Fig. 3. Statistical models.

In separate suppressed-intercept linear mixed effects models, we tested which brain regions hold stable representations of mental state concepts across (a) targets or (b) modalities. (c) The second-order neural RSM (Fig. 2c) was the dependent variable in each mixed-effects model with individual predictors for same and different targets/modalities, and random effects for subject, study, and modality/target. (d) Because we suppressed the intercept (set equal to 0), the different targets/modalities coefficient (blue with asterisk) reflected the mean correlation values of cells that compare first-order RSMs from different targets/modalities (dark cells). That is, it measured the average similarity of mental state representations across targets/modalities. If this coefficient was significantly greater than 0, then the relevant brain region showed stable structures of mental state representation across targets/modalities.