Bonferroni's bound: A control of significance level errors in speech pathology and audiology research

Abstract

Many studies in behavioral sciences, such as speech pathology and audiology, involve statistical hypothesis testing. Repeated tests are made, for example, of judge reliability in assessing the disorder, or within subject variability, or between subject comparisons over several measures of the disorder or types of treatment. If the error rate of the statistical test is only controlled for each individual test, the overall error rate is magnified and the chance of reporting a significant result where none exists arises. This paper addresses this potential problem, by noting some common procedures that inherently guard against this pitfall, and suggesting a simple, albeit conservative, solution for other cases.