Prerpints.org logo
Preprint
Article

Impact Analysis of Road Infrastructure and Traffic Control on Injury Severity of Single- and Multi-vehicle Crashes

This version is not peer-reviewed.

Submitted:

27 June 2023

Posted:

27 June 2023

You are already at the latest version

Abstract
Single and multi-vehicle crashes are a significant issue that causes economic and social costs and has therefore gained attention. This study explored the factors associated with injury severity for both single- and multi-vehicle crashes using over 550,000 crash data points in Japan between 2019 and 2021. We identified the determinants of road infrastructure and traffic control while controlling for driver, vehicle, environmental, and accident characteristics by applying ordered logit and bias-reduced binomial regression models. Our findings are as follows. 1) Traffic control variables did not affect single-vehicle crashes. 2) Guardrails had a higher severity in both single-vehicle crashes and multi-vehicle crashes at intersections. 3) The impact of the centerline differed between intersections and non-intersections for multi-vehicle crashes.  These results of this study provide transportation agencies with important guidance as to the road infrastructure and transport control.
Keywords: 
injury severit
;  
road infrastructure
;  
traffic control
;  
singe- and multi-vehicle crash
;  
population density
Subject: 
Engineering  -   Safety, Risk, Reliability and Quality

1. Introduction

The effective treatment of road accidents and the enhancement of road safety are major concerns for societies owing to the loss of human lives, economic costs, and social impact. Accident severity has gained the attention of many researchers, and transportation practitioners have made tremendous efforts to improve road safety [1].
While advancements owing in airbags, sensing technology, road structures, and traffic signal control have contributed to reducing the number of driver and passenger fatalities in traffic accidents in the United States, Europe, and Japan, the ratio of driver and passenger fatalities to the total number of traffic fatalities remains high at over 50%. In this study, we analysed the factors associated with the severity of both single- and multi-vehicle crashes.
Numerous studies have recently analysed the factors that impact injury severity under limited conditions (see Table 1). Wu et al. (2014) highlighted significant differences in the causal attributes determining driver injury severity between single- and multi-vehicle crashes [2]. They analysed accident type and weather conditions (visibility) for both types of crashes. In analysing multi-vehicle crashes, it is essential to differentiate between intersections and non-intersections [3]. Furthermore, a comparative analysis was performed on single-vehicle crashes in urban and rural areas. Additionally, a multi-vehicle accident analysis that includes motorcycles, which have a relatively high risk, was conducted.
Table 1. Previous studies [2,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28].
Table 1. Previous studies [2,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28].
Conditions Single-vehicle Multi(Two)-vehicle
Intersection Sharafeldin et al. (2022a) [4]
Yuan et al. (2022) [5]
Accident type Zhou and Chin (2019) [6]
Khan and Vachal (2020) [7]		 Wang and Abdel-Aty (2008) [8]
Liu and Fan (2020) [9]
Zhang et al. (2021) [10]
Yaman et al. (2022) [11]
Sharafeldin et al. (2022b) [12]
Weather
(Visibility)		 Naik et al. (2016) [13]
Li et al. (2018) [14]
Li et al. (2019) [15]
Cai et al. (2021) [16]		 Mphekgwana (2022) [17]
Vehicle type
 Zou et al (2017) [18]
Agrawal et al. (2019) [19]
Wahab and Jiang (2019) [20]
Yang et al.(2019) [21]
Champahom et al. (2022) [22]
Region
(Urban/rural)		 Wu et al. (2016) [23]
Comparison Wu et al. (2014) [2], Rezapour et al. (2018) [24],
Ma et al. (2023) [25]
Road barrier Li et al. (2018) [26], Russo & Savolainen (2018) [27],
Molan et al. (2020) [28]
The following six factors have been identified as potential influential factors [29].
  • Driver: age, sex, drug/alcohol impairment, seat belt use.
  • Vehicle type
  • Environment: weather, time of day, day of the week, region (urban/rural), land use
  • Accident status: rollover, collision mark of crashed vehicle, airbag activation
  • Road structure: centreline, boundary between sidewalk and roadway, road alignment (curve and slope), number of lanes, etc.
  • Traffic control: speed limit, traffic signals, stop signs, zone 30 etc.
Out of these six factors, policymakers cannot control the first four. In this study, we focused on the road structure and traffic control.
This section discusses the impact of road structures on the injury severity in single-vehicle crashes. Wu et al. (2016) analysed driver injury severity in single-vehicle crashes on rural and urban roads [23]. Using 2010–11 accident data from New Mexico, USA, they found that curved, multi-lane roads in urban areas, in addition to alcohol impairment, female and senior drivers, rain conditions, no-passing zones in rural areas, and drug-impaired drivers in urban areas, increased severity. Li et al. (2018) and Moran et al. (2020) analysed the injury severity of the boundary between sidewalks and roadways including various road barriers and found different impacts based on barrier and vehicle types [26,27].
Using data from four US states, Li et al. (2019) investigated driver injury severity in rural single-vehicle crashes under rainy conditions [15]. They demonstrated that curved, on-grade, multi-lane roads significantly increased the probability of incapacitating injuries, controlling for driver, vehicle, and environmental factors. Additionally, signal control was found to be an influencing factor. Zhou and Chin (2019) analysed the factors influencing single-loss-of-control single-vehicle crashes for two- and four-wheeled vehicles in Singapore [6]. They showed that the type of median lane and high-speed limit roads had different influences on riders and drivers in terms of injury severity.
Some recent studies have investigated the impact of speed limits on injury severity in traffic accidents. For instance, Khan and Vachal (2020) examined the factors influencing the severity of single-vehicle rollover crashes, including variables such as environment, driver, vehicle type, and accident location [7]. They found that higher speed limits were associated with more severe injuries. Similarly, Zhang et al. (2021) focused on left-turn accidents and found that high-speed limits and protected left-turn signals were related to increased injury severity [10]. Gong et al. (2022) identified several risk factors for injury severity, including speed limit, driver age and gender, seatbelt use, speeding, and vehicle type [30].
Additionally, Sharafeldin et al., (2022a) analysed the factors affecting the severity of two-vehicle crashes and found that urban and signalized intersections were associated with reduced severity levels and higher pavement friction was associated with less severe crashes [4].
Several studies have compared single and multi-vehicle crashes, such as those by Wu et al. (2014) and Rezapour et al. (2018) [2,24]. Ma et al. (2023) found that time, road, speed, lighting, and weather correlate positively with single-vehicle crash injury severity but negatively with multi-vehicle crash injury severity [25]. Factors such as area, location, and angle are significant only for single-vehicle crashes, whereas the day, interference, and wind are significant only for multi-vehicle crashes. However, it is assumed that the parameters are the same at intersections and non-intersections, except for the intercept. Thus, there is a need for further studies that compare single-vehicle crashes, which mostly occur at non-intersections.
However, there are limited analyses that focus on road structures and traffic control, as well as differences between single- and multi-vehicle crashes at non-intersections and intersections.
Hence, this study aims to identify the factors that influence road structures and traffic control in single and multi-vehicle crashes. It also compares the common and different factors in single- and multi-vehicle crashes at non-intersections and investigates the differences in influencing factors between intersections and non-intersections for multi-vehicle crashes. There are three novelties in this research. First, the road structure and traffic regulation are analyzed using the eight variables described later. More than 550,000 single- and multi-vehicle crashes in Japan are analyzed. And, we compared the factors affecting the severity between single- and multi-vehicle crashes and between intersections and non-intersection sections for multi-vehicle crashes.

