1. Introduction
Safety is essential for railway development, and accident prevention is the primary concern in railway operations. Although large-scale railway accidents, such as collisions between trains or derailments of trains, easily attract the attention of the media and the public, the actual occurrence of such incidents is rare. According to European Union statistics, train–train collisions and derailments account for only 5% of all railway accidents [
1]. In contrast, accidents caused by pedestrian intrusion onto the tracks constitute a significant percentage of railway accidents and should warrant more attention. Previous studies conducted mainly in Western countries have shown that train–pedestrian collisions have become the leading cause of casualties in railway accidents [
2,
3,
4,
5]. For instance, in the United Kingdom, the number of deaths from illegal intrusions of pedestrians onto railways is three times larger than other types of railway-related deaths [
6]. In Sweden, approximately 80–100 people are killed each year in train–pedestrian collisions [
7], and in the United States, the number is about 500 [
8].
The severity and impact of train–pedestrian collisions are generally greater than those of road collisions. Due to their high speed, long braking distance and inability to change directions, trains are more likely to cause severe injuries and deaths than vehicles on the road. In the United States, the casualty ratio was 54.1% in 2021 [
8], and in Canada, this figure was as high as 67.2% [
9]. In addition, since the scene of train–pedestrian collisions may be very dreadful, people involved in such accidents, such as train drivers, passengers, bystanders and rescuers, might experience particular psychological trauma from which they find it difficult to recover [
10,
11,
12,
13,
14].
Due to the dreadful consequences and the prevalence of pedestrian intrusions, it is necessary to identify the causes of these collisions so that administrative personnel can take appropriate preventive measures. To date, most studies on this issue have used accident record data and descriptive analysis to reveal certain influential factors. These factors can be categorized in four groups: the characteristics of pedestrians, the time of the collision, the structure of the track and the type of train. Because the demographic information of the victims and the time of the collisions are always included in surveys, these two factors are most commonly analyzed.
Among the characteristics of the pedestrians in these cases, one consistent finding across different studies has been that men are more likely than women to be involved in such collisions [
15,
16,
17,
18]. In terms of age, most studies have found that young adults are more likely to be involved than other age groups. For example, Savage’s studied 471 train–pedestrian collisions in the United States in 2005 and pointed out that people aged 16–45 constituted the majority of such collisions [
17]. Mohanty and Patnaik’s research on 88 collisions in India showed that pedestrians between 21 and 40 years old were the most common victims of train–pedestrian collisions [
19]. In Finland, the most vulnerable age group was even younger at 10–29 years old [
18].
Another important pedestrian characteristic is their intention. While many pedestrians involved in these collisions were mere trespassers, others may have intentionally used the railway as a means to commit suicide. In some countries, the rate of suicide is very high. For instance, in Sweden, researchers can distinguish whether an incident was a suicide on the basis of a police report [
16]. In another study, researchers analyzed train suicides on an international level and found that train suicides accounted for 1–12% of all suicides [
20]. Although there are established criteria for judging railway suicides, it is still difficult to distinguish a suicide attempt from other types of intentions because most collision records provide insufficient information [
18]. Precrash behaviors, if accurately documented, can help identify an individual’s purpose. For example, lying on the track is typically considered a pre-suicide behavior, while jumping off the track is not. Most studies that have used this type of data have found that walking, sitting or lying on the track were the most frequent behaviors in cases of collisions [
21]. Even though these behaviors are more dangerous than others, it is still difficult to determine whether they can be regarded as precise proxies of suicide. As the records are based mainly on the train drivers’ narratives, it is difficult to verify whether “lying on the track” describes a scene accurately. Even if it is correct, understanding whether an individual instance is a suicide attempt or an involuntary fall is not an easy task.
Additionally, few studies have examined other pedestrian features. In two studies of situations in the United States, most train–pedestrian collisions were found to be related to alcohol or drug usage [
22,
23]. In another study, Pelletier found that most victims were area residents rather than homeless people or temporary residents, suggesting that loitering may not have been the cause of the collisions [
5].
When analyzing the times of such occurrences, current findings can be organized into the time of year, day of the week or date. In terms of the time of year, some studies have found that such collisions are more likely to occur in spring and summer, possibly as a result of increased outdoor pedestrian activities [
16,
24]. However, a study in Finland found no difference among the twelve months [
18]. For the day of the week, studies have found that collisions are more likely to occur in the second half of the week (Thursday to Sunday) or on weekends, showing the influence of typical working habits in industrialized countries. However, collisions in New Zealand are evenly distributed [
25].
