Safety Science 53 (2013) 8295

Contents lists available at SciVerse ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/ssci

Bayesian network analysis of safety culture and organizational culture in a nuclear power plant
Susana Garca-Herrero a,, M.A. Mariscal a,1, J.M. Gutirrez b, Antonio Toca-Otero c,2
a

Escuela Politcnica Superior, Universidad de Burgos, Avda. Cantabria, s/n, 09006 Burgos, Spain Instituto de Fsica de Cantabria (CSIC-UC), Avda. de los Castros, s/n, 39005 Santander, Spain c Santa Ma de Garoa NPP, Valle de Tobalina, 09212 Burgos, Spain
b

a r t i c l e

i n f o

a b s t r a c t
Many high-hazard industries around the world have explicitly recognized the critical role that human, management and organizational risk factors play in major accidents. The ndings of accident investigations and risk assessments demonstrate a growing recognition that the cultural context of work practices may inuence safety just as much as technology. The objective of this paper is to establish a relationship between the concepts of safety culture and organizational culture in a Nuclear Power Plant (NPP). This study permits the identication and quantication of the possible mechanisms for improving the safety culture in the NPP acting on organizational culture. It therefore provides a methodology to identify potential strategies for safety improvement. Probabilistic (Bayesian) Networks (BNs) have been used to determine the relationships between the organizational culture and safety culture in a quantitative form. To this aim, we considered data from a survey conducted of every employee at a Spanish NPP. The resulting data-driven models allow us to establish the probabilistic relationship among organizational culture factors, including the 12 OCI (Organizational Culture Inventory) scales, that have an inuence on safety culture. The study yielded a ranking of organizational cultures that can be used to improve safety culture in a NPP. 2012 Elsevier Ltd. All rights reserved.

Article history: Received 24 February 2012 Received in revised form 23 August 2012 Accepted 2 September 2012 Available online 27 October 2012 Keywords: Safety culture Organizational culture Organizational culture inventory Bayesian network Nuclear industry

1. Introduction Many high-hazard industries around the world have explicitly recognized the critical role that human, management and organizational risk factors play in major accidents (Booth and Lee, 1995; Oien, 2001). An analysis of the main accidents that have taken place throughout history shows that these events cannot be explained by random equipment failures alone, but also by a combination of human and organizational factors. Some of these accidents from the late 1970s and the 1980s include: Three Mile Island accident in 1979 (Kondo, 1996; Le Bot, 2004; Maddox and Muto, 1999); Bhopal gas tragedy on 3 December 1984 (Shrivastava, 1994); Space Shuttle Challenger disaster on 28 January 1986 (Winsor, 1988, 1989); the Chernobyl disaster on 26 April 1986 (Medvedev, 1991; Meshkati, 2007; Misumi et al., 1999; Stanton, 1996);
Corresponding author. Tel.: +34 947259084; fax: +34 947258910.
E-mail addresses: susanagh@ubu.es (S. Garca-Herrero), mariscal@ubu.es (M.A. Mariscal), manuel.gutierrez@unican.es (J.M. Gutirrez), antonio.toca@nuclenor.es (A. Toca-Otero). URLs: http://www.ubu.es/es (S. Garca-Herrero), http://www.ubu.es/es (M.A. Mariscal), http://www.ifca.unican.es (J.M. Gutirrez), http://nuclenor.org (A. TocaOtero). 1 Tel.: +34 947259084; fax: +34 947258910. 2 Tel.: +34 947349400. 0925-7535/$ - see front matter 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ssci.2012.09.004

and the explosion onboard the Piper Alpha oil platform in 1988 (Moore and Bea, 1993; Pate-Cornell, 1993). That is why the incorporation of organizational factors into risk management, measurement and control models took on such importance in the 1990s. For example, PSA, Probabilistic Safety Assessment, is a methodology for quantifying risk in industrial domains. Traditional PSA (Rasmussen, 1997, 1975) does not explicitly account for the inuence of organizational factors on accident risk. After conducting an analysis of the main accidents, however, the incorporation of organizational factors into PSA has been addressed by various researchers. Embrey (1992) developed a model based on Bayesian Network (BN) for the inclusion of organizational factors in PSA as applied to the rail transportation sector. (Davoudian et al., 1994a,b) developed an approach, applicable to Nuclear Power Plants (NPPs), called work processes analysis model (WPAM). The WPAM uses a set of twenty organizational factors developed for the Nuclear Regulatory Commission (Jacobs and Haber, 1994). The ASRM (Luxhoj, 2004) utilizes the Human Factors Analysis and Classication System (HFACS). The omega-factor approach is a method that models organizational failures and their inuence on NPP safety (Mosleh et al., 1997; Mosleh and Goldfeiz, 1999). The Socio-Technical Risk Analysis (SoTeRiA) extends the PRA framework to include the effects of organizational factors as the fundamental causes of accidents and incidents. This framework

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

83

integrates the technical system risk models with the social (safety culture and safety climate) and structural (safety practices) aspects of safety prediction models (Mohaghegh et al., 2009; Mohaghegh and Mosleh, 2009a,b). Finally, Lger and Weber (2009) developed a method for risk assessment considering three main aspects on the system resources: technical, human, and organizational; the integration is based on system knowledge structuring and its unied modeling by means of BN. The ndings of accident investigations and risk assessments evidence a growing recognition that the cultural context of work practices may inuence safety just as much as technology (Antonsen, 2009; Goh et al., 2010). The assumed link between culture and safety, epitomized through the concept of safety culture, has been the subject of intense research in recent years (Antonsen, 2009; Guldenmund, 2007; Kettunen et al., 2007; Mengolini and Debarberis, 2007). The term safety culture was introduced into the nuclear industry by the IAEAs International Nuclear Safety Advisory Group after the Chernobyl accident to denote the management and organizational factors that are important to safety (INSAG, 1986). But safety culture may not capture all of the management and organizational factors that are important to safe plant operation (Sorensen, 2002). The major problem with most existing safety culture models is that they are not integrated into general models of organization and of organizational culture (Grote and Kunzler, 2000). Organizational culture is a concept often used to describe shared corporate values that affect and inuence members attitudes and behaviors. In the literature, no attempt is made to link or integrate safety culture with organizational culture. The goal of this paper, then, is to establish a relationship between these two concepts of safety culture and organizational culture, in order to determine how to improve safety culture by altering organizational aspects. The methodology used to achieve this aim relies on Probabilistic (Bayesian) Networks (BNs). Currently, BNs are being applied in different research related to safety (Galan et al., 2007; Garca-Herrero et al., 2012; Li et al., 2012; Mariscal Saldaria et al., 2012; Zhao et al., 2012). For instance, Zhou et al. (2008) proposed a BN model to establish a probabilistic relational network among causal factors, including safety climate factors and personal experience factors, which exert inuences on human safety behavior. McCabe et al. (2008) demonstrated using BNs that the higher the work pressure, the higher the interpersonal conict. They also showed that lowquality leadership was most strongly associated with work-related health problems and accidents. Martn et al. (2009) used BNs to analyze the factors affecting the performance of tasks that involve a high risk of falls from ladders or from other auxiliary equipment. This enabled them to identify the circumstances that have the greatest bearing on workplace accidents during these activities, such as the adoption of incorrect work postures, the duration of tasks and a workers inadequate knowledge of safety regulations. Focusing on people and organizations, the paper by Ren et al. (2008) aims to contribute to offshore safety assessments by proposing a methodology to model causal relationships with a BN capable of providing graphical inter-relationships and of calculating numerical values for the likelihood of each failure event occurring. Bayesian inference mechanisms also make it possible to monitor how a safety situation changes when information ow travel forwards and backwards within the networks. In this paper we analyze the relationships between organizational and safety cultures in a nuclear power plant (Santa Mara de Garoa-Nuclenor S.A., Spain) using Bayesian networks. Section 2 denes the concept of organizational culture and describes the organizational culture questionnaire used in the study, that is, the Organizational Culture Inventory (OCI). Section 3 denes the concept of safety culture and explains the questionnaire used to

