UC Berkeley

Pavement deterioration models are an important part of any pavement management system. Many of these models suffer from endogeneity bias because of the inclusion of independent variables correlated with unobserved factors, which are captured by the model's error terms. Examples of such endogenous variables include pavement overlay thickness and maintenance and rehabilitation activities, both of which are not randomly chosen but are in fact decision variables selected by pavement engineers based on field conditions. Inclusion of these variables in a pavement deterioration model can result in biased and inconsistent model parameter estimates, leading to incorrect insights. Previous research has shown that continuous endogenous variables, such as pavement overlay thickness, can be corrected using auxiliary models to replace the endogenous variable with an instrumented variable that has lower correlation with the unobserved error term. Discrete endogenous variables, such as the type of maintenance and rehabilitation activities, have been accounted for by modeling the likelihood of each potential outcome and developing individual deterioration models for each of the potential responses. This paper proposes an alternative approach to accommodate discrete endogenous variables-the selectivity correction method-that allows a single model to incorporate the impacts of all discrete choices. This approach is applied to develop a pavement-roughness progression model that incorporates both continuous and discrete endogenous variables using field data from Washington State. The result is a roughness progression model with consistent parameter estimates, which have more realistic values than those obtained in previous studies that used the same data.