Figure 7. No evidence for a trade-off between context models.

(A) Trade-off across subjects: testing the hypothesis that subjects differ in which context model they rely on. Each plot compares the predictive power of two context models in the mid/posterior superior temporal gyrus region of interest, each dot representing % explained for one subject. The line represents a linear regression with 95% bootstrap confidence interval (Waskom, 2021). None of the pairwise comparisons exhibits a negative correlation that would be evidence for a trade-off between reliance on different context models. Data from Figure 4—source data 2. (B) Trade-off over time: testing the hypothesis that subjects alternate over time in which context model they rely on. Each dot represents the partial correlation over time between the predictive power of two context models for one subject, controlling for predictive power of the full model. Correlations are shown separately for the left and the right hemisphere (L/R). Stars correspond to a one-sample t-tests of the null hypothesis that the average r across subjects is 0, that is, that the two context models are unrelated over time. None of the context models exhibited a significant negative correlation that would be evidence for a trade-off over time.