|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-05-051
Date: October 2005
Crash Cost Estimates by Maximum Police-Reported Injury Severity Within Selected Crash Geometries
PDF Version (424 KB)
PDF files can be viewed with the Acrobat® Reader®
The results of the analyses are included in Microsoft® Excel tables found in the appendix. They are organized into the following six categories or levels:
At each level, in addition to estimates for individual KABCO levels and combinations, crash cost estimates are also included for two additional categories-"Injured, severity unknown," which means there was at least one injury in the crash, but the severity was not recorded in the police files, and "Unknown severity," which means no injury severities were provided on the police report. These cost categories are not expected to be used very often, but they are included for completeness.
The output is presented in tabular form in appendix B. The title of each tables provides the number of the Level (e.g., "Level 1...") and a designation of whether the estimates are categorized by speed limit (e.g., "Level 1 w SL") or not (e.g., "Level 1 no SL"). An example of the top portion of the "Level 2 w SL" output is shown in table 1 below. The first column, which is a crash geometry number, and the columns labeled "Maximum Injsev Code" are included to assist the user in later sorts of the data. The remaining columns headings are self-explanatory.
In the more detailed levels such as Levels 1 and 2, one finding that appears somewhat counter-intuitive is that the cost estimates for the same crash injury level within the same crash type are sometimes greater for the lower speed limits. For example, in the table below, the human capital and comprehensive cost estimates for a K+A injury crash at the lower speed limit (row three) is greater than for the comparable crash at the higher speed limit in row eight (i.e., $576,985 versus $425,414 for mean comprehensive costs). This resulted from the fact that the cost for the A-injury crash at the lower speed limit was greater than the cost for the A-injury crash at the higher speed limit. Examination of the base data indicated that this may be a function of the fact that lower speed limits are generally in urban areas, where there may be more occupants (and younger occupants) in the involved vehicles (or more or younger pedestrians in the same crash). It is noted, however, that this pattern does not hold for all crash types even at the lower levels. This means that there are other unknown factors at work in the database used in the cost development. The user will note that this counter-intuitive finding can be overcome by using costs with combined speed limits, or using higher-level cost (e.g., Level 3 estimates include fewer of these counter-intuitive findings than Level 2 estimates, which have less than in Level 1).
Small Samples and Outliers
Note that some of the rows in the table are color-coded S, I, and N. All three codes are included as "flags" to the user that these estimates are felt to be less accurate than estimates in other rows. The S-coded rows indicate estimates that were derived from small sample sizes. For example, the second S-coded row in the table (i.e., the sixth row of data in the table) indicates that there were only five observations (i.e., crashes in the CDS files used) where a no-injury pedestrian crash occurred at an intersection with a speed limit of 80 km/h (50 mi/h) or greater. A decision was made to flag fatal crash cost estimates where less than five observations were present, and to flag estimates in all other injury levels where less than 10 observations were present. In these rows, only the mean crash cost estimates are included. The standard deviations and confidence intervals are omitted from the output since these are felt to be virtually meaningless given the small sample sizes. Suggestions to the user for dealing with these questionable estimates are included in the next section.
The I-coded rows indicate what are felt to be "illogical values" or "outliers" in the data-cells with ample sample sizes, but where the cost for a given injury level is an outlier when compared with either other costs within the same crash type (e.g., a B+C cost that is greater than an A-injury cost for a given crash type), or when compared to costs of different crash types at the same injury level (e.g., a no-injury cost that is much greater than all other no-injury costs). These illogical estimates were identified by looking at patterns of costs in similar severity levels or crash types. For example, the first I-coded row in table 1 (i.e., the tenth row of data in the table) indicates a very high cost per crash when compared to other no-injury level pedestrian crashes and other no-injury level crashes in general. Additional examination of the cost-development data indicated that some of these outliers might be due to erroneous coding by the police officer (e.g., in one case, a "no-injury" pedestrian crash was found to have two rather severe injuries.) Since it was not possible to examine each illogical finding in detail, they were flagged for the user's benefit. Again, suggestions for dealing with these are found in the next section.
In addition, there are crash types in the NASS data used to develop these estimates where no fatal crashes were present. For example, if the comparable "Level 1 w SL" table had been presented here, the user would note the absence of a crash cost estimate for fatal crashes within the Type 3, single-vehicle animal crashes. No such fatal crash existed in the NASS data used to develop these estimates. As a result, the estimate for "K+A" crashes in the final row of table 1 below does not have a fatal crash cost component, and is less accurate than similar combined costs where both K and A crashes existed in the NASS data. All combined estimates (e.g., K+A, K+A+B, K+A+B+C) with no fatal component are coded N in the tables.
Table 1. Level 2 crash cost estimates categorized by speed limit
45 mi/h = 72 km/h 50 mi/h=80 km/h
While these flagged estimates do exist, in general, most estimates are felt to be stable and usable in analysis. Many of the small sample estimates are for "unknown severity" conditions, where the officer either failed to code the injury level or simple coded it as "injured" without a specific level provided. As noted earlier, these categories are not likely to be used very often in subsequent analyses.
Suggestions for Handling Flagged Estimates
There are at least four alternative "corrections" a user could consider when a pertinent cost estimate is flagged for sample size or as an outlier or questionable combined-severity estimate.
Finally, it might appear that a fifth option would be for researchers to develop a customized cost specific analysis using a weighted combination of estimates provided. This should not be done. To combine different estimates (e.g., combine a K estimate and an A estimate into a K+A estimate), it is necessary to weight the individual estimates by the national estimates of the number of applicable crashes in each cell. The sample sizes provided in the output under "Observ" represent the number of raw cases in the NASS files used to develop the estimate provided. (See appendix B.) They do not represent the extrapolation of this raw frequency into a national estimate. (Pacific Institute for Research and Evaluation (PIRE) used the extrapolated national estimates in developing the combined estimates in the appendix tables.)
Topics: research, safety, stop red light running program
Keywords: research, safety, crash geometries, red light running
TRT Terms: traffic accidents, accident data, cost estimating