Reckless Driving Crash Risks: Analysis of Contributing Factors

Abstract

Reckless driving activities (e.g., aggressive, speeding, distracted, and impaired driving) pose serious threats to roadway safety, particularly on high-exposure and complex road networks. Understanding the factors that contribute to reckless driving is critical for developing targeted interventions and improving roadway safety. To that end, this study employs negative binomial regression models to estimate the frequency of different reckless crashes and identify key factors contributing to reckless driving crashes across various roadway types in Wisconsin, United States. Using data from over 29,000 reckless driving crashes that resulted in injuries or fatalities between 2017 and 2021, the results reveal that reckless driving crash risks are strongly influenced by exposure-related variables, such as annual average daily traffic and segment length, and other roadway features. Reckless driving crashes increase with annual average daily traffic; however, the impact of this is greatest on aggressive driving crash frequency and weakest on impaired crash frequency. Conversely, wider shoulders or the presence of a shoulder and higher posted speeds are associated with a reduced crash risk on all types of roadways; yet their effectiveness may vary depending on roadway type and reckless driving activity. These findings suggest that roadway-specific design strategies—such as improving shoulder width, refining lane configurations, and considering speed limit context—can play a critical role in mitigating reckless driving behaviors and enhancing traffic safety.

Keywords

reckless driving safety negative binomial

Introduction

In recent years, there has been a marked increase in reckless driving crashes across the United States (U.S.), including speeding and other aggressive driving behaviors. According to data from the U.S. Department of Transportation (U.S. DOT) and the NHTSA, reckless driving has contributed to a sustained rise in traffic fatalities. Speed-related fatalities increased by over 15% between 2019 and 2022, highlighting a growing public safety concern ( 1 ). These reckless driving behaviors place a significant burden on individuals, families, and society.

NHTSA defines several unsafe driving behaviors as risky driving, including drunk driving, drug-impaired driving, distracted driving, speeding, and drowsy driving ( 2 ). These behaviors represent different forms of reckless driving that significantly increase the likelihood of crashes and the injury severity of crashes. Recognizing these various types is essential for understanding how each dangerous behavior contributes to crash risk and for developing targeted prevention strategies to improve road safety.

Existing research on reckless driving has primarily focused on individual behaviors rather than comprehensive analysis. Most studies examine how speeding affects crash frequency and injury severity, consistently finding that speeding is associated with higher crash frequency ( 3 ) and greater injury severities (4 –8). Speeding-related crashes are more likely at intersections and under poor road conditions ( 7 , 9 ), while injury severity increases on dry roads and divided roadways ( 5 ). Specifically, speeding crashes have a higher association with fatal or severe injury crashes when involving young or middle-aged drivers, while older adults have a higher association with non-capacitating injuries in reckless crashes ( 10 ). Research on drowsy driving shows that at-fault drowsy drivers experience more severe injuries than non-at-fault drivers ( 6 ), and their physical condition increases crash likelihood and severity ( 11 ). Crashes because of alcohol impairment typically result in more severe injuries ( 12 ) and occur more frequently in rural areas, on weekdays, and during late evening or early morning hours ( 7 ). Aggressive driving behavior is positively associated with crash injury severity levels, especially for young drivers who are between 16 and 20 years old or driving with any age passengers ( 13 ).

Few studies comprehensively examine multiple reckless driving behaviors simultaneously. Some research defines reckless driving as including several behaviors while others rely on crash report designations ( 14 , 15 ). Generally, reckless driving increases both crash likelihood and injury severity ( 6 , 8 , 12 , 16 ), with effects amplified under less than ideal road conditions ( 5 ). McCombs et al. ( 14 ) found that signalized intersection density and midblock crossings increase fatal and serious injury crashes, while roads with more than four lanes and fewer horizontal curves show reduced crash risk. Patwary and Khattak ( 8 ) identified spatial variations across counties in reckless driving crash frequency and severity in pre- and during pandemic times: the increase in speeding, reckless, and alcohol-impaired driving behavior is positively associated with higher injury severity at the county level. Arvin and Khattak ( 17 ) examined how impairment and distraction duration influence crash risk. Their findings indicate that alcohol and drug impairment increase the likelihood of crash or near-crash involvement by 34%, and that different types of distraction are associated with varying levels of risk ( 17 ). Similarly, Islam ( 18 ) showed that the severity of distracted driving crashes differs significantly across vehicle types and roadway contexts, suggesting that the effects of reckless behaviors are not uniform but depend on the driving environment. While these studies provide valuable insights into individual reckless driving behaviors, they are largely limited to single behavior analyses, specific crash severity outcomes, and few have simultaneously examined multiple reckless driving crash types across different roadway facility types. These gaps in literature limit our understanding of how different forms of reckless behavior contribute to crash risk and how these effects are shaped by roadway characteristics. In particular, most existing studies do not develop facility-specific models that distinguish how roadway features influence various reckless driving crash types across different contexts. Therefore, comprehensive studies that simultaneously examine multiple reckless behaviors across diverse roadway settings, while incorporating facility-specific distinctions, are needed to better capture these complex interactions and inform more effective safety interventions.

A recent Wisconsin Department of Transportation (WisDOT) funded study ( 19 ) examined these reckless driving crash risks on Wisconsin roadways and serves as the foundation for the analysis presented in this paper. The objective of this paper is to provide a comprehensive understanding of the impact of reckless driving on the safety of Wisconsin roadways. Using roadway characteristics and crash data from across the state, statistical models are developed to predict the occurrence of various types of reckless driving-related crashes on different roadway facilities. These models are then integrated into a network screening tool designed to identify locations with the highest risk of reckless driving activity.

The remainder of this paper is organized as follows. The following section summarizes the data that were available and collected as a part of this study. The next section describes the negative binomial (NB) methodology that was used. Then a summary of the crash risk model estimation results is given. Finally, concluding remarks are provided.

Data

Reckless driving crash information was obtained through the WisTransPortal maintained by the Wisconsin Traffic Operations and Safety Laboratory (TOPS Lab) ( 20 ). This database contains information on crashes on Wisconsin state roads, including the location of each crash, vehicles involved, and general crash attributes. We identified four categories of reckless driving crashes that occurred between 2017 and 2021, as follows.

Aggressive driving

Speeding

Distracted driving

Impaired driving

Table 1 summarizes the variables used for identifying reckless driving behavior in Wisconsin for this analysis.

Table 1.

