[Previous: Data Input Requirements] [Table Of Contents] [Next: CPM Output]

3.  Crash Prediction Procedural Elements

This section describes the main procedural elements that the CPM uses to estimate the crash frequency for a given project. These elements, which comprise the crash prediction algorithm, are as follows: segmentation procedure, base models, AMFs, calibration factors, and the Empirical Bayes Procedure for crash prediction when site-specific crash history data are available.

3.1  Segmentation Procedure

The crash prediction algorithm first divides the project within the specified evaluation bounds into homogeneous analysis sections (homogeneous highway segments and intersections). Estimates of crash frequency are made on the individual homogeneous analysis sections that make up the project. The total prediction for the project is the sum of the individual predictions for each of the homogeneous analysis sections. The segmentation procedure is described in detail in Steps 2-4 of the Crash Prediction Algorithm.

3.2  Base Models (Segments and Intersections)

The base model for highway segments is the best available regression model for predicting the total crash frequency of a homogeneous highway segment on a rural two-lane highway. The base model predicts the total expected crash frequency on the highway segment during a specified time period as a function of the highway segment's traffic volume, geometry, and traffic control. The specific regression model used by the CPM is presented in Step 7 of the Crash Prediction Algorithm.

Separate base models have been developed for three-leg intersections with minor-road STOP control, four-leg intersections with minor-road STOP control, and four-leg signalized intersections. These models are available in the Report No. FHWA-RD-99-207. The specific regression models used by the CPM for intersections are presented in Step 7 of the Crash Prediction Algorithm.

3.3  AMFs (Segments and Intersections)

AMFs are multiplicative factors used to adjust the expected base crash frequency for the effect of individual geometric design and traffic control features. Each AMF is formulated so that the nominal or base condition is represented by an AMF of 1.00. Conditions associated with higher crash experience than the nominal or base condition will have AMFs greater than 1.00 and conditions associated with lower crash experience than the nominal or base condition will have AMFs less than 1.00. The AMFs are applied in Step 9 of the Crash Prediction Algorithm. For details, see AMFs for Highway Segments, and AMFs for Intersections.

3.4  Calibration Factors

A key element of the crash prediction algorithm is a set of calibration factors that allow individual highway agencies to tailor the safety prediction to their own local conditions. Geometric design factors and ADTs are accounted for in the crash prediction algorithm and do not require calibration. However, there are factors that lead to differences in reported crash frequencies between highway agencies in different geographical areas that are not accounted for by the crash prediction algorithm. These include:

The calibration procedure is intended to account for these differences and provide crash predictions that are comparable to the estimates that a highway agency would obtain from its own crash records system.

The values of these calibration factors can be viewed and edited in the Administration Tool (see System Administration). The calibration factors include:

The calibration factors Cr, Ci1, Ci2, and Ci3 are initially set to the default value of 1.00.

Appendix C of Report No. FHWA-RD-99-207, Prediction of the Expected Safety Performance of Rural Two-Lane Highways,presents a process that can be used by highway agencies to develop calibration factors appropriate for their own local conditions. There are spreadsheets available through the Administration Tool to perform the calibration process externally. The results can then be implemented in the CPM.

3.5  Crash Prediction When Site-Specific Crash History Data are Available

When starting an Evaluation, users have the option of specifying whether or not to use crash history data in the crash prediction algorithm. When users select this option, the algorithm incorporates an Empirical Bayes (EB) procedure for combining expected crash frequencies (estimated using the base models and AMFs) with site-specific crash history data.

This section briefly explains the advantages of using the EB procedure and factors that should be considered in deciding whether or not the EB procedure should be applied in a particular evaluation. Report No. FHWA-RD-99-207, Prediction of the Expected Safety Performance of Rural Two-Lane Highways, provides additional details on the EB procedure.

3.5.1  Caution on Use of CPM Results When Site-Specific Crash History Data Are Available

The estimated crash frequencies derived from the base models and AMFs have the advantage of being based upon data from many locations and representing the long-term average crash frequency that would be expected among sites with similar geometry and traffic characteristics. These estimates do not, however, account for all of the factors that cause expected frequencies at individual locations to vary about the average of all similar locations. The advantage of using the EB procedure is to improve the accuracy of estimates for an individual location by factoring in the actual crash history of the location being evaluated.

One might ask, Why not rely solely upon site-specific crash history data? Although crash history data are an important indicator of the safety performance of a highway, they suffer from the weakness of being highly variable. Given this high variability, it is difficult to estimate the long-term expected crash frequency using the relatively few years of crash data generally available. This variability is an issue for two-lane rural highways where crashes are rare and many locations experience no crashes over a period of several years. If a location has experienced no crashes during the past several years, it is certainly not correct to think that it will never experience a crash (i.e., that its long-term average crash frequency will be zero, which is what relying solely upon a few years of crash history data would suggest).

In theory, when enough years of crash history data are available, it would be more accurate to appropriately combine estimates from the base models and AMFs with site-specific crash history data than to rely on either the model estimates or the site-specific data alone. The EB procedure determines the statistically appropriate weighting of estimates from the base models and AMFs and site-specific crash history data.

3.5.2  Situations in Which the EB Procedure Should and Should Not Be Applied

If the project being evaluated involves an existing highway for which sufficient crash history data are available, then the EB procedure is an option that the user should consider. For projects involving highways on new locations, there is no relevant crash history and, therefore, use of the EB procedure is not an option.

For projects involving existing highways, two primary factors should be considered in determining whether or not to select the evaluation option to use crash history data: (1) the availability of a sufficiently large sample of crash history data, and (2) the relevance of the crash history on the existing highway to the project alternatives being evaluated.

The availability of a sufficiently large sample of data needs to be considered because of the high variability in crash data on two-lane rural highways. IHSDM specifies at least 2 years of crash history data to use the EB procedure.

The relevance of the crash history on the existing highway to the project alternatives being evaluated must also be considered. The EB procedure considers both the existing highway's and the proposed alternatives' geometric design and traffic control. When projects involve several alternatives, comparisons are generally made among each alternatives' estimated crash frequency. To ensure comparability among the estimates for individual alternatives, the EB procedure should be applied either in all alternatives or in none of the alternatives. Therefore, if the crash history on the existing highway is not relevant to one or more of the project alternatives, then it should not be applied in any of the alternatives being evaluated.

Judgment is required to assess the relevance of the crash history on the existing highway. The fundamental issue is whether or not the crash history on the existing highway would be indicative of the crash experience that should be expected in the future after a design alternative is implemented. The following guidance is provided on project types in which it would be appropriate and, then, in which it would not be appropriate to apply the EB procedure.

The existing highway's crash history is generally relevant and, therefore, use of the EB procedure is considered appropriate in the following situations:

The existing highway's crash history may not be relevant and, therefore, use of the EB procedure may not be appropriate involving significant changes to alignment geometry and/or intersection configuration. Examples include:

3.5.3  Example to Illustrate When to Use/Not Use Historical Crash Data

Conditions for appropriate use of historical crash data to predict future crashes are somewhat subjective, and require user judgment. The following example is provided to help determine when use of historical crash data is appropriate:

The safety effect of spot improvements (e.g., flattening a single horizontal curve) may be examined in the context of an extended length of highway. For a given spot improvement, the new alignment might differ substantially from the old alignment (e.g., if a curve is flattened or replaced by a tangent, the horizontal alignment centerline for the limits of the improvement will be shifted). However, if when an extended length of highway is considered, the percentage of alignment changed is small, the use of historical crash data might be appropriate.



[Previous: Data Input Requirements] [Top] [Next: CPM Output]