Internal validation of near-crashes in naturalistic driving studies: a continuous and multivariate approach
Journal article, 2014

Large naturalistic driving studies give extremely detailed insight into how traffic accidents happen and what causes them. However, even in very large studies there are only relatively few crashes. Hence one additionally selects and studies crash surrogates, so called “near-crashes”, i.e. situations when a crash almost happened. The selection procedures invariably entail severe risks of causing bias. In this paper we use extreme value statistics to develop two methods to study the extent and form of this bias. The methods are applied to a large naturalistic driving study, the 100-car study. Both methods identified a severe discrepancy between the rear-striking near-crashes and the rear-striking crashes. Perhaps surprisingly, one conclusion is that, for rear-striking and in this study, the crashes have little relevance for increasing traffic safety. We believe substantial efforts should be made to develop statistical methods for using near-crashes and crashes in future large naturalistic driving studies (such as the SHRP2 study).

Selection bias

Naturalistic driving study

Crash surrogate

Extreme value statistics

Traffic safety

Rear-ending crash


Jenny Jonasson

Vehicle and Traffic Safety Centre at Chalmers

University of Gothenburg

Chalmers, Mathematical Sciences, Mathematical Statistics

Holger Rootzen

Chalmers, Mathematical Sciences, Mathematical Statistics

University of Gothenburg

Accident Analysis and Prevention

0001-4575 (ISSN)

Vol. 62 102-109

Areas of Advance



Basic sciences

Subject Categories

Vehicle Engineering

Probability Theory and Statistics



More information