Abstract
Safety culture is a very complex phenomenon due to its intangible nature. It is tough to measure and express it with numerical values, as there is no simple indicator to measure it. This paper presents a fuzzy inference system that measures the safety culture. First of all, a safety culture assessment questionnaire is developed by utilizing related literature. The initial questionnaire had 29 items. The questionnaire is applied to 259 employees within the gun manufacturing factory. After making an exploratory factor analysis, the questionnaire is based on five factors with 25 items. The safety culture indicators are defined as; safety follow-up audit reporting, employees’ self-awareness, operational safety commitment, management’s safety commitment, safety orientedness. Normality, reliability, and correlation analysis are performed. Then a fuzzy model is constructed with five inputs and one output. The inputs are the five factors mentioned above, and the output generated is the safety culture result, which is between 0-1. The presented fuzzy model produces reliable results indicating the safety culture level from the employees’ eyes. Beyond exploring the employees’ safety culture, the proposed model can easily be understood by the practitioners from various sectors. Furthermore, the model is straightforward to customize for various fields of industry.
Introduction
The term ‘safety culture’ has been a crucial concept for high-risk industries since it is defined by the International Atomic Energy Agency (IAEA) after the Chernobyl Nuclear Disaster in 1986. The International Nuclear Safety Advisory Group (INSAG) of IAEA presented a report for nuclear installation. The report mentioned the absence of a safety culture, which prepared the suitable conditions for the nuclear disaster. The report significantly influenced the occupational safety field [1].
Although the term ‘safety culture’ was born in the nuclear power generation sector, it is widely accepted by all other industries such as; aerospace, aviation, defense, mining, construction, power generation, logistics, manufacturing, and all kind of process industries. Today, not only the high-risk sectors but also the relatively low-risk business areas are also focusing on developing a positive safety culture in their workplace. Although it is increasingly accepted, safety culture is not a tangible term, so it is impossible to measure it with a culture meter; hence this study is presented. The studies on safety culture offer various methods to measure this phenomenon. The measurement can be made via some interviews with the managers and employees. The other option can be by performing some surveys on employees using questionnaires. Feedback acquired from the questionnaires could help understand employees’ opinions, perceptions, and attitudes [2].
The common idea in different studies is to define some indicators for safety culture. A survey by Grecco et al. proposes six indicators to measure the safety culture: top-level commitment to safety, organizational learning, organizational flexibility, awareness, just culture, and emergency preparedness [2]. Kim et al. present an assessment of safety culture where four indicators, a norm system, a safety management system, safety culture awareness of workers, and worker behavior, are utilized [3]. Warszawska and Kraslawski employ an assessment tree method for safety culture, which proposes the indicators as; knowledge and skills, awareness, information flow, monitoring-control and supervision, management commitment, and continuous improvement [4]. According to the S.L. Morrow et al., there are nine indicators of safety culture. These nine indicators are stated as management commitment to safety, willingness to raise concerns, decision-making, supervisor responsibility for safety, questioning attitude, safety communication, personal responsibility for safety, prioritizing safety, training quality [5]. In another study done by CS do Nascimento et al., twelve indicators (priority given to safety, allocation of resources, roles & responsibilities, safety commitment, qualification & personnel size, communication, relationship with superiors and regulators, feasibility & processes, documentation & procedures, work conditions, organizational learning, and, internal & external evaluations) are presented. [6].
The management’s safety commitment is the critical attribute for strong leadership for successful safety activities. The absence of management support will cause a fail in safety activities. So the first thing is a management style, which encourages the employees for a safer workplace [7, 8]. The management’s role is including but not limited to; acting as a leader in all safety activities, promote people who put effort into developing the positive safety culture, making arrangements for good safety practices, and providing resources for safety activities. Management is the shipmaster of the organization that defines the direction. The organization can be directed either towards a positive or poor safety culture. Thus, we need to state that it is critical for creating a positive safety culture within the organization. It helps to lower the number of incidents and minimize losses. On the other hand, organizations with a poor safety culture can experience environmental, financial, and social losses, severe injuries, and even multiple fatalities. Management’s safety commitment is expected to reflect the operations, and this aspect is called operational safety commitment [9].
