Log-linear modeling is advanced as a procedure to identify factors that underlie the relative frequency of occurrence of various accident characteristics, such as accident type, location, and severity. The method is centered on the estimation of saturated log-linear models for pairs of accident variables and determination of indices of association between categories of the variables. Using data drawn from more than 9, 000 truck-involved accidents that occurred over a 2-year period on freeways in three metropolitan counties in southern california, the method is demonstrated by analyzing accident characteristics both by type and by freeway route segment. Accidents by collision type are analyzed relative to characteristics such as the primary collision factor, the location of the accident, the time period, road conditions, and weather conditions. Differences among 38 specific freeway segments in terms of accident characteristics are also analyzed. The results of the analyses indicate that the method is a useful tool in uncovering underlying patterns in accident characteristics.
Abstract