schultzprioritizingsafetyfundingusing2025.md 48 KB
category: literaturenote citekey: schultzprioritizingsafetyfundingusing2025 title: Prioritizing safety funding using severity weighted risk scores authors: "Schultz, Grant G.; Aristizabal, Tomas Barriga; Ritchie, Jace; Warr, Richard L." year: 2025 date: 2025-00-00 2025 doi: 10.1016/j.aap.2025.108083 publication: "Accident Analysis & Prevention" url: "https://doi.org/10.1016/j.aap.2025.108083" zotero_key: M3KNPD5E zotero_storage: BK5X55RB collections: transpordi_planeerimine folder: 001_artiklid firstAuthor: "Schultz, Grant G."
status: converted
Contents lists available at ScienceDirect
Accident Analysis and Prevention
journal homepage: www.elsevier.com/locate/aap
Grant G. Schultz a,* , Tomas Barriga Aristizabal b , Jace Ritchie c , Richard L. Warr d
- a Department of Civil and Construction Engineering, Brigham Young University, 430 EB, Provo, UT 84602, USA
- b AECOM, 756 E Winchester St., Salt Lake City, UT 84107, USA
- c Mathematics and Statistics Department, Utah State University, 3900 Old Main Hill, Logan, UT 84322, USA
- d Department of Statistics, Brigham Young University, 2152 WVB Provo, UT, USA
ARTICLE INFO
Keywords: Weighted risk score Crash-based statistical analysis Severity distribution Safety improvements Safety data analytics Safety modeling
ABSTRACT
Limited funding availability requires government agencies to focus transportation funding on locations most in need of safety improvements. The Two-Output Model for Safety (TOMS) was created to prioritize segments and intersections for safety analysis. The TOMS compiles input data, prepares the data files so that the format and content are consistent, and then two different processes occur. The first is to segment roadways based on five variables: average annual daily traffic, functional class, lanes, speed limit, and urban code. The second is to assign the physical characteristics of the roadway to each individual intersection. TOMS outputs a segment file and an intersection file used for statistical analysis and are the "two outputs" referenced by the name of the model. The segments and intersections are analyzed using severity and total number of crashes at the sites. An excess weighted risk score was developed using an equivalent property damage only value to analyze the severity and number of crashes concurrently. The segments and intersections with the highest excess weighted risk scores are prioritized as locations for safety funding. A report compiler is then executed to create two-page safety reports that contain roadway and crash information organized in a manner that allows governing agencies to identify how many crashes are occurring at a site and the manner of collision for the crashes. The research presented in this paper shows that the simultaneous use of intersection and segment analysis combined with excess weighted risk scores can provide insight into the prioritization of safety funding.
1. Introduction
Because transportation agencies have limited budgets, it is important for these agencies to prioritize safety funding and identify locations that will benefit the most from safety improvements. By combining roadway and crash data, both roadway segments and intersections can be analyzed to prioritize funding. These segments and intersections can then be evaluated by comparing the expected crash frequency to the observed crash frequency, after which improvements to those sites can be determined. Those areas with the largest difference between expected and observed crash frequency in comparison to similar sites, are the priority areas for safety funding.
Prioritizing safety funding in Utah was the subject of a study of a joint effort between the Utah Department of Transportation (UDOT) and the Brigham Young University (BYU) Civil and Construction Engineering and Statistics Departments. Previous studies on this subject resulted in the emergence of two tools to assist in the prioritization of safety
funding and proposing potential countermeasures. These are the Crash Analysis Methodology for Segments (CAMS) (Schultz et al., 2020; Schultz et al., 2023) and the Intersection Safety Analysis Methodology (ISAM) (Schultz et al., 2020; Schultz et al., 2018; Gibbons et al., 2019). These tools were used to perform analysis of all crashes on highways within the state of Utah, the CAMS focusing on segments and the ISAM focusing on intersections. The tools were developed so that crashes do not overlap between the two tools. Technical reports were created with the results from both tools and sent to the UDOT Traffic and Safety Division with roadway and crash information, as well as proposed countermeasures.
The methodologies for CAMS and ISAM each used two different statistical models, one accounting for total crashes named the Prediction model, and another accounting for severity of crashes, named the Severity model. The two statistical models have been used to prioritize locations for safety funding, but some locations identified in one model would not be identified in the other. These methodologies did not
E-mail addresses: gschultz@byu.edu (G.G. Schultz), warr@stat.byu.edu (R.L. Warr).
☆ This article is part of a special issue entitled: 'RSS 2024' published in Accident Analysis and Prevention.
* Corresponding author.
provide a way to analyze crash count and severity concurrently.
