next up previous contents index
Next: Generalizing Tests of Significance Up: Test of Significance involving Previous: Two-tailed t-test

The t-test is not universally appropriate

There are situations where the t-test does not work well. The t-test tends to be appropriate whenever the sample size is reasonably large, say $n\geq 25$. For sample sizes smaller than that, the t-test is appropriate only when the population histogram looks normal in the first place. When the sample size is small and the population histogram does not look normal, alternative tests called nonparametric tests are advised.