U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
TECHBRIEF |
This techbrief is an archived publication and may contain dated technical, contact, and link information |
Publication Number: FHWA-HRT-15-038 Date: August 2015 |
Publication Number: FHWA-HRT-15-038 Date: August 2015 |
PDF Version (2.37 MB)
PDF files can be viewed with the Acrobat® Reader®
FHWA Publication No.: FHWA-HRT-15-038 |
Access management is the process that provides (or manages) access to land development while simultaneously preserving the flow of traffic on the surrounding road network for safety, capacity, and speed. Access management provides important benefits to the transportation system. These benefits have been increasingly recognized at all levels of government, and a growing number of States, cities, counties, and planning regions are managing access by requiring driveway permit applications and establishing where new access should be allowed. They are also closing, consolidating, or improving driveways, median openings, and intersections as part of their access management implementation strategy. However, these decisions are often challenged for various reasons.
Additional information is needed to help guide decisions related to access management. This information will help agencies better explain the safety and operational benefits of their policies and practices. Previous studies and empirical evidence have shown positive operational and safety benefits associated with good access management practices. While the operational effects of access management have been investigated quantitatively through different modeling and analysis approaches, there have been few scientifically rigorous evaluations to quantify safety effectiveness, particularly for corridor access management. The Federal Highway Administration initiated this study to help fill some of the research gaps, namely quantifying the safety impacts of corridor access management decisions.
The objective of this research was to develop corridor-level crash prediction models to evaluate the potential safety effects of access management strategies. Agencies can apply the algorithms to assess the safety impacts of their decisions related to access management.
The intent of this study was to focus on corridors based on functional classification, area type, and land use. All corridors included in this study are functionally classified as arterials and fall under one of nine area type/land use scenarios. Table 1 identifies the nine area type/land use categories and provides a definition for each area type.
Area Type | Land Use |
Urban: Metropolitan area with population of at least 250,000. |
Residential |
Commercial | |
Mixed Use | |
Suburban: Nearby areas with population of 50,000 to 250,000. |
Residential |
Commercial | |
Mixed Use | |
Urbanizing: Areas with build-out plans to reach or exceed population of 50,000. |
Residential |
Commercial | |
Mixed Use |
Residential and commercial areas are characterized by the type of development but are also differentiated by the type and distribution of vehicle types accessing the areas. Residential areas serve mainly passenger cars, and commercial areas serve a larger proportion of heavy vehicles. Commercial areas are generally defined as those areas with office buildings and other businesses that operate primarily during normal business hours on weekdays. Commercial areas, as defined in this study, do not include large shopping centers (e.g., malls) that have a larger percentage of trips on the weekends. Mixed-use area types are defined as those areas with a balanced mix of both commercial and residential establishments and access points. Figure 1 and figure 2 provide examples of corridors included in the study.
Table 2 identifies the access management strategies considered in this research project and notes the related access management safety principle.
Access Management Strategy/Policy | Limit Conflicts | Separate Conflicts | Reduce Conflicts |
ACCESS SPACING | |||
Unsignalized access spacing | ● | ||
Traffic signal spacing | ● | ||
Interchange crossroad spacing | ● | ||
Corner clearance | ● | ||
ROADWAY CROSS-SECTION | |||
Median type: TWLTL | ● | ||
Median type: Non-traversable median | ● | ||
Median type: Replace TWLTL with non-traversable median | ● |
● | |
Directional median opening | ● | ||
Median opening spacing | ● | ||
PROPERTY ACCESS | |||
Frontage/backage roads | ● |
● | |
Internal cross-connectivity | ● |
● |
TWLTL = Two-way left-turn lane.
Methodology
An observational cross-sectional study design was employed for this project. The safety impact of a given feature can be derived from a cross-sectional study by comparing the safety of a group of sites with that feature to the safety of a group of sites without that feature. Multiple variable regression models were used to estimate the effects of one feature while controlling for other characteristics that vary among the sites. These cross-sectional models are also called crash prediction models, which are mathematical equations that relate crash frequency to site characteristics. While cross-sectional models provide a means to estimate the safety impacts of access management strategies, potential issues need to be addressed. The following potential sources of bias were identified in this study with an explanation of how they were addressed or dismissed:
Data Collection
Detailed data were collected for more than 600 mi of corridors across four regions of the United States. The regions included North Carolina (Raleigh, Cary, and Wake Forest), Minnesota (St. Paul and Minneapolis), Northern California (Oakland, Sacramento, San Francisco, and San Jose), and Southern California (Los Angeles and San Diego). This section identifies the procedures for selecting corridors, collecting and verifying data, and merging the various sources of data for analysis.
