Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/hindawi-publishing-corporation/modeling-the-motorcycle-crash-severity-on-nonintersection-urban-CBgvqgtEG4
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- Publisher
- Hindawi Publishing Corporation
- ISSN
- 0197-6729
- eISSN
- 2042-3195
- DOI
- 10.1155/2023/2250590
- Publisher site
- See Article on Publisher Site
Abstract
Hindawi Journal of Advanced Transportation Volume 2023, Article ID 2250590, 12 pages https://doi.org/10.1155/2023/2250590 Research Article Modeling the Motorcycle Crash Severity on Nonintersection Urban Roadways in the Australian State of Victoria Using a Random Parameters Logit Model 1 1 2 Mohammad Ali Seyfi , Kayvan Aghabayk , Amir Mohammad Karimi Mamaghan , and Nirajan Shiwakoti School of Civil Engineering, College of Engineering, University of Tehran, Tehran, Iran School of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, Stockholm, Sweden School of Engineering, RMIT University, Melbourne, Australia Correspondence should be addressed to Nirajan Shiwakoti; nirajan.shiwakoti@rmit.edu.au Received 7 November 2022; Revised 30 April 2023; Accepted 12 May 2023; Published 25 May 2023 Academic Editor: Yanyong Guo Copyright © 2023 Mohammad Ali Seyf et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Due to a lack of physical protection and balance, motorcycle riders are one of the most vulnerable road users and are more likely to sufer severe injuries than motorists. Between 2009 and 2020, about 60% of motorcycle crashes occurred on nonintersection urban roadways in Victoria, Australia. While considerable research on intersections and their infuence on the severity of motorcycle crashes has been conducted, there are limited studies on motorcycle crashes on nonintersection roadways. Since gathering all information from every motorcycle crash may not be possible, heterogeneity can arise from unobserved factors and cause problems in developing reliable crash severity models. Terefore, this study aims to investigate the factors contributing to motorcycle crash severity on Victorian nonintersection urban roadways while considering the heterogeneity of factors. A total of 10,897 nonintersection motorcycles crash data from the beginning of 2009 to November 2020 in the State of Victoria, Australia, were analyzed. A random parameters (mixed) logit model (RPL) was used for evaluating motorcycle crashes. Te severity of motorcycle crashes was divided into three categories: fatal injury, serious injury, and minor injury. Also, marginal efects were calculated to see how each parameter estimate afects crash severity outcomes. Te RPL model results showed that some factors increased the likelihood of fatal injuries. Tese factors included not wearing a helmet, being in the older rider age group, riding during the early morning or midnight hours, weekend motorcycle use, riding in the early morning or midnight hours (00:00–6:29 A.M), and insufcient lighting (dark and dusk/dawn). Also, the following factors enhanced the probability of serious injuries: having a pillion passenger, having a motorcycle age of more than 7years, riding at higher speed limits (more than 50km/h) or during peak hours in the morning (6:30−8:59 A.M), and being in the younger age group (less than 26years old). Te fndings from this study are valuable resources for road safety policy managers to develop efective strategies for improving motorcyclists’ safety at nonintersections. Tis may include improving the light conditions at nonintersection, encouraging the motorcyclist to maintain motorcycles regularly, and educating the motorcyclist to wear a helmet, avoid distractions, and ride responsibly on the weekends. motorcyclists, cyclists, and pedestrians [1]. Road trafc crash 1. Introduction fatalities are forecasted to be the third leading cause of Trafc crashes have caused notable fnancial, physical, and mortality by 2030 [2]. Te safety of vulnerable road users has emotional pain for families and society. Every year, nearly received more attention recently and is seen as an important 1.3million people are killed and 20 to 50million people are social concern. As a result, considerable eforts are needed to injured in road crashes worldwide. About 0.65million of improve the safety of vulnerable road users, including those fatalities are vulnerable road users, including motorcycles [3]. 2 Journal of Advanced Transportation variables. Te authors in [23] examined motorcycle crashes Motorcycles are less expensive vehicles to acquire and ride than others, especially when fuel costs are high [4]. at intersections in the State of Victoria between 2006 and 2018, taking into account the characteristics of diferent Further, motorcycles’ mobility benefts, like their small size [5], make them a practical choice in congested urban areas types of intersections. Various factors were considered in the with limited space. Teir popularity as recreational vehicles study: motorcyclist age and sex, helmet use, certifcate of [6] and easy maneuverability [7] have made them an eco- status, weather, pavement conditions, lighting conditions, nomical and afordable mode of transport. In recent years, time of the crash, day of the crash, single-vehicle or mul- demand for motorcycles has increased in Australia and its tivehicle type crash, time of the crash, and impact and southeastern State of Victoria. Indeed, motorcycle and condition of the asphalt pavement. By using the multinomial logit model, the study concluded that motorcyclists with an scooter sales in Australia climbed by 6.2% in 2020, reaching 87.453 units, the largest in the previous fve years. Te age over 55, morning rush hour crashes, weekend crashes, midnight/early morning crashes, multivehicle crashes, t- upward trend continued in 2021 [8]. Motorcyclists are considered vulnerable road users, and intersections, crashes in towns and rural areas, stop or give-way intersections, roundabouts, and uncontrolled in- they are more prone to fatal or serious injury due to a lack of physical protection for the rider [9] and less conspicuity [10] tersections increase the risk of fatal injury. In contrast, it was than other motorized vehicles. Considering per vehicle mile observed that female motorcyclists, snowy/stormy/foggy or traveled in 2020, motorcyclists were approximately 28times wet weather, nighttime rush hours, and unpaved roads all more likely than passenger car occupants to die in a motor reduced the risk of fatal motorcycle crashes and lowered vehicle crash and were four times more likely to be mortality. While considerable research on intersections and injured [11]. their infuence on the severity of motorcycle crashes has been conducted [24–26], there have been fewer studies on In 2019, an average of 10% of persons in the State of Victoria, Australia, held a motorcycle licence, while mo- motorcycle crashes on nonintersection roadways. Paying more attention to crashes at intersections was not for torcyclists or pillion passengers were involved in around 17% of fatal crashes. In addition, compared to 2018, there motorcycles only. According to a study that investigated intersection and nonintersection crashes [27], 26% of private was a 25% rise in crashes and fatalities in 2019. However, the increasing trend in the number of crashes in 2020 compared car intersection crashes were found to be severe, compared to 2019 decreased (it may be due to Victoria’s strict to 58.7% of nonintersection crashes. Also [28], claimed that COVID-19 lockdown in 2020), but it continued to increase the fatality rate of crashes that happened at nonintersection in 2021 compared to 2020. Motorcycle crashes increased was substantially higher than the rate at the intersection. It is faster than all other road users in this state from 2020 to to be noted that about 60% of urban motorcycle crashes 2021 [8]. between 2009 and 2021 in the State of Victoria occurred in nonintersections [29]. One of the challenges in the trafc safety domain is to minimize the number of fatalities and injuries and lessen Some of the factors infuencing the likelihood of a crash and the severity of the resulting injury may not be available crash severity levels. Several modeling approaches have been used in recent decades to predict the severity of motorcyclist in a crash database. Tese variables (which comprise un- injuries. Te