Time Varying Covariates in Markov Latent Class Analysis: Some Problems and Solutions
Marcus Berozfsky, Paul P. Biemer, William D. Kalsbeek · 2010
Markov latent class analysis (MLCA) is a modeling technique for panel or longitudinal data that can be used to estimate the classification error rates for categorical outcomes with categorical predictors (i.e., false positive and false negative rates for dichotomous items) when gold standard measurements are not available. Because panel surveys track respondents over time, explanatory variables (called grouping variables) can be either time varying or time invariant (static). Time varying grouping variables can be useful in explaining differences in the latent construct. However, they generate a large number of model parameters that can cause problems with data sparseness, make model diagnostics invalid, and model convergence less reliable. This paper discusses alternative coding schemes for time varying grouping variables and proposes a set of procedures for determining the best coding scheme for a particular set of data. This process is then illustrated using data from the National Crime Victimization Survey (NCVS). We found that for the NCVS, when parsimony is taken into account, a coding scheme that uses fewer model parameters has better fit than the more traditional coding scheme and another alternative and does not negatively affect the estimates of the classification error.