Analysis of collision injuries with consideration of selectivity bias in linked police-hospital data
The Maximum Abbreviated Injury Scale (MAIS), which is utilized in linked police-hospital data, is a better estimation of severity than the KABCO scale. However, the issues of sample selection should be taken into consideration while using the linked police-hospital data that generates MAIS. Past studies have overlooked this issue in the injury severity models involving MAIS. A bivariate sample selection model is the established method for mitigating the selection bias. This study conducted a Monte Carlo simulation to investigate the sample selection issues in police-hospital linked data. Three alternative model specifications for a bivariate ordered probit model were compared with the univariate ordered probit model. The parameters were compared at different censoring levels, and at different correlations between the errors in sample selection and outcome equations. The results show that the univariate model computed biased estimates and the magnitude of bias increased with higher levels of censoring and correlation between the errors. Pedestrian injury severity analysis in Indiana was demonstrated as a case study. Certain important factors, such as pedestrian actions, weather variables, road type, and functional classification, were confirmed in the case study. The injury analysis was also extended to injury by body regions. The results of this study can assist to precisely estimate injury outcome by hospital data; provide a better understanding of factors affecting different body parts; and help comprehend some relevant updating process for the KABCO or MAIS injury scales.
Tarko, Purdue University.
Statistics|Civil engineering|Transportation planning
Off-Campus Purdue Users:
To access this dissertation, please log in to our