The Fatality Analysis Reporting System (FARS) is the National Highway Traffic Safety Administration's census of motor vehicle crash fatalities in the United States and its territories. This paper seeks to explain the implementation of the re-coding process, in which previously entered cases were sampled and re-coded by select, experienced FARS analysts to check the accuracy and performance of the original analysts. The goal of the re-coding process is to enhance the quality control process to maintain accurate, complete, and timely FARS data. To test the feasibility of the re-coding process, 80 cases were selected by a random case generator tool for a pilot study. The pilot study results indicated that when all the re-coders coded the same value, there was an 85.4-percent agreement with the original coded value. The pilot study proved projected timelines, sample sizes, and the re-coding processes to be practical and the first system-wide re-coding process was conducted. Overall, the first system-wide re-code was successful and results indicated that there was 83-percent agreement between the re-coder’s values and the original coded values for the Accident, Driver, and Person level variables. At the vehicle level, the re-coder’s matched the original coding 98 percent of the time. Both the pilot study and the first system-wide sample had over 80 percent of cross-coder agreement, an acceptable level of uniformity. The re-coding process also identified complex data elements that require analytical skill to discern and code correctly. (Author/publisher)
Abstract