The first goal of this research was to improve the CAMS and ISAM to allow them to be more effective in assisting in safety prioritization and implementing countermeasures. CAMS and ISAM were built within a Macro-Enabled Workbook using Visual Basic for Applications (VBA) code to compile the data necessary to perform the analysis. These two models were rebuilt and combined as the Two-Output Model for Safety (TOMS) in R. R is a programming language built for statistical computing that allows for more efficient and consistent data management resulting in this singular model. The creation of TOMS ensures tool improvements are incorporated including adjustments to safety prioritization analysis, changes to the ranking procedure of sites, and modifications to the "intersection related" criteria using UDOT guidelines.
Another goal of this research was to develop segment and intersection severity distributions. Previously, the CAMS and ISAM models provided prioritization rankings based on total crashes and injury crashes separately. The new Joint Statistical Model (JSM) creates a single ranking that compares the number and severity of observed crashes to the expected crashes. The results from the JSM are used in developing segment and intersection severity distributions which are delivered to UDOT to assist in analyzing Utah roads.
UDOT provided the most recent roadway and crash data for use in TOMS and the JSM including the biannual asset inventory data and the real-time updates of crash data. Reports generated from the analysis help the UDOT regions determine which areas need to be prioritized for safety funding. Before the reports were submitted, they were screened by the UDOT Traffic and Safety Division and only reports for suitable sites were delivered.
The scope of this project was to combine the existing CAMS and ISAM within TOMS. Also, combining the Prediction and Severity models into the JSM helped UDOT analyze crash count and severity concurrently and provide segment and intersection severity distributions for use in safety analysis. The data from the TOMS is input into the JSM, and the results are input into the report compiler to generate two-page safety reports for sites selected by UDOT to deliver to their respective regions for safety prioritization.
The following sections outline the methodology, analysis results, discussion of the results, and conclusions of the research.
2. Methodology
This section describes the methodology and processes used to compile the segment and intersection datasets and assign crashes to the datasets to create the input file for the statistical model. A brief overview of statistical model used to prioritize safety funding as well as creating severity distributions will also be discussed.
2.1. Roadway and crash data
Twelve roadway files are used within TOMS: Routes, Average Annual Daily Traffic (AADT), Functional Class, Speed Limit, Urban Code, Lanes, Intersections, Medians, Driveways, Shoulders, Utah Transit Authority (UTA) Stops, and Schools. The UDOT Open Data Portal (Utah Department of Transportation (UDOT), 2024) provided seven of these files (Routes, AADT, Functional Class, Lanes, Medians, Driveways, Shoulders). The files are exported directly from the website with data reported according to the Adjusted Linear Referencing System (ALRS) that UDOT implemented in the summer of 2020. The Speed Limit, Urban Code, and Intersections files used in this research were provided directly from UDOT as they included additional data not accessible on the website. The Utah Geospatial Resource Center (UGRC, 2024) provided the UTA Stops and Schools files.
Five principal datasets: AADT, Functional Class, Lanes, Speed Limit, and Urban Code were used to create homogenous segments which were analyzed in the research. These variables have been used in similar research dating back to 2012, where the researchers established a
Table 1 Updated Crash Costs for Use in Benefit Calculations (2020 Dollars).
| Severity | Severity No. |
Crash Cost Weighted |
Crash Cost Unweighted |
Unweighted EPDO |
|---|---|---|---|---|
| K (Fatal) | 5 | \$ 3,078,500 | \$14,010,300 | 828 |
| A (Suspected Serious Injury) |
4 | \$ 3,078,500 | \$ 805,800 | 47 |
| B (Suspected Minor Injury) |
3 | \$ 264,000 | \$ 264,200 | 16 |
| C (Possible Injury) |
2 | \$ 148,000 | \$ 148,000 | 9 |
| O (PDO) | 1 | \$ 17,000 | \$ 17,000 | 1 |
| KA (Severe) | 5,4 | \$ 3,078,500 | N/A | 182 |
| KAB (Injury) | 5,4,3 | \$ 806,700 | N/A | 48 |
| KABC (Anticipated Injury) |
5,4,3,2 | \$ 415,100 | N/A | 24 |
| KABCO (Total Crashes) |
5,4,3,2,1 | \$ 134,300 | N/A | 8 |
Table 2 Area of Influence of Intersections.
| Intersection Type | Area of Influence (ft) |
|---|---|
| Signal Control | 300 |
| Minor Leg Stop Control | 150 |
| All Way Stop Control | 100 |
| Yield Control | 100 |
| Uncontrolled | 100 |
| Roundabout | 300 |
| Offset Left-turn (CFI) | 400 |
| Median Thru-U Turn | 400 |
| R-Cut | 400 |
| SPUI | 500 |
| DDI | 400 |
| Active Transportation Only | 100 |
| Railroad Crossing | 100 |
framework for crash data analysis that included four roadway characteristics for homogeneous segmentation: AADT, functional class, number of through lanes, and speed limit (Schultz et al., 2012). Beginning in 2013, the research team added urban code as a fifth segmentation variable (Schultz et al., 2013). The details on the segmentation process and the variables used from each of these datasets is available in the literature (Schultz et al., 2012; Schultz et al., 2013).
