519 – Contributed Oral Poster Presentations: Social Statistics Section
A Simulation Study of the Independent Means T-Test, Satterthwaite's Approximate T-Test, and the Trimmed T-Test Under Normal and Non-Normal Distributions
Anh P. Kellermann
University of South Florida
Diep Thi Nguyen
University of South Florida
Patricia RodrÃguez de Gil
University of South Florida
Eun Sook Kim
University of South Florida
Jeffrey D. Kromrey
University of South Florida
While the independent means t-test is popular for testing the equality of two population means, it is sensitive to violations of the population normality and homogeneity of variance assumptions. Satterthwaite's approximate t-test and Yuen's trimmed t-test are recommended as robust alternatives which relax those assumptions. This simulation study compared the performance of the t-test, Satterthwaite's t-test, and the trimmed t-test under normal and non-normal distributions. The two latter tests were conducted both unconditionally and conditionally on a preliminary test of variances. Simulation conditions manipulated were total sample size, group sample size ratio, population variance ratio, population effect size, alpha sets for both the treatment effect and group variances tests, and population distribution shape. As expected, the independent means t-test showed great dispersion of Type I error rates. The other tests (both conditional and unconditional) evidenced notable improvement in Type I error control relative to the independent means t-test. Power comparisons (for conditions in which Type I error control was adequate) were used to identify the most powerful test among the set.