Identifying Candidate Corridors
State and local agencies were contacted to solicit candidate corridors for inclusion in the study. Guidance was provided on what constituted suitable corridors to assist the State and local agencies with this process. The critical factors for corridor selection included the following:
Collecting Highway Safety Information System (HSIS) Data
HSIS contains readily available crash, roadway, and traffic volume data for select States. By design, the three States included in this study are all members of HSIS. The HSIS guidebooks were examined, and any potentially useful variables were requested. At the time of the data request, the most recent year of available data was 2008. As such, the study period for this project was from 2006 to 2008.
Corridor Screening
Three rounds of screening were employed to ensure that there were no major construction activities or changes along the corridors during the study period. Initial screening was conducted by the participating agencies through a review of construction records. The study team performed a second phase of screening using HSIS data, comparing specific variables across years to detect changes that would indicate construction activity (e.g., number of lanes, lane width, shoulder width, median type, median width, and mileposts). The team performed a third round of screening using historical aerial imagery. They identified high-resolution aerial imagery for the identified corridors from the United States Geological Survey National Seamless Server. By comparing historical aerial images with current conditions, the team was able to identify where changes had taken place during the study period.
Supplemental Data Collection and Verification
The data obtained through HSIS were verified and augmented with additional data from video collected during field visits and aerial imagery. Detailed information such as lighting presence, visual clutter, and posted speed limit were obtained from the field videos. Aerial imagery was used to verify the land use, number of through lanes, and median type for each corridor. Aerial imagery was also used to collect information that was unavailable from HSIS such as the frontage type, presence of frontage or backage road, extent of internal cross-connectivity, condition of pavement markings, and access points (location, type, and density).
Setup of Geographic Information System (GIS) Database to Facilitate Data Collection
ArcGIS™ feature classes were created for signalized intersections, unsignalized intersections, driveways, and medians. This process allowed data collectors to insert symbols representing these objects on the aerial images of the corridors. Data fields were created for each object so its characteristics could be noted. The characteristics collected for each object are summarized in table 3.
Object Type | Characteristics |
Driveways | · Type (commercial or residential) · Movements permitted (limited movement or full movement) |
Median openings and crossovers | · Presence of left-turn lane |
Unsignalized intersections | · Type (two-way stop control, all-way stop control, or roundabout) · Presence of left-turn lane(s) on mainline · Presence of right-turn lane(s) on mainline · Presence of left-turn lane(s) on cross street · Presence of right-turn lane(s) on cross street · Movements permitted (right in right out, left from major only, or full) · Maximum number of through lanes on the cross street |
Signalized intersections | · Number of approaches · Presence of left-turn lane(s) on mainline · Presence of right-turn lane(s) on mainline · Presence of left-turn lane(s) on cross street · Presence of right-turn lane(s) on cross street · Presence of non-traditional accommodation of left turns · Maximum number of through lanes on the cross street |
Figure 3 provides an example of these objects for a 1-mi section of a study corridor (California Route 1). In total, the study corridors contained more than 1,500 signalized intersections, 3,500 unsignalized intersections, and 15,000 driveways.
Post-Processing
Data were obtained in various formats. Some information was provided in Microsoft® Word documents (e.g., area type, land use, and frontage type), specifying the beginning and ending points of each corridor. HSIS data were provided in Microsoft® Excel format. General corridor information and specific attributes for signalized intersections were identified in the video logs. Other information was identified and stored in the form of ArcGIS™ feature datasets (e.g., intersections and driveways). Due to the multiple formats, a post-processing step was required to transform these data sources into a well-integrated database to serve as the basis for statistical analysis. The following tasks were performed as part of the post-processing:
The segmentation process required that links be combined into study corridors to achieve a reasonable length for analysis. In this way, some variables were summed over all links making up a study corridor (e.g., number of driveways). In other cases, new variables reflecting the percentage of the total length were created (e.g., number of lanes). The average annual daily traffic (AADT) and percentage truck variables were calculated as weighted averages, weighting by the lengths of the links within a corridor.
