The use of questionnaires in safety culture resear

research –an evaluation

Frank W.Guldenmund *

Safety Science Group,Delft University of Technology,Jaffalaan 5,2628BX Delft,The Netherlands

Abstract

Questionnaires have not been particularly successful in exposing the core of an organisational safety culture.This is clear both from the factors found and the relations between these and safety indicators.The factors primarily seem to denote an overall evaluation of management,which does not say much about cultural basic assumptions.In addition,methodology requires that levels of the-ory and measurement are properly recognised and distinguished.That is,measurements made at one level cannot be employed at other levels just like that unless certain conditions are met.

Safety management has been described through nine separate processes that together encompass the safety management system (SMS)of an organisation.Policies developed at the organisational level shape the organisational context and working conditions of the group and individual levels and therefore also attitudes within the organisation.The questionnaires seem to expose only those attitudes that are shared throughout the whole of the organisation.The workforce could very well recognise the safety policies of higher management as concern for their well-being and the overall value attached to safety.Pictured this way,safety climate (attitudes)and safety culture are not sep-arate entities but rather different approaches towards the same goal of determining the importance of safety within an organisation.

Ó2007Elsevier Ltd.All rights reserved.

Keywords:Safety climate;Safety culture;Questionnaire research;Safety management

0925-7535/$-see front matter Ó2007Elsevier Ltd.All rights reserved.doi:10.1016/j.ssci.2007.04.006

*Tel.:+310152788053.

E-mail address:

f.w.guldenmund@tudelft.nl

Safety Science 45(2007)

723–743

1.Introduction

In the past years considerable effort has been put into the construction of a valid and reliable safety climate questionnaire.In safety culture research a(safety climate)question-naire has been the predominant measurement instrument(Collins and Gadd,2002;Gul-denmund,2000).This popularity is not surprising,it being a quick but also‘dirty’instrument.Because of this‘dirtiness’a safety climate survey only gives an inkling of what a particular safety culture might be about.The challenge is,of course,to develop a ques-tionnaire that yields just enough relevant and valid information–the trusted‘wetfinger’to find out from which way the wind blows–to decide whether and possibly where any cor-rective measures or actions are opportune.The key words here are relevant and valid.

When developing a safety climate questionnaire two avenues of possibilities can be taken.First,a descriptive model of safety climate can be used as a starting point–a nor-mative or theoretical approach.Second,results of previous research can be combined to construct a new questionnaire–a more pragmatic approach.In an earlier paper(Gulden-mund,2000),an overview is given of both the models devised for safety climate and the scales resulting from safety climate questionnaire research.Basically,the models provide a global taxonomy for the(safety)attitude objects in question,whereas the scales give more precise descriptions of these attitudes.

The purpose of this paper is to look at these taxonomies and attitude objects anew and to propose a common basis that might explain the patterns of shared attitudes found in safety climate research.Furthermore,a list of safety management processes will be put for-ward that is considered to be fundamental in creating,continuing,changing and hence evaluating safety climate.With this proposal the paper changes focus from what has been measured to what should be measured.

2.Current state of affairs

Safety climate should be distinguished from safety culture research,where the former is a manifestation or‘snapshot’(Flin et al.,2000)of the latter.Climate is reflected in the ‘workforce’s perceptions of the organisational atmosphere’(ibid.,p.178),i.e.climate is more superficial and transient than culture.Following Schein(1992)the essence of culture is reserved for the core of an organisation’s culture,which has to be deciphered from many sources,(organisational)climate amongst these.Whether this distinction is purely aca-demic or also has some conceptual texture to it,I will return to later on.

Safety climate is generally explored through a questionnaire survey within the target organisation or parts of it.The questionnaires are often composed of series of thematic questions that tap people’s evaluations of various aspects considered to be relevant for safety(climate);several authors have enumerated many such researches(Cooper,2000; Cox and Flin,1998;Flin et al.,2000;Guldenmund,2000;Mearns et al.,2003;Williamson et al.,1997).After the survey the researchers process the data,which could involve the construction of two or more scales through principal components analysis(PCA)orfind-ing the scores of the organisation on such scales,which have already been established in previous research.A third variant of a climate survey could be confirmatory in that the researchers try to confirm the scales found in earlier research.All three types of research approaches are to be found in the literature,resulting in many,many scales(Flin et al., 2000;Guldenmund,2000).Afinal endeavour of the research team might be tofind theF.W.Guldenmund/Safety Science45(2007)723–743725

correlation between one or more scales and some criterion variable,such as accidents or safety-related behaviour,possibly with reference to different subgroups.

My2000paper(Guldenmund,2000)reviewed research into safety culture and safety climate up to1997and proposed a guiding framework for future studies.The paper by Flin et al.(2000)complemented this review with an overview of generalised scales distilled from the plethora of scales found in safety climate research.Additionally,Cooper(2000) put forward a model to‘measure and analyse’safety culture,which was linked interest-ingly to the framework presented in Guldenmund(2000).Also,Neal et al.(2000)and Thompson et al.(1998)proposed synthesising models bringing together several aspects of the researchfield.

Papers appearing after this wave of frameworks,syntheses and generalisations referred to these publications in their introductory paragraphs but then started offon their own course(e.g.Glendon and Litherland,2001;Reiman and Oedewald,2004).In2002Collins and Gadd again reviewed the safety culture researchfield,but did not report any signifi-cant advances since the previous surge of papers(which actually are all from or around 1998,which is when the thematic sessions during the I.C.O.H.conference in Amsterdam and the A.P.A.symposium in San Francisco were held).In effect the followingfindings currently still hold:

1.There is a large variety in factors(dimensions,scales,facets)that make up the safety

climate concept(Collins and Gadd,2002;Flin et al.,2000;Guldenmund,2000).How-ever,these can be brought down to a limited number of‘themes’(Collins and Gadd, 2002;Flin et al.,2000).

2.A factor pertaining to‘management’pops up in the analyses about75%of the time and

a factor‘safety system’in about two-thirds of the studies(Collins and Gadd,2002;Flin

et al.,2000).

