Bayesian Estimation of Fuel Economy Potential Due to Technology Improvements

Richard Andrews, James Berger, and Murray Smith

We address the question of prediction of the amount of improvement in fuel economy of vehicles that can be achieved through further incorporation of existing technology. The data used is EPA fuel economy data, which also includes quite complete descriptions of the vehicles tested. A Bayesian analysis is implemented using hierarchical mixed models and Gibbs sampling. Models are developed, in part, through engineering reasoning. The analysis includes a very large subjective elicitation of high-dimensional distributions, and considers decision-theoretic prediction. Postscript File (742kB)