Some features associated with the locations of collisions also played an important role. Interestingly, many researchers have focused on collisions at railway level crossings because level crossings are where people can legally cross the railway and are frequently the sites of accidents. Therefore, many studies have focused on analyzing the frequency and severity of collisions at level crossings [
26,
27] and proposed reasonable improvement measures [
28].
Table 1 is a partial summary of the references cited in this article. Although there are several existing works on pedestrian–train collision accidents and their causes, many issues have not been fully studied and resolved.
First, most of the published papers focus only on situations in Western developed countries. However, railroads also play an important role in developing countries, where train–pedestrian conflicts can be more severe because of denser populations and fewer protective facilities. Among developing countries, China is worth investigating for several reasons. First, as a major railroad country, China has the second-longest railway mileage in the world with 146,000 km and is still expanding its networks. The operational speed of the trains has become increasingly faster. Second, as one of the most populous countries in the world, China has experienced dramatic industrialization and urbanization in recent decades. As a result, there are serious conflicts in the use of land. Previously uninhabited areas around railways have been developed due to economic growth; thus, new railways must cut across certain habitual local routes, such as people’s living areas and farms. All these issues, which may differ from those examined in Western societies, may increase the likelihood of train–pedestrian conflicts. Skládaná et al.’s research on the Czech Republic explored a similar situation in Eastern Europe [
29].
Third, although many studies have investigated risk-contributing factors in the personal and temporal domains, how the characteristics of places and trains can influence train–pedestrian collisions has not been fully explored. Some protective facilities (fences) have been important in preventing collisions in small-scale experimental studies, but no research has investigated their effects based on a large set of data. While certain places, such as level crossings, are considered very dangerous, factors that might influence collisions on the track have not been fully studied, probably due to the difficulty in obtaining relevant data. However, evidence accumulated from road collision research has shown that road characteristics such as slope and curvature can influence the control of trains or detection of potential trespassers [
30,
31,
32]. Although previous studies have presented no direct evidence yet, we believe these properties might have certain effects.
Finally, some papers have compared the observed distribution of collisions with the baseline distribution in interpreting the data. For example, when concluding whether men are more likely to become involved in a collision than women, we should compare the observed distribution in the collision report with that in the regional population. This comparison is particularly important for studying situations in a society such as China because it has witnessed major changes in demographics.
To fill these research gaps, we examine train–pedestrian collisions in China for the first time. Our analysis was based on a 10-year collection period of collisions in Southwest China, a crowded, mountainous and inland region with ethnic, geographic and economic diversity. To depict a full picture of these records, we also obtained other data types from various sources. We compared the demographic data in the acquired forms with those of the general social survey to better understand collision characteristics in China. In addition, we collected structural information about the tracks from maintenance departments in order to explore the influence of track characteristics, including the slope, curvature and presence of fences.
4. Discussion
The main purpose of this study is to describe the basic characteristics of train–pedestrian collisions in the Greater Sichuan-Chongqing area of China and provide possible suggestions for future research and practice. Using collision records from 2011 to 2020 in this area, we analyzed how collision occurrence and severity were distributed across pedestrian-, time-, train- and track-related characteristics. Several findings are worth discussing.
First, there were 2090 train–pedestrian collisions in this area over the selected decade, resulting in a total of 963 people injured and 1173 people killed. Given the large population in this area (millions), the annual collision rate per million people was 1.4. It should be noted that this rate is lower than that in other countries, such as Finland (11.2) [
34], the United States (3.1) [
8,
35], and the European Union (6.2) [
36,
37]. In addition, the fatality rate was 54.9%, which is similar to that of the United States (54.10%) [
8] and slightly lower than that of Canada (67.20%) [
9], emphasizing the severe consequences of such collisions across different countries.
Regarding pedestrian-related characteristics, we found that men were more likely than women to be involved in train–pedestrian conflicts. This finding is consistent with previous studies in other countries [
18]. A possible explanation is that men are more risk-seeking due to male hormones such as androgen [
38]. Further analysis corroborated this explanation because the sex differences diminished as the victims became older, as illustrated in
Figure 6. As the secretion of sex hormones decreases during the aging process, men and women become equally likely to be involved in such collisions.
