Biased minus standard reward prediction errors

Contributed by johannes.algermissen on Sept. 23, 2021

Collection: Biased credit assignment in motivational learning biases arises through prefrontal influences on striatal learning

Description: Parametric regressor: reward prediction errors computed with a standard Rescorla Wagner model. This is GLM1. The GLM contains the following 10 regressors: 1-4) 4 regressors crossing the performed action (Go/ NoGo) with the valence of the cue (Win/ Avoid). At the time of cue onset. 5) Response hand: +1 for left and response, 0 for no response, -1 for right hand response. At the time of cue onset. 6) Incorrect response. At the time of responses. 7) Outcome Onset (any outcome). At the time of outcomes. 8) Standard reward prediction errors computed with the standard Rescorla-Wagner learning model. 9) Difference between biased and standard reward prediction errors, respectively computed with a) a Rescorla-Wagner models assuming an increased learning rate for rewarded Go responses and a decreased learning rate after punished NoGo responses, and b) a standard Rescorla-Wagner learning model. At the time of outcomes. 10) Invalid outcomes (non-instructed key pressed, returning error message). At the time of outcomes. This contrast just reflects the difference of biased minus standard reward prediction errors (regressor 9). A conjunction of this contrast and the standard reward prediction error contrast captures regions for which BOLD signal is significantly better explained by biased prediction errors compared to standard prediction errors (see approach in Wittmann et al., 2006; Daw et al., 2011).

Tags: reward value valence learning outcome reinforcement bias

Task View 3D View