Statistical Modeling in Pavement Management

Abstract

Statistical analysis is one of the key aspects of engineering studies by researchers and transportation agencies. It is common to use regression analysis to develop a relationship between a response and one or more regressor variables. However, in many instances correlations between different variables are suggested and used for the prediction of an event without proper evaluation or interpretation of the developed models. This procedure may lead to gross errors, particularly when many senior engineers are retiring and leaving behind successors with limited skills. The issue is further complicated because engineers receive limited exposure to statistics during their undergraduate program and consequently must learn on the job; this situation can result in some key aspects to statistical analysis being missed. For example, the coefficient of determination, which can be meaningless in many circumstances, is used by many people to evaluate the adequacy of a model. Points of intersection of two or more trend lines are sometimes taken as the optimum value of various criteria without considering practical or engineering aspects. This paper focuses on some important aspects of statistical analysis and shows through various pavement engineering and management models how a model and resulting conclusion can be misleading. Proper evaluation of the causal relationship among variables and interpretation of the model are important to support a model's usefulness as well as to draw meaningful conclusions from it. A simple and concise approach is presented to aid researchers and practitioners with regression modeling.