In-depth accident investigations: comparison of self-reported and reconstructed driving speeds

In-depth accident investigation programs generally involve a kinematic reconstruction of each crash, which makes it possible to estimate the speed of the vehicles involved at the time preceding the accident. In some cases, however, such a reconstruction is not possible due to the lack of sufficient material clues. Another possibility is to ask drivers involved in the crash an estimate of their driving speed just before the accident. To what extent can these self-reports be used as reasonable substitutes for estimates obtained from reconstruction? This paper presents a comparison of self-reported driving speeds with reconstructed driving speed, using a sample of cases where a robust reconstruction is possible. This sample comprises 73 cases from the in-depth accident study carried out by the French Institute of Science and Technology for Transport, Development and Networks at Salon de Provence (France). The self-reported speeds were drawn from the thorough drivers’ interviews carried out by psychologists just after the accident. The comparison between self-reported speed (VSR) and speed estimated from kinematic reconstruction (VKR) shows a good linear correlation, with a coefficient of determination R2 of 0.79. The average of the differences (VSRi – VKRi) is –0.53 km/h. However, the absolute differences |VSRi – VKRi| are larger than 5 km/h in 55% of the cases studied. Further analyses on the conditions that influence the difference between reported and reconstructed speed suggest that male drivers tend to overstate their speed (as compared to the speed estimated from reconstruction). A tendency toward understating is found when the actual speed, as estimated from reconstruction, exceeds the legal speed limit. Overall, although the correlation obtained is clearly better than correlations between reported and observed speeds mentioned by other authors, self-reported speeds do not seem to be very reliable substitutes for speed data based on material clues.