Quantitative Safety Analyses for Highway Applications (2025)
Chapter: 7 Future HSM-Related Research Needs
CHAPTER 7
Future HSM-Related Research Needs
This chapter presents future HSM-related research needs identified in the development of the HSM2 in NCHRP Project 17-71A. These research needs may be considered in formulating future HSM-related research programs. It may also be determined that some of these issues or topics are better addressed through training or other types of HSM-related materials, such as a practitioner’s guide. The research needs are organized by HSM2 parts and chapters. Some research needs relevant to several chapters appear multiple times.
The research needs presented here are not necessarily the only research needed to develop the next update of the HSM. Rather, the research needs presented here are those that were identified in the performance of the research and external stakeholder review processes conducted within NCHRP Project 17-71A. A current NCHRP project to create a road map leading to the next HSM edition is expected to expand and prioritize the list of future research needs.
HSM2 Part A—Fundamentals
HSM2 Chapter 2—Road Safety Principles
A key limitation of many HSM2 crash prediction procedures (with the exception of the pedestrian and bicycle crash prediction procedures) is the lack of a quantitative effect of motor-vehicle traffic speed on crash likelihood and severity. Research is needed to develop such a quantitative speed effect and incorporate it into crash prediction methods. Negative binomial regression modeling has failed to provide such an effect because speed is correlated with many other factors in the models. Speed could potentially be incorporated into the procedures as an AF using models developed by Nilsson (2004) or Elvik (2019). Alternative approaches may be identified and considered in the research, such as work by Kloeden et al. (1997, 2001, 2002) and pedestrian-specific work by Rosen et al. (2011). The pedestrian and bicycle models based on the Nilsson models could also be updated.
In HSM2 it was decided to incorporate a brief section on the safe-system approach in Chapter 2. Research is needed to more fully explain how the HSM supports the safe-system approach and integrate safe-system concepts throughout the manual.
HSM2 Chapter 3—Human Factors
There were several topics that emerged from the reviewer comments and through the development of the human-factors chapter that could be addressed in future research. In particular, research is needed to identify how some of the key HSM chapters would benefit from additional examples, case studies, and text discussions describing human-factors research and applications. These topics are really about where and how to improve the implementation of human factors
across the broad range of activities associated with highway safety. Specifically, what resources would help traffic safety professionals better understand human factors and more confidently apply human-factors principles to their day-to-day activities?
HSM2 Chapter 4—Pedestrians and Bicyclists
Research is needed on how to resolve the issue of underreporting of pedestrian and bicycle collisions.
The HSM2 Part C predictive methods for pedestrian collisions are not intended to provide estimates for pedestrian collisions involving (a) police officers standing in the traveled way for enforcement activities, (b) construction or maintenance workers in the traveled way in work zones, and (c) vehicle occupants whose vehicles break down and who leave their vehicle to work on repairs. These three types of pedestrian collisions are not addressed by the predictive method because pedestrian volume counts for such activities are seldom available and because the engineering countermeasures typically programmed using the HSM2 Part C procedures do not address these types of crashes. Research is needed to determine the extent of these collisions and what type of remedies can be used to mitigate these types of collisions.
HSM2 Part B—Roadway Safety Management Process
There were several items noted by HSM2 Part B chapter reviewers that were either items with limited research that could not be added to HSM2 or were items that required additional time and resources to address. One overarching item that should be considered for the third edition of the HSM (HSM3) is consolidating Chapters 7 and 8. This change would require changes to the Introduction to Part B. Another overarching item that should be considered for HSM3 is incorporating safe-system approach concepts throughout the roadway safety management process.
HSM2 Chapter 5—Areawide Approach to Roadway Safety Management
Research is needed on how to apply the crash prediction models to scenarios that are outside the bounds of the independent variables. This is a topic that expands beyond this chapter. Another potential research need related to this chapter is the development of CMFs that can be applied in an areawide analysis.
