No evidence that nonincentivized behavioral interventions effectively mitigate climate change after adjusting for publication bias

Abstract

Behavioral interventions on citizens are often promoted as a low-cost route to induce environmentally friendly behavior, yet published estimates of their effectiveness are highly variable and prone to selective reporting. We reanalyzed the evidence of nonincentivized behavioral interventions on citizens. We applied robust Bayesian meta-analysis (RoBMA), averaging across a full set of publication bias-adjusted models, to the 144 effect estimates (91 studies) compiled by Nisa et al. (2019). After accounting for publication bias and model uncertainty using multilevel RoBMA, the data strongly favor a zero average effect. The posterior probability that the meta-analytic mean equals zero is 0.984, and the Bayes factor comparing a zero mean to a nonzero mean is BF01 = 63.5. Accordingly, the previously reported mean benefit of behavioral interventions on households and individuals may largely reflect publication bias and potentially other small-study effects. There is evidence for small between-study heterogeneity, indicating that some specific interventions might have an effect. These results suggest that, on average, behavioral interventions without incentives on households and individuals are unlikely to deliver material climate benefits.