Average Treatment Effect for Modeling Maintenance Work

The evaluation of the effect of maintenance and rehabilitation (M&R) activities plays a critical role in pavement management systems. A quantification of the treatment effect usually resorts to simple statistical regression analysis. However, these analyses often suffer from two major problems: (1) sample selectivity bias due to the nonrandom nature of treatment applied to pavement sections and (2) inaccurate or erroneous representation of the treatment effect due to the difficulty in determining the timing between M&R activity and performance inspection. In this paper, an econometric technique, average treatment effect (ATE) modeling, is applied to address these concerns. In addition to its capability in addressing the two aforementioned problems, ATE modeling includes other desirable properties: (1) for any section individual, the treatment effect can be estimated regardless of whether or not it received treatment and (2) the treatment effect for specific sections of interest can be evaluated. A case study with focus on highway pavement overlay treatment is presented. The treatment effect on varying flexible pavement structure is investigated. A structured econometric ATE model is constructed and estimated, and a traditional regression model is presented for comparison. The comparison between the two models suggests that the ATE model is more accurate and informative in addressing the pavement maintenance treatment effect. The results can lead to profound implications for M&R decision-making and optimal budget allocation in pavement design and management.