Factors Influencing Escalator-Related Incidents in China: A Systematic Analysis Using ISM-DEMATEL Method

Escalator-related incidents (EIs) have recently resulted in serious injuries and even deaths. Given the frequency and severity of EIs, a systematic exploration of factors influencing EIs is critical in order to identify preventive measures. Twenty-two factors influencing EIs were identified by analyzing 213 EI cases in China and related literatures. A combination of the Interpretive Structural Modeling (ISM) and Decision Making Trial and Evaluation Laboratory (DEMATEL) methods were utilized to establish a hierarchical structure of the influencing factors and to distinguish cause factors and effect factors. The results show: (i) behavior, emergency plan, safety rules, safety supervision, information exchange, safety culture, and safety education are the most important factors influencing EIs; (ii) safety education is the cause factor imposing the greatest influence on other factors while behavior is the effect factor that is the most influenced; and (iii) the structure of influencing factors has five hierarchies, and factors in the root cause layer are settings and components, safety rules, safety supervision, safety culture, and safety education. Management priority should be given according to the hierarchy level, and the interaction of factors should be considered when taking preventive measures. The corresponding five-layer countermeasures are proposed to reduce escalator-related injuries.


Introduction
An escalator is a type of vertical transportation in the form of moving staircases. Along with common transportation instruments like subways and trains, escalators are a special and important part of the public transportation system that enable people to travel to their destinations. However, the dynamic interaction of escalator-to-passenger, passenger-to-passenger, passenger-to-environment, and passenger-to-management forms a complex escalator-passenger-environment-management (EPEM) system. Safety risks in this system are unpredictable, hard to identify and difficult to control, increasing the likelihood of EIs. The recent emergence of EIs in various countries has brought many negative effects on public safety and social sustainability. On 23 October 2018, an escalator incident in a subway station in Rome, Italy injured more than 20 passengers. According to General Administration of Quality Supervision Inspection and Quarantine (AQSIQ) statistics, 56 serious EIs occurred in China in 2017 and 31 occurred in 2018. EIs have consistently been the leading cause of injuries and deaths related to special equipment from 2008 to 2018 in China. A number of serious EIs were reported by mainstream media in recent years and are listed in Table 1. Preventing the occurrence of EIs has become an issue worthy of attention. Statistical studies on EIs were conducted to analyze the frequency distribution of factors such as cause, hazard pattern, injured body part, gender, and age. Xing et al. identified the main causes of EIs in subway stations in Guangzhou, China as failing to keep balance, carrying out other tasks, not holding the handrail and unhealthy passengers [1]. O'Neil et al. found that escalator-related injuries among old adults in the U.S. were mainly caused by a slip, trip, or fall [2]. Chi et al. investigated 194 EIs cases and suggested that strict design code and in-depth incident investigation are effective measures to prevent incidents [3]. McGeehan et al. studied 26,000 escalator-related injuries of children in the U.S. and found that falls and entrapments were the crucial causes of injuries [4]. Additionally, some studies examining EIs were either case studies or focused on children and old adults [5][6][7].
Simulation methods were utilized to study the evolution process of EIs in a virtual world [8]. Li et al. used simulation tools to study the impact of key factors on group stampede risk during escalator transfer [9]. They concluded that the propagation speed of the incident was always faster than the recovery rate, and timely taking emergency measures can reduce the severity of the accident. Li et al. analyzed the impact of contributing factors on individual stampede probability, and the impact of four factors were queued in descend order: pedestrian traffic, picking-up duration, pedestrian velocity, and escalator velocity [10]. Kauffmann and Kikuchi calculated the practical capacity of escalators through simulation based on passenger behavioral rules and found that prohibiting walking on the escalator can improve evacuation efficiency in emergency scenarios by reducing variability in the escalator system [11].
In the EPEM system, however, EIs are much more likely to be caused by multiple influencing factors rather than one single factor. The dynamic interaction of factors usually constitutes multiple influencing loops to form a complex structure. Influencing factors of EIs have not been adequately analyzed through a traditional linear analysis or simulation method mentioned above. Further in-depth understanding of the influencing mechanism of the factors is crucial for EPEM system safety.
ISM is an analytical method exploring complex socio-economic system problems. It is also an important structural modeling technique for system analysis [12][13][14]. DEMATEL is an approach using graph and matrix theory to analyze system factors [15]. ISM emphasizes establishing hierarchical structure models of complicated systems according to relationships of system factors, while DEMATEL focuses on identifying key factors of systems and distinguishing cause and effect factors. The integrated ISM-DEMETAL method has been widely used to overcome the weaknesses of applying either alone as they are complementary in function. Wang et al. defined the factors of mining safety and revealed the influencing mechanisms by using ISM-DEMETAL method [16]. Fu et al. clarified system structure of communication networks through ISM-DEMETAL and entropy [17]. Shen et al. constructed the multi-level hierarchical structure of factors affecting distributed natural gas-combined cooling, heating, and power systems in China [18]. Mousavizade and Shakibazad studied the critical success factors of knowledge management in Iranian urban water and sewage companies [19]. In this paper, the ISM-DEMETAL method was adopted to analyze the influencing factors of EIs extracted from 213 EIs cases in China and related literatures. This paper systematically analyzed influencing factors of EIs in China and accordingly proposed preventive measures, providing theoretical support for safety management of escalator passengers. The main aims of this paper were: (a) to identify the influencing factors of EIs and distinguish the importance and degree of influence of each factor; (b) to establish the hierarchical structure of the factors and analyze the influencing mechanism; and (c) to correspondingly propose preventive measures to reduce escalator-related injuries.

