(A) Left: Psychometric curves for all trials from the baseline, saline, and muscimol conditions. Only L/R: trials from earlier in the shaping procedure were delivered, in which puffs were presented exclusively on the correct side, guide puffs were delivered following the cue period, and free rewards were delivered on the correct side. Performance was measured according to the side that the mice licked first, just as they do to indicate their decision in the main task. Right: Regression analysis as in Figure 1C for all trials in the indicated conditions. Error bars: 95% confidence interval. (B) Quantification of lick rate and decision latency (time from port arrival to decision lick) for all mice in all conditions. Boxes: lower and upper quartiles, line: median, whiskers: 5th to 95th percentile. Within-subject differences of rate and latency across control (baseline/saline) and muscimol conditions, and across left (L) and right (R) choices, were not significant (p>0.05, two-tailed paired t-test). As in all analyses, trials with no decision lick were excluded; this accounted for 2.0 ± 2.7% of trials in the baseline condition, and 2.6 ± 5.1% of trials in the muscimol condition (mean ±std over sessions), and was not significantly different across the baseline and muscimol conditions (p=0.54, two-tailed t-test). (C) Best-fit coefficients from the behavioral regression model. Each data point corresponds to one pair of sessions (colored by animal identity), corresponding to a muscimol inactivation session and the control session directly preceding it. Error bars indicate standard error of the fit coefficients. The evidence sensitivity parameter was significantly reduced in the muscimol condition relative to control, and the success history and failure history fit parameters were significantly increased relative to control (p<0.05, bootstrap). For the latter three parameters, one data point is omitted from display because the best-fit standard error was uninterpretably large (>8 standard deviations from others). (D) Psychometric curves for all trials from the five individual subjects in the control and muscimol conditions (one panel per subject). Curves show fits to a four-parameter logistic function (see Methods). In top-left panel, L and R indicate additional sessions in which muscimol was delivered only to the left (L) or right (R) cerebellum, to determine whether vertical shifts in the psychometric curve can be explained by left-right asymmetry of injections. Error bars: 95% confidence interval.