For intersections, the intersections dataset was used and data from the five principal datasets along with any other data were added to each of the intersections. The intersection file was merged with data from all other datasets to create the intersection roadway file. To facilitate assigning crashes to the intersection, the UDOT area of influence by intersection type was used to identify which crashes to include in the intersection analysis. Crashes that fell within the area of influence were assigned to the intersection. Table 2 provides the area of influence assigned to each intersection type provided by UDOT Traffic and Safety. The actual determination of the intersection area of influence was not part of the scope of the research.
Four crash data files are used within TOMS: Crash, Location, Rollups, and Vehicle. These files were provided by UDOT Traffic and Safety along with a unique crash ID for each file so that they can be joined for the analysis. While the unique crash ID pertains to an actual crash that occurred along Utah roadways, the crash ID was only used for crash safety analysis and not made available to the public. An attribute called "Intersection Related" in the crash data was used to assign crashes to intersections or segments. For example, any crash that occurs within 100 feet of an All-Way Stop Control intersection and is marked as "Intersection Related" would be assigned to that intersection. UDOT has a procedure that they use wherein each of the crash reports is double checked for accuracy and a determination is made regarding "Intersection Related" in the report. Fig. 1 outlines the process to combine the
Fig. 1. Outline of process to combine roadway and crash files.
roadway and crash files.
2.2. Weighted risk scores
In previous years, the BYU-UDOT team developed a crash severity model called the Utah Crash Severity Model (UCSM) for segments and the Utah Intersection Crash Severity Model (UICSM) for intersections (Schultz et al., 2020; Schultz et al., 2023; Schultz et al., 2018; Gibbons et al., 2019). These models evolved into the CAMS-S for segments and the updated UICSM for intersections but used the same methodology for evaluating crash severity. The CAMS-S model was created to identify segments that may not necessarily have an unusually large number of crashes, but that have an unusually high proportion of injury crashes. In other words, the model answers the question, "If a crash was to occur on any segment, which segments are most likely to experience an injury crash?" This method works reasonably well for network screening, but it does not assign weight to crashes by their severity as recommended in the Highway Safety Manual (HSM) (AASHTO, 2010). One method of severity weighting given in the HSM is to estimate the monetary cost associated with crashes of various severities. This is usually done by assigning an estimated cost of damages to each crash severity, and then assigning an equivalent property damage only (EPDO) score based on the relative cost to the average cost of property damage only (PDO) crashes.
One way to model EPDO crashes is to apply the predictive method described in the HSM. This is reliable for modeling EPDO crashes because it uses statistical regression to account for regression to the mean bias. Therefore, the only thing which sets this method apart from the predictive method for total crashes is determining the weights of each crash severity. This is done by determining the average societal cost for each crash severity and dividing this with the average cost of PDO crashes. The actual weights for different crash severities may vary depending on location and the criteria used. Table 1 shows the weighted and unweighted crash costs for UDOT in 2021 (Utah Department of Transportation (UDOT), 2021). This EPDO method has the advantage of giving a severity score to each roadway, but it has the disadvantage of a higher ranking for sites with a small number of severe crashes. This is the case for most severity-weighted ranking procedures, so it is valid to have separate models for total crashes and severity-weighted crashes.
This research uses an estimated monetary cost associated with crashes to account for the significance of high severity crashes on roadways. With the addition of an excess weighted risk score (EWRS) based on the unweighted crash cost, the two previous models are
improved and combined. Within this paper, the phrase "roadway site" will be used to refer to both segments and intersections. The count model looks at the total number of crashes occurring at a roadway site. The severity model looks at how many of the total crashes are severe at a roadway site.
2.3. Statistical model
When modeling crash counts and severity, a model that evaluates both at the same time while also introducing a correlation term between the models can be useful. Such an approach to modeling has been used to improve upon previous models utilizing negative binomial crash count models and ordered logit fractional split severity models (Afghari et al., 2020; Yasmin and Eluru, 2018). This approach results in more accurate crash counts of both total and severe crashes compared to common models that do not use the correlation term (Afghari et al., 2020; Yasmin and Eluru, 2018). However, the proportion of severe crashes did not show significant differences and might be worse than if the severity were calculated from a model that independently calculated crash severity and crash count (Yasmin and Eluru, 2018). It is possible that other widespread models could be used to achieve higher levels of accuracy for severity proportions.
