(A), Identification of the optimal Δbias for an example session using logistic fits.
For each reward context (blue for LR-Left and red for LR-Right), RTrial was computed as a function of bias values sampled uniformly over a broad range, given the session-specific sensitivities, lapse rate, coherences, and large:small reward ratio. The optimal Δbias was defined as the difference between the bias values with the maximal RTrial for the two reward contexts. The fitted Δbias was defined as the difference between the fitted bias values for the two reward contexts. (B) The optimal bias decreases with increasing sensitivity. The example heatmap shows normalized RTrial as a function of sensitivity and bias values in the LR-Right blocks, assuming the same coherence levels as used for the monkeys and a large:small reward ratio of 2.3. The black curve indicates the optimal bias values for a given sensitivity value. (C) Scatterplots of optimal Δbiases obtained via the procedure described above as a function of sensitivity for each of the two reward contexts. Same format as Figure 2C Solid lines indicate significant partial Spearman correlation after accounting for changes in reward ratio across sessions (p<0.05). Note that the scatterplots of the monkeys’ Δbiases and sensitivities in Figure 2C also show negative correlations, similar to this pattern.