Summary Statistics
The result of the data collection and post-processing totalled 245 corridors representing over 600 mi and approximately 6,500 crashes. Table 4 presents the corridor mileage by area type and land use. Table 5 presents a summary of the crashes that occurred along the study corridors from 2006 to 2008. The summary statistics are based on corridor-level data and represent the crash density (i.e., number of crashes per mile per year) for various crash types. The crash types are defined in the Results section.
Scenario | Commercial | Mixed Use | Residential | Subtotal |
Urban | 79.1 | 92.4 | 48.7 | 220.2 |
Suburban | 64.3 | 119.7 | 57.0 | 241.0 |
Urbanizing | 63.8 | 31.9 | 62.4 | 158.3 |
Subtotal | 207.4 | 244.1 | 168.2 | 619.5 |
Crash Type | Region | Corridors | Minimum | Maximum | Mean | Standard Deviation |
Total |
North Carolina | 74 | 0.84 | 195.31 | 28.57 | 25.91 |
Northern California | 61 | 0.18 | 64.52 | 20.16 | 15.27 | |
Southern California | 51 | 1.14 | 108.99 | 23.29 | 19.66 | |
Minnesota | 59 | 3.10 | 140.26 | 33.32 | 27.55 | |
Injury |
North Carolina | 74 | 0.00 | 40.74 | 7.22 | 5.70 |
Northern California | 61 | 0.00 | 24.29 | 8.34 | 6.35 | |
Southern California | 51 | 0.33 | 33.61 | 10.34 | 6.55 | |
Minnesota | 59 | 1.19 | 52.39 | 11.07 | 10.80 | |
Rear-end |
North Carolina | 74 | 0.00 | 69.63 | 12.91 | 11.76 |
Northern California | 61 | 0.00 | 25.57 | 8.82 | 7.07 | |
Southern California | 51 | 0.00 | 43.60 | 9.60 | 9.27 | |
Minnesota | 59 | 0.33 | 67.83 | 16.13 | 16.33 | |
Right-angle |
North Carolina | 74 | 0.00 | 52.10 | 4.31 | 6.39 |
Northern California | 61 | 0.00 | 16.67 | 2.32 | 3.09 | |
Southern California | 51 | 0.00 | 14.99 | 2.46 | 2.77 | |
Minnesota | 59 | 0.36 | 27.64 | 5.72 | 6.10 | |
Turning |
North Carolina | 74 | 0.00 | 24.44 | 4.94 | 4.73 |
Northern California | 61 | 0.00 | 17.18 | 5.23 | 4.23 | |
Southern California | 51 | 0.33 | 35.88 | 7.19 | 6.29 | |
Minnesota | 59 | 0.00 | 23.93 | 2.49 | 3.52 |
Analysis
GLM techniques were applied to estimate the models. A negative binomial error structure was specified, following the state-of-the-art in modeling crash data. The negative binomial structure is now recognized as being more appropriate for crash counts than the normal distribution that is assumed in conventional regression modeling. Crash counts per year by crash type were used as estimates of the dependent variable, while corresponding roadway characteristics and traffic volume data were used as the independent variables.
The first step in the analysis process was to develop a model using only the AADT as a predictor variable and both the number of years and corridor length as offset variables. The general form of this model is given by figure 4.
The next step was to investigate additional variables. This investigation involved entering each variable one at a time such that only AADT and the new variable of interest were included. The estimated parameter and its standard error were examined to determine the following:
Alternate model forms were explored using the procedure described by Hauer and Bamfo.(2) It was determined that the exponential model form is appropriate due to its flexibility, and this form was retained for development of the final models.
Pearson correlation statistics were computed for each independent variable and all crash types per mile-year. The correlation matrix was not the primary driver of model-building but helped to identify those variables most associated with the different crash types. The correlation matrix also helped identify independent variables that were highly correlated. High correlation between independent variables can be problematic in developing models. Specifically, the inclusion of highly correlated variables can lead to illogical results. While this issue can be avoided by omitting a highly correlated variable, this omission limits the practicality of the results when determining the safety impacts of the omitted variable. To overcome this challenge, a series of models were estimated with various combinations of variables, which helped address issues related to correlation and provide information for all variables of interest.