As mentioned, the count model looks at the total number of crashes occurring at a roadway site, while the severity model looks at how many of the total crashes are severe. The count model is derived from the techniques described in the literature where a joint model was used to identify road segments with high risk of fatal and serious injury crashes (Afghari et al., 2020). The count model is a negative binomial regression model. The functions used within this model are shown in Eq. (1) where i represents the number of years, j represents a roadway site (segment/ intersection), Xij is the total number of crashes at a site, uij is the mean number of crashes at a site, wij is a matrix of the attributes of the site used to estimate crash counts, β represents the average effects of the attributes on the mean, uij is a matrix of the attributes of the site used to estimate both crash counts and crash severity, and γ represents the average effects of the attributes in uij on the mean.
Xij NB(μij,r)
$$P(X{ij} = x) = \left(\frac{x+r-1}{x}\right) \left(\frac{\mu{ij}}{\mu{ij}+r}\right)^x \left(\frac{r}{\mu{ij}+r}\right)^r \tag{1}$$
Table 3 Top 10 Segment Results.
| StateRank | Region | MainRoute | County | FunctionalClass | | Observed Crashes (2016–2020) | | | | | | | | |-----------|--------|-----------|-----------|--------------------------|-------|------------------------------|----------|---|---|---|---|--|--| | | | | | | Total | Injury(3–5) | Severity | | | | | | | | | | | | | | | 1 | 2 | 3 | 4 | 5 | | | | 1 | 2 | 0080 | Summit | Interstate | 75 | 17 | 50 | 8 | 9 | 4 | 4 | | | | 2 | 4 | 0089 | Kane | Other Principal Arterial | 66 | 8 | 49 | 9 | 3 | 1 | 4 | | | | 3 | 2 | 0080 | Tooele | Interstate | 26 | 7 | 13 | 6 | 3 | 1 | 3 | | | | 4 | 4 | 0006 | Emery | Other Principal Arterial | 10 | 5 | 5 | 0 | 2 | 0 | 3 | | | | 5 | 2 | 0036 | Tooele | Major Collector | 19 | 6 | 11 | 2 | 2 | 1 | 3 | | | | 6 | 4 | 0163 | San Juan | Major Collector | 9 | 4 | 4 | 1 | 1 | 0 | 3 | | | | 7 | 1 | 0084 | Box Elder | Interstate | 36 | 8 | 27 | 1 | 3 | 2 | 3 | | | | 8 | 1 | 0084 | Box Elder | Interstate | 32 | 6 | 19 | 7 | 2 | 1 | 3 | | | | 9 | 4 | 0089 | Sanpete | Other Principal Arterial | 41 | 10 | 27 | 4 | 5 | 2 | 3 | | | | 10 | 1 | 0126 | Davis | Other Principal Arterial | 39 | 11 | 24 | 4 | 7 | 1 | 3 | | |
$$\log(\mu{ij}) = \mathbf{w}{ij}\boldsymbol{\beta} + \mathbf{u}_{ij}\boldsymbol{\gamma}$$
The count model looks at roadway and crash attributes that contribute to the total number of crashes occurring at roadway sites with similar characteristics and predicts the number of crashes for these roadway sites. The number of crashes is converted to an EWRS, and using the total number of crashes that occur at the roadway site, the actual EWRS is compared and used in the final ranking. The severity model is an ordinal multinomial model as shown in Eq. (2) where Yij is the severity of a crash at a site, ω is the standard cumulative logistic function, ρij is the latent variable used to estimate severity, θy is the adjustment to the intercept for severity y, zij is a matrix of the attributes of the site used to estimate crash severity, δ is the average effects of the attributes in zij on the latent variable, and the remaining variables are as defined previously.
$$P(Y{ij} \le y) = \omega(-\rho{ij} + \theta_y)$$ (2)
$$\rho{ij} = \mathbf{z}{ij} \boldsymbol{\delta} + \mathbf{u}_{ij} \boldsymbol{\gamma}$$
The severity model analyzes roadway and crash attributes that contribute to the different severities of crashes occurring at roadway sites with similar characteristics and predicts the severity of crashes for these roadway sites. The crashes are converted to an EWRS which is used in the final ranking.
The joint model is run for both segments and intersections using the statistical input created from combining the roadway and crash data. The sites are compared by their EWRS based on the joint count and severity model. The EWRS from the joint count and severity model are used and those roadway sites with the highest EWRS are ranked the highest. After the statistical analysis is complete, a list of all segments and intersections is created with the ranking of all sites and output as a. csv file. The lists of top 10 segments, signalized intersections, and unsignalized intersections by region are generated for UDOT to use in prioritizing safety funding.
Severity is an important factor when prioritizing projects for safety funding. Instead of using two different models, one accounting for total number of crashes and another for the total number of severe crashes, the two models were combined using an EWRS. All segments and intersections were analyzed using this method. Those with the highest EWRS were marked as higher priority for safety funding and can be screened accordingly.
