Robust Fisher's Exact Test
Also known as: mid-p Fisher's exact test, robust exact test for contingency tables, conditional robust Fisher test, Fisher mid-p test
The robust Fisher's exact test extends Fisher's classic exact test for contingency tables by applying conservative-correcting adjustments — most commonly the mid-p correction — to reduce the extreme conservatism of the standard exact test. This produces better-calibrated Type I error rates while maintaining validity in small and sparse samples.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the robust Fisher's exact test when: (1) you have a small or sparse contingency table (expected cell counts below 5 in any cell) where chi-square approximations are invalid; (2) you find the standard Fisher's exact test too conservative for your application and want better power without violating validity; (3) marginal totals are considered fixed by design (e.g., case-control studies). Do NOT use it as a substitute for chi-square when all expected frequencies exceed 5 — standard chi-square or Fisher's exact test suffice. Avoid when the unconditional approach (Barnard's test) is more natural because row totals are random rather than fixed.
Strengths & limitations
- Maintains validity for very small samples and sparse tables where chi-square approximations fail.
- Reduces the excessive conservatism of the standard Fisher's exact test, recovering lost statistical power.
- No distributional assumptions about the data beyond the conditioning on fixed marginals.
- The mid-p correction is widely accepted and implemented in major software packages (R, SAS, Stata).
- Particularly well-suited for case-control and other designs with fixed marginal totals.
- The mid-p test is not strictly conservative: the Type I error rate may slightly exceed the nominal level in some configurations.
- Conditioning on fixed margins can be controversial when marginals are random — Barnard's unconditional test may be preferable.
- Computationally intensive for large tables or large sample sizes, though this is usually not a practical concern with modern software.
- Less familiar to reviewers than standard Fisher's exact test; justification may be needed in applied publications.
Frequently asked
When should I prefer the robust Fisher test over the standard Fisher's exact test?
When you are concerned that the standard Fisher test is too conservative — rejecting too rarely — you can use the mid-p version to gain statistical power. This is most relevant when you have reason to believe the true effect exists but the standard test fails to detect it at borderline significance.
Is the mid-p correction the same as a continuity correction?
No. The Yates continuity correction is applied to the chi-square test statistic to approximate the exact distribution. The mid-p correction is applied directly to the hypergeometric tail probability in Fisher's exact framework, halving the probability mass assigned to the observed outcome. They address different problems and should not be used together.
Does the mid-p test still control Type I error?
The mid-p test is not strictly conservative: in some configurations the true Type I error rate can slightly exceed the nominal level. However, the exceedance is typically very small in practice, and for most applications the improved power calibration justifies the minor loss of strict conservatism.
What is the difference between Barnard's test and the robust Fisher test?
Both are designed to improve on standard Fisher's exact test, but they differ in their conditioning principle. Fisher's test (and its mid-p variant) conditions on both marginal totals being fixed — appropriate for stratified designs. Barnard's unconditional test does not condition on one set of margins and generally has more power when those margins are random.
What effect size should I report alongside this test?
Report the odds ratio (OR) with its 95% confidence interval. The conditional maximum likelihood OR is consistent with the Fisher exact framework; mid-p confidence intervals for the OR are also available in R (exact2x2 package) and provide intervals with improved coverage properties.
Sources
- Agresti, A. (2002). Categorical Data Analysis (2nd ed.). Wiley-Interscience. ISBN: 978-0471360933
- Lancaster, H. O. (1961). Significance tests in discrete distributions. Journal of the American Statistical Association, 56(294), 223–234. DOI: 10.1080/01621459.1961.10482105 ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Fisher's Exact Test. ScholarGate. https://scholargate.app/en/statistics/robust-fishers-exact-test
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Chi-square testStatistics↔ compare
- Fisher's exact testStatistics↔ compare
- Robust chi-square testStatistics↔ compare