The area between this curve and the mean performance accuracy is a measure of performance fluctuation over the microstimulation session: if performance stayed close to average all the way through the session this area would be very small, whereas if performance fluctuated greatly this area would be large as the smoothed performance line moves away from the mean.
To estimate the amount of performance fluctuation over the microstimulation session at a given site, the area between the smoothed performance curve and the horizontal line indicating the mean overall performance over that session was calculated using trapezoidal numerical integration implemented by MATLAB's trapz function (see Figure 8 figure supplement 1A).
Secondly, we separated trials into "good" and "bad" task engagement epochs depending upon whether they were located in a (>30 trial) time-window of performance accuracy that was either higher or lower, respectively, than the overall average performance accuracy over the microstimulation session at each site.
However, a time-series analysis of performance over the course of the microstimulation experimental sessions showed that there was no correlation between the performance fluctuation and the differential effect of microstimulation by reward.
First, we measured the extent of fluctuation in smoothed performance accuracy (a measure of task engagement) over the microstimulation session at each site, by calculating a sliding average of performance accuracy over time windows of 30 trials.
These fluctuations could therefore potentially underlie the observed association between large reward trials and better performance, and the observed effects on the microstimulation shift.
In Figure 8 figure supplement 1, we analyse the effect of reward in good and bad performance epochs within microstimulation sessions, with the same result (please also see our response to point 1).
(B ) There is no correlation between normalised reward effect size and the amount of performance fluctuation across microstimulation sites, either for animals separately or together (Pearson's product moment correlation: p >> 0.05 in all cases), suggesting that performance fluctuation cannot explain the reward effect on microstimulation.
This could be solved by adding such information to Table 2. 5) Unexplored trends in the data: There's a puzzle in that the bulk of the cases where the reward manipulation had a big effect were cases where the microstimulation effect was weak and the performance was good.
The microstimulation effect may wane over time in some experiments (Salzman et al., 1992 ). Microstimulation might be less effective near the end of each experimental session, simultaneously with the occurrence of a greater proportion of large reward trials near the end of the session – for example, owing to a training effect that increases the number of consecutive correct responses.