3. Analysis and results
The TOMS output allows UDOT region engineers to review and determine which areas of roadway need improvement. After the data compilation and statistical analysis, two-page safety reports are created for the highest priority locations in each UDOT region. These reports contain tables of roadway and crash data to help summarize each location and assist in identifying countermeasures for improvement. This section will review the high priority segments and intersections on Utah roadways. The following subsections describe the results for roadway segments, then intersections, the development of severity distributions, and applications of this research. All results have been truncated in this paper. The full analysis results are available in the literature (Schultz et al., 2023).
3.1. Roadway sites
Table 3 contains the results of the statistical analysis conducted on the compiled segments file. The state rank, region, main route, county, and functional class information of the segment are given in the first five columns while the crash information is given in the next seven. The state ranking is determined based on the EWRS for each segment with crash data from 2016 to 2020. The top-ranking segments ranged from having nine total crashes to 75 total crashes in the five-year period from 2016 to 2020. All the segments have had three or more fatal crashes within the
Table 4 Top 10 Intersection Results.
| StateRank | Region | TrafficControl | MainRoute | County | Int.Type | Observed Crashes (2016–2020) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Total | Injury(3–5) | Severity | ||||||||||
| 1 | 2 | 3 | 4 | 5 | ||||||||
| 1 | 1 | Signalized | 0165 | Cache | 4-LEG | 31 | 4 | 19 | 8 | 1 | 0 | 3 |
| 2 | 1 | Unsignalized | 0235 | Weber | 3-LEG | 6 | 4 | 1 | 1 | 2 | 0 | 2 |
| 3 | 1 | Unsignalized | 0091 | Cache | 4-LEG | 12 | 4 | 8 | 0 | 1 | 1 | 2 |
| 4 | 2 | Signalized | 0173 | Salt Lake | 4-LEG | 37 | 8 | 19 | 10 | 3 | 3 | 2 |
| 5 | 2 | Signalized | 0173 | Salt Lake | 4-LEG | 22 | 6 | 10 | 6 | 4 | 0 | 2 |
| 6 | 1 | Unsignalized | 0126 | Weber | 3-LEG | 8 | 2 | 3 | 3 | 0 | 0 | 2 |
| 7 | 1 | Signalized | 0126 | Davis | 4-LEG | 195 | 33 | 120 | 42 | 27 | 4 | 2 |
| 8 | 2 | Signalized | 0171 | Salt Lake | CFI | 195 | 20 | 119 | 56 | 17 | 1 | 2 |
| 9 | 2 | Signalized | 0089 | Salt Lake | 4-LEG | 20 | 7 | 5 | 8 | 3 | 3 | 1 |
| 10 | 1 | Signalized | 0204 | Weber | 4-LEG | 76 | 30 | 32 | 14 | 27 | 2 | 1 |
Table 5 Average Crash Severity by Segment Type.
| Segment Type (#) | Severity | Severity 4 and 595 % Uncertainty Interval | ||||
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | ||
| Urban | ||||||
| 2-Lane (1091) | 72.88 % | 16.05 % | 8.51 % | 2.01 % | 0.55 % | (2.26 %, 2.89 %) |
| 3-Lane w/ TWLTL (53) | 71.90 % | 16.57 % | 8.84 % | 2.12 % | 0.57 % | (1.06 %, 4.55 %) |
| 4-Lane Divided (371) | 73.49 % | 15.76 % | 8.27 % | 1.95 % | 0.53 % | (2.26 %, 2.71 %) |
| 4-Lane Undivided (303) | 72.73 % | 16.11 % | 8.57 % | 2.03 % | 0.56 % | (2.13 %, 3.09 %) |
| 5-Lane w/ TWLTL (23) | 69.40 % | 17.75 % | 9.83 % | 2.36 % | 0.66 % | (1.28 %, 5.05 %) |
| 6-lane Divided (214) | 72.69 % | 16.16 % | 8.57 % | 2.02 % | 0.56 % | (2.21 %, 2.98 %) |
| 6-Lane Undivided (58) | 70.72 % | 17.14 % | 9.30 % | 2.22 % | 0.62 % | (2.03 %, 3.69 %) |
| Freeway 4-Lane (230) | 74.01 % | 15.49 % | 8.08 % | 1.90 % | 0.52 % | (1.96 %, 2.90 %) |
| Freeway 6-Lane w/ HOV (56) | 73.67 % | 15.66 % | 8.21 % | 1.93 % | 0.53 % | (2.06 %, 2.87 %) |
| Freeway 6-Lane w/o HOV (174) | 73.83 % | 15.59 % | 8.15 % | 1.91 % | 0.52 % | (2.11 %, 2.78 %) |
| Freeway 8-Lane w/ HOV (111) | 73.36 % | 15.82 % | 8.32 % | 1.96 % | 0.54 % | (2.22 %, 2.78 %) |
| Freeway 8-Lane w/o HOV (72) | 74.31 % | 15.33 % | 7.98 % | 1.87 % | 0.51 % | (1.90 %, 2.87 %) |