HSM2 Chapter 6—Network Screening
A topic that can be expanded in HSM3 is providing discussion on how to rank and prioritize projects from different peer groups (e.g., 4-leg signalized intersection and freeway segments) when there is a limited budget. How will an agency prioritize improvements when considering multiple facility types? Another topic that can be discussed in HSM3, perhaps as part of an example, is how to interpret the different rankings when different methods are used for ranking sites. Another topic that could potentially be addressed in Chapter 6 is an approach to network screening based on driver workload [refer to the final report of NCHRP Project 22-45, NCHRP Research Report 1111/BTSCRP ResearchReport 12: Diagnostic Assessment and Countermeasure Selection: A Toolbox for Traffic Safety Practitioners (Campbell et al., 2024), for quantitative approaches to estimating driver workload].
HSM2 Chapter 7—Diagnosis
As discussed in the introduction to this section, consideration should be made for Chapters 7 and 8 to be consolidated in HSM3. At the time of this writing, NCHRP Project 22-45, “Informing the Selection of Countermeasures by Evaluating, Analyzing, and Diagnosing Contributing Factors
That Lead to Crashes,” had been completed, but publication of the research report was pending. HSM3 should evaluate inclusion of material from NCHRP Project 22-45 (now published as Campbell et al., 2024) into Chapter 7 and Chapter 8. A potential research need related to this chapter is how to incorporate non-crash–based data sources into the diagnosis process, such as near-miss reports and conflict studies. Diagnosis of pedestrian and bicycle collisions should also be expanded in HSM3. For example, a discussion on how the built environment leads to certain walking and biking conditions could be included.
HSM2 Chapter 8—Countermeasure Selection
As discussed in the introduction to this chapter, consideration should be made for Chapters 7 and 8 to be consolidated in HSM3. At the time of this writing, NCHRP Project 22-45, “Informing the Selection of Countermeasures by Evaluating, Analyzing, and Diagnosing Contributing Factors That Lead to Crashes,” had been completed, but publication of the research report was pending. HSM3 should evaluate inclusion of material from NCHRP Project 22-45 (Campbell et al., 2024) into Chapter 7 and Chapter 8. Improvements can be made to the presentation of contributing factors. For example, the bulleted lists could be consolidated into tables. The chapter could be expanded to discuss how to select effective treatments that are appropriate for specific contexts. This could be aligned with the new context classifications planned for inclusion in the next edition of the AASHTO Green Book being developed under NCHRP Project 7-29, “Development of the 8th Edition of AASHTO’s A Policy on the Geometric Design of Highways and Streets (Green Book).”
HSM2 Chapter 9—Economic Appraisal
Discussion on methods for crash prediction was expanded in HSM2. However, HSM3 could expand on these methods further by providing more examples. Another topic that could be expanded in HSM3 is how to incorporate nonmonetary considerations into countermeasure selection. This topic could potentially be tied back to Chapter 8.
HSM2 Chapter 10—Project Prioritization
Chapter 10 could be expanded in HSM3 to provide examples of how to perform the optimization methods.
HSM2 Chapter 11—Countermeasure Effectiveness Evaluation
Sample problems in Chapter 11 should be expanded in HSM3 if the sample problems are intended to teach the concepts rather than just demonstrate one particular use case.
HSM2 Chapter 12—Systemic Approach to Roadway Safety Management
Chapter 12 could be expanded in future HSM editions to include discussion on using nontraditional data sources such as social media, video, and crowdsourced data. Chapter 12 could also be expanded by pulling in more material from the literature on both crash-based and non-crash–based systemic analyses.
HSM2 Part C—Predictive Methods
It would be desirable in future research to refit, rather than just calibrate, every SPF in the HSM from crash data for a common state. If an appropriate state with data for every facility type cannot be identified, an alternative approach would be to refit every roadway segment SPF with data for one state and every intersection or ramp terminal SPF with data for another state.
HSM2 Chapter 13—General Concepts for Applying the Part C Predictive Methods
Research is needed to compare the validity of the predictive method when conducting alternative analysis, specifically when changing facility types. Research is also needed on the impact of crash prediction as it relates to underreporting of crashes, and in particular as it relates to pedestrian and bicycle collisions.
