Figure 1.
Dynamics of behavioral performance in the detection task is clustered, autocorrelated, and scale-free. A, A representative series of responses with “runs” of consecutive detected (hits; blue) and undetected (misses; red) stimuli. B, Run probability as a function of run length (top) shows that runs of ∼15–100 s are more abundant in real data (black line) than in random data (gray line). The run-length probability follows power-law scaling for runs longer than three trials. The p value (bottom; paired Wilcoxon signed-rank test) is for the difference between real and random data. Gray lines indicate the uncorrected (p = 0.05) and Bonferroni-corrected (n = 6) significance levels. C, Subjects' responses (grand average, black line, top) have stronger autocorrelations than randomized data (gray line). The difference between measured and random data is significant up to the time lag of ∼170 s (bottom; gray line, p = 0.05, paired Wilcoxon signed-rank test). D, DFA reveals that behavioral data display robust scale-free dynamics. The scaling exponent averaged across all sessions is α = 0.71 ± 0.11 (mean ± SD; thin gray lines). For the grand-average data, αμ = 0.65 (black line). For uncorrelated noise, αref = 0.5 (thick gray line).