Data Resource
News is an original record with authoritative description, which provides the latest statistics and the detailed information [20]. A Web crawler tool was adopted to obtain Chinese news reports of EIs published between September 2003 and February 2019. Data was recorded from news sites including China News, China Daily, Xinhua Net, People Net, Baidu News, Sina News, CNTV and Youth News [21]. "Escalator injury", "escalator incident" and "escalator death" were used as subject terms to conduct the retrieval of related news. After dealing with duplicate data, a total of 213 cases of EIs were collected.

Influencing Factors
According to the analysis of 213 cases in China, each EI was the result of multiple factors, not a single factor. The multiple factors make the EPEM system uncertain and opaque overall, reducing the ability of managers to reduce risk and exacerbating the consequences [22]. Based on collected cases and related literatures, influencing factors of EIs were identified and described in four main areas, i.e., passenger, escalator, environment, and management.

Passenger Factors
Basir et al. conducted research on escalator incidents and found that the main factors in contributing to incidents were passenger factors which are dynamic and complicated [23]. Traditional linear risk model is insufficient to explore the impact of passenger factors on incidents [24]. In this paper passenger factors were analyzed in two dimensions: passenger characteristic and passenger behavior. Passenger characteristic reflects reaction capacity and action capability of the passenger, which affect the safety level of the EPEM system. Passenger characteristic includes physical state such as age and gender, and other features such as mental condition, clothing and safety education level. Escalator-related injuries are believed to be closely related to passenger age, gender and mental condition [1]. Passenger behavior impacts public transportation safety, which has been widely discussed in previous studies [25][26][27][28][29]. Passenger factors were comprehensively considered from eight perspectives: age and gender, mental or physical condition, outfits and belongings, behavior, knowledge about escalators, safety awareness, safety cognition, and safety skill.

Escalator Factors
Escalator factors refer to the factors related to the operation of escalators which might have encouraged, facilitated, or enabled the incident to happen, including type, size, materials, service term, labeling, safety devices, malfunctions and maintenance. Al-Sharif et al. found that escalator braking should be set at 1.16 m·s −2 in order to eliminate the risk of passenger falls caused by unplanned stoppages [30]. Escalator factors were analyzed from four aspects including settings and components, operating status, safety devices, and maintenance.

Environmental Factors
Environmental factors are the factors related to the surroundings of escalators which might have contributed to EIs. The operating environment of escalator is rather complex and has many aspects, such as location, surrounding object, ground, lighting and visibility, and signage. By analyzing previous research results, installation location, size of space, signage, and physical environment were selected as the main environmental factors to be analyzed.