HSM2 Chapter 14—Predictive Method for Rural Two-Lane, Two-Way Roads
A key limitation of many HSM2 crash prediction procedures (with the exception of the pedestrian and bicycle crash prediction procedures) is the lack of a quantitative effect of motor-vehicle traffic speed on crash likelihood and severity. Research is needed to develop such a quantitative speed effect and incorporate it into crash prediction methods. Negative binomial regression modeling has failed to provide such an effect because speed is correlated with many other factors in the models. Speed could potentially be incorporated into the procedures as an AF using models developed by Nilsson (2004) or Elvik (2019). Alternative approaches may be identified and considered in the research, such as work by Kloeden et al. (1997, 2001, 2002) and pedestrian-specific work by Rosen et al. (2011). The pedestrian and bicycle models based on the Nilsson models could also be updated.
The effect on pedestrian collisions of right-turn lanes provided for motor vehicles at intersections is considered in Chapter 14 in assessing the number of lanes to be crossed by pedestrians in crossing maneuvers on an intersection leg, but not in other respects. Research is needed to quantify other effects that the presence of a right-turn lane may have on the likelihood and severity of pedestrian collisions. The effects of right-turn lanes on bicycle collisions should also be considered.
The crash prediction models for pedestrian and bicycle collisions in Chapter 14 do not address prediction of pedestrian and bicycle collisions at roundabouts. At present there are no definitive research findings concerning the likelihood and severity of pedestrian and bicycle collisions at roundabouts. Research is needed to address this issue for pedestrian and bicycle collisions at both single-lane and multilane roundabouts. The pedestrian-related research should focus primarily on collisions involving pedestrians crossing the roundabout legs. The bicycle-related research should address all bicycle movements through a roundabout, with emphasis on potential conflicts between motor vehicles and bicyclists at roundabout entrances and exits and on the circulating roadway between entrances and exits. The research on single-lane roundabouts is relevant to HSM2 Chapter 14.
iRAP is updating the pedestrian and bicycle crash prediction procedures that were used in NCHRP Project 17-84 as the basis for the pedestrian and bicycle crash prediction models that appear in HSM2 Chapter 14. This iRAP effort should be monitored to identify future changes to the pedestrian and bicycle crash prediction procedures that should be incorporated into the HSM, including any potential changes in crash likelihood factors, crash severity factors, motor-vehicle speed factors, motor-vehicle traffic volume factors, and pedestrian and bicycle flow factors. Issues of particular interest may include introducing some variability in the crash severity factor and other sensitivities to the presence of right- and left-turn lanes, in addition to increasing the pedestrian crossing distance.
One reviewer of the HSM2 procedures expressed a wish for pedestrian and bicycle crash prediction procedures with fewer factors than those presented in Chapter 14. Nearly all of the factors included in the pedestrian and bicycle crash prediction methods address issues of direct
interest to transportation agencies, so it is not certain that simplified procedures would serve users effectively. However, it may be useful to address possible simplification of the pedestrian and bicycle crash prediction procedures in future research.
One reviewer of the HSM2 procedures suggested that the predictive method for pedestrian collisions be modified to incorporate separate crash severity distributions for pedestrian movements along the road and pedestrian movements crossing the road. Another reviewer suggested that those same crash severity distributions for pedestrian collisions be modified to vary with motor-vehicle speed as well.
One reviewer of the HSM2 procedures suggested that the values of the crash likelihood AFs for the effect of intersection type on pedestrian collisions involving pedestrian movements crossing intersection legs be modified to vary not just by number of intersection legs but also by type of traffic control and presence of left-turn lanes.
HSM2 Chapter 15—Predictive Method for Rural Multilane Highways
A key limitation of many HSM2 crash prediction procedures (with the exception of the pedestrian and bicycle crash prediction procedures) is the lack of a quantitative effect of motor-vehicle traffic speed on crash likelihood and severity. Research is needed to develop such a quantitative speed effect and incorporate it into crash prediction methods. Negative binomial regression modeling has failed to provide such an effect because speed is correlated with many other factors in the models. Speed could potentially be incorporated into the procedures as an AF using models developed by Nilsson (2004) or Elvik (2019). Alternative approaches may be identified and considered in the research, such as work by Kloeden et al. (1997, 2001, 2002) and pedestrian-specific work by Rosen et al. (2011). The pedestrian and bicycle models based on the Nilsson models could also be updated.
