A public highway–rail grade crossing (HRGC), also known as a level crossing, is an intersection where a road or highway and a railroad meet on equal ground. This design permits the crossing of vehicles and trains. These crossings have been designed specifically to facilitate the safe passage of vehicles over railroad tracks. To enhance safety, public HRGCs are typically equipped with warning signs, pavement markings, and, in some cases, traffic control devices, such as flashing lights, gates, and bells. These safety measures are intended to alert drivers to approaching trains, ensuring they arrive at a complete stop and yield the right-of-way before proceeding across the tracks. However, it is essential to note that many of these safety traffic control devices are absent in developing nations, posing a significant safety risk (Tjahjono et al., 2019).

While numerous developed nations have made substantial advancements in reducing road traffic fatalities in recent years, the progress remains highly variable on a global scale. The risk of road traffic fatalities is significantly elevated in low- and middle-income countries, with an average rate of 27.5 per 100,000 population, in contrast to high-income countries, where the average rate stands at 8.3 per 100,000 population (WHO, 2018). Thailand, classified as a middle-income and developing nation, grapples with considerable economic and emotional burdens attributable to road accidents, characterized by a death rate of 32.8 per 100,000 population (WHO, 2018). According to data from the Department of Highways (DOH) in 2004, the total costs associated with traffic accidents in Thailand amounted to 153,755 million baht (approximately 3,460 million USD). Notably, the total costs related to highway-rail grade crossing (HRGC) accidents for the year 2004 were 2,482 million baht, which increased to 4,344 million baht (approximately US$140 million) due to economic growth, representing roughly 0.4 percent of Thailand’s GDP in 2011 (Settasuwacha et al., 2012). This is primarily attributed to the human toll resulting from fatalities and severe injuries. Consequently, it is crucial to identify and comprehend the risk factors associated with fatal and severe crashes, with the aim of devising effective policies and implementation strategies to mitigate these incidents.

Accidents at HRGCs in Thailand have become a significant safety issue, resulting in a considerable number of crashes and deaths over time (Settasuwacha et al., 2012). Several factors are responsible for these accidents, such as insufficient warning systems, driver irresponsibility, visibility issues, train speed, and a lack of education and awareness about the problem. Although HRGC accidents constitute a relatively minor percentage of total accidents compared with other types of collisions, the severe outcomes typically associated with train involvement often command considerable public concern and media coverage.

Investigating the risk factors that affect the severity of crash injuries in public HRGC is vital to devise successful strategies to reduce fatalities and injuries at these sites. The risk analysis associated with the severity of collision injuries on HRGCs may be more nuanced than on conventional roads. This complexity arises from the intricate relationships between road users and the distinct environment of the highway–railroad intersection.

Several past studies have investigated the factors contributing to HRGC accidents. For instance, Yan et al. (2010) conducted an analysis using Federal Railroad Administration HRGC crash data spanning from 1980 to 2005. Their study compared the annual crash rate before and after installing stop signs at passive crossings, revealing that such installations significantly decreased the annual crash rates. Eluru et al. (2012) used a latent segmentation-based ordered logit model to examine the elements that influence the severity of injuries drivers sustain in vehicle-train collisions at United States HRGCs. They found that the severity of the injury was affected by several key factors, such as the timing of the accident, the driver’s age, weather conditions (e.g., snow or rain), the vehicle’s participation in the crash, and the driver’s actions leading up to the collision.

Meanwhile, Hao and Daniel (2013) analyzed HRGC crashes in the United States from 1997 to 2006, using a conventional ordered probit model. Their findings revealed that adverse weather conditions, reasonably favorable visibility conditions, vehicle speeds over 50 miles per hour (mph), and train speeds exceeding 50 mph at the time of collision were associated with an increased probability of injury or death. Moreover, Hao and Daniel (2014) used an ordered probit model to identify risk factors contributing to injury severity at railroad grade crossings, taking active controls (gates and flashing lights) and passive controls (cross bucks and stop signs) into account. Using crash data from the United States Federal Railway Administration, they identified a number of significant factors that affected the severity of driver injuries at both active and passive highway–rail crossings. These factors included time (rush hour), visibility, vehicle speed, train speed, driver age, area type, traffic volume, and road surface conditions.

