This article discusses ways in which randomized controlled trials do not accurately measure the impact of HIV behavioral interventions. This is because: 1. Such trials measure the wrong outcomes. Behavior change may have little to do with changes in HIV incidence since behavior change in events between HIV-concordant people have no impact on incidence. Even more important, the comparison of HIV incidence rates between study arms of individual-level RCTs does not measure the true outcome of interest-whether or not the intervention reduces HIV transmission at the community level. This is because this comparison cannot measure the extent to which the intervention stops transmission by HIV-infected people in the study to those outside it. (And this is made even worse if HIV-infected are excluded from the evaluation of the intervention.) 2. There are potential harms implicit in most cognitively oriented behavioral interventions that are not measured in current practice and may not be measurable using RCTs. Intervention trials often reinforce norms and values of individual self-protection. They rarely if ever measure whether doing this reduces community trust, solidarity, cohesion, organization, or activism in ways that might facilitate HIV transmission. 3. Many interventions are not best conceived of as interventions with individuals but rather with networks, cultures of risks, or communities. As such, randomizing individuals leads to effective interventions that diffuse protection through a community; but these are evaluated as ineffective because the changes diffuse to the control arm, which leads to systematic and erroneous reductions in the evaluated effectiveness as RCTs measure it. The paper ends by discussing research designs that are superior to individual-level RCTs at measuring whether an intervention reduces or increases new HIV transmission.
Keywords: HIV transmission; RCTs; behavioral interventions; effectiveness; randomized controlled trials; reducing infection; research design.