ERIC Number: ED227141
Record Type: RIE
Publication Date: 1982-Aug
Reference Count: 0
Validation in Group Membership Prediction.
Huberty, Carl J; Smith, Janet C.
Predictive discriminant analysis involves a technique used in multivariate classification, i.e., in predicting membership in well-defined groups for units on which multiple measures are available. The validation (assessment) of group membership predictions pertains to two problems: estimating true proportions of correct classifications (i.e., hit rates) and determining, on the basis of one data set, a classification rule to use with other data sets. Two approaches to the hit rate problem are the shrinkage formula and external analysis. The formula approach has been developed only for the two-group case. Of the two external analysis methods, holdout and leave-one-out, the latter is preferred. The form of a classification rule considered is a set of k linear classification functions (LCF's). The problem of determining a classification rule has two dimensions: determining LCF weight estimates, and determining which predictors to delete from consideration in the final formulation of the rule. Jackknife methodology is used to obtain LCF weight estimates and to assess the variability of the estimates. This assessment leads to approximate t statistics which may be used to determine which of the predictors, if any, to delete. (Author/PN)
Publication Type: Speeches/Meeting Papers; Reports - Research
Education Level: N/A
Authoring Institution: N/A
Identifiers: Linear Classification Functions; Multiple Measures Approach
Note: Paper presented at the Annual Meeting of the American Psychological Association (Washington, DC, August 23-28, 1982).