Part C of the HSM provides a predictive method to estimate average crash frequencies for both existing and proposed future conditions of a roadway network, facility, or individual site. This predictive method not only estimates total crashes over a given time but can also be used to estimate average crash frequencies by crash severity and collision type. It is recommended that these estimates be determined for existing conditions, alternatives to the existing conditions, or for proposed new roadways. A roadway network can be divided into individual sites of either homogenous roadway segments or intersections to run the predictive method.

The estimated average crash frequency of an individual facility type is based on the traffic volumes, geometric design features, and traffic control type for that site. Predictive models were developed to estimate average crash frequencies for a variety of facility types using regression models. These regression models were developed from data of similar sites across the country and are known as safety performance functions (SPFs) in the HSM. SPFs have been developed for specific facility types under what are referred to as “base conditions,” the most common conditions for similar sites across the country. For

example

,

Table 5-1

shows a sample of the base conditions for rural multilane highways.

The HSM has developed the SPFs for the following facility types:

Table 5-2

and

Table 5-3

demonstrate the various facility types that SPFs were developed for as part of the HSM.

A simplified

example

of how SPFs were developed for the HSM is illustrated in

Figure 5-3

. This figure illustrates how crash data from across the country was used in the development of the SPFs located in the HSM.

Adjustments to the predicted crashes, determined by using an SPF, may account for the differences between the base conditions used in developing the SPF and the site-specific conditions.

Part C CMFs are used to account for these differences when site-specific conditions vary from the base conditions. These CMFs were developed specifically for individual SPFs and are intended to be applied only to the crash prediction results of that particular SPF. The SPF-related CMFs are different than the CMFs on the CMF Clearinghouse website. The Highway Safety Manual 2nd Edition (HSM2) is expected to rename SPF-related CMFs to AFs to reduce the confusion between Part C CMFs developed for a specific SPF and CMFs located on the Clearinghouse website.

A calibration factor (C) may also be used to make agency-specific adjustments to the predicted crash totals. This accounts for the differences between the agency or agencies for which the models were developed and the agency/agencies for which the predictive model is being applied. Calibration factors to the HSM predictive method calculations should be applied whenever possible. Each agency is responsible for developing their own calibration factors for specific facility types.

The HSM Part C predictive method is summarized in the following general equation:

Where:

N

Predicted

= predicted average crash frequency for a specific site type

N

SPFX

= predicted average crash frequency for the base conditions of the SPF for a specific site type

CMF

1x

= CMF specific to the SPF for a specific site type

C

x

= calibration factor to adjust the specific SPF to the local conditions for a specific site type

x

= specific site type

Because crash totals fluctuate over time, it is difficult to know whether changes in the observed crash totals are due to changes in site conditions or are due to natural fluctuations. A period with a high number of observed crashes is statistically likely to be followed by a period with a low number of observed crashes. The opposite of this tendency also applies; it is probable that a period with low crash totals will be followed by a period with high crash totals. This tendency to regress to the mean or average is known as regression-to-the-mean (RTM). Failure to account for the effects of RTM introduces the potential for RTM bias, also known as selection bias. RTM bias results in overestimating or underestimating the effectiveness of a treatment. The effects of RTM bias are illustrated in

Figure 5-4

. For existing sites, facilities, or roadways, the Empirical Bayes (EB) method can be applied within the predictive method to account for both the predicted average crash frequency and the observed crash frequency. This method accounts for the reliability of a particular SPF and RTM bias.

The application of the EB method within the predictive method results in an expected average crash frequency.

Figure 5-5

illustrates the relationship between observed, predicted, and expected crashes when performing the predictive method for a particular SPF. See

Appendix F, Section 2 – NCHRP 17-50 Project Example of HSM in Design

for more details on how expected crashes can be estimated.

Table 5-4

identifies the scenarios when the EB method is applicable and not applicable, as identified in the HSM.

The HSM predictive method is summarized as an 18-step process used to estimate average crash frequencies for a roadway network, facility, or site and is illustrated in

Figure 5-6

. An example of applying the HSM Part C predictive method is provided in

Appendix F, Section 2 – NCHRP 17-50 Project Example of HSM in Design

.