| Freeway 10-Lane w/ HOV (54) | 72.45 % | 16.28 % | 8.66 % | 2.05 % | 0.56 % | (2.24 %, 2.99 %) |
| Freeway 10-Lane w/o HOV (14) | 74.50 % | 15.24 % | 7.90 % | 1.85 % | 0.51 % | (1.22 %, 3.69 %) |
| Freeway 12-Lane (8) | 73.12 % | 15.93 % | 8.42 % | 1.99 % | 0.54 % | (1.29 %, 3.95 %) |
| Rural | ||||||
| 2-lane (2724) | 73.37 % | 15.82 % | 8.31 % | 1.96 % | 0.54 % | (2.28 %, 2.71 %) |
| Multilane Divided (57) | 76.21 % | 14.35 % | 7.29 % | 1.69 % | 0.46 % | (1.57 %, 2.78 %) |
| Multilane Undivided (201) | 74.62 % | 15.18 % | 7.86 % | 1.84 % | 0.50 % | (1.53 %, 3.24 %) |
| Freeway 4-Lane (454) | 74.06 % | 15.47 % | 8.06 % | 1.89 % | 0.52 % | (2.09 %, 2.73 %) |
| Freeway 6-Lane (47) | 75.55 % | 14.69 % | 7.53 % | 1.75 % | 0.48 % | (1.52 %, 3.02 %) |
Table 6 Average Crash Severity by Intersection Type.
| Intersection Type(#) | Severity | | Severity 4 and 595 % Uncertainty Interval | | | | | |------------------------------|----------|---------|-------------------------------------------|--------|--------|------------------|--| | | 1 | 2 | 3 | 4 | 5 | | | | Urban | | | | | | | | | Signalized, 4-Leg (988) | 63.50 % | 21.61 % | 12.31 % | 2.24 % | 0.34 % | (2.34 %, 2.80 %) | | | Signalized, 3-Leg (134) | 62.69 % | 21.99 % | 12.66 % | 2.31 % | 0.35 % | (2.03 %, 3.33 %) | | | 2-Way Stop, 4-Leg (916) | 63.62 % | 21.57 % | 12.25 % | 2.22 % | 0.34 % | (2.12 %, 3.00 %) | | | Minor Leg Stop, 3-Leg (2459) | 63.05 % | 21.82 % | 12.50 % | 2.28 % | 0.35 % | (2.24 %, 3.01 %) | | | Rural | | | | | | | | | Signalized, 4-Leg (18) | 65.30 % | 20.81 % | 11.51 % | 2.06 % | 0.32 % | (0.00 %, 5.45 %) | | | 2-Way Stop, 4-Leg (1048) | 62.92 % | 21.84 % | 12.59 % | 2.30 % | 0.35 % | (1.45 %, 4.03 %) | | | Minor Leg Stop, 3-Leg (2528) | 61.28 % | 22.60 % | 13.29 % | 2.45 % | 0.38 % | (1.75 %, 4.04 %) | |
study period. Eight of the top 10 segments have a functional class of either Interstate or Other Principal Arterial. Region 3 is the only one without a segment in the top 10 while the other three regions have at least three segments in the ranking.
Table 4 contains the results of the statistical analysis conducted on the compiled intersections file. The state rank, region, traffic control device, main route, county, and intersection type information of the segment are given in the first six columns, while the crash information is shown in the following seven columns. The state ranking is determined by the EWRS for each intersection with crash data from 2016 to 2020. The top 10 intersections are not spread equally across the regions with most of the prioritized intersections occurring in Region 1. Because the EWRS is a combination of the costs associated with both total crashes and severe crashes, there are instances where there are more total crashes for segments or intersections with a higher ranking. However, the highest ranked sites consistently include the greatest number of fatal crashes (severity 5). The top-ranking intersections ranged from having six crashes to 195 total crashes in the five-year period from 2016 to 2020. All the top ranked intersections have at least one fatal crash within the study period. UDOT checked to verify that no recent project that may have improved safety conditions and no planned or upcoming projects were already being considered for these sites. All remaining roadway sites were recommended for further consideration.
The results from the segment and intersection analysis are useful to
UDOT region engineers in helping them review and determine which roadways and intersections across the state should be prioritized for safety funding. A ranking based on economic impact improves the significance of the model results. The list of both segments and intersections was provided to UDOT, and with their feedback, the reports from the screened list were generated and sent out to the regions for review.
