Linear discriminant analysis versus logistic regression: A comparison of classification errors in the two-group case

Classification studies are important for practitioners who need to identify individuals for specialized treatment or intervention. When interventions are irreversible or misclassifications are costly, information about the proficiency of different classification procedures becomes invaluable. This study furnishes information about the relative accuracy of two widely used classification procedures, linear discriminant analysis and logistic regression, under various commonly encountered and interacting conditions. Monte Carlo simulation was used to manipulate four factors under multivariate normality: equality of covariance matrices, degree of group separation, sample size, and prior probabilities. Three criterion measures were employed: total, small-group, and large-group classification error. Interactions of these between factors with two within factors, cut-score and method of classification, were of primary interest.