2. Materials and Methods

2.1. Materials

This study utilizes crash data obtained from the National Police Agency (NPA) of Japan. The dataset consists of 995,611 traffic crashes that occurred between 2019 and 2021, which includes 8,500 fatalities. These data provide information on the severity of the crashes, as well as variables related to road infrastructure, and traffic control at crash locations. These variables were first published in 2020. For the purposes of this study, 11,882 single-vehicle and 554,365 multi-vehicle crashes were identified.
Regarding single-vehicle crashes, driver injuries were categorised as fatality (788 cases), injury (6,719 cases), and no injury (4,375 cases). The fatality rate for drivers in single-vehicle crashes was high, with 6.6% of all single-vehicle crashes resulting in driver fatalities compared to 0.9% for all crashes. In this study, only single-vehicle crashes that occurred at non-intersections were analysed since the number of single-vehicle accidents at intersections is relatively low.
In multi-vehicle crashes, injury levels were separated into fatality (1,468 cases), injury (410,510 cases), and no injury (142,387 cases). Because the number of fatalities was quite low in this study, injury severity was defined as either fatality or not.
Table 2 presents the descriptive statistics for the 18 categorical variables selected for this dataset and for the population within 500 m of the crash location. The variables common to both kinds of accidents were: (1) centreline, (2) boundary between sidewalk and roadway, (3) road alignment curve, (4) road alignment slope, (5) speed limit, (6) zone-30-policy, (7) traffic signal, (8) stop sign, (9) land use, (10) population density (population within 500-m radius), (11) weather, (12) time period, (13) day type, (14) collision marks of a crashed vehicle, (15) airbag activation.
Table 2. Variables and their distributions.
Table 2. Variables and their distributions.
Crash type Single Two Single Two
Variable and category N % N % Variable and category N % N %
Road infrastructure Environment
Centerline Weather
No 3,356 29.9 140,462 25.3 Clear 7,401 62.3 361,933 65.3
Paint 6,435 54.2 294,000 53.0 Cloudy 2,634 22.2 113,893 20.5
Median 1,699 16.2 112,458 20.3 Bad 1,847 15.5 78,539 14.2
Other (a) 192 1.4 7,445 1.3 Time period
Boundary between sidewalk and roadway After dawn 447 2.5 19,431 3.5
Curb 5,735 48.3 370,920 66.9 Daytime 6,774 54.2 361,971 65.3
Guard rail 1,311 11.0 46,302 8.4 Before dusk 561 40.5 36,643 6.6
White line 2,438 20.5 79,742 14.4 After dusk 592 7.5 41,678 7.5
No 2,398 20.2 57,401 10.4 Nighttime 3,206 40.5 87,912 15.9
Road alignment- Curve Before dawn 302 2.8 6,730 1.2
Straight 8,725 73.5 531,416 95.9 Day type
Inside 1,845 15.5 10,611 1.9 Weekday 3,901 32.8 151,176 27.3
Outside 1,312 11.0 12,338 2.2 Holiday, weekend 7,981 67.2 403,189 72.7
Road alignment- Slope Vehicle and driver
Flat 9,254 77.9 507,036 91.5 Vehicle type
Up 1,022 8.6 20,102 3.6 Cars 4,107 34.6 - -
Down 1,606 13.5 27,227 4.9 Kei cars 2,341 19.7 - -
Traffic control Large truck 337 2.8 - -
Stop sign Small/Medium truck 1,035 8.7 - -
Yes 57,874 10.4 Motorcycle 126+cc 1,383 11.6 - -
No 150,386 27.1 Motorcycle –125cc 2,679 22.5 - -
Not applicable (b) 346,105 62.4 Driver age
Speed limit (km/h) 16–24 2,284 19.2 - -
20,30 1,683 14.2 43,656 7.9 25–34 1,376 11.6 - -
40 3,349 28.2 170,695 30.8 35–44 1,401 11.8 - -
50 2,587 21.8 153,493 27.7 45–54 1,862 15.7 - -
60 4,263 35.9 186,521 33.6 55–64 1,768 14.9 - -
Traffic signal 65–74 1,884 15.9 - -
Three-light 760 6.4 145,607 26.3 75– 1,307 11.0 - -
Pedestrian-controlled (c) 41 0.3 6,525 1.2 Accident type
Pedestrian-vehicle separated 20 0.2 2,711 0.5 Airbag
Flashing 9 0.1 4,582 0.8 Activated 2,740 23.1 154,396 27.9
None 11,052 93.0 394,940 71.2 Non-activated/Unsupported 9,142 76.9 399,969 72.1
Zone-30-policy Collision marks of a crashed car
Yes 97 0.8 4,381 0.8 Front 4,596 38.7 246,896 44.5
No 11,785 99.2 549,984 99.2 Right 1,245 10.5 29,366 5.3
Environment Rear 437 3.7 107,615 19.4
Land uses Left 1,467 12.3 37,839 6.8
Urban- DID 4,780 40.2 251,700 45.4 diagonally right front 841 7.1 66,243 11.9
Urban- nonDID 2,669 22.5 177,870 32.1 diagonally left front 1,887 15.9 58,712 10.6
Rural 4,433 37.3 124,795 22.5 No 1,409 11.9 7,694 1.4
Crash location
