The analysis of the functional imaging commenced with single subject analyses. We created subject-specific design matrices containing the following regressors: (a) 3 regressors encoding the average BOLD response at each of the 3 states (2 choice states, 1 outcome state), (b) 2 regressors encoding the model-derived prediction error signals (RPE and SPE) modeled at the time of state 2 and the outcome, (c) 2 regressors of model-derived value signals modeled at the time of state 1 and state 2 (not further analyzed in this paper), (d) a nuisance partition containing regressors modeling the individual scans that were identified as contaminated by movement and unphysiological global signal change (see Image Processing above), (e) a nuisance partition containing 6 regressors that encoded the movement displacement as estimated from the affine part of the image realignment procedure. Because subjects did not receive any reward information in session 1, we only included the state prediction error signal; all other model-derived variable were 0 throughout the entire session 1 because of the lacking reward information. Both error signals were enter unorthgonalized into the 1st level design matrices.
