An Investigation of the Large-Sample/Small-Sample Approach to the One-Sample Test for a Mean (Sigma Unknown)

In this article we investigate the large-sample/small-sample approach to the one-sample test for a mean when the variance is unknown, using the probability of a Type I error as the criterion of interest. We show that in most cases using a t-test (t critical value) provides a more robust test than does using the z-test (standard normal critical value). The only case in which z has some advantage is when using a small sample from a parent population with extremely high kurtosis or with skewness in the direction of the rejection region tail. The implications for teaching the large-sample/small-sample approach in introductory statistics classes are discussed in light of these findings.