Variables Indicating Reckless Driving Behavior from the Wisconsin DT4000 Crash Report

Reckless driving behavior category	Variables indicating reckless driving behavior	Description
Aggressive	TRUE if for any person, DRVRPC[1,2][A,B,C,D]: has one or more occurrences from Tier 1, two or more occurrences from Tier 2, or three or more occurrences from Tiers 2 or 3, where Tier 1: • AR—Operated motor vehicle in aggressive/reckless manner • RAC—Racing Tier 2: • SPD—Exceed speed limit • TFC—Speed too fast • FTC—Following too close • IOR—Improper overtaking/passing right • IOL—Improper overtaking/passing left • ID—Operated motor vehicle in inattentive, careless, or erratic manner Tier 3: • FTY—Failed to yield right of way • FVC—Failure to control • DRED—Disregarded red light • DSS—Disregarded stop sign • DTC—Disregarded other traffic control • DRM—Disregarded other road markings	The actions by the driver that may have contributed to the crash, based on the judgment of the law enforcement officer investigating the crash
Speeding	SPEEDFLAG	Flag indicating whether speed was a factor in a crash
Distracted	DISTFLAG	Flag indicating whether a crash involved distracted or inattentive driving
Distracted	DNMFTR[1,2][A,B]: SLEEP—Asleep or Fatigued	Any relevant condition of the individual (motorist or non-motorist) that is directly related to the crash
Impaired	DRUGLFAG	Indicates whether law enforcement suspected that at least one driver or non-motorist involved in the crash had used drugs
	ALCFLAG	Indicates whether law enforcement suspected that at least one driver or non-motorist involved in the crash had used alcohol. This includes both alcohol use under the legal limit and at or over the legal limit
	DNMFTR[1,2][A,B]: • UI MDA—Under the influence of medication/drugs/alcohol • PHY IMP—Physically impaired • SICK—Ill (sick), fainted CONF—Confused or disoriented (non-lucid)	Any relevant condition of the individual (motorist or non-motorist) that is directly related to the crash

Each crash had a unique identifier and contained information on the location of the crash, injury severity level, and other general attributes. Five unique severity levels were present in the data, as follows.

K—Fatal injury

A—Incapacitating injury

B—Non-incapacitating injury

C—Possible injury

O—No apparent injury

A unique crash identifier was used to link qualifying reckless driving crashes from 2017 to 2021 to quantify the magnitude of reckless driving occurrences along the Wisconsin State Trunk Highway Network (STHN). Table 2 summarizes the statistics of different reckless driving-related crashes by injury level included in the study. The combination of all injury crashes into a single crash frequency measure was necessitated by sample size constraints, as disaggregating crashes by severity across five roadway types and four reckless driving categories resulted in insufficient observations in several subgroups for statistically reliable estimation. To generate more reliable reckless driving crash risk model results, reckless driving crash risk models for each reckless driving category at the KABC level were developed to represent all crashes that resulted in an apparent injury or fatality.

Table 2.

Summary of Reckless Driving Crashes by Injury Level

Type	K	A	B	C	O	KABC
Speeding	322	1,483	4,879	4,197	25,358	10,881
Distracted	161	915	4,084	4,238	17,083	9,398
Impaired	384	1,144	2,169	1,265	5,751	4,962
Aggressive	172	584	1,558	1,480	5,956	3,794
Total	1,039	4,126	12,690	11,180	54,148	29,035

Note: Injury levels: K = fatal injury; A = incapacitating injury; B = non-incapacitating injury; C = possible injury; O = no apparent injury.

Roadway characteristics information for segments in the Wisconsin STHN was provided directly by WisDOT. The Wisconsin STHN database contained a wide variety of roadway characteristics that can be linked to the roadway segment base file with unique roadway segment IDs. Specific data elements associated with each roadway segment in Wisconsin were as follows.

Number of lanes

Travel lane width

Shoulder width and presence

Posted speed

Horizontal curvature status

Segment length

Roadway separation type

Given the missing urban and rural code definitions from the Wisconsin STHN database, the existing Highway Capacity Manual (21) facility type (HCMTYPE) variable was used to classify roadway segments. The following categories were available for this variable.

FRE: Basic freeway segments.

MLT: Multilane highway segments that have four or six lanes, and posted speed limits >40 mph and signal spacing >2 mi apart.

URB: Highway segments that have an urban cross-section, or segments that have signal spacing of less than 2 mi apart and are within city or village limits.

TWO: Two-lane highway segments that have a two-lane undivided rural cross-section.

Urban highway segments were further categorized as divided or undivided, based on roadway separation types for analysis purposes and based on differences in safety performance of reckless driving crashes that were observed. Roadway types classified as MLT and TWO were considered to represent rural roadway types.

Average annual daily traffic (AADT) data from 2017 to 2021 was obtained directly from WisDOT. Since the AADT volumes are provided for both directions, the directional or adjusted AADT was obtained for divided road segments by dividing the reported AADT by two. This adjusted AADT value is used for modeling reckless driving crash risk. Table 3 summarizes the data by roadway categories.

Table 3.

Summary of Risk Data by Roadway Categories

Roadway category	Total segments (#)	Relative frequency by segment (%)	Total mileage (mi)	Relative frequency by mileage (%)	Total crashes (#)*	Relative frequency by segment type (%)
Basic freeway	3,071	15.4	2,557	17.8	26,564	31.1
Multilane highway	2,476	12.4	1,874	13.1	7,873	9.2
Two-lane highway	10,404	52.0	8,435	58.8	22,595	26.4
Undivided urban highway	1,990	9.9	574	4.0	10,447	12.2
Divided urban highway	3,065	15.3	901	6.3	18,060	21.1
Total	20,006	100.0	14,341	100.0	85,539	100.0

* The total crash numbers in this table reflect all crashes that occurred on all highway segments within the Wisconsin State Trunk Highway Network, including one-way two-lane roadway segments and roadway segments with incomplete data.

Lastly, two new variables were generated using existing roadway characteristic information for analysis purposes, as follows.

Average shoulder width: obtained by summing the right shoulder width with the left shoulder width divided by two.

Average lane width: obtained by dividing the total traveled way width by the number of lanes.

No other pre-analysis processing or manipulation of the crash or roadway data was performed; all variables were used as provided by WisDOT and the WisTransPortal database.

Methodology

NB regression models were employed to estimate reckless driving crash risks. NB regression is a count-based modeling technique appropriate for dependent variables that take on non-negative integer values ( 22 ); it is widely used in crash data analysis and is consistent with the models developed in the Highway Safety Manual ( 21 ) for safety management decision-making ( 23 ). NB regression is particularly well suited for this application because it accounts for overdispersion, a common characteristic of crash datasets in which the variance exceeds the mean ( 24 , 25 ). All statistical models were estimated using R statistical software (version 4.5.1) with the MASS package for NB regression models.

