(A, B) For each condition, trials were divided into three groups according to saccade latency. Then, the data were normalized for each neuron, aligned with saccade initiation, and were shifted in time so that the times of saccades (colored vertical lines) were placed at the mean saccade latencies relative to the cue onset (right panels). On the left panels, data of the population activity were aligned with the cue onset (vertical black line). Inverted triangles indicate the time when the traces of normalized neuronal firing rate started to diverge as detected by repeated measures ANOVAs (p<0.01 for consecutive 40 ms). The baseline fluctuation (SD) of normalized neuronal activity was comparable between the recording sites (unpaired t-test, p=0.84). (C) Onset of trial-by-trial variation relative to the cue (left panel) or saccade initiation (right). Each point indicates the data derived from the analysis shown in (A) and (B). Error bars with three tick marks denote 2.5, 50 and 97.5 percentile of the results of the bootstrap analysis. Note that the trial-by-trial variation started earlier in the cerebellum than the striatum for medium and long intervals.