Randomized controlled trials (RCTs) provide a methodological touchstone for empirical research across many scientific disciplines. Whether medicine, research in public health, or development economics, RCTs often represent one compelling way to take a setting high on external validity and draw strong causal inferences about the effects of some intervention.
Our claim is that interventions specifically designed to modify attitudes and behavior can have up to six different types of effects. Interventions of this sort are common in RCTs related to public health and development economics, but RCTs typically examine only one of the six effects, namely the effect on the behavior of primary interest among people having direct experience with the intervention. We will call this the “direct effect” of the intervention.
Though the direct effect of an intervention is certainly important, our goal is to examine systematically the other five effects a behavioral intervention can have. We will call these effects the “indirect effects” of the intervention. As explained below, indirect effects occur for at least two generic reasons. First, if an intervention leads a given individual to change her behavior in one decision-making domain, this change could easily produce changes in behavior in other decision-making domains for the same individual. Second, if an intervention changes one person’s behavior, the influence that person has on others may also lead other people to change their behaviors. These two classes of mechanism generate the five indirect effects at this center of our study. Most RCTs ignore indirect effects. Those that consider indirect effects do so in an ad hoc and incomplete fashion. Our study aims to show how a complete analysis of any behavioral RCT can and should systematically examine both the direct effect and all five indirect effects of the interventions at the heart of the RCT. Only then can the researcher claim to understand what the intervention actually does, which is essential information when deciding whether to scale up the intervention in society at large.