majority of the literature looked at the severity observed heterogeneity) can cause variance in the infuence of the crash by categorizing the causes into the follows: of observable variables on crash risk and severity [30]. As human factors such as the rider’s sociodemographic char- a result, a thorough investigation of the variables infuencing acteristics, vehicle features such as motorcycle age or size, the severity of motorcycle crashes, considering the hetero- environmental conditions such as weather conditions, geneity of the parameters, is required. roadway characteristics such as the width of the lane or Tis paper’s main contribution lies in investigating the median, and crash attribute such as the type of collision factors that contribute to the severity of motorcycle crashes on nonintersection urban roadways in Victoria while con- [12, 13]. Te following factors were discovered to be linked with sidering the heterogeneity of these factors. Non-intersection crashes refer to those that occur on any segment of motorcyclist injury severity in past studies: not wearing a helmet [14–17]; riding in a high-speed limit zone [14–16]; a roadway or transportation facility that is not located at an age of motorcyclist [16, 17]; sex [12]; light condition [16, 18]; intersection. Te study’s focus on considering the impact of licence status [19]; intersection characteristics [20]; function heterogeneity by employing the random parameters logit of roadway [16, 21]; road geometry [22]; and presence of model sets it apart from other works in the feld. Neglecting a pillion passenger [21]. to account for data heterogeneity in crash severity analysis Studies have yielded diferent outcomes regarding the can lead to biased outcomes, fawed conclusions, inaccurate safety policies, forecasting errors, and wastage of resources importance of various contributing factors afecting a partic- ular motorcycle crash type. Te variances in driver behaviors, [31]. Terefore, the paper’s contributions to the literature include exploring the infuence of nonintersection urban vehicle features, and trafc and road characteristics have been identifed as reasons for these discrepancies. Further, the sta- factors on motorcycle crash severity and considering het- erogeneity in the analysis to get accurate results. Several tistical procedures employed difer in various studies [9]. Some research focused solely on intersections, while studies [32, 33] have demonstrated the strengths and ac- others employed statistical models of discrete choice and curacy of random parameter models in trafc crash analysis machine learning models to investigate the contributing compared to the fxed-parameter model. Journal of Advanced Transportation 3 a result, using an ordered model for all ordered data is not 2. Methodology recommended [40]. To address the issue of within-crash correlations and quantify uncertainty in the data, the authors Te severity of crashes generally has discrete results. Terefore, statistical analysis, like discrete choice and in [26, 41] employed Bayesian models to analyze data and econometric models, as well as conventional and practical identify factors contributing to motorcycle crashes. Te machine learning methods [34], has been used to solve this MNL model, on the other hand, is prone to violating the problem. Among the discrete choice models, many studies independence of irrelevant alternatives (IIA) criterion (the have investigated the crash severity with various nominal or standard multinomial logit model, which assumes that the ordinal and ordered or unordered models, including probit error terms (ε ) are independently distributed among al- ij and logit models [35]. ternate outcomes) [16]. Te authors [9] estimated a nested Choosing an analytical model has always been difcult logit (NL) model and heterogeneous choice models to for trafc safety researchers since each model has benefts overcome these restrictions. Also, the authors in [42] stated and drawbacks, requiring the researcher’s ultimate decision that the NL model is not as fexible as heterogeneous models to be based on the dataset’s features [35]. For example, [36] such as the RPL model and that the NL model is more found that when deciding between the ordered probit, sensitive to the problem of data underreporting. random parameter logit (RPL), and multinomial logit In addition to the models discussed earlier, other less (MNL) models, the RPL model is preferable when the commonly utilized nonheterogeneity models exist, including sample size is large. Furthermore, RPL models outperformed ordered probit [43, 44], binary logit [45], and data mining MNL in terms of statistical performance with the same techniques [46–48]. Tese models ofer alternative ap- sample size [37]. Traditional crash severity studies are based proaches to analyze and interpret data related to motorcylist on information gathered after a crash. In many aspects, this crashes, with each having its own advantages and limitations. is quite restrictive. First, many minor crashes are not For instance, ordered probit models can account for the recorded, resulting in the loss of potentially crucial data and ordinal nature of injury severity outcomes, while binary logit biasing the data towards the more recorded levels. Second, models are useful when predicting the probability of a crash many signifcant factors that afect crash occurrence and resulting in an injury or noninjury. Data mining techniques severity (for example, vehicle speed, driver braking and can also be employed to identify patterns and associations maneuvering responses, etc.) are not recorded. Tis results within large crash datasets. As such, researchers may consider in signifcant unobserved heterogeneity that afects the these alternative models when exploring the complexity of model and prevents important information from being used motorcylist crash-related data. to make signifcant new inferences. Tird, police-reported Random parameters models, also known as random ef- injury severity indicators (no injury, possible injury, obvious fects models, are a useful tool for modeling unobserved injury, crippling injury, fatality) are based on observations at heterogeneity in crash data. Unlike fxed models that assume the crash scene and are subject to change until a medical constant parameter estimates across all observations, RPL diagnosis is completed [38]. So, several unobserved factors models allow for variation in the efects of parameters across infuence the severity of individual crashes, and acquiring diferent observations. As a result, RPL models have been comprehensive data on each crash can be difcult or im- found to have better model ft and more accurate predictions possible. By avoiding those unobserved factors (also known than fxed models. For example, the full Bayesian random as unobserved heterogeneity), it may result in erroneous parameters logistic regression approach [49] and the random parameter estimations and conclusions [30]. parameters multivariate Tobit model [41] have both out- Te MNL model was used by [15], who argued that MNL performed fxed models. Another advantage of heterogeneity is an efective method for studying variables that contribute models like RPL models is that they are not afected by the to the severity of motorcycling injuries. Since crash severity independence of irrelevant alternatives problem, which can be is frequently defned by natural ordering (from low to high a signifcant issue in fxed models such as the MNL model severity), an ordered probability model, such as an ordered [50]. Tis problem arises when adding an irrelevant alter- logit or ordered probit, may be suitable for analysis [39]. To native to a choice set that afects the probability of choosing account for the ordinal nature of the severity outcomes, one of the other alternatives. In the context of crash data, this researchers like [21] employed an ordered probit model to could occur if a crash’s outcome depends not only on the analyze injury severity and motorcycle damage severity in characteristics of the crash but also on the presence of other motorcycle crashes. Te ordered models, however, have nearby crashes. Since RPL models allow for individual- some drawbacks: Te frst issue is the underreporting of low- specifc variations in parameters, they can account for such severity