Bayesian exchangeability, benefit transfer, and research efficiency

Abstract

We offer an economic model of the policymaker's site‐ or time‐specific benefit estimate extrapolation problem when she must weigh the potential gains from an increase in the accuracy and the precision of her agents' estimates against the costs of conducting their assessments. If Bayesian exchangeability is treated as a maintained hypothesis, we suggest that empirical Bayes estimators offer a powerful way to increase the economic efficiency of extrapolation. Finally, we employ a hedonic study of pollution control benefits to illustrate a Bayesian diagnostic that allows the hypothesis of exchangeability to be tested rather than taken as maintained. The power of the diagnostic arises from its ability to identify those sources of parameter variability most likely to discourage extrapolations. Copyright 1992 by the American Geophysical Union.