At the time of this writing, the United States and the European Union have entered into negotiation of the Transatlantic Trade and Investment Partnership (TTIP). This agreement is designed to reduce barriers to trade between the two economic units. One barrier to trade is the differing safety standards testing and requirements for vehicles sold in the EU and the US. Testing the same make/model under both regimens and adapting design to each can be expensive, and negotiation of common standards may be difficult and time-consuming. An alternative to item-by-item harmonization is mutual recognition, an approach that has been implemented to some degree in the airline domain. Under this solution, vehicles that meet EU regulations would be recognized for sale in the US, and vehicles that meet US regulations would be recognized for sale in the EU. To justify mutual recognition, it would be helpful (or possibly even necessary) to demonstrate that safety in EU- and US-regulated vehicles is essentially equivalent. The TTIP trade negotiations prompted the current research project to analyse crash data to compare the crash injury risk of US and EU vehicles. In Phase 1 of the project, a methodology was proposed to investigate the hypothesis that vehicles meeting EU safety standards would perform equivalently to US-regulated vehicles in the US driving environment, and that vehicles meeting US safety standards would perform equivalently to EU-regulated vehicles in the EU driving environment. In Phase 2, the analysis was carried out. This document contains a description of the Phase 1 methodology as implemented (this was done with only minor changes) and presentation of the results of the Phase 2 analysis. A key challenge in evaluating safety performance for EU- and US-regulated passenger vehicles is that the two types of vehicles are driven in different driving environments, and crash datasets contain events involving only one group of vehicles. Thus, crash datasets represent the combination of risk and exposure for a given environment and vehicle population. Risk is the probability of injury or crash involvement given a particular set of circumstances, while exposure is the particular collection of those circumstances. If a vehicle is moved to a different driving environment, its risk characteristics are carried with it, but the exposure to different crash characteristics changes with the change in environment. To answer the question posed, we must separate risk from exposure. Because EU vehicles and US vehicles are separated geographically, their risk is represented with a statistical model, which is then applied to the other region’s exposure population. The risk model based on EU vehicle performance can be applied to the US crash environment and compared to the performance of US vehicles in the US crash environment, and vice versa. As the risk models generated from each region are applied to both regions’ environments, the question is then asked: What is the evidence that vehicle safety performance is (or is not) essentially equivalent? In this project, analysis of crashworthiness and crash avoidance are performed separately, as the relevant datasets and outcomes are different. In-depth crash databases with harmonized injury outcomes are needed to assess crashworthiness, defined as a risk of injury given that a crash occurred. Databases of police-reported crashes and exposure data are needed for crash avoidance, defined as the risk of a crash occurring. The methods section of this report contains details on datasets, treatment of the data (inclusion criteria and variable definitions), and analytical methods. Some statistical details are included in appendices. The approach for analysing crashworthiness uses three methods to better understand the comparison between the two vehicle groups. The first method tests the basic hypothesis that the two best-fit risk models (one for EU-regulated vehicles and one for US-regulated vehicles) are the same. The second method applies the two Method 1 risk models side-by-side to the EU crash data, which represent the EU driving environment, and again to the US crash data, which represent the US driving environment. This creates two separate direct comparisons of risk, which allows for a more detailed look at the groups of crashes (such as frontal or side impacts or rollovers) for which predicted injury risk is similar or different within each environment. Finally, the third method compares the overall weight of evidence for models that predict some risk difference vs. models that predict no risk difference. This approach uses Bayes Factors to compare evidence for two hypotheses and does not depend on the single best-fit model. As with the second method, the comparisons are done separately for the EU crash population and the US crash population. The methods also include description of how crash avoidance was considered. Data in the relevant EU and US datasets were only sufficient to address two crash avoidance countermeasures: headlamps (in relation with pedestrian crashes at nighttime versus daytime) and mirrors (where the analysis is based on the proportion of lane-change/merge crashes to the driver’s side versus the passenger’s side). (Author/publisher)
Samenvatting