Social Work Research Applications of Instrumental Variable Estimation: Designs That Address Compliance, Dosage, and Mediation

Objective: Social work researchers conducting randomized experiments of innovative interventions commonly face the problem of imperfect treatment compliance, questions about the relationship between dose and response, or the causal effect of intermediate processes. These are important issues, and social work researchers require the appropriate statistical tools in such scenarios. Identifying the appropriate approach is crucial because the applicability of statistical tools varies by design and the nature of the effect to be estimated. Description: One method for estimating a treatment on its treated effect is to use the ratio of the intent-to-treat effect to the proportion compliant with treatment assignment. A framework derived from instrumental variable estimation and the Rubin causal model extends this method to situations including dose-response and intermediate processes. We discuss the assumptions underlying these techniques, present an in-depth description of how these assumptions enable causal inferences, and explain the key differences between compliance, dosage, and intermediate processes. We then discuss the method in the wider context of research design and the need for formative evaluation and flexibility.