Piotr Radkiewicz
Institute for Social Studies, University of Warsaw
Abstract:
This article is dedicated to the discriminant analysis – a statistical method that allows to test differences between groups of observations (two or more) based on a set of selected independent variables (predictors). It may be effectively applied to various fields of social sciences and practice (psychology, sociology, political science, economy, law). Linear combination of independent variables obtained on the basis of the discriminant analysis model serves as a criterion of assigning observations to groups. Information carried by independent variables is saved in a synthetic form as discriminant function scores. Discriminant analysis may have two goals: discrimination (separation) and classification (allocation). In the first case, the researcher tries to explain causes of differences between groups of observations by making use of their characteristics available as disciriminating variables. In the second case, the researcher seeks to find a mathematical equation that combines group characteristics of observations in order to effectively predict the unknown group category to which the observation belongs. The article consists of two parts. The first part contains a general description of the statistical model, the second includes two empirical examples of its application for two and for four groups of observations.
Keywords: discriminant analysis, grouping variable, linear discriminant function, group centroids, classification, observed and predicted group membership
Cite this article as:
Radkiewicz, P. (2010). Discriminant analysis. Basic assumptions and applications in social research. Psychologia Społeczna, 14, 142–161.