Selectivity bias in modeling highway pavement maintenance effectiveness

In selecting the type of maintenance or rehabilitation (M+R) activity to apply to highway pavement sections, engineers rely on several items of information. Foremost among these is the effectiveness of the various M+R activities in slowing pavement deterioration and improving its condition. A number of researchers have sought to quantify the effectiveness of M+R activities by developing separate models of pavement deterioration for each type of M+R activity. Often, the results of such analyses have not been satisfactory. For example, some researchers have reported models with poor fits to data or counterintuitive signs of important variable coefficients. These results may be due to the fact that the sample used for model estimation was self-selected. The observations used to analyze the effectiveness of a certain M+R activity are not representative of the population of highway pavement sections, because these observations consist mainly of sections for which that particular activity was believed to be most effective. This paper presents a structured econometric approach for estimating the effectiveness of pavement M+R activities. This approach consists of a discrete model describing the choice of M+R activity by the highway agency, and a set of continuous models representing pavement response, with one equation for each activity. This model system accounts for the self-selected nature of the sample, and consequently obtains consistent model parameter estimates. A case study that demonstrates the applicability and value of the methodology is presented. The data set used for model estimation was provided by the Indiana Department of Transportation.