Employees’ self-awareness represents the level of consciousness of employees about the risks. Awareness is directly proportional to the safety culture [10]. Various researches mention about the awareness as a key attribute for zero incidences and higher safety performance. Industrial organizations have multiple risks within the workplace. Employees are expected to be aware of the risks and act according to this awareness. In some countries and regular safety training, employees are supposed to take the safety awareness education [2, 11–13].
On the other hand, safety orientedness occurs, which indicates the focality of safety in all activities. There is one simple question before the execution of the job, ‘Is this safe?’ If people are asking this question, this could show the level of safety orientedness. In safety oriented workplaces, all the business is planned with giving priority to safety. In those workplaces, safety is the first within all the operation instructions and work orders. Nobody is allowed to act in an unsafe way. The higher the safety orientedness, the better the safety [14, 15].
As mentioned above and many more, such as safety follow-up, audit, and reporting activities, the aspects are crucial for developing a safety culture within the organization. Thus, every kind of safety issues or any of the smallest near-miss cases must be followed carefully. People must be encouraged to report the incidents instead of sweeping under the carpet. According to Warszawska and Kraslawski, measuring the effectiveness of safety activities, follow-up, monitoring, and audit activities are necessary [4].
This paper aims to introduce a fuzzy model in order to measure the safety culture. A safety culture assessment questionnaire (SCAQ) is utilized to collect the fuzzy model’s input data. Besides, the questionnaire let the authors explore the above introduced 5 factors (safety culture indicators), which are utilized as the presented fuzzy model’s inputs.
This paper is structured as follows: Section 2 presents the materials and methods employed in this work. Section 3 reveals the results obtained and Section 4 concludes this paper.
Materials and methods
As the purpose of this paper is to provide a practical approach to measure the safety culture perceptions of the employees within a gun manufacturing environment, a SCAQ is prepared and applied in a factory where gun parts are manufactured. The data is statistically processed. According to the exploratory factor analysis (EFA), five factors are determined. These factors are the indicators of the safety culture and are used as inputs within the fuzzy model. The proposed fuzzy model processes the inputs and provides a safety culture result. Figure 1 summarizes the fundamental steps of this work.

Steps of the work.
Zohar proposes a way of developing a safety culture questionnaire, including the indicators of safety culture [16]. Various studies are dealing with the assessment of safety culture [17–21]. In addition to the assessment of safety culture, developing the questionnaire itself is another aspect of the work. There are some questionnaires within various studies which would be identical to the purpose of our research [22–25]. However, the handicap of applying a questionnaire directly from the literature is not appropriate and even not a correct action due to the existence of cultural differences. Since it is aimed to use this newly designed questionnaire in Turkey, the authors were very attentive during the development process [26, 27]. So, the questions were structured to ensure that the participants understood each of them with crystal clarity. Various questionnaires of the above literature are also comprehensively examined; thus, it is aimed to go beyond capturing the values of an organization that can be easily assessed via surveys.
Therefore, the proposed questionnaire aimed to answer the fundamental question of “how can we gather information on safety culture?". The questions are adapted by utilizing them so that various aspects considered directly asked for employees’ perceptions & needed their honest, expert opinion. So, questions are prepared in Turkish and translated for this manuscript by the authors as they spent years within the manufacturing & chemical sectors shop floor. The place where the questionnaire was applied is a gun manufacturing factory. Two of the authors are chemical engineers. One of the authors is a mechanical engineer who is very familiar with the shop floor’s risks and working environment. This advantage lets the authors translate and prepare the questions very clearly and understood by the participants. The initial questionnaire is shown in Table 1.
Safety Culture Assessment Questionnaire
Safety Culture Assessment Questionnaire
The SCAQ is carefully utilized to collect the data. The questionnaire had 29 questions, initially. The research is conducted under the permission of company management and the company’s safety specialists’ supervision.
There is a text at the top of the questionnaire; “We aim to assess your safety culture perceptions within your company. So please read each question carefully and provide an answer which best expresses your opinion. The demographic questions are prepared to understand the structure of the population rather than disclosing your identity. You are not supposed to disclose your identity. We will keep all of your answers anonymous. So please feel free while answering all the questions. Please provide only one answer for each question. Your opinions and contributions are very much appreciated. Thank you.” In a questionnaire, all answers staying anonymous are essential for employees, as most worry about losing their job because of their responses.