3.Most studies have not been able to replicate a factor solution from a previous study,not

even within the same type of company(Brown and Holmes,1986;Cox and Flin,1998;

Coyle et al.,1995;DeDobbeleer and Be´land,1991;Flin et al.,2000),but see(Glendon and Litherland,2001)for more confirmatoryfindings.

4.Safety climate and‘safety performance’are weakly(.20)related at best(Clarke,2006).

To summarise,although more research effort has been put into the safety climate concept since the special issues of Work and Stress(1998)and Safety Science(2000)thefield is as fragmented and misunderstood as it was left by the reviewers and framework builders more than six years ago.

3.A closer look at questionnaires

In the introduction to this paper questionnaires have been called both quick and dirty. Self-administered questionnaires can be distributed among large groups of people in a rel-atively short period of time fairly easily,hence the term quick.However,the possibilities to control unwanted influences affecting the responses are limited and therefore these include a lot of random‘noise’,hence the term dirty.Thefirst characteristic actually cancels the effect of the latter since the‘dirtiness’in the responses is averaged out over the large num-ber of responses,provided,of course,that the unwanted influences are unsystematic and normally distributed.726 F.W.Guldenmund/Safety Science45(2007)723–743

These are not the only methodological reasons for doing a large survey.Questionnaires often also provide instant quantified results that enable the researcher to produce medians or means,compare subgroups and benchmark these.Obviously,a self-administered ques-tionnaire is a valuable tool in(social scientific)research.

In organisational culture research,however,certain conditions apply which might make the self-administered questionnaire less useful.For instance,culture is,by definition, something that is shared between people and the variance created by the dirtiness of ques-tionnaires may obscure the assumptions they share.The degree of variance per se defines the extent to which assumptions are commonly held amongst the members of a group. Moreover,within organisations the groups we can assume to have a common culture are often not large enough to average out the random influences.Furthermore,the scales that are used to record the responses(Likert scales,preference scales,indices of impor-tance or significance)are assumed to be at the(quasi-)interval measurement level,but this is at least doubtful.1This principally means that calculating means,variances,correlations and other linear transformations is not allowed.Again,with large populations this would not be such a problem,but within most studies of organisations,so far published in the safety climate literature,it probably is.Basically,in survey research one is caught between the theoretical demands of statistics(heterogeneous normally distributed variables around a single mean obtained from large populations)and the theoretical requirements of culture ([strong]convictions shared by groups or categories of people,which are small enough to interact and create a culture about safety or any other related topic).

Yet another source of confusion in safety climate research is the distinction that is made between perceptions and attitudes.Perceptions seem to be regarded as‘descriptive’and referring to‘external objects’whereas attitudes are considered personal‘evaluations’of the same objects;e.g.see Williamson et al.(1997),Glendon and Litherland(2001).It could,however,very well be argued that these perceptions are infused with the attitudes that underlie them,in that perceptions are not mere descriptions but,rather,evaluations of what people see around them.Consequently,perceptions reflect attitudes(Eagly and Chaiken,1993).My point here is that safety climate research is basically attitude research.2 Importantly,in linking safety climate research to attitude research the theoretical develop-ment in the latterfield could readily be applied also in thefield of safety climate;see Fiske and Taylor(1991)or Eagly and Chaiken(1993)for impressive overviews on theoretical and practical aspects of attitudes.

To make matters even more complex,with regard to culture the organisation cannot be considered a closed system.That is,not only local conditions within the organisation determine the culture of its members.Actually,when a company has not experienced any serious problems during its existence there probably will not be a typical culture (Schein,1992);its culture will be determined largely by external(national,regional)con-ditions and the(educational,social-economic,religious)background of its workforce (Guldenmund et al.,2006).

1Interval level measurement implies that the psychometric distances between categories of the scale are all the same,i.e.the distance between strongly agree and agree is the same as disagree and neither agree,nor disagree.It is indeed doubtful that this is actually the case.

2Attitudes are defined as a‘psychological tendency that is expressed by evaluating a particular entity with some degree of favor or disfavor’(Eagly and Chaiken,1993).F.W.Guldenmund/Safety Science45(2007)723–743727

So,what kind of information do we collect with questionnaires?Although we intend to uncover an underlying trait called culture,the questionnaires invite respondents to espouse rationalisations,aspirations,cognitions or attitudes at best,that is,the very thing called espoused values by Schein(1992).Obviously,one could still argue that behind all these espoused values the‘true’shared values,if any,hide,but it takes a lot of deciphering and a creative analyst to uncover these.Hence,we are stuck with a set of factors and scores on them but we do not know what they really mean or imply.We maybe have an answer to the what?-question but we certainly do not know why.Basically,we are back where we started from with trying tofigure out why this company shows these artefacts and expresses these espoused values.Or,put in another way,survey research does not yield processed climate or culture results but rather provides another source of raw data to extract an organisational culture from.

4.Safety climate structure

4.1.Introduction

In my2000paper(Guldenmund,2000)I proposed four principal‘attitude objects’with regard to safety climate–hardware/physical environment,software,people and risk–these partly being taken from Cox and Cox(1991).However,this classification is too coarse and unspecific to be of any use.

Flin et al.(2000)identify in their review paper six‘themes’common to18papers they have scrutinised:i.e.management(72%),safety system(67%),risk(67%),work pressure (33%),competence(33%)and procedures/rules.Although a little more extended,this tax-onomy still does not put a sufficient handle on the concept of safety climate.

Instead,I would like to classify the scales found according to the organisational level at which the processes they refer to,take place.This classification provides some insight into the information that surveys might actually have collected.Firstly,however,the organisa-tional levels will be outlined.

4.2.Organisational levels

Generally,in organisations several structural levels can be distinguished,based on the different types of processes taking place at these levels.For instance,in a report for the Dutch government de Boer and van Drunen(2003)list four levels at which behaviour within organisations can be observed–macro-organisational,micro-organisational,indi-vidual-rational and individual-perceptual.Also Hofmann et al.(1995)distinguish macro, micro-and individual behavioural levels with regard to safety(performance).I would like to adopt these levels for the present taxonomy,re-labelling them for the current purpose organisational,group and individual.

