Empirical Comparison of Statistical Pavement Performance Models

We conduct an empirical comparison of nine representative statistical pavement performance models, which have been estimated using serviceability data from the AASHO Road Test. The purpose of the study is to understand the effect of different statistical assumptions and estimation techniques on the models' predictive capabilities. The study consists of using the models to predict the serviceability of a common subset of pavements from the test over a two-year forecast horizon. Comparison of the models is carried out using both aggregate- and disaggregate-level measures. The main insight that stems from our study is that models that account for heterogeneity lead to improved predictive capabilities. This is particularly important because the data were obtained from a controlled experiment. The study further shows that disaggreggate measures are more useful in testing performance models because they are better indicators of the ability to capture the variability in data. Throughout the paper, we also discuss other model characteristics and issues that can be useful in the future development and evaluation of performance models.