Comparison of online and in-lab learning measurements.

We took measurements for n = 16 randomly selected subtasks from Experiment 1 in a group of in-lab human subjects (n = 4) that used a chinrest and calibrated monitor setup. In A, we show that the overall accuracy of these in-lab subjects fell within the empirical support of the subject distribution from our online experiments. In B, we show that patterns of average accuracy (over subtasks) were tightly correlated between the in-lab and online populations (Spearman’s ρ = 0.94; see Section 1.1 in S1 Appendix). Errorbars are SEM (simple bootstrap over subjects). (TIF)