The company where the questionnaire was performed has 342 employees (the number of total employees is valid at the date of this research and may change within time), and 259 of them willingly accepted to participate in the study where this number also reflects the canceled improperly filled questionnaires. Thus, the participation rate is found to be 75.7%. The research is performed on Sep 27, 2019.
A 5-point Likert scale was used to collect the required responses from the participants. According to the Likert scale; 5 points mean that participant strongly agrees with the question, where 1 point means strongly disagree. Descriptive statistics (mean, standard deviation, skewness, and kurtosis) of individual items on the safety culture questionnaire were obtained. The factorial structure was tested through EFA using principal components as the extraction method and applying a direct oblimin rotation criterion [28]. The reliability is also tested with Cronbach’s alpha coefficient. The analyses were carried out using SPSS 22.
Descriptive statistics
The items in the questionnaire showed standard deviations between 0.594 and 1.055 values. Mostly±1 value is excellent for the kurtosis; however, ±2 value would be acceptable [29]. The average skewness and kurtosis values are 0.88 and 1.22, respectively. The 8th and 18th questions are exceeding this limit slightly, and the 21st question is very close. Therefore, the authors decided to keep these questions within the questionnaire. The mean, standard deviation, skewness, and kurtosis values are presented in Table 2.
Descriptive Statistics
Descriptive Statistics
The factorial structure was tested through EFA using principal components as the extraction method and applying a direct oblimin rotation criterion. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was 0.92, and the Bartlett test of sphericity was statistically significant (p < 0.01), indicating the suitability of these data for factor analytic procedures. Four of the questions were removed (Q4, Q5, Q6, Q7), as they did not load uniquely into any one factor. Also, these items had low communality scores. The results presented here related to the EFA are conducted with the 25 remaining items. The five factors identified (Eigenvalue > 1), comprising 25 of the original 29 items, accounted for 66% of the common variance. The first factor is named as safety follow-up, audit, reporting. The second factor is named as employees’ self-awareness. The third factor is named as operational safety commitment. The fourth factor is named as management’s safety commitment. The fifth factor is named as safety orientedness. The factor structure is presented in Table 3, and factor loadings below 0.40 are not given. Communalities are shown in brackets.
Exploratory Factor Analysis Results
Exploratory Factor Analysis Results
Extraction Method: Principal Component Analysis. Rotation Method: Oblimin with Kaiser Normalization. a. Rotation converged in 35 iterations.
Values of 0.70 or more indicate acceptable reliability for Cronbach’s alpha coefficient [30]. All the reliability analysis results are higher than 0.70 and presented in Table 4, in addition to the factor names, mean values, and standard deviation values. Correlations should be higher than 0.30, and Pearson correlations are presented in the same table [31]. All correlations are significant at p < 0.01 (2-tailed).
Reliability Analysis
Reliability Analysis
T-tests are performed according to some demographic information, and some results are obtained according to the safety culture indicators. The job positions of the employees are defined as blue-collar (N = 220) and white-collar (N = 39). T-tests revealed significant differences between white-collar employees and the blue-collar employees, with higher mean values for blue-collar employees in employees’ self awareness indicator (meanbluecollars = 4.19, meanwhitecollars = 3.91, t = 2.98, p < 0.01) and safety orientedness indicator (meanbluecollars = 3.84, meanwhitecollars = 3.61, t = 2.11, p < 0.01).
According to the gender of the employees, T-tests revealed significant differences between male (N = 227) and female (N = 32) employees, with higher mean values for females in operational safety commitment indicator (meanfemale = 4.03, meanmale = 3.64, t = –3.06, p < 0.01) and safety orientedness indicator (meanfemale = 4.04, meanmale = 3.77, t = –3.16, p < 0.01).
According to the employees’ marital status, there is no significant difference in any of the five indicators between single employees (N = 85) and the rest (N = 174).
According to the question “Did you have any occupational disease or work accident in this workplace?", we collected the responses as yes (N = 33) and no (N = 226). We collected higher mean values for employees who says no for safety follow-up, audit, reporting indicator (meanno = 3.88, meanyes = 3.52, t = –2.12, p < 0.01) and operational safety commitment indicator (meanno = 3.73, meanyes = 3.43, t = –2.40, p < 0.01).
