Repeated Measurement Data Analysis in Pavement Deterioration Modeling

This paper considers the nature of repeated measurement data under the framework of the mixed-effects approach with an extension of an empirical nonlinear model of rutting progression on asphalt concrete pavements. The parameters estimates using the mixed-effects approach, which allow some parameters to vary randomly, are found to differ significantly from ordinary least-squares (OLS) and random-effects (RE) estimates. OLS ignores unobserved heterogeneity whereas RE attempts to capture it through the model's constant term. However, unobserved pavement sections' heterogeneity may manifest not only through model intercepts but through other model parameters as well, which makes the mixed-effects approach more appropriate for the analysis. The results obtained with the mixed-effects approach are more in accordance with a priori expectations. The paper also presents important modifications to the specification of an empirical rutting model that deals with different load magnitudes. The modified specification is found to be appropriate for modeling material hardening with loading. The results point out a potential problem with the traditional approach of dealing with mixed traffic loads in the estimation of models for predicting rutting progression.