When analyzing the effect of age, we found that the oldest group (61–70) was the largest group of victims. This finding differed from those of studies in Western countries, where most victims were young and middle-aged people [
18,
19,
24]. The difference cannot be explained by age composition because although the aging population is rapidly increasing in China, the ratio of senior citizens is still lower in the Chinese population than in most Western societies [
39]. One possible reason is that the rapid expansion of the railway might cause certain conflicts with the habits of local people [
29]. Whereas younger people might adapt to these changes very quickly, older people are less likely to change their trip habits and thus are more likely to be involved in this kind of collision. Future research could examine whether older people in this area are less aware of new changes in the infrastructure or are more likely to cling to their old habits at higher risk. If so, it might be important to design age-friendly educational campaigns to prevent such collisions.
In terms of precrash behavior, we found that the leading behaviors before collisions were running across the tracks, followed by walking along the tracks, either on the line or in the station. This finding is consistent with previous research in Western countries [
21]. Further analyses showed that the relationship between precrash behaviors and collision severity differed across locations. When the collisions occurred in the station, the fatality rate was approximately 58.70% and reached the highest level when the victims ran across the tracks (64.70%). However, when they happened on the line, the overall fatality rate was slightly higher (65.80%), and the death rate was highest when the victims were lying on the tracks (82.60%). We believe the reason is that in the station, railway personnel can prevent people lying on the tracks from being hurt. Since the lying action is generally time-consuming, there would be enough time to notify the driver to brake or to forcibly remove people from the tracks. However, when the behaviors happen along the line, there are no personnel to prevent them, which could be the reason for the difference in fatality rates for people lying on the tracks. On the other hand, if a person runs across the tracks, railway personnel are unlikely to have time to respond, whether they observe the behavior or not. This reduces the difference in the fatality rate across locations.
For time-related characteristics, we first found that the collision rate in this area was decreasing annually. One possible reason might be the increased use of fences. Although we do not have actual statistics for the yearly installation of new protective equipment in this area, there has been a continuing effort to improve the protective infrastructure. According to Chinese government documents, since 2014, railway lines with a design speed of more than 120 km per hour have been completely closed through measures such as closed facilities and warning signs [
40]. Previous studies have suggested that fencing is among the most effective measures to prevent pedestrian intrusions [
41], and we believe that such measures are equally effective in China. However, the death rate in collisions appears to be increasing, and we have a few guesses about this trend. First of all, with the development of technology, the speed of trains has gradually increased during this decade, which may increase the severity of collisions. However, we cannot prove this assertion using our own data directly because the speeds in the early years were not fully recorded, as mentioned in a previous response. Secondly, more protective nets were installed after 2014. While this can reduce the number of less motivated trespassers (e.g., who only want to take a short cut), it might not reduce the number of very motivated intruders. For example, some people may deliberately break the protective net to commit suicide or just show off. As these behaviors are more related to severe consequences, the increase of death rate is reasonable. A minor but also possible reason is that the driver may reduce their degree of vigilance when driving along the line with protective nets installed. They would not expect people to appear on the track and have slower reactions.
In analyzing the monthly distribution, we found that collisions did not occur uniformly throughout the year. We observed that the collision occurrence reached its lowest value in February and its highest value in December. Previous studies have suggested that low temperatures and the Chinese spring festival might reduce the time that people spend outdoors. This may explain why collisions occurred with the lowest frequency in February. However, the highest occurrence rate in December cannot be explained. A possible interpretation is a special weather condition in these mountainous areas called swarm fog, which happens mostly in December. As this kind of fog can reduce the visibility range for drivers and pedestrians, it might increase the occurrence of collisions in this month [
42]. It might be fruitful for future studies to examine this possible influential factor by incorporating experts in meteorology.
We further found that collisions occurred equally across days of the week. Notably, this finding is different from the findings of studies in Western countries. While researchers from different countries will have different results, according to national rail collision data in the United States and previous research, the risk is higher on Fridays and Saturdays, and 49.2% of suicides and 65.7% of accidents in Finland occurred on weekends (Friday to Sunday) [
18]. This difference can be explained by the difference in the data of the population engaged in agriculture. Office workers do not need to work on weekends and can move around at will, while agricultural workers carry out agricultural activities according to the solar and weather conditions. There are no fixed rest days within a week, so the collision data within a week will not be concentrated on a particular day.