The general form of the crash frequency model estimated for roadway segments is presented in Equation (1). These models were used to quantify the influence of traffic and various roadway characteristics on the risk of crashes within each of the four identified reckless driving categories:

\begin{matrix} N_{i, risk}^{k} = \ln (AADT)^{β_{AADT}} \times \ln (L)^{β_{Length}} \times e^{β_{0}} \times e^{\sum x_{ij} β_{j}} \\ = \ln (AADT)^{β_{AADT}} \times \ln (L)^{β_{Length}} \times e^{β_{0}} \times e^{x_{i 1} β_{1}} \\ \times e^{x_{i 2} β_{2}} \times \dots \times e^{x_{iJ} β_{J}} \end{matrix}

(1)

where

$N_{i, risk}^{k}$ : Predicted frequency of a reckless driving crash type $k$ on segment i (crashes/year)

$\ln (AADT)$ : Natural log of annual average daily traffic on segment i (vehicles per day)

$β_{AADT}$ : Estimated coefficient for traffic volume

$β_{Length}$ : Estimated coefficient for segment length

$\ln (L)$ : Natural log of length of segment (mi)

$β_{0}$ : Regression intercept

$x_{ij}, β_{j}$ : Explanatory variables and corresponding coefficients related to roadway design and traffic characteristics

The models incorporated variables commonly associated with reckless driving behaviors, including posted speed, lane width, shoulder width, and other roadway characteristics. Segment length (L) was included as an estimated parameter rather than a proportional constant because preliminary analysis indicated that crash frequency was not strictly proportional to segment length, a finding consistent with unobserved factors that may correlate with segment characteristics. This approach improves model fit and prediction accuracy.

The influence of independent variables on reckless driving crash frequency was interpreted using elasticities. These values represent the responsiveness of the predicted crash frequency to a marginal change in an explanatory variable. For continuous variables the elasticity is defined as:

E_{X_{ijk}}^{λ_{ij}} = \frac{\partial λ_{ij} / λ_{ij}}{\partial X_{ijk} / X_{ijk}} = \frac{\partial λ_{ij}}{\partial X_{ijk}} \times \frac{X_{ijk}}{λ_{ij}}

(2)

Depending on how the variable is modeled (log-log or log-linear), elasticity simplifies as follows:

Log-log form (e.g., AADT):

E_{X_{ijk}}^{λ_{ij}} = β_{k}

(3)

Log-linear form (e.g., average shoulder width):

E_{X_{ijk}}^{λ_{ij}} = β_{k} X_{ijk}

(4)

For indicator (binary) variables (such as the posted speed limit being greater than some threshold value), pseudo-elasticity was used to estimate the percentage change in crash frequency when the variable switches from 0 to 1:

E_{X_{ijk}}^{λ_{ij}} = \exp (β_{k}) - 1

(5)

These elasticities allow for meaningful interpretation of the risk associated with individual roadway or environmental features in relation to different types of reckless driving crashes.

Results

This section presents the estimated models for reckless driving obtained at the KABC level for each roadway type.

Basic Freeways

Table 4 provides a summary of reckless crash frequency models developed for roadway segments categorized as basic freeways. The table gives the coefficient estimates, their significance levels, the overdispersion parameter, the log-likelihood value, and the elasticity values for the models. Cumulative residual (CURE) plots were used to assess the model fit to the observed data as described in Hauer ( 26 ). CURE plots graphically represent the cumulative residuals (or the differences between predicted and observed values) against any variable of interest. In this case, different types of predicted reckless crash frequency are used in the CURE plots as the independent variables to describe the cumulative residuals. The plots therefore provide insights into how reasonable the selected functional forms are through the visual depiction of the goodness of fit. Generally, a good CURE plot is one in which the cumulative residuals oscillate around zero and do not exceed the two standard deviation limits ( 27 ). The CURE plots generated for all 20 estimated models in this study mostly confirmed that the models fit the observed data well without significant overfitting because cumulative residuals generally stayed within the 95% confidence interval developed assuming a random walk process. An example of aggressive driving crashes on different roadways is provided in Figure 1 for illustrative purposes. In practical terms, these results suggest that freeway segments with heavier traffic volumes and more lanes experience more reckless driving crashes, as expected because of increased vehicle exposure and interactions. However, design features matter significantly: wider shoulders provide drivers with more space to recover from errors, reducing crash risk by approximately 43% for aggressive driving when shoulder width increases by 1% from the median value of 7 ft. Similarly, higher speed limits (65 mph or above) are associated with approximately 28% fewer aggressive crashes, likely because these roads are built to higher safety standards with features like wider lanes, gentler curves, and better sight distances that collectively reduce crash opportunities.

Table 4.

Summary of Reckless Driving Crash Frequency Models Developed for Basic Highways

Variable	Aggressive crash frequency		Speeding crash frequency		Distracted crash frequency		Impaired crash frequency
Variable	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity
Constant	−11.8381 (***)	NA	−6.9485 (***)	NA	−9.6469 (***)	NA	−5.9711 (***)	NA
Natural log of adjusted average AADT	1.1690 (***)	1.1690	0.8166 (***)	0.8166	1.0170 (***)	1.0170	0.5806 (***)	0.5806
Natural log of segment length (mi)	0.8682 (***)	0.8682	0.8292 (***)	0.8292	0.8529 (***)	0.8529	0.7964 (***)	0.7964
Average shoulder width	−0.0620 (***)	−0.4339 (7)	NA	NA	−0.0430 (***)	−0.3012 (7)	−0.0069	−0.0485 (7)
Average shoulder width greater than or equal to 4 ft	NA	NA	−0.1647 (**)	−0.1518	NA	NA	NA	NA
Posted speed 65 mph or above	−0.3331 (**)	−0.2833	−0.7516 (***)	−0.5284	−0.5407 (***)	−0.4213	−0.8776 (***)	−0.5842
Number of lanes three or above	0.4189 (***)	0.5203	0.4659 (***)	0.5934	NA	NA	0.4423 (***)	0.5562
Inverse of overdispersion parameter	1.9980		2.1860		2.6880		2.1790
2xlog-likelihood value	−3,809.531		−7,835.036		−5,120.047		−3,840.989
Pseudo R²	0.1552		0.1568		0.1344		0.1212
AIC	3,823.53		7,849.04		5,132.05		3,854.99
BIC	3,865.25		7,890.76		5,167.81		3,896.71

Note: All models are based on observations from each roadway category. In the coefficient columns, values in parentheses indicate statistical significance levels, as marked by symbols: * for p < .05, ** for p < .01, and *** for p < .001. In the elasticity columns, values in parentheses indicate the median of a continuous variable. AADT = average annual daily traffic; R² = coefficient of determination; AIC = Akaike information criterion; BIC = Bayesian information criterion; NA = not available.

Figure 1.

CURE plots for aggressive crash frequency on basic freeway segment, multilane highways, two-lane roads, urban divided highways, and urban undivided highways.

The coefficient estimate for a given variable shown in Table 4 for each model provides the relationship between that variable and the associated reckless driving crash frequency: positive values represent factors associated with increased crash risk, while negative values represent factors associated with decreased crash risk in each reckless driving category. These coefficient estimates generally align with expectations. Crash frequencies are mostly expected to increase with vehicle exposure, and both traffic volume (number of vehicles that travel on the segment) and segment length (the amount of travel on the segment) increase exposure. The number of lanes is associated with increased reckless driving crashes, which seems reasonable because more lanes typically mean more interactions with other vehicles and occurrences of reckless driving behaviors. In contrast, shoulder width is usually negatively associated with most reckless driving behaviors, probably because wider shoulders offer more space for vehicles to recover after deviating from the travel lane. Roadway segments with higher posted speeds are associated with fewer reckless driving crashes. Although this may seem counterintuitive, such roads are usually designed with more conservative safety features. Moreover, higher travel speeds can help limit opportunities for aggressive or abrupt driving maneuvers.