asses safety culture. The methodology is illustrated in section 4. In order to quantitatively establish the relationship of the model, a survey of every employee at the nuclear power plant was conducted in June of 2007; in section 4.1 data acquisition is explained. The survey consisted of two parts: the rst part included 120 questions related to organizational culture, taken from the Organizational Culture Inventory (OCI) developed by Human Synergistics International (Cooke and Lafferty, 1987); and the second part, on safety culture, included 35 questions (written by the authors) based on the ve components of safety culture dened by the International Atomic Energy Agency (IAEA). The study used probabilistic Bayesian network models to analyze the inuence of organizational cultures on safety culture. In section 4.2 Bayesian network models are briey described. Section 5 shows the results and establish the probabilistic relationship among organizational culture factors, including the 12 OCI scales that have an inuence on safety culture. Finally conclusions are developed in the last section; in summary, this study allows us to identify those steps to take so as to improve the safety culture at a nuclear power plant. 2. Organizational culture Before dening the concept of organizational culture, we must establish the concept of culture. Hofstede (1990) denes culture as a collective mental approach that distinguishes the members of one group or category from those of another. Culture is acquired, not inherited, and stems from the social environments effect on the individual, and not from his genes. Therefore, culture can be acted upon, evaluated and improved. Schein (1992) makes reference to the set of values, needs, expectations, beliefs and norms that are accepted and practiced by cultures, and distinguishes among several levels of culture: basic assumptions, values and ideologies, artifacts (slang, stories, rituals and decoration) and practices. Artifacts and practices express managerial values and ideologies. The concept of organizational culture and/or climate gained much attention in the 1970s and 1980s. In the 1970s, much research was undertaken under the title of organizational climate, with the term culture replacing climate in the 1980s. Uttal (1983) denes the concept of Organizational Culture as a system of shared values (what is important) and beliefs (how things work) that interact with a companys people, organizational structures, and control systems to produce behavioral norms (the way we do things around here). 2.1. Organizational culture questionnaires The four major self-reporting measures for organizational culture are compared (by correlation and factor analysis) by Xenikou and Furnham (1996). These four questionnaires are: the Organizational Culture Inventory (OCI), developed by Cooke and Lafferty (1987); the Organizational Beliefs Questionnaire (OBQ), developed by Sashkin (1983); the Corporate Culture Survey (CCS) by (Glaser, 1983); and the Culture Gap Survey (CGS) by Kilmann and Saxton (1983). The study set out to compare and contrast these different, but supposedly equivalent, questionnaire measures of culture by correlation and factor analysis. The correlation analysis showed the convergent validity of the questionnaires, and the factor analysis yielded six factors (accounting for 70,6% of the total variance) providing a framework of the organizational culture dimensions. In the model of Xenikou and Furnham the rst factor is labeled openness to change in a cooperative culture and contained four subscales of the OCI, two subscales from the CGS, and one subscale of the OBQ, and accounted for 33.3% of the variance; the four subscales of the OCI are humanistic orientation, afliation, achievement and self-actualization, these are the four constructive

84

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

cultural norms. The second factor accounted for 12.2% of the variance, is called task-oriented organizational growth and is identied principally by six subscales of the OBQ. The third factor, called the human factor in a bureaucratic culture, was bipolar and accounted for 7.9% of the total variance; it is composed by four subscales from OCI and the CGS subscale concerning personal freedom in the workplace; the four subscales of OCI are the four passive/ defensive cultural norms: approval, conventionality, dependence and avoidance. The four factor accounted for 5.8% of the variance and contained all the subscales of the CCS, which suggest that instead of measuring four different dimensions of culture, all the subscales are closely related; these factor was not included in the dimensional model of Xenikou and Furnham because is probable that the questionnaire measuring artifacts and not behavioral norms and values. The fth factor contained four subscales from OCI and accounts for 4.4% of total variance; these four subscales seemed to focus on power, competition, negativism toward the ideas of others (oppositional) and desire to avoid all mistakes (perfectionistic), these are the four aggressive/defensive cultural norms; it can be argued that the factor concerns mainly negativism and resistance to new ideas. The sixth factor account for only 3.9% of the common factor variance and contained a combination of subscales from the CGS and the OBQ; it was labeled as positive social relations in the workplace (Xenikou and Furnham, 1996). The study of Xenikou and Furnham could be used to recommend some measures over others. The OCI had the best internal reliability and its subscales loaded on four of the six dimensions while the CCS had the poorest alphas and all were highly intercorrelated loading on one factor. Consequently, according to Xenikou and Furnham (1996), of the four questionnaires mentioned, OCI is the most reliable measure of organizational culture. The organizational culture questionnaire used in our study is the Organizational Culture Inventory. It was developed by Human Synergistics International (Cooke and Lafferty, 1987), and is a validated survey used for organizational consulting and change purposes. Since its introduction in prototype form in 1983, the inventory has been used by thousands of organizations and been completed by over two million respondents worldwide (Balthazard et al., 2006). The OCI has been used in a wide array of organizations, including manufacturing and high-technology rms, research and development laboratories, schools and universities, governments, hospitals, etc. It has been administered for a variety of purposes, some of which are: as an outcome indicator (Balthazard and Cooke, 2004); to measure organizational culture in manufacturing rms and examine the relationship between the type of culture and the rms performance on quality measures (Rastrick and Corbett, 1998); to measure organizational culture in hospitals (Callen et al., 2009; Keenan et al., 1998; McDaniel and Stumpf, 1993; Seago, 1997, 2000); to predict the kind of leadership style that characterizes an organizations culture (Eppard, 2004); to investigate organizational culture in small, community-based residential services providers for people with intellectual disabilities (Gillett and Stenfert-Kroese, 2003). This wide range of applications has produced an extensive information base regarding the ways in which culture operates in different types of organizations (Balthazard et al., 2006). 2.2. Organizational culture inventory The OCI questionnaire focuses on the behavioral norms and expectations associated with the values shared by an organizations members. In doing so, it provides a way to address those cultural aspects that have a signicant impact on the activities of the employees and on the functioning of the organization. The OCI questionnaire provides an insight into what the staff believes is expected from it in relation to a series of aspects. In other words,

what behaviors, forms of thinking and of handling problems, ways of carrying out tasks, etc., are promoted and thus characterize the organization. To do this, the questionnaire asks the extent to which the organization expects or requires its members to act in a specic manner. Respondents indicate their degree of agreement with a series of statements using a 15 scale, (1) being not at all, (2) somewhat, (3) moderately, (4) mostly and (5) strongly. The OCI distinguishes among three general styles of organizational cultures: constructive, passive/defensive and aggressive/ defensive. The OCI further proposes 12 norms or types of organizational cultures (Fig. 1). The three styles and 12 norms that comprise the OCI are described below: The four constructive cultural norms are: A humanistic-encouraging culture (OCI1) characterizes organizations that are managed in a participative and person-centered way. Members are expected to be supportive, constructive, and open to inuence in their dealings with one another. (Helping others to grow and develop; taking time with people.) An afliative culture (OCI2) characterizes organizations that place a high priority on constructive interpersonal relationships. Members are expected to be friendly, open, and sensitive to the satisfaction of their work group. (Dealing with others in a friendly way; sharing feelings and thoughts.) An achievement culture (OCI11) characterizes organizations that do things well and value members who set and accomplish their own goals. Members of these organizations set challenging but realistic goals, establish plans to reach these goals, and pursue them with enthusiasm. (Pursuing a standard of excellence; openly showing enthusiasm.) A self -actualizing culture (OCI12) characterizes organizations that value creativity, quality over quantity, and both task accomplishment and individual growth. Members of these organizations are encouraged to gain enjoyment from their work, develop themselves, and take on new and interesting activities. (Thinking in unique and independent ways; doing even simple tasks well.) The four passive/defensive cultural norms are:

Fig. 1. Human Synergistics Circumplex. (Cooke and Lafferty, 1987).

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

85

An approval culture (OCI3) describes organizations in which conicts are avoided and interpersonal relationships are pleasant-at least supercially. Members feel that they should agree with, gain the approval of, and be liked by others. (Making sure people accept you; going along with others.) A conventional culture (OCI4) is descriptive of organizations that are conservative, traditional, and bureaucratically controlled. Members are expected to conform, follow the rules, and make a good impression. (Always following policies and practices; tting into the mold.) A dependent culture (OCI5) is descriptive of organizations that are hierarchically controlled and non-participative. Centralized decision-making in such organizations leads members to do only what they are told and to clear all decisions with superiors. (Pleasing those in positions of authority; doing what is expected.) An avoidance culture (OCI6) characterizes organizations that fail to reward success but nevertheless punish mistakes. This negative reward system leads members to shift responsibilities to others and avoid any possibility of being blamed for a mistake. (Waiting for others to act rst; taking few chances.) The four aggressive/defensive culture norms are: An oppositional culture (OCI7) describes organizations in which confrontation prevails and negativism is rewarded. Members gain status and inuence by being critical and thus are reinforced to oppose the ideas of others and to make safe (but ineffectual) decisions. (Pointing out aws; being hard to impress.) A power culture (OCI8) is descriptive of non-participative organizations structured on the basis of the authority inherent in members positions. Members believe they will be rewarded for taking charge, controlling subordinates and, at the same time, being responsive to the demands of superiors. (Building up ones power base; motivating others any way necessary.) A competitive culture (OCI9) is one in which winning is valued and members are rewarded for outperforming one another. People in such organizations operate in a win- lose framework and believe they must work against (rather than with) their peers to be noticed. (Turning the job into a contest; never appearing to lose.) A competence/perfectionistic culture (OCI10) characterizes organizations in which perfectionism, persistence, and hard work are valued. Members feel they must avoid all mistakes, keep track of everything, and work long hours to attain narrowly dened objectives. (Doing things perfectly; keeping on top of everything.) Specically, the Organizational Culture Inventory contains 120 items designed to produce the 12 OCI norms. So each norm is measured using 10 items. The Organizational Culture Inventory shows high levels of internal consistency. Empirical support for the 12 clusters, and therefore the construct validity of the inventory, is provided by the results of principal components analyses presented elsewhere, e.g. (Cooke and Rousseau, 1988), (Cooke and Szumal, 1993), (Xenikou and Furnham, 1996). The OCI is the questionnaire used in this study to measure organizational culture. In addition to the 120 items, the OCI questionnaire allows for an extra 40 questions that are prepared expressly for each organization. In our case, this space was used to present the 35 questions involving the ve safety culture dimensions. This feature was the basis for our study and allowed us to include both concepts of interest, the organizational culture and the safety culture, in the same questionnaire.

