Issues in the statistical modeling of data with application to the skid resistance data
Previous research studies have indicated that polish value may not be a good predictor of actual skid performance of pavements. In this study, data on skid performance was used to build a model to predict skid numbers at 40 mph for different pavements in different climates and different regions of Texas. The data provided for analysis included mean British pendulum number, texture measurements, vehicle passes per lane, polish value, LA abrasion test results, magnesium sulfate test results, acid insoluble residue test results, chmate region and the mixture of aggregate in the pavement. This data was given for the years 1995 and 1996. Since there is no theoretical justification to fit a specific model to the SN(40) data, the strategy adopted for this analysis was to use the method of empirical model building to develop a suitable model for this data set. The techniques of exploratory data analysis are employed to get some ideas about the type of models that may be suitable for this data. The strategy at this stage was to choose the simplest candidate model and then use diagnostic techniques to improve the previous model. After improvement of each stage, diagnostic analyses were performed to check model adequacy or lack thereof.