NotesFAQContact Us
Collection
Advanced
Search Tips
ERIC Number: ED432591
Record Type: Non-Journal
Publication Date: 1999-Apr
Pages: 23
Abstractor: N/A
Reference Count: N/A
ISBN: N/A
ISSN: N/A
Comparing the Classification Accuracy among Nonparametric, Parametric Discriminant Analysis and Logistic Regression Methods.
Ferrer, Alvaro J. Arce; Wang, Lin
This study compared the classification performance among parametric discriminant analysis, nonparametric discriminant analysis, and logistic regression in a two-group classification application. Field data from an organizational survey were analyzed and bootstrapped for additional exploration. The data were observed to depart from multivariate normality; neither the group sizes in the sample nor the covariance matrices of the two groups were equal. A crossed design of classification function by prior probability was implemented for over 244 bootstrap samples. The classification error rates for each group and the total sample were gathered for each cell of the design matrix. The major findings of this study are: (1) nonparametric discriminant functions and logistic regression performed below expectations from theory; (2) the choice of prior probabilities influenced the classification performance for the smaller and the larger group, but not for the total sample; and (3) minimization of error rates for one group implied an increment in the error rate for the other group, or vice versa. The findings do not demonstrate the expected theoretical strength of nonparametric discriminant functions when applied to data with nonnormality and unequal covariance matrices. No consistent superiority was observed in logistic regression and quadratic discriminant function over the linear discriminant function. This indicates a more complicated situation than that portrayed in previous studies on the applications of discriminant functions and logistic regression for classification purposes. (Contains 7 tables and 22 references.) (Author)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A