MerQuR+Statistics Glossary

Statistics Glossary → Unconditional exact tests (Barnard, Boschloo)

Unconditional exact tests (Barnard, Boschloo)concept

Exact tests for comparing two independent proportions in a 2×2 table that fix only the group sizes and are usually more powerful than Fisher's test.

Fisher's exact test treats both the row and column totals as fixed (it is a conditional test). In many two-group studies, however, only the group sizes are set in advance; the total number of successes is random. Unconditional tests fit this situation by assuming two independent binomial distributions and taking the worst-case (largest) p value over all possible values of the unknown common success probability.

Barnard's test uses a Wald or score statistic; Boschloo's test uses the p value of Fisher's test as its test statistic and is always at least as powerful as Fisher's test. Both keep the Type I error below α in small samples while reducing the excessive conservatism of Fisher's test. The price is heavier computation; as the sample grows, the results approach those of the chi-square test.

An everyday example

A new treatment was tried in 12 patients and 9 recovered; 3 of the 12 control patients recovered. With such a small sample the chi-square approximation is unreliable. Fisher's test tends to overstate the p value for this difference, whereas Boschloo's test, which accounts for the group sizes being fixed in advance, usually gives a smaller p.

In MerQur

The 'Barnard / Boschloo' checkbox of Fisher's Exact Test adds the unconditional exact tests for two independent binomial samples with fixed row totals.

Analyses where this concept appears

Fisher's Exact TestMultiple Imputation

Related concepts

Fisher's exact testExact test and exact p-value