Case-control analysis in highway safety: Accounting for sites with multiple crashes.

There is an increased interest in the use of epidemiological methods in highway safety analysis. The case-control and cohort methods are commonly used in the epidemiological field to identify risk factors and quantify the risk or odds of disease given certain characteristics and factors related to an individual. This same concept can be applied to highway safety where the entity of interest is a roadway segment or intersection (rather than a person) and the risk factors of interest are the operational and geometric characteristics of a given roadway. One criticism of the use of these methods in highway safety is that they have not accounted for the difference between sites with single and multiple crashes. In the medical field, a disease either occurs or it does not; multiple occurrences are generally not an issue. In the highway safety field, it is necessary to evaluate the safety of a given site while accounting for multiple crashes. Otherwise, the analysis may underestimate the safety effects of a given factor. This paper explores the use of the case-control method in highway safety and two variations to account for sites with multiple crashes. Specifically, the paper presents two alternative methods for defining cases in a case-control study and compares the results in a case study. The first alternative defines a separate case for each crash in a given study period, thereby increasing the weight of the associated roadway characteristics in the analysis. The second alternative defines entire crash categories as cases (sites with one crash, sites with two crashes, etc.) and analyzes each group separately in comparison to sites with no crashes. The results are also compared to a "typical" case-control application, where the cases are simply defined as any entity that experiences at least one crash and controls are those entities without a crash in a given period. In a "typical" case-control design, the attributes associated with single-crash segments are weighted the same as the attributes of segments with multiple crashes. The results support the hypothesis that the "typical" case-control design may underestimate the safety effects of a given factor compared to methods that account for sites with multiple crashes. Compared to the first alternative case definition (where multiple crash segments represent multiple cases) the results from the "typical" case-control design are less pronounced (i.e., closer to unity). The second alternative (where case definitions are constructed for various crash categories and analyzed separately) provides further evidence that sites with single and multiple crashes should not be grouped together in a case-control analysis. This paper indicates a clear need to differentiate sites with single and multiple crashes in a case-control analysis. While the results suggest that sites with multiple crashes can be accounted for using a case-control design, further research is needed to determine the optimal method for addressing this issue. This paper provides a starting point for that research.