crashes like minor injury or property damage only unobserved heterogeneity and provide more accurate esti- (which is common in crash reports) which can lead to mates of the efects of diferent factors on crash outcomes. inaccuracies in parameter estimates or in ordered models. Other models have also been proposed that address the Te second issue is that ordered models either raise the problem of unobserved heterogeneity in common discrete probability of fatality (and hence reduce the probability of choice models for predicting the severity of crashes [51]. minor injury) or reduce the probability of fatality (and In recent years, heterogeneous studies on crash injury subsequently increase the probability of minor injury). Tese severity analysis of various road users have been employed: models also can not consider a feature that raises or reduces RPL (also known as the mixed logit model) [51–55], random both fatality and minor injury simultaneously [16]. As parameters ordered probability models [56–58], latent-class 4 Journal of Advanced Transportation models [14, 51, 59], Markov switching models with random EXPβ X j ij parameters [60], and bivariate/multivariate models with P (j) � . (2) EXPβ X ∀J j ij random parameters [61]. However, directly comparing the performance of these By allowing β to vary across observed crash dataset models using numerical measures such as the likelihood features, the RPL model formulation addresses the un- ratio test is not suitable, as it fails to capture the complexity observed heterogeneity issue. Te outcome constants and β and subtleties of heterogeneity modeling in this domain. parts can be fxed or randomly distributed with fxed means Terefore, researchers typically rely on theoretical insights across all parameters. As a result, a mixing distribution is and empirical evidence to evaluate the efectiveness of dif- used to calculate crash severity probability [65]. ferent heterogeneity models for crash data analysis. Further, latent class models and Markov switching models are limited EXPβ X j ij P (j|φ) � ⎰ fβ φdβ . (3) in their ability to extend beyond classes. In addition, when i j j EXPβ X ∀i j ij comparing RPL with latent class models, latent class models will eventually require a probability model such as MNL or Here P (j|φ) is the probability of injury severity j even RPL to determine their probability [30]. Based on [51], conditional on f(β |φ) and f(β |φ) is the density function j j the RPL model prediction of probabilities for all three levels of β and φ is a known vector of parameters (mean and of injury severity was more ft to the data (on average) than variance) that defne the density function. X is avector of ij the latent class model prediction of probabilities. Eventually, observable features (such as crash/roadway/environment/ our choice between heterogeneity models was RPL or mixed driver/vehicle-specifc parameters) that infuence the injury logit models. Also, the RPL method has been employed more result j for crash i. RPL probabilities are a weighted average than others in prior research to address the problem of of various values ofβ across crashes, with some components unobserved heterogeneity in the severity of crashes among of the vector β will be fxed and others being randomly diferent road users [30, 51]. distributed. If the parameters are random, the density Te severity of the crashes was divided into 3 categories: function f(β |φ) generates the RPL weights. Normal, log- minor injury (bruising, contusions, unconsciousness, dis- normal, triangular, and uniform distributions are consid- comfort, or complaints of pain soreness), serious injury (i.e., ered for the functional form of the parameter density hospitalization), and fatal injury (i.e., killed in crash place or functions. Tere is no mathematical proof that one distri- death within 30days after crash). Te authors in [62] pointed bution is better than the other. Te statistical ft of the model out around 30years ago that not all trafc crashes were can be used to make the fnal choice of the random pa- reportable and that not all reportable crashes were really rameter [12]. Te current study examined all four distri- recorded. Property damage only (PDO) crashes are more butions for random parameters, and the normal distribution likely to be unreported (according to [63], about half of them had a higher log-likelihood than the others. Also, according are likely to be unreported). As they constitute a small to past studies, the normal distribution provided the best ft proportion of the data, they were not chosen to give for crash injury severity data [51, 53, 54]. a severity level. Using the maximum likelihood method and 500 Halton RPL is an advanced statistical model that builds upon draws [66], the RPL model is calculated in the NLOGIT MNL, a traditional modeling technique. RPL ofers the software. Te marginal efects are also computed to make the advantage of incorporating a distribution function for the estimated results easier to comprehend. Te marginal efects coefcients that describe the impact of unobserved factors. are the diferences in estimated probability when indicator Tis means that RPL can provide a more nuanced and variables are changed from zero to one. unbiased analysis than MNL. In contrast, MNL is limited in P (j) that its β coefcients always have constant values, and it E � P j|X � 1 − P j|X � 0 . (4) X i ik i ik ik cannot address heterogeneous efects. Terefore, RPL is an ideal approach for researchers and analysts who need to We reviewed past studies on crash severity heterogeneity to determine the randomness of the parameters. Te basis account for the complex interplay of various factors and their efects on outcomes. By using RPL, they can gain for decision-making was the signifcance of the standard a more accurate understanding of the underlying patterns deviation of each random parameter [67]. and make informed decisions based on the insights gained from their analyses. 3. Data Te RPL model utilized in this research is shown as follows. After preprocessing the motorcycle crash data in the State of Te severity function indicates that each individual crash Victoria obtained from VicRoads [29], 10,897 data related i refects injury severity j as follows: only to urban nonintersection motorcyclist crashes (not pillion passenger information) from the beginning of 2009 to November 2020 were selected for analysis. Note that Vic- U � β X + ε . (1) ij j ij ij Roads is a statutory corporation in the State of Victoria, Te approach proposed in [64] is used to calculate the Australia, responsible for driver licencing and vehicle reg- probability of a crash i resulting in driver injury severity j istration and is a part of the Department of Transport, given as follows: Victoria. VicRoads provides crash statistics to researchers Journal of Advanced Transportation 5 3.1.4. Helmet Use. Wearing a helmet has always been an and other stakeholders to assist with education and research and develop road safety programs and initiatives. efective way to reduce the severity of motorcycle crashes. When an appropriate protective helmet was used, the usage Te data were divided into three groups: fatal injury, serious injury, and minor injury. Crashes that occurred in of the helmet was classifed as “Yes”; when the motorcyclist the intersection-afected region, which had previously been did not wear a helmet, it was categorized as “No.” analyzed, were then removed. Te quantity and percentage of the variables that may impact the crash severity are 3.2. Vehicle Factor depicted in Table 1 based on the level of injury sufered by the motorcyclists. 