The AF for the effect of median width on roadway segment crashes on rural multilane divided highways (nonfreeways) in Chapter 15 does not include values for median widths of less than 10 ft. Research is needed to determine whether the tabulated AF value of 1.04 for 10-ft medians may be applied to roadway segments with median widths of less than 10 ft or whether some alternative AF value applies.
The effect on pedestrian collisions of right-turn lanes provided for motor vehicles at intersections is considered in Chapter 15 in assessing the number of lanes to be crossed by pedestrians in crossing maneuvers on an intersection leg, but not in other respects. Research is needed to quantify other effects that the presence of a right-turn lane may have on the likelihood and severity of pedestrian collisions. The effects of right-turn lanes on bicycle collisions should also be considered.
The crash prediction models for pedestrian and bicycle collisions in Chapter 15 do not address prediction of pedestrian and bicycle collisions at roundabouts. At present there are no definitive research findings concerning the likelihood and severity of pedestrian and bicycle collisions at roundabouts. Research is needed to address this issue for pedestrian and bicycle crashes at both single-lane and multilane roundabouts. The pedestrian-related research should focus primarily on collisions involving pedestrians crossing the roundabout legs. The bicycle-related research should address bicycle movements through a roundabout, with emphasis on potential conflicts between motor vehicles and bicyclists at roundabout entrances and exits and on the circulating roadway between entrances and exits. The research on multilane roundabouts is relevant to HSM2 Chapter 15.
iRAP is updating the pedestrian and bicycle crash prediction procedures that were used in NCHRP Project 17-84 as the basis for the pedestrian and bicycle crash prediction models that appear in HSM2 Chapter 15. This iRAP effort should be monitored to identify future changes to
the pedestrian and bicycle crash prediction procedures that should be incorporated into the HSM, including any potential changes in crash likelihood factors, crash severity factors, motor-vehicle speed factors, motor-vehicle traffic volume factors, and pedestrian and bicycle flow factors. Issues of particular interest may include introducing some variability in the crash severity factor and other sensitivities to the presence of right- and left-turn lanes, in addition to increasing the pedestrian crossing distance.
One reviewer of the HSM2 procedures expressed a wish for pedestrian and bicycle crash prediction procedures with fewer factors than those presented in Chapter 15. Nearly all of the factors included in the pedestrian and bicycle crash prediction methods address issues of direct interest to transportation agencies, so it is not certain that simplified procedures would serve users effectively. However, it may be useful to address possible simplification of the pedestrian and bicycle crash prediction procedures in future research.
Table 15-64 was developed in NCHRP Project 17-84 and presents typical proportions of pedestrian crashes at intersections on rural multilane highways by injury severity level. The values for four-leg intersections presented in the table have an unusual pattern in that the proportion of C-injury crashes is substantially less than the proportion of A- and B-injury crashes, rather than substantially larger as would be expected. These values should be reexamined in future research on pedestrian collisions.
One reviewer of the HSM2 procedures suggested that the values of the crash likelihood AFs for the effect of intersection type on pedestrian collisions involving pedestrian movements crossing intersection legs be modified to vary not just by number of intersection legs but also by type of traffic control and presence of left-turn lanes.
One reviewer suggested that a predictive method be developed for ramp terminals at which the crossroad is a rural multilane highway.
HSM2 Chapter 16—Predictive Method for Urban and Suburban Arterials
A key limitation of many HSM2 crash prediction procedures (with the exception of the pedestrian and bicycle crash prediction procedures) is the lack of a quantitative effect of motor-vehicle traffic speed on crash likelihood and severity. Research is needed to develop such a quantitative speed effect and incorporate it into crash prediction methods. Negative binomial regression modeling has failed to provide such an effect because speed is correlated with many other factors in the models. Speed could potentially be incorporated into the procedures as an AF using models developed by Nilsson (2004) or Elvik (2019). Alternative approaches may be identified and considered in the research, such as work by Kloeden et al. (1997, 2001, 2002) and pedestrian-specific work by Rosen et al. (2011). The pedestrian and bicycle models based on the Nilsson models could also be updated.
