Page 18 - tmp_Neat

severity injury or non-injury to fatal injury. On the other hand, the unordered multinomial

discrete outcome does not consider ordinal severity injury datasets.

Generally, the injury severity of crashes is categorised as fatal, incapacitating injury, non-
incapacitating injury, possible injury, and property damage only. In general, a sample size

smaller than 1000 should not be used for crash severity model development, and sample sizes
should be larger than 1000 for the ordered logit model and 2000 for the multinomial logit

model (Ye & Lord 2014).

A list of crash severity models which focus on heavy vehicle and non-heavy vehicle crashes
is shown in Table 2.1.

Table 2.1 Summary of existing crash severity models focusing on heavy-vehicle crashes

Crash Severity Models

Models applied in studies focusing on Models applied in studies focusing on non-
heavy vehicle crashes
heavy vehicle crashes

Ordered logit and ordered probit Binary logit and probit models

Nested logit Bayesian hierarchical binary logit

Simultaneous logit model Bivariate and multivariate binary outcome

Multinomial logit Copula-based multivariate ordered probit

Heteroskedastisc ordered probit and logit Bivariate ordered probit model

Mixed logit Generalized ordered logit

Partial proportional odds Bayesian ordered probit, mixed generalized

ordered logit and mixed ordered logit

Classification and regression tree Markov switching multinomial logit

Sequential logit and probit outcome

Artificial neural network

7