Generalized ordered logit analysis on the injury severity of traffic crashes by older drivers: the case of South Korea

ABSTRACT With a case of South Korea, this study attempts to analyze how the severity of traffic crashes for which older drivers are at fault is associated with road/environmental/seasonal/weather conditions, driver’s attributes, and crash/violation types. It analyzes the ordinal variable of injury severity by specifying conventionally used ordered logit and its statistical alternatives, multinomial logit, generalized ordered logit, partial proportional odds models. Based on the assumptions and fits of the models, this study finds that relative to the poor, but conventional ordered logit model, the best-fit generalized ordered logit model reports more accurate results in terms of the coefficient significance, direction, and magnitude. Among its major findings, although the severity decreases by year (conventional model), it does not apply to the minor injury level (alternative). The higher severity in the picnic season (conventional) is also limited to the level of minor injury or less. The times of the day at which the severity is higher (4–6) and lower (8–21) (conventional) are specific to fatal and serious injuries, respectively (alternative). Several variables have double-edged effects. Single-vehicle crashes do not always result in higher severity. They also reduce it to the lowest report-only level. Likewise, age and alcohol use change the severity to the most severe fatal level as well as to the most minor report-only level. These results are compared with the accumulated findings of crash frequency studies and interpreted in relation to older drivers’ high cautiousness and low physical/cognitive ability.