that were correctly recalled at Day 1 over restudied pairs) for unrelated pairs (yellow) than for related pairs (green). Across all panels, open circles reflect means of individual participants (N = 80), with connecting lines showing within-subject differences across conditions. Error bars reflect standard error of the mean. Symbols reflect statistically significant differences across conditions using Holm-Bonferroni corrections for multiple comparisons (*p < 0.05, **p < 0.01, ***p < 0.001).
