Session: Innovative Causal Inference Approaches and Applications in Prevention Science: A Focus on Propensity Score Matching and Cace (Society for Prevention Research 23rd Annual Meeting)

Many studies in the field of prevention science aim to understand issues of causality. However, researchers are often limited in their ability to design studies in a way in which causal inferences can be drawn. A number of methods have been developed to assist researchers in estimating causal effects. This presentation provides an overview of 2 types of approaches to causal inference: propensity score matching and the complier-average treatment effect (CACE) approach. Specifically, the first paper focuses on an assumption of existing propensity score methods, which is that the covariates are measured without error; however, covariate measurement error is common, especially in many prevention science research studies where constructs (e.g., mental health problems) cannot be measured directly. The purpose of this first talk is to suggest an alternative propensity score matching approach called Multiple Imputation using External Calibration (MIEC), which addresses some of the issues associated with covariate measurement error. Findings from a series of simulation studies are presented. The second study also uses propensity score modeling, but multilevel data in the context of a randomized controlled trial. This study seeks to identify the optimal form of the propensity score model, given the multilevel nature of the intervention. The authors present findings from a simulation study comparing propensity score models for a continuous mediator using a multi-level model, a single-level model with fixed effects, and a single-level model without fixed effects. The third study focuses on CACE approach for examining the impact of an intervention within the context of a randomized controlled trial. Specifically, this study uses data from a classroom-based preventive intervention in elementary schools, in which there was variation in the implementation of the program. This paper first presents findings from a traditional intent-to-treat (ITT) approach, which estimated the effect of being assigned to the treatment condition. However, in the case of low teacher compliance, the effect of being assigned to treatment may substantially differ from the effects on those who are assigned and participate and ITT effects may be underestimated (Stuart, Perry, Le, & Ialongo, 2008). The paper then presents findings from CACE, which provides an estimate of the treatment effect accounting for noncompliance, thereby showing that larger effect sizes in the CACE analysis than in the ITT. Implications of the 3 papers are considered by a leading prevention science methodologist, who highlights strengths and limitations of both propensity score matching and CACE.