Liu et al. (2015) compared the safety outcomes of crashes at HRGCs with passive controls (e.g., Crossbucks and Stop signs) and those with active controls (e.g., flashing lights, gates, audible warnings, and highway signals). Their study found that drivers are more likely to stop at crossings with gates, flashing lights, and audible warnings, leading to less serious injuries. Fan et al. (2015) employed a multinomial logit model to discern crucial factors that influence variations in injury severity, and to investigate the influence of these explanatory variables on three different levels of severity of vehicle-related accidents that occur at HRGCs in the United States. Their findings revealed that when high-speed rail equipment collided with a vehicle, the probability of a fatal outcome increased significantly. Additionally, their analysis indicated that pickup trucks and surfaces made of concrete or rubber were more prone to involvement in more severe accidents. In yet another study, Hao et al. (2015) employed an ordered probit model to explore the variables that influence driver injury severity at HRGCs, considering variations in age and gender among United States motor vehicle drivers. Their findings indicated that older drivers are more likely to sustain fatal injuries when driving passively in open areas, particularly in adverse weather conditions. Conversely, they found that younger male drivers were more susceptible to severe injuries during peak traffic hours while driving at high speeds, and at unpaved HRGCs with passive control.

In Hao et al. (2016a) study, ordered probit models were employed to investigate the factors that influence the severity of injuries among motor vehicle drivers involved in HRGC accidents, specifically focusing on differentiating between time-of-day effects. The findings reveal that the severity of injury among motor vehicle drivers is significantly increased during the a.m. peak, p.m. peak, and p.m. off-peak periods compared with other time intervals. On a different note, Ghomi et al. (2016) employed ordered probit models, association rules, and classification and regression tree algorithms to highlight the primary factors related to injury severity for vulnerable road users in accidents at HRGCs in the United States. Their study found that train speed significantly influences injury severity, and older road users have a higher probability of fatal accidents than their younger counterparts. In a different context, Laapotti (2016) examined fatal HRGC accidents in Finland spanning from 1991 to 2011 and highlighted the effectiveness of active warning devices in reducing fatal crashes. Hao et al. (2016c) assessed the severity of driver injuries sustained in truck accidents at HRGCs in the United States. According to their analysis, specific characteristics of truck driver behavior, including driving under the influence of fatigue during peak hours, emerged as statistically significant predictors of increased injury severity. Wang et al. (2016) used geographically weighted regression to study injuries from rail-trespassing crashes along United States railway tracks. They discovered that people lying or sleeping on or near the tracks were more likely to sustain injuries. Meanwhile, Kang and Khattak (2017) investigated the severity of crashes reported at HRGCs by employing suitable data clustering techniques to account for the unobserved heterogeneity. Their findings underscored a significant correlation between higher train speeds and increased severity of injuries in train-vehicle collisions.

Using the mixed logit model analysis, Mannering et al. (2016) conducted a study to investigate the determinants of driver-injury severity at HRGCs in the United States, considering the presence or absence of aggressive driving behaviors. Younger male drivers who exhibit aggressive driving behaviors are more likely to sustain severe injuries, particularly during peak hours, especially in the morning peak (6–9 a.m.), and in open space areas. Khan and Khattak (2018) used the mixed logit model to identify factors contributing to the severity of injuries sustained by truck and truck-trailer drivers involved in United States crashes. They found higher train speeds, drivers ignoring crossing gates, older driver age, crashes in rural areas, situations where the train hit the truck/truck-trailer, and crashes at crossings with a crossing angle of 60–90° all correlated with more severe injuries. Zhao and Khattak (2018) used a binary logit model of random parameters to study the impact of driver inattention on injury severity in crashes near HRGCs in Nebraska, United States. According to their research, distracted driving causes more severe injuries than attentive driving. In two-vehicle collisions where at least one driver was not paying attention, the likelihood of at least one driver being injured increased by 14.6%.