Population density Mean s.d, Mean s.d. Non-intersections 11,882 100.0 346,105 62.4
Population within 500-meter radius 3,929 5,698 4,050 3,829 Intersections 0 0 208,260 37.6
(a) Particular processing such as "High-brightness paints," "Postcones," and "Chatter bars."
(b) This refers to cases where the crash occurred at a location other than within intersections, or where the vehicle type is unknown.
(c) A traffic signal where the pedestrian signal turns green only when a button is pressed. In the UK, pedestrian crossings with this type of signal are called “pelican crossings”.
Variables related to road infrastructure (variables (1)–(4)) and traffic control (variables (5)–(8)) were categorised in detail. This study considers these nine variables.
For (3) road alignment curve, "Inside" refers to driving on the inside of the curve, and "Outside" refers to driving on the outside of the curve, as vehicles in Japan drive on the left side of the road. In multi-vehicle crashes, data recorded the direction of the curve as seen from the perspective of the at-fault driver. The same applies to road alignment slopes.
The speed limit variable (5) in this study pertains to non-highway roads in Japan, with speed limits ranging from 20 to 60 km/h. Expressways with speeds exceeding 60 km/h were not included in the data. The (6) zone-30-policy is a traffic safety measure implemented in residential areas surrounded by arterial roads, with a speed limit of 30 km/h to ensure the safety of pedestrians and cyclists. For non-intersection accidents, (7) traffic signal data were recorded only for accidents within 30 m of a pedestrian crossing at an intersection, where traffic signals were used in combination with other safety measures to restrict vehicle speed. The variable (9) land uses in this study used population density within a 500-m radius, with "DID" standing for "Densely Inhabited District," referring to census tracts with a population density of 4,000 or more people per square kilometre and a population of 5,000 or more people. Table 2 provides details on the categories for each variable.
Additionally, to study driver characteristics in single-vehicle crashes, we considered vehicle type (16) and driver age (17) as variables. For multi-vehicle crashes, we used (18) vehicle type combination (hit and being hit) and (19) driver age combination (hit and being hit) as additional variables. For multi-vehicle crashes, we built a separate model based on the crash location (intersection and non-intersection).
Table 3 shows the descriptive statistics for the variables used in this study. Age groups were classified as (0–24, 25–64, and 65–), and vehicle types were classified as passenger cars (cars and kei cars), large trucks, small/medium trucks, and motorcycles. For each category, the attribute before the "×" represents the perpetrator, and the attribute after the "×" represents the victim.
Table 3. Driver age and vehicle type combination of multi-vehicle crash.
Table 3. Driver age and vehicle type combination of multi-vehicle crash.
Variable and category N % Variable and Category N %
Driver age combination Vehicle type combination
16–24 × 16–24 9,716 1.8 (C-C) Passenger car × Passenger car 304,732 55.0
16–24 × 25–64 57,532 10.4 (C-M) Passenger car × Motorcycle 86,652 15.6
16-24 × 65– 2,727 0.5 (C-ST) Passenger car × Small/Medium truck 78,516 14.2
25–64 × 16–24 63,005 11.4 (LT-C) Large truck × Passenger car 30,426 5.5
25–64 × 25–64 362,344 65.4 (ST-M) Small/Medium truck × Motorcycle 14,700 2.7
25–64 × 65– 24,420 4.4 (ST-C) Small/Medium truck × Passenger car 11,428 2.1
65– × 16–24 3,864 0.7 (ST-ST) Small/Medium truck ×
Small/Medium truck		 8,680 1.6
65– × 25–64 28,470 5.1 (LT-ST) Large truck × Small/Medium truck 6,308 1.1
65– × 65– 2,287 0.4 (LT-M) Large truck × Motorcycle 5,386 1.0
(M-M) Motorcycle × Motorcycle 4,913 0.9
(LT-LT) Large truck × Large truck 2,624 0.5
We did not observe accidents in which the vehicle type combination of the perpetrator and victim was motorcycle × (large truck, small/medium truck, passenger car), or small/medium truck × large truck. The number of passenger car × large truck combinations was also very low at 130. Therefore, we included vehicle-type combinations as variables in 11 categories, excluding the above five categories.