3. Safety culture The term of safety culture was rst used by IAEA (INSAG, 1991) for high risks industries. This term is not exclusive to the nuclear industry; the concept is very popular across a broad spectrum of industries with different types of accident hazards, work site activities, and risk scenarios (Baram and Schoebel, 2007): Air Trafc control (Ek et al., 2007; Gill and Shergill, 2004), aircraft maintenance (Atak and Kingma, 2011), offshore (Kalpee, 2008), chemical industry (Antonsen, 2009; Frank, 2007), construction (Camino Lpez et al., 2011; Ling et al., 2009; Lpez et al., 2008; Trner and Pousette, 2009), mining (Sanmiquel et al., 2010; Zhang et al., 2011) and shipping (Havold and Nesset, 2009).

3.1. Denition of safety culture It is acknowledged that there is no shared and validated approach on how to dene, assess and measure safety culture (Guldenmund, 2000; Hale, 2000; Sorensen, 2002). Indeed, due to its interdisciplinary nature, the concept tends to be of interest to different academic disciplines, from the social sciences (sociology, psychology or anthropology) to the more technical disciplines (maintenance, reliability engineering or systems safety). Consequently, the lack of a unanimous consensus on the concept is understandable (Mengolini and Debarberis, 2007). As a result, there are numerous denitions of safety culture in the academic safety literature (see Cooper (2000), Guldenmund (2000) and Sorensen (2002). Uttal (1983), for example, denes it as shared values and beliefs that interact with an organizations structures and control systems to produce behavioral norms. Turner et al. (1989) denes it as the set of beliefs, norms, attitudes, roles, and social and technical practices that are concerned with minimizing the exposure of employees, managers, customers and members of the public to conditions considered dangerous or injurious. According to Cooper (1998), the safety culture is the product of multiple goal-directed interactions between people (psychological), jobs (behavioral) and the organization (situational). The denition of safety culture proposed by ACSNI (The Advisory Committee on the Safety of Nuclear Installations) is the product of individual and group values, attitudes, perceptions, competencies and patterns of behavior that determine the commitment to and the style and prociency of an organizations health and safety management (Lee and Harrison, 2000). The culture of safety observed by Rochlin (1999) is a dynamic and emergent property not just at the individual, but also at the intergroup, level. How the operators operate safely is a property of the interactions, rituals, and myths of the social structure and beliefs of the entire organization, or at least of a large segment of it. There is also an on-going debate involving the terms safety climate (attitudes) and safety culture. There are authors who clearly differentiate between the two (Wiegmann et al., 2004), and others who consider them not as separate entities (Cox and Flin, 1998; Flin et al., 2000), but rather as different approaches towards the same goal of determining the importance of safety within an organization (Guldenmund, 2007; Hale, 2000). The denition of safety culture adopted here is the one proposed by the International Atomic Energy Agency (IAEA) in the INSAG-4 report (INSAG, 1991). Safety culture is dened as that assembly of characteristics and attitudes in organizations and individuals which establishes that, as an overriding priority, nuclear plant safety issues receive the attention warranted by their signicance. Organizations typically go through three phases in developing and strengthening safety culture: control of safety, guarantee of safety and total safety (Garca Herrero et al., 2002). These three

86

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

stages have also been identied by the IAEA in the nuclear industry (IAEA, 1998): Stage 1. Safety is compliance driven and is based mainly on rules and regulations. At this stage, safety is seen as a technical issue, whereby compliance with externally imposed rules and regulations is considered adequate for safety. In the rst stage, improvements are often gained primarily by improving the engineered safeguards of the plant in line with, for example, the principles contained in INSAG-12 (INSAG, 1999a) (the revised version of INSAG-3), and introducing basic systems and procedures to control hazards. Stage 2. Good safety performance becomes an organizational goal and is dealt with primarily in terms of safety targets or goals. This stage involves the use of a framework such as that presented in INSAG-13 (INSAG, 1999b). Stage 3. Safety is seen as a continuing process of improvement to which everyone can contribute. An example of the workers participation to improve safety culture, in an NPP, is presented in the publications (Mariscal-Saldaa et al., 2009; Mariscal et al., 2012). 3.2. Assess safety culture Nowadays, improving and maintaining the safety culture is a goal at every nuclear power plant. Attaining this improvement requires the establishing mechanisms to assess safety culture. These evaluation mechanisms have evolved in recent years. The INSAG-4 document (INSAG, 1991) includes identifying attributes and 143 INSAG questions to judge the effectiveness of safety culture. Following the publication of INSAG-4, the International Atomic Energy Agency (IAEA) published the document ASCOT Guidelines (Assessment of Safety Culture in Organizations Team Guidelines) (IAEA, 1996), which summarizes the concept of safety culture and describes a process for assessing safety culture. These guidelines restate and expand on the INSAG questions with 300 Guide Questions. The IAEA subsequently developed other documents that describe the processes for evaluating the safety culture, such as the IAEA-TEC-DOC-1321 Self Assessment of Safety Culture in nuclear installations: highlights and good practices(IAEA, 2002b) and the IAEA-TEC-DOC-1329 Safety Culture in nuclear installations: Guidance for use in the enhancement of safety culture (IAEA, 2002a). These guides compile various techniques for evaluating the safety culture, including the revision of documentation, interview techniques, observation techniques, and personnel questionnaires. Moreover, the provision of peer review and advisory services by international experts has proven to be an important mechanism for drawing on the nuclear safety experience which is available worldwide (Rosen, 1997). Also, a number of organizations, such as the World Association of Nuclear Operators (WANOs), have established expert review mechanisms that are now in use. The frequently used IAEA services are: the Operational Safety Review Team (OSART) missions for assessing plant management and operational safety performance; the Assessment of Safety Signicant Events Team (ASSET) service for examining operational incidents and long-term operating histories; the Engineering Safety Review Services (ESRSs) to review sitting and external hazards issues as well as re safety, computer-based safety systems and overall safety design features; and the SCART (Safety Culture Assessment Review Team) Mission. 3.3. The dimensions of safety culture Another important aspect that has also evolved over time is the identication of characteristics involving the safety culture con-

cept. For example, in 2002, the IAEA published the document INSAG-15 (INSAG, 2002) Key Practical Issues in Strengthening Safety Culture, which species and details seven key issues related to the safety culture: commitment, use of procedures, conservative decision making, a reporting culture, challenging unsafe acts and conditions, the learning organization and underpinning issues (communication, clear priorities, and organization). WANOs (World Association of Nuclear Operators) published the document GL-2006-02/January 2006, which includes the following characteristics: everyone is personally responsible for nuclear safety, leaders demonstrate commitment to safety, trust permeates the organization, decision-making reects safety rst, nuclear technology is recognized as special and unique, a questioning attitude is cultivated, organizational learning is embraced and nuclear safety undergoes constant examination. The characteristics adopted here, to describe a strong safety culture, are the established by the International Atomic Energy Agency (IAEA) (IAEA, 2006).These characteristics are: A. B. C. D. E. Safety is a clearly recognized value. Accountability for safety is clear. Safety is integrated into all the activities in the organization. Leadership for safety is clear. Safety is learning driven.

3.4. Safety culture questionnaire For this study, a safety culture questionnaire, based on those ve characteristics, was designed. The questionnaire includes 35 questions divided into ve groups: seven questions on each of the ve dimensions of safety culture. All of the questions are closed and each question has a ve-point Likert scale (15) expressed as strongly disagree, disagree, neither agree nor disagree, agree, and strongly agree. This safety culture questionnaire stemmed from a previous research project between Nuclenor and the University of Burgos (Mariscal Saldaa et al., 2007). The internal consistency of the safety culture questionnaire is measured by using Cronbachs a (Cronbach, 1951), in which scale reliability is dened as the extent to which a measure produces similar results over different data collection instances. With all 35 items, the resulting Cronbach alpha was 0.955, which was judged to be very good (the value generally accepted as the minimum desired value is 0.7). 4. Methodology 4.1. Data acquisition A survey was conducted to quantitatively establish the relationships between organizational culture and safety culture. The questionnaire was designed to ensure both anonymity and that all participants would understand the questions. As a result, 292 surveys were collected from a total of 323 workers, corresponding to 90.4% coverage (89.1% in the Sta Mara de Garoa nuclear power plant and 96.4% in the Santander administrative ofces). The survey consisted of 120 questions on organizational culture (referred to as OCI), 35 questions on safety culture (referred to as OSI), and ve additional demographic questions. Each of the 120 OCI questions was a discrete variable with ve possible responses (from 1 to 5, the value increasing according to the degree of agreement with the question); these variables are named V1V120. In addition, there are 12 discrete variables for safety culture, representing the twelve types of organizational culture: Humanistic, Afliative, Achievement, Self-actualization, Approval, Conventional, Dependent, Avoidance, Oppositional, Power, Competitive,

