Table 3. Predicting RT with a GLM that also includes whether the object selected on the preceding trial was present or absent.
We used a GLM to predict the normalized RT as a function of the absolute difference in the estimated (subjective) probability of reward of the two objects presented on a given trial, the trial number within a block of the experiment, the difference between BIC per trial (BICp) based on the best feature-based and object-based models for a given subject, the reward outcome on the previous trial, presence of an object that was selected on the previous trial (prev. selected present), and the interaction of the last two predictors. Reported values are the normalized regression coefficients (±s.e.m.), p-values for each coefficient (two-sided t-test), and adjusted R-squared for each experiment.
| Regressor | Abs. difference in subjective reward prob. | Trial number | BICp (Ft)–BICp (Obj) | Reward outcome on prev. trial | Prev. selected present | Reward outcome on prev. trial x prev. selected present | R2 |
|---|---|---|---|---|---|---|---|
| Exp. 1 | -0.12±0.006 (p = 10−16) | -0.16±0.005 (p = 10−16) | -0.02±0.005 (p = 10−5) | -0.05±0.005 (p = 10−16) | -0.05±0.005 (p = 10−16) | -0.05±0.006 (p = 10−16) | 0.045 |
| Exp. 2 | -0.13±0.008 (p = 10−16) | -0.12±0.008 (p = 10−16) | -0.03±0.008 (p = 10−5) | -0.06±0.008 (p = 10−16) | -0.05±0.008 (p = 4.3*10−9) | -0.07±0.008 (p = 10−16) | 0.037 |
| Exp. 3 | -0.21±-0.008 (p = 10−16) | -0.13±0.008 (p = 10−16) | -0.06±0.008 (p = 0.005) | -0.05±0.008 (p = 10−9) | -0.05±0.008 (p = 6.6*10−9) | -0.03±0.008 (p = 10−4) | 0.074 |
| Exp. 4 | -0.18±-0.007 (p = 10−16) | -0.20±0.007 (p = 10−16) | -0.01±0.007 (p = 0.03) | -0.03±0.007 (p = 10−5) | -0.02±0.007 (p = 2*10−4) | -0.02±0.007 (p = 0.002) | 0.089 |