Extended Data Fig. 2: Reason for Z-scoring to fit the reversal points.

a–c) Three example neurons’ firing rate plotted as a function of time since cue onset, and split according to the four value levels, showing that some neurons increase their firing relative to baseline pre-stimulus firing rate for all reward levels (a), others increase or decrease it depending on the reward level (b) and others decrease it for all reward levels (c). Shaded regions denote SEM. The reason for Z-scoring in our data is as follows. In dopamine neurons, it appears that any firing rate deviation from baseline activity (that is, pre-stimulus onset activity) is signalling a reward prediction error. This is not true for cortical neurons, which may, for example, increase (as in a) or decrease (c) their firing to all reward levels. If this is the case, then deviation from baseline cannot be assumed to denote an RPE. That some neurons either increase or decrease their firing to all reward levels is indicative of the heterogenous coding schemes evident in PFC neurons. Given this, we can isolate the component of the activity that is associated with RPE by calculating z-scores, and using deviation from mean firing to capture the same effect and compute reversal points. Therefore our reversal point measure captures, for each neuron, the relative differences in responses to different reward levels (that is, the non-linearity) that indicates optimism, rather than being affected by overall shifts in firing. These reversal points, that is, the value at which the neuron firing reverses from below to above the mean firing in the epoch, are an index of neuron optimism; the higher the reversal point, the more optimistic the neuron, and neurons with reversal points above 2.5 are optimistic and below 2.5 are pessimistic (Methods).