Management Factors
When implementing complex system safety control, human factors, management factors, and organizational factors should be fully considered. Organizational management, safety culture, and employee participation are important to prevent safety incidents in companies [31]. Management errors repeatedly appear in safety incidents. Komljenovic et al. found that analyzing management weakness is useful for probing underlying factors of safety incidents [32]. These studies suggest the importance of management factors in incident prevention. Safety performance has a close relationship with safety culture [33]. Nordlöf et al. concluded that safety culture is closely related to occupational health and safety management based on the statistical analysis [34]. This paper analyzed the effect of safety management from six aspects including safety rules, safety supervision, information exchange, safety culture, emergency plan, and safety education. Twenty-two influencing factors of EIs are summarized and listed in Table 3.

S 12 Maintenance
The maintenance aspect covers engineering areas such as the periodic preventative maintenance and remedial works as well as items such as cleaning (to prevent accumulation of dust and dirt that can lead to escalator fires).

Method
ISM-DEMATEL is a systems analysis approach utilizing matrix and graph theory to analyze the relationship between system factors. The combination of ISM and DEMATEL can not only identify the influence degree and importance of factors in the EPEM system, but also comprehensively analyze the hierarchical structure of the system. The framework of the ISM-DEMATEL method is illustrated in Figure 1. The Equations (1)- (12) for calculating the matrices are listed below. The detailed steps of ISM-DEMATEL method can be found from the related literatures [16,18].

Establishment of Direct Influence Matrix
Based on case study and literature analysis, 22 factors influencing EIs, denoted as S 1 , S 2 , S 3 , . . . , S n were identified in Section 2.
Influence relationships between these factors are determined through Delphi method. A 3-point Likert-type scale was designed. Twenty questionnaires were collected from safety management experts in universities, safety managers in subway stations and escalator manufacturers in Wuhan. Questionnaire data were averaged, and the average values were rounded-off to determine the measure of the direct relationship between every two factors. The direct influence matrix X was obtained and is shown in Table S1 in Supplementary Materials. x ij = 0 indicates that factor S i has no influence on factor S j . x i j = 1 suggests that S i has a weak influence on S j and x ij = 2 means that S i has a strong influence on S j . x ij = x ji = 0 when i = j.

Establishment of Comprehensive Influence Matrix
A direct influence matrix reflects the direct relationship between influencing factors. However, considering just the direct relationship is inadequate, since multiple factors affect the safety risk in the EPEM system. A comprehensive influence matrix reflects the comprehensive relationship among system factors, including both direct and indirect relations. The comprehensive influence matrix T was calculated and is reported in Table S2.

Calculation of Influencing Degree, Influenced Degree, Centrality, and Causality
Based on the comprehensive influence matrix, influencing degree, influenced degree, centrality and causality of each factor were calculated via Equations (4)- (7). The results are listed in Table 4.
The cause-and-effect relation diagram of the influencing factors is drawn using n i as the x-axis and m i as the y-axis, as shown in Figure 2.
The influencing degree and influenced degree of the factors clearly seen from Table 4. Eight factors have large influencing degrees, including safety education S 22 , safety rules S 17 , settings and components S 9 , safety supervision S 18 , safety culture S 20 , safety awareness S 6 , information exchange S 19 , and knowledge about escalators S 5 . These factors are not clearly visible but are very important, as they can affect escalator passenger safety by influencing other factors. Actions should be taken with respect to these factors so as to make a positive change in other factors and eventually lead to a higher safety level of the EPEM system. In the influenced degree ranking, behavior S 4 , emergency plan S 21 , maintenance S 12 , safety cognition S 7 , safety devices S 11 , and operating status S 10 are the top six factors, indicating that enhancing their influence has a direct impact on improving passenger safety.
Centrality m i reflects the position and importance of the factor S i in the system. Behavior S 4 is the most important factor, followed by emergency plan S 21 , safety rules S 17 , safety supervision S 18 , information exchange S 19 , safety culture S 20 , safety education S 22 , and safety awareness S 6 . The centrality value of these factors is greater than 2.10 in Table 4, and these factors occupy the top region of Figure 2. Two of them are passenger factors and the remaining four are management factors, suggesting that strengthening safety management and enhancing safety awareness are prime means of controlling EI risk.  The causality value of the cause factors is greater than 0, and these factors are on the right region of Figure 2. Half of the 22 factors are cause factors, including safety education S 2 , settings and components S 9 , safety rules S 17 , safety supervision S 18 , installation location S 13 , size of space S 14 , safety culture S 20 , knowledge about escalators S 5 , physical environment S 16 , safety awareness S 6 , and information exchange S 19 , while the rest are effect factors. The cause factors are of great importance for system safety, and investment in them tends to create compound returns.

