SAMPLE SIZE DETERMINATION BASED ON MID-P VALUE FOR USE WITH THE TESTING IN 2×2 COMPARATIVE TRIALS

Abstract

In testing two binomial probabilities Fisher's exact test is known to be too conservative. As a consequence, it requires a larger sample size than expected to detect statistical significance. The use of mid-P value instead of the ordinary P-value in the testing would be one possibility to resolve such conservatism. This article investigates how much sample size can be reduced by adopting the mid-P value. It is shown that the sample sizes are almost comparable to those of unconditional tests. A table of sample sizes is given for the reasonable range of parameters to achieve the power of 0.8 or more when the significance level is 0.05.