Literaturnachweis - Detailanzeige
Autor/in | Daniel, Larry G. |
---|---|
Titel | Use of the Jackknife Statistic To Establish the External Validity of Discriminant Analysis Results. |
Quelle | (1989), (22 Seiten)
PDF als Volltext |
Sprache | englisch |
Dokumenttyp | gedruckt; online; Monographie |
Schlagwörter | Discriminant Analysis; Educational Research; Estimation (Mathematics); Monte Carlo Methods; Research Methodology |
Abstract | That the jackknifing technique is superior to traditional techniques for assessing the external validity of statistical results of discriminant analysis is defended. Traditional approaches assessed include: (1) the empirical method, in which the discriminant function coefficients (DFCs) obtained in a given analysis are applied to predict group membership in the same sample used for deriving the DFCs; (2) the "holdout" method, in which statistical results are cross-validated by random splitting of the original sample into a group for deriving the discriminant function and a group for cross-validating it; (3) the Monte Carlo method; and (4) the random assignment method, whereby discriminant functions are computed based on repeated random assignment of actual cases from the original sample to groups. The jackknife statistic (JS) is similar to the "U-method" and focuses on the stability of the DFCs obtained in the original analysis. One case or subset of cases is eliminated from the original data set, and the discriminant function is computed using the remaining observations. The procedure is repeated, with each individual observation or unique subgroup, in turn, omitted. At each step, pseudo-values are computed, based on computation of the original and cases-minus-one DFCs. The values are averaged to provide a jackknifed estimate of the DFCs. A data set shows the JS's value and is used to assess the stability of the jackknifed DFCs for three predictors of teachers' (N=69) level of experience. The JS may be used to reduce bias in an estimator that is attributable to artifacts of the sample used. Since jackknife methods minimize sample splitting via sample omission and reuse, they are particularly useful with small samples. Four data tables are provided. (TJH) |
Erfasst von | ERIC (Education Resources Information Center), Washington, DC |