Publication Bias in Measuring the Income Elasticity of Water Demand

We present the first meta-analysis of the income elasticity of water demand that accounts for the effects of publication selection (the preferential reporting of estimates that are intuitive and statistically significant). Paradoxically, more affected by publication selection are the otherwise preferable estimates that control for endogeneity. Because such estimates tend to be smaller and less precise, they are often statistically insignificant, which leads to more intense specification searching and bias. Correcting simultaneously for publication and endogeneity biases, we find that the underlying elasticity is approximately 0.15 or less. The result is robust to controlling for 30 other characteristics of the estimates and using Bayesian model averaging to account for model uncertainty. The differences in the reported estimates are systematically driven by differences in the tariff structure, regional coverage, data granularity, and control for temperature in the demand equation.