2.2. Methods

In this study, we utilized ordered logit and bias-reduced binomial regression models to analyse the severity of injuries for single- and multi-vehicle crashes. We employed ordered logit models to assess the connections between the dependent variables, which were categorical and ordered, such as "fatalities," "injuries," and "no injuries," and a range of independent variables. This entry specifically pertains to models in which the outcomes can be ordered. An ordered logit model estimates the underlying score by a linear function of the independent variables and a set of cutoff points.
The probability of observing outcome i corresponds to the probability that the estimated linear function, plus random error, is within the range of the cutoff points estimated for the outcome. Its equation can be given as follows:
Pr(outcomej = i) = Pr(κi−1 < β1x1j + β2x2j + · · · + βkxkj + uj ≤ κi)
where uj is assumed to be logistically distributed in ordered logit. In either case, we estimate the coefficients β1, β2, . . ., βk along with the cutoff points κ1, κ2, . . . , κk−1, where k is the number of possible outcomes. κ0 is taken as −∞, and κk is taken as +∞. This is a direct generalisation of the ordinary two-outcome logit model [30].
For the multi-vehicle crash analysis, we applied a bias-reduced logistic regression [31]. Conventional logistic regression analysis causes a first-order term in the asymptotic bias of the maximum likelihood (ML)coefficient estimates. The logistic regression model P ( y i = 1 | x i ) = ( 1 + e x p ( − x i β ) ) − 1 with i = 1, . . . , N associates a binary outcome y i ∈ {0, 1} with a vector of covariate values x i = (1, xi1, . . . , xip) using a (p + 1)-dimensional vector of regression parameters β = (β0, β1, . . . , βp) ′ . The ML estimate β ^ M L is given by the parameter vector maximising the log-likelihood function l(β) and is usually derived by solving the score equations ∂l/∂βr = 0 with r = 0, . . . , p. For ML estimates, the proportion of observed events is equal to the average predicted probability. This can be observed in the explicit form of the ML score function ∂l/∂βr = ∑ i y i − π i x i r where π i = ( 1 + e x p ( − x i β ) ) − 1 denotes the predicted probability for the ith observation.
Firth (1993) showed that penalising the likelihood function using the Jeffreys invariant prior removes the first-order term from the asymptotic bias of the ML coefficient estimates [32]. Jeffreys invariant prior is given by | I ( β ) | 1 / 2 = | X ' W X | 1 / 2 , where I(β) is the Fisher information matrix, X is the design matrix, and W is the diagonal matrix diag(πi(1− πi)). Estimates of coefficients β ^ F L for Firth-type penalised logistic regression (FL) can be found by solving the corresponding modified score equations:
∂l/∂βr = ∑ i = 1 N y i − π i + h i ( 1 2 − π i ) x i r = 0 , r = 0, . . . p,
where hi is the ith diagonal element of the hat matrix W 1 / 2 X | X ' W X | 1 / 2 X ' W 1 / 2 .
Moreover, we incorporated the Bayesian information criterion (BIC) for selecting the most appropriate model. We also included an interaction term to account for potential variations in the impact of a factor, based on the surrounding environment.

3. Results

3.1. Single-vehicle crash

Table 4 illustrates the findings from the ordered logistic regression analysis of the single-vehicle crashes. Our analysis revealed that road alignment curve, boundary between sidewalk and roadway, time period, vehicle type, driver age, airbags, primary collision marks, and population density were significant variables impacting fatality risk. Most single-vehicle crashes are attributable to driver error or violations of laws and regulations. ¸Conversely, the risk of death is lower for "Daytime," "Before dusk," and "Populated areas." Surprisingly, night-time does not increase the risk of mortality. Furthermore, the interaction term analysis suggests that the fatality risk from single motorcycle crashes is higher in populated areas.
Table 4. Estimation results of single vehicle crashes (ordered logistic regression).
Table 4. Estimation results of single vehicle crashes (ordered logistic regression).
Variable Category Coef. t value Fatality n
Road alignment- Curve (ref) Straight 418 8,725
Inside 0.23*** 3.70 209 1,845
Outside 0.29*** 4.12 161 1,312
Boundary
between sidewalk and roadway		 (ref) Curb 353 5,735
Guardrail 0.18*** 2.58 97 1,311
White line 0.23*** 4.06 184 2,438
No 0.15*** 2.69 154 2,398
Time period (ref) After dawn 37 447
Daytime −0.32*** −2.91 384 6,774
Before dusk −0.44*** −3.06 25 561
After dusk −0.25 −1.78 29 592
Nighttime 0.06 0.50 287 3,206
Before dawn −0.06 −0.34 26 302
Population density In logarithm −0.14*** −10.31 - -
Airbag (ref) Activated 313 2,740
Daytime −0.99*** −17.32 475 9,142
Primary collision mark (ref) Front 472 4,596
Right −0.17*** −2.22 52 1,245
Rear −1.01*** −8.08 21 437
Left −0.26*** −3.58 85 1,467
Diagonally right front −0.17*** −1.99 72 841
Diagonally left front −0.71*** −11.17 75 1,887
No −1.39*** −15.82 11 1,409
Vehicle type (ref) Cars 158 4,107
Kei cars 1.01*** 16.73 183 2,341
Large truck 0.73*** 5.67 41 337
Small/Medium truck 1.10*** 13.67 108 1,035
Motorcycle 126+cc 2.59*** 15.11 168 1,383
Motorcycle –125cc 2.30*** 12.85 130 2,679
Driver age (ref) 16–24 114 2,284
25–34 0.52*** 6.70 76 1,376
35–44 0.62*** 7.86 85 1,401
45–54 0.67*** 9.07 126 1,862
55–64 0.60*** 7.99 115 1,768
65–74 0.72*** 9.73 113 1,884
75– 1.22*** 14.55 159 1,307
Interaction term Vehicle type: ”Motorcycle” ×
Population density (in logarithm)		 0.10*** 4.44
Intercept Fatality | Injury −0.96*** −5.90
Injury | No injury 3.23*** 19.19
BIC 15,989
BIC (null) 20,698
Number of observations 11,882

