Inconsistencies of Ordered and Unordered Probability Models for Pedestrian Injury Severity

You are here: Homepage > Search

Inconsistencies of Ordered and Unordered Probability Models for Pedestrian Injury Severity

Kwigizile, Valerian / Sando, Thobias / Chimba, Deo

Data on crashes between a single vehicle and a pedestrian recorded in Florida from 2004 to 2008 were used to identify the factors affecting the level of pedestrian injury severity, given that an accident had occurred and to assess the consistency of the ordered (ordered probit) and the unordered (multinomial logit) models. Both models were applied to the same data set. For the impact of individual variables on the levels of pedestrian injury severity to be discerned for the ordered probit and the multinomial logit models, the marginal effects were calculated. The results of a comparison of the two models indicated that the two models were consistent when they suggested the impact of individual factors on the lowest and the highest levels of injury severity (no injury or a possible injury and fatal injury, respectively) but suggested opposing impacts for some factors on intermediate levels of injury severity (nonincapacitating and incapacitating injuries). Such an inconsistency has implications for pedestrian safety measures and policies that are based on the models. Therefore, cautious selection of ordered and unordered probability models should be exercised with the use of a trade-off between recognition of the ordered pedestrian injury outcomes and loss of the flexibility in specification offered by unordered probability models. However, because the models are consistent for determination of the impact of variables on the lowest and the highest outcomes, pedestrian safety measures and policies should be derived on the basis of these outcomes.