According to the question “Did you have any near-miss in this workplace?”, we collected the responses as yes (N = 54) and no (N = 205), with higher mean values for employees who says no for safety follow-up, audit, reporting indicator (meanno = 3.93, meanyes = 3.48, t = –3.75, p < 0.01), employees’ self awareness indicator (meanno = 4.23, meanyes = 3.83, t = –5.17, p < 0.01), operational safety commitment indicator (meanno = 3.77, meanyes = 3.41, t = –3.25, p < 0.01), management’s safety commitment indicator (meanno = 3.97, meanyes = 3.61, t = –2.95, p < 0.01) and safety orientedness indicator (meanno = 3.89, meanyes = 3.49, t = –3.52, p < 0.01).
Fuzzy logic
Logic-based systems include two values as true and false, which are 1 and 0. This is sometimes inadequate for describing human reasoning in some complex situations, as there are many values between 1 and 0. That’s why fuzzy sets propose linguistic variables and membership functions (MF’s) to explain these situations [32, 33]. This case may be simply explained with an example. According to the crisp logic, if we measure a cup of water’s temperature, it is either hot or cold. But if the water is warm, then it is a problem to decide whether it will be classified as hot or cold. Fuzzy logic proposes a solution to this problem and lets us work beyond the crisp 0 and 1’s.
According to Zadeh’s fuzzy sets [32], the MF of a classic set A is,
and MF of a fuzzy set
where a triangular fuzzy number
then the MF will be denoted as;
This paper employs triangular fuzzy numbers. MF’s are defined according to these triangular fuzzy numbers [36]. Figure 2 shows the graphic demonsration of triangular MF of fuzzy number

Triangular MF of fuzzy number
Safety follow-up audit reporting, employees’ self-awareness, operational safety commitment, management’s safety commitment, and safety orientedness are the inputs of this model, and the output is safety culture. Each input is defined with MF’s [37]. Three fuzzy numbers are defined with their linguistic terms. First is “poor”, the second is “medial,” and the third one is “good”. The range of MF’s is set between 1 to 5. But, it is desired to get the output between 0 to 1. So the output function range is arranged between 0 to 1. Then five linguistic terms are defined for output MF’s. These linguistic terms are “very poor”, “poor”, “medial”, “good”, and “very good”. The input and output MF’s are shown in Figs. 3 and 4, respectively. The users of this model may choose to customize the number of MF’s. For instance, the user may provide 4 or 5 input MF’s instead of 3. The same approach is also valid for output. This would help the users to increase the flexibility and accuracy of the model.

Input MF’s.

Output MF.
Once MF’s are completed, the fuzzy rules are defined according to the if-then logic (i.e., If input1 is P and input2 is Q and input3 is R and input4 is S and input5 is T then output is W). The output will vary according to the sum of the inputs. The sum of the five inputs would be 15 as a maximum and five as a minimum value like below.
Let the lowest point be 1×5 = 5 (This may occur if all the MF’s are provided with 1 point.)
Let the highest point be 3×5 = 15 (This may occur if all the MF’s are provided with 3 points.)
Mamdani type inference is adopted to aggregate the fuzzy rules, and the centroid method is adopted for defuzzification. The question is to decide which input combination will result in which output value. So we need to define a rule to form the rule combinations. In most cases, the rules are determined according to expert opinions. However, the proposed procedure is designated by the authors’ opinions as an expert with years of experience. A method is followed to generate the 243 (35) rules, which are given in Equation (5).
In order to arrange the rules according to the Equation (5), a Matlab code is prepared, and rules are generated automatically. The code is given in Fig. 5.

Matlab code for automatically generating the rules.