3.2.1. Motorcycle’s Age. The diference between the date of Some important characteristics, such as motorcycle type the crash and the date of manufacturing of the motorcycle was or engine size, alcohol-impaired riding, licence status, trip used to calculate the motorcycle age. Based on the previous purpose, and annual average daily trafc (AADT), which research on the Victorian motorcycles [69], motorcycles had have been demonstrated to infuence crash severity in earlier a median age of 5years and a mean age of 7years. In addition, research, were not examined in this analysis. Tis was due to our preprocessed data showed that the mean age of motor- a database absence or missing data. cycles was 6.84years. So, we divided vehicle age into two Te correlation between each characteristic was frst categories: under 7years and more than 7years. determined, and among those with a clear connection, one was kept while the others were eliminated. Te analysis was 3.3. Temporal Factors performed using the following features: 3.3.1. Day of the Week. We divided the days of the week into weekdays and weekends [70]. Even if there are no features 3.1. Human Factors specifcally relevant to the trip’s purpose, the day of the week could be relevant. We consider it as a heterogeneous feature, 3.1.1. Sex. We categorized it into two groups: male and since there are behavioral diferences between recreational female after excluding information on motorcyclists with riding and commuting or working purposes. unknown or missing gender data . Although there are clear physiological diferences between male and female (justi- fying the use of an indicator variable like 1 for females and 3.3.2. Time of the Day. Based on earlier research [68], we 0 for males), there is also a great variation in the same classifed the time period of the crash into fve groups based gender, such as diferences in height, weight, bone density, on the volume of trafc: early morning or midnight hours (00: levels of caution, and other factors that researchers are 00–6:29 A.M), morning peak hours (6:30−8:59 A.M), day of- usually unaware of [30]. Te impact of rider sex on crash peak hours (9:00 A.M–14:59 P.M), evening peak hours (15: data has been considered a heterogeneous variable, as the 00–18:29 P.M), and night of-peak hours (18:30− 23:59 P.M). efect of the same gender on crash outcomes may vary across diferent crashes. 3.4. Environmental and Roadway Factors 3.4.1. Light Condition. We classifed light conditions into 3.1.2. Age Group. We divided the age group into three three categories based on the amount of lightness available [71]: categories, as in previous classifcations [40, 68]: under 1-light (daytime or nighttime with sufcient light in the crash 26years old, between 26 and 59years old, and above 59years area) 2-dark (a nighttime crash at a location with insufcient old. Age is linked to a person’s physical characteristics, as lighting or when the lights were turned of at the time of the well as their reaction times, risk-taking behavior, and other crash) and 3-dusk/dawn. Because the time of day was classifed factors that may impact the severity of an injury. However, based on trafc volume, there was no high (5.66%) correlation because age is only a proxy for these elements (which re- between the time of day and light conditions. searchers are unable to see and measure), the impact of age on injury severity may difer among people of the same age, 3.4.2. Road Surface Condition. We divided road surface as age is frequently used as an age group indication variable [38]. Considering that, despite the obvious diferences be- conditions into two categories: normal (dry) and slippery (wet/snowy/icy/muddy) [72, 73]. tween young and old, a middle-aged person might still exhibit youthful capabilities with good nutrition and exer- cise. Alternatively, a young person may not be energetic 3.4.3. Road Surface Type. Weconsideredthe roadsurface type because of factors such as sickness or inactivity. It has been as paved and unpaved (including gravel roads) surfaces [68]. tried to be viewed as a heterogeneous variable as a consequence. 3.5. Crash Factors 3.5.1. Speed Limit Zones. Te authors in [69] reported that 3.1.3. Having a Pillion Passenger. Motorcycle crashes with at least one pillion passenger and motorcycle crashes without Victorian motorcycle riders are more likely than drivers of a pillion passenger have been separated into two all other types of vehicles to exceed the legal speed limit, categories [21]. particularly in areas where the speed limit is greater than 6 Journal of Advanced Transportation Table 1: Descriptive statistics of independent factors. Variables Fatal injury Serious injury Minor injury Total Sex Male 322 (3.22%) 4734 (47.37%) 4936 (49.39%) 9992 (91.69%) Female 14 (1.54%) 387 (42.76%) 504 (55.69%) 905 (8.30%) Age group <26years old 62 (2.21%) 1247 (44.48%) 1494 (53.30%) 2803 (25.72%) 26–59years old 237 (3.23%) 3454 (47.17%) 3631 (49.59%) 7322 (67.19%) >59years old 37 (4.79%) 420 (54.40%) 315 (40.80%) 772 (7.08%) Helmet use Yes 312 (2.98%) 4877 (46.63%) 5269 (50.38%) 10458 (95.97%) No 24 (5.46%) 244 (55.58%) 171 (38.95%) 439 (4.02%) Speed limit zones ≤50km/h 39 (1.83%) 864 (40.56%) 1227 (57.60%) 2130 (19.54%) >50km/h 297 (3.38%) 4257 (48.55%) 4213 (48.05%) 8767 (80.45%) Motorcycle’s age ≤7years old 173 (2.56%) 3112 (46.06%) 3471 (51.37%) 4141 (61.99%) >7years old 163 (3.93%) 2009 (48.51%) 1969 (47.54%) 6756 (38.00%) Day of the week Weekdays 175 (2.63%) 3014 (45.30%) 3464 (52.06%) 6653 (61.05%) Weekends 161 (3.79%) 2107 (49.64%) 1976 (46.55%) 4244 (38.94%) Time of day 00:00–6:29 A.M 34 (7.02%) 223 (46.07%) 227 (46.90%) 484 (4.44%) 6:30−8:59 A.M 24 (2.42%) 407 (41.11%) 559 (56.46%) 990 (9.08%) 9:00 A.M–14:59 P.M 154 (3.14%) 2278 (46.55%) 2461 (50.29%) 4893 (44.90%) 15:00–18:29 P.M 82 (2.40%) 1638 (48.10%) 1685 (49.48%) 3405 (31.24%) 18:30−23:59 P.M 42 (3.73%) 575 (51.11%) 508 (45.15%) 1125 (10.32%) Having a pillion passenger Yes 30 (5.85%) 267 (52.14%) 215 (41.99%) 512 (4.69%) No 306 (2.94%) 4854 (46.74%) 5225 (50.31%) 10358 (95.31%) Light condition Light 291 (3.03%) 4468 (46.65%) 4818 (50.30%) 9577 (87.88%) Dark 28 (5.87%) 250 (52.41%) 199 (41.71%) 477 (4.37%) Dusk/dawn 17 (2.01%) 403 (47.80%) 423 (50.17%) 843 (7.73%) Road surface condition Dry 315 (3.41%) 4447 (48.26%) 4451 (48.31%) 9213 (84.54%) Slippery 21 (1.24%) 674 (40.02%) 989 (58.72%) 1684 (15.45%) Road surface type Paved 319 (3.45%) 4365 (47.21%) 4561 (49.33%) 9245 (84.83%) Unpaved 17 (1.02%) 756 (45.76%) 897 (53.20%) 1652 (15.16%) Number of vehicles Single-vehicle crash 118 (1.95%) 2884 (47.82%) 3028 (50.21%) 6030 (55.33%) Multivehicle crash 218 (4.47%) 2237 (45.96%) 2412 (49.55%) 4867 (44.66%) Te bold values specifcally represent the total percentages that have been appropriately mentioned in the third line of the results and discussion section. 50km/h. We divided speed limits into two categories: up to total percentages are bolded in Table 1) was chosen as the 50km/h (including 50km/h) and above 50km/h, as other reference. Te estimation results of the RPL model for Victorian crash severity studies suggested [68, 74]. motorcycle crashes on urban nonintersection Victorian roadways are shown in Table 2. Te calculated coefcients, chi-square, log-likelihood, McFadden R-squared, marginal 3.5.2. Number of Vehicles. Tis variable is classifed into two efects, and p values are shown in this table. For each injury types based on the number of vehicles involved: single- level, the variable’s coefcient was calculated using the vehicle crashes and multiple-vehicle crashes [23]. maximum likelihood method. As indicated in the table, the fnal model’s outcome shows a better statistical ft than the 4. Results and Discussion initial testing models (log-likelihood � −7632.736, chi- Logistic regression typically chooses the reference category squared �6034.515, McFadden pseudo-R-squared �0.378). for features and severities based on the quantity of each According to [75], McFadden’s pseudo-R-squared value category and prefers it to be the normative category. between 0.2 and 0.4 indicates an excellent ft. Te Terefore, the most frequent value in each variable (whose marginal efects in the mentioned categories show the Journal of Advanced Transportation 7 Table 2: Te RPL model’s output for motorcycle crashes on urban nonintersection roadways in Victoria. Marginal efects (multiplied by a factor of 100) Variable Coefcient Standard error p value Fatal injury (FI) Serious injury (SI) Minor injury (MI) Fatal injury (FI) ∗∗∗ Constant 4.288 0.61706 <0.001 ∗∗∗ Slippery road −2.794 0.39521 <0.001 −33.64 31.62 2.02 ∗∗ No helmet use 1.802 0.53450 0.031 32.94 31.37 1.56 ∗∗∗ Female rider −1.019 0.27028 <0.001 −1.37 13.07 8.30 ∗∗ Standard deviation for “female rider” (normally distributed) 1.410 0.57571 0.0143 ∗∗∗ Weekend trip 3.617 1.14493 <0.001 10.34 −6.98 −3.36 ∗∗∗ Standard deviation for “weekend trip” (Normally distributed) 4.297 1.04271 <0.001 ∗∗ Unpaved road −0.493 0.20910 0.0183 −6.27 4.74 1.52 ∗∗∗ Dark light condition 1.314 0.32611 <0.001 10.55 −5.02 −5.52 ∗∗ Dusk/dawn light condition 0.644 0.26263 0.0141 3.52 −1.74 −1.77 ∗∗∗ Rider’s age above 59years (old rider) 2.517 0.23892 <0.001 7.35 −5.68 −1.66 Standard deviation for “old rider” (normally distributed) 1.485 0.86342 0.0854 ∗∗ Early morning or midnight hours (00:00–6:29 A.M) 0.269 0.13328 0.0430 12.03 −0.88 −11.15 Serious injury (SI) ∗∗∗ Constant 1.147 0.32622 <0.001 ∗∗∗ Pillion passenger presence 1.230 0.29835 <0.001 −9.25 14.20 −4.95 ∗∗ Motorcycle’s age above 7years 1.214 0.71447 0.0291 −4.14 13.85 −9.71 ∗∗∗ Speed limit above 50km/h 2.183 0.46550 <0.001 −3.82 4.14 −0.32 ∗∗ Morning peak hours (6:30−8:59 A.M) 0.299 0.15088 0.0471 −3.55 3.86 −0.30 ∗∗∗ Rider’s age under 26years (young rider) −0.504 0.18809 0.0078 2.16 −2.27 0.10 Standard deviation for “young rider” (normally distributed) 0.85865 0.46907 0.0672 ∗∗∗ ∗∗ ∗ , , and ⇒signifcance at 1%, 5%, and 10% level. RPL model statistics: log-likelihood function � −7632.73696. McFadden pseudo-R-squared �0.37899. Chi-squared �6034.51582. Te bold values of the marginal efects at a specifc level of severity indicate their signifcance solely at that particular level. 