Title:  Robust Means Modeling with k-Independent Groups.

 

Abstract: Analysis of variance (ANOVA) is a commonly used test in the social sciences. For ANOVA to be properly implemented, two important assumptions must be met: normality and homogeneity of variance. These assumptions, however, are rarely satisfied in social science data. Many alternatives to the traditional ANOVA exist (e.g, Welch, 1951), but a promising alternative called robust means modeling (RMM) has recently been investigated (Fan & Hancock, 2012). RMM is built upon a structural equation modeling framework, and incorporates procedures to control for non-normality and heterogeneity of variance. This paper extends previous work on RMM using new conditions and estimators to compare RMM’s Type I error rates and power to traditional and modified ANOVA-based statistics. We also extended the RMM research to include pairwise comparison tests using the RMM approach. The talk will be separated into two parts; the first will discuss the results of the omnibus ANOVA-based and RMM methods, while the second will compare the results of the differing multiple comparison strategies. Preliminary results suggest that RMM methods tend to perform better than traditional methods, but have liberal error rates with small sample sizes and may suffer from convergence problems.