Within a day, collisions were more likely to happen in the daytime and reached the highest rate in the morning (10:00–11:00 a.m.) and the lowest late at night (3:00–4:00 a.m.). Such a pattern seems to be very similar to the human circadian rhythm, which governs the outdoor activities of pedestrians. However, the train schedule is a confounding variable here. As more trains run in the daytime than in the night, it is difficult to draw a firm conclusion that pedestrians’ travel behaviors are the only cause of such an hourly distribution. Future studies may further examine this possibility by controlling the quantity of train flow.
Regarding train-related characteristics, train types were found to play an important role. We discovered that freight trains were involved in most of the collisions (0.20%), followed by regular-speed passenger trains (37.10%). This pattern is very different from that in Western countries, where commuter trains were most likely to cause collisions [
24]. This ratio is in accordance with the proportion of freight trains to all running trains (66.90%) [
43], which is associated with the manufacturing and logistic power of China. It is worth noting that although high-speed passenger trains (HSPTs) represent a substantial proportion of all running trains (10–22%) according to two different sets of statistics [
44,
45], the number of collisions caused by HSPTs is very low (0.50%). The reason is that high-speed rail lines are built on a highly elevated base, and protective nets have been installed throughout the entirety of the lines, making it difficult for pedestrians to intrude.
We also found that train speed can influence collision severity. Victims are less likely to survive when trains run at higher speeds. Although this finding is seemingly quite commonsensical, to the best of our knowledge, we are the first to report the actual relationship between train speed and collision severity. As we also suffer from the problems of limited data, we must be aware that any conclusions about the speed might be confined to situations of recent years; future studies may use a new apparatus to accurately record the speed before the collision and use this finding to enact new speed-limit regulations in areas where collisions happen more frequently.
Regarding track-related characteristics, when analyzing the location, we found that more collisions occurred along sections between stations, and 10% of collisions occurred within stations. However, the mileage of the line owned by stations represents far less than 10% of the mileage of the entire line. Although the absolute number of collisions occurring at stations is low, the danger level is worthy of attention. This danger may be because, despite the management and monitoring of staff in stations, pedestrians can still intrude onto the track without any hindrance. Moreover, in stations, the times that pedestrians and trains arrive at the track are almost the same. When a train is coming, the density of pedestrians is much higher in the station than along the line.
Finally, regarding level-crossing collisions, in the past ten years, only two collisions occurred at level crossings within the jurisdiction of the China Railway Chengdu Group Co., Ltd. Due to the small number of collisions, there is not enough data to analyze them individually. This is very different from the relevant conclusions in other countries. In the United States, a total of 1749 collisions occurred at road-rail level crossings in 2021 [
46]. An Australian study found that collisions at level crossings cause significant economic losses of more than AUD
$ 116 million per year [
47]. In China, there have been almost no collisions at railway level crossings; from the documents we obtained from the China Railway Chengdu Group Co., Ltd., we found there are 48 level crossings along all the tracks in this area (17,893.88 km). Therefore, the number of level crossings per 100 km is 0.268. According to public data in the United States, the United States has a total of 203,778 level crossings (about 250,000 km) [
48], and the number of level crossings per 100 km is 81.5, which is much higher than China. This can prove that measures to reduce level-crossing accident rates, such as reestablishing crossings and building underpass tunnels or pedestrian bridges, have been effective [
49].
We might be the first to examine the influence of curvature and slope on train–pedestrian collisions, and we obtained some interesting findings. First, by comparing the collision distribution from the baseline (the actual proportion of different curvatures), we found that collisions were more likely to occur on sharper bends. The reason might be that a large curvature (small curve radius) may make it challenging for drivers to control the train and reduce the visibility range. Although studies in automobile driving have suggested that large curvatures may increase the workload of car drivers [
28], few studies have examined their possible influence on train drivers. Future studies may use train simulators to further explore such influences. Second, in terms of track slope, we found that collisions were more likely to occur on gentle slopes, either upward or downward, than on level tracks. However, the collision rate dropped again on the steepest slopes. This cannot be explained by the fact that steeper roads cause greater psychomotor demand for drivers. One possible explanation might be that drivers may have different risk perceptions and vigilance levels when facing different slopes [
50]. When the slopes are gentle, although they still cause certain problems in controlling the train, the drivers may not be aware of the risk factor. However, when the slopes are very steep, the drivers will be highly attentive and use the highest standard to ensure safety when operating on these more dangerous sections (e.g., reducing the speed of the train). Future studies may examine drivers’ vigilance level and risk perception on different slopes to test this explanation. If this is the case, it might be useful to develop certain training or warning systems to improve drivers’ safety awareness.