The popular opinion leader (POL) intervention aims to increase safe sex norms among a target population. It is based on the diffusion of innovation theory, which predicts that an idea can spread through a population or social system and be adopted by the social system (LaMorte, 2019). This idea then becomes a social norm, which in turn may influence behaviors. The POL intervention can be tailored to specific groups, which allows the challenges for the group to be addressed directly. This intervention has been implemented in populations such as male sex workers, migrant MSM (men who have sex with men), and women. The POL intervention was originally designed for MSM and tested in the United States among white men. The positive results in the United States have led some to try the intervention in other countries.
Among the included studies in our ongoing rapid systematic review of social norms-based interventions, we found two studies that evaluate POL interventions implemented in low- and middle-income countries (LMICs). Caceres et al. (2010) report the results of a multi-country evaluation (China, India, Peru, Russia and Zimbabwe), while Duan et al. (2013) evaluate a different POL implementation in China. The interventions utilize POLs in similar ways, so we review them together in this post.
The POL interventions in these studies
In China, Duan et al. recruited 20 POLs every three months based on surveys of MSM and trained them in four sessions. Study facilitators trained POLs on strategies to advocate for safe sex behaviors, communication and challenges when advocating to MSM peers. POLs delivered prevention messages at MSM venues, such as bars, nightclubs and other entertainment venues. Meetings were held every three months to provide updates, discuss challenges and share strategies. The control city took part in standard HIV prevention activities.
The study findings
The behavioral outcome for Caceres et al. is the proportion of people reporting unprotected sex with non-spousal/non live-in partners. In all five countries, there was a decrease (in some cases quite large) in reported unprotected sex in both the intervention and control groups from baseline to endline 24 months later. However, only in India was there a statistically significant difference in the change between intervention and control venues, and in the India case, it was actually the control venues that experienced the greater decrease. In other words, this study found null results from the intervention.
Duan et al. use the occurrence of unprotected sex as the primary behavioral outcome. There was significant pre-post decrease in having unprotected sex with a primary or casual partner in the intervention group. There was no change in this behavior in the control group. Condom use with primary and casual partners during last sex increased in the intervention group and decreased in the control group. We note that although the authors calculate statistical significance for pre-post differences within the intervention and control groups, they only compare differences between the intervention and control groups narratively, that is, they do not conduct statistical difference-in-difference tests. However, the differences between the intervention and control groups are quite large.
How strong are these findings?
While the Duan et al. study does find the intervention to be effective, we should view these findings with some caution. While the authors claim random assignment, the study includes only two clusters, that is, there were just two cities with one assigned to the intervention and one to the control. The authors collected data from different samples at baseline and endline for each city. The article reports statistically significant differences in almost all demographic characteristics between baseline and endline for the intervention samples and the control samples. It is also clear from the table in the article that there are large differences in demographic characteristics between the intervention samples and the control samples. For example, at baseline, 60.5% of the intervention sample and 47.5% of the control sample have college education or higher. At endline, 46.5% of the intervention sample and 28.5% of the control sample have college education or higher. These large differences likely arise from the snowball sampling method used to recruit survey participants for each sample; those who took the survey received 20 yuan RMB for up to three peers they recruited to complete the survey. It appears that the authors do not control for these demographic variables in their logistic regressions.
The Duan et al. study also measures two statistically significant negative outcomes for the control city, which raises additional doubts about the comparability of these small samples.
What can we take away from these studies?
However, the absence of evidence for the effectiveness of this norms-based intervention on measured behaviors is consistent with the findings presented here from another subset of studies from the rapid systematic review. As that post concludes, the limitations of the current evidence base for norms-based approaches to behavior change does not mean that norms are not a crucial element for understanding and changing behaviors, especially behaviors with a social aspect. It means that we still have much to learn about how to design effective norms-based programs to achieve detectable changes in behaviors, especially over a set period of time.
Photo credit: Thomas Barwick/Getty