3.2. Severity distributions
Using TOMS and the JSM, severity distributions were derived from the statewide data. Table 5 shows the predicted average crash severity by segment type from 2016-2020. Table 6 shows the predicted average crash severity by intersection type for 2016–2020. Segments and intersections were split into urban and rural (with urban roadway sites appearing first). The first column lists the roadway site type and below it, in parentheses, are the number of miles of that segment type or total number of intersections of that type that were analyzed. The next five columns present the predicted percentage of crash severity, from PDO to fatal crashes. The final column is a 95 percent uncertainty interval on the predicted percentage of severe crashes (severity 4 and 5) for that roadway site type.
One observation that was made for both the segment and intersection analysis is that the proportion of predicted rural severe crashes is
Fig. 2. Example two-page safety report.
biased low. While this is unexpected, it does not mean the model should not be used. The statistical model should have a near zero bias and is accounting for the roadway characteristics, so it is not expected to see much of a change when those characteristics change. The hierarchical nature of the statistical model attempts to link the multiple populations and tie them closer to each other (e.g., statistical borrowing of strength). Usually this is a desirable model characteristic; however, in this case it had an undesirable effect. Via a hypothesis test, the data provides strong evidence that the proportion of rural severe crashes is different (and higher) than the proportion of urban severe crashes. The model does not fully capture that effect.
4. Discussion
Instead of using two different models, one accounting for total number of crashes and another for the total number of severe crashes, the two models were combined using an EWRS to provide a more thorough evaluation of safety. The resulting TOMS output allows UDOT region traffic engineers to review and determine which areas of roadway need improvement. All segments and intersections were analyzed using this method. Those with the highest EWRS were marked as high priority locations.
After the data compilation and statistical analysis, the model results; including the top 10 segments, signalized intersections, and unsignalized intersections, from each region were provided to UDOT. UDOT screened the results to verify that no recent projects that may have improved safety conditions and/or no planned or upcoming projects were already being considered for these sites. All remaining roadway sites were recommended for further consideration. For this iteration, there were 26 segments, 20 signalized intersections, and 30 unsignalized intersections selected for a total of 76 roadway sites to evaluate.
Two-page safety reports were created using a report compiler for the roadway segments and intersections of interest in each UDOT region. The report compiler uses the results from the statistical analysis and a parameters file produced by the research team to create roadway site reports that describe the crashes that occur at the site. A list of possible countermeasures is automatically generated depending on the crash characteristics and then screened and selected by the research team. While countermeasures are suggested, they are not meant to be definitive solutions. The two-page reports contain tables of roadway and crash data that assist region traffic engineers in quickly identifying intersection characteristics, crash count and severity, crash factors, manner of collision data, and possible countermeasures as shown in Fig. 2.
UDOT implements the results of this research by screening the results, selecting sites that fall under their criteria, distributing the twopage safety reports to the regions, and using the reports to prioritize safety projects across the state using the EWRS from the research. Along with the reports, the severity distributions assist UDOT and the regions as they analyze segments and intersections and determine if the proportion of severe crashes occurring at a site is significantly higher (or lower) than other similar sites. These distributions provide an overarching metric that may be useful in identifying high crash locations for safety funding prioritization even without a two-page safety report.
5. Conclusions
The goal of Zero Fatalities is a high priority for UDOT. Efforts are made to reduce the number of fatalities, injuries, and crashes on Utah roads every year. To aid engineers in selecting the sites most in need of attention and improvements, UDOT has teamed up with BYU researchers in a series of safety-focused research projects. Building off previous safety analysis research, TOMS provides a new way for the state of Utah to identify and prioritize segment and intersection safety improvement projects. Along with improvements to the methodology and statistical analysis, TOMS uses a joint model to rank segments and intersections based on the highest EWRS. UDOT can better focus their efforts and manage their budget on projects that are of highest concern across the regions with the identifying and ranking of these segments and intersections.
The TOMS is a model built within R that allows for the compilation of roadway and crash data to conduct a statistical analysis that will identify high priority locations and help UDOT region engineers determine how best to address safety concerns. Sixteen data files are used to create the combined crash and roadway data for analysis. Twelve contain roadway data while the remaining four contain crash data. The entire process is conducted from a R markdown file in which the functions are sourced, the data are read in, the data preparation scripts are run, and the compiling script creates the two output files. Instead of using two different models (one for total crashes and one for severe crashes), the two models are combined using an EWRS. Those with the highest EWRS are flagged for further evaluation. Those locations that are determined to be of interest are then summarized in a two-page safety report and evaluated for safety funding prioritization.