In their study using latent class clustering, Zhao et al. (2019) used latent class clustering to investigate factors associated with the severity of pedestrian injuries in train-pedestrian collisions at HRGCs, using United States Federal Railroad Administration data. Their research showed that variables such as freight train involvement, direct impact between the train and the pedestrian, the lack of flashing lights and warnings at crossings, rural locations, lower visibility conditions, and older pedestrians all increase the severity of pedestrian injuries. Tjahjono et al. (2019) examined road-railway level crossing crashes in Indonesia between 2013 and 2016, using a traditional ordered logit modeling framework. Their analysis revealed associations between fatal crashes and male drivers, rainy weather, and low traffic volume conditions. Using the mixed logit modeling framework, Khales et al. (2020) analyzed 10 years of crash data, spanning from 2008 to 2017, involving HRGCs in the United States. The results revealed that the crash-contributing factors differed between crossings equipped with active and passive warning devices. In a more recent study, Ahmed et al. (2023) applied a random parameter model, considering variations in means and variances, to identify contributing factors to the severity of injuries in accidents at HRGCs. They found that crashes on main tracks had a higher likelihood of injury and death compared to those on tracks on the yard, siding, or industry. Furthermore, obstructed views of the rail track for drivers were associated with a higher chance of fatalities.

All coefficients of the explanatory variables were allowed to vary across the crash population during estimation. Following this step, only two variables exhibited significant random parameters: “Traffic sign 1” and “Lack/no traffic device.” Both variables exhibited substantial standard deviations, indicating significant varying effects across the crash population. Among the distributions considered for random parameters, the normal distribution was determined to provide the optimal statistical fit, outperforming the triangular, uniform, and lognormal distributions. A likelihood ratio test was conducted to determine whether these two variables should be included in the model as random parameters to determine if their inclusion would lead to a significantly improved model fit compared to a fixed-effect model. The test is conducted as follows (Washington et al., 2020): χ 2 = − 2   L L β f i x e d − e f f e c t − L L β r a n d o m − e f f e c t

Table 2 presents a statistical fit comparison between fixed- and random-effect models, employing the likelihood ratio test, ρ², and the Akaike Information Criterion (AIC) (Se et al., 2021a). As shown in Table 2, using “traffic sign 1” and “Lack/no traffic device” as random parameters produces a higher ρ² and a lower AIC value, indicating a better goodness of fit compared to using them as fixed parameters. The ρ² value of 0.182 in the random parameter model falls within an acceptable range, consistent with previous studies on the severity of crash injuries (Alnawmasi and Mannering, 2019; Alogaili and Mannering, 2022). Furthermore, the likelihood ratio test results demonstrated that the random parameter models for both factors were statistically superior to the fixed-effect models with a confidence level exceeding 98%. In summary, incorporating random parameters can significantly improve statistical fit, consistent with the findings of previous studies (Ahmed et al., 2023; Khales et al., 2020; Khan and Khattak, 2018).

After identifying the random parameters, the study investigated whether other fixed parameters affected their distribution, including their mean and variance. After conducting individual tests for each parameter, the study did not identify any significant heterogeneity in the means or variances of these parameters. Consequently, the model estimation reverted to the standard mixed logit model.

