Addressing social justice and population heterogeneity: An applied introduction to latent class

Identifying discrete groups within a heterogeneous population represents a major hurdle for public health researchers and policy makers. Across a population, individuals can respond quite differently to a set of questions about their health. However, a finite number of response patterns often occurs for the question set. With the appropriate statistical model, one can cluster individuals into heterogeneous classes based on different response patterns. In addition to using the response patterns, one can include demographic and other characteristics (covariates) to help predict class membership. Latent class mixture models seek to identify heterogeneous discrete classes of meaningful patterns in responses and to identify the likelihood, based on other variables, that an individual belongs to one class as opposed to another.

To illustrate this technique from an applied perspective, I use the patterns of responses associated with the reasons why parents of children with special health care needs (CSHCN) delay or forgo getting necessary health care and include a set of covariates to model class membership. I present the results from one of the primary software packages that fit latent class mixture models and show how three distinct classes appear. One class particularly represents a major challenge to policy makers, a class of CSHCN who generally endorse all reasons for delaying or forgoing care, who generally come from Hispanic homes, and generally do not have insurance. I conclude the example by demonstrating how latent class mixture models can address issues of social justice generally and within a population of vulnerable children specifically.