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

87

and Perfectionistic (referred to as OCI1, . . . , OCI12, respectively). Each of these variables is associated with ten out of the 120 questions and its value is dened as the mean of the corresponding ten questions; moreover, a discretization of these variables in ve states with equal frequency was considered in this work. The 35 safety culture questions are also discrete variables, with increasing agreement values ranging from 1 to 5, corresponding to ve different safety dimensions (Recognized Value, Accountability, Integrated, Leadership, Learning). The value for each of these dimensions was computed by aggregating the corresponding seven answers. Finally, in order to simplify the analysis, a single safety variable was considered as the mean of the ve previous dimensions and a discretization in ve equal-frequency classes was also considered as in the previous case. The resulting variable is referred to as SC (Safety Culture). Overall, in this paper we consider 120 discrete OCI variables, Vi, as well as 12 OCI aggregated variables, OCIi (Humanistic, Afliative, Achievement, Self-actualization, Approval, Conventional, Dependent, Avoidance, Oppositional, Power, Competitive, and Perfectionistic), and a single discrete variable on safety culture (SC). 4.2. Probabilistic (Bayesian) networks Probabilistic network models (Castillo et al., 1997) are popular multivariate modeling and analysis data mining techniques for extracting knowledge from databases containing information for a given problem (e.g., transactions of a company or surveys of a population). These techniques are based on sound statistical learning algorithms for automatically extracting knowledge (a joint probabilistic model in this case) from the raw data, thus allowing for probabilistic inference. The basic idea behind probabilistic networks is to encode dependencies among a set of stochastic variables using a graphical representation (a graph) which is easy to understand and interpret. There is a node in the graph for each variable, and edges connect attributes which are directly dependent on each other. The graph denes the decomposition of the high-dimensional Joint Probability Distribution (JPD), containing an intractable number of parameters, into a number of low-dimensional local distributions (marginal or conditional) by means of an appropriate factorization. The edges within the graph also indicate the paths along which to propagate evidence when inferences are to be drawn in the light of new ndings (for instance, when the value for some of the OCI variables is known).

A Bayesian Network (BN) is a special and powerful type of probabilistic network which is based on directed acyclic graphs, leading to simple factorizations of the JPD that are compatible with the dependency structure encoded by the graph. For instance, let us consider the subset of variables shown in the graph in Fig. 2, where the variables are represented pictorially by a set of nodes one node for each variable (in this example we consider SC and the 12 OCI aggregated variables), and the directed links representing relationships. If there is an arrow from node A to node B, we say that A is a parent of B or, equivalently, B is a child of A. For instance, in Fig. 2, the node OCI 1(Humanistic Style) is a parent of SC (Safety Culture), and OCI 1 is a child of OCI 2, OCI 11 and OCI 12. The set of parents of a node Vi is denoted by Pi. Directed graphs provide a simple denition of independence (d-separation) based on the existence, or not, of certain paths between the variables. The direction of the links provides an extra degree of freedom to provide expressiveness to these models and cannot be interpreted as a causal relationship (all the relationships are symmetrical in a probabilistic dependency model); for a detailed introduction to probabilistic network models see Castillo et al. (1997). In the case of Bayesian networks, the dependency/independency structure displayed by an acyclic directed graph for a set of variables X = {X1, . . . , Xn} can be translated to the Joint Probability Distribution (JPD) of the variables, p(x) = p(x1, . . . , xn), factorized as a product of several conditional/marginal distributions as follows:

px Pi pxi jpi

Depending on the discrete or continuous character of the variables, the conditional probabilities in Eq. (1) are specic parametric families. In this paper, we consider the variables to be discrete and, then, the conditional probability of each variable given its parents is multinomial (specied by the probability tables of the probabilities associated with the different combinations of values for the variables involved). Note that the number of parameters required to specify a full-dependence JPD would involve a huge number of parameters; in contrast, Bayesian network models dramatically reduce the number of parameters (by means of the existing independencies), which are given by a set of conditional probability tables, one for each variable conditioned to its parents set. Besides the modeling and diagnosis properties mentioned above, Bayesian networks provide a sound framework for probabilistic reasoning when new evidence, or information about the state of some variables, is known; for instance, in a particular application we may know that the humanistic style is high (e.g. OCI1 = 5

Fig. 2. Example of a directed acyclic graph (the graphical structure of a Bayesian network) for safety culture and twelve aggregated OCI variables.

88

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

in Fig. 2). Probabilistic reasoning is done by efciently updating the marginal probabilities p(Xi = xi) of the non-evidential nodes conditioned to the evidence e: p(xi|e), for all Xi R E. A comparison of the marginal and conditional probabilities shows the effect of evidence on the remaining model variables, allowing for the exploration and analysis of the survey. From a practical point of view, the application of BNs in realworld problems depends on the probability of inferring both the graphical structure and the parameters of the probability tables directly from data (automatic learning). Several methods have been introduced for learning the graphical structure (structure learning) and estimating probabilities (parametric learning) from data. Among the available structural learning algorithms, the methodology known as constraint satisfaction explores the possible dependencies and independencies among sets of variables by applying appropriate statistical tests (e.g. v2) to the available data; in this case, the graph is formed by the aggregation of the local dependencies (we refer the reader to Neapolitan 2003 for further details on these methods). One of these algorithms, a version of the original PC method named NPC, is included in a number of tools, such as (Hugin, 2010), used in this work to learn the Bayesian networks from the survey data.

5. Results As a rst example, Fig. 2 shows a Bayesian network tted to the survey data described in Section 4 and that considers the twelve aggregated OCI variables and the safety culture (SC). The Hugin software NPC learning algorithm was used in this case, considering a minimum condence threshold for the dependency relationships of 0.95, i.e. only the signicant dependencies are included in the graph. Both the graph and the corresponding probabilities in Eq. (1) are estimated from the set of data given by the surveys (a total of 292 cases). Thus, a cross-validation of the model is required in order to assess the robustness of the results, discarding overtting. As a measure of model performance, we considered the predictive skill of the model (the Bayesian network) by estimating how well the target variable SC is predicted, given the values of the rest of variables OCI1, . . . , OCI12. Note that the Bayesian network allows us to efciently compute an estimation of p(sc|OCI1 = oci1, . . . , OCI12 = oci12) for each of the cases and, therefore, a performance measure could be directly obtained by comparing this probabilistic prediction with the real SC value in the case considered. Thus, for the rst case (e.g. SC = 5, OCI1 = 4, . . . , OCI12 = 3) a high probability p(SC = 5|OCI1 = 4, . . . , OCI12 = 3) would indicate a good prediction of this particular case; however, assessing the performance of probabilistic forecasts is not straightforward. Relative Operating Characteristics (Rochlin) is a popular measure for the forecast skill of a probabilistic prediction, which compares the hits and failures of the deterministic forecasts given by a range of different probability thresholds (see (Mason, 1982)) for more details). ROC areas range between 0.0 and 1.0, with a score of 0.5 indicating zero forecast skill, and areas greater than 0.5 indicating useful forecasts. The ROC area provides an estimation of forecast skill for specic categories and so were calculated for the ve categories separately for the sample of 292 cases available in our study. In particular, for the two extreme cases SC = 1 and SC = 5, we obtained ROC areas of 0.797 and 0.772, respectively. Moreover, in order to validate these results and to assess the robustness to overtting, we performed a Leave One Out cross validation in which one out of the 292 cases was labeled as test and removed from the sample, training the Bayesian network (both the graph and the resulting probabilities) with the 291 remaining cases. Afterwards, the conditional probability of SC for the test case was computed and compared with the real value. This process was

