2019 CSCE Annual Conference - Laval (Greater Montreal) Conference
Dr. Susan Tighe, CPATT - University of Waterloo
Maintaining the safety of the highways in the closures during the construction, maintenance, and rehabilitation activities is crucial. Throughout the years, several studies have been conducted to identify the influences of various factors on the injury-severity level of the collisions occurring on the highways. This study collected historical data from four different US states (New York, Pennsylvania, Illinois, and Michigan), between 2014 and 2016. Selected states typically have similar weather condition, pavement condition, and construction policies and regulations as Ontario, Canada. Authors believe because of stated similarities between these US states and Ontario, the spatial transferability of the developed statistical models is expected.
Developing statistical models are one of the tools which can identify the factors which significantly have an influence on the injury-severity level. Also, the clear and practical results of these models could be adopted by the contractors in order to improve the level of safety in the work zones. In recent years, several complicated and advanced statistical models were applied to injury-severity data. The main concern related to these types of models is that they are not time efficient, practical, and sometimes hard to interpret by non-statistical experts. Also, these models are intended to have the issue of overfitting which limits the ability of the models to predict future events. In addition, the fact that fixed parameter models are not accounting for unobserved heterogeneity is not neglectable. Therefore, this paper aims to apply the random parameter concept to some of the well-known statistical models in order to overcome issues from both unnecessarily complicated models and statistically insufficient methodologies. Random parameter ordered Probit, random parameter ordered Logit, and random parameter ordered Gompertz will be developed to address stated issues. Then, a comparative assessment among all ordered models will be conducted to investigate which one of these models have a statistical dominance over the other ones. In addition, these models will be used to develop an Excel-based system which can be adapted by non-statistical experts to operate, understand, and plan.
This study intended to present a straightforward methodology, which addresses previous concerns in this field. An Excel-based system as the fruit of this study can interpret the effect of each significant factor as well as predicting the possible severity level of future collisions in the work zones.