Analysis of Human Factors Relationship in Hazardous Chemical Storage Accidents

Human factors are important causes of hazardous chemical storage accidents, and clarifying the relationship between human factors can help to identify the logical chain between unsafe behaviors and influential factors in accidents. Therefore, the human factor relationship of hazardous chemical storage accidents was studied in this paper. First, the human factors analysis and classification system (HFACS), which originated from accident analysis in the aviation field, was introduced. Since some items were designed for aviation accident analysis, such as the item “Crew Resource Management”, it is not fully applicable to the analysis of hazardous chemical storage accidents. Therefore, this article introduced some modifications and changes to make the HFACS model suitable for the analysis of hazardous chemical storage accidents. Based on the improved HFACS model, 42 hazardous chemicals storage accidents were analyzed, and the causes were classified. After analysis, we found that under the HFACS framework, the most frequent cause of accidents is resource management, followed by violations and inadequate supervision, and finally the organizational process and technological environment. Finally, according to the statistical results for the various causes of accidents obtained from the improved HFACS analysis, the chi-square test and odds ratio analysis were used to further explore the relevance of human factors in hazardous chemical storage accidents. The 16 groups of significant causal relationships among the four levels of factors include resource management and inadequate supervision, planned inappropriate operations and technological environment, inadequate supervision and physical/mental limitations, and technological environment and skill-based errors, among others.


Introduction
Hazardous chemical warehouses and storage tanks are locations for the storage and maintenance of hazardous chemicals such as chemical raw materials, chemical drugs, chemical reagents, pesticides, etc. Because of the large quantity, variety and high risk of hazardous chemicals, the potential storage hazards can exceed those of the production, transportation and use of hazardous chemicals. Human factors play an important role in the occurrence of hazardous chemical accidents in China [1,2]. Therefore, it is necessary to analyze and identify human factors in hazardous chemical storage accidents. In addition, this paper continued the research in reference [3], combining fault tree analysis (FTA) with human factors analysis and classification system (HFACS) model to analyze multiple hazardous chemical storage accidents. Based on this approach, we further explored the relationship between human factors in hazardous chemical storage accidents using the chi-square test and odds ratio analysis. According to the above methods, this paper realized the quantitative analysis of human factors.
It should be noted that the human factors studied in this paper not only refer to individual behaviors that directly lead to an accident, but also include other organizational factors such as organizational supervision and resource management, because individual people are not isolated and act as members of an organization. As a result, individual behavior is affected by other people, technology and the organization, and these factors restrict and influence each other. Therefore, the study of human factors should consider individual factors as well as organizational factors related to human behaviors.
Currently, certain human factors and proposed relevant models are available, including the software-hardware-environment-liveware (SHEL) model [4], Swiss cheese model [5] and the HFACS model [6]. Among these, the HFACS model has been widely recognized and adopted in many industries. Dekker noted that the HFACS model is the most powerful tool for human factors analysis of various accidents [7]. In the field of aviation, Shappell analyzed the data from 1020 aviation accidents in the United States and found that the majority of accidents were caused by the aircrew and environment, and the number of accidents related to supervision and organizational reasons was significantly reduced [8]. Daramola used the HFACS model to analyze aviation accidents in Nigeria and concluded that the most common causes of accidents were skill-based errors, the physical environment and inadequate supervision. Supervision violation to crew resource management to decision errors was considered the most likely path to accidents [9]. Michal et al. used the accident analysis method combining HFACS with a systems-theory accident model and processes (STAMP) to analyze the Überlingen air accident and confirmed the feasibility of the STAMP-HFACS analysis method [10]. Rashid et al. proposed the Human Factors Analysis and Classification System-Maintenance Extension (HFACS-ME) model for helicopter maintenance accidents and statistically analyzed 58 helicopter maintenance accidents to study the survival rate of helicopter maintenance accidents and the distribution of accident severity [11].
The application of HFACS in accident research also includes coal mining, maritime, medical, railway, chemical and other industries. For example, Patterson and Shappell used the HFACS-Mining Industry (HFACS-MI) model to analyze 508 coal mine accidents in Queensland and concluded that skill-based errors are the most common unsafe behavior, with no significant difference between different types of mines [12]. Chauvin et al. analyzed the human factors and organizational factors of ship collision accidents in Britain and Canada using the improved HFACS model [13]. The analysis showed that most collision accidents were caused by decision errors. Baysari et al. analyzed railway accidents in Australia using the HFACS and Technique for the Retrospective and Predictive Analysis of Cognitive Errors (TRACEr) methods and suggested the effectiveness of the two methods [14]. However, each tool seems to ignore certain important factors related to the occurrence of errors. Cohen et al. used HFACS-Healthcare to identify systemic vulnerabilities during surgery [15]. Hale et al. used the HFACS model to analyze 26 fatal building accidents and found deficiencies in planning and risk assessment, hardware design, purchase and installation, and contracting strategies [16]. In the chemical industry, Gong and Fan analyzed the "11·13" explosion accident at the PETROCHINA Jilin petrochemical biphenyl factory using HFACS and classified the human factors that led to the accident, confirming the usefulness and feasibility of the HFACS for accident analysis in the chemical industry [17]. Zhou et al. improved the items of HFACS and used the improved HFACS to analyze the "8·12" Tianjin Binhai New Area explosion accident [18]. That research showed that the interaction between different levels of human factors in the Ruihai company led to the accident, and the accident investigation report displayed limitations in the identification of human factors and guidance for similar accident prevention. A review of the main relevant information of the HFACS is shown in Table 1. Deficiencies in planning and risk assessment, hardware design, purchase and installation, and contracting strategies during building construction were found [16] Gong and Fan Chemical industry HFACS analysis method Classified the human factors that led to the accident, confirming the usefulness and feasibility of the HFACS for accident analysis in the chemical industry [17] Zhou et al.

