Using Ecometric Data to Explore Sources of Cross-Site Impact Variance in Multi-Site Trials

Abstract

A new method is proposed to explore sources of cross-site impact variance in multi-site trials of social interventions. With this approach, aggregate reports from participants in the treatment arm about the treatment experience are used to define potential mediators and aggregate reports from participants in the control arm about the outcome of interest are used to remove confounding due to unmeasured local contextual factors. Particular attention is paid to correcting bias due to measurement error in the mediator. Asymptotic theory for the validity of this approach is provided. Two example applications are given, one simulated and one empirical.