Following the development of preliminary models, feedback was requested from a steering committee as to which of the variables were most desired from a practical perspective. Not all variables could be included in the models due to sample size limitations and correlation between potential explanatory variables. As such, the steering committee was asked to identify the explanatory variables that would be most useful to practitioners. The following variables were indicated to be most important for practical use according to the feedback:
All of these variables were included in various models except for posted speed limit. Vehicle speed was related to the severity of a crash, but the posted speed limit was not included in these models because it was statistically insignificant after accounting for other variables. Posted speed tends to be highly correlated with other variables such as access density and frontage type, which is likely why it could not be included in the final models. It is also possible that posted speed is not providing an accurate representation of the actual speeds (i.e., operating speed may be a better alternative for capturing the impacts of speed).
Results
One or more models were successfully calibrated for each land use type and crash type combination. The three land use types are mixed use, commercial, and residential. The crash types include total, injury, turning, rear-end, and right-angle crashes. Note that individual models by crash type cannot be summed to estimate total crashes. Also, each State has specific crash codes, and, as such, the definitions vary slightly. The crash types are identified in table 6 with the associated definitions for each region.
Crash Type | Definition |
Total | · All regions: Defined as all crashes |
Injury | · All regions: Defined as KABC on KABCO scale |
Turning | · California: Defined as any involved vehicle making a turn · Minnesota: Defined as left turn or right turn · North Carolina: Defined as rear-end turn, left-turn same roadway, left-turn different roadway, right-turn same roadway, right-turn different roadway* |
Rear-end | · California and Minnesota: Defined as rear-end · North Carolina: Defined as rear-end slow or stop and rear-end turn* |
Right-angle | · California: Defined as broadside and no vehicle was turning · Minnesota: Defined as right angle · North Carolina: Defined as angle |
KABCO = KABCO injury severity scale, where K = Fatal, A = Incapacitating injury, B = Non-incapacitating injury,
C = Possible injury, and O = Property damage only.
*North Carolina crashes coded as rear-end turn crashes are included in both rear-end and turning crashes. Because the specific crash types cannot be summed to get total crashes, it was decided that double-counting should not pose a problem for the crash type models.
In the modeling phase, the treatment of area type and regional variables required further resolution. Within each land use type (i.e., mixed use, commercial, and residential), each corridor was identified as being located within an urban, suburban, or urbanizing area. All area types were combined within the respective land use type in order to develop reliable models. A factor variable was included in each model to account for any differences due to area type, but the differences were minor and statistically insignificant. This is not to say that there was no difference in crash patterns by area type, but the data did not allow this relationship to be quantified. It is also likely that area type is better described by other variables in the model. For example, the traffic volume, number of lanes, access density, and frontage development can be used to describe the characteristics of a corridor and are more quantitative than defining a corridor as urban, suburban, or urbanizing. As such, area type was not included as an independent variable in the final models.
An indicator variable was included in each model to identify the region in which the corridor is located (i.e., North Carolina, Minnesota, Northern California, or Southern California). This variable accounts for differences between regions such as those related to crash reporting practices, driver demographics, weather, and other non-access-related factors affecting reported crashes. The factors for Northern and Southern California were similar and sufficiently close to be considered as one region. Similarly, the factors for Minnesota and North Carolina were sufficiently similar to consider them as one region. The aggregate regions help to increase sample sizes within the models (i.e., two regions instead of four) and reflect the similarities between the aggregated regions. Note that the models presented in this section include a variable to identify the applicable region. Users should select an applicable region based on a comparison between the corridor of interest and the summary statistics in the full report, and not based on geographic proximity.
The models in this section are presented in one of two forms. In most cases, the model form is represented by figure 5. In these cases, the result is expressed as crashes per mile per year. In other cases, the traffic volume variable is statistically insignificant, indicating a linear relationship between traffic volume and crashes. In these limited cases, the model form is reduced to figure 6, and the result is expressed as crashes per million vehicle-miles.
