Binary predictors used in logistic regressions predicting subjects choices (e.g., Figures 2B and 2C) are shown in Table 2. The two-step task lagged logistic regression used predictors Choice, Outcome, Transition and Transition-outcome interaction at lags 1, 2, 3-4, 5-8, 8-12 (where lag 3-4 etc. means the sum of the individual trial predictors over the specified range of lags) and predictors Bias: top/bottom, and Bias:clockwise/counter-clockwise. The Correct predictors was included in the previous trial regression to prevent correlations across trials from causing spurious loading on the Transition-outcome interaction predictor (see Akam et. al. 2015 for discussion). It was not included in the lagged regression as here the effect of earlier trials is accounted for by the lagged predictors. For the two-step task logistic regressions, the first 20 trials after each reversal in the transition probabilities was exclude from the analysis as it is ambiguous which transitions are common and rare at this point. This resulted in ∼9% of trials being excluded.