With the organisational level I refer to(behavioural)processes taking place at higher organisational levels–i.e.plant management,the management team or senior manage-ment.There is an abundance of literature on the importance of such processes at this level for safety performance,see for instance Collins and Gadd(2002)for a review.

The group level refers to(behavioural)processes within groups or teams the respondent works in and belongs to,including the team leader or supervisor(the importance of the728 F.W.Guldenmund/Safety Science45(2007)723–743

role of supervisors is stressed in several publications,e.g.Simard and Marchand(1996)or Flin et al.(2000)).

Finally,following de Boer and van Drunen(2003),wefind both rational and perceptual processes influencing behaviour at the individual level,i.e.processes particular to the respondent.Combining this with what has been said about attitudes above part of the pro-cesses at this level could be considered attitudinal processes.

Now,we will look more closely at the processes taking place at each level(see Table1). The nine dimensions used as structure here are adapted from those developed in Delft in research on auditing of management(see,e.g.Guldenmund et al.,2006;Hale et al.,1999). At the highest,organisational,level the processes carried out here can be equated with safety management efforts at,what has been called,the‘policy’or strategic level.What is decided at this level of the organisation in terms of policies,ambitions,strategies,goals, targets,means and so on,is passed on to the next,or group,level,where such decisions are usually divided among groups or individuals and worked out in detail,often by technical or staffservices.The important point here to observe is that how these processes are detailed is both dependent on what is outlined at the level above,as well as on the circum-stances the particular groups operate in.This means that the very generic outcomes of the processes defined above,become concrete functions and activities(Table1third column) of groups and teams.

Clearly,what can be defined in most global terms at the highest level of the organisation becomes much more detailed and nuanced at the level below.Moreover,this is also the actual environment of the worker and this is what(s)he has primary knowledge of,expe-riences with and feelings and,therefore,attitudes about.

Finally,getting at the lowest level the information that has been processed(i.e.specified or detailed)at the group level becomes reality for the individual worker during a particular task or job(Table1fourth column).

The inventory of all these processes has been so detailed to make one point very clear: the(very)general objects that are the subject of policies at the organisational level(com-petence,procedures,hardware)become very specific matters(knowledge about a particu-lar piece of hardware,specific procedures to do a certain job,etc.)at the individual level. So what is quite coherent at the organisational level becomes quite fragmented at the level of individuals.

5.Organisational levels and attitude objects

5.1.Organisational level

The nine dimensions defined above and processed at each of the three organisational levels could also be envisioned as things people have actually done or perceived being done,or have knowledge or particular feelings about.To put it another way,these pro-cesses could also function as attitude inputs or objects and,hence,produce evaluative responses.

Referring again to Table1,this also provides a rough overview of how policies that are promulgated at the highest level of the organisation,trickle down through the other levels to arrive at the primary process as specific conditions under which individual tasks are per-formed.What is important to appreciate is that policies propagated at the organisational level furnish what could be considered the Gestalt of safety within the organisation–i.e.

Table1

Management systems and their manifestation at different organisational levels

Process Organisational level Group level Individual level

Risks Policies with regard to

which,when and how risks

are evaluated(i.e.what the

present dangers and risks are

and how they should be

perceived and controlled)Risks and dangers present in

the primary tasks performed

by a group

Risks and dangers associated

with the particular task(and

location)at hand

Hardware

design and

layout Policies with regard to how

identified risks are avoided,

reduced or controlled by

design or layout(i.e.policies

determining the choice of

barriers that control present

dangers and risks)

Barriers encountered/used

by the work group to control

the risks they face during the

execution of their primary

tasks

Particular barriers present,

used and/or operative during

the execution of a particular

task

Maintenance Policies with regard to

maintenance and inspection

(e.g.when[preventive vs

corrective],how often,etc.)Maintenance and/or

inspection tasks carried out

by the group(if relevant).

Maintenance and inspection

plans

Particular inspection/

maintenance tasks carried

out

Procedures Policies with regard to

formal procedures and

instructions(balanced by

what is trained and how

workers are recruited and

selected).Basically,policies

determining the structure of

work Detailed rules,procedures

and safety instructions

related to work group

functions and tasks

Presence of rules,procedures

and work instructions for a

particular task and their

level of detail

Manpower

planning Policies with regard to

(minimum)manning levels

during(a)normal

operations,(b)process

disturbances and(c)

emergency situations(i.e.

basically policies

determining work pressure)

Work schedules and

manning levels:people of

work group assigned to

tasks,shifts,rosters.

Planning of personnel for

major maintenance stops

Time and people available to

carry out a particular job

Competence Policies determining what

competencies(skills,

knowledge,physique)are

required to perform any task

(balanced by what is written

down in formal procedures)Function and task

descriptions and

requirements.The qualities

(skills,knowledge,physique)

present in the work group

Particular task requirements

in terms of knowledge

(plant,process,instruments,

tools)and physical abilities

Commitment Policies with regard to

supervision and motivation

of the workforce and the

adherence to rules(e.g.what

is acceptable behaviour and

how deviations should be

corrected)Motivation or incentive

programs and ways of

supervision.Adequate and

dedicated behaviour shown

in relation to the primary

tasks of the work group

(either execution or

supervision)

Adequate and dedicated

behaviour shown in relation

to this task.Supervision

and/or active behavioural

modification programs

functioning during the

execution of a particular task

(continued on next page)

F.W.Guldenmund/Safety Science45(2007)723–743729

what is considered safe,acceptable and controllable.Put yet in another way –this is the stuffthat beliefs and convictions are made of.

It has already been pointed out in several reviews (e.g.Cox and Flin,1998;Flin et al.,2000;Guldenmund,2000),that both the questionnaires as well as the target populations in safety climate research differ significantly.Although this might explain some of the varia-tion in the factor solutions found,it is nevertheless very striking that many studies report finding similar or comparable ‘management’and ‘safety system’factors (Flin et al.,2000).Obviously,these two factors would compare well with the nine processes defined above.In Table 2the scales that have been found in previous safety climate studies are labelled somewhat tentatively according to the level and the attitude object (i.e.one or more of the nine processes)the scale seems to denote.In this table only studies are used that mention either the eigenvalues 3or percentage of explained variance (or both)resulting from the analysis (which is usually PCA)as well as the questions that make up the scale.Also included are the alpha-coefficients,which are a measure of the internal consistency of a particular scale.

