Access Type

Open Access Dissertation

Date of Award

January 2015

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Education Evaluation and Research

First Advisor

Shlomo S. Sawilowsky

Abstract

The author examined how, in the context of experimental design, one might become aware of the Behrens-Fisher problem (heteroscedasticity) in order to apply an approximate solution, such as the Yuen's statistic (1974). It was expected that both the Mood-Westenberg dispersion test (1948) and the Siegel-Tukey test (1960) would remain robust with respect to Type I and Type II error properties (and associated power levels) for detecting variance changes when their assumptions of equal means was slightly violated (i.e., the Behrens-Fisher problem). With the use of Monte Carlo Simulations, the author reviewed 34,606 permutations composed of interactions between various sample sizes, alpha levels, distributions/data sets, variance changes and means shifts. While the Mood-Westenberg (1948) and Siegel-Tukey (1960) tests both remained robust under certain conditions with respect to Type I and II error properties, the Siegel-Tukey test (1960) was by far the most robust of the two statistics, able to handle a more diverse set of conditions and would therefore be the statistic of choice in identifying the Behrens-Fisher problem.

Share

COinS