The vast majority of the safety studies concerning added freeway lanes are based on pre-project versus post-project comparisons of accident rates for the project area only. There are two potential problems with such an approach: First, accurate traffic volume data are required for both the pre-project and post-project periods in order to provide exposure measures in the calculation of accident rates. Experience has shown that the accuracy of traffic volume data is often suspect, particularly in the period prior to reconstruction of a roadway section; repairs to induction-loop counting devices are often delayed to coincide with the construction involving the lane addition. Consequently, in many cases traffic counts in the pre-project period, and sometimes counts in both the pre- and post-project periods, are based on projections from actual counts taken years before. Because added capacity can induce latent demand for travel on the affected section of freeway, the pre-project and post-project periods can be substantially different than predicted by ordinary projection methods. Inaccuracies in the denominators of aggregate accident rate statistics can lead to false conclusions regarding project safety. Comparisons with "control" sections of roadway can be used in the absence of good exposure data.
The second problem with aggregate comparisons of accident rates in the project section for pre-project versus post-project periods is that the additional capacity might affect accident risks in adjacent sections of roadway. If this occurs, it is potentially misleading to define the roadway only within the project area as the spatial unit of analysis in the accident rate comparisons. This precludes the ability to analyze possible spatial redistributions of accident locations due to the changes in the characteristics of traffic congestion from the pre-project to post-project periods. Moreover, the selection of the specific area of roadway for project location might be related to its accident history. The treated section of roadway might be a “blackspot" for a number of reasons, but the random nature of accident occurrence is a component in the historical process. It has been shown that a form of selectivity bias called "regression to the mean" (Hauer, 1980a, 1980b) can lead to overestimation of the reductions in accident rates resulting from safety-improvement projects. Such bias can also lead to overestimation of accident migration from treated (project) to non-treated (adjacent) roadway sections (Stein, 1984; McGuigan, 1985).
Particularly with regard to added freeway lanes, there might be a migration of accidents from the project area and from upstream of the project area to bottlenecks downstream of the project area, due to the partial relief of congestion in one area and increased traffic flow to another area of congestion. Similarly, there might be a relief of congestion in areas upstream of the project area. Previous studies have recognized the need for "influence" areas (e.g., Urbanik and Bonilla, 1987), but it is difficult to assess the nature of any accident "migration" in terms of aggregate accident rate statistics. The present study attempts to characterize influences through a disaggregate spatial analysis of accident locations.
