Validation of two types of models are discussed: functional and structural. The former are predictive, the latter presume to represent reality with great accuracy. To validate a model, its predictions must be compared with observable data, which calls for accurate use of statistics. The application of inferential statistics is briefly discussed, along with the problems of experimental design.
The author concludes that modeling can be an effective tool in automotive safety research, but it must be carefully coordinated with accident data collection. It is in the latter area that more work is needed.