What can be clearly seen from Table 2is that the most important factors –i.e.,those having the highest eigenvalues and,hence,percentage of explained variance –have acquired the label ‘organisation’.4As it is,these factors account for the bulk of the vari-ance within a data set and their constituting variables (the questionnaire items)therefore have the highest correlations amongst themselves.

The issue of aggregation of questionnaire data have been raised more than once,see for instance Guldenmund (2000)or Mearns et al.(2003).To produce their factor solutions all

Table 1(continued )Process Organisational level Group level

Individual level

Communication

Policies with regard to effective collaboration and interaction of (groups)of people (i.e.policies

determining who should talk with whom about what)

Communication (between members of the work group)related to the primary tasks:meetings,hand-over procedures,permits-to-work (PTWs),field vs.panel operator routines

Types of communication involved with executing a particular task (e.g.hand-over,PTW,

communication with control room or colleagues)

Monitoring and change

Policies with regard to performance indicators and adjustments to the technology or safety management system

Registration of

performance indicators by/of work group members:audits,inspections,checklists,to-do lists,accident/incident reporting.Change

management procedures

Possibility and means to give feedback on

deviations that occurred during the execution of a particular task and

proposals to change work methods or layout

3

Eigenvalues result from the decomposition of a correlation or co-variance matrix and which is part of the principal component analysis (PCA)Basically they express the amount of variance a particular dimension accounts for.Eigenvalues are always presented in descending order.4

Some original labels are rather misleading in that they employ words like ‘personal’or ‘motivation’.Inspection of the constituting questions revealed that these factors actually refer to organisation level attitude objects.

730

F.W.Guldenmund /Safety Science 45(2007)723–743

Research teams,factors and associated statistics and organisational level

Research team Year Safety climate scale ev%Variance a Level

Zohar1980Importance of safety training

programs

9.8440.9Organisation(competence)

Management attitudes towards

safety

4.6319.3Organisation(risk)

Effects of safe conduct on

promotion

2.5310.6Organisation(commitment)

Level of risk at work place 2.349.7Group(risk,hardware)

Effects of required work pace on

safety

1.66 6.9Organisation/group(availability)

Status of safety officer 1.17 4.8Organisation(commitment)

Effects of safe conduct on social

status

1.07 4.4Group(commitment,competence)

Status of safety committee0.84 3.4Organisation(commitment)

Cox and Cox1991Personal scepticism 3.3419.00Organisation/group(commitment)

Individual responsibility 1.8410.00Individual(commitment,competence)

Safeness of work environment 1.518.00Group(risk,hardware)

Effectiveness of arrangements

for safety

1.257.00Organisation/group(hardware)

Personal immunity 1.06 6.00Individual(commitment,competence) Cabrera and Isla1994(?)Company policies towards

safety

–38.9Organisation(risk)

Emphasis on productivity versus

safety

– 6.4Organisation(commitment,availability)

Group attitudes towards safety– 5.00Group(commitment)

Specific strategies of prevention– 3.9Organisation(risk,hardware)

Safety level perceived in the

airport

– 3.7Group/individual(risk,hardware)

Safety level perceived on the job– 3.6Group/individual(risk,hardware)

(continued on next page)F.W. Guldenmund / Safety Science 45 (2007) 723–743 731Table2(continued)

Research team Year Safety climate scale ev%variance a Level

Coyle and Sleeman1995Maintenance and management

issues(organisation1)–49.12Organisation/group(a.o.maintenance,

commitment,procedures)

Company policy (organisation1)–11.27Organisation(a.o.commitment,

competence,monitoring and change)

Accountability(organisation1)–9.51Organisation(commitment,procedures,

monitoring and change)

Training and management issues

(organisation1)

–8.56Organisation(competence,procedures)

Work environment

(organisation1)

–7.87Group(hardware,commitment)

Policy/procedures

(organisation1)

–7.39Organisation(hardware,procedures)

Personal authority

(organisation1)

– 6.27Individual(competence,commitment)

Work environment (organisation2)–71.19Organisation/group(a.o.maintenance,

commitment,procedures)

Personal authority (organisation2)–17.62Organisation(commitment,monitoring and

change)

Training and enforcement of policy(organisation2)–11.19Organisation/Individual(procedures,

hardware,competence,commitment)

Berends1996Confidence in the arrangements

for safety

11.1220.20.83Organisation(hardware)

Compliance with safe working

practices

3.35 6.10.82Group(commitment)

Perceived priority given to safety 2.4 4.40.8Organisation(commitment)

Own active effort put in safety

matters

2.08

3.80.8Group/individual(commitment)

Communication about safety 1.96 3.60.82Group(communication)

Williamson,Feyer, Cairns and Biancotti 1996Personal motivation for safety–18.000.86Organisation/group(commitment) Positive safety practice–11.90.84Organisation/group(commitment)

Risk justification–7.20.79Individual(competence,hardware)

732

F.W.

Guldenmund

/

Safety

Science

45

(2007)

723–743Fatalism– 5.90.65Group/individual(commitment,monitoring

and change)

Optimism– 4.20.39Group(commitment,hardware)

Glendon and Litherland2001Communication and support 5.18.30.93Organisation/group(a.o.communication,

commitment,availability)

Adequacy of procedures 4.413.70.92Organisation/group(procedures)

Work pressure 4.1713.000.Organisation(availability)

Personal protective equipment (PPE)3.2310.10.86Group/individual(hardware,monitoring

and change)

Relationships 2.37.20.82Organisation/group(commitment)

Safety rules 2.18 6.80.72Group(procedures)

O’Toole2002Management’s commitment to

safety

–11.94Organisation(commitment)

Safety supervisory process–9.72Group(commitment)

Emergency response–7.59Organisation(risk,hardware,procedures)

Education and knowledge– 5.62Organisation(competence)

Employee involvement and

commitment

– 5.48Organisation/group(commitment)

