Temporal precision is not an artifact of higher signal for high-reward responses.

(A) An outline of the null hypothesis using a data set in which low-reward responses had substantially lower amplitudes. Left panel: Scatterplot of response amplitude versus time per trial, colored by reward size. Gray and white shading indicates quintiles along the amplitude axis, combining high (“hi”) and low (“lo”) rewards. Right panel: 2 standard deviation width of response time distribution in each quintile. The time axis is scaled to match that for the left panel. These 2 standard deviation widths (“2x StdDev”) increased progressively for lower response amplitudes, which were also more dominated by low-reward trials. The null hypothesis is that this covariance alone gives low-reward trials larger timing scatter. Arrows mark the y-axis locations indicating median amplitudes of quintiles (data in S12 Data). (B) Plot of response amplitude versus timing for a large data set (1,285 correct trials; 629 low reward; 626 high reward) separated into quartiles by response amplitude (gray/white shading). Each quartile also roughly matched for numbers of low- versus high-reward trials (data in S13 Data). (C1, C2, D1, D2, E1, E2, F1, F2) Pairs of distributions of response amplitude and timing per quartile separated by reward size. The numbers “N” in parentheses in (C1–F1) indicate numbers of high- and low-reward trials. The high-reward responses are significantly more precise than the corresponding low-reward ones in each quartile despite the similarity in response amplitudes (C2–F2). All distributions are shown as “violin plots” using the same conventions as Fig 3A and 3B. Panels (C1–F1) share a common abscissa scale, as do panels (C2–F2). The “NTrials” label for the ordinate is shown only for (F1) to avoid clutter. p-Values, bootstrap, 10,000 resamples (data in S14 Data).