Figure 3.

Logistic regression indicates different adaptation patterns to probabilistic reversal learning. (A) Model predictions show that the logistic regression model fits real data (red: the model prediction, blue: actual choices, green: rewarded levers). (B) Regression coefficient values from the logistic regression, with green and red lines indicating the reward predictors and non-reward predictors, respectively. Higher regression coefficients indicate a greater likelihood of selecting the previously rewarded lever in the current trial. Rewards from −2 trials back had significantly more weight on the choices of control animals after introducing the probabilistic condition (one-tailed paired sample t-test, −2 trials back; reward predictor, non-probabilistic vs. probabilistic reversal learning, t(5) = 2.532, p = 0.026). Negative feedback (no reward) from −1 trial back had significantly less influence on the choices of experimental animals in the probabilistic condition (one-tailed paired sample t-test, −1 trial back; non-reward predictor, non-probabilistic vs. probabilistic reversal learning, t(5) = 2.246, p = 0.037).