Off-the-job safety– 3.Not covered

Drugs and alcohol– 3.04Organisation(commitment,procedures) Itoh,Andersen and Seki2003Motivation–25.80.3Organisation/group(commitment,

competence,procedures)

Satisfaction with own

competence

–8.50.901Organisation/Individual(competence)

Safety awareness of operation– 6.40.323Group(risk,competence)

Morale– 5.30.312Organisation(commitment)

Satisfaction with manual and checklists – 5.10.737Organisation(procedures,monitoring and

change)

Satisfaction with management

system

– 4.50.553Organisation(procedures,commitment)

Trust in management–40.573Organisation(commitment,monitoring and

change)F.W. Guldenmund / Safety Science 45 (2007) 723–743 733734 F.W.Guldenmund/Safety Science45(2007)723–743

studies aggregate their data to the level of the organisation.While this might be meaning-ful for some issues,other issues are best explored at the group or even individual level. However,because of the small amount of data a PCA might not be feasible then.In addi-tion,as long as correlations between variables function as input for the analysis methods, another issue can arise.For instance,when there are two tight groups with very opposed points of view–this is not an uncommon situation in some organisations–the overall cor-relation(i.e.aggregated correlation)between the variables expressing these views will be low,while the within group correlations will be high.These variables will also show a bimodal distribution.Again,the analysis methods will not work well for these variables and they will not contribute much,if at all,to the scales identified.Yet they could indicate important cultural differences between groups within one organisation.

What this tells us is that substantial correlations will only be found among variables that are unimodally distributed and also(are perceived to)have meaning at that level of measurement and aggregation.Only then might such high correlations reveal shared notions in the(aggregated)population.As the analysis of the research in Table2shows, these are often shared notions about the products of processes operating at the organisa-tional level(e.g.organisational context,working conditions,knowledge,capacities,etc.). Most likely these factors therefore provide an evaluation of higher management per se and,in the absence of actual knowledge or experience,are primarily based on an affective stance towards this management.5Similar observations about the nature of these manage-ment factors have been offered by Cabrera et al.(1997),Mearns et al.(2003)and Cabrera and Isla(1998).

Another problem exposed here that mars all attitude-related research is that people will almost always express an attitude when asked about it(Fiske and Taylor,1991,p.520). Schuman and Presser(1981)report an interestingfinding in this respect(p.148ff.).In a nationwide survey US respondents were asked to agree or disagree with a particular act about to pass congress.While this particular law proposal was hardly even known, answers to this question were nevertheless collected.It appeared,however,that these answers correlated significantly with a question about the economic performance of the current administration.This relationship is explained as expressing a more general confi-dence in the current government.Such a general trust would provide a positive attitude towards any bill they propose.

Related causal reasoning seems to underlie both structural models of Thompson et al. (1998)and Neal et al.(2000).These papers propose levels that can be identified as the organisation,group and individual levels,where the factors pertaining to the organisation level reflect general ideas about management and the group and individual levels refer to concrete knowledge,conditions and practices.Interestingly,at the latter two levels the word‘safety’also becomes more prominent.

Summarising,the dominant factors found in most safety climate research can be ascribed to mostly affective evaluations of the workforce about its management at an organisational level of aggregation.It has been further argued that nine generic safety management processes at the organisational level shape these notions.

5Behavioural and cognitive as well as affective processes can function as inputs to attitude formation.In the absence of both behavioural experience and actual knowledge of what goes on at the organisational level,it is therefore argued that affective processes play a prominent role in attitude formation among those at lower organisational levels.

F.W.Guldenmund/Safety Science45(2007)723–743735

5.2.Group level

Although processes at the organisational level deliver the context and conditions for all activities at lower levels,the results of these processes are still rather unspecific.Additional inputs for attitude formation are provided at the group level where the organisational level outputs are worked out in detail and divided among,and processed by groups or individ-uals.In specifying and executing these processes the shopfloor starts to get conscious of the policies and provisions of its management.The role of the supervisor in this process has been stressed in several publications(Collins and Gadd,2002;Hofmann and Stetzer, 1998;Sorensen,2002;Zohar,2002)and(s)he is added emphatically to this level.The supervisor often explains,justifies and applies(or downplays and undermines)higher man-agement’s policies to and on the workfloor and is therefore responsible for the notions that live among the workers about its management,or,for example,the cause of accidents (Gherardi et al.,1998;Hofmann and Stetzer,1998).Clearly,when policies are not put into operation,workers develop other notions about safety then when they are.For instance, when policies are sold as‘compliance’to the workforce the main message becomes to obey and,therefore,please your boss(‘comply to the rules’)rather than doing yourself a favour (‘return home unharmed’).This particular relationship between organisational,group and individual ideas and values is quite important but a somewhat underdevelopedfield in terms of research.The important role of supervisors as the tender of organisational culture in creating congruence by mixing organisation,group and individual interests into a mean-ingful whole cannot,however,be overstated.

What might be hypothesised here is that,when significant correlations appear between variables that refer to group level attitude objects in data sets that are aggregated to the level of the organisation–for instance,questions about colleagues,rule following,work-ing conditions–this might reflect the existence of shared attitudes at both the organisa-tional and group level,meaning that there is some evidence of congruence between these two levels.Such congruence is indicative of an organisational culture,e.g.see Schein (1992,1996).Nevertheless,such congruence is never a given,which means that within an organisation multiple sub-cultures canflourish.However,being both a shared and learned phenomenon,such sub-cultures can basically exist only within interacting groups or groups with a comparative formal educational background(Jones and James,1979; Schein,1996).

Summarising,when factor solutions from aggregated data sets provide factors referring to group level attitude objects,it is hypothesised that these factors express a certain degree of congruence within the organisation,with regard to that factor.In other words,groups within the organisation will show comparable attitudes.