The elasticity values quantify the amount of “risk” associated with each risk factor included in the model. Specifically, each value represents the relevant increase in crash frequency associated with a change in a given variable, referred to hereafter as “crash risk”. Positive values represent an increase in crash risk associated with an increase in that variable (e.g., positive correlation), whereas negative values represent a decline in crash risk associated with an increase in that variable (e.g., negative correlation). Continuous variables that are not expressed in log form are evaluated at the median values for these variables. While elasticity values may vary at different points along these variables’ distributions, the reported estimates offer a useful indication of the strength of each variable’s relationship with reckless driving crash risk. Despite being continuous variables, AADT and segment length are log-transformed in the model, so their elasticity values are constant and equal to their respective model coefficients. For binary (indicator) variables, the elasticity represents the expected change in crash frequency when the variable shifts from 0 to 1. The elasticities can be interpreted as follows. A 1% change in AADT (log variable) along a basic freeway segment is associated with a 1.17% increase in aggressive crash frequency along that segment. For the average shoulder width (continuous variable), the elasticity is provided at the median value observed in the data: a 1% change in average shoulder width (for the “average” roadway segment with average shoulder width of 7 ft) would be associated with a 0.4399% decrease in aggressive crash frequency along that segment. Lastly, the presence of three or more travel lanes (indicator variable) is associated with a 52% increase in aggressive crash frequency. Other variables can be interpreted in a similar manner.

To put these numbers in perspective, on basic freeways a 10% increase in traffic volume is associated with an 11.7% increase in aggressive crashes, but only a 5.8% increase in impaired driving crashes. This suggests that congestion and vehicle interactions play a larger role in triggering aggressive behaviors than impaired driving. Wider shoulder widths are particularly effective in reducing aggressive driving crashes: a segment with 8-ft shoulders instead of 6-ft shoulders would be expected to have about 13% fewer aggressive driving crashes, providing clear guidance for roadway designers on the safety benefits of wider shoulders. While higher posted speeds are associated with decreases in all reckless driving activities, the decrease is less for aggressive driving relative to other crash types, perhaps indicating that speeding and impaired driving are less prominent on high-speed freeways relative to distracted and aggressive driving behaviors. The especially large reduction in speeding-related crashes on high-speed roads is in line with expectations, as the likelihood of a vehicle exceeding the speed limit decreases on higher-speed roads.

Multilane Highways

Table 5 provides a summary of reckless crash frequency models developed for roadway segments categorized as multilane highways. As shown, factors associated with increased risk include the following.

AADT

Roadway segment length

Roadway segments with greater lane widths (e.g., significant for aggressive and distracted crash frequencies when greater than or equal to 12.5 ft)

Factors associated with reduced risk include the following.

Average shoulder width (distracted crash frequency only)

Roadway segments with higher posted speeds (not significant for speeding crash frequency)

These relationships between roadway characteristics and reckless driving crash frequencies on multilane highways are generally similar to these on basic freeways. Besides crash frequencies’ positive relationship with exposure (AADT and segment length), greater lane widths are also positively associated with aggressive and distracted crash frequencies. Wider lanes could foster a false sense of safety, prompting drivers to take greater risks, like frequent and aggressive lane changes and weaving or careless maneuvers, and believing they have extra space to avoid crashes. Similar to reckless driving crashes on basic highways, higher posted speeds are negatively related to most reckless driving crash frequencies on multilane highways. However, average shoulder width is only significantly associated with reduced risks in distracted driving crash frequency.

Table 5.

Summary of Reckless Driving Crash Frequency Models Developed for Multilane Highways

Variable	Aggressive crash frequency		Speeding crash frequency		Distracted crash frequency		Impaired crash frequency
Variable	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity
Constant	−13.544 (***)	NA	−11.332 (***)	NA	−12.1958 (***)	NA	−10.0193 (***)	NA
Natural log of adjusted average AADT	1.3509 (***)	1.3509	1.2161 (***)	1.2161	1.3435 (***)	1.3435	0.9950 (***)	0.9950
Natural log of segment length (mi)	0.7158 (***)	0.7158	0.7007 (***)	0.7007	0.5849 (***)	0.5849	0.5712 (***)	0.5712
Average lane width greater than or equal to 12.5 ft	0.6000 (**)	0.8222	NA	NA	0.4009 (**)	0.4932	NA	NA
Average lane width greater than or equal to 13 ft	NA	NA	0.2345	0.2642	NA	NA	NA	NA
Average shoulder width	NA	NA	NA	NA	−0.0532 (*)	−0.1597 (3)	NA	NA
Posted speed 65 mph or above	−0.4623 (***)	−0.3702	−0.0569	−0.0553	−0.6774 (***)	−0.4921	−0.5688 (***)	−0.4338
Inverse of overdispersion parameter	1.009		1.671		1.178		1.751
2xlog-likelihood value	−1,802.631		−3,783.925		−3,458.821		−2,224.664
Pseudo R²	0.0829		0.0793		0.0913		0.0603
AIC	1,814.63		3,795.93		3,472.82		2,234.66
BIC	1,849.21		3,830.5		3,513.16		2,263.48

As with freeway segments, aggressive driving crashes increase more rapidly with traffic volumes than other crash types. However, higher posted speeds appear to be associated with a larger reduction in aggressive, distracted, and impaired driving crashes relative to speeding-related crashes. From a safety planning perspective, these findings suggest that multilane highways may benefit from targeted countermeasures. Lane widths of 12.5 ft or greater are associated with approximately 82% more aggressive crashes and 64% more distracted crashes, suggesting that while wider lanes provide operational benefits, they may inadvertently encourage riskier driving by creating a perception of extra maneuvering space. In contrast, posted speeds of 65 mph or above reduce aggressive crashes by about 37%, indicating that higher-speed multilane facilities with appropriate geometric design provide operating environments that inherently reduce driving-related crashes.

Urban Divided Highways

Table 6 provides a summary of reckless crash frequency models developed for roadway segments categorized as urban divided highways. As shown, factors associated with increased risk include the following.

AADT

Roadway segment length

Factors associated with reduced risk include the following.

Roadway with shoulders (aggressive and speeding crash frequencies only)

Roadway segments with higher posted speeds (not significant for speeding crash frequency)

Crash frequencies also have positive relationships with exposure (AADT and segment length) for all types of reckless driving crashes on urban divided highways. Roadways with shoulders are associated with reduced risks for aggressive and speeding crash frequencies. Though higher posted speeds are generally negatively associated with reckless driving crash risks, their effect in reducing crash risks becomes more pronounced as the posted speed increases. Higher posted speeds could help reduce reckless driving behaviors by encouraging more uniform traffic flow while supporting high traffic volumes on urban divided highways.

Table 6.