Establishment of Reachability Matrix
The reachability matrix K was obtained via the total influence matrix and the threshold value λ. By substituting λ into Equation (9), the weak relationships of the factors were eliminated. The reachability matrix was constructed and is reported in Table S3.

Establishment of Hierarchical Structure Model
Reachable set R i , antecedent set A i and collective set C i of each factor were obtained via Equations (10)- (12). The results are represented in Table 5. In Table 5, factors S 3 , S 4 , S 10 and S 11 , Ci = Ri (i = 3, 4, 10 and 11), S 3 , S 4 , S 10 , and S 11 were classified into L 1 . The i-th (i = 3, 4, 10, and 11) rows and columns were deleted to create a new reachability matrix. For factors S 15 and S 16 in this new matrix, Ci = Ri (i = 15 and 16), then S 15 and S 16 were classified into L 2 . This step was repeated until all influencing factors in the EPEM system were classified into 5 layers: L 1 , L 2 , L 3 , L 4 , and L 5 . L 1 denotes the direct cause layer with four factors such as outfits and belongings S 3 , behavior S 4 , operating status S 10 and safety devices S 11 . Signage S 15 and physical environment S 16 are in the indirect cause layer L 2 . L 3 and L 4 are the transition cause layers. L 3 has eight influencing factors including age and gender S 1 , mental or physical condition S 2 , safety cognition S 7 , safety skill S 8 , maintenance S 12 , installation location S 13 , size of space S 14 and emergency plan S 21 . L 4 consists of three factors including knowledge about escalators S 5 , safety awareness S 6 and information exchange S 19 . L 5 is the root cause layer, including five factors that are settings and components S 9 , safety rules S 17 , safety supervision S 18 , safety culture S 20 , and safety education S 22 .
Finally, a hierarchical structure model was established and is illustrated in Figure 3, which clearly reflects the hierarchical structure of the influencing factors and the related influencing mechanisms. (1) Management priority and preventive measures should be arranged according to the hierarchy level. The rule is that the higher the hierarchy level, the more attention deserved. Root cause factors in L 5 are the most fundamental and important factors in the EPEM system. Accordingly, optimization of escalator settings and components, strengthening of safety supervision, and implementation of safety education must be emphasized as the top priority for creating long-term system safety. Transition factors in L 4 and L 3 connect the root cause layer and the direct cause layer in the system. Enhancing these factors helps to form a positive loop for higher safety levels. Outfits and belongings S 3 , behavior S 4 , operating status S 10 and safety devices S 11 are the direct causes of EIs. These factors should be closely monitored and promptly identified with effective and timely responses so as to prevent the occurrence of EIs.
(2) Factors influencing system safety are interactive, and the interaction should be considered when implementing preventive measures. Factors directly or indirectly influence the factors in lower levels. Factors in the same level also influence each other. In particular, four environmental factors in L 2 and L 3 include installation location S 13 , size of space S 14 , signage S 15 , and physical environment S 16 influence the factors in L 1 but are less influenced by other factors. Special attentions should be paid to these factors, and targeted measures should be conducted to mitigate the environment risk.
(3) Factors belonging to the same category are distributed on the same or adjacent hierarchies. Four of management factors include safety rules S 17 , safety supervision S 18 , safety culture S 20 , and safety education S 22 are in L 5 , and information exchange S 19 and emergency plan S 21 are in L 4 and L 3, respectively. Passenger characteristic factors are distributed in L 4 and L 3 , and passenger behavior factors in L 1 . Environmental factors are located in L 3 and L 2 . The exception is escalator factors. One is distributed in L 4 , two in L 3 , and two in L 1 . A general rule is that management factors affect passenger characteristic factors and environmental factors, and characteristic factors and environmental factors affect passenger behavior factors and escalator factors.

Academic Implications
The hierarchical structure of factors influencing EIs were deduced using the proposed method. With DEMATEL, influencing factors were divided into cause factors and effect factors, and the importance of all factors is identified. With ISM, the nonlinear and complicated EPEN system was decomposed into five hierarchies according to the relationships of influencing factors.
The distinguishing features of the ISM-DEMATEL method are that it is systematic and quantitative. First, the method stems from system theory, and the analysis is summarized in a hierarchical structure model. The five hierarchies of factors provide a holistic scenario for understanding the EPEN system. Second, the analysis is based solely on mathematical equations.
Additionally, the influencing factors in this paper were determined based on both news reports cases and related literatures. The combination of two data sources to identify factors is more objective and authentic than reliance on single data source.

