Open Access Dissertation
Date of Award
Education Evaluation and Research
Both discriminant function analysis (DFA) and logistic regression (LR) are used to classify subjects into a category/group based upon several explanatory variables (Liong & Foo, 2013). Although the two procedures are generally related, there is no clear advice in the statistical literature on when to use DFA vs. LR, although LR appears to be preferred due to the claim that its underlying assumptions are more easily met (Liong & Foo, 2013). Although DFA and LR use different methods to accomplish their objectives, they can answer the same research questions (Antonogeorgos et al., 2009). This facilitates a practical comparison of their outcome to identify the differences and/or similarities of the two methods. Therefore, the purpose of this study is to compare the operating characteristics of DFA and LR when using dichotomous predictors, with a particular application of prison data. This analysis used the two multivariate statistical methods to evaluate if the mentally ill will be housed in prison by identifying the predictors. The outcome showed that although the assumptions normality and homogeneity of variance were violated, the results of both methods were virtually identical, and showed a disproportionate number of the mentally ill being incarcerated in the United States penal system. This suggests DFA is robust to violations of normality and homogeneity of variance/covariance. Therefore, for this analysis, the performance and results of DFA and LR are comparable, even when the assumptions are violated.
King, Mona, "A Comparison Of Discriminant Function Analysis And Logistic Regression By Categorizing The Incarcerated Mentally Ill" (2018). Wayne State University Dissertations. 2039.