Summary of Reckless Driving Crash Frequency Models Developed for Urban Divided Highways

Variable	Aggressive crash frequency		Speeding crash frequency		Distracted crash frequency		Impaired crash frequency
Variable	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity
Constant	−14.7673 (***)	NA	−12.1074 (***)	NA	−12.499 (***)	NA	−9.2732 (***)	NA
Natural log of adjusted average AADT	1.6253 (***)	1.6253	1.3920 (***)	1.3920	1.4774 (***)	1.4774	0.9971 (***)	0.9971
Natural log of segment length (mi)	0.7327 (***)	0.7327	0.6934 (***)	0.6934	0.6631 (***)	0.6631	0.6514 (***)	0.6514
Average shoulder width is not zero	−0.3307 (**)	−0.2816	−0.2713 (***)	−0.2376	NA	NA	NA	NA
Posted speed 25 mph or below	NA	NA	−0.3710 (**)	−0.3100	NA	NA	NA	NA
Posted speed 30 mph or above	NA	NA	NA	NA	−0.3523 (***)	−0.2969	NA	NA
Posted speed 40 mph or above	−0.4641 (***)	−0.3713	NA	NA	NA	NA	NA	NA
Posted speed 45 mph or above	NA	NA	−0.4135 (***)	−0.3386	NA	NA	−0.4984 (***)	−0.3925
Inverse of overdispersion parameter	1.480		1.684		1.193		2.102
2xlog-likelihood value	−3,547.454		−5,143.050		−6,007.791		−3,612.821
Pseudo R²	0.1457		0.1461		0.1471		0.0989
AIC	3,559.45		5,157.05		6,017.79		3,622.82
BIC	3,595.17		5,198.71		6,047.55		3,652.58

Note: All models are based on observations from each roadway category. In the coefficient columns, values in parentheses indicate statistical significance levels, as marked by symbols: ** for p < .01, and *** for p < .001. In the elasticity columns, values in parentheses indicate the median of a continuous variable. AADT = average annual daily traffic; R² = coefficient of determination; AIC = Akaike information criterion; BIC = Bayesian information criterion; NA = not available.

As with the models in the previous sections, traffic volumes have a greater impact on aggressive driving crashes than on other crash types. Higher speed limits consistently reduce crashes across multiple types—aggressive, speeding, and impaired. However, speeding crashes decrease with lower posted speeds, indicating that drivers may be more compliant in low-speed zones. Distracted crashes are uniquely sensitive to moderate–high speeds, suggesting that distraction may be less tolerated in faster environments. For urban divided highways, the presence of shoulders emerges as a critical safety feature, reducing aggressive driving crashes by approximately 28% and speeding-related crashes by about 24%. This finding has direct implications for urban highway retrofit projects, where adding or widening shoulders could significantly improve safety. The graduated effect of posted speeds (with larger reductions at higher speed thresholds) suggests that speed limit policies should be coordinated with geometric improvements to maximize safety benefits.

Urban Undivided Highways

Table 7 provides a summary of reckless crash frequency models developed for roadway segments categorized as urban undivided highways. As shown, factors associated with increased risk include the following.

AADT

Roadway segment length

Factors associated with reduced risk include the following.

Roadway with shoulders (not significant for speeding crash frequency)

Roadway segments with higher posted speeds (not significant for aggressive crash frequency)

Crash frequencies also have positive relationships with exposure (AADT and segment length) for all types of reckless driving crashes on urban undivided highways. Roadways with shoulders are associated with reduced risks for most reckless driving crash frequencies except for speeding crashes. Higher posted speeds are generally negatively associated with reckless driving crash risks, except for aggressive crashes. The complexity of undivided urban highways means that the impact of posted speed on reckless driving crash frequency is smaller than on divided highways.

Table 7.

Summary of Reckless Driving Crash Frequency Models Developed for Urban Undivided Highways

Variable	Aggressive crash frequency		Speeding crash frequency		Distracted crash frequency		Impaired crash frequency
Variable	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity
Constant	−13.9500 (***)	NA	−11.0937 (***)	NA	−11.9038 (***)	NA	−10.1008 (***)	NA
Natural log of adjusted average AADT	1.5025 (***)	1.5025	1.2703 (***)	1.2703	1.4288 (***)	1.4288	1.1472 (***)	1.1472
Natural log of segment length (mi)	0.7201 (***)	0.7201	0.8647 (***)	0.8647	0.9118 (***)	0.9118	0.8581 (***)	0.8581
Average shoulder width is not zero	−0.5759 (**)	−0.4378	−0.2032	−0.1839	−0.2971 (**)	−0.2570	−0.5134 (**)	−0.4015
Posted speed 35 mph or above	−0.1418	−0.1322	−0.2944 (*)	−0.2550	−0.3086 (***)	−0.2655	−0.326 (*)	−0.2782
Inverse of overdispersion parameter	0.952		1.085		2.443		1.093
2xlog-likelihood value	−1,802.207		−2,586.673		−3,393.049		−2,324.914
Pseudo R²	0.1296		0.1223		0.1696		0.1168
AIC	1,814.21		2,598.67		3,405.05		2,336.91
BIC	1,846.98		2,631.45		3,437.83		2,369.69

Similar to urban divided highways, on urban undivided highways, the impact of traffic volumes on aggressive driving crashes is greater than on other crash types. The presence of shoulders has a greater impact on reducing aggressive and impaired crashes. This may be because shoulders provide additional recovery space for impaired drivers or allow aggressive maneuvers to correct errors or avoid conflicts, thereby mitigating the severity and frequency of these crash types. Higher posted speeds do not have a statistically significant impact on aggressive crashes on urban undivided highways. This may be because aggressive crashes are influenced more by driver behavior and congestion, making them less sensitive to posted speeds. Urban undivided highways present unique safety challenges because of their constrained geometry and mixed traffic. The strong protective effect of presence of shoulders (reducing aggressive driving crashes by 44% and impaired crashes by 40%) highlights the value of even modest shoulder additions on these constrained facilities. However, the limited impact of posted speeds on aggressive driving crashes suggests that geometric complexity and driver interactions, rather than speed alone, drive aggressive behavior patterns on these roadways.

Two-Lane Highways

Table 8 provides a summary of reckless crash frequency models developed for roadway segments categorized as two-lane highways. As shown, factors associated with increased risk include the following.

AADT

Roadway segment length

Roadway is undivided (not significant for speeding crash frequency)

Factors associated with reduced risk include the following.

Roadway with greater shoulder widths (e.g., significant for aggressive crash frequency when greater than or equal to 3 ft)

Roadway with lane widths greater than or equal to 11.5 ft

No curves on roadway with posted speed of 40 mph or below (only for speeding crash frequency)

Roadway segments with higher posted speeds

Crash frequencies also have positive relationships with exposure (AADT and segment length) for all types of reckless driving crashes on two-lane highways. Additionally, undivided roadways are positively related to reckless driving crash risks. The combination of limited space, opposing traffic flow, and unpredictable vehicle movements on undivided two-lane highways creates conditions where reckless behaviors like speeding, tailgating, and unsafe passing are more likely to result in crashes. Roadways with shoulders and the negative associations of higher posted speeds with reckless crash frequencies are only significant for aggressive crashes. It could be that greater shoulders offer more space for aggressive maneuvers, while roads with higher speeds are designed with more conservative measures. Wider lanes could provide more space for error correction; therefore, they can help reduce certain reckless driving crash risks. Moreover, roads without horizontal curves can limit opportunities for high-speed maneuvers to reduce speeding crash risks.

