Safety Evaluation of Access Management Policies and Techniques
CHAPTER 13. CONCLUSIONS
This research was performed to develop corridor-level crash prediction models to estimate and analyze the safety effects of selected AM techniques for different area types, land uses, roadway variables, and traffic volumes. More than 600 mi of detailed corridor data were collected across four regions of the United States to facilitate the model estimation process. It was not possible to develop a single model for each crash type and land use scenario because of the strong correlations among many of the variables of interest. As a result, 41 crash prediction models were estimated for specific land use and crash type scenarios. In most cases, multiple models are presented for each land use and crash type scenario; the alternate models contain subsets of AM strategies in an attempt to account for strong correlations among variables. A four-step process is provided to guide users through the model selection and application process, but it is envisioned that a simple software tool will be developed to simplify this process based on the functional specifications. Several sample problems are also provided to illustrate the various uses of the models and to demonstrate the model selection and application process.
These models represent the first of their kind for evaluating the safety effects of AM strategies at the corridor level based on national data. Although the results of this research will help to advance the knowledge base and state of the practice, the crash prediction models are not without limitations, including the following:
- Omitted variables. Ideally, a single-crash prediction model would include all desired variables of interest. This was not a preferred option in this study because of the strong correlation among several of the independent variables. To overcome issues related to correlation, all variables could not be included in a single model. Other variables were omitted because of illogical effects and lack of statistical significance. As a result, most models have few variables, and median type is not represented in most models.
- Inability to quantify effects of turning restrictions. Detailed data were collected to identify the type of access points (e.g., residential versus commercial driveway) and the associated turning restrictions (e.g., full-movement, right-in/right-out, and left-from-major-only). Incorporating this information in the models proved difficult. Variables were created to represent these characteristics at the corridor level, but the results were not statistically significant. Although detailed data are available for each point, the models were not developed to assess the impacts of individual points (i.e., a specific driveway or intersection). Therefore, differences between full- and limited-movement access points and between three-legged and four-legged intersections are not clear from these models.
- Lack of volumes on cross streets and driveways. Traffic volume is a key variable in predicting crashes. The objective of this study was to develop corridor-level crash prediction models, so a weighted average of the traffic volume along the corridor was used to account for exposure. The major road volume was included, but the minor road volume and driveway volumes were not included.
- Inability to quantify effects of interchange cross-road spacing. Detailed data were collected to represent various characteristics of interchange crossroads (e.g., distance from ramp terminal to nearest turning opportunity); however, relatively few interchanges were included in the dataset, and the results were not statistically significant.
Based on the results of this research and lessons learned during the completion of the study, there are several opportunities for future research as follows:
- Increase sample size and regional diversity. There is an opportunity to increase the number of sites and years of data in the database. Increasing the sample size will likely improve the models and allow for additional analysis of the variables of interest. Specifically, this effort could focus on resolving the shortcomings noted previously.
- Corroborate results. Cross-sectional methods are useful for developing crash prediction models, but there are several sources of potential bias as discussed in chapter 3. Rigorous before–after studies are preferred for estimating the effects of an individual strategy (e.g., AM characteristic). There is an opportunity to corroborate the results of these crash prediction models by collecting additional data to undertake before–after evaluations of each individual strategy.
- Separate models for nondriveway and driveway crashes. This study estimated models for a variety of crash types, including total, injury, turning, rear-end, and right-angle. It may be of interest to estimate additional models to explore the effects of specific AM strategies on driveway and nondriveway crashes. This was not possible as part of this study because of the lack of specific information in the crash data (i.e., California does not indicate driveway-related crashes). Additional research could investigate the suitability of developing these separate models while considering the potential for extensive geographic diversity in how driveway and nondriveway crashes are defined.
- Develop Highway Safety Manual–type algorithms. The AASHTO Highway Safety Manual provides methods for estimating the expected number of crashes for individual intersections and homogeneous segments.(20) These estimates can be combined to estimate the crashes for a given corridor. The AASHTO Highway Safety Manual uses a system of base models to predict crashes for an average scenario, and adjustment factors (i.e., CMFs) are used to adjust the base predictions to reflect actual conditions.(20) The models developed in this study are corridor-level models, but there may be an opportunity to use these as base models for average conditions and apply corridor-level adjustment factors to reflect actual corridor conditions. Additional research could investigate the suitability of using these models for this purpose.