Chemical industry HFACS analysis method
The interaction between different levels of human factors in the Ruihai company led to the accident, and the accident investigation report displayed limitations in the identification of human factors and guidance for similar accident prevention [18] The HFACS model addresses the defects of the Reason model and gives a specific definition of the loopholes in each layer of the Reason model, which is more conducive to the study of the classification and mechanism of human factors. However, there are many reasons that exist for hazardous chemical storage accidents [2,19,20], and qualitative analysis of the causes of hazardous chemical storage accidents alone is not sufficient to ensure the effectiveness of the analysis results. Therefore, it is necessary to use other methods to quantitatively analyze the cause of an accident. For this reason, this paper collected information from 42 hazardous chemical storage accidents. The collected accident data were summarized, including the date of accident, enterprises, type of accident and number of deaths. The main sources of accident data include the national and local emergency management departments at all levels, the websites of local governments, the official website of the China Chemical Safety Association, and the chemical registration center of the Ministry of Emergency Management of the People's Republic of China. Then, according to the characteristics of hazardous chemical storage accidents, the HFACS model was modified to make it more suitable for the analysis of hazardous chemical storage accidents. Second, the improved HFACS model was applied to accident analysis. In this way, the causes and high-frequency human factors of accidents under the HFACS framework were obtained. Finally, a chi-square test and odds ratio analysis were used to test the significance and relevance between the four levels of factors under the improved HFACS framework.