3.2. Multi-vehicle crash

Table 5 presents the outcomes of the bias-reduced logistic regression analysis of multi-vehicle crashes. Our analysis revealed that in both intersection and non-intersection models, the centerline, speed limit, time period, airbag, primary collision mark, driver age combination, vehicle type combination, and population density were significant variables impacting fatality risk. In the intersection crash model, only the boundary between sidewalk and roadway, stop signs, and day types were identified as factors influencing the risk of fatalities. In contrast, only the road alignment curve was identified as a significant variable in the non-intersection crash model. Moreover, several road infrastructure and traffic control variables substantially affected the risk of fatalities compared to single-vehicle crashes.
Table 5. Estimation results of multi-vehicle crashes (bias-reduced logistic regression).
Table 5. Estimation results of multi-vehicle crashes (bias-reduced logistic regression).
Crash location (fatality rate) Intersections (0.4%) Non-intersections (0.2%)
Variable Category Coef. Z value fatality n Coef. Z value fatality n
Centreline (ref) No 426 96,560 50 43,902
Paint −0.57*** −7.06 272 88,239 1.17*** 7.62 640 205,761
Median −1.05*** −7.46 62 21,751 0.78*** 4.38 113 90,707
Other −0.96*** −2.03 4 1,710 1.35*** 5.07 23 5,735
Road alignment- curve (ref) Straight - - - - 533 328,059
Inside - - - - 0.54*** 3.88 68 8,019
Outside - - - - 1.04*** 9.11 225 10,027
Boundary
between sidewalk
and roadway		 (ref) Curb 456 125,566 - - - -
Guardrail 1.07*** 11.04 154 14,075 - - - -
White line −0.50*** −4.13 88 38,578 - - - -
No −0.86*** −6.19 66 30,041 - - - -
Stop sign (ref) Yes 103 57,874 - - - -
No 0.51*** 4.55 661 150,386 - - - -
Speed limit
(km/h)		 (ref) 20,30 37 27,306 13 16,350
40 0.44*** 2.41 192 61,068 0.59*** 2.09 190 109,627
50 0.78*** 4.25 205 37,583 0.92*** 3.26 328 115,910
60 0.85*** 4.84 330 82,303 1.02*** 3.61 295 104,218
Time period (ref) After dawn 39 7,817 43 11,614
Daytime −0.62*** 3.63 396 135,390 −0.42*** −2.50 477 226,581
Before dusk −0.51*** −2.19 34 13,868 −0.29 −1.24 37 22,775
After dusk −0.43 −1.95 42 14,174 −0.35 −1.50 35 27,504
Night-time 0.37*** 2.10 227 34,171 0.51*** 2.83 198 53,741
Before dawn 0.30 1.16 26 2,840 0.73*** 3.07 36 3,890
Day type (ref) Weekday
Weekends/Holiday		 527 150,614 - - - -
0.26*** 3.23 237 57,646 - - - -
Population density in logarithm −0.38*** −16.75 - - −0.28*** −14.86 - -
Airbag (ref) Activated 670 87,255 692 67,141
Non-activated/Unsupported −1.73*** −12.59 94 121,005 −2.34*** −21.06 134 278,964
Primary collision mark (ref) Front - - - - 482 105,635
Right - - - - –0.56*** –4.04 67 18,213
Rear - - - - –1.49*** –10.43 60 178,370
Left - - - - –0.38*** –2.03 33 11,433
Diagonally right front - - - - 0.20 1.86 158 18,313
Diagonally left front - - - - –0.77*** –3.07 17 11,587
No - - - - –0.28 –0.85 9 2,554
Vehicle type combination (ref) (C-C) 89 104,980 154 199,752
(C-M) 1.41*** 10.76 355 49,760 0.34*** 2.49 168 36,892
(C-ST) 0.85*** 4.61 46 24,664 0.47*** 3.18 70 53,852
(LT-C) 1.91*** 11.26 62 8,015 1.86*** 15.84 174 22,411
(ST-M) 1.86*** 11.55 88 8,262 0.79*** –4.20 4 6,438
(ST-C) 0.79 1.93 6 3,327 –0.28 –0.63 5 8,101
(ST-ST) 1.09*** 2.47 5 2,365 0.54 1.41 7 6,315
(LT-ST) 2.73*** 10.45 19 1,496 2.77*** 17.73 75 4,812
(LT-M) 3.02*** 18.09 85 2,276 2.24*** –14.21 97 3,110
(M-M) 0.40 0.99 6 2,712 –0.09 –0.23 7 2,201
(LT-LT) 2.41*** 4.30 3 403 2.37*** 10.14 25 2,221
Driver’s age combination (ref) 16–24 × 16–24 18 3,746 27 5,970
16–24 × 25–64 0.23 0.87 63 19,175 –0.07 –0.23 1424 38,357
16–24 × 65– 1.75*** 4.82 14 1,239 1.68*** 4.05 12 1,488
25–64 × 16–24 –0.20 –0.79 104 25,673 –0.33 –1.08 91 37,332
25–64 × 25–64 –0.01 –0.05 388 130,856 0.08 0.27 495 231,488
25–64 × 65– 1.50*** 5.77 120 11,103 1.31*** 4.26 119 13,317
65– × 16–24 0.29 –0.60 5 2,049 2.03 1.42 0 1,815
65– × 25–64 0.40 1.39 41 13,243 0.66*** 1.98 29 15,227
65– × 65– 1.45*** 3.77 11 1,176 1.99*** 4.91 20 1,111
Interaction term Curve: “Outside” ×
Vehicle type: “Motorcycle”		 0.69*** 3.82 73 1,642
Intercept −3.66*** −9.14 - - −5.10*** −10.78 - -
BIC 8,628 8,456
BIC (null) 10,106 11,637
Number of observatios 208,260 346,105
Our study identified several road infrastructure characteristics that impact fatality risk differently. Specifically, "Curved road" (non-intersection), "Guard rail" (intersection), and "No centreline" (intersections) increased the fatality risk, while "Median" (non-intersection) decreased the fatality risk. In terms of intersection crashes, "No stop signs" was identified as a significant factor contributing to higher fatality risk. Additionally, speed limit was found to be a significant variable for both crash locations. Surprisingly, our analysis did not find any effect of traffic signals on fatality risk.
In our study, we also identified other factors associated with a lower or higher fatality risk. The factors that lowered the fatality risk included "Daytime," "Before dusk" (intersections), and "Populated areas." Moreover, factors that raised the fatality risk included "Weekends/Holiday," "Night time," "Before dawn" (non-intersections only), "Airbag Activated," "Perpetrator is Large trucks," "Motorcycle involvement," and "Front collision marks" (non-intersections only). Moreover, motorcyclists were found to be at an even higher risk of death during "Outside curves," which was shown as an interaction term.
Our study found that driver age had a similar impact on the severity of multi-vehicle crashes as in single-vehicle accidents. Specifically, the severity of crash was found to increase as age increases. However, we also found that the impact was not the same between hitting drivers and drivers who were hit. In particular, the risk of death was estimated to be higher in the case of a hit than in the case of a collision.
Furthermore, we found that a higher population density was associated with decreased injury severity in both single- and multi-vehicle crashes. We believe that this may be due to population density acting as a proxy variable for vehicle speed; specifically, the higher the population density, the higher the traffic volume and intersection density, which may result in lower vehicle speeds and, consequently, lower injury severity in the event of a crash.

