A Poisson Method of Controlling the Maximum Tolerable Number of Type I Errors
The problem of controlling the risk of occurrence of at least one Type I Error in a family of n statistical tests has been discussed extensively in psychological literature. However, the more general problem of controlling the probability of occurrence of more than some maximum (not necessarily zero) tolerable number ( xm) of Type I Errors in such a family appears to have received little attention. The present paper presents a simple Poisson approximation to the significance level P( EI) which should be used per test, to achieve this goal, in a family of n independent tests. The cases of equal and unequal significance levels for the n tests are discussed. Relative merits and limitations of the Poisson and Bonferroni methods of controlling the number of Type I Errors are examined, and application of the Poisson method to tests of orthogonal contrasts in analysis of variance, multiple tests of hypotheses in single studies, and multiple tests of hypotheses in literature reviews, are discussed.