Improving the HFACS Model
Based on the Reason model, HFACS defines the dominant and implicit factors that cause accidents in the Reason model and describes four levels of human error: (1) Unsafe Acts, (2) Preconditions for Unsafe Acts, (3) Unsafe Supervision, and (4) Organizational Influences [6]. However, some items in the original HFACS framework might not correspond to the causes of hazardous chemical storage accidents, e.g., "Crew Resource Management", "Routine Violation", "Exceptional Violation" and other subcategories. Therefore, combined with the characteristics of hazardous chemical storage accidents, this paper made appropriate improvements to the original HFACS model. In this manner, a modified HFACS model was established which is more suitable for the analysis of hazardous chemical storage accidents.
"Crew Resource Management" in the original HFACS model is changed to "Communication and Coordination". Crew resource management is a professional term used in the field of aviation and usually refers to problems such as poor information communication and lack of team cooperation between the aircraft and air traffic control during task execution. Thus, in the HFACS, "Crew Resource Management" essentially refers to the problem of communication and coordination. In the storage of hazardous chemicals, if the information exchange among the superiors, subordinates, or employees of the enterprise is poor and the cooperation between teams is ineffective, unsafe behaviors also occur. Therefore, "Crew Resource Management" in the original HFACS model was changed to "Communication and Coordination". In addition, from the hazardous chemical storage accident investigation report, it is impossible to determine whether front-line employee violations are "routine" or "exceptional". As a result, this paper combined the two types of violations into one type: Violations. The specific meanings of the items in the HFACS model of hazardous chemical storage accidents are identified in Figure 1.

Chi-Square Test and Odds Ratio Analysis
After using the improved HFACS model to analyze the frequency of each accident cause, we used the chi-square test (χ 2 ) and odds ratio analysis to analyze the relevance among the four levels of factors in the HFACS framework. In statistics, the χ 2 test is often applied for relevancy testing of nonparametric data variables and the analysis of fixed type data. The odds ratio (OR) is used to measure the characteristic value of the relevance between occurrences of attribute A and attribute B in a specific group [21].

Chi-Square Test
First, we used the χ 2 test to analyze whether a significant causal relationship exists between different factors at the upper and lower levels of the improved HFACS model. The original hypothesis (H0) was proposed: there is no significant causal relationship between the upper and lower level factors in the improved HFACS model. The alternative hypothesis (H1) was also proposed: there is a significant causal relationship between the upper and lower level factors in the improved HFACS model. Because only two factors at a time are selected for correlation analysis, the relevant frequency statistics are calculated in the form of a 2 × 2 contingency table, and the χ 2 value was calculated. The 2 × 2 contingency table for the calculation of the χ 2 value is shown in Table 2.

Chi-Square Test and Odds Ratio Analysis
After using the improved HFACS model to analyze the frequency of each accident cause, we used the chi-square test (χ 2 ) and odds ratio analysis to analyze the relevance among the four levels of factors in the HFACS framework. In statistics, the χ 2 test is often applied for relevancy testing of nonparametric data variables and the analysis of fixed type data. The odds ratio (OR) is used to measure the characteristic value of the relevance between occurrences of attribute A and attribute B in a specific group [21].

Chi-Square Test
First, we used the χ 2 test to analyze whether a significant causal relationship exists between different factors at the upper and lower levels of the improved HFACS model. The original hypothesis (H0) was proposed: there is no significant causal relationship between the upper and lower level factors in the improved HFACS model. The alternative hypothesis (H1) was also proposed: there is a significant causal relationship between the upper and lower level factors in the improved HFACS model. Because only two factors at a time are selected for correlation analysis, the relevant frequency statistics are calculated in the form of a 2 × 2 contingency table, and the χ 2 value was calculated. The 2 × 2 contingency table for the calculation of the χ 2 value is shown in Table 2. Table 2. Calculation of the chi-square (χ 2 ) value: 2 × 2 contingency table.

