By including additional predictors in the regression, which in this example relate to the value of the actions taken, state reached on the trial, and the recent rate of rewards over the past 8 trials (see Blanco-Pozo et al.102for details), we can both resolve the influence of these different behavioural variables at different timepoints and remove the spurious loading on the outcome predictor before the outcome cue, as variance in the signal due to reward expectation is now captured by the other predictors.
