Bias of Odds Ratio Estimate in Fisher's Exact Test
Xiaofeng Steven LiuABSTRACT
Objectives
The odds ratio estimate in Fisher's exact test can overestimate the parameter. A simple computer simulation can easily reveal the positive bias of the odds ratio estimate from Fisher's exact test. Bootstrap can facilitate bias correction for the odds ratio estimate.
Methods
The bias can be estimated, using bootstrap samples and the original sample to approximate the expectation of the odds ratio estimator and the true parameter value—their difference is the bias. Here, the bias is computed from the underlying distribution, conditional on the exclusion of zero cells in sampling, to avoid the infinite expectation.
Results
A study of depression is used to demonstrate how to use bootstrap to correct the bias in an odds ratio estimate based on Fisher's exact test.
Conclusions
Bootstrapping can easily estimate and correct the bias of an odds ratio estimate in Fisher's exact test. The results suggest that bootstrapping is sensitive enough to detect even a small bias.