5.3.Individual level

According to the framework by de Boer and van Drunen(2003)both rational and per-ceptual processes influence behaviour at the individual level.Perceptual processes are more of a skill-and rule-based nature whereas the rational processes are typically knowledge-based.The safety management processes‘competence’and‘procedures’typically shape all skill-,rule and knowledge-based activities(see Table1),and both should be well-bal-anced with regard to what has to be trained and well-known versus what should be done according to detailed instructions along with checklists or more sophisticated means of736 F.W.Guldenmund/Safety Science45(2007)723–743

decision support.Obviously,this balance is a matter of choice because an organisation can choose to train its people very well and leave the procedures on the shelf,or instruct its people how to use the procedures.

The aspect of competence itself has a large culturalflavour to it,as culture is learned and competence is mostly learned too,that is,apart from its physical/anthropometrical component.National cultural values are generally acquired early in life at home and at school(Hofstede,1991)whereas more practical input is provided later in life at work. The cultural values passed on through education are often more binding than the ones acquired at work through interaction,e.g.see Jones and James(1979)for the interesting finding on several navy ships where functional groups(e.g.navigation,maintenance,mis-siles)showed more perceptual cohesion than structural groups(e.g.ships and various groups on ships).With regard to organisations Schein(1996)also distinguishes executives from engineers from operators–functional groups obviously with a different educational background.

An individual level variable that has shown predictive value with regard to injury rates in hospitals is a good previous safety record of employees(Vredenburgh,2002).Employ-ees who are engaged on such a record have significantly fewer injuries.Obviously,such employees have good competence(skills or rules),which apparently can be transposed effectively to other hospitals.

Walker has been doing interesting work in Australia on psychological contracts between individual workers and their employers(Walker and Hutton,2005).A psycholog-ical contract contains implicit information about mutual expectations(especially obliga-tions)the employee has of his or her employer.When such expectations become shared between the members of a group,this may be either a structural or a functional group, they might become part of the culture this group has,i.e.become shared basic assumptions and convictions the group has about its management.The shared set of beliefs regarding reciprocal obligations may be viewed as a group level psychological contract or a norma-tive contract(Rousseau,1995).

However,within the context of culture research individual level variables do not have much meaning as the level of theory of the two levels differs.The level of theory pertains to the level at which generalisations will be made and it often also determines the level of measurement–i.e.the source level of the data.Whether individual level questions are used to construct higher-level climate dimensions should largely depend on empirical character-istics of the data.If there is sufficient perceptual agreement at a higher level of aggregation, then these aggregated values might be used.Suitable statistics for assessing perceptual agreement are the CCI(1)and CCI(2)6(Glick,1985;James,1982;Jones and James,1979).

Moreover,variables defined at the group or organisational level usually do not work very well at the individual level.This pertains to the issue of aggregation bias(James, 1982),which will be taken up below.Nevertheless,most questionnaires applied in safety climate research use questions defined at the individual(or group)level and these show up in primary factors in aggregated data sets.It has been argued above that these factors represent most likely evaluations of management per se.

6Intraclass correlation.ICC(1)denotes agreement among individuals and ICC(2)pertains to the reliability of group means(James,1982).F.W.Guldenmund/Safety Science45(2007)723–743737

6.The organisational context of culture

Within organisations three major forces are operating at the same time on(the behav-iour of)the people who work there.These generic forces are structure,culture and pro-cesses and they are dynamically interrelated,that is,the particular strength of each force is determined by the other two.This actually means that these forces are also func-tionally related in that their strengths are both meaningful and significant with regard to each other and,hence,with regard to the organisation.Together they provide the context in which behaviour,and therefore also safety related behaviour,takes place.

The structure primarily outlines the formal organisation,i.e.the proposed allocation of power and responsibilities(horizontal and vertical differentiation)and the mechanisms of communication,coordination and control.This defines how the organisational mission should be accomplished and by whom.The culture is the basic assumptions,the underly-ing convictions.For instance–‘We need a lot of supervisors because our people need to be watched constantly’.Such a conviction you willfind back in the structure of the organi-sation and therefore also on the workfloor,e.g.in the number of supervisors and where they are situated(organisationally and physically).The processes are the actual primary and supporting(incl.management)processes going on in the entire organisation;in this example the process of supervision,aimed at ensuring commitment and the reduction of violations.This might be according to structure,but this does not have to be the case. For instance,(some)supervisors do not watch constantly,or do not correct workers, although they see them make mistakes or violations.The reason for this might be struc-tural–the wrong man in the right place–or cultural–the convictions of a group of super-visors and/or their workforce do not match up to the structure.Fig.1depicts these three major forces.

One important implication of thefigure is that any organisation’s culture cannot be iso-lated from its structure or processes.In carrying out the processes and coping with diffi-culties groups of people develop a culture,either despite of or because of some particular structure.This provides a strong argument for a holistic exploration of culture, including all three elements discussed above.

738 F.W.Guldenmund/Safety Science45(2007)723–743

Within this triangle the nine management processes defined above are formulated at the organisational level,use the structural component of the triangle and shape the conditions for the primary business process(es)and related operational processes at the levels below. Safety climate is of course part of the culture component and is,evidently,influenced by both structure and processes.This triangle again illustrates that shared organisational atti-tudes(culture)might be largely determined by organisational level processes and struc-tures whereas specific local conditions(structure)and processes might shape group and individual level attitudes(culture).

7.Safety outcomes

The whole point of constructing climate scores is their assumed influence on safety related behaviour or safety outcomes.As can be seen from Table3,the relationship between climate factors and several criterion variables(often safety related behaviour)is far from convincing.Again,the issue of aggregation might play a role here,along with issues regarding levels of theory and measurement.

When scores are aggregated to the level of the organisation,individual influences dis-appear and the overall mean becomes the replacement score.However,when this aggre-gated score is combined with values still at the individual level,an uneven mix is obtained which often even produces inflated results because of‘aggregation bias’(James, 1982).

Moreover,as has been argued previously,thefindings from climate research might very well represent general attitudes towards management and its perceived influence on work-ing conditions rather than an evaluation of the conditions themselves and it may not make much sense to correlate general notions about management with safety performance indi-cators in the form of output variables(behaviour or accidents).It is therefore important when doing research that recognises multiple levels of theory that these levels are also appreciated and,furthermore,assessed with appropriate source data.

