| This study initially investigates the relationship between the as-built roughness of asphaltic concrete pavement and construction characteristics factors using panel (longitudinal) data analysis. This panel study utilizes as-built roughness measurements and pavement and contractor's characteristics for reconstructed, replaced, and resurfaced projects in Wisconsin from 1996 through 2000. Several panel models, fixed and random effects models, were developed from the basic model proposed in the study. The results indicate that some pavement and construction characteristics significantly affect as-built pavement roughness. The findings also reveal that the history of qualities from past pavement projects is a statistically significant factor that affects as-built roughness and can possibly be converted to contractor's accumulated performance knowledge on pavement construction.; In order to develop a dynamic prediction model for as-built roughness, this study investigates the impacts of geographic location (urban and rural) and construction type (reconstruction, resurfacing, and replacement) on as-built roughness using explanatory data analysis (EDA) and panel data analysis (PDA). Sets of roughness and other data for reconstructed, replaced, and resurfaced projects in Wisconsin from 1998 and 2000 were used for the analysis. This research defines the significant factors and quantifies the interactive impacts of geographic location and construction type on as-built IRI (International Roughness Index). The results show that geographic location is strongly significant and panel data analysis has the better performance in tracing these impacts than cross-sectional regression analysis. With the results from two preliminary analyses, the study develops a dynamic prediction model for as-built IRI by utilizing the random effects model that makes the inferences possible about a population larger than the sample. In-sample and out-of sample specifications during model development process make it possible to validate the prediction model using Akaike's Information Criterion (AIC) and mean absolute percent error (MAPE). The results show that a dynamic model developed in the study predicts the as-built IRI within the acceptable MAPE range of 16.09 percents. |
