Using Confidence Intervals to Test Hypotheses

Abstract

When separate 100(1-α) percent confidence intervals on two population means overlap, the means may or may not be significantly different at the 100α percent level. This phenomenon is explored analytically and numerical examples are given.