High-Level Factors Row Sum Exist None
Exist In Table 2, nij represents the actual observed value, i.e., the actual statistical value. fij represents the theoretical observation value, which means the expected value under the assumption that the two variables are uncorrelated. It should be noted that the meanings of n 11 , n 12 , n 21 and n 22 are as follows: (1) if the tested high-level factors and low-level factors occur at the same time in an accident, it is recorded as one time, and the cumulative value is n 11 ; (2) if the tested high-level factors in an accident do not appear but the low-level factors appear, it is recorded as one time, and the cumulative value is n 12 ; (3) if the tested high-level factors in an accident appear but the low-level factors do not appear, it is recorded as one time, and the cumulative value is n 21 ; (4) if the high-level factors and low-level factors are not found in an accident, it is recorded as one time, and the cumulative value is n 22 . In particular, for a 2 × 2 contingency table, if A, B, C and D represent the actual observation times n 11 , n 12 , n 21 and n 22 respectively in four cells, then the chi-square value can be calculated by the following formula: The p value can be obtained by looking up the value in the table when the degree of freedom (df) = 1. The p value has the following statistical significance: when p > 0.05, we should accept the original hypothesis (H0) and reject the alternative hypothesis (H1), and when p < 0.05, we should reject the original hypothesis (H0) and accept the alternative hypothesis (H1).

Odds Ratio Analysis
For a 2 × 2 contingency table, the formula for calculating the OR value is given as follows: and the relationship between the occurrence of upper factors and the occurrence of lower factors in the HFACS model is determined after obtaining the OR value: when the OR value is greater than 1, it indicates that the occurrence of upper factors in the HFACS model can increase the occurrence possibility of lower factors, and when the OR value is less than 1, it indicates that the occurrence of upper factors cannot increase the occurrence possibility of lower factors.

Analysis of Hazardous Chemical Storage Accidents
To more clearly show the causes of hazardous chemical storage accidents and their logical relationships with each other, we restored the development and evolution process of accidents using a combination of the fault tree analysis (FTA) method and the HFACS model to analyze the accidents in depth. The detailed analysis steps are given in the literature [3]. We used this method to analyze 42 hazardous chemical storage accidents. The details of the accidents are shown in Table 3.  After the analysis, the accident causes were classified and statistically analyzed. The frequency and percentage of each accident cause were obtained under the improved HFACS framework, as shown in Table 4.

Using χ 2 /OR to Analyze the Relevance of Human Factors in Hazardous Chemical Storage Accidents
To conduct a quantitative study of human factors in hazardous chemical storage accidents, the relevance among the factors in 42 accident cases was further studied based on the HFACS analysis. In this paper, a chi-square test and odds ratio analysis method were used to test the significance and relevance of four level factors in the improved HFACS framework, so as to realize the key step of human factor quantitative analysis.
Taking the calculation of the relevance between Communication and Coordination and Decision Errors in the improved HFACS model as an example, we calculated the χ 2 value and OR value. Original hypothesis (H0): there is no significant causal relationship between Communication and Coordination and Decision Errors. Alternative hypothesis (H1): there is a significant causal relationship between Communication and Coordination and Decision Errors. Table 5 shows the statistical results of the frequency of occurrence in accidents. The χ 2 value and OR value were calculated using Equations (1) and (2). By calculation, χ 2 = 4.582 and OR = 4.000 > 1, and by df = 1, we can obtain p = 0.032 < 0.05. Therefore, we rejected H0 and accepted H1. This means that there was a significant causal relationship between Communication and Coordination and Decision Errors. In addition, the OR value greater than 1 indicated that the occurrence of Communication and Coordination can increase the possibility of Decision Errors.
We used the above method to analyze the relevance among the four levels of factors in the improved HFACS model. We screened the causal relationships between different levels of factors that satisfy p < 0.05 and OR > 1, and eliminated the causal relationships of human factors that did not meet the conditions, e.g., Resource Management and Planned Inappropriate Operations, Organizational Climate and Failure to Correct Problem, Inadequate Supervision and Technological Environment, Failure to Correct Problem and Physical Environment, Personal Readiness and Perceptual Errors, etc. Thus, we obtained the results shown in Table 6. According to the inspection results, the causal relationship diagram of human factors in hazardous chemical storage accidents was obtained, as shown in Figure 2. value greater than 1 indicated that the occurrence of Communication and Coordination can increase the possibility of Decision Errors. We used the above method to analyze the relevance among the four levels of factors in the improved HFACS model. We screened the causal relationships between different levels of factors that satisfy p < 0.05 and OR > 1, and eliminated the causal relationships of human factors that did not meet the conditions, e.g., Resource Management and Planned Inappropriate Operations, Organizational Climate and Failure to Correct Problem, Inadequate Supervision and Technological Environment, Failure to Correct Problem and Physical Environment, Personal Readiness and Perceptual Errors, etc. Thus, we obtained the results shown in Table 6. According to the inspection results, the causal relationship diagram of human factors in hazardous chemical storage accidents was obtained, as shown in Figure 2.