4. Discussion

4.1. Comparison of single- and multi-vehicle crashes at non-intersection

Our study found that injury severity was higher outside curves, which is consistent with previous research [2,9]. Excessive speed can make it difficult for drivers to navigate curves, leading to severe accidents. As a result, reducing speed limits and implementing devices to slow down vehicles can be effective measures for preventing serious accidents.
The analysis found that the speed limit was a significant variable affecting the severity of multi-vehicle crashes, but traffic control measures such as speed limit and Zone 30) were not associated with the severity of single-vehicle crashes. However, this dataset did not include information on the speed at the time of collision; therefore, further investigation is needed to better understand the relationship between speed and crash severity. Additionally, it was found that the severity of single-vehicle crashes cannot be controlled by traffic control measures in Japan. Furthermore, the study revealed that the absence of a curb to separate pedestrians and vehicles increases the severity of crashes, which is often the case on rural roads with few pedestrians or cyclists.
In multi-vehicle crashes, a centerline was found to be a significant factor associated with higher severity, possibly due to the increased impact during collisions or collision with the median strip as a physical barrier. Both single- and multi-vehicle crashes have a high fatality risk at night or before dawn. As reported in previous studies, the elderly and motorcycles are at a higher risk of accidents.

4.2. Comparison of intersection and non-intersection of multi-vehicle crashes

Regardless of whether the crash occurs at an intersection or a non-intersection, the higher the speed limit, the higher the severity of the crash. This finding is similar to that of a previous study.
An intersection with a centreline was found to have a reduced risk of injury severity, while the risk increased for non-intersections. This may be due to the difference in speed between the two vehicles at the time of collision. In a collision at an intersection, one vehicle is likely to stop or move at a lower speed, whereas, on a single road, both vehicles may be moving at high speeds. Interestingly, this result is consistent with the findings of a study on pedestrian-vehicle crashes [33]. The presence of a centreline was also found to decrease injury severity in bicycle-vehicle crashes, both at intersections and non-intersections [34].　
The location of a crash was found to be a significant variable for both single- and multi-vehicle crashes. The severity was higher at intersections than at non-intersections, and further investigation is needed to determine the reason for this difference. Additionally, the type of barrier used in the roadway was also found to be a significant variable in determining the severity of a crash. Li et al. (2018) found that hitting a guardrail reduced the probability of fatal and severe injury and that a strong post-W-beam guardrail resulted in significantly more fatal and severe crashes than a low-tension cable system [26]. Russo & Savolainen (2018) showed the severity varies across the different barrier types [27]. Moran et al. (2020) suggested that a rigid barrier might be more effective for reducing truck crash severity than a guardrail barrier. Further investigation is required, including barrier type and crash location [28].
Furthermore, as in previous studies, the risk of death was found to be higher for elderly individuals involved in motorcycle crashes.

5. Conclusions

This study identified the factors contributing to road structure and traffic control in single- and multi-vehicle crashes. This study is the first result of analysis using a dataset of over 550,000 recorded accidents in Japan, applying the ordered logit and bias-reduced logistic regression models and controlling for various driver, vehicle, environment, and accident type characteristics. Then we compared the common and distinct factors of single- and multi-vehicle crashes at non-intersections and investigated the differences in influencing factors between intersections and non-intersections for multi-vehicle crashes. We confirmed that: 1) traffic control variables (such as speed limit and Zone 30) did not affect single-vehicle crashes, 2) guardrails caused high-severity crashes in single- and multi-vehicle crashes at intersections, 3) the impact of the centreline differed between intersections and non-intersections for multi-vehicle crashes, 4) population density had the negative impacts on crash severity.
The results of this study provide transportation agencies with important guidance as to the center line and boundary between sidewalk and roadway and stop signs. Especially, reducing the vehicle speed is crucial in reducing injury severity. However, our study suggests that it may be challenging to lower the severity of single-vehicle crashes through traffic control measures alone. In an attempt to address this issue, a potential solution is to implement optical illusion, as proposed by Pan et al. (2022) [35]. A Computational Illusion Team (2013) has already introduced six optical illusions that can be used, such as: anamorphosis, image bump, changing the interval of repeated patterns, slanted lines inside lane lines, speed-reduction markers, and melody roads [36]. However, it should be noted that using optical illusions may only deceive the observer's eye and may not necessarily result in reduced crash severity. Future studies should investigate the cost-effectiveness of these devices in reducing crash severity.
However, there are several limitations to this study. Firstly, data on driver violations of the law were not included, which has been identified as a significant influencing factor in many previous studies. Because this study focuses solely on road structure and traffic control, if there is a correlation between these variables and driver violations, the obtained parameters may be biased. Secondly, the impact of guardrails on injury severity requires further analysis. Thirdly, machine learning, which does not assume a parametric distribution, has been used as an estimation method in recent years [37]. Finally, Behnood and Mannering (2015) indicate that although data from different years share some common features, the model specifications and estimated parameters are not temporally stable [38]. Therefore, it is challenging to compare the analysis models and verify the stability of the parameters.