The effect on pedestrian collisions of right-turn lanes provided for motor vehicles at intersections is considered in Chapter 16 in assessing the number of lanes to be crossed by pedestrians in crossing maneuvers on an intersection leg, but not in other respects. Research is needed to quantify other effects that the presence of a right-turn lane may have on the likelihood and severity of pedestrian collisions. The effects of right-turn lanes on bicycle collisions should also be considered.
iRAP is updating the pedestrian and bicycle crash prediction procedures that were used in NCHRP Project 17-84 as the basis for the pedestrian and bicycle crash prediction models that appear in HSM2 Chapter 16. This iRAP effort should be monitored to identify future changes to the pedestrian and bicycle crash prediction procedures that should be incorporated into the HSM,
including any potential changes in crash likelihood factors, crash severity factors, motor-vehicle speed factors, motor-vehicle traffic volume factors, and pedestrian and bicycle flow factors. Issues of particular interest may include introducing some variability in the crash severity factor and other sensitivities to the presence of right- and left-turn lanes, in addition to increasing the pedestrian crossing distance.
Further research to confirm or revise the value of the crash likelihood adjustment factor for extra-wide outside lanes (≥14 ft) shown in Table 16-96 may be needed.
The adjustment factor for on-street parking in Section 16.7.1.1 (see Equation 16-60) does not address on-street parking with a bicycle lane outside the parking spaces or inside the parking spaces. Research to develop values for such a factor is needed.
One reviewer of the HSM2 procedures expressed a wish for pedestrian and bicycle crash prediction procedures with fewer factors than those presented in Chapter 16. Nearly all of the factors included in the pedestrian and bicycle crash prediction methods address issues of direct interest to transportation agencies, so it is not certain that simplified procedures would serve users effectively. However, it may be useful to address possible simplification of the pedestrian and bicycle crash prediction procedures in future research.
The crash prediction models for pedestrian and bicycle crashes in Chapter 16 do not address prediction of pedestrian and bicycle collisions at roundabouts. At present there are no definitive research findings concerning the likelihood and severity of pedestrian and bicycle collisions at roundabouts. Research is needed to address this issue for pedestrian and bicycle collisions at both single-lane and multilane roundabouts. The pedestrian-related research should focus primarily on crashes involving pedestrians crossing the roundabout legs. The bicycle-related research should address bicycle movements through a roundabout, with emphasis on potential conflicts between motor vehicles and bicyclists at roundabout entrances and exits and on the circulating roadway between entrances and exits. The research on both single-lane and multilane roundabouts is relevant to HSM2 Chapter 16.
One reviewer of the HSM2 procedures suggested that the values of the crash likelihood AFs for the effect of intersection type on pedestrian collisions involving pedestrian movements crossing intersection legs be modified to vary not just by number of intersection legs but also by type of traffic control and presence of left-turn lanes.
One reviewer identified an issue that the predictive method for pedestrian collisions does not explicitly address the frequency of collisions related to pedestrians playing in the road, lying in the road, or standing in the road. These types of collisions have not been addressed in the current procedures primarily because of the lack of exposure data concerning how many pedestrians take such actions. Potential methods for obtaining such exposure data should be considered.
One reviewer suggested that the HSM2 predictive method for bicycle collisions be expanded to address the configuration with a bicycle lane on the roadway but outside of the parking lane.
As part of NCHRP Project 15-66, a methodology was developed for evaluating the operational performance of arterial weaving sections (Elefteriadou et al. 2024). The goal is to incorporate this new methodology into the next edition of the Highway Capacity Manual. As agencies look to implement this new methodology for evaluating the operational performance of arterial weaving sections, they will also be interested in evaluating the safety performance of arterial weaving sections. However, currently there is no defined approach in the HSM (or other sources) for evaluating the safety performance of arterial weaving sections. Research is needed to develop a methodology that predicts the safety performance of arterial weaving sections and that is suitable for inclusion in the HSM.