Managerial Implications
The hierarchical structure provides a visualization of interrelationships and interdependences among the influencing factors of EIs. It can serve as a useful reference for escalator incident prevention.
The corresponding five-layer countermeasures are proposed to reduce escalator-related injuries.
(1) The top layer measures are to implement safety education and to create a safety climate. A prime example is a more-than-ten-year-old education program called the "Safe-T rider program" in the U.S. that educates school children about how to behave on elevators and escalators. Government management departments of safety in China can consider conducting a similar national escalator safety education program utilizing multi-media and field education modules. Additionally, the public can be educated through publicity campaigns and educational videos. Educational videos and public service announcements on escalator safety can be filmed and broadcasted on mainstream media and social media such as TV channels and Weibo. These efforts should target children and elders. Escalator safety knowledge should be compiled into the safety education textbook for primary students. For elders, watching safety education performances organized by communities is a relaxing and effective way to learn about escalator-related knowledge.
(2) The fourth layer measures are standardizing escalator safety rules and strengthening safety supervision. New standard rules of riding escalators should be formulated and implemented. Escalator safety rule-makers should consist of professionals with escalator safety knowledge and practice, such as escalator technicians, safety managers, safety experts and academics. The formulation of safety rules should refer to the internationally established safety rules for escalators and, more importantly, should be based on the characteristics of passengers and incidents found in safety management practice in China. For example, the rule of walking on the left and standing to the right that originated in the U.K. has been accepted and practiced by escalator passengers worldwide. However, this rule has recently been banned in some cities in China, like Shanghai, Nanjing and Guangzhou, as the practice proven to be unsafe for passengers and damaging to the escalator. Under these circumstances, prohibition of walking on escalators should be written into the safety rules and implemented nationwide. Ensure passengers understand and follow safety rules through communication and supervision.
(3) The third layer measures are to optimize escalator settings and components and to strengthen maintenance. For escalator manufacturers, the construction and installation of escalators must strictly accord with the relevant international standards. The investment in research and development of safety devices should be encouraged since escalator safety could be much improved through the use of high-quality safety devices and equipment. Narrowing the gap between the step and balustrade skirt is another improvement worthy of investment that can reduce the occurrence of the common injury of entrapment. For escalator owners, maintenance on escalators should occur frequently. Quality escalator combined with a robust maintenance program should enable the equipment to last for thirty years. Reliable maintenance helps to reduce overall operating costs and reduce the risk to escalator passengers.
(4) The second layer measures focus on enhancing the management of the environment around escalators. When installing escalators, the specific location of the escalator should be optimized to provide a comfortable and safe environment for passengers. Wet floors and obstacles should be cleaned up in a timely manner. Signage with safety rules should be placed at the entrance to every escalator. Broadcasting escalator safety messages can alert passengers and enhance their safety.
(5) The first layer measures focus on correcting passenger behavior. Passengers are supposed to actively participate in safety education program, learn more escalator safety knowledge and behave properly on escalators. Some of the escalator injuries, such as stumbles and entrapments, are age-specific. Hence, the preventive measures should also be age-specific. Young children should be supervised by parents while riding escalators. They also need to be informed that escalators are dangerous and are not places for jumping or playing. Elders, unhealthy passengers and passengers carrying large luggage or strollers should take elevators instead of escalators. Passengers who wear high-heel shoes, Crocs, long scarves, and long shirts should be on high alert to avoid being entrapped.

Conclusions
This study revealed the five-hierarchy structure of factors influencing EIs in China. Importantly, root cause factors, including settings and components, safety rules, safety supervision, safety culture, and safety education, were identified and should be given priority when taking preventive measures. The hierarchical structure model suggests that occurrence of EIs can be reduced in the short-term by correcting passenger behavior and in the long-term by enhancing safety education. The proposed five-layer preventive measures provide both theoretical and practical support for managing the safety of escalator passengers. The findings contribute to passenger safety research by offering a deeper understanding of interactive influencing mechanisms of the EPEM system.