Assessing reliability of categorical substance use measures with latent class analysis.

This article illustrates the use of the latent class model to identify classes of individuals and to assess the psychometric reliability of categorical items. The latent class model is a categorical latent variable model used to identify homogeneous classes of respondents such that class membership accounts for item responses. The assessment of measurement reliability comes directly from the estimates of the model. Although not based on classical test theory, the reliability assessment procedures described here answer the same question-that is, how consistent or dependable is measurement? The goal is to identify reliable indicators of a characteristic by examining measurement error and the inter-relatedness of the items. Methods for estimating the reliability of individual items as well as sets of items are presented. These methods are illustrated with data on cigarette smoking from a national sample of adolescents. By using the procedures described here, researchers are able to determine: (1). which classes of people are measured well and which are not; (2). which items perform well and which do not; and (3). whether items need to be altered or added in order to measure and identify particular classes better.