repeated considering each of the cases once as a test sample, and the resulting probabilistic forecasts were joined and validated together, providing a robust assessment of the model performance. In this case, the ROC areas for the two extreme cases SC = 1 and SC = 5 were 0.727 and 0.716, respectively, which are comparable (differences lower than 10%) to the previous results, obtained for the whole sample. The same conclusions were obtained when performing the same crossvalidation experiment for the different models shown in this paper (not shown). Therefore, for simplicity we show the results of the different analysis considering the whole sample of data to train the Bayesian models. From a qualitative point of view, the graph in Fig. 2 shows that, in the NPP studied, two different OCI norms independently inuencing SC. On the one hand, the constructive group, formed by the Humanistic, Afliative, Achievement and Self-actualizing norms, with the Humanistic style playing a central role (among the remaining styles and also with SC). On the other hand, we nd the group of defensive styles formed by the Approval, Conventional, Dependent, Avoidance, Oppositional, Power, Competitive, and Perfectionistic norms. The defensive styles are further divided into two groups, those inversely related to safety culture (Avoidance and Competitive), and those presenting no clear relationship with safety culture (see Section 5.2 for more details). These ndings are in agreement with previous results. In general, the literature suggests that the constructive style is the preferred culture for good performance at both the individual and organizational levels (Balthazard et al., 2006; McDaniel and Stumpf, 1993; Rastrick and Corbett, 1998). According to Cooke and Rousseau (1988), norms for AggressiveDefensive and PassiveDefensive are loosely linked in certain settings. In the NPP studied, we also nd these relationships between the two styles (PassiveDefensive and AggressiveDefensive). See, for example, in the graph of Fig. 2, the direct connections between OCI3 (an AggressiveDefensive norm) and OCI7, OCI8 and OCI9 (Passive Defensive norms). The graph in Fig. 2 shows the global relationships among the different OCI norms and the safety culture. In this NPP, only two norms are directly connected to SC, while the other norms are indirectly (conditionally) connected to this variable. Although this is the best way to jointly model all these variables, it is also interesting to analyze the relationships between the different norms and SC one by one. To this end we use the original 120 OCI variables and analyze the inuence of each of these defensive and constructive OCI norms on safety culture separately. 5.1. Relationship between safety culture and OCI constructive style As an illustrative example of constructive styles, in this section we consider the humanistic culture and present the corresponding modeling and sensitivity results. Fig. 3 shows the Bayesian network formed by SC and the ten OCI variables related to the humanistic culture: V2 worrying about the needs of others, V3 involving others in those decisions that affect them, V4 resolving conicts constructively, V5 being supportive of others, V18 helping others excel and develop, V19 knowing how to listen, V20 rewarding others, V34 devoting time to people, V35 encouraging others, V50 helping others to think of themselves. This graph shows that all these variables form a joint model (none of the variables is disconnected from the others) and that the direct connection between SC and OCI is given through V4, through which the remaining OCI variables are indirectly related (and through V3 and V5, etc.). In order to quantify these relationships, Table 1 shows the sensitivity of SC to changes in the different OCI variables Vi. To this end, an evidence Vi = 5 (maximum value for the corresponding variable) is considered, and the relative probability increment is com-

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

89

Fig. 3. Relationship between the humanistic culture and safety culture.

puted as DP(SC = j) = (P(SC = j|Vi = 5) P(SC = j))/P(SC = j). Note that the marginal distribution is uniform, P(SC = j) = 0.2 for all j, by construction. For instance, the relative probability increment produced in the maximum state of SC by V4 is higher than 100%, increasing from a marginal value of 0.2 to a conditional value of 0.42. Moreover, the minimum state of SC decreases from 0.20 to 0.02 in this case (with a relative 88.48% reduction). Adjusting the humanistic culture has a signicant effect on safety culture. For example, note the large inuence of variables V2, V3, V4, V5 and V20 on safety culture: worrying about the needs of others, involving others in those decisions that affect them, resolving conicts constructively, being supportive of others, rewarding others. Furthermore, it has also been shown that the remaining constructive styles, afliative, achievement and selfactualizing, have a direct bearing on safety culture. Following a similar procedure with the remaining OCIs, those variables that most inuence the afliative culture (those resulting in SC variations in excess of 30%) are: V61 (collaborating with others), V62 (treating others in a friendly and pleasant manner), V63 (thinking in terms of satisfying the group), V64 (showing concern for people), and V106 (behaving tactfully). The organization must therefore promote relationships and collaboration among employees so as to enhance the safety culture.

Within the achievement culture, variables V42 (adhering to standards of excellence) and V59 (knowing the companys activity) stand out. In the case of the self-actualizing culture, the most inuential variables are: V90 (resisting conformism), V105 (communicating ideas), V117 (being open with oneself) and V118 (enjoying work).

5.2. Relationship between safety culture and OCI defensive styles (passivedefensive and aggressivedefensive) When doing the same analysis for OCI defensive styles, we found a weak relationship with SC. Moreover, for some cultures, all the OCI variables are disconnected from the SC: the approval culture, the conventional culture, the power culture, the competitive culture. In the dependent culture, only one variable exhibited an inuence in favor of safety culture, namely V45 (reviewing decisions with superiors). This action contributes to a signicant improvement in safety culture. As an illustrative example, in this case, we consider the relationships between the avoidance culture and SC. The BN that analyze these relation is show in Fig. A1 in Appendix A. Those actions that hamper the safety culture are: V72 (favoring popular over necessary decisions), V74 (delegating responsibilities to others), V87

Table 1 Relative probability increment of SC given the evidence of different OCI humanistic variables. SC variation for evidence Vi = 5 Vi V2 V3 V4 V5 V18 V19 V20 V34 V35 V50

Table 2 Variables with the greatest inuence on SC arranged by OCI. Cultural norms Variables 2, 3, 4, 5, 20 61, 62, 63, 64, 106 45 87, 88 1, 16, 32, 47 114 42, 59 90, 105, 117, 118

DP (SC = 1) (%)
39.33 70.3 88.48 47.97 16.54 21.12 39.39 12.77 15.69 12.63

DP (SC = 2) (%)
5.07 11.03 10.7 4.39 2.54 3.36 6.48 1.87 2.29 1.88

DP (SC = 3) (%)
0.44 7.73 18.23 0.8 0.75 0.66 0.43 0.77 0.89 0.77

DP (SC = 4) (%)
10.16 13.9 11.16 12.11 5.05 6.31 11.27 4.01 4.89 3.98

DP (SC = 5) (%)
34.05 74.05 105.44 40.48 12.93 17.08 33.37 9.6 11.91 9.51

1. Humanistic 2. Afliative 3. Approval 4. Conventional 5. Dependent 6. Avoidance 7. Oppositional 8. Power 9. Competitive 10. Perfectionistic 11. Achievement 12. Selfactualizing

90

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

Fig. 4. Global model. Bayesian network relating the most inuential SC variables.

(keeping out of sight when difcult situations arise), and V88 (never be the one blamed for mistakes). For this case, the variations in SC were calculated considering evidence Vi = 1 and not Vi = 5, as in the other cases, since these variables are inversely related to safety culture. Thus, when a value of 100% is assigned to variable Vi in answer 1, the variations in SC shows good results (Table A1 in Appendix A shows these results). For instance, when the value of V87 (keeping out of sight when difTable 3 Relative variation in variables for maximizing SC. Vi variation for evidence SC = 5 Vi V1 V2 V3 V4 V5 V16 V20 V32 V47 V59 V61 V62 V63 V64 V87 V88 V105

DP (Vi = 1)
47.81 30.71 42.81 14.41 20.29 0.61 4.37 21.16 38.21 7.07 0 18.6 4.22 5.37 62.16 37.73 2.11

DP (Vi = 2)
37.79 25.03 25.78 15.31 16.53 2.21 4.05 44.38 12.09 18.24 23.78 13.53 8.01 5.54 78.85 2.36 2.52

DP (Vi = 3)
32.05 7.01 9.27 61.16 3.36 2.69 1.43 40.96 35.57 15.2 30.32 9.36 4.26 2.29 66.06 24.9 0.62

DP (Vi = 4)
12.2 4.66 7.99 9.78 2.87 0.49 5.96 40.38 0.66 4.6 7.07 2.66 0.42 1.71 55.78 2.36 0.13

DP (Vi = 5)
41.94 28.97 23.23 59.85 15.42 3.6 4.17 136.5 13.13 9.8 20.74 15.11 9.38 5.77 100 37.62 3.06

cult situations arise) is the smallest one (V87 = 1, i.e. NOT keeping out of sight when difcult situations arise) the probability of SC = 5 increases to 0.31 (a relative increase of 50.29%). For this culture the most inuential variables are V87 and V88. Lastly, within the defensive styles, two, oppositional and perfectionistic, require a detailed analysis. Some of the variables improve SC and others go against it. Let us consider, for example, the Bayesian network for the oppositional culture (Fig. B1 in Appendix B). This network has two groups of variables: those favoring the SC and those that hamper it. The variables that have a direct inuence on safety culture are: V1 (pointing out aws), V16 (adopting an impartial and completely objective attitude) and V32 (looking for errors). Those that have an adverse inuence are: V47 (remaining on the sidelines) and, to a lesser extent, V48 (refusing criticism). The sensitivity analysis for those variables that contribute positively to improving the SC is shown in Table B1 in Appendix B. For example, if answer 5 for variable V32 (looking for errors) were 100%, the SC values would be 2.61 for state 1 (that is, an 85.28% decrease) and 46.6 for state 5 (a 134.71% increase) (see Fig. B2 in Appendix B). Let us now consider the results of the sensitivity analysis for those variables with a negative inuence on the SC. In these cases we force answer 1 to be 100% for each of the variables so as to determine the improvement in SC. For example, as shown in Table B2 in Appendix B, if answer 1 were 100% for variable V47
Table A1 Relative variations in SC. Variables for Avoidance culture. SC variation for evidence Vi = 1 Vi V72 V74 V87 V88

An analysis of the above table shows that the variables with the greatest inuence on improving SC are as follows: 1 Humanistic; V2 Worrying about the needs of others; V3 Involving others in those decisions that affect them; V4 Resolving conicts constructively; V5 Being supportive of others; 2 Afliative; V61 Collaborating with others; V62 Treating others in a friendly and pleasant manner; V63 Thinking in terms of satisfying the group; 6 Avoidance; V87 Not keeping out of sight when difcult situations arise; V88 Not avoiding being blamed for mistakes; 7 Oppositional; V1 Pointing out aws; V32 Looking for errors; V47 Not remaining on the sidelines; 11 Achievement; V59 Knowing the companys activity; 12 Self-actualizing; V105 Communicating ideas.

