Lecturer: professor Ted Gerber, Department of Sociology, University of Wisconsin-Madison firstname.lastname@example.org
Latent class models specify a categorical latent (unobservable) variable that explains relationships among categorical observed variables in a manner akin to factor analysis. They have broad applications in problems where the researcher postulates the population is divided into a finite set of latent discrete categories that have implications for joint distributions of observed variables: for example, “ideology” might be construed as a set of underlying discrete dispositions, each of which implies particular responses on observable opinion questions, or “social class” might be a set of discrete categories of occupations that predict their extent of autonomy, relations to the means of production, or implied human capital requirements. Latent class models are, in fact, a special instance of a broader set of models, often called “finite mixture models,” which postulate that a population consists of distinct underlying subpopulations whose characteristics and membership cannot be unobserved. We will examine some specifications of finite mixture models for which, in contrast to standard latent class models, there are no observable indicators of membership in the latent classes. These models have been applied most extensively in studies in the sociology of crime and deviance.
The course presents the basic logic of each type of model, including statistical assumptions, alternative parameterizations, principles for making inferences and assessing model fit, and parallels with more familiar regression analysis. Then we examine different subsets of each type of models. Students will learn the basic principles of research design using these models, how to estimate them using Stata 15.0, and principles for evaluating different models, interpreting results, and presenting findings.
Readings include introductory methodological treatments and illustrative substantive articles drawn from family demography, social stratification, criminology, and public opinion research. Throughout, the emphasis is on social science applications of the models rather than on statistical properties and derivations. Sessions will consist of lectures, lab time for software instruction and in-class exercises, and discussions of readings. The discussions will focus on the methodological rationale underlying the substantive claims made by the authors of the articles.
To gain the most from the course, participants should have some familiarity with social science statistical methods (in particular, multivariate regression and techniques for analyzing multiway-contingency tables.)
Day 1 (November 19, 12:00-16:00, room S-338): Latent class analysis: introduction
READ: Han and Hong 2011; Magidson and Vermunt 2004
Day 2 (November 20: 12:00-16:00, room S-338): Latent class analysis: estimation and interpretation
READ: Gerber 2000; van Gaalen and Dykstra 2006
Day 3 (November 21: 14:00-17:00, room S-338): Latent class analysis: applications and extensions
READ: Evans et al. 1998; Petev 2013
Day 4 (November 22: 9:00-12:00, room S-338): Finite Mixture models: overview and estimation
READ: Stata Corp 2017; D’Unger et al. 1998; McDermott and Nagin 2001; O’Rand and Hamil-Luker 2005
The course is financially supported by European Regional Development Fund (Project TU TEE - Tallinn University as a Promoter of Intelligent Lifestyle)