.png)
139 – Inference and Variance Estimation with Complex Survey Data
Parametric Tests for Two Population Means: An Empirical Comparison of Type I Error Control and Statistical Power
Patricia Rodriguez de Gil
University of South Florida
Yi-Hsin Chen
University of South Florida
Eun Sook Kim
University of South Florida
Diep Nguyen
University of South Florida
Anh Kellermann
University of South Florida
Aarti Bellara
University of South Florida
Jeffrey D. Kromrey
University of South Florida
A simulation study was conducted to explore the Type I error rates and statistical power of the independent means t-test, Satterthwaite's approximate t-test, and the conditional t-test based on a preliminary test of variances. Factors manipulated were total sample size (10, 20, 50, 100, 200, 300, 400), sample size ratio between the groups (1:1, 2:3, 1:4), population variance ratio (1:1, 1:2, 1:4, 1:8, 1:12, 1:16, 1:20), population effect size (0, .2, .5, .8), and alpha for both the test of treatment effect and the test of variances. Normal population distributions and distributions with varying kurtosis and skewness values (?1=1.00, ?2=3.00; ?=1:50, ?2=5:00; ?=2.00, ?2=6.00; ?1=0.00, ?2=25.00) were included. As expected, the independent means t-test showed great dispersion of Type I error control. The Satterthwaite t-test provided adequate Type I error control in nearly all conditions and the conditional t-test evidenced notable improvement in Type I error control relative to the independent means t-test as the level of alpha for the test of variances increased. This study provides guidance on the proper use of parametric tests with nonnormal, heteroscedastic populations.