DP (SC = 1) (%)
9.43 13.7 23.57 16.99

DP (SC = 2) (%)
12.87 9.47 21.39 12.93

DP (SC = 3) (%)
0.14 3 10.12 5.92

DP (SC = 4) (%)
1.24 0.74 2.29 0.93

DP (SC = 5) (%)
19.65 24.15 50.29 33.27

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

91

Fig. A1. Bayesian network to analyze the relationships between avoidance culture and safety culture.

Fig. B1. Relationship between the Oppositional culture and safety culture.

(remaining on the sidelines), the SC values would be as follows: 10.06 in state 1 (which implies a 40.25% drop) and 30.57 in state 5 (which represents an increase of 53.99%). As a result of the above simulation, we can see that the oppositional culture does not lend itself to an aggregate analysis since some of the questions have a positive and others a negative inuence. Improving the safety culture requires handling each variable appropriately. The same value in the oppositional culture does not imply a similar effect on safety culture; rather, it depends on how said value was obtained (individual values for the ten questions). The perfectionistic culture exhibits similar behavior to the oppositional culture (see Fig. C1 in Appendix C). In that the variables that affect safety culture positively are V114 (being diligent and persistent), V113 (giving the impression of being competent and independent) and V82 (never being wrong). Variable V96 (setting overly demanding and unrealistic goals), on the other hand, has a negative effect on SC. The remaining variables have no inuence on SC.

The sensitivity analysis of the variables that contribute positively to improving SC are shown in Table C1 in Appendix C. Also, in Appendix C, Table C2 shows how the variable V96 (setting overly demanding and unrealistic goals) has a negative inuence on SC. The sensitivity analysis on variable V96 has been made forcing the answer 1100%, because V96 has a negative inuence on SC. The sensitivity analysis of the perfectionistic culture shows that diligence and persistence (V114) is the type of behavior that most favors a safety culture.

5.3. Analysis of the OCI variables with the greatest inuence on safety culture A detailed study of the 120 OCI variables revealed that those with the greatest inuence on safety culture, that is, those causing variations in the SC results in excess of 30%, are as shown in Table 2:

92

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

Fig. B2. Conditional probabilities of SC for V32 = 5.

Table B1 Relative variations in SC. Oppositional culture variables with a positive inuence. SC variation for evidence Vi = 5 Vi V32 V1 V16

DP (SC = 1) (%)
85.28 39.26 40.04

DP (SC = 2) (%)
36.78 17.08 50.79

DP (SC = 3) (%)
51.82 7.33 9.67

DP (SC = 4) (%)
32.44 10.12 18.17

DP (SC = 5) (%)
134.71 48.68 114.24

Table B2 Relative variations in SC. Oppositional culture variables with a negative inuence. SC variation for evidence Vi = 1 Vi V47 V48

DP (SC = 1) (%)
40.25 26.75

DP (SC = 2) (%)
11.38 8.4

DP (SC = 3) (%)
48.28 17.83

DP (SC = 4) (%)
23.89 13.5

DP (SC = 5) (%)
53.99 21.57

Since the ultimate goal is to improve the safety culture, a global model was constructed, with every style, that included the safety culture variable and the most relevant OCI variables (see Fig. 4). Conducting a sensitivity analysis of this network, and assigning safety culture the highest value possible (100% in answer 5), reveals the variations in the variables related with SC. Table 3 shows the relative variation as a percent of the variables with respect to the initial situation (also see Fig. C1b and c). For example, the distribution for V1, which is (state 1 = 1.38, state 2 = 3.77, state 3 = 20.24, state 4 = 41.42, state 5 = 32.85), changes to (state 1 = 0,72, state 2 = 2,35, state 3 = 13,75, state 4 = 36.37, state 5 = 46.63) given an evidence of SC = 5, which then corresponds to a relative change of (47.81, 37.79, 32.05, 12.2, 41,94) respectively.

6. Conclusions The study presented used probabilistic Bayesian network models to analyze the inuence exerted by each of the organizational cultures on safety culture in a NPP. First, it seems that the constructive styles are those that have the greatest inuence on the safety culture results. In contrast, the defensive styles do not, in general, exhibit a clear relationship with safety culture. This does not mean that we should forget the defensive style to improve safety culture; the detailed analysis, made with each norm, and above all the global model carried out in this study shows how defensive styles also contribute to improving safety culture. The detailed studies of the four constructive cultural norms (humanistic, afliative, achievement and self-actualizing) reect

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295

93

Fig. C1. Relationship between the Perfectionistic culture and safety culture.

Table C1 Relative variations in SC. Perfectionistic culture variables with a positive inuence. Answer 5 for variable V114 was forced to be 100%. SC variation for evidence Vi = 5 Vi V114 V113 V82

DP (SC = 1) (%)
50.88 3.45 8.9

DP (SC = 2) (%)
60.43 15.95 9.75

DP (SC = 3) (%)
8.1 2.5 5

DP (SC = 4) (%)
26.32 2.95 2.3

DP (SC = 5) (%)
137.4 12.95 11.35

Table C2 Relative variations in SC. Perfectionistic culture variables with a negative inuence. SC variation for evidence Vi = 1 Vi V96

DP (SC = 1) (%)
16.58

DP (SC = 2) (%)
3.66

DP (SC = 3) (%)
16.25

DP (SC = 4) (%)
1.9

DP (SC = 5) (%)
4.37

the specic actions that serve to enhance the safety culture in the NPP. Thus, in the humanistic culture, we note the following actions: worrying about the needs of others, involving others in the decisions that affect them, resolving conicts constructively and being supportive of others and rewarding others. The most inuential behaviors in the afliative culture are: collaborating with others, treating others in a friendly and pleasant manner, thinking in terms of satisfying the group, showing concern for people and behaving tactfully. In the achievement culture, adhering to standards of excellence and knowing the companys activity are vital to improving the SC. Lastly, in the self-actualizing culture, the behaviors that reinforce the SC are resisting conformism, communicating ideas, being open with oneself and enjoying work. Analyzing the defensive styles even further, we note the existence of four different behaviors. First, the cultural norms that do not correlate: the approval culture, the conventional culture, the power culture and the competitive culture. Second, those that show how specic actions enhance the safety culture, such as the dependent culture. Third include those that are inversely related to safety culture, such as the avoidance culture. And lastly involves those featuring variables that favor and oppose the safety culture, such as the oppositional and perfectionistic culture. Below are the specied defensive actions that enhance or harm the safety culture in the NPP. In the dependent culture, reviewing decisions with superiors enhances the safety culture. The avoid-

ance culture exhibits an inverse relationship with safety culture, which are realized in actions such as: keeping out of sight when difcult situations arise and never being the one blamed for mistakes. Within the oppositional culture, looking for errors, adopting an impartial and completely objective attitude, and pointing out aws, help to improve the safety culture; on the other hand, remaining on the sidelines is a type of behavior that does not favor the safety culture. In the perfectionistic culture a behavior that is directly related to the safety cultures is diligence and persistence. Finally, a global model has been constructed in order to identify the behaviors that play the greatest role in improving SC. This model involves all of the organizational cultural norms and includes the most relevant variables having thus been identied in the detailed studies. These most inuential behaviors on SC, classied by cultures, are specied below. Within the humanistic culture are: resolving conicts constructively, involving others in those decisions that affect them, worrying about the needs of others and being supportive of them. In the afliative culture the following stand out: collaborating with others, treating others in a friendly and pleasant manner and thinking in terms of satisfying the group. Negative inuences in the avoidance culture include: keeping out of sight when difcult situations arise and avoiding being blamed for mistakes. In the oppositional culture, looking for errors, pointing out aws and not remaining on the sidelines have an exceedingly high inuence on SC. Within the achievement