8.Conclusion and discussion

This paper has tried to explain two striking results from safety climate research:

1.Analyses provide many different factors that are hard to replicate.

2.Most analyses produce one or several higher management related or organisational fac-

tors that account for most of the variance in the data.

Attitudes towards several generic safety management processes can account for these results while these processes shape the context and working conditions for the entire orga-nisation.Nine such processes have been defined,which together cover the full safety man-agement spectrum(Guldenmund et al.,2006).It depends on the questionnaire,the primary business process and the target population how many of these safety management processes will be recognised in the factors resulting from a data analysis.Furthermore,it is argued that these factors probably represent an evaluation of management and the way they consider and handle safety.

However,not only work-related processes but also values are operative at all levels of the organisation.De Boer and van Drunen(2003)refer to relevant work by Schwarz

Table 3

Research teams,scales and relationships with criterion variables

Research team

Year Safety climate scale Criterion variable Coefficient Zohar

1980Overall (sum)safety climate score Safety inspectors’rankings of safety practices and accident prevention programs of participating plants r s =.90(chemical)r s =.80(metal)r s =.50(textile)Cox and Cox

1991See Table 2None Cabrera and Isla

1994(?)Company policies towards safety Comparisons are made between a fuelling company,airport authority and handling division all active at an airport.Rankings of these companies on safety level,safety climate and safety attitude are similar.Age and time at company are also contributing variables Emphasis on productivity versus safety Group attitudes towards safety Specific strategies of prevention Safety level perceived in the airport Safety level perceived on the job Coyle and Sleeman

1995See Table 2None Berends

1996Compliance with safe working practices Accident rate r part =À.45(n.s.)Communication about safety Response rate (on questionnaire)r part =À.32(n.s.)Williamson,Feyer,Cairns

and Biancotti

1996Personal motivation for safety Dangers in workplace (yes/no)?F =3.73(5%)Accident while working (yes/no)?F =1.94(n.s.)Positive safety practice Dangers in workplace (yes/no)?F =56.3(0.01%)Accident while working (yes/no)?F =10.5(0.1%)Risk justification Dangers in workplace (yes/no)?F =6.49(1%)Accident while working (yes/no)?F =1.62(n.s.)Fatalism Dangers in workplace (yes/no)?F =1.06(n.s.)Accident while working (yes/no)?F =0.4(n.s.)Optimism Dangers in workplace (yes/no)?F =19.5(0.01%)Accident while working (yes/no)?F =19.6(0.1%)Glendon and Litherland 2001Communication and support

Percentage safe behaviour n.s.Adequacy of procedures

Percentage safe behaviour n.s.Work pressure

Percentage safe behaviour n.s.Personal protective equipment (PPE)

Percentage safe behaviour n.s.Relationships

Percentage safe behaviour n.s.Safety rules Percentage safe behaviour n.s.

(continued on next page ) F.W.Guldenmund /Safety Science 45(2007)723–743739

Research team Year Safety climate scale Criterion variable Coefficient O’Toole2002Management’s commitment to safety Deviation from‘norm’+(n.s.)

Safety supervisory process Deviation from‘norm’+(sign.)

Emergency response Deviation from‘norm’À(sign.)

Education and knowledge Deviation from‘norm’+(n.s.)

Employee involvement and commitment Deviation from‘norm’À(n.s.)

Off-the-job safety Deviation from‘norm’+(n.s.)

Drugs and alcohol Deviation from‘norm’À(sign.) Itoh,Andersen and Seki2003Motivation Weighted accident/incident rate Trend

Morale Weighted accident/incident rate Trend 740 F.W.

Guldenmund

/

Safety Science 45 (2007) 723–743

F.W.Guldenmund/Safety Science45(2007)723–743741 (Schwartz,1992;Schwartz and Sagie,2000)and also Reiman and Oedewald(2004)refer to a‘value framework’,that they have adapted from Cameron and Quinn.In addition,the International Atomic Energy Agency(IAEA)emphasise in their SCART7-guidelines the importance of‘safety as a value’being present throughout the whole of the nuclear instal-lation(International Atomic Energy Agency,2005).

While values regarding safety seem to be important they have not been successfully assessed through self-administered questionnaires yet.Neither Reiman and Oedewald (2004)nor Berends(1996)succeeded in operationalising values or beliefs,respectively.It might very well be that safety value assessment simply requires a different research strategy.

Furthermore,the extent to which safety indeed is a value in any organisation is largely determined by the decisions and choices made at the organisational level,which start to take shape–that is,to manifest themselves–at the group level.Put in this way,(safety) climate,attitudes and(safety)culture now begin to share common borders.Particular manifestations of(un)safety visible to people at work in their groups or on their own will be,in the end,understood by them as how concerned management is with their well-being. As such safety is constructed and re-constructed not only on a daily experiential basis (cognition,behaviour)but also from the general notions of concern(affection)that‘trickle down’(DeJoy,2005)from above.

Following this line of argument it would make much sense to consider the outputs of the nine management processes not only as safety attitude objects for individual and group level evaluations but also as themes expressing(management)values regarding safety.In thefirst case the nine management process dimensions can be used as objects to generate questions to elicit the cultural dimensions,which influence whether the processes operate successfully or not.In the latter case the(self-administered)questionnaire should perhaps be used with more reservation,as an approach using a similar set of themes has not been so successful in exposing values in the past(see above).It is probably more efficient to inte-grate the value aspect in an audit tool aimed at assessing the quality of the nine manage-ment processes.Quite comparable reasoning can be found in Grote and Ku¨nzler(2000), although they primarily rely on questionnaires to elicit aspects of organisational culture and it is also not clear whether their levels of theory and measurement sufficiently overlap. Nevertheless,their approach is exemplary for the one advocated here,that is,assessing organisational and safety culture within the context of organisational processes and struc-ture,which both shape and are shaped by the present culture.

Acknowledgements

The author wishes to thank Michael Baram,Markus Scho¨bel and especially Andrew Hale for their useful comments on earlier drafts of this paper.

References

Berends,J.J.,1996.On the measurement of safety culture(unpublished graduation report).Eindhoven University of Technology,Eindhoven.

