Citation

Keall MD, Newstead SV. Traffic Injury Prev. 2016; 17(2): 151-158.

Affiliation

a Wellington School of Medicine and Health Sciences, Otago University , PO Box 7343, Wellington South , New Zealand.

Copyright

(Copyright © 2016, Informa - Taylor and Francis Group)

DOI

10.1080/15389588.2015.1045064

PMID

26043773

Abstract

OBJECTIVE: Vehicle safety rating systems aim firstly to inform consumers about safe vehicle choices and secondly to encourage vehicle manufacturers to aspire to safer levels of vehicle performance. Primary rating systems (that measure the ability of a vehicle to assist the driver in avoiding crashes) have not been developed for a variety of reasons, mainly associated with the difficult task of disassociating driver behaviour and vehicle exposure characteristics from the estimation of crash involvement risk specific to a given vehicle. The aim of the current study was to explore different approaches to primary safety estimation, identifying which approaches (if any) may be most valid and most practical, given typical data that may be available for producing ratings.

METHODS: Data analysed consisted of crash data and motor vehicle registration data for the period 2003 to 2012: 21,643,864 observations (representing vehicle-years) and 135,578 crashed vehicles. Various logistic models were tested as a means to estimate primary safety: conditional models (conditioning on the vehicle owner over all vehicles owned); full models not conditioned on the owner, with all available owner and vehicle data; reduced models with few variables; induced exposure models; models that synthesised elements from the latter two models.

RESULTS: It was found that excluding young drivers (aged 25 and under) from all primary safety estimates attenuated some high risks estimated for make/model combinations favoured by young people. The conditional model had clear biases that made it unsuitable. Estimates from a reduced model based just on crash rates per year (but including an owner location variable) produced estimates that were generally similar to the full model, although there was more spread in the estimates. The best replication of the full model estimates was generated by a synthesis of the reduced model and an induced exposure model.

CONCLUSIONS: This study compared approaches to estimating primary safety that could mimic an analysis based on a very rich data set, using variables that are commonly available when registered fleet data are linked to crash data. This exploratory study has highlighted promising avenues for developing primary safety rating systems for vehicle makes and models.

Language: en