Network-Level Pavement Performance Prediction Model Incorporating Censored Data

Abstract

A methodology is presented for network-level pavement performance prediction that incorporates censored condition data. Censoring occurs when the duration at a specific condition level is not completely observed. This happens when pavement condition is improved and for the duration of the latest condition rating on file for each highway section. Pavement condition history files may contain significant quantities of censored data, yet such data typically are excluded when performance curves are developed. As a result, estimated condition durations and corresponding deterioration curves include deterioration rates that are greater, sometimes substantially greater, than those actually observed. The primary purpose for developing the presented methodology is to correct this shortcoming. Methodology development was facilitated with the use of a comprehensive information basis containing up to 20 years of historical pavement condition data for approximately 19,000 highway sections maintained by the New York State Department of Transportation. Durations at each condition rating were determined for each highway section over the 20-year period, with distinctions made between censored and uncensored observations. A modeling approach, with probability plotting and parameter estimation, was developed that resulted in performance curves. Differences in pavement performance based on geographic region were also investigated. From results obtained with the developed methodology, the main conclusion of this study is that accommodating censored data in pavement performance prediction models not only is feasible but better describes actual performance than if the data were simply excluded from the analysis.