Accounting for dependence among study results in Meta-Analysis: methodology and applications to the valuation and use of natural resources

Meta-analysis refers to the statistical analysis of empirical estimates obtained in previous studies, and is increasingly used in environmental and natural resource economics as a complement to a state-of-the-art literature review. The occurrence of dependence or auto-correlation among study results, for multiple estimates from the same study or for estimates from different studies, is a compelling problem that is usually ignored. This paper suggests that autocorrelation tests and estimators developed for other types of data constitute an appropriate solution to measuring and remedying dependence in meta-analysis. Moreover, visualization by means of a scatterplot provides a useful tool for the interpretation of dependence, and helps to detect outliers. The paper provides illustrations of the techniques through meta-analyses on the valuation of wetlands and the price elasticity of residential water demand. The applications show that between-study dependence is usually sufficiently modeled by means of variability in study characteristics. Ignoring within-study dependence, however, can result in biased estimators and makes inferences from meta-analyses imprecise in size and significante.