Statistical Analysis of In-Service Pavement Performance Data for LTPP SPS-1 and SPS-2 Experiments

Observational or experimental studies are designed to investigate the effects of various factors on a response variable. This distinction is important because the latter studies are assumed to provide a firmer basis for establishing cause-and-effect relationships. However, experimental studies involving in-service pavement sections present certain concerns in statistical analyses, which are addressed in this paper. The challenges presented by the in-service pavements data included: (1) outlier issues; (2) quantification of performance; and (3) the lack of measurable distresses due to the “young” age of test sections. Experiment-related issues included: (1) wide variation in traffic levels and ages among the test sites and (2) an unbalanced distribution of test sites among climatic zones and subgrade types. The importance of selecting appropriate analytical methods for obtaining reliable results is discussed in this paper. Though most of the methods that were applied for the analyses are well established, the choice of magnitude—versus frequency-based methods was driven by the extent and occurrence of distresses. Based on the data, frequency-based methods such as linear discriminant analysis and binary logistic regression lend themselves well to explaining trends associated with distresses with reasonable occurrence but lower magnitude while a magnitude-based method like analysis of variance is more appropriate for evaluating distresses with high numbers of occurrence and magnitude.