The results of the mixed logit estimate are detailed in Table 3. Regarding temporal characteristics, the results of the coefficient estimation in Table 3 indicate that accidents at HRGCs occurring between midnight and 6 a.m. are more likely to result in severe injuries. Furthermore, crashes between 6 a.m. and midday also positively correlated with PDO crashes. These findings align with logical explanations. During these hours, reduced visibility due to darkness can pose challenges to drivers and train operators in detecting obstacles and responding promptly. In addition, drivers may experience increased fatigue during the late night and early morning hours, potentially altering their reaction times and decision-making abilities. Another contributing factor could be the reduced traffic volume during these times, leading to higher vehicle speeds and a false sense of security among drivers, thus increasing the likelihood of serious accidents. According to previous studies, accidents on railway tracks at night or in darkness tend to result in more severe injuries (Eluru et al., 2012; Hao and Daniel, 2014).

Our findings indicate that accidents at HRGCs involving pickup trucks and heavy trucks are significantly more likely to result in severe injuries and fatalities. Previous studies also observed a similar trend with pickup truck crashes (Fan et al., 2015); however, heavy truck crashes were less likely to result in fatalities (Yan et al., 2010; Hao and Daniel, 2014; Hao et al., 2016b). This could be because pickup trucks and heavy trucks are generally larger and heavier than passenger cars. In collisions with trains, the size and mass of these vehicles can cause more significant damage and a greater risk of severe injuries or fatalities. In addition, passenger cars often have safety features and designs that better protect occupants in collisions. On the contrary, pickup trucks and large trucks may have fewer safety features and less effective crash protection, making their occupants more vulnerable. In the event of a collision, cargo carried by large trucks may shift or spill, resulting in more severe consequences. Typically, larger and heavier vehicles require a greater distance to come to a complete stop. If a vehicle cannot stop in time at a railroad crossing, the collision with a train is more likely to be severe.

Severe injuries and PDO crashes were positively correlated with crashes involving motorcycles as the secondary party. This finding differs somewhat from previous research, such as the 2014 study by Hao and Daniel (2014) in the United States, which found that motorcycle collisions at railroad crossings were more likely to result in severe or fatal accidents. The distinct sociocultural and contextual factors surrounding motorcycle use in Thailand versus the United States may account for this disparity. In the United States, motorcycles are frequently used for recreational purposes, and riders frequently operate motorcycles with larger engines. These powerful motorcycles can attain higher speeds and their riders may be more inclined to ride at substantial velocities, potentially resulting in more severe accidents in railway crossing collisions. On the contrary, in Thailand, motorcycles serve as versatile and ubiquitous modes of transportation for a wide range of daily activities. They are commonly used for family runs, commuting to work or school, and navigating congested urban areas. In addition, many motorcycles in Thailand are equipped with smaller engines, which tend to have lower maximum speeds than their larger engine counterparts. These contextual differences likely influence variations in accident outcomes in motorcycle usage patterns and vehicle specifications. To illustrate the disparity in motorcycle-involved collisions, it is necessary to consider the frequency of these incidents in each setting. In the United States, motorcycle-related accidents at railway crossings account for less than 0.5% of the total number of railway crossing accidents, as reported by Hao and Daniel (2014). In contrast, the prevalence of motorcycle-related railway crossing accidents in Thailand is significantly higher, accounting for a substantial 24.2% of all railway crossing collisions. Consequently, the disparity in motorcycle-involved collisions at railroad crossings between the two nations highlights the importance of considering local contextual factors and usage patterns when interpreting crash data. It also highlights the need for tailored safety measures and interventions in each region to address motorcycle use’s unique characteristics and risks at railway crossings.

