Publication Detail

Assessing Motor Carrier Driving Risk Using Cox's Semiparametric Model with Multiple Stop Effects

UCD-ITS-RR-92-10

Research Report

Suggested Citation:
Yang, Chun-Zin and Paul P. Jovanis (1992) Assessing Motor Carrier Driving Risk Using Cox's Semiparametric Model with Multiple Stop Effects. Institute of Transportation Studies, University of California, Davis, Research Report UCD-ITS-RR-92-10

Interstate motor carriers are subject to limitations on the hours that their drivers may be on duty and driving. They include a requirement that a driver be off duty for a minimum of 8 hours after driving for 10 hours or being on duty for 15 hours. There are also cumulative restrictions for on-duty time over several days: 70 hours on duty in 8 days for carriers operating 7 days a week and 60 hours in 7 days for those operating 5 days a week. These limitations, referred to as the hours-of-service regulations, were initiated in the 1930's. Since then the U. S. highway system has changed dramatically as has the nature of the trucking business and the technology of the vehicles. Despite these change, there have been rather limited attempts to assess the safety implications of the hours of service for contemporary conditions.

Pioneering research was conducted in this case in the 1970's. An example with U-shape risk curve from the research is illustrative of several important points. The vertical scale can be thought of as the probability of an accident while the horizontal scale is hours driving. The curve, drawn by hand and fit by eye, indicates an increase in risk beyond about 5 hours driving with very sharp increase in the last hour. The data points are constructed by taking the ratio of the observed number of accidents to the expected number where the expected number is taken from data indicating the length of the trip (with no accident). By assuming that accidents should be proportional to number of trucks in each time interval, an "expected" function is derived. Dividing it into the observed number should reveal if the number is greater than expected.

This clear use of data results in a trend line using accident data only, a major advantage. It relies on knowing the expected length of each trip, but know that, the data and curves can be derived. Note also that the approach accounts for what we will call the "survival effect" of all trips, including those with accidents. That is, a driver who has a accident in the 5th hour successfully completes the first 4.

Accurate models of accident risk and hours driving must take account of this survival effect. The weakness in the approach are the lack of statistical fit and inference and the fact that the sample contains accident-involved drivers only. Interestingly, data obtained from other firms reported in the same reference do not indicate the same trend and, in fact, in some cases reveal a decline in risk with hours driving. Despite these criticisms, the research by Harris and Mackie remains among the most important linking motor carrier accident risk and continuous driving.

Recent important research has examined sleeper berth operations and hours of service violations. Others have not adequately included the survival effect in their analysis, invalidating their conclusions concerning driving hours.

The groundbreaking experimental study of Mackie and Miller stands out as the most important research in the area to date. Their experimental results, while cargo loading effects were also considered, illustrate findings for regular and irregular schedules over several days using lane keeping as a performance measure. Similar results were obtained for other physiological measures.

Interestingly, the most frequent significant decline in performance occurred when the cumulative hours on duty over 8 days exceeded 70. This means the greatest declines occurred outside of the legal driver hours, perhaps indicating that imposed enforcement was a more effective accident deterrent than altering driving hours of service regulations.

Recent research has successfully used cluster analysis to extract limited sets of multiday (7 days) driving patterns from samples of accident and non-accident data. One paper then used the output of the cluster analysis as the input to a logistic regression model with discrete outcomes of accident or non-accident. Another paper extends the logistic regression model to include several covariate interaction terms with hours driven along with multiday clusters.