Creator

Raj K. Chandran

Advisor

Mundfrom, Dan J.

Committee Member

Perrett, Jamis J.

Committee Member

Schaffer, Jay R,

Committee Member

Heiny, Robert L.

Department

Applied Statistics & Research Methods

Institution

University of Northern Colorado

Type of Resources

Text

Place of Publication

Greeley (Colo.)

Publisher

University of Northern Colorado

Date Created

5-1-2009

Genre

Thesis

Extent

199 pages

Digital Origin

Born digital

Description

This study examined the effectiveness of stepwise discriminant analysis (SWDA) using the F-statistic and Partial R-square criterion as a follow up analysis to a significant MANOVA. Monte Carlo simulations were conducted, and 7,128 scenarios were examined using different combinations of levels of number of MANOVA dependent variables, sample size, population correlation matrices, effect sizes, alpha significance levels and Partial R-square correlations. The two group case of MANOVA was considered, and simulations were run under the assumptions of multivariate normality, homogeneity of variance-covariance matrices, and linearity among all pairs of predictors within each group. This study has shown that SWDA is a viable option as a follow up analysis to a significant MANOVA if the correct conditions are met. It was found that SWDA performs well when the number of dependent variables with significantly differing means in each group is held low. Based on the results SWDA performs best when the number of significant dependent variables is three or less. Additionally, SWDA only works well when correlations between dependent variables are quite low. If correlations between dependent variables are held low, then SWDA can be used in situations where there are three dependent variables or less. SWDA can be used in situations where there are more than three dependent variables, but the number of significant dependent variables must be below four in order for SWDA to perform well. Another procedure could be used to gauge what that may be, then SWDA could be employed if the correct conditions are met. Because SWDA only works well when low correlations between dependent variables are present, it could be combined with another procedure, perhaps descriptive discriminant analysis to supplement situations when higher correlations are found. This dissertation has shown however, that using several univariate F-tests, also known as the "protected" F-test, should not be used after a significant MANOVA and SWDA should be used instead if the correct conditions are met.

Notes

[Released from embargo]

Degree type

PhD

Degree Name

Doctoral

Language

English

Local Identifiers

Chandran_unco_0161N_10000

Rights Statement

Copyright is held by author.

COinS