8 Journal of Advanced Transportation that age-related alterations in body structure can result in impact of switching from the reference category (normal and maximum) to the mentioned category at each injury level. a heightened risk of injury severity for older people. For example, the fatal injury marginal efect of the female rider indicator � −21.37% means if the rider’s sex switches 4.1.3. Day of the Week. Furthermore, when engaged in from male to female, the probability of a fatal injury reduces weekend crashes (Saturday and Sunday), motorcyclists were by 21.37%. shown to have a larger risk of fatal injury (fatal injury Te estimation fndings reveal that the following factors marginal efect �10.34%), and it was signifcant with a p value increase the probability of a fatal injury: not wearing a helmet, <0.001. Tis conclusion is consistent with previous studies old riders (more than 59years old), riding during early or [14, 80]. Tis might be because motorcyclists prefer to take midnight hours (00:00–6:29 A.M), weekend motorcycle use, recreational road trips on weekends (because of the holidays) and insufcient lighting (dark and dusk/dawn). And having when trafc fow is low; so motorcycle riders may engage in a pillion passenger, a motorcycle age of more than 7years, risky behaviors such as drifting, careless riding, and speeding, riding at higher speed limits (more than 50km/h) or during thereby increasing the likelihood of fatal injuries [81]. peak hours in the morning (6:30−8:59 A.M), and being in the younger age group (less than 26years old), increase the 4.2. Fixed Variables probability of a serious injury. Our conclusions are based on the features that are present in the dataset. Te other im- 4.2.1. Helmet Use. Motorcyclistswho donot weara helmetare portant factors like motorcyclist experience or licence status 32.94% (fatal injury marginal efect in the not wearing helmet were not available in the dataset. As such, we recommend category) more likely to be in fatal crashes. Tis observation is considering these factors (if known) in future studies. consistent with earlier research fndings, where it has been found that wearing a safety helmet reduces the chance of fatal and incapacitating injuries in motorcycle crashes [18, 82, 83]. 4.1. Random Variables. According to the RPL model results and the signifcance of standard deviations, the following Helmet use has been shown to minimize the fatality risk of motorcyclists by 2.12 times [84]. Among the several variables variables were chosen as heterogeneous variables. reviewed in this article, not wearing a helmet seems to be the leading cause of the increasing probability of fatality. 4.1.1. Sex. According to the RPL model’s coefcients and p values (Table 2), the female rider indicator is signifcant in fatal 4.2.2. Speed Limit Zones. As reported by [69], motorcycles injury. Table 2 also shows the marginal efects of the female were statistically more likely than other vehicles to exceed rider indicator. Based on the results of the marginal efects, the speed limit by more than 10km/h in speed limit zones of female motorcyclists had a lower risk of fatal injury (fatal injury 60, 80, 90, and 100km/h, but not in speed limit zones of 40 marginal efect � −21.37%). In previous studies, it has been and 50km/h, in the State of Victoria. According to our found that male motorcyclists are more likely to die or have research results, speed limit zones above 50km/h were serious injuries in motorcycle crashes [24, 76]. Tese results signifcant with a 99% confdence interval in the serious may originate from male motorcyclists being more likely to injury level, and serious injury probability increased by ride at faster speeds, engage in unsafe driving behaviors, and 4.14% in this category. Our fnding is in line with prior disregard trafc laws [77]. However, contrary to the afore- research fndings that concluded that in motorcycle crashes, mentioned fndings, [78] revealed a reduction in the likelihood riding at higher speed limits causes severe injury levels of severe injury for male riders involved in motorcycle crashes. [14, 15]. 4.1.2. Age Group. With a 99% confdence interval, the 4.2.3. Motorcycle Age. According to the RPL model results, younger age group (under 26years old) is signifcant in the age of the motorcycle indicator appeared to be signif- serious injury level, and the older age group (above 59years cant, with at least a 95% confdence interval. Riding on old) is signifcant in fatal injury level. Te results also in- a motorcycle that is more than 7years old raises the chances dicate that motorcyclists above the age of 59 (fatal injury of serious injury by 13.85%. Tis result is similar to [16], marginal efect �7.35) are more likely to experience a fatal which found that newer motorcycles (5years) are less likely injury. And motorcyclist under the age of 26years (serious to be involved in serious crashes, and each 1% rise in injury marginal efect � −2.27%) are less likely to be involved motorcycle age translates into a 0.6% drop in the probability in a serious injury crash and, therefore, more likely to ex- of a no-injury crash (and thus a higher likelihood of other perience a fatal or minor injury. Our fnding is in line with crash-injury-severity types). most past studies that concluded that injury severity in motorcycle crashes increases with increase in the rider’s age [15, 17, 18]. According to research conducted by [79], as the 4.2.4. Time of Day. Early morning or midnight hours (00: body ages, there is a decline in bone density and strength, 00–6:29 A.M) and morning peak hours (6:30−8:59 A.M) were statistically signifcant in diferent severities, with a 95% modifcations in the distribution of subcutaneous and vis- ceral fat, and a decrease in the fexibility of the chest wall. confdence interval. Early morning or midnight hours (00: Tese changes can lead to more severe injuries following 00–6:29 A.M) raised the risk of fatal injury by 12.03. It was exposure to trauma. Moreover, the study by [39] highlighted also observed that there is a 3.86% increase in the risk of Journal of Advanced Transportation 9 serious injury during the next time interval (6:30−8:59 A.M). parameters. To achieve this aim, 10,897 data on motorcycle Tere is a diference in the outcome of this variable. Based on crashes that occurred at urban nonintersection in the State of most previous studies, riding a motorcycle in the early Victoria, Australia, were evaluated using the RPL model morning hours raises the risk of fatality and incapacitating from 2009 to November 2020. injuries [21, 24, 25]. Te authors in [85] also concluded that Motorcycle crashes were classifed as having three levels of-peak hours increase fatality. of severity: minor injuries, serious injuries, and fatal injuries. Te risk factors were the motorcyclist’s age and gender, speed limit zone, time of day, day of week, helmet use, 4.2.5. Having a Pillion Passenger. When a motorcycle has motorcycle age, presence of a pillion passenger, road surface a pillion passenger, serious injuries are 14.20% more likely to condition, type of road surface, light condition, and the occur. Tis fnding is consistent with earlier research. Te number of vehicles involved