Retrospective and Prospective Evaluation

Methods for evaluating vehicle-based safety measures can be categorized into prospective and retrospective [11-15]. The main difference is the time of the evaluation regarding the development process and/or life cycle of the measure in question [12], Prospective analysis can be used from very early stages of development on (without the necessity of having a fuUy developed measure), and retrospective analysis can be used once a measure has been developed (and usually has already been in the market for a given span of time) [15]. 

Retrospective analysis mainly uses real accident data and evaluates existing measures with respect to a safety statistic. A common procedure is to define two groups in the accident data, one with the measure in question and the other without. The two groups are then compared searching for changes in characteristic values of the statistic [14, 15]. 

There seems to be a consensus in the literature concerning the power of this approach as being both very important [14] and impressive [9]. Analysis of existing real-world accident data is even sometimes regarded as [a]n ideal method to assess the safety impact of advanced safety technologies [10]. The most prominent example for retrospective evaluation is the Electronic Stability Program (ESP) 

The importance of controlling confounders and interaction effects shall be pointed out by an example. Using the retrospective approach, one study evaluated the effects of xenon headlamps on accidents in Germany [20] on basis of the federal accident statistics. As a result, introducing xenon in 100 % of passenger vehicles would lead to a decrease in 6 % of all accidents and 18 % of all fatalities. The study claims that all possible confounders were taken into account and do not bias the results [20], The possible confounders cited, such as exposure time of vehicles, driver behavior 

2 State of Scientific and Technical Knowledge on Pre-crash Evaluation 

---

As both retrospective and prospective evaluation methods are based in many cases on accident data, a short summary concerning possibilities and limitations of accident data as well as other data sources is given in the following. These general findings have effects on the validity of each method discussed below and are not dependent on the specific method used. 

---