Publication:
Statistical Models of Accidents
on Interchange Ramps & Speed-Change Lanes
FHWA-RD-97-106
Contact:
Joe Bared
(Joe.Bared@fhwa.dot.gov)
Abstract:
The objective of this research was to develop statistical models for defining the relationship between traffic accidents and highway geometric design elements and traffic volumes for interchange ramps and speed-change lanes. The data base used to develop the
models consisted of data for interchange ramps and speed-change lanes in the State of Washington and was obtained from the
FHWA Highway Safety Information System. Additional geometric design features were obtained from the review of interchange
diagrams. Data on other geometric design features, such as the ramp grades and horizontal curvature, were collected for a
sample of ramps from aerial photographs and other existing highway agency files.
The statistical modeling approaches used in the research included Poisson and negative binomial regression. Regression models
to determine relationships between accidents and the geometric design and traffic volume characteristics of ramps were difficult to
develop because the observed accident frequencies for most ramps and speed-change lanes are very low. The regression models
developed, based on the negative binomial distribution, explained between 10 and 42 percent of the variability in the accident data,
with the negative binomial distribution providing a poor to moderate fit to the data. However, most of that variability was explained
by ramp Annual Average Daily Traffic (AADT). Other variables found to be significant in some models included mainline freeway
AADT, area type (rural/urban), ramp type (on/off), ramp configuration, and combined length of ramp and speed-change lane.
The best models obtained for predicting accident frequencies were those obtained when modeling the combined accident
frequency for an entire ramp, together with its adjacent speed-change lanes. These models provided a better fit than separate
models for ramps and speed-change lanes. Models developed to predict total accidents generally performed slightly better than
did models to predict fatal and injury accidents.
Back to Safety Research