Results Analysis
According to the four levels of the improved HFACS framework (including Organizational Influences, Unsafe Supervision, Preconditions for Unsafe Acts, Unsafe Acts) and the results of the χ 2 test and OR analysis, the results of quantitative calculation were analyzed as follows.

Defect of Organizational Influences
Organizational influences include Resource Management, Organizational Climate and Organizational Process. From Table 5, inadequate Resource Management has the greatest impact on Inadequate Supervision. In other words, if resource management of a hazardous chemical storage enterprise is poor, the possibility of insufficient supervision will increase to 20.7 times (OR = 20.667). Resource management loopholes are primarily reflected in human resources, equipment and facilities resources, funds, and other aspects. The specific performance issues are unreasonable personnel allocation, lack of a qualification examination system for special operation personnel, poor quality of safety management personnel, insufficient equipment and facilities, or quality defects. The poor organizational climate also leads to the occurrence of supervisory violations and inadequate supervision. A poor organizational climate will increase the possibility of Inadequate Supervision to 4.5 times (OR = 4.455) and the possibility of Supervisory Violations to 7.8 times (OR = 7.800). Poor organizational climate includes insufficient safety investment, insufficient risk management policies, "focusing on efficiency, ignoring safety", and poor safety culture.
In addition, Organizational Process loopholes have a significant impact on Planned Inappropriate Operations and Supervisory Violations in Unsafe Supervision. Organizational process loopholes will increase the probability of Planned Inappropriate Operations to 6.2 times (OR = 6.231) and the probability of Supervisory Violations to 4.8 times (OR = 4.848). Organizational process loopholes are mainly reflected in systems, procedures, production supervision and other aspects. Examples of this include where the enterprise has not formulated a specific safety management system or the system is incomplete, the regulatory system has loopholes, the organization and management of the site are disordered, and relevant operation instructions are lacking.

Unsafe Supervision
Unsafe Supervision includes Inadequate Supervision, Planned Inappropriate Operations, Failure to Correct Problem and Supervisory Violations. It can be observed from Table 5  Planned Inappropriate Operations has a significant impact on the Technological Environment and Communication and Coordination in the Preconditions for Unsafe Acts level. Planned inappropriate operations will increase the possibility of a poor technological environment to 4.7 times (OR = 4.722) and poor communication and coordination to 4.3 times (OR = 4.333). Planned inappropriate operations is manifested as improper collocation between team members or authorization of unqualified team members for work, resulting in insufficient cooperation and communication among team members. In addition, planned inappropriate operations also refers to the improper allocation of resources, and thus it might lead to differences in equipment allocation among different teams, resulting in the risk of a poor technological environment. Failure to correct the problem will increase the probability of technological environmental problems to 4.3 times (OR = 4.275). Failure to correct the problem refers to the failure of the supervisor to find problems or correct the problems in time in the hazardous chemical storage process, resulting in the continuous existence of hazards. Poor technological environment refers to equipment and facility failures, lack of protective devices, lack of electronic monitoring facilities, unreasonable control design, etc. Therefore, it is easy to increase the risk of a poor technological environment if problems are not found or not solved in time, or hazards are not investigated adequately.