in the crash. authors in [21] stated that the severity of the injury increases Te marginal efect of parameters was calculated to when a pillion passenger is present, but the intensity of the determine the signifcance of the factors. According to the damages reduces. Tis might be due to the fact that there fndings, the following factors enhanced the probability of were at least two persons involved in the event, which in- fatal injury: not wearing a helmet, being in an older age creases the probability of the crash being categorized as group, riding during early morning or midnight hours, having a higher severity level. Also the distractions may arise weekend motorcycle use, and insufcient lighting (dark and from talking and paying attention to the pillion passenger. dusk/dawn). Also, having a pillion passenger, the motorcycle age of more than 7years, riding at higher speed limits (more than 50km/h) or during peak hours in the morning (6:30−8: 4.2.6. Light Condition. Dark and dusk/dawn indicators were 59 A.M), and being in the younger age group (less than signifcant in fatal injury severity with 99% and 95% conf- 26years old) were found to increase the chance of serious dence intervals, respectively. In addition, from the result of injury. the marginal efect, it can be concluded that light conditions Although the number of motorcyclists who did not wear in the categories of dark (fatal injury marginal efect �10.55%) and dusk/dawn (fatal injury marginal efect �3.52%) increase a helmet was notably lower than the number of those who did, not wearing a helmet appeared to be the most critical the probability of fatal injury in motorcycle crashes. Even cause of increasing fatalities. It shows that educating mo- though various fndings have been stated in the literature torcyclists on the importance of wearing a helmet and ap- regarding the efect of light conditions, this study’s fnding is plying strict enforcement measures for not wearing a helmet consistent with the majority of them [9, 18, 82]. is important to reduce fatalities. By comparing our fndings with those by [23], it can be 4.2.7. Road Surface Condition. Slippery road surface con- concluded that older age, unpaved surfaces, riding in the ditions were shown to reduce the likelihood of a fatality by early morning or midnight hours, and weekend motorcycle 33.64%. Tis is consistent with past studies [15, 21, 86] use are important risk factors for fatal injury in both in- fndings that wet roadway conditions increase the chance of tersection crashes and nonintersection crashes. However, no injury in motorcycle crashes. Under slippery conditions, factors like helmet use and lighting conditions are crucial in although motorcycle riders are more likely to be involved in nonintersection crashes. Tese variances in the results may a crash, they may be riding cautiously or reducing speed so have been due to the present study’s analysis of unobserved that the injury severity decreases. A study on driving be- heterogeneity of data or by variations in rider behaviors havior has shown that under rainy conditions, most drivers between junction and nonintersection situations. For ex- tend to decrease their speed and drive cautiously [87]. ample, intersections are usually in better lighting conditions, even if the lamps are of. Moreover, intersections are gen- erally exposed to conficting trafc fows, and motorcyclists 4.2.8. Road Surface Type. Te indicator variable “unpaved may be more cautious when approaching them, reducing roads” was found to be signifcant, with a 95% confdence their speed accordingly. Consequently, if they are involved interval in serious injury. According to the results of the in any crash at intersections, the severity of head injuries marginal efects, it reduced the risk of a fatal injury by 6.27%. may be less than when they are speeding away from in- Tis is consistent with [88], fnding that urban motorcycle tersections. Tese variances in the results may have been due crashes on unpaved roads were more likely to result in to the present study’s analysis of unobserved data or by incapacitating injury than fatal injury. Similar to the slippery variations in rider behaviors between junction and non- road condition argument, it is possible that on gravel and intersection situations. Furthermore, the intersection study unpaved roads, motorcyclists are more careful than normal, found that the number of vehicles involved in crashes was resulting in fewer fatalities, yet the road condition causes a signifcant factor in intersection crashes but not in non- more serious and minor injury crashes. intersection crashes, regardless of the severity level of the crash. One possible explanation for this discrepancy is the 5. Conclusion intersection’s characteristics, such as the presence of con- Tis study aimed to analyze the factors contributing to ficting trafc fows and therefore a higher possiblity of multivehicle crashes. Intersection studies have also reported motorcycle crash severity on Victorian urban non- intersection roadways while considering the heterogeneity of that other intersection characteristics, such as unpaved 10 Journal of Advanced Transportation Asian Countries–Case of Taiwan, Citeseer, Princeton, NJ, intersections, intersections controlled with stop or give-way USA, 2003. signs, and T-intersections, were signifcant variables for [8] Transport Accident Commission, “Lives lost-Annual,” 2022, severe injuries in intersection crashes. Intersection variables, https://www.tac.vic.gov.au/road-safety/statistics/lives-lost- however, are not related to nonintersection crashes. annual. Terefore, it seems crucial to consider the diferent char- [9] S. M. Rifaat, R. Tay, and A. De Barros, “Severity of motorcycle acteristics of intersection and nonintersection crashes and crashes in Calgary,” Accident Analysis and Prevention, vol. 49, unobserved heterogeneity of crash data when analyzing the pp. 44–49, 2012. factors contributing to crash severity. [10] P. Gershon, N. Ben-Asher, and D. Shinar, “Attention and Te current study was limited by the data available on the search conspicuity of motorcycles as a function of their visual Vic-Roads database. New parameters (for example, mo- context,” Accident Analysis and Prevention, vol. 44, no. 1, torcyclist experience and licence, and intoxicated riders) pp. 97–103, 2012. may aid in investigating more precise and comprehensive [11] National Highway Trafc Safety Administration, “Motorcycle safety,” 2020, https://www.nhtsa.gov/road-safety/ studies. One potential avenue for future research would be to motorcycles. consider the AADT and trafc composition, particularly [12] M. Waseem, A. Ahmed, and T. U. Saeed, “Factors afecting heavy vehicle percentage, as they have been found to sig- motorcyclists’ injury severities: an empirical assessment using nifcantly impact trafc characteristics and increase the random parameters logit model with heterogeneity in means likelihood of severe injury or fatality [89]. Te fndings from and variances,” Accident Analysis and Prevention, vol. 123, this study may assist Victorian road safety policy managers pp. 12–19, 2019. in picking efective strategies for improving motorcyclists’ [13] N. Parishad, K. Aghabayk, and M. Palassi, “Assessing risk safety at urban nonintersections. Tis may include im- factors associated with motorcycle crash severity in mashhad, proving the light conditions at nonintersection, encouraging Iran,” International Journal of Transportation Engineering, the motorcyclist to maintain motorcycles regularly, and vol. 10, no. 2, pp. 1041–1054, 2022. educating the motorcyclist to avoid distractions (e.g., con- [14] M. S. Shaheed and K. Gkritza, “A latent class analysis of versation with pillion passengers while riding) and ride single-vehicle motorcycle crash severity outcomes,” Analytic Methods in Accident Research, vol. 2, pp. 30–38, 2014. responsibly on the weekends (holidays). [15] V. Shankar and F. Mannering, “An exploratory multinomial logit analysis of single-vehicle motorcycle accident severity,” Data Availability Journal of Safety Research, vol. 27, no. 3, pp. 183–194, 1996. [16] P. Savolainen and F. Mannering, “Probabilistic models of Data can be made available by contacting the second co- motorcyclists’ injury severities in single-and multi-vehicle author, Dr. Kayvan Aghabayk (kayvan.aghabayk@ut.ac.ir). crashes,” Accident Analysis and Prevention, vol. 39, no. 5, pp. 955–963, 2007. Conflicts of Interest [17] W. H. Schneider and P. T. Savolainen, “Comparison of se- verity of motorcyclist injury by crash types,” Transportation Te authors declare that they have no conficts of interest. Research Record, vol. 2265, no. 1, pp. 70–80, 2011. [18] M. De Lapparent, “Empirical Bayesian analysis of accident severity for motorcyclists in large French urban areas,” Ac- References cident Analysis and Prevention, vol. 38, no. 2, pp. 260–268, [1] World Health Organization (WHO), “Road trafc injuries,” [19] R. Dandona, G. A. Kumar, and L. Dandona, “Risky behavior 2021, https://www.who.int/news-room/fact-sheets/detail/ of drivers of motorized two wheeled vehicles in India,”Journal road-trafc-injuries. of Safety Research, vol. 37, no. 2, pp. 149–158, 2006. [2] C. D. Mathers and D. Loncar, “Projections of global mortality [20] M. Abdel-Aty and J. Keller, “Exploring the overall and specifc and burden of disease from 2002 to 2030,” PLoS Medicine, crash severity levels at signalized intersections,” Accident vol. 3, no. 11, p. e442, 2006. Analysis and Prevention, vol. 37, no. 3, pp. 417–425, 2005. [3] P. Wu, X. Meng, and L. Song, “Bayesian space–time modeling [21] M. A. Quddus, R. B. Noland, and H. C. Chin, “An analysis of of bicycle and pedestrian crash risk by injury severity levels to motorcycle injury and vehicle damage severity using ordered explore the long-term spatiotemporal efects,” Physica A: probit models,” Journal of Safety Research, vol. 33, no. 4, Statistical Mechanics and Its Applications, vol. 581, Article ID pp. 445–462, 2002. 126171, 2021. [22] M. M. Abdul Manan, A. Varhelyi, ´ A. K. Çelik, and [4] M. D. Keall and S. Newstead, “Analysis of factors that increase H. H. Hashim, “Road characteristics and environment factors motorcycle rider risk compared to car driver risk,” Accident associated with motorcycle fatal crashes in Malaysia,” IATSS Analysis and Prevention, vol. 49, pp. 23–29, 2012. Research, vol. 42, no. 4, pp. 207–220, 2018. [5] Y. Guo, T. Sayed, and M. H. Zaki, “Evaluating the safety impacts of powered two wheelers on a shared roadway in [23] M. Abrari Vajari, K. Aghabayk, M. Sadeghian, and N. Shiwakoti, “A multinomial logit model of motorcycle crash China using automated video analysis,” Journal of Trans- portation Safety and Security, vol.11, no. 4, pp. 414–429, 2019. severity at Australian intersections,” Journal of Safety Re- search, vol. 73, pp. 17–24, 2020. [6] T. Allen, S. Newstead, M. Lenne´ et al., “Contributing factors to motorcycle injury crashes in Victoria, Australia,” Trans- [24] C.-W. Pai and W. Saleh, “Modelling motorcyclist injury se- verity by various crash types at T-junctions in the UK,” Safety portation Research Part F: Trafc Psychology and Behaviour, vol. 45, pp. 157–168, 2017. Science, vol. 46, no. 8, pp. 1234–1247, 2008. [25] C.-W. Pai, “Motorcyclist injury severity in angle crashes at T- [7] T.-P. Hsu, E. A. F. M. Sadullah, and I. N. X. Dao, A Com- parison Study on Motorcycle Trafc Development in Some junctions: identifying signifcant factors and analysing what Journal of Advanced Transportation 11 made motorists fail to yield to motorcycles,” Safety Science, [42] S. Patil, S. R. Geedipally, and D. Lord, “Analysis of crash vol. 47, no. 8, pp. 1097–1106, 2009. severities using nested logit model—accounting for the [26] M. M. Haque, H. C. Chin, and H. Huang, “Applying Bayesian underreporting of crashes,” Accident Analysis and Prevention, hierarchical models to examine motorcycle crashes at sig- vol. 45, pp. 646–653, 2012. nalized intersections,” Accident Analysis and Prevention, [43] R. A. Blackman and N. L. Haworth, “Comparison of moped, vol. 42, no. 1, pp. 203–212, 2010. scooter and motorcycle crash risk and crash severity,” Acci- [27] A. S. Al-Ghamdi, “Analysis of trafc accidents at urban in- dent Analysis and Prevention, vol. 57, pp. 1–9, 2013. tersections in Riyadh,” Accident Analysis and Prevention, [44] Y. Chung, T.-J. Song, and B.-J. Yoon, “Injury severity in vol. 35, no. 5, pp. 717–724, 2003. delivery-motorcycle to vehicle crashes in the Seoul metro- [28] B. Qiu and W. Fan, “Mixed logit models for examining pe- politan area,” Accident Analysis and Prevention, vol. 62, destrian injury severities at intersection and non-intersection pp. 79–86, 2014. locations,” Journal of Transportation Safety and Security, [45] S. Cafso, G. La Cava, and G. Pappalardo, “A logistic model for vol. 14, no. 8, pp. 1333–1357, 2022. Powered Two-Wheelers crash in Italy,” Procedia-social and [29] VicRoads, Crash Statistics, VicRoads, Melbourne, Australia, behavioral sciences, vol. 53, pp. 880–889, 2012. [46] A. Tavakoli Kashani, R. Rabieyan, and M. M. Besharati, “A [30] F. L. Mannering, V. Shankar, and C. R. Bhat, “Unobserved data mining approach to investigate the factors infuencing heterogeneity and the statistical analysis of highway accident the crash severity of motorcycle pillion passengers,” Journal of data,” Analytic methods in accident research, vol. 11, pp. 1–16, Safety Research, vol. 51, pp. 93–98, 2014. [47] S. Kumar and D. Toshniwal, “Severity analysis of powered two [31] Z. Christoforou, S. Cohen, and M. G. Karlaftis, “Vehicle wheeler trafc accidents in Uttarakhand, India,” European occupant injury severity on highways: an empirical in- transport research review, vol. 9, no. 2, pp. 24–10, 2017. vestigation,” Accident Analysis and Prevention, vol. 42, no. 6, [48] A. Montella, M. Aria, A. D’Ambrosio, and F. Mauriello, pp. 1606–1620, 2010. “Analysis of powered two-wheeler crashes in Italy by clas- [32] Y. Guo, Y. Wu, J. Lu, and J. Zhou, “Modeling the unobserved sifcation trees and rules discovery,” Accident Analysis and heterogeneity in e-bike collision severity using full Bayesian Prevention, vol. 49, pp. 58–72, 2012. random parameters multinomial logit regression,” Sustain- [49] Y. Guo, Z. Li, Y. Wu, and C. Xu, “Exploring unobserved ability, vol. 11, no. 7, p. 2071, 2019. heterogeneity in bicyclists’ red-light running behaviors at [33] Y. Guo, P. Liu, Y. Wu, and J. Chen, “Evaluating how right- diferent crossing facilities,” Accident Analysis and Prevention, turn treatments afect right-turn-on-red conficts at signalized vol. 115, pp. 118–127, 2018. intersections,” Journal of Transportation Safety and Security, [50] S. Washington, Statistical and Econometric Methods for vol. 12, no. 3, pp. 419–440, 2020. Transportation Data Analysis, Chapman and Hall/CRC, Boca [34] X. Wen, Y. Xie, L. Jiang, Z. Pu, and T. Ge, “Applications of Raton, FL, USA, 2020. machine learning methods in trafc crash severity modelling: [51] D. M. Cerwick, K. Gkritza, M. S. Shaheed, and Z. Hans, “A current status and future directions,” Transport Reviews, comparison of the mixed logit and latent class methods for vol. 41, no. 6, pp. 855–879, 2021. crash severity analysis,” Analytic Methods in Accident Re- [35] P. T. Savolainen, F. L. Mannering, D. Lord, and M. A. Quddus, search, vol. 3-4, pp. 11–27, 2014. “Te statistical analysis of highway crash-injury severities: [52] J.-K. Kim, G. F. Ulfarsson, V. N. Shankar, and S. Kim, “Age a review and assessment of methodological alternatives,” and pedestrian injury severity in motor-vehicle crashes: Accident Analysis and Prevention, vol. 43, no. 5, pp. 1666– a heteroskedastic logit analysis,” Accident Analysis and Pre- 1676, 2011. vention, vol. 40, no. 5, pp. 1695–1702, 2008. [36] F. Ye and D. Lord, “Comparing three commonly used crash [53] J. C. Milton, V. N. Shankar, and F. L. Mannering, “Highway severity models on sample size requirements: multinomial accident severities and the mixed logit model: an exploratory logit, ordered probit and mixed logit models,” Analytic empirical analysis,” Accident Analysis and Prevention, vol. 40, methods in accident research, vol. 1, pp. 72–85, 2014. no. 1, pp. 260–266, 2008. [37] Q. Hou, X. Huo, J. Leng, and Y. Cheng, “Examination of [54] D. N. Moore, W. H. Schneider, P. T. Savolainen, and driver injury severity in freeway single-vehicle crashes using M. Farzaneh, “Mixed logit analysis of bicyclist injury severity a mixed logit model with heterogeneity-in-means,” Physica A: resulting from motor vehicle crashes at intersection and non- Statistical Mechanics and Its Applications, vol. 531, Article ID intersection locations,” Accident Analysis and Prevention, 