Meta-Modeling and Benefit Transfer: The Empirical Relevance of Source-Consistency in Welfare Measures

A common assumption within the meta-analysis and benefit transfer literature is that the validity of benefit predictions depends on the utility-theoretic consistency of welfare measures in underlying source studies. However, to date there exists little evidence as to the empirical relevance of this proposition in terms of the accuracy or efficiency of predicted benefits. Using Bayesian Model Search techniques we examine whether different portions of metadata, distinguished by underlying welfare construct, share common willingness-to-pay distributions. Applying our algorithm to two separate meta-datasets we find strong evidence of information sharing across welfare categories for a large subset of contexts. For cases where information sharing is indicated, substantial efficiency gains in predicted benefits can be achieved by pooling the underlying data.