Where:
In either case, the same general procedure is followed to select an appropriate model and compute the predicted crashes. Further discussion of model selection and related examples are provided in the full report. The model coefficients and dispersion parameters are provided in table 7 through table 9. The dispersion parameter (k) is provided to help select an appropriate model when multiple options are available. The following factors (in priority order) may be considered in the selection of an appropriate model if more than one option is available:
Variables | Total | Injury | Turning |
Rear-End | Right-Angle | |||||||
1 | 2 | 3 | 1 | 2 | 1 | 2 | 3 | 1 | 1 | 2 | 3 | |
Intercept | -3.1845 (1.9550) | -3.2905 (1.8743) | -0.8926 (0.5021) | -3.5700 (1.7816) | -1.7775 (0.5964) | -2.1083 (0.4338) | -2.0792 (0.3963) | -0.4146 (0.7632) | -3.3091 (0.6700) | -5.8048 (1.9472) | -5.2671 (2.1768) | -2.1485 (0.6851) |
Region | 1.1410 (0.2316) | 1.0533 (0.2086) | 0.6166 (0.1013) | 0.5695 (0.1980) | 0.2465 (0.0931) | 0.9647 (0.2843) | 0.8015 (0.2354) | -0.3163 (0.1301) | 0.8113 (0.1136) | 1.8390 (0.2616) | 1.2134 (0.2457) | 1.2344 (0.1377) |
AADT | 0.5187 (0.1819) | 0.5266 (0.1738) | 0.3766 (0.0468) | 0.5010 (0.1659) | 0.3880 (0.0558) | 0.2179 (0.0729) | 0.5015 (0.0618) | 0.4656 (0.1856) | 0.5678 (0.2103) | 0.2433 (0.0648) | ||
ACCDENS | 0.0053 (0.0044) | 0.0088 (0.0061) | 0.0112 (0.0051) | |||||||||
MEDOPDENS | 0.1901 (0.0884) | |||||||||||
PROPDIV | -0.4710 (0.3461) | |||||||||||
PROPFULLDEV | 0.6787 (0.1846) | |||||||||||
PROPLANE1 | -0.5185 (0.3789) | -0.6376 (0.3796) | -0.5814 (0.3582) | -0.6623 (0.1404) | -0.5548 (0.1713) | |||||||
PROPNODEV | -0.4252 (0.2268) | -0.3159 (0.2201) | -0.5890 (0.2827) | |||||||||
PROPVC | ||||||||||||
PROPTWLTL | ||||||||||||
SIGDENS | 0.1095 (0.0607) | 0.0957 (0.0594) | 0.1239 (0.0556) | 0.1865 (0.0754) | 0.1797 (0.0742) | 0.0621 (0.0380) | 0.2284 (0.0637) | |||||
UNSIGDENS | 0.0471 (0.0224) | 0.0582 (0.0323) | ||||||||||
Dispersion (k) | 0.5073 | 0.4897 | 0.5165 | 0.4248 | 0.4151 | 0.7920 | 0.7780 | 0.7791 | 0.6098 | 0.5585 | 0.6796 | 0.7674 |
Variables | Total | Injury | Turning |
Rear-End | Right-Angle | ||||||
1 | 2 | 1 | 2 | 3 | 4 | 1 | 2 | 1 | 1 | 2 | |
Intercept | -0.7017 (0.6873) | -0.6854 (0.5010) | -2.0602 (0.7991) | -0.9792 (0.8386) | 0.2127 (0.7288) | -1.9690 (0.5862) | -0.9816 (0.9366) | 0.0085 (1.1277) | -3.0651 (0.6691) | -1.6746 (0.9312) | -1.9023 (0.6838) |
Region | 0.8353 (0.1883) | 0.6166 (0.1013) | 0.4672 (0.1815) | 0.2383 (0.1497) | 0.6769 (0.1559) | 0.3056 (0.0923) | -0.2548 (0.2101) | 0.8113 (0.1136) | 1.4756 (0.2388) | 1.2344 (0.1377) | |
AADT | 0.3094 (0.0660) | 0.3766 (0.0468) | 0.3649 (0.0766) | 0.3225 (0.0797) | 0.2705 (0.0697) | 0.3751 (0.0548) | 0.1650 (0.0960) | 0.1947 (0.1068) | 0.5015 (0.0618) | 0.1238 (0.0912) | 0.2433 (0.0648) |
