Evidence building and information accumulation: Using the bayesian paradigm to advance child welfare intervention research

Objective: Intervention evaluation typically follows a frequentist paradigm: New analyses are conducted for each subsequent study, and findings are then used to improve policy practice. This approach largely ignores data from prior studies, leading to information loss and incomplete or inaccurate conclusions. Unlike the frequentist paradigm, the Bayesian paradigm uses formative data (as pri-ors), which can be updated with the summative data, thus building on existing evidence about an intervention’s effectiveness. Method: This article uses data from the Safe Families for Children randomized controlled trial to illustrate how the Bayesian paradigm incorporates prior evidence at the formative phase with data at the summative phase to provide a more comprehensive analysis. This approach is consistent with the scientific principle of evidence building. We compare the merits of each paradigm on two evaluation criteria: (a) p-values from a chi-square test, and (b) the probability that the intervention is superior to the comparison group on three outcome variables (protective custody, deflection from foster care, and whether repeat victimization occurred). Results: The Bayesian paradigm consistently outperformed the frequentist paradigm. Conclusion: The Bayesian paradigm is superior to the frequentist paradigm in demonstrating the effectiveness of an intervention, as evidenced by smaller p-values and a higher probability that the intervention group outperformed the comparison group.