HSM2 Chapter 17—Predictive Method for Directional Freeway Segments
The results of NCHRP Project 17-89 concerning the effect of PTSU on freeway segment crashes were not incorporated into the HSM2 because NCHRP Project 17-89 developed this effect as an adjustment to a new model for predicting freeway segment crashes rather than the existing predictive model that is used in the HSM1 and, in a modified form, in the HSM2. NCHRP Project 17-71A concluded that using two different predictive models for freeway segment crashes, one for freeway segments with PTSU and another for freeway segments without PTSU, was likely to lead to incompatible results that would not reliably indicate the effect on crashes of PTSU. Reviewers of the NCHRP Project 17-89 results also identified issues with the range of values of the PTSU adjustment. Thus, further research on the effect of PTSU on crashes is needed to develop this effect in a suitable form for incorporation in the HSM.
The results of NCHRP Project 17-89A concerning the effect of managed lanes, specifically HOV/HOT lanes, on freeway segment crashes were not incorporated into the HSM2. The NCHRP Project 17-89A results were not incorporated into the HSM2 because NCHRP Project 17-89A developed a predictive model for crashes on HOV/HOT lanes independently of the existing HSM1 model for freeway segment crashes, so no reliable method was provided to quantify the effect on crashes of adding HOV/HOT lanes to a freeway. Since the existing freeway segment model and the HOV/HOT lane model were developed independently, there would be no way to separate the effects of adding HOV/HOT lanes from differences between the two models. Further research is needed to develop a predictive method that explicitly quantifies adding HOV/HOT lanes to an existing freeway.
One reviewer has questioned the findings of NCHRP Project 17-45, found in NCHRP Web-Only Document 306: Safety Prediction Methodology and Analysis Tool for Freeways and Interchanges (Bonneson et al., 2012), incorporated into both the HSM1 and HSM2, that (a) freeway weaving sections generally have fewer predicted crashes than similar freeway segments containing entrance and exit ramps and (b) that there is no factor included in the predictive method for weaving sections representing the effect on predicted crashes for the length of weaving section. NCHRP Project 17-71A had no basis for changing these aspects of the predictive method in the absence of new research on these issues. These issues should be addressed in any future research to revise HSM2 Chapter 17.
Another reviewer questioned why, based on the findings of NCHRP Project 17-45, which are incorporated into both the HSM1 and HSM2, the same SDFs are applicable to determining the proportions of specific crash severity levels for both multiple- and single-vehicle crashes. The reviewer expressed the viewpoint that the proportions of specific crash severity levels would likely differ between multiple- and single-vehicle crashes. NCHRP Project 17-71A had no basis for changing this aspect of the predictive method in the absence of new research on this issue. This issue should be addressed in any future research to revise HSM2 Chapter 17.
The consideration of shoulder rumble strips has been omitted from the HSM2 Chapter 17 procedure because the draft procedure may, in some cases, predict higher crash frequencies with shoulder rumble strips present than without them. This result was considered to be counterintuitive. The effectiveness of shoulder rumble strips on directional freeway segments should be better quantified and incorporated into a future HSM edition.
One reviewer suggested that the effect of barriers on severe motor-vehicle crashes be investigated further to avoid predicted crash frequencies for conditions with barriers present for which the estimate of severe crash frequency is higher than the estimate for severe crashes without a barrier present. This will be a challenging issue to find a general solution for because the difference in crash frequency between the two conditions depends on the condition behind the barrier.
Further research is suggested on predicting crashes at system interchanges where mainline lanes split, get dropped, or get added.
HSM2 Chapter 18—Predictive Method for Ramps
Research is needed to develop a crash prediction model for crossroad ramp terminals with all-way stop control. Research is also needed to develop a crash prediction model for the ramps and ramp terminals at diverging diamond interchanges, and research is needed to incorporate the effect of right-turn acceleration lanes in predictive models for ramp terminals. Research is needed as well to address closely spaced arterial intersections on the crossroad in predictive models for ramp terminals.
HSM2 Part D—Crash Modification Factors
HSM2 Chapter 19—Selecting CMFs
Research is needed to better understand how long a CMF is valid/reliable and at what point practitioners need to start considering a CMF to be outdated.
HSM2 Chapter 20—Applying CMFs
Research is needed to combine, as appropriate, some of the CMFs in the FHWA CMF Clearinghouse for the same treatment so that practitioners do not need to wrestle with how to combine them. Research is also needed to develop practical guidance on the maximum number of CMFs that should be used.