Table 8.

Summary of Reckless Driving Crash Frequency Models Developed for Two-Lane Highways

Variable	Aggressive crash frequency		Speeding crash frequency		Distracted crash frequency		Impaired crash frequency
Variable	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity	Coefficient	Elasticity
Constant	−9.7572 (***)	NA	−6.1055 (***)	NA	−9.6766 (***)	NA	−6.3627 (***)	NA
Natural log of adjusted average AADT	0.9302 (***)	0.9302	0.7519 (***)	0.7519	1.0318 (***)	1.0318	0.6236 (***)	0.6236
Natural log of segment length (mi)	0.5433 (***)	0.5433	0.7843 (***)	0.7843	0.5284 (***)	0.5284	0.7523 (***)	0.7523
Roadway is undivided	0.6426 (**)	0.9013	0.2126	0.2369	0.5003 (***)	0.6492	0.6363 (***)	0.8894
Posted speed above 45 mph	−0.2702 (*)	−0.2368	NA	NA	NA	NA	NA	NA
Posted speed 55 mph or above	NA	NA	−0.1649 (*)	−0.1520	−0.1968 (**)	−0.1787	−0.2597 (**)	−0.2287
Average shoulder width is greater than or equal to 3 ft	−0.2670 (*)	−0.2344	NA	NA	NA	NA	NA	NA
Average shoulder width is greater than or equal to 5 ft	NA	NA	−0.1202	−0.1132	NA	NA	NA	NA
Average lane width is greater than or equal to 11.5 ft	NA	NA	−0.4336 (***)	−0.3518	NA	NA	−0.4611 (***)	−0.3694
No curves on roadway with posted speed 40 mph or below	NA	NA	−0.5431 (***)	−0.4190	NA	NA	NA	NA
Inverse of overdispersion parameter	1.463		2.400		2.596		2.432
2xlog-likelihood value	−6,332.844		−13,439.428		−12,820.624		−10,183.795
Pseudo R²	0.0762		0.0879		0.1107		0.0711
AIC	6,346.84		13,457.43		12,812.20		10,197.80
BIC	6,397.47		13,522.51		12,855.59		10,248.42

Traffic volumes are associated with greater increases in distracted crashes. Furthermore, undivided roads are more likely to have more aggressive, distracted, and impaired crashes. This may be because of increased conflict points that create more complex driving situations, increasing risky behaviors and the likelihood of crashes, especially when drivers are distracted or impaired. Wider shoulders are associated with a greater reduction in aggressive crashes, while wider lanes provide speeding and impaired drivers with more space to maneuver safely. Additionally, the absence of curves on roads with relatively lower speeds may reduce the likelihood of speeding crashes because of the lower risks of speeding. Two-lane highways, particularly undivided configurations, show the highest sensitivity to geometric features. Undivided roads experience approximately 90% more aggressive driving crashes and 89% more impaired driving crashes compared with divided configurations, underscoring the safety benefits of median separation where feasible. The finding that wider lanes (11.5 ft or greater) reduce speeding-related and impaired driving crashes by about 35%–37% provides clear design guidance: on constrained two-lane highways where widening is considered, the safety benefits may justify the investment, particularly on high-crash corridors.

The results presented above are generally consistent with findings from previous safety studies on reckless driving. The positive relationship between traffic volumes and crash frequency aligns with established literature ( 3 , 14 ), confirming that exposure remains a dominant predictor of crash risk across all reckless driving categories and facility types. The protective effect of wider shoulders observed in our models is consistent with existing studies that narrower shoulder widths are associated with increased crash risks in distracted driving crashes ( 5 ). The positive association between the number of lanes and reckless driving crash frequency observed in this study is also consistent with existing findings ( 14 ). Moreover, the positive association between undivided two-lane highway segments and certain reckless driving crash risks is consistent with the findings of existing studies ( 4 , 14 ), suggesting that the absence of physical median separation (undivided configuration) could increase the frequency of reckless driving crashes. However, the present study finds that a higher posted speed is consistently associated with fewer reckless driving crashes across all roadway types and in most reckless driving crash categories, a finding that contrasts with the broader literature where higher speed limits were associated with higher reckless crash frequency ( 3 , 4 , 8 ). The only partial alignment with the present study’s finding comes from Patwary and Khattak ( 8 ), who acknowledged finding a negative association between speed limits and crash frequency in certain Tennessee, U.S., counties in their study, attributing it to the increased design standards of higher-speed highways and lower traffic volumes. The observed negative relationship between higher posted speeds and reckless crash frequency in this study probably also reflects the influence of roadway design, as higher-speed facilities are typically engineered with more robust safety features, such as wider lanes, gentler curves, improved sight distances, as well as controlled access, which together could help mitigate crash risk despite increased speeds.

The present study contributes to the literature by modeling different types of reckless driving crash frequency across five distinct roadway types, providing specific evidence on how geometric factors shape reckless crash risks. This study also provides new insights into the relationship between posted speeds and reckless crash frequency, revealing a pattern that contrasts with prevailing findings in the literature and suggests an inverse or more nuanced association than previously examined and reported.

Concluding Remarks

This study used NB regression models to estimate crash frequencies and identify the significant factors associated with reckless driving in Wisconsin, U.S., roadways. Reckless driving crash frequencies vary considerably across different roadway types, influenced by both exposure-related factors and specific roadway design features. The results indicate the following.

Basic freeways: Higher traffic volumes, longer segments, and the presence of three or more lanes increase reckless crash risk. In contrast, wider shoulders and higher posted speeds are associated with reduced crash frequencies. These effects are consistent across aggressive, distracted, speeding, and impaired crashes, with wider shoulders particularly effective in reducing speeding-related crashes.

Multilane highways: Exposure variables are positively associated with crash frequency. Additionally, wider lanes are linked to increased aggressive and distracted crash frequencies, probably because of a false sense of safety that encourages riskier maneuvers. Shoulder width has a protective effect but only for distracted driving crashes.

Urban divided highways: Exposure plays a key role in increased crash risk. Shoulders are associated with fewer aggressive and speeding crashes, while higher posted speeds tend to reduce crash frequency, particularly when speed limits are higher. These roadways may benefit from better design standards that support safer driving behaviors.

Urban undivided highways: Crash risk increases with exposure, and although shoulders help reduce risk for most reckless behaviors, they are not significant for speeding crashes. Higher posted speeds generally reduce reckless crash risk, but the effect is less pronounced because of the complexity and limited maneuvering space on these roadways.