ACCDENS | 0.0069 (0.0048) | 0.0085 (0.0047) | 0.0110 (0.0052) | 0.0165 (0.0064) | |||||||
MEDOPDENS | |||||||||||
PROPDIV | |||||||||||
PROPFULLDEV | 0.6787 (0.1846) | ||||||||||
PROPLANE1 | -0.6047 (0.2631) | -0.6244 (0.2566) | -0.5245 (0.1430) | -0.7328 (0.3577) | -0.5548 (0.1713) | ||||||
PROPNODEV | -0.4252 (0.2268) | -0.6472 (0.3040) | -0.6967 (0.4150) | ||||||||
PROPVC | 0.5421 (0.1990) | ||||||||||
PROPTWLTL | |||||||||||
SIGDENS | 0.1002 (0.0523) | 0.0566 (0.0512) | 0.1075 (0.0300) | 0.1995 (0.0660) | 0.0621 (0.0380) | 0.1532 (0.0658) | |||||
UNSIGDENS | |||||||||||
Dispersion (k) | 0.4890 | 0.5165 | 0.4406 | 0.4228 | 0.4739 | 0.3951 | 0.7140 | 0.7802 | 0.6098 | 0.7288 | 0.7674 |
Variables | Total | Injury | Turning | Rear-End | Right-Angle | ||||||||
1 | 2 | 3 | 1 | 2 | 1 | 2 | 3 | 1 | 2 | 3 | 1 | 2 | |
Intercept | -0.5615 (0.7076) | -1.3644 (0.4953) | -1.1048 (0.4876) | -2.7357 (0.8556) | -2.7379 (0.9147) | -1.1275 (1.1225) | -0.9528 (0.7286) | -0.7154 (0.7477) | -3.8941 (0.9816) | -2.6180 (1.0221) | -3.3056 (0.6549) | -1.4079 (1.0732) | -2.1173 (0.6540) |
Region | 0.4443 (0.1533) | 0.6850 (0.1107) | 0.6166 (0.1013) | 0.1656 (0.1423) | 0.2303 (0.1603) | -0.6520 (0.2073) | -0.1651 (0.1339) | -0.3163 (0.1301) | 0.5803 (0.1984) | 0.5406 (0.1865) | 0.8113 (0.1136) | 0.8858 (0.2180) | 1.1970 (0.1314) |
AADT | 0.3094 (0.0673) | 0.3883 (0.0463) | 0.3766 (0.0468) | 0.4189 (0.0820) | 0.4615 (0.0867) | 0.1826 (0.1059) | 0.1759 (0.0708) | 0.2179 (0.0729) | 0.5392 (0.0945) | 0.4782 (0.0967) | 0.5015 (0.0618) | 0.1332 (0.1051) | 0.1768 (0.0639) |
ACCDENS | 0.0032 (0.0022) | 0.0052 (0.0028) | 0.0044 (0.0028) | ||||||||||
MEDOPDENS | |||||||||||||
PROPDIV | |||||||||||||
PROPFULLDEV | 0.3371 (0.2317) | 0.3720 (0.2273) | 0.4295 (0.3125) | ||||||||||
PROPLANE1 | -0.5479 (0.1702) | -0.4040 (0.1669) | -0.6125 (0.1715) | -0.8174 (0.2078) | -0.5548 (0.1713) | -0.3633 (0.2383) | |||||||
PROPNODEV | -0.4252 (0.2268) | -0.5890 (0.2827) | |||||||||||
PROPVC | |||||||||||||
PROPTWLTL | -0.5600 (0.2439) | ||||||||||||
SIGDENS | 0.1262 (0.0629) | 0.2081 (0.0539) | 0.2244 (0.0818) | 0.1821 (0.0426) | 0.1675 (0.0864) | 0.0621 (0.0380) | 0.2267 (0.0750) | 0.2084 (0.0390) | |||||
UNSIGDENS | 0.0635 (0.0283) | ||||||||||||
Dispersion (k) | 0.3277 | 0.5181 | 0.5165 | 0.2663 | 0.3220 | 0.5792 | 0.7030 | 0.7791 | 0.5541 | 0.4803 | 0.6098 | 0.5555 | 0.6790 |
Once a model is selected, the parameter estimates and the characteristics of the corridor of interest are used to compute the predicted crashes for the corridor. The following example provides sample results and a demonstration of how a given model can be used to compute the predicted crashes. Further details and examples are provided in the full report.
