Experimental Design for Studies Probing Multilevel Mediation

A common line of inquiry in prevention research studies is to investigate the intermediate behaviors or attitudes that transmit the effects of an intervention on an outcome. In many instances, these types of investigations take place within multilevel contexts and, as a result, potentially include the examination of not only individual behaviors but also the collective behavior of individuals within a cluster. The predominant approach in these types of investigations is to test the sequence of relationships connecting an intervention, mediator, and outcome through a multilevel mediation framework (Pituch & Stapleton, 2012; Zhang, Zyphur, & Preacher, 2009). The goals of such studies are typically to unpack the processes whereby an intervention impacts an individual- or cluster-level mediator variable lying along the intervention-outcome pathway and how those changes in the mediator are then translated into changes in an outcome. Such inquiries advance scientific theory by building a multilayered body of evidence regarding if, how, and why an intervention impacts an outcome. Investigation of the transmission channels of an intervention can help develop component-specific evidence regarding the theory of action guiding that intervention as well as evidence regarding the effectiveness of the entire system or intervention. For these reasons, (multilevel) mediational analyses have been widely used across many disciplines.

Despite the critical role mediation analyses play in prevention research and in many other areas, literature guiding researchers as to the intentional and strategic design of studies probing multilevel mediation has been very limited. In this study, we derive and develop a flexible framework to direct the effective and efficient design of multilevel mediation studies under a wide range of considerations. We highlight and delineate strategies for eight of these considerations: (1) level of mediator (group- and/or individual-level), (2) latent variables (e.g., use of structural equation models), (3) optimal sample allocation (e.g., should we sample more groups or individual within groups), (4) moderated mediation (e.g., interactions between treatment and mediator), (5) multiple mediators (e.g., multiple parallel or sequential mediators), (6) three-level mediation (e.g., introduction of multiple levels of nesting), (7) sensitivity of derivations to parameter value misspecifications (e.g., underestimates of the intraclass correlation coefficient), and (8) relative performance of asymptotic, re-sampling, and Bayesian tests. The framework is implemented in the R library BLINDED and the free BLINDED software.