Which Results Survive Correction? Find Out in Minutes

How it works

The analysis validates your p-values (dropping any missing or outside 0-1), then applies two standard corrections from base R's p.adjust: Holm-Bonferroni, a step-down procedure that controls the family-wise error rate (the chance of even one false positive), and Benjamini-Hochberg, a step-up procedure that controls the false discovery rate (the expected fraction of false positives among your significant results). Every test gets both adjusted p-values and a significance verdict at alpha = 0.05, and a 10-bin p-value histogram diagnoses whether your results look like real effects or noise.

Use it any time you ran more than a handful of hypothesis tests — A/B variants, metrics, subgroups, biomarkers — and want to know which findings hold up. Holm answers 'which results can I state with almost no risk of any false positive?'; BH answers 'which results can I pursue while keeping the false-positive fraction low?'

Not for computing the p-values themselves (run the underlying tests first), and not a substitute for pre-registered analysis — correcting after cherry-picking which tests to include still biases results.

What data do you need?

One row per test: the test's name and its raw p-value. For example, a batch of experiment results:

Minimum 5 rows · Best with 10-1,000 tests

What's in the report?

Standard-library analysis: you ran many statistical tests — which results are still significant after correction? Upload your p-values and get Holm-Bonferroni and Benjamini-Hochberg adjusted p-values side by side, a count of what survives each method, and a p-value distribution diagnostic that hints whether your significant results reflect real effects or multiple-testing luck.