NotesFAQContact Us
Search Tips
ERIC Number: ED420716
Record Type: Non-Journal
Publication Date: 1998-Apr
Pages: 28
Abstractor: N/A
Reference Count: N/A
Type I Error Control of Two-Group Multivariate Tests on Means under Conditions of Heterogeneous Correlation Structure.
Fouladi, Rachel T.
A variety of approaches have been suggested by which to assess the equality of population mean vectors under conditions of population covariance matrix homogeneity and heterogeneity. The nonrobustness of commonly used multivariate tests of means to population covariance matrix heterogeneity has been long documented. However, most studies have examined the performance characteristics of the statistical procedures under conditions of heterogeneous covariance structure by simulating heterogeneity in the structure of the variances. The only study that examined performance under heterogeneous covariance structure by simulating heterogeneity in the correlations concluded that there was little difference in the performance characteristics of standard multivariate means under conditions of variance homogeneity and correlation heterogeneity (T. Beasley and J. Sheehan, 1994); this study, however, only examined the performance of the procedures under equal sample sizes. This paper assesses the Type I error control of standard and alternative multivariate tests of means under homogeneous and heterogeneous correlation structure for a full range of sample size conditions. This paper focuses on the performance of multivariate tests on means in the two-group case. A Monte Carlo simulation experiment was conducted. Findings show that the "F" based on Hotelling's T-squared is robust to between groups differences in correlation matrices under equal and unequal sample size conditions as long as the difference in the magnitude of the correlations is not extremely large, no matter what the sample size conditions or the number of variables under study. Differences between the performance profiles of the standard multivariate means test procedure and available alternative procedures are discussed. An appendix provides a chart of observed percent bias on Type I error control of multivariate tests on means. (Contains 5 tables and 22 references.) (Author/SLD)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A