A Brief Description of
Discriminant analysis is a statistical technique with increasing applications in political science. Recent studies that employ discriminant techniques include analysis of floating voters ( Kim 1994), party identification ( Rose and Mishler 1998), and political cleavages ( Knutsen 1989). My use of discriminant analysis in this book follows Knutsen ( 1989) analysis of cleavage dimensions in Norway and Western Europe. The main statistical notions and assumptions are drawn from Klecka ( 1980).
Discriminant analysis is a multivariate statistical technique used to study differences between two or more groups on the basis of several discriminating variables. The groups are mutually exclusive categories of a nominallevel variable. In the study of political cleavages developed in this book, this nominal-level variable is individual party preference. The discriminating variables are measurements that are expected to distinguish between (or classify) the groups. In this book the discriminating variables are attitudinal issue dimensions--postmodern-fundamentalist, left-right materialism, democratic-authoritarian, liberal fundamentalist, attitudes toward reform--and measures of the individual's age, income, religiosity, region where he or she lives, language, etc. The results from discriminant analysis shown in this book include two main pieces of information: (a) the canonical discriminant functions and (b) the group centroids.
The canonical discriminant functions are linear combinations of the discriminating variables. They indicate which variables are the most powerful discriminators of the groups. The discriminant functions in the tables used throughout this book display two types of coefficients for each discriminating variable: a standardized discriminant coefficient and a structural coefficient (which is shown in parentheses). The standardized discriminant coefficient indicates the relative contribution of a variable in determining the discriminant function. The higher the coefficient of a variable, the more