Improved Imputation of Missing Pavement Performance Data Using Auxiliary Variables

Abstract

Missing data in pavement condition and performance records of pavement management systems (PMS) are commonly encountered in practice. Imputation of missing data is often required in the analysis of pavement performance and decision making for maintenance and management of pavement networks. The traditional methods of handling missing data by pavement engineering professionals include deletion of affected records, and imputation of missing data by means of interpolation substitution, mean substitution, or regression substitution. Today, the advancement of computer technology has permitted the use of computationally complex stochastic methods of multiple imputation to improve the accuracy of imputed data. This study proposes an improved multiple imputation approach using a joint multivariate model on the understanding that, besides pavement performance data, a typical PMS database also collects data of related pavement properties and nonpavement variables such as traffic and weather conditions. The proposed approach develops an imputation strategy that uses selected pavement properties and nonpavement data as auxiliary variables in the multiple imputation analysis for missing pavement performance data. The theoretical basis and imputation procedure of the proposed approach are first presented, followed by a case study using the Long-Term Pavement Performance (LTPP) database to illustrate the choice of auxiliary variables and the steps involved in imputing missing rutting and roughness data respectively. The merits of the proposed approach are demonstrated by comparing the imputed results with actual measured data and by comparing the results with those obtained using the multiple imputation method without the inclusion of auxiliary variables.