Estimation of truncated ordinal regression models

A major obstacle to the regression analysis of road traffic fatality data is that data is typically only recorded for accidents where at least one fatality occurs. Examples are the FARS database in the USA and the Fatal File of the Federal Office of Road Safety. Data of this type is called group truncated data. A regression technique should allow for this truncation if it is to avoid serious biases. Two existing methods are conditional logistic regression (CLR) and double pair comparisons (DPC). This report discusses the implementation of a new procedure called truncated ordinal regression (TOR). The technique is more general and efficient than the existing methods. It allows for an ordinal scale of injury such as: uninjured, moderate injury, severe injury, dead. The software consists of an Splus interface to a suite of C routines. Help files, installation scripts and an example are provided with the software. As a more complicated example of its use, TOR is applied to the Fatal File in Appendix B.