Introduction to Bayesian Modeling in Intervention Research

Background:  Drawing on theory and prior research, intervention research focuses on the design and development social and health programs. A central feature of intervention research is the sequential testing of programs, where knowledge about an intervention accrues over time from single-group studies to complex randomized control trials (e.g., cluster randomized designs). Based on sample sizes from traditional power estimates, intervention researchers typically recruit study participants, and run trials. When a study is completed, statistical analyses are undertaken and recommendations are made. This is called a frequentist approach. In the analysis, the frequentist approach ignores developing knowledge on effects of interventions and, because analyses are not conditioned on prior information, they are inefficient. A logical extension of the frequentist approach is to incorporate prior knowledge into analyses by using Bayesian modeling.

Purpose: The aim of this workshop is to demonstrate Bayesian modeling, which has the potential to produce more efficient and more powerful statistical analysis through the incorporation of prior information – a posterior distribution – into the data likelihood function.

Contents: This workshop will: (a) review five steps in intervention research; (b) discuss traditional research design with effect size, power analysis and sample-size determinations, and standard analytics such as regression, multi-level models, and SEM; and (c) introduce and illustrate the Bayesian modeling for incorporating prior knowledge in the data likelihood function. Traditional and Bayesian analytic approaches will be compared

Pedagogical Techniques: This workshop will use a PowerPoint presentation to review intervention research methods and traditional statistical principles along with the Bayesian modeling perspective. Bayesian modeling will be demonstrated in the free software R, and participants will receive instructional handouts showing R code for Bayesian models.

Significance:  This workshop aims to introduce Bayesian modeling in intervention research, a topic that we think will be of great interest to many social workers. Case examples will illustrate Bayesian modeling in the free software R. This workshop will help to improve the rigor of quantitative research in social work, and, more distally, contribute to advances in research for social work practice.