Unobserved confounders cannot explain over-crediting in avoided deforestation carbon projects

In ecology and conservation, a growing number of studies seek to draw causal inference using quasi-experimental designs. Despite the risk of omitted variable bias from such designs, the degree to which results are sensitive to unobserved confounders is seldom assessed. Here, to demonstrate the value of such sensitivity analyses, we use the controversy surrounding whether projects aimed at reducing emissions from deforestation and forest degradation (REDD+) overestimated their effectiveness (resulting in too many credits being sold). Verifiers of REDD+ credits have argued that independent quasi-experimental analyses of REDD+ projects are flawed because they omit site-specific drivers of deforestation. If these drivers also affect where REDD+ projects are established (that is, projects target areas facing threat), omitting them will tend to underestimate the deforestation that projects avoided. We revisit a global sample of 44 REDD+ projects and show that while some projects reduced deforestation, over-crediting was rife. Crucially, we explore the sensitivity of these results to unobserved confounders and demonstrate that unobserved local drivers of both deforestation and REDD+ locations are unlikely to fully account for reported over-crediting. Assessing sensitivity to unobserved confounders remains uncommon in ecology and conservation but should become standard practice where causal conclusions are based on controlling for confounders.