Preconditions for Unsafe Acts
Preconditions for Unsafe Acts includes the seven aspects of Physical Environment, Technological Environment, Adverse Mental States, Adverse Physiological States, Physical/Mental Limitations, Communication and Coordination, Personal Readiness, among others. However, for the 42 hazardous chemical storage accidents collected, only Technological Environment, Physical/Mental Limitations, Communication and Coordination, and Personal Readiness have a significant impact on Unsafe Acts. Among these, the lack of personal readiness is one of the main reasons for unsafe acts, especially for skill-based errors and violations. The lack of employee personal readiness can increase the probability of skill-based errors to four times (OR = 4.000) and increase the probability of violations to 13 times (OR = 13.000). Personal readiness refers to a lack of knowledge and skills for the related hazardous chemicals or a lack of physical strength and energy of the front-line workers before work. The main manifestations are insufficient knowledge about hazardous chemicals, a lack of mastery of skills required by the position, failure to wear personal protective equipment (PPE), insufficient rest, etc. Therefore, the lack of personal readiness will inevitably increase the possibility of skill-based errors and violations.
Physical/Mental Limitations and Communication and Coordination have a significant impact on the occurrence of Decision Errors. Physical or mental limitations can increase the probability of decision errors to 4.1 times (OR = 4.083). Communication and coordination can increase the probability of decision errors to 4 times (OR = 4.000). Physical/mental limitation refers to a lack of experience and the ability of employees to function in complex situations. Communication and coordination refers to insufficient cooperation among team members and lack of information exchange between superiors and subordinates. If these two factors are defective, it will inevitably lead to decision errors under different situations. Decision errors mainly refer to emergency judgment errors, emergency response errors, improper selection, problem handling errors, etc. In addition, the technological environment also has a significant impact on skill-based errors (OR = 9.000). A poor technological environment may lead workers to be unfamiliar with important equipment and ignore operational details, resulting in skill-based errors.

Unsafe Acts
Unsafe Acts includes Skill-based Errors, Decision Errors, Perceptual Errors and Violations. According to the statistics of unsafe acts in 42 hazardous chemicals storage accidents, violations by front-line workers account for the largest proportion, reaching 85.714%, followed by decision errors and skill-based errors, accounting for 52.381% and 47.619% respectively, and finally perceptual errors, accounting for 7.143%. Violations mainly refer to the violation of the existing rules and various safety operating procedures and risky operations of front-line workers.
Decision errors refer to the errors caused by improper employee estimation of the situation, including three main types of errors in emergency situations: process errors, selection errors and problem-solving errors. Skill-based error refers to mistakes in skill-related behaviors of employees, mainly including poor operation technology, blind operation blind, improper use of PPE, etc. Perceptual errors are manifested by individual cognition and actual conditions such as visual errors, information understanding errors, wrong directions, etc.
According to the above analysis, the main factors leading to unsafe acts of employees are personal readiness, communication and coordination, and technological environment, whereas unsafe acts are primarily reflected in violations and decision errors. Therefore, managers should enhance training on professional knowledge and skills for front-line operators, improve the professional development of employees, and assure good job preparation to reduce the possibility of unsafe behaviors.

Conclusions
This paper collected the investigation reports from hazardous chemical storage accidents that occurred in China during 2010-2019 in order to establish an improved HFACS model suitable for the analysis of hazardous chemical storage accidents. Through the improved HFACS model analysis, chi-square test and odds ratio analysis, we obtain the frequency of each accident cause and the causal relationships among the four levels of factors in the improved HFACS model. The conclusions are given as follows:

1.
A modified HFACS model was established for human factors analysis of hazardous chemical storage accidents. Some items of the original HFACS model were not fully applicable to the analysis of hazardous chemical storage accidents. Therefore, according to the actual situation and characteristics of the collected hazardous chemical storage accidents, an HFACS model suitable for the analysis of hazardous chemical storage accidents was established.

2.
The high-frequency human factors in hazardous chemical storage accidents were obtained.