Two-lane highways: Exposure and the lack of a median (undivided configuration) are major contributors to reckless driving crash risk. Shoulder width and lane width help reduce crash frequency, particularly for aggressive driving. Roads with no curves and low posted speeds are associated with reduced speeding crashes, probably because of fewer opportunities for acceleration.

Across all five roadway types, exposure variables (AADT and segment length) were consistently significant predictors of reckless driving crash frequency across all facility types and crash categories, confirming exposure as a universal risk factor. Higher posted speeds were negatively associated with crash frequency on nearly all facility types, likely reflecting conservative geometric design standards on higher-speed facilities rather than a protective effect of high posted speeds. Nevertheless, several factors exhibited facility-specific effects: lane width was only significant on multilane and two-lane highways; shoulder presence was particularly significant on urban highways (both divided and undivided); and the number of lanes was significant only on basic freeways. Among crash categories, aggressive driving crash frequency showed the strongest sensitivity to AADT across all facility types, while impaired driving crashes were comparatively less responsive to traffic volume but more sensitive to roadway separation (divided versus undivided) on two-lane highways.

These findings extend existing literature ( 3 , 7 , 8 , 14 ) by quantifying how roadway characteristics differentially affect reckless driving-related crashes. Our results suggest several targeted interventions: prioritize shoulder improvements on freeways and urban highways (28%–44% crash reduction); reconsider lane width standards on multilane facilities where wider lanes may encourage risky driving; install medians on two-lane highways where undivided configurations show 90% higher aggressive driving crash rates; and maintain enhanced geometric design on higher-speed facilities. The counterintuitive negative relationship between higher posted speeds and crash frequency probably reflects that higher-speed roadways are designed with more conservative safety features (e.g., wider lanes, gentler curves, better sight distances, and controlled access) that collectively reduce crash risk despite higher speeds. Furthermore, these facilities typically have fewer conflict points and more homogeneous traffic flows, limiting opportunities for reckless behaviors to result in crashes compared with lower-speed roads with frequent intersections, driveways, and mixed traffic conditions. This finding is consistent with existing studies in which higher speed limits (45 mph or above) were associated with fewer crashes ( 28 , 29 ), suggesting that posted speed limits serve as a proxy for overall roadway design quality rather than being a direct causal factor in crash occurrence.

Overall, exposure (measured by AADT and segment length) is a consistent predictor of increased reckless driving crash frequency across all roadway types. Shoulder presence and wider lanes can reduce risk on certain highway types, especially for aggressive or distracted driving. Meanwhile, higher posted speeds are often associated with reduced crash frequency, possibly reflecting safer and more conservative road design rather than behavioral change. The NB regression model was selected in this study for being consistent with the Highway Safety Manual ( 21 ) safety performance function (SPF) estimation and practical application. While NB regression models were appropriate for this analysis because of the over-dispersed nature of crash count data, future research could explore more advanced modeling approaches. For example, random parameters models may better capture unobserved heterogeneity across roadway segments and provide additional insights. While the 2017–2021 analysis period may not capture the most recent trends, the 5-year window provides sufficient sample sizes for reliable model estimation across all crash types and roadway categories. We recommend that models be periodically recalibrated as newer data become available to ensure continued relevance. Future research should also consider dividing the datasets into calibration and validation subsets to strengthen the generalizability of the identified risk factors. Another limitation of this study is the reliance on police-reported crash data for classifying reckless driving behavior types. Reporting practices may vary across officers, and some overlap between categories (e.g., aggressive driving and speeding) is inherent in the data. Future research could explore methods to account for potential misclassification in crash type coding.

However, the developed SPFs can guide targeted safety improvements by serving as a proactive network screening tool. The models generate predicted crash frequencies for each reckless driving category based on observed AADT, segment length, and roadway characteristics, which can then be compared with observed crash counts to identify over-performing segments where reckless driving crashes occur at a higher rate than predicted. These flagged locations become priority candidates for targeted safety countermeasures, enabling agencies to direct limited resources toward segments with the greatest potential for reckless driving crash reduction. An example of this is provided in Figure 2, which identifies locations of predicted aggressive crashes across the Wisconsin, U.S., highway network by mapping the magnitude of predicted crashes by segments using ArcGIS (geographic information system). These risk predictions can be integrated into state DOT systemic safety programs to proactively identify and screen high-risk corridors, enabling agencies to prioritize countermeasures, such as shoulder widening, lane configuration modifications, or targeted enforcement, based on predicted reckless driving crash frequency rather than waiting for crash history to accumulate.

Figure 2.

Map showing predicted aggressive driving-related crashes by segment.

Footnotes

Acknowledgements

The authors would like to thank the Wisconsin Department of Transportation for funding this work and providing the necessary data that were used in this study.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: S. Guler, V. Gayah; data collection: X. Gu; analysis and interpretation of results: X. Gu, S. Guler, V. Gayah; draft manuscript preparation: X. Gu, S. Guler, V. Gayah. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: S. Ilgin Guler and Vikash Gayah are members of the Transportation Research Record’s Editorial Board.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Wisconsin Department of Transportation (WisDOT) through project ID: 0092-24-11 and the United States Department of Transportation (U.S. DOT) in the interest of information exchange.

ORCID iDs

S. Ilgin Guler

Vikash V. Gayah

Data Accessibility Statement

The crash data used in this study are publicly available through the Wisconsin Traffic Operations and Safety Laboratory (TOPS Lab) Crash Database Query Tool at . The roadway characteristics data that support the findings of this study are not publicly available because they were provided by WisDOT specifically for this research project. Roadway data may be available from the corresponding author on reasonable request and with permission of WisDOT.

The material or information published is the result of research done under the auspices of the Department. The content of this paper reflects the views of the authors, who are responsible for the correct use of brand names, and for the accuracy, analysis, and any inferences drawn from the information or material published. WisDOT and FHWA (U.S. DOT) assume no liability for its contents or use thereof. This paper does not endorse or approve any commercial product, even though trade names may be cited, does not reflect official views or policies of the Department or FHWA (U.S. DOT), and does not constitute a standard specification or regulation of the Department or FHWA.

References

National Highway Traffic Safety Administration. Speeding-Related Fatalities Reach 14-Year High: Speeding Catches Up with You. National Highway Traffic Safety Administration, July 25, 2023. https://www.nhtsa.gov/press-releases/speed-campaign-speeding-fatalities-14-year-high.

National Highway Traffic Safety Administration. Risky Driving. National Highway Traffic Safety Administration, U.S. Department of Transportation. Accessed July 1, 2025. https://www.nhtsa.gov/risky-driving.

Chu

H.-C.

An Investigation of the Risk Factors Causing Severe Injuries in Crashes Involving Gravel Trucks. Traffic Injury Prevention, Vol. 13, No. 4, 2012, pp. 355–363. https://doi.org/10.1080/15389588.2012.654545.

