Premium
On the Validation of Discriminant Functions: An Empirical Analysis Using Event Related Potentials
Author(s) -
Daruna Jorge H.,
Karrer Rathe
Publication year - 1981
Publication title -
psychophysiology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.661
H-Index - 156
eISSN - 1469-8986
pISSN - 0048-5772
DOI - 10.1111/j.1469-8986.1981.tb01548.x
Subject(s) - discriminant function analysis , linear discriminant analysis , discriminant , jackknife resampling , optimal discriminant analysis , psychology , set (abstract data type) , psychophysiology , statistics , function (biology) , cross validation , artificial intelligence , data set , computer science , mathematics , psychiatry , estimator , evolutionary biology , biology , programming language
Discriminant analysis is becoming increasingly popular in psychophysiology. Appropriate use of this technique requires computation of the discriminant function using one set of data followed by its validation using a new data set (i.e., cross‐validation). Recent studies have relied on an alternative method of validation referred to as the jackknife or leave‐one‐out procedure. This study examined the appropriateness of such a procedure in instances like those typical of psychophysiology, where only a subset of the available variables is used in the discriminant function. Results are presented which indicate that the jackknife procedure can be misleading when the variables entering the discriminant function are selected on the basis of differences between the groups (i.e., stepwise discriminant analysis). It is concluded that under such circumstances cross‐validation is the more appropriate procedure.