Publication Detail
Modeling the Effect of Driver Service Hours on Motor Carrier Accident Risk Using Time Dependent Logistic Regression
|
UCD-ITS-RR-93-04 Research Report |
Suggested Citation:
Lin, Tzuoo-Ding, Paul P. Jovanis, Chun-Zin Yang (1993) Modeling the Effect of Driver Service Hours on Motor Carrier Accident Risk Using Time Dependent Logistic Regression. Institute of Transportation Studies, University of California, Davis, Research Report UCD-ITS-RR-93-04
Presented at the Transportation Research Board 72nd Annual Meeting
A time dependent logistic regression model has been formulated to assess the safety of motor carrier operations. The model is flexible, allowing the inclusion of time independent covariates, time main effects and time-related interactions. The model is used to test the safety implications of current U.S. D.G.T. driving hours of service policies using a data set from a national less than truckload carrier. The model estimates the probability of having an accident at time interval t, subject to surviving (i.e. not having an accident) before that time interval. Individual accidents are statistically compared to a random sample of individual non-accident trips by estimating a logistic regression model with two outcomes: an accident or non-accident. Covariates tested in the model include consecutive driving time, multiday driving pattern over a 7 day period (including time of day of driving and days driving within the 7 day period), driver age, driver experience and hours off-duty prior to the trip of interest.
Using accident and non-accident data for 1984, from one national less than truckload carrier, eight logistic regression models are estimated which include time independent effects (i.e. age, experience, multiday driving pattern and off-duty time prior to the trip of interest), main time effects (the driving time) and a series of time-related interactions. Driving time has the strongest direct effect on accident risk. The first four hours consistently have the lowest accident risk and are indistinguishable from each other. Accident risk increases significantly after the fourth hour, by approximately 50% or more, until the seventh hour. The eighth and ninth hours show a further increase. approximately 80% and 130% higher than the first 4 hours. Multiday driving patterns did have an effect on subsequent accident risk but the effect was not nearly as strong as for driving time. Daytime driving, particularly in the three days prior to the day of interest (pattern 2), results in a significantly lower risk of an accident on the subsequent day. Six driving patterns have accident risk about 40-50% higher than pattern 2; one of these was infrequently scheduled drivers. Of the remaining 5 multiday patterns, 4 involve some type of night driving. This implies a generally elevated collision risk at night.
A time dependent logistic regression model has been formulated to assess the safety of motor carrier operations. The model is flexible, allowing the inclusion of time independent covariates, time main effects and time-related interactions. The model is used to test the safety implications of current U.S. D.G.T. driving hours of service policies using a data set from a national less than truckload carrier. The model estimates the probability of having an accident at time interval t, subject to surviving (i.e. not having an accident) before that time interval. Individual accidents are statistically compared to a random sample of individual non-accident trips by estimating a logistic regression model with two outcomes: an accident or non-accident. Covariates tested in the model include consecutive driving time, multiday driving pattern over a 7 day period (including time of day of driving and days driving within the 7 day period), driver age, driver experience and hours off-duty prior to the trip of interest.
Using accident and non-accident data for 1984, from one national less than truckload carrier, eight logistic regression models are estimated which include time independent effects (i.e. age, experience, multiday driving pattern and off-duty time prior to the trip of interest), main time effects (the driving time) and a series of time-related interactions. Driving time has the strongest direct effect on accident risk. The first four hours consistently have the lowest accident risk and are indistinguishable from each other. Accident risk increases significantly after the fourth hour, by approximately 50% or more, until the seventh hour. The eighth and ninth hours show a further increase. approximately 80% and 130% higher than the first 4 hours. Multiday driving patterns did have an effect on subsequent accident risk but the effect was not nearly as strong as for driving time. Daytime driving, particularly in the three days prior to the day of interest (pattern 2), results in a significantly lower risk of an accident on the subsequent day. Six driving patterns have accident risk about 40-50% higher than pattern 2; one of these was infrequently scheduled drivers. Of the remaining 5 multiday patterns, 4 involve some type of night driving. This implies a generally elevated collision risk at night.