94

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295 Cooper, M.D., 2000. Towards a model of safety culture. Safety Science 36, 111136. s stone or man of straw? Work and Cox, S., Flin, R., 1998. Safety culture: philosopher Stress 12, 189201. Cronbach, L.J., 1951. Coefcient alpha and the internal structure of tests. Psychometrika 16, 297334. Davoudian, K., Wu, J.-S., Apostolakis, G., 1994a. Incorporating organizational factors into risk assessment through the analysis of work processes. Reliability Engineering & System Safety 45, 85105. Davoudian, K., Wu, J.-S., Apostolakis, G., 1994b. The work process analysis model (WPAM). Reliability Engineering & System Safety 45, 107125. Ek, A., Akselsson, R., Arvidsson, M., Johansson, C.R., 2007. Safety culture in Swedish air trafc control. Safety Science 45, 791811. Embrey, D.E., 1992. Incorporating management and organizational-factors into probabilistic safety assessment. Reliability Engineering & System Safety 38, 199208. Eppard, R.G., 2004. Transformational and transactional leadership styles as they predict constructive culture and defensive culture. Virginia Polytechnic Institute and State University, United States Virginia, p. 129. Flin, R., Mearns, K., OConnor, P., Bryden, R., 2000. Measuring safety climate: identifying the common features. Safety Science 34, 177192. Frank, W.L., 2007. Process safety culture in the CCPS risk based process safety model. Process Safety Progress 26, 203208. Galan, S.F., Mosleh, A., Izquierdo, J.M., 2007. Incorporating organizational factors into probabilistic safety assessment of nuclear power plants through canonical probabilistic models. Reliability Engineering & System Safety 92, 11311138. Garca-Herrero, S., Mariscal, M.A., Garca-Rodrguez, J., Ritzel, D.O., 2012. Working conditions, psychological/physical symptoms and occupational accidents. Bayesian network models. Safety Science 50, 17601774. Garca Herrero, S., Mariscal Saldaa, M.A., Manzanedo del Campo, M.A., Ritzel, D.O., 2002. From the traditional concept of safety management to safety integrated with quality. Journal of Safety Research 33, 120. Gill, G.K., Shergill, G.S., 2004. Perceptions of safety management and safety culture in the aviation industry in New Zealand. Journal of Air Transport Management 10, 231237. Gillett, E., Stenfert-Kroese, B., 2003. Investigating organizational culture: a comparison of a High- and a Low-performing residential unit for people with intellectual disabilities. Journal of Applied Research in Intellectual Disabilities 16, 279284. Glaser, R., 1983. The Corporate Culture Survey. Organizational Design and Development, Bryn Mawr, PA. Goh, Y.M., Brown, H., Spickett, J., 2010. Applying systems thinking concepts in the analysis of major incidents and safety culture. Safety Science 48, 302 309. Grote, G., Kunzler, C., 2000. Diagnosis of safety culture in safety management audits. Safety Science 34, 131150. Guldenmund, F.W., 2000. The nature of safety culture: a review of theory and research. Safety Science 34, 215257. Guldenmund, F.W., 2007. The use of questionnaires in safety culture research - an evaluation. Safety Science 45, 723743. Hale, A.R., 2000. Cultures confusions. Safety Science 34, 114. Havold, J.I., Nesset, E., 2009. From safety culture to safety orientation: validation and simplication of a safety orientation scale using a sample of seafarers working for Norwegian ship owners. Safety Science 47, 305326. Hofstede, G., 1990. Cultures and Organization: Software of the Mind. McGraw-Hill, London. Hugin, 2010. <http://www.hugin.com>. IAEA, 1996. ASCOT Guidelines: Guidelines for Organizational Self-Assessment of Safety Culture for Reviews by the Assessment of Safety Culture in Organizations Team, IAEA-TECDOC-860. International Atomic Energy Agency, Vienna. IAEA, 1998. Developing Safety Culture in Nuclear Activities: Practical Suggestions to Assist Progress, Safety Reports Series. No. 11, IAEA, International Atomic Energy Agency, Vienna. IAEA, 2002a. Safety culture in nuclear installations. Guiadance for Use in the Enhancement of Safety Culture. IAEA-TECDOC-1329. International Atomic Energy Agency, Vienna. IAEA, 2002b. Self-assesment of safety culture in nuclear installations: highlights ad good practices. IAEA-TECDOC-1321. International Atomic Energy Agency, Vienna. IAEA, 2006. Application of the Management System for Facilities and Activities Safety Guide Safety Standards Series No. GS-G-3.1. International Atomic Energy Agency, Vienna. INSAG, 1986. Sumary Report on the post-Accident Review Meeting on the Chernobyl Accident. (Safety Series No. 75-INSAG-1). International Nuclear Safety Advisory Group. International Atomic Energy Agency, Vienna. INSAG, 1991. Safety Culture (Safety Series No 75-INSAG-4). International Nuclear Safety Advisory Group, International Atomic Energy Agency, Vienna. INSAG, 1999a. Basic Safety Principles for Nuclear Power Plants (75-INSAG-3 Rev. 1), INSAG Series No. 12. International Atomic Energy Agency, Vienna. INSAG, 1999b. Management of Operational Safety in Nuclear Power Plants, INSAG Series No. 13. International Atomic Energy Agency, Vienna. INSAG, 2002. Key Practical Issues in Strengthening Safety Culture (Safety Series No 75 -INSAG-15). International Nuclear Safety Advisory Group, International Atomic Energy Agency. Jacobs, R., Haber, S., 1994. Organizational Processes and Nuclear-Power-Plant Safety. Reliability Engineering & System Safety 45, 7583.

culture includes knowing the companys activity. And lastly, in the self-actualizing culture, communicating ideas. The principal advances of knowledge made with the approach, developed in the present research, are the identication of the three principal norms and behaviors necessaries to achieve maximum results in safety culture. These behaviors, classied in order of importance, are: looking for errors, not keeping out of sight when difcult situations arise, and resolving conicts constructively. In addition the model shows others behaviors that also contribute to improve safety culture. Taking steps to improve these behaviors has been identied as paving the way to enhancing safety culture at a NPP. For future research it would be advisable to compare safety culture and organizational culture in others companies. It would be interesting to know if the organizational cultural norms have the same inuence in safety culture in chemical, manufacturing, mining and others industries. Acknowledgments The authors would like to express their thanks to all the personnel of the NPP for their participation, motivation and enthusiasm. Thanks also to the anonymous reviewers for their valuable and helpful suggestions for enriching the quality of the paper. Appendix A. Avoidance culture See Table A1 and Fig. A1. Appendix B. Oppositional culture See Figs. B1 and B2, Tables B1 and B2. Appendix C. Perfectionistic culture See Fig. C1 and Tables C1 and C2. References
Antonsen, S., 2009. The relationship between culture and safety on offshore supply vessels. Safety Science 47, 11181128. Atak, A., Kingma, S., 2011. Safety culture in an aircraft maintenance organisation: a view from the inside. Safety Science 49, 268278. Balthazard, P.A., Cooke, R.A., 2004. Organizational culture and knowledge management success: assessing the behavior-performance continuum. In: Sprague, R.H., Jr (Ed.), Proceedings of the Hawaii International Conference on System Sciences. Big Island, HI, pp. 37413750. Balthazard, P.A., Cooke, R.A., Potter, R.E., 2006. Dysfunctional culture, dysfunctional organization: capturing the behavioral norms that form organizational culture and drive performance. Journal of Managerial Psychology 21, 709732. Baram, M., Schoebel, M., 2007. Editorialsafety culture and behavioral change at the workplace. Safety Science 45, 631636. Booth, R.T., Lee, T.R., 1995. The role of human-factors and safety culture in safety management. Proceedings of the Institution of Mechanical Engineers Part B Journal of Engineering Manufacture 209, 393400. Callen, J., Braithwaite, J., Westbrook, J.I., 2009. The importance of medical and nursing sub-cultures in the implementation of clinical information systems. Methods of information in medicine 48, 196202. Camino Lpez, M.A., Fontaneda, I., Gonzlez Alcntara, O.J., Ritzel, D.O., 2011. The special severity of occupational accidents in the afternoon: The lunch effect. Accident Analysis and Prevention 43, 11041116. Castillo, E., Gutirrez, J.M., Hadi, A.S., 1997. Sensitivity analysis in discrete Bayesian networks. IEEE Transactions on Systems, Man, and Cybernetics Part A: Systems and Humans. 27, 412423. Cooke, R.A., Lafferty, J.C., 1987. Organizational Culture Inventory. Human Synergistics, Plymouth, MI. Cooke, R.A., Rousseau, D.M., 1988. Behavioral norms and expectations: a quantitative approach to the assessment of organizational culture. Group and Organization Studies 13, 245273. Cooke, R.A., Szumal, J.L., 1993. Measuring normative beliefs and shared behavioral expectations in organizations: the reliability and validity of the organizational culture inventory. Psychological Reports 72, 12991330. Cooper, D., 1998. Improving Safety Culture: A Practical Guide. Wiley, Chichester.