Example
Assume an analyst would like to predict the total number of crashes for a mixed-use corridor in North Carolina, and model 1 for total crashes is selected from table 7. The corridor is described by the following characteristics: AADT of 25,000 vehicles per day, 10 signalized intersections, 30 unsignalized intersections, and 80 driveways. The total corridor length is 2.5 mi, of which 0.625 mi is two lanes. Figure 5 presented the model form and sample estimates for the example problem.
The intercept and region are constants, and the region coefficient is included in this case since the corridor of interest is similar to the corridors in North Carolina. The intercept and region coefficients are summed and included in the first exponential term of the model. The traffic volume is identified for the scenario of interest and input as the AADT term in the model. The coefficient for traffic volume is identified from the table and input as the b term in the model. The appropriate values for relevant access management variables are then identified and input as the X1 – Xn terms in the model. Finally, the corresponding coefficients for access management variables are identified from the table and input as the c1 to cn terms in the model. (Note that the coefficients c1 to cn correspond to predictor variables X1 to Xn, respectively.)
In this case, the model coefficients from table 7, model 1, are intercept (-3.1845), region (1.1410), AADT (0.5187), ACCDENS (0.0053), SIGDENS (0.1095), and PROPLANE1
(-0.5185). The AADT is 25,000 vehicles per day. Predictor variables X1 to Xn are defined as follows:
The predicted number of crashes per mile per year is computed as seen in figure 5. With the values used in model 1, crashes per mile per year would be computed as seen in figure7:
In the previous example, a model was selected and applied to predict the number of total crashes for a corridor with specific characteristics. This example is just one potential use of the corridor crash prediction models. The following discussion identifies the two basic applications of the corridor crash prediction models from this research:
The full report provides additional guidance on the selection and application of the most appropriate models depending on the intended application. Six typical scenarios are discussed in detail, and sample problems are provided in the full report to further illustrate the application of models in the six scenarios. The six scenarios include the following:
The following guiding principles are common to all scenarios.
The results of this project will help users better understand the safety implications of their decisions related to access management. Specifically, users can apply the models to assess the relative safety effects of one or more contemplated strategies (or combinations), or they can compare the benefits and costs of two or more alternative strategies (or combinations). It is recommended that a safety evaluation software tool be developed to help users select and apply an appropriate model or set of models. Functional specifications were developed as part of this project to facilitate the development of such a tool. The specifications include a detailed description of the model selection process and identify the required and optional inputs as well as default values for the various scenarios included in this study.
This research was performed to develop corridor-level crash prediction models to estimate and analyze the safety effects of selected access management 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 due to 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 in which the alternate models contain subsets of access management strategies in an attempt to account for strong correlations among variables. A four-step process is provided in the full report to guide users through the model selection and application process, but it is envisioned that a basic software tool will be developed to simplify this process based on functional specifications. Sample problems are provided in the full report 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 access management strategies at the corridor level based on national data. While the results of this research will help advance the knowledge-base and state of the practice in access management, the crash prediction models are not without limitations. Specific limitations of the models include the following:
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:
Researchers—This study was performed by Vanasse Hangen Brustlin, Inc. For more information about this research, contact Wei Zhang, FHWA Project Manager, HRDS-10 at (202) 493-3317, wei.zhang@dot.gov.
Distribution—This TechBrief is being distributed according to a standard distribution. Direct distribution is being made to the Divisions and Resource Center.
Availability—The report may be obtained from the FHWA Product Distribution Center by e-mail to report.center@fhwa.dot.gov, fax to (301) 577-1421, phone to (301) 577-0818, or online at http://www.tfhrc.gov/safety.
Key Words—Access management, safety analysis, crash prediction models.
Notice—This document is disseminated under the sponsorship of the U.S. Department of Transportation in the interest of information exchange. The U.S. Government assumes no liability for the use of the information contained in this document. The U.S. Government does not endorse products or manufacturers. Trademarks or manufacturers’ names appear in this report only because they are considered essential to the objective of the document.
Quality Assurance Statement—The Federal Highway Administration (FHWA) provides high-quality information to serve the Government, industry, and public in a manner that promotes public understanding. Standards and policies are used to ensure and maximize the quality, objectivity, utility, and integrity of its information. FHWA periodically reviews quality issues and adjusts its programs and processes to ensure continuous quality improvement.