Although factors such as the number of railway lanes, the type of area, and the lanes of the roadway were not significantly associated with the severity of the accident injuries, several variables within the group of traffic signs were found to significantly influence the severity of the crash outcomes. Specifically, railway crossings with traffic sign 2 (sign indicating railroad crossing without barriers), traffic sign 10 (hazard marker), and traffic sign 14 (Traverse rumble strips) (see Supplementary Figure S1), were found to be significantly and positively associated with PDO accidents. However, traffic sign 1 (warning for a single-track railroad crossing) was determined to be a random parameter for injury crashes. The normal distribution of this random parameter (mean = 2.046 and standard deviation = 6.265) revealed that 63% of the collisions at intersections with traffic sign 1 were more likely to result in injury collisions, while 37% were more likely to result in fatal injury crashes. Conversely, railway crossings equipped with traffic sign 9, denoting a no-passing zone, were positively associated with fatal injury crashes. When we investigate the dynamics of these crossings, a plausible explanation emerges for this seemingly counterintuitive result. Traffic sign 9 is typically installed at railway crossings in areas with high-speed traffic and frequent overtaking maneuvers. This signage serves as a regulatory measure aimed at improving safety by prohibiting overtaking near the crossing, where doing so could significantly increase the risk of collisions with oncoming trains. However, the mere presence of traffic sign 9 may not be sufficient to effectively deter drivers from overtaking or force them to reduce their speed when approaching the railway crossing. Given these complexities, the positive association between traffic signal nine and fatal injury crashes underscores the need for a multifaceted approach to railway crossing safety.

In terms of types of traffic control, the results showed that railway crossings with vertical moving barriers were positively associated with injury crashes. Furthermore, intersection type B (intersections equipped only with a warning post and a horizontal crossing barrier) was statistically and positively associated with fatal accidents. A variable representing railroad crossings without traffic control devices generated a significant random parameter with a mean of −0.775 and a standard deviation of 4,709. Further interpretation of these distributions reveals that 57% of crashes at these locations were more likely to result in injuries or PDO crashes, whereas 43% of crashes at railroad crossings without traffic control devices are positively associated with fatal injuries. A possible explanation is that railroad crossings lacking traffic control devices frequently lack visual or audible signals to warn drivers and pedestrians of an approaching train. Without warning signals or gates, drivers and pedestrians often have insufficient time to react and safely clear the tracks when a train approaches. Additionally, the absence of traffic control devices may result in risky behavior, such as attempting to outrun an approaching train, underestimating the train’s speed, or disregarding safety precautions. When a collision occurs, these behaviors can have disastrous effects. This finding highlights the critical significance of traffic control devices, such as warning signals, gates, and flashing lights, in reducing the severity of accidents at railroad crossings.

The results of the model employed in this study highlight critical factors that impact the severity of driver injuries at HRGCs in Thailand. These insights are invaluable in suggesting potential strategies for accident prevention and injury reduction. The following section elaborates on potential countermeasures and interventions.

Concerning the time of day, the coefficient of the result indicated that crashes occurring between midnight and 6 a.m. were more likely to result in severe injuries. A possible recommendation is to improve visibility at railway grade crossings during night by installing efficient lighting systems. This measure could alert drivers to the presence of crossings and improve their ability to spot approaching trains.

Regarding vehicle types, pickup cars and truck drivers were found to have a higher risk of fatal injuries at railway grade crossings. A potential recommendation is to develop specific education and training programs specifically for truck and pickup drivers, focusing on safety measures and best practices at HRGCs. This could involve raising awareness of associated risks, underlining the importance of stopping and looking for oncoming trains, and providing guidance on safe maneuvering techniques.

Taking into account the association between traffic signs and crash severity, the study found that the presence of hazard markers and rumble strips is positively correlated with PDO crashes, and the probability of fatal outcomes decreases. Therefore, a possible recommendation is to ensure the placement of hazard markers, such as reflectors or delineators, in suitable locations near railway grade crossings. These markers improve visibility and provide clear signals of the presence of a crossing, reducing the likelihood of fatal and injury-related crashes. They could be installed on roads edges, medians, or other strategic locations to effectively alert drivers. In addition, it is recommended to install traverse rumble strips leading to railway grade crossings. These textured or elevated strips on the road generate vibrations and auditory warnings when vehicles approach them, prompting drivers to slow down and exercise caution. This can help decrease the vehicle’s speed when approaching the crossing and lower the probability of accidents. Furthermore, it is suggested to collaborate with relevant transportation agencies to develop guidelines and standards for installing and maintaining hazard markers and traversing rumble strips at railway grade crossings. This would encourage a uniform implementation across different regions and ensure adherence to safety regulations.