S. Garca-Herrero et al. / Safety Science 53 (2013) 8295 Kalpee, T., 2008. The journey towards world class safety performance bpTT. In: Society of Petroleum Engineers 9th International Conference on Health, Safety and Environment in Oil and Gas Exploration and Production 2008 In Search of Sustainable Excellence, Nice, pp. 22962300. Keenan, G.M., Cooke, R., Hillis, S.L., 1998. Norms and nurse management of conicts: keys to understanding nurse-physician collaboration. Research in Nursing and Health 21, 5972. Kettunen, J., Reiman, T., Wahlstrom, B., 2007. Safety management challenges and tensions in the European nuclear power industry. Scandinavian Journal of Management 23, 424444. Kilmann, R.H., Saxton, M.J., 1983. The Kilman-Saxton Culture-Gap Survey. Organizational Designs Consultans, Pittsburgh, PA. Kondo, J., 1996. The spirit of safety: oriental safety culture. Nuclear Engineering and Design 165, 281287. Le Bot, P., 2004. Human reliability data, human error and accident models illustration through the Three Mile Island accident analysis. Reliability Engineering and System Safety 83, 153167. Lee, T., Harrison, K., 2000. Assessing safety culture in nuclear power stations. Safety Science 34, 6197. Leger, A., Weber, P., Levrat, E., Duval, C., Farret, R., Iung, B., 2009. Methodological developments for probabilistic risk analyses of socio-technical systems. Proceeding of the Institution of Mechanical Engineers Part 0, Journal of Risk and Reliability 223 (O4), 313332. Li, P.-c., Chen, G.-h., Dai, L.-c., Zhang, L., 2012. A fuzzy Bayesian network approach to improve the quantication of organizational inuences in HRA frameworks. Safety Science 50, 15691583. Ling, F.Y.Y., Liu, M., Woo, Y.C., 2009. Construction fatalities in Singapore. International Journal of Project Management 27, 717726. Lpez, M.A.C., Ritzel, D.O., Fontaneda, I., Alcantara, O.J.G., 2008. Construction industry accidents in Spain. Journal of Safety Research 39, 497507. Luxhoj, J., 2004. Building a Safety Risk Management System: A Proof-of-Concept Prototype. FAA/NASA Risk Analysis Workshop. Maddox, M.E., Muto, W.H., 1999. Three Mile Island: the human side. Ergonomics in Design 7, 612. Mariscal-Saldaa, M.A., Garca-Herrero, S., Toca-Otero, A., 2009. Assessing safety culture using RADAR matrix. Dyna 84, 327336. Mariscal, M.A., Garca Herrero, S., Toca Otero, A., 2012. Assessing safety culture in the Spanish nuclear industry through the use of working groups. Safety Science 50, 12371246. Mariscal Saldaa, M.A., Garca Herrero, S., Dez Simn, E., 2007. Preparacin, aplicacin y valoracin de cuestionarios sobre cultura de seguridad nuclear. Ref. 2007/00053/001. NUCLENOR, S.A., Spain. Mariscal Saldaria, M.A., Garcia-Herrero, S., Garcia-Rodriguez, J., Ritzel, D., 2012. The impact of occupational hazard information on employee health and safety: an analysis by professional sectors in Spain. The International Electronic Journal of Health Education 15, 8398. Martn, J.E., Rivas, T., Matas, J.M., Taboada, J., Argelles, A., 2009. A Bayesian network analysis of workplace accidents caused by falls from a height. Safety Science 47, 206214. Mason, I., 1982. A model for assessment of weather forecasts. Australian Meteorological Magazine 30, 291303. McCabe, B., Loughlin, C., Munteanu, R., Tucker, S., Lam, A., 2008. Individual safety and health outcomes in the construction industry. Canadian Journal of Civil Engineering 35, 14551467. McDaniel, C., Stumpf, L., 1993. The organizational culture: Implications for nursing service. Journal of Nursing Administration 23, 5460. Medvedev, G., 1991. The Truth About Chernobyl. Basics Books, New York. Mengolini, A., Debarberis, L., 2007. Safety culture enhancement through the implementation of IAEA guidelines. Reliability Engineering & System Safety 92, 520529. Meshkati, N., 2007. Lessons of the Chernobyl nuclear accident for sustainable energy generation: creation of the safety culture in nuclear power plants around the world. Energy Sources Part A Recovery Utilization and Environmental Effects 29, 807815. Misumi, J., Wilpert, B., Miller, R., 1999. Nuclear Safety: A Human Factors Perspective. Taylor & Francis, London. Mohaghegh, Z., Kazemi, R., Mosleh, A., 2009. Incorporating organizational factors into Probabilistic Risk Assessment (PRA) of complex socio-technical systems: a hybrid technique formalization. Reliability Engineering and System Safety 94, 10001018. Mohaghegh, Z., Mosleh, A., 2009a. Incorporating organizational factors into probabilistic risk assessment of complex socio-technical systems: principles and theoretical foundations. Safety Science 47, 11391158. Mohaghegh, Z., Mosleh, A., 2009b. Measurement techniques for organizational safety causal models: characterization and suggestions for enhancements. Safety Science 47, 13981409.

95

Moore, W.H., Bea, R.G., 1993. Human and organizational error in operations of marine systems: occidental piper alpha. In: Proceedings of the International Conference on Offshore Mechanics and Arctic Engineering OMAE. Publ by ASME, Glasgow, Scotland, Engl, pp. 2129. Mosleh, A., Goldfeiz, E., Shen, S., 1997. The w-factor approach for modeling the inuence of organizational factors in probabilistic safety assessment. In: IEEE Sixth Annual Human Factors Meeting. Orlando, FL, pp. 9/18-19/23. Mosleh, A., Goldfeiz, E.B., 1999. An approach for assessing the impact of organizational factors on risk, Technical Research Report CTRS-B3-Center for Technology Risk Studies, University of Maryland. Oien, K., 2001. A framework for the establishment of organizational risk indicators. Reliability Engineering and System Safety 74, 147167. Pate-Cornell, M.E., 1993. Learning from the Piper alpha accident: a postmortem analysis of technical and organizational factors. Risk Analysis 13, 215232. Rasmussen, J., 1997. Risk management in a dynamic society: a modelling problem. Safety Science 27, 183213. Rasmussen, N.C., 1975. Reactor Safety Study, In: Commision, N.R. (Ed.), WASH-1400. Washington, DC, US. Rastrick, K., Corbett, L.M., 1998. Organizational culture and quality performance: a New Zealand study. In: Anon (Ed.), Proceedings Annual Meeting of the Decision Sciences Institute. Decis Sci Inst, San Diego, CA, USA, p. 6. Ren, J., Jenkinson, I., Wang, J., Xu, D.L., Yang, J.B., 2008. A methodology to model causal relationships on offshore safety assessment focusing on human and organizational factors. Journal of Safety Research 39, 87100. Rochlin, G.I., 1999. Safe operation as a social construct. Ergonomics 42, 1549 1560. Rosen, M., 1997. Towards a global nuclear safety culture. Nuclear Energy-Journal of the British Nuclear Energy Society 36, 287289. Sanmiquel, L., Freijo, M., Edo, J., Rossell, J.M., 2010. Analysis of work related accidents in the Spanish mining sector from 19822006. Journal of Safety Research 41, 17. Sashkin, M., 1983. Pillars of Excellence-the Organizational Beliefs Questionnaire. Organizational Desing and Development ed., Organizational Desing and Development, King of Prussia, PA. Schein, E.H., 1992. Organizational Culture and Leadership: A Dynamic View. JosseyBass, San Francisco. Seago, J.A., 1997. Organizational Culture in Hospitals: Issues in Measurement. Journal of Nursing Measurement 5, 165178. Seago, J.A., 2000. Registered nurses, unlicensed assistive personnel, and organizational culture in hospitals. Journal of Nursing Administration 30, 278286. Shrivastava, P., 1994. Technological and organizational roots of industrial crises: lessons from Exxon Valdez and Bhopal. Technological Forecasting and Social Change 45, 237253. Sorensen, J.N., 2002. Safety culture: a survey of the state-of-the-art. Reliability Engineering & System Safety 76, 189204. Stanton, N.A., 1996. Human Factors in Nuclear Safety. CRC. Trner, M., Pousette, A., 2009. Safety in construction a comprehensive description of the characteristics of high safety standards in construction work, from the combined perspective of supervisors and experienced workers. Journal of Safety Research 40, 399409. Turner, B.A., Pidgeon, N., Blockley, D., Toft, B., 1989. Safety Culture: Its Importance in Future Risk Management. Second World Bank Workshop on Safety Control and Risk Management, Karstad, Sweden. Uttal, B., 1983. The Corporate Culture Vultures. Fortune 108, 6672. Wiegmann, D.A., Zhang, H., Von Thaden, T.L., Sharma, G., Gibbons, A.M., 2004. Safety culture: an integrative review. International Journal of Aviation Psychology 14, 117134. Winsor, D.A., 1988. Communication failures contributing to the Challenger accident: an example for technical communicators. IEEE Transactions on Professional Communications 31, 101107. Winsor, D.A., 1989. Challenger: a case of failure to communicate. CHEMTECH 19, 528535. Xenikou, A., Furnham, A., 1996. A correlational and factor analytic study of four questionnaire measures of organizational culture. Human Relations 49, 349 371. Zhang, X.M., Liu, Y.J., Yao, Q.G., 2011. AHP-fuzzy-based coal mine enterprise safety culture evaluation model analysis. In: International Conference on Engineering and Business Management. Sci Res Publ, Inc-Srp, Irvin, pp. 163166. Zhao, L., Wang, X., Qian, Y., 2012. Analysis of factors that inuence hazardous material transportation accidents based on Bayesian networks: a case study in China. Safety Science 50, 10491055. Zhou, Q., Fang, D., Wang, X., 2008. A method to identify strategies for the improvement of human safety behavior by considering safety climate and personal experience. Safety Science 46, 14061419.