UDOT has implemented the results of this research by screening the results, selecting sites that fall under their evaluation criteria, distributing the two-page safety reports to the regions, and using the reports to prioritize safety projects across the state. Along with the reports, the severity distributions were generated to assist UDOT and the regions in analyzing segments and intersections to determine if the proportion of severe crashes occurring at a site is significantly higher than other similar sites. These distributions provide an overarching metric that is also useful in identifying high crash locations.
The TOMS has significantly improved previous models to assist the state of Utah in prioritizing locations for safety funding. By using severity weighted risk scores, TOMS can be used to analyze crash count and severity concurrently with a singular statistical model. This research allows for more data to be analyzed in the future to further improve traffic safety in the state and aid UDOT in their goal of reaching Zero Fatalities.
Funding sources
This work was supported by the Utah Department of Transportation [Contract no: 21-8977].
CRediT authorship contribution statement
Grant G. Schultz: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Validation, Writing – review & editing. Tomas Barriga Aristizabal: Formal analysis, Investigation, Methodology, Visualization, Writing – original draft. Jace Ritchie: Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft. Richard L. Warr: Conceptualization, Supervision, Validation, Writing – review & editing.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors acknowledge the Utah Department of Transportation (UDOT) for funding this research, and the following individuals from UDOT on the Technical Advisory Committee for helping to guide the research: Jeff Lewis, Robert Miles, Ivana Vladisavljevic, Travis Jensen, Dallas Wall, Clancy Black, Charles Allen, and Samuel Runyan. This report may be protected under 23 USC 409.
The authors alone are responsible for the preparation and accuracy of the information, data, analysis, discussions, recommendations, and conclusions presented herein. The contents do not necessarily reflect the
views, opinions, endorsements, or policies of the Utah Department of Transportation or the U.S. Department of Transportation. The Utah Department of Transportation makes no representation or warranty of any kind, and assumes no liability therefore.
Data availability
The authors do not have permission to share data.
References
- American Association of State Highway and Transportation Officials (AASHTO), 2010. Highway Safety Manual, first ed., Washington, DC.
- Afghari, A., Haque, M., Washington, S., 2020. Applying a joint model of crash count and crash severity to identify road segments with high risk of fatal and serious injury crashes. Accid. Anal. Prevent. 144.
- Gibbons, J.D., Schultz, G.G., Saito, M., 2019. A methodology for analyzing intersection safety. In: Noyce, D.A. (Ed.), International Conference on Transportation and Development 2019: Innovation and Sustainaiblity in Smart Mobility and Smart Cities. ASCE, Reston, VA, pp. 143–153.
Schultz, G.G., Johnson, E.S., Black, C.W., Francom, D., Saito, M., 2012. Traffic and safety statewide model and GIS modeling. Publication UT-12.06, Utah Department Transportation Traffic and Safety, Research and Innovation Divisions, Salt Lake City,
Schultz, G.G., Farnsworth, J.S., Roundy, R.S., Saito, M., Reese, C.S., Briggs, T., 2013. Hot spot identification and analysis methodology. Publication UT-13.15, Utah Department Transportation Traffic and Safety, Research and Innovation Divisions, Salt Lake City, UT.
Schultz, G.G., Gibbons, J.D., Saito, M., Clegg, B.W., 2018. Intersection safety analysis methodology for Utah roadways. Publication UT-18.06, Utah Department of Transportation Traffic and Safety, Research and Innovation Divisions, Salt Lake City,
Schultz, G.G., Lunt, C.C., Pew, T., Warr, R.L., Heaton, M.J., 2020. Segment and intersection methodologies for Utah highways. Publication UT-20.04, Utah Department of Transportation Traffic and Safety, Research and Innovation Divisions, Salt Lake City, UT.
Schultz, G.G., Lunt, C.C., Pew, T., Warr, R.L., 2023. Using complementary intersection and segment analyses to identify crash hot spots. Saf. Sci. 163, 106121. ISSN 0925- 7535.
Schultz, G.G., Barriga Aristizabal, T., Ritchie, J., Warr, R.L., 2023. Using severity weighted risk scores to prioritize safety funding in Utah. Publication UT-23.15, Utah Department Transportation Traffic and Safety, Research and Innovation Divisions, Salt Lake City, UT.
Utah Geospatial Resource Center (UGRC), 2024. Utah Geospatial Resource Center. <httt ps://gis.utah.gov> Accessed May 18, 2024.
Utah Department of Transportation (UDOT), 2021. Update of Crash Costs Based on Severity. Salt Lake City, UT.
Utah Department of Transportation (UDOT), 2024. UDOT Open Data Portal. <https://da ta-uplan.opendata.arcgis.com/> Accessed May 8, 2024.
Yasmin, S., Eluru, N., 2018. A joint econometric framework for modeling crash counts by severity. Transportmetrica A: Transp. Sci. 14 (3), 230–255.