Concerning crossings located in areas prone to vehicle overtaking, it is recommended that, along with the “no overtaking zone” sign (traffic sign 9), additional warning signs be installed at the railway crossing. These may include signs that indicate an upcoming railway crossing, emphasizing the need to slow down, exercise caution, and resist oncoming trains. Consideration should also be given to installing physical barriers, such as bollards or delineators, to dissuade drivers from undertaking overtaking maneuvers near the railway crossing. Such barriers can establish both a visual and a physical deterrent that discourages overtaking and urges drivers to maintain safe speeds and position on the road.

Regarding other traffic control devices, crossing intersections equipped only with warning posts or horizontal barriers and those lacking traffic control devices were positively associated with a higher risk of fatal accidents. As a remedy, it is recommended to reassess intersections that currently only employ warning posts or cross horizontal barriers and consider the installation of traffic signals. Traffic signals can effectively regulate vehicle flow and provide clear directions on right-of-way, thus reducing the risk of collisions and improving safety for all road users. In situations where traffic signals may not be necessary, implementing stop signs at intersections should be considered. Stop signs control vehicular movement by forcing drivers to stop completely and yield the right-of-way to the train, consequently improving safety and reducing the risk of serious accidents. Finally, it may also be crucial to conduct safety audits at intersections lacking traffic control devices, to identify potential hazards, and propose suitable safety measures. Safety audits can evaluate sightlines, traffic volumes, and crash history to decide on the most suitable traffic control devices or engineering improvements for each intersection.

This research explores risk factors related to the severity of injuries sustained in crashes at railway grade crossings in Thailand. Using a decade’s worth of crash data, from 2012 to 2022, this study sought to identify key risk factors. For this purpose, a random parameter model was utilized, taking into account potential heterogeneity in means and variances. This modeling technique enables the integration of unobserved characteristics into the analysis. The study evaluated three levels of injury severity: PDO, injury, and fatal. A variety of potential risk factors were considered, including the time of the accident, the type of vehicle involved, the number of rail lanes, the type of area, the number of vehicle lanes, the type of railway sleepers, the types of traffic signals installed, visibility conditions, control device types, and types of railway intersections.

The analysis of marginal effects has revealed several factors strongly associated with severe or fatal crashes. These factors include accidents that occur between midnight and 6 a.m., the involvement of pickup trucks or cars, the presence of traffic sign 9 (indicating a no overtaking zone), Intersection Type B1 (featuring warning posts and horizontal crossing barriers), and intersections lacking traffic control devices. On the contrary, factors positively correlated with PDO crashes (no injury) including the presence of traffic sign 2 (indicating a railroad crossing without barriers), traffic sign 10 (signifying hazard markers), and traffic sign 14 (indicating traverse rumble strips). In particular, a new finding in this study is that the effects of pickup cars, trucks, motorcycles, and the traffic sign for a “no overtaking zone” yielded contradictory results compared to previous studies. This discrepancy can be attributed to sociocultural and contextual factors that influence crash characteristics in developing countries such as Thailand as opposed to developed nations such as the United States.

In summary, this study looks at the factors contributing to the severity of driver injuries in accidents at railway grade crossings and discusses potential recommendations to improve traffic safety at HRGCs. The findings offer valuable information for transportation engineers addressing safety concerns at these crossings.

As with any study, there are limitations to this paper. First, the available data for this study lack crucial variables that may influence crash severity outcomes, such as driver characteristics, varying train speeds, varying road speeds, and others. Therefore, relevant authorities should make greater efforts to collect more exhaustive and detailed datasets to facilitate more comprehensive studies. Second, recording specific crash locations would be advantageous for conducting network analysis and creating geographical maps that define risks and safety, thereby enhancing the depth of research in this field.