Fig. 4

Applicability of benchmarked methods to various test statistics and case study datasets. a The top panel displays various null and non-null test statistic distributions used in simulations, with the corresponding observed FDR shown in the bottom panel. Note that although fdrreg-t requires normally distributed test statistics, it is included in the evaluation of t₁₁ and t₅ statistics to illustrate the effect of a heavy tailed distribution. In addition, neither ashq nor fdrreg-t is evaluated on χ² statistics, as they violate the assumptions of the method. b Proportion of maximum rejections (color) for each dataset and informative covariate (column, grouped by case study) and FDR correction method (row). In each column, the maximum proportion of rejections out of the total possible number of comparisons is displayed. Where ashq and fdrreg-t could not be applied to a case study to due violation of assumptions, the cell is colored in gray. Where the method lfdr was not applied due to practical limitations on the number of tests, the cell is colored grey and marked with “*.” The informative covariate used in each case study is listed in Table 1. For case studies with more than one covariate, the covariate is denoted in the x-axis labels