Author Contributions

Conceptualization, M.T.; methodology, M.T.; software, Y.S.; validation, M.T.; formal analysis, M.T.; investigation, Y.S.; resources, M.T.; data curation, Y.S.; writing—original draft preparation, Y.S.; writing—review and editing, M.T.; visualization, Y.S.; supervision, M.T.; project administration, M.T.; funding acquisition, M.T.

Funding

This research received a grant from Chuo University Joint Research Grant.

Data Availability Statement

You can access the data from the following website. https://www.npa.go.jp/publications/statistics/koutsuu/opendata/index_opendata.html

Acknowledgments

In this section, you can acknowledge any support given which is not covered by the author contribution or funding sections. This may include administrative and technical support, or donations in kind (e.g., materials used for experiments).

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. George, Y.; Athanasios, T.; George, P. Investigation of road accident severity per vehicle type. Transportation Research Procedia 2017, 25, 2076–2083. [Google Scholar] [CrossRef]
  2. Wu, Q.; Chen, F.; Zhang, G.; Liu, X.C.; Wang, H.; Bogus, S.M. Mixed logit model-based driver injury severity investigations in single- and multi-vehicle crashes on rural two-lane highways. Accident Analysis & Prevention 2014, 72, 105–115. [Google Scholar] [CrossRef]
  3. Vajari, M.A.; Aghabayk, K.; Sadeghian, M.; Shiwakoti, N. A multinomial logit model of motorcycle crash severity at Australian intersections. Journal of Safety Research 2020, 73, 17–24. [Google Scholar] [CrossRef] [PubMed]
  4. Sharafeldin, M.; Farid, A.; Ksaibati, K. A Random Parameters Approach to Investigate Injury Severity of Two-Vehicle Crashes at Intersections. Sustainability 2022, 14, 13821. [Google Scholar] [CrossRef]
  5. Yuan, R.; Gan, J.; Peng, Z.; Xiang, Q. Injury severity analysis of two-vehicle crashes at unsignalized intersections using mixed logit models. International Journal of Injury Control Safety Promotion 2022, 29, 348–359. [Google Scholar] [CrossRef]
  6. Zhou, M.; Chin, C.H. Factors affecting the injury severity of out-of-control single-vehicle crashes in Singapore. Accident Analysis & Prevention 2019, 124, 104–112. [Google Scholar] [CrossRef]
  7. Khan, U.I.; Vachal, K. Factors affecting injury severity of single-vehicle rollover crashes in the United States. Traffic Injury Prevention, 2020, 21, 66–71. [Google Scholar] [CrossRef]
  8. Wang, X.; Adbel-Aty, M. Analysis of left-turn crash injury severity by conflicting pattern using partial proportional odds models. Accident Analysis & Prevention 2008, 40, 1674–1682. [Google Scholar] [CrossRef]
  9. Liu, P.; Fan, W.D. Exploring injury severity in head-on crashes using latent class clustering analysis and mixed logit model: A case study of North Carolina. Accident Analysis & Prevention 2020, 135, 105388. [Google Scholar] [CrossRef]
  10. Zhang, Z.; Yang, R.; Yuan, Y.; Blackwelder, G.; Yang, X.T. Examining driver injury severity in left-turn crashes using hierarchical ordered probit models. Traffic Injury Prevention 2021, 22, 57–62. [Google Scholar] [CrossRef]
  11. Yaman, T.T.; Bilgic, E.; Esen, M.F. Analysis of traffic accidents with fuzzy and crisp data mining techniques to identify factors affecting injury severity. Journal of Intelligent and Fuzzy Systems 2022, 42, 575–592. [Google Scholar] [CrossRef]
  12. Sharafeldin, M.; Farid, A.; Ksaibati, K. Injury Severity Analysis of Rear-End Crashes at Signalized Intersections. Sustainability 2022, 14, 13858. [Google Scholar] [CrossRef]
  13. Naik, B.; Tung, L.; Zhao, S.; Khattak, J.A. Weather impacts on single-vehicle truck crash injury severity. Journal of Safety Research 2016, 58, 57–65. [Google Scholar] [CrossRef] [PubMed]
  14. Li, N.; Park, B.B.; Lambert, H.J. Effect of guardrail on reducing fatal and severe injuries on freeways: Real-world crash data analysis and performance assessment. Journal of Transportation Safety & Security 2018, 10, 455–470. [Google Scholar] [CrossRef]
  15. Li, Z.; Ci, Y.; Chen, C.; Zhang, G.; Wu, Q.; Qian (Sean), Z.; Prevedouros, D.P.; Ma, T.D. Investigation of driver injury severities in rural single-vehicle crashes under rain conditions using mixed logit and latent class models. Accident Analysis & Prevention 2019, 124, 219–229. [Google Scholar] [CrossRef]
  16. Cai, Z.; Wei, F.; Wang, Z.; Guo, Y.; Chen, L.; Li, X. Modeling of Low Visibility-Related Rural Single-Vehicle Crashes Considering Unobserved Heterogeneity and Spatial Correlation. Sustainability, 2021, 13, 7438. [Google Scholar] [CrossRef]
  17. Mphekgwana, P.M. Influence of Environmental Factors on Injury Severity Using Ordered Logit Regression Model in Limpopo Province, South Africa. Journal of Environmental and Public Health 2022, 625168. [Google Scholar] [CrossRef]
  18. Zou, W.; Wang, X.K.; Zhang, D.P. Truck crash severity in New York City: An investigation of the spatial and the time of day effects. Accident Analysis & Prevention 2017, 99 Pt A, 2490261. [Google Scholar] [CrossRef]
  19. Agrawal, V.; Chatterjee, S.; Mitra, S. Crash Severity Analysis through nonparametric machine learning methods. Journal of Eastern Asia Society for Transportation Studies 2019, 13, 2614–29. [Google Scholar]