121760, 2019. vol. 43, no. 3, pp. 621–630, 2011. [38] F. L. Mannering and C. R. Bhat, “Analytic methods in accident [55] P. C. Anastasopoulos and F. L. Mannering, “An empirical research: methodological Frontier and future directions,” assessment of fxed and random parameter logit models using Analytic methods in accident research, vol. 1, pp. 1–22, 2014. crash-and non-crash-specifc injury data,” Accident Analysis [39] S. Islam and F. Mannering, “Driver aging and its efect on and Prevention, vol. 43, no. 3, pp. 1140–1147, 2011. male and female single-vehicle accident injuries: some ad- [56] N. Eluru and C. R. Bhat, “A joint econometric analysis of seat ditional evidence,” Journal of Safety Research, vol. 37, no. 3, belt use and crash-related injury severity,” Accident Analysis pp. 267–276, 2006. and Prevention, vol. 39, no. 5, pp. 1037–1049, 2007. [40] A. K. Celik and E. Oktay, “A multinomial logit analysis of risk [57] R. Paleti, N. Eluru, and C. R. Bhat, “Examining the infuence factors infuencing road trafc injury severities in the Erzu- of aggressive driving behavior on driver injury severity in rum and Kars Provinces of Turkey,” Accident Analysis and trafc crashes,” Accident Analysis and Prevention, vol. 42, Prevention, vol. 72, pp. 66–77, 2014. [41] Y. Guo, Z. Li, P. Liu, and Y. Wu, “Modeling correlation and no. 6, pp. 1839–1854, 2010. heterogeneity in crash rates by collision types using full [58] S. Yasmin, N. Eluru, and A. R. Pinjari, “Analyzing the con- Bayesian random parameters multivariate Tobit model,” tinuum of fatal crashes: a generalized ordered approach,” Accident Analysis and Prevention, vol. 128, pp. 164–174, 2019. Analytic methods in accident research, vol. 7, pp. 1–15, 2015. 12 Journal of Advanced Transportation [59] Y. Xie, K. Zhao, and N. Huynh, “Analysis of driver injury [75] J. J. Louviere, D. A. Hensher, and J. D. Swait, Stated Choice severity in rural single-vehicle crashes,” Accident Analysis and Methods: Analysis and Applications, Cambridge University Press, Cambridge, UK, 2000. Prevention, vol. 47, pp. 36–44, 2012. [60] N. V. Malyshkina and F. L. Mannering, “Markov switching [76] Z. Wang, C. Lee, and P.-S. Lin, “Modeling injury severity of single-motorcycle crashes on curved roadway segments,” in multinomial logit model: an application to accident-injury Proceedings of the 93rd Annual Meeting of the Transportation severities,” Accident Analysis and Prevention, vol. 41, no. 4, Research Board, Washington, DC, USA, December 2014. pp. 829–838, 2009. [77] L. T. Truong, H. T. Nguyen, and C. De Gruyter, “Correlations [61] B. J. Russo, P. T. Savolainen, W. H. Schneider, and between mobile phone use and other risky behaviours while P. C. Anastasopoulos, “Comparison of factors afecting injury riding a motorcycle,” Accident Analysis and Prevention, severity in angle collisions by fault status using a random vol. 118, pp. 125–130, 2018. parameters bivariate ordered probit model,” Analytic methods [78] C. Xin, Z. Wang, C. Lee, and P. S. Lin, “Modeling safety efects in accident research, vol. 2, pp. 21–29, 2014. of horizontal curve design on injury severity of single- [62] E. Hauer and A. Hakkert, “Extent and some implications of motorcycle crashes with mixed-efects logistic model,” incomplete accident reporting,” Transportation Research Transportation Research Record, vol. 2637, no. 1, pp. 38–46, Record, vol. 1185, no. 1-10, p. 17, 1988. [63] F. Ye and D. Lord, “Investigation of efects of underreporting [79] T. L. Jackson and M. J. Mello, “Injury patterns and severity crash data on three commonly used trafc crash severity among motorcyclists treated in US emergency departments, models: multinomial logit, ordered probit, and mixed logit,” 2001–2008: a comparison of younger and older riders,” Injury Transportation Research Record, vol. 2241, no. 1, pp. 51–58, Prevention, vol. 19, no. 5, pp. 297–302, 2013. [80] M. Islam, “Te efect of motorcyclists’ age on injury severities [64] D. McFadden and K. Train, “Mixed MNL models for discrete in single-motorcycle crashes with unobserved heterogeneity,” response,” Journal of Applied Econometrics, vol. 15, no. 5, Journal of Safety Research, vol. 77, pp. 125–138, 2021. pp. 447–470, 2000. [81] A. Pervez, J. Lee, H. Huang, and X. Zhai, “What factors would [65] K. E. Train, Discrete Choice Methods with Simulation, Cam- make single-vehicle motorcycle crashes fatal? Empirical evi- bridge University Press, Cambridge, UK, 2009. dence from Pakistan,” International Journal of Environmental [66] J. Damsere-Derry, E. K. Adanu, T. K. Ojo, and E. F. Sam, Research and Public Health, vol. 19, no. 10, p. 5813, 2022. “Injury-severity analysis of intercity bus crashes in Ghana: [82] S. R. Geedipally, P. A. Turner, and S. Patil, “Analysis of a random parameters multinomial logit with heterogeneity in motorcycle crashes in Texas with multinomial logit model,” means and variances approach,” Accident Analysis and Pre- Transportation Research Record, vol. 2265, no. 1, pp. 62–69, vention, vol. 160, Article ID 106323, 2021. [67] N. Alnawmasi and F. Mannering, “A temporal assessment of [83] F. Chang, M. Li, P. Xu, H. Zhou, M. Haque, and H. Huang, distracted driving injury severities using alternate “Injury severity of motorcycle riders involved in trafc crashes unobserved-heterogeneity modeling approaches,” Analytic in Hunan, China: a mixed ordered logit approach,” In- methods in accident research, vol. 34, Article ID 100216, 2022. ternational Journal of Environmental Research and Public [68] M. Abrari Vajari, K. Aghabayk, M. Sadeghian, and Health, vol. 13, no. 7, p. 714, 2016. S. Moridpour, “Modelling the injury severity of heavy vehicle [84] J. H. Salum, A. E. Kitali, H. Bwire, T. Sando, and P. Alluri, crashes in Australia,” Iranian Journal of Science and Tech- “Severity of motorcycle crashes in Dar es Salaam, Tanzania,” nology, Transactions of Civil Engineering, vol. 46, no. 2, Trafc Injury Prevention, vol. 20, no. 2, pp. 189–195, 2019. pp. 1635–1644, 2021. [85] A. Pervez, J. Lee, and H. Huang, “Identifying factors con- [69] T. Allen, R. McClure, S. V. Newstead et al., “Exposure factors tributing to the motorcycle crash severity in Pakistan,”Journal of Victoria’s active motorcycle feet related to serious injury of Advanced Transportation, vol. 2021, Article ID 6636130, 10 pages, 2021. crash risk,” Trafc Injury Prevention, vol. 17, no. 8, pp. 870– [86] S. Jung, Q. Xiao, and Y. Yoon, “Evaluation of motorcycle 877, 2016. safety strategies using the severity of injuries,” Accident [70] Y. Li, L. Song, and W. D. Fan, “Day-of-the-week variations Analysis and Prevention, vol. 59, pp. 357–364, 2013. and temporal instability of factors infuencing pedestrian [87] V. Bakhshi, K. Aghabayk, N. Parishad, and N. Shiwakoti, injury severity in pedestrian-vehicle crashes: a random pa- “Evaluating rainy weather efects on driving behaviour di- rameters logit approach with heterogeneity in means and mensions of driving behaviour questionnaire,” Journal of variances,” Analytic methods in accident research, vol. 29, Advanced Transportation, vol. 2022, Article ID 6000715, Article ID 100152, 2021. 10 pages, 2022. [71] S. A. Samerei, K. Aghabayk, A. Mohammadi, and [88] W. Agyemang, E. K. Adanu, and S. Jones, “Understanding the N. Shiwakoti, “Data mining approach to model bus crash factors that are associated with motorcycle crash severity in severity in Australia,” Journal of Safety Research, vol. 76, rural and urban areas of Ghana,” Journal of Advanced pp. 73–82, 2021. Transportation, vol. 2021, Article ID 6336517, 11 pages, 2021. [72] Speedytests, “What does the Slippery sign mean?,” 2022, [89] S. Moridpour, E. Mazloumi, and M. Mesbah, “Impact of heavy https://speedytests.co.uk/blog/what-does-the-slippery-road- vehicles on surrounding trafc characteristics,” Journal of sign-mean. Advanced Transportation, vol. 49, no. 4, pp. 535–552, 2015. [73] P. Tomas, R. Welsh, K. Folla, and A. Laiou, “Recommen- dations for a common data collection system and defnitions,” SaferAfrica project Deliverable, vol. 4, 2018. [74] S. A. Samerei, K. Aghabayk, N. Shiwakoti, and S. Karimi, “Modelling bus-pedestrian crash severity in the state of Victoria, Australia,” International Journal of Injury Control and Safety Promotion, vol. 28, no. 2, pp. 233–242, 2021.
Journal
Journal of Advanced Transportation – Hindawi Publishing Corporation
Published: May 25, 2023