Statistical findings
According to the job position, employees’ self-awareness (second indicator) and safety orientedness (fifth indicator) indicators revealed significant differences with higher values of blue-collar employees. According to the gender; operational safety commitment (third indicator) and safety orientedness (fifth indicator) showed significant differences with female employees’ higher values. No significance was seen according to marital status. The question of experiencing a disease or accident (did you have any occupational disease or work accident in this workplace?); the employees who say no, revealed substantial differences in safety follow-up audit reporting (first indicator) and operational safety commitment. According to the question, having a near-miss within this workplace (did you have any near-miss in this workplace?), the employees who say no provided higher scores in all five indicators. According to these results, the people who do not have a near-miss, accident, or disease experience are likely to be more optimistic about the safety culture and provide higher points. Also, blue-collar employees and female employees tend to be more optimistic. This can be simply explained as; the employees who had a near-miss, accident, or disease are more careful and have a questioning attitude. That’s why people who had a near-miss, accident, or illness are more niggard while replying to the questions. This is the same for males while responding to the questions, which we think that the limited number of female employees, and most of them are working in slightly low risky positions. And lastly, the blue-collar employees provide higher points than the white-collar employees. This could be because of the education level of white-collar employees. As white-collar employees have higher educational levels, they may have a more questioning attitude, which results in lower points. Or we can simply say that they have a good safety culture perception about their workplace.
Fuzzy model results
The safety culture points are calculated by using the presented fuzzy model. The total safety culture point is found as 0.734 within the gun manufacturing factory. In order to get more results, the model is used for various demographic data. According to gender, the results are 0.726 and 0.925 for males and females, respectively. According to marital status, the results are 0.717 and 0.780 for married and single employees, respectively. According to the job status, the results are 0.746 and 0.704 for blue-collar and white-collar employees, respectively. According to the occupational disease and work accident history (did you have any occupational disease or work accident in this workplace?), the participants who say ‘yes’ had a result of 0.642 points. The participants who say ‘no’ had a result of 0.763 points. Similarly, according to the near-miss history (did you have any near-miss in this workplace?), the participants who say ‘yes’ had a result of 0.638 points where the participants who say ‘no’ had a result 0.807 points. The researches on safety culture take the acceptable level of safety culture as 0.6, and the results reveal that the general state of the safety culture within the workplace is acceptable [2, 38].
Conclusions
This paper proposes a fuzzy logic approach for measuring the safety culture within the workplace. For that purpose, a SCAQ is presented within this paper and utilized to collect data about the safety culture perceptions of the gun manufacturing factory employees. After making the statistical analysis of the data, 5 factors are determined and it is seen that, the structure of the SCAQ is consistent. The results revealed that; items in the questionnaire seem to have acceptable inter-item consistency, and they also have acceptable validity in measuring what they are intended to measure within the workplace. Then a fuzzy model developed with 5 inputs and 1 output. The fuzzy model is a reliable tool for measuring the safety culture perceptions. The model can be easily modified, based on the user needs. Professionals from various sectors can easily understand how it works and customize it according to their needs. They may increase or decrease the number of inputs if they choose to collect the data with another questionnaire. They may increase or decrease the number of the membership functions and may change the values of them.
One of the important reason for using such a fuzzy model for measuring the safety culture is; its flexibility. Once the data is collected; according to the needs, users may customize the membership functions according to their industry-specific needs and employ the model. The statistical results’ itself are rigid most of the time. This circumstance does not let the safety practitioners to see the alternatives while making their decisions. However, proposed fuzzy model produces very smooth and reliable results once it is constructed. If the safety expert wants to put the scale in a different level and decide to work with another questionnaire, which has 6 or 7 factors, then it is easy to adapt the fuzzy model for 6 or 7 inputs. Also, this SCAQ may be applied periodically within the same workplace to assess the improvements within the safety culture; as long as the final shape of the membership functions and parameters are defined as the same between the two periodical applications.
The limitation of our study is being performed within only one gun factory. The reason for this is due to time and money concerns in addition to the difficulty of finding a second similar workplace. Another limitation is; the nature of the safety culture concept itself. There are many industries with various risks, and it is almost impossible to develop a universal questionnaire that will be valid for all business areas.
Future studies may be performed within other gun manufacturing installations and may focus on improving safety performance with safety culture. The model may be extended to apply in other sectors as well. Before the application, it is strongly advised to understand the risks of the specific sector and customize the SCAQ and the model according to the particular sector’s needs.
Footnotes
Acknowledgments
We are grateful to the gun factory, as they let us in, for this research. The authors do not declare the name of the company due to the confidentiality request of the company. We thank Mr. Abdullah Özgülen for his contributions to the statistical calculations of this research.
