CHAPTER 6

Lessons Learned in HSM2 Part C Development

The HSM1 and HSM2 have each been developed through a series of independently managed research projects. Each project has generally had a separate research contractor and a separate oversight panel. Once these independent research projects were complete or nearly complete, both the HSM1 and HSM2 have had a production contractor assemble the various independent research project results into a nearly final HSM version that was ready for review and balloting by AASHTO. This same approach appears likely to continue for future HSM editions. The challenges in finalizing the HSM2 Part C in NCHRP Project 17-71A, and to a lesser extent the challenges encountered in NCHRP Project 17-36 in finalizing the HSM1, indicate a need for closer coordination between the independent research projects conducted for future HSM editions.

Several issues in research conducted for the HSM2 Part C were noted in Chapter 5. Some of these issues were resolved in NCHRP Project 17-71A, while others could not be resolved and resulted in research conducted for the HSM2 not being used in the HSM2. Lessons learned in the process of assembling HSM2 Part C are described here:

• Research to develop material for inclusion in HSM Part C chapters should not focus solely on developing SPFs for specific facility types without also considering what AFs can be used with those SPFs. If a new SPF is intended to replace an existing SPF in a Part C procedure, the research project that develops the SPF should assess whether the AFs used with the existing SPF can also be used with the new SPF or whether changes to the AFs are needed. If changes to the AFs are needed, those changes should be made as part of the same research project. If a new SPF is developed for a facility type not currently included in the HSM, the research should also assess which AFs should be used with the new SPF and, where appropriate, develop the appropriate AFs. A key element of compatibility between specific SPFs and AFs is sharing common base conditions. To summarize, it should be recognized that SPFs without AFs do not constitute a complete crash prediction method; therefore, research for HSM Part C chapters should provide SPFs and AFs that can be used together.
• Research to develop CMFs intended for use as AFs in HSM Part C procedures should also assess which SPFs the new AFs may appropriately be used with. A new CMF cannot appropriately be incorporated into HSM2 Part C as an AF unless the research verifies that the AF is suitable for use with a specific SPF (or several SPFs). A key element of compatibility between specific SPFs and AFs is sharing common base conditions. CMFs whose compatibility with specific SPFs has not been established are more appropriate for incorporation in the FHWA CMF Clearinghouse than in HSM2 Part C.
• Each research team developing a new crash prediction method or revising an existing crash prediction method for HSM2 Part C should be responsible for ensuring that their method is compatible with and can provide results that can be used with and compared to predicted values from existing HSM methods for facility types that constitute likely design alternatives for the facility type addressed by the new or revised method. If the results of the new or revised

• method are not capable of being compared with results for other facility types, use of the new or revised method should not be recommended.
• Each new or revised crash prediction method should be based on the same or compatible definitions of crash severity levels, crash and collision types, and crash assignments to roadway segments versus intersections and freeway mainline and speed-change-lane segments versus ramps as the existing HSM methods for facility types that constitute likely design alternatives for the facility type addressed by the new or revised method. If this is not done, crash predictions obtained with the new SPFs or AFs may not be suitable for comparison with crash predictions made with other existing HSM2 Part C procedures. For example, in one recent research project, the research team made a decision to exclude animal-related crashes from the data they used to develop SPFs. As a result, the SPFs they developed were not compatible with other existing SPFs in the same chapter and with the existing AFs intended for use with the new SPFs, all of which were developed using data that included animal-related crashes. Adjustments to compensate for this incompatibility were made in NCHRP Project 17-71A. The same recent project also developed prediction models for specific crash types using unique collision type definitions that differed from the crash-type definitions based on MMUCC (NHTSA, 2017). It was not possible to resolve this discrepancy in NCHRP Project 17-71A, so the new collision type prediction models were omitted from HSM2 Part C. In summary, research projects intended to develop crash prediction methods for HSM2 Part C should use definitions of crash severity levels and collision types that are compatible with existing HSM2 Part C methods. If a logical reason to change a definition is identified, then the research should provide replacements for all potentially affected prediction models and tabulated values.
• Each research team that develops revised SPFs intended to replace existing SPFs for specific facility types in HSM2 Part C should compare the existing and revised SPFs and demonstrate that the revised SPFs provide realistic and reasonable results and are an improvement on the existing SPFs. There is no reason to replace an existing SPF unless the revised SPF is a demonstrable improvement over the existing SPF. The final report of the research project that develops the revised SPFs should document this comparison with graphs showing the existing and revised SPFs and should make a formal recommendation concerning use of the revised SPFs based on this comparison. There is no single method that is always suitable for making such comparisons. The appropriate approach for making such comparisons should be selected based on results from related research, including comparisons between the facility type of interest and other facility types, and professional judgment.
• Each research team that develops a predictive method for a facility type not currently addressed in HSM2 Part C should be responsible for comparing the results of the new predictive method to predicted values from existing HSM methods for facility types that constitute likely design alternatives for the facility type addressed by the new method and demonstrating that the new method provides reasonable results in comparison to those other facility types. The reasonableness of the comparison could be based on existing CMFs for conversion between facility types or on existing cross-sectional comparisons in published literature or previous transportation agency studies. If the results from the new method are not reasonable in comparison to the results for other facility types, the research team should either find ways to modify the new method until it does provide reasonable results in comparison to other facility types or recommend that the existing method be retained. The final report of the research project that develops the new method should present graphs documenting the comparison of its results to predictions for other facility types and should make a formal recommendation concerning use of the revised SPFs based on this comparison. If it is determined that the original models are flawed or outdated, the researchers should say that and base their recommendation on that.
• Each research team should conduct sensitivity analysis of the model output using ranges of the variable inputs and validate results with crash records for various and realistic vehicle and

• vulnerable user volumes and design element alternatives. Sample problems should be developed with reasonable ranges of input variable values for use in the HSM or other AASHTO or FHWA documents and in training development.

The issues that have resulted in the lessons described here have arisen because, in the past, each research contractor and oversight panel has operated independently without much central coordination of key decisions. There may be valid reasons for research teams developing HSM materials to deviate from some of the suggestions presented, but this should not be done without consideration of the effect of the decision on the usefulness of the HSM as a whole, and there should be some mechanism to verify that such deviations are appropriate. It would be useful to designate a specific group that either a research team or a project panel could ask to review issues that arise in HSM development and provide central coordination before decisions to deviate from HSM precedents are made. Possibly the AASHTO HSM Steering Committee could serve this function. The primary need is to provide central coordination at the time decisions are made during the research rather than after the research is completed.