Designing Multisite Experiments Probing Mediation

• In Event: Hilton Roundtable Session 1
  In Roundtable Session: Methodological Advances in Experimental Design

Abstract

We develop a flexible framework to guide the statistical design of multisite mediation experiments. Despite the widespread use of multisite and multisite cluster randomized designs and the consistent presence of mediation models in the theories of action underlying most education studies, literature has offered little guidance on the design of such studies. We identify and develop key design considerations and track statistical power and requisite sample size under a wide range of settings. We highlight five of these considerations: (1) level of mediator (site/group- and/or individual-level), (2) optimal sample allocation under a given budget, (3) moderated mediation, (4) multiple mediators, and (5) three-level multisite cluster-randomized mediation. The framework is implemented in the R package BLINDED and the free BLINDED software.

Author

• Benjamin Kelcey, University of Cincinnati