  20. Wahab, L.; Jiang, H. A multinomial logit analysis of factors associated with severity of motorcycle crashes in Ghana. Traffic Injury Prevention 2019, 20, 521–527. [Google Scholar] [CrossRef]
  21. Yang, J.; Ren, P.; Ando, R. Examining Drivers’ Injury Severity of Two-Vehicle Crashes between Passenger Cars and Trucks Considering Vehicle Types. Asian Transport Studies 2019, 5, 720–733. [Google Scholar]
  22. Champahom, T.; Wisutwattanasak, P.; Chanpariyavatevong, K.; Laddawan, N.; Jomnonkwao, S.; Ratanavaraha, V. Factors affecting severity of motorcycle accidents on Thailand's arterial roads: Multiple correspondence analysis and ordered logistics regression approaches. IATSS Research 2022, 46, 101–111. [Google Scholar] [CrossRef]
  23. Wu, Q.; Zhang, G.; Zhu, X.; Liu, C.X.; Tarefder, R. Analysis of driver injury severity in single-vehicle crashes on rural and urban roadways. Accident Analysis & Prevention 2016, 94, 35–45. [Google Scholar] [CrossRef]
  24. Rezapour, M.; Moomen, M.; Ksaibati, K. Ordered logistic models of influencing factors on injury severity of single and multiple-vehicle downgrade crashes: A case study in Wyoming. Journal of Safety Research 2019, 68, 107–118. [Google Scholar] [CrossRef]
  25. Ma, J.; Ren, G.; Li, H.; Wang, S.; Yu, J. Characterizing the differences of injury severity between single-vehicle and multi-vehicle crashes in China. Journal of Transportation Safety & Security 2023, 15, 314–334. [Google Scholar] [CrossRef]
  26. Li, Z.; Chen, C.; Wu, Q.; Zhang, G.; Liu, C.; Prevedouros, D.P.; Ma, T.D. Exploring driver injury severity patterns and causes in low visibility related single-vehicle crashes using a finite mixture random parameters model. Analytic Methods in Accident Research 2018, 20, 1–14. [Google Scholar] [CrossRef]
  27. Russo, J.B.; Savolainen TP, A. comparison of freeway median crash frequency, severity, and barrier strike outcomes by median barrier type. Accident Analysis & Prevention 2018, 117, 216–224. [Google Scholar] [CrossRef]
  28. Moran, M.A.; Rezapour, M.; Ksaibati, K. Modeling the impact of various variables on severity of crashes involving traffic barriers. Journal of Transportation Safety & Security 2020, 12, 800–817. [Google Scholar] [CrossRef]
  29. Bhuiyan, H.; Ara, J.; Hasib, K.M. Crash severity analysis and risk factors identification based on an alternate data source: a case study of developing country. Science Report 2022, 12, 21243. [Google Scholar] [CrossRef]
  30. Gong, H.; Fu, T.; Sun, Y.; Guo, Z.; Cong, L.; Hu, W.; Ling, Z. Two-vehicle driver-injury severity: A multivariate random parameters logit approach. Analytical Methods in Accident Research 2022, 33, 100190. [Google Scholar] [CrossRef]
  31. Kosmidis, I. Bias Reduction in Binomial-Response Generalized Linear Models, 2021. https://cran.r-project.org/web/packages/brglm/index.html (accessed 09 April 2023). 09 April.
  32. Firth, D. Bias reduction of maximum likelihood estimates. Biometrika, 1993, 80, 27–38. [Google Scholar] [CrossRef]
  33. Tanishita, M.; Sekiguchi, Y.; Sunaga, D. Impact analysis of road infrastructure and traffic control on severity of pedestrian–vehicle crashes at intersections and non-intersections using bias-reduced logistic regression. IATSS Research 2023. in Press. [Google Scholar] [CrossRef]
  34. Sekiguchi, Y.; Tanishita, M.; Sunaga, D. Characteristics of cyclist crashes using polytomous latent class analysis and bias-reduced logistic regression. Sustainability 2022, 14, 5497. [Google Scholar] [CrossRef]
  35. Pan, F.; Wu, Q.; Wang, Z.; Wang, L.; Zhang, L.; Li, M. Effectiveness Evaluation of Optical Illusion Deceleration Markings for a V-Shaped Undersea Tunnel Based on the Set Pair Analysis Method and the Technique for Order Preference by Similarity to Ideal Solution Theory. Transportation Research Record 2022, 0. [Google Scholar] [CrossRef]
  36. Computational Illusion Team (Team Leader, Kokichi Sugihara), Alliance for Breakthrough between Mathematics and Sciences, Japan Science and Technology Agency CREST Project. Optical Illusions on Roads and Measures for Their Reduction, 2013. http://compillusion.mims.meiji.ac.jp/pdf/roadillusions_eng.pdf (Access: 15 April, 2033). 1: (Access.
  37. Chiou, Y.C.; Fu, C.; Ke, C.Y. Modelling two-vehicle crash severity by generalized estimating equations. Accident Analysis & Prevention 2020, 148, 105841. [Google Scholar] [CrossRef]
  38. Behnood, A.; Mannering, L.F. The temporal stability of factors affecting driver-injury severities in single-vehicle crashes: Some empirical evidence. Analytic Methods in Accident Research 2015, 8, 7–32. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.

Downloads

84

Views

34

Comments

0

Subscription

Notify me about updates to this article or when a peer-reviewed version is published.

Email

Prerpints.org logo

Preprints.org is a free preprint server supported by MDPI in Basel, Switzerland.

facebook logotwitter logolinkedin logo
MDPI Initiatives
Important Links
Subscribe

Disclaimer

Terms of Use

Privacy Policy

© 2025 MDPI (Basel, Switzerland) unless otherwise stated