Hasan

A. S.

Jalayer

Das

Kabir

M. A. B.

Application of Machine Learning Models and SHAP to Examine Crashes Involving Young Drivers in New Jersey. International Journal of Transportation Science and Technology, Vol. 14, 2024, pp. 156–170. https://doi.org/10.1016/j.ijtst.2023.04.005.

Islam

M. A.

Dinar

Evaluation and Spatial Analysis of Road Accidents in Bangladesh: An Emerging and Alarming Issue. Transportation in Developing Economies, Vol. 7, No. 1, 2021, p. 10. https://doi.org/10.1007/s40890-021-00118-3.

Duddu

V. R.

Penmetsa

Pulugurtha

S. S.

Modeling and Comparing Injury Severity of At-Fault and Not At-Fault Drivers in Crashes. Accident Analysis & Prevention, Vol. 120, 2018, pp. 55–63. https://doi.org/10.1016/j.aap.2018.07.036.

Love

Rowland

Davey

Exactly How Dangerous Is Drink Driving? An Examination of Vehicle Crash Data to Identify the Comparative Risks of Alcohol-Related Crashes. Crime Prevention and Community Safety, Vol. 25, No. 2, 2023, pp. 131–147. https://doi.org/10.1057/s41300-023-00172-6.

Patwary

A. L.

Khattak

A. J.

Crash Harm Before and During the COVID-19 Pandemic: Evidence for Spatial Heterogeneity in Tennessee. Accident Analysis & Prevention, Vol. 183, 2023, p. 106988. https://doi.org/10.1016/j.aap.2023.106988.

Kamrani

Arvin

Khattak

A. J.

Extracting Useful Information from Basic Safety Message Data: An Empirical Study of Driving Volatility Measures and Crash Frequency at Intersections. Transportation Research Record: Journal of the Transportation Research Board, 2018. 2672: 290–301.

10.

Gong

Fan

W. D.

Modeling Single-Vehicle Run-Off-Road Crash Severity in Rural Areas: Accounting for Unobserved Heterogeneity and Age Difference. Accident Analysis & Prevention, Vol. 101, 2017, pp. 124–134. https://doi.org/10.1016/j.aap.2017.02.014.

11.

Penmetsa

Pulugurtha

S. S.

Duddu

V. R.

Examining Injury Severity of Not-At-Fault Drivers in Two-Vehicle Crashes. Transportation Research Record: Journal of the Transportation Research Board, 2017. 2659: 164–173.

12.

Weiss

H. B.

Kaplan

Prato

C. G.

Analysis of Factors Associated with Injury Severity in Crashes Involving Young New Zealand Drivers. Accident Analysis & Prevention, Vol. 65, 2014, pp. 142–155. https://doi.org/10.1016/j.aap.2013.12.020.

13.

Paleti

Eluru

Bhat

C. R.

Examining the Influence of Aggressive Driving Behavior on Driver Injury Severity in Traffic Crashes. Accident Analysis & Prevention, Vol. 42, No. 6, 2010, pp. 1839–1854. https://doi.org/10.1016/j.aap.2010.05.005.

14.

McCombs

Al-Deek

Sandt

Carrick

Identifying Important Roadway Corridor Factors Which Affect Crashes Influenced by Unsafe Driving Behaviors to Help Law Enforcement Proactively Reduce Crashes. Transportation Research Record: Journal of the Transportation Research Board, 2024. 2678: 706–719.

15.

Awadalla

D. M.

De Albuquerque

F. D. B.

Fatal Road Crashes in the Emirate of Abu Dhabi: Contributing Factors and Data-Driven Safety Recommendations. Transportation Research Procedia, Vol. 52, 2021, pp. 260–267. https://doi.org/10.1016/j.trpro.2021.01.030.

16.

Lastrucci

Innocenti

Lorini

Berti

Silvestri

Lazzeretti

Voller

Bonaccorsi

The Prevalence of Several Risky Driving Behaviors and Associated Crash Risk in Adolescents: A Population-Based Study of Tuscany Region. International Journal of Public Health, Vol. 67, 2022, p. 1604582. https://doi.org/10.3389/ijph.2022.1604582.

17.

Arvin

Khattak

A. J.

Driving Impairments and Duration of Distractions: Assessing Crash Risk by Harnessing Microscopic Naturalistic Driving Data. Accident Analysis & Prevention, Vol. 146, 2020, p. 105733. https://doi.org/10.1016/j.aap.2020.105733.

18.

Islam

Unraveling the Differences in Distracted Driving Injury Severities in Passenger Car, Sport Utility Vehicle, Pickup Truck, and Minivan Crashes. Accident Analysis & Prevention, Vol. 196, 2024, p. 107444. https://doi.org/10.1016/j.aap.2023.107444.

19.

Gayah

V. V.

Guler

S. I.

Zhang

Reckless Driving Crash Risks on Wisconsin Roadways. Wisconsin Department of Transportation Research Report 0092-24-11, 2025. https://wisconsindot.gov/documents2/research/0092-24-11-final-report.pdf.

20.

Wisconsin Traffic Operations and Safety Laboratory. Crash Database Query Tool. https://transportal.cee.wisc.edu/services/crash-data/. Accessed July 9, 2025.

21.

Highway Capacity Manual. Transportation Research Board, Washington, D.C., Vol. 2, No. 1, 2000, p. 1.

22.

Shankar

V. N.

Albin

R. B.

Milton

J. C.

Mannering

F. L.

Evaluating Median Crossover Likelihoods with Clustered Accident Counts: An Empirical Inquiry Using the Random Effects Negative Binomial Model. Transportation Research Record: Journal of the Transportation Research Board, 1998. 1635: 44–48.

23.

AASHTO. Highway Safety Manual, 1st ed. American Association of State Highway and Transportation Officials, Washington, D.C., 2014.

24.

Geedipally

S. R.

Lord

Dhavala

S. S.

The Negative Binomial Lindley Generalized Linear Model: Characteristics and Application Using Crash Data. Accident Analysis & Prevention, Vol. 45, No. 2, 2012, pp. 258–265. https://doi.org/10.1016/j.aap.2011.07.012.

25.

Hilbe

J. M.

Negative Binomial Regression, 2nd ed. Cambridge University Press, Cambridge, UK and New York, 2011.

26.

Hauer

The Art of Regression Modeling in Road Safety, Vol. 38. Springer, New York, NY, 2015.

27.

Lyon

Persaud

Gross

The Calibrator – An SPF Calibration and Assessment Tool User Guide. Report Number: FHWA-SA-17-016. 2016.

28.

Liu

Zhao

Sharma

Multivariate Random Parameters Zero-Inflated Negative Binomial Regression for Analysing Urban Midblock Crashes. Analytical Methods in Accident Research, Vol. 17, 2018, pp. 32–46. https://doi.org/10.1016/j.amar.2018.03.001.

29.

Sengupta

Gayah

V. V.

Donnell

E. T.

Examining the Impacts of Crash Data Aggregation on SPF Estimation. Accident Analysis & Prevention, Vol. 160, 2021, p. 106313. https://doi.org/10.1016/j.aap.2021.106313.