82 Transportation Research Record: Journal of the Transportation Research Board, No. 2432, Transportation Research Board of the National Academies, Washington, D.C., 2014, pp. 82–90. DOI: 10.3141/2432-10 Freeway incidents are major sources of nonrecurrent congestion, and resultant secondary crashes can prolong traffic impact and increase costs. Research on secondary crashes to support statewide transportation system management has been limited. In this study, a two-phase automated proce- dure was developed to identify secondary crashes on large-scale regional transportation systems. In the first phase, a crash-pairing algorithm was developed to extract spatially and temporally nearby crash pairs. The accuracy and efficiency of the algorithm were validated by comparing it to an ArcGIS-based program. In the second phase, two filters were proposed to reduce the crash pairs for secondary crash identification: the first filter selected crash pairs whose earlier crashes were on mainline highways; the second filter selected crash pairs whose later crashes happened within the dynamic impact areas (i.e., backup queues) of the earlier crashes. Shock- wave theory was used to model the dynamic impact of a primary incident. The two-phase procedure used a linear referencing system for crash local- ization and can be applied to any regional transportation system with a similar data structure. A case study using 2010 data was conducted on nearly 1,500 mi of freeways in Wisconsin. Among the crash pairs produced by the two-phase procedure, 73 secondary crashes were confirmed with police reports. Preliminary analyses showed that (a) secondary crashes occurring in the same traffic direction as the primary incidents were about three times as frequent as secondary crashes in the opposing direction, and (b) two-vehicle rear-end collisions, multiple-vehicle rear-end collisions, and sideswipes were three major types of secondary crashes (about 84%). The annual cost of congestion in the United States reportedly exceeds $120 billion (1). Freeway incidents are major sources of nonrecurrent congestion, and resulting secondary crashes can prolong traffic impact and increase costs. A secondary crash is an undesirable consequence of a primary incident. More formally, according to the FHWA, “sec- ondary crashes are those that occur within the time of detection of the primary incident where a collision occurs either (a) within the incident scene or (b) within the queue, including the opposite direction, result- ing from the original incident” (2). Existing studies have shown the extended traffic impact and the economic costs of secondary crashes (3–5). Reducing the chances of secondary crashes becomes a major consideration in the dispatch plans of traffic incident management agencies (6, 7 ). Despite various findings on secondary crashes, most existing studies were limited by scope. Many were conducted on only one or two sample freeways or a short segment of highway; other studies extended the scope to freeways but considered a small regional scale. Only two studies were performed on a large scale that involved state- wide highway systems. One major reason for such scope constraints was the challenge of identifying secondary crashes. To identify secondary crashes accurately, most existing studies considered the dynamic features of the traffic impact caused by the primary incidents. Thus, the study scopes were limited to highway facilities for which high-resolution traffic data were available for dynamic analyses. In addition, modeling the dynamic impact of primary incidents required considerable computational efforts, which for a statewide transporta- tion system could be intolerable or even infeasible. Previous studies considering statewide highway systems did not consider the dynamic impact of primary incidents. In summary, none of the previous studies investigated secondary crashes on a statewide transportation system while considering the dynamic impact of primary incidents. To fill the research gap identified above, the current study develops a two-phase automatic procedure. In the first phase, spatially and tempo- rally nearby crash pairs (up to custom static thresholds) are extracted from a large network on the basis of a crash-pairing algorithm. The accuracy and the efficiency of this algorithm were validated. In the sec- ond phase, two filters are used to select crash pairs that are more likely to be primary–secondary crash pairs. One of the filters uses shockwave theory to evaluate the dynamic traffic impact of the primary incidents. At the end of the two-phase procedure, manual review of identified police reports is needed to confirm actual secondary crashes. However, the number of crash reports to review is considerably less. LITERATURE REVIEW Secondary crashes have been observed to be one of the notable conse- quences of freeway incidents. Early in the 1970s, Owens conducted an on-the-spot study of traffic incidents on a 21-km (13-mi) stretch of motorway in England during peak hours and found that 32.5% of the observed crashes were related to primary incidents (8). In recent decades, the development of intelligent transportation systems has made a variety of transportation data easier to access, which in turn has encouraged researchers to revisit secondary crashes. In earlier studies (3–5, 9–17 ), an incident was identified as a secondary crash as long as it occurred within a rectangular time–space window that originated from another incident. For example, Raub classified an incident as a secondary crash if it happened within 1,600 m upstream of another incident and no later than 15 min after that incident was cleared (9, 10). This type of method was called the “static threshold” in the sense that it considered the spatial impact range of a primary incident to be consistent throughout a certain period. However, the impact of a traffic incident is typically dynamic with respect to time. Identification of Secondary Crashes on a Large-Scale Highway System Dongxi Zheng, Madhav V. Chitturi, Andrea R. Bill, and David A. Noyce D. Zheng, Room 1249A; M. V. Chitturi and A. R. Bill, Room B243; and D. A. Noyce, Room 1204, Traffic Operations and Safety Laboratory, Department of Civil and Environment Engineering, University of Wisconsin–Madison, Engineer- ing Hall, 1415 Engineering Drive, Madison, WI 53706. Corresponding author: D. Zheng, dzheng3@wisc.edu. 2014 PATRICIA F. WALLER AWARD: Outstanding Paper on Safety and System Users