Comparison of Predictive Cost Models for Bridge Replacement Projects

Abstract

Predicting the cost of bridge replacement is essential to estimate present and future bridge funding needs. This paper presents a multilevel predictive cost model for bridge replacement projects with random effects. Both Frequentist and Bayesian approaches were implemented, and the results are compared with the ordinary least square methods (OLS). The data set used in the analysis included 190 bridge replacement projects completed between 1994 and 2008 in Oregon. The results from the analyses indicate that traditional regression models fail to fully capture the effect of certain attributes. The results of this study would indicate that alternative analytical approaches will provide more accurate predictions of costs. Specifically, as a result of this research, two different approaches, which account for both interaction and hierarchical effects for predicting bridge replacement costs, are identified.