Fig. 2

Relative type 1 error rates (false positive rates) for the 3 tests of interest. Relative rates are plotted on the log2 scale, for the 4 different testing thresholds at each of the 3 sample sizes examined. Values greater than 0 indicate type 1 error rates higher than nominal levels (e.g. a value of 1 means the observed rate was double what one would expect), while values less than 0 indicate conservative behavior (e.g. -1 means observed rate was half of what is expected). Values in the figure are based on the average rate calculated across the 10 data sets simulated at each sample size. Supplementary Fig. 5 shows the variability in observed rates for each of the methods. Results are consistent across the data sets, indicating the mean of the results is an accurate reflection of each method’s performance