Factors Affecting Multiple Regression and Discriminant Analysis with a Dichotomous Dependent Variable: Prediction, Explanation, and Classification

Social science researchers use multiple regression and/or discriminant analysis to predict, explain, or classify with a dichotomous dependent variable. This study outlines the steps a researcher should take to analyze data using each method, the concerns and emphasis of each method, and the issues involved. Basically, both statistical methodologies are affected by the magnitude of intercorrelation, marginal dependency, multivariate normality, variance-covariance in groups, prior probability of group membership, and misclassification. The usefulness of these two multivariable statistical methods can be enhanced if we better understand how these assumptions and issues impact their outcomes. ocial science research studies are inevitably multivariable and correlational, thus non-experimental in nature and based on meeting certain underlying assumptions (Newman, 1989). Multivariable techniques examine the relationship among several variables using a correlation matrix or corresponding variance-covariance matrix. Multiple regression and related discriminant analysis with a dichotomous dependent variable therefore uses the correlation or variance-covariance among a set of variables to predict, explain, or classify. A first step in using either method is to first determine if the sample data contains meaningful relationships, hence a test of multivariable independence. The multivariable test of independenc