The simulation results are summarized in Fig. 2 and Supplementary Table 1. We assessed the methods on three important tasks: distinguishing non-null units from null units (Fig. 2a–b); distinguishing non-null effects (i.e., non-null unit-condition pairs) from null effects (Fig. 2c–d); and estimating effect sizes (Fig. 2e–f). (We included metasoft results for the first task only because it does not provide multivariate effect estimates, and because it is unclear how best to combine unit-level significance (p value) and condition-specific significance (m value) into a significance measure for each unit-condition pair.)
