Session: Estimating the Causal Effect of a Latent Class Treatment On Distal Behavioral and Health Outcomes (Society for Prevention Research 21st Annual Meeting)

Causal inference methods, including propensity score techniques, are being used with increasing frequency in prevention science to estimate the causal effect of an observed intervention or exposure on an outcome. However, to date it has not been possible to use these methods when the exposure of interest is a latent class variable measured by several indicators. This symposium introduces, extends, and applies a new solution to the methodological challenge of estimating the causal effect of latent class membership on a distal outcome.

In the context of three empirical examples, this symposium will present an introduction to the latent class causal analysis method proposed by Schafer and Kang (2010). The three talks comprising this symposium represent the first-ever applications of this imputation-based approach to real data. Originally proposed for continuous outcomes, and extended here for binary and count outcomes, latent class causal analysis involves specifying a model for selection into the latent classes and a model for the imputation of potential outcomes. This is conceptually similar to creating a propensity score model.

All three talks focus on providing a solid understanding of this new methodology in the context of addressing key questions about the etiology of substance use and the prevention of related negative outcomes. The first speaker will introduce the latent class causal analysis method for continuous outcomes, and will extend this method to binary and count outcomes in the context of estimating the causal effect of adolescent smoking behavior patterns on young adult regular smoking. The second speaker will continue this discussion by using the method to determine the effectiveness of patterns of non-randomized substance abuse treatment components on substance abuse outcomes. The third speaker will examine the causal effect of adolescent substance use behavior patterns on depression symptoms during emerging adulthood, while also conducting a brief software demonstration of the implementation of the method in the R package LCCA. The discussant is a world-renowned expert in statistical methods for prevention research; he will bring the three talks together by highlighting methodological implications of this approach and opportunities for prevention scientists.