7Safety Culture Assessment Research Team.A temporary multidisciplined team assembled by the IAEA to assesses the safety culture of nuclear installations on their request.742 F.W.Guldenmund/Safety Science45(2007)723–743

Brown,R.L.,Holmes,H.,1986.The use of a factor-analytic procedure for assessing the validity of an employee safety climate model.Accident Analysis&Prevention18(6),455–470.

Cabrera,D.D.,Isla,R.(Eds.),1998.The Role of Safety Climate in a Safety Management System.Pergamon Press,Oxford.

Cabrera,D.D.,Isla,R.,Vilela,L.D.,1997.An evaluation of safety climate in ground handling activities.Paper presented at the IASC-97,International Aviation Safety Conference,The Netherlands,August27–29. Clarke,S.,2006.Contrasting perceptual,attitudinal and dispositional approaches to accident involvement in the workplace.Safety Science44,537–550.

Collins,A.M.,Gadd,S.,2002.Safety culture:A Review of the Literature.Health and Safety Laboratory,Human Factors Group,Sheffield.

Cooper,M.D.,2000.Towards a model for safety culture.Safety Science36(2),111–136.

Cox,S.,Cox,T.,1991.The structure of employee attitudes to safety:a European example.Work&Stress5(2), 93–106.

Cox,S.,Flin,R.,1998.Safety culture.Philosophers’stone or man of straw?Work&Stress12(3),1–201. Coyle,I.R.,Sleeman,S.D.,Adams,N.,1995.Safety climate.Journal of Safety Research26(4),247–254.

De Boer,J.,van Drunen,M.,2003.Occupational safety from a behavioural perspective(No.E-03/12),Institute for Environmental Studies(IVM),Vrije Universiteit,Amsterdam.

DeDobbeleer,N.,Be´land,F.,1991.A safety climate measure for construction sites.Journal of Safety Research 22,97–103.

DeJoy,D.M.,2005.Behavior change versus culture change:divergent approaches to managing workplace safety.

Safety Science43(2),105–129.

Eagly,A.H.,Chaiken,S.,1993.The Psychology of Attitudes.Harcourt Brace Jovanovich,Fort Worth. Fiske,S.T.,Taylor,S.E.,1991.Social Cognition,second ed.McGraw-Hill,New York.

Flin,R.,Mearns,K.,O’Connor,P.,Bryden,R.,2000.Measuring safety climate:identifying the common features.

Safety Science34,177–192.

Gherardi,S.,Nicolini,D.,Odella,F.,1998.What do you mean by safety?Conflicting perspectives on accident causation and safety management in a constructionfirm.Journal of Contingencies and Crisis Management6

(4),202–213.

Glendon,A.I.,Litherland,D.K.,2001.Safety climate factors,group differences and safety behaviour in road construction.Safety Science39,157–188.

Glick,W.H.,1985.Conceptualizing and measuring organizational and psychological climate:pitfalls in multilevel research.Academy of Management Review10(3),601–616.

Grote,G.,Ku¨nzler,C.,2000.Diagnosis of safety culture in safety management audits.Safety Science34(1–3), 131–150.

Guldenmund,F.W.,2000.The nature of safety culture:a review of theory and research.Safety Science34(1–3), 215–257.

Guldenmund,F.W.,Ellenbroek,M.,van den Hende,R.,2006.Organisatiecultuuronderzoek in een oer-Hollands bedrijf(Organisational culture research in a true Dutch company.In Dutch).Tijdschrift voor Toegepaste Arbowetenschap19(2),24–32.

Guldenmund,F.W.,Hale,A.R.,Goossens,L.H.J.,Betten,J.,Duijm,N.J.,2006.The development of an audit technique to assess the quality of safety barrier management.Journal of Hazardous Materials130(3),234–241.

Hale, A.R.,Kirwan, B.,Guldenmund, F.W.,1999.Capturing the river:multi-level modelling of safety management.In:Misumi,J.,Wilpert,B.,Miller,R.(Eds.),Nuclear Safety:A Human Factors Perspective.

Taylor&Francis,London.

Hofmann,D.A.,Stetzer,A.,1998.The role of safety climate and communication in accident interpretation: implications for learning from negative events.Academy of Management Journal41(6),4–657. Hofmann,D.A.,Jacobs,R.,Landy,F.,1995.High reliability process industries:individual,micro,and macro-organizational influences on safety performance.Journal of Safety Research26(3(Fall)),131–149. Hofstede,G.R.,1991.Cultures and Organisations:Software of the Mind.McGraw-Hill,London. International Atomic Energy Agency,2005.SCART guidelines.Guidelines for safety culture assessment review teams.Unpublished report.IAEA,Vienna.

James,L.R.,1982.Aggregation bias in estimates of perceptual agreement.Journal of Applied Psychology67(2), 219–229.

Jones,A.P.,James,L.R.,1979.Psychological climate:dimensions and relationships of individual and aggregated work environment perceptions.Organizational Behavior and Human Performance23,201–250.F.W.Guldenmund/Safety Science45(2007)723–743743 Mearns,K.,Whitaker,S.M.,Flin,R.,2003.Safety climate,safety management practice and safety performance in offshore environments.Safety Science41,1–680.

Neal,A.,Griffin,M.A.,Hart,P.M.,2000.The impact of organizational climate on safety climate and individual behavior.Safety Science34,99–109.

Reiman,T.,Oedewald,P.,2004.Measuring maintenance culture and maintenance core task with culture-questionnaire–a case study in the power industry.Safety Science42(9),859–8.

Rousseau, D.M.,1995.Psychological Contracts in Organizations:Understanding Written and Unwritten Agreements.Sage Publications,California.

Schein,E.H.,1992.Organizational Culture and Leadership,second ed.Jossey-Bass,San Francisco.

Schein,E.H.,1996.Three cultures of management:the key to organisational learning.Sloan Management Review 1996(Fall),9–20.

Schuman,H.,Presser,S.,1981.Questions and Answers in Attitude Surveys.Academic Press,Orlando. Schwartz,S.H.,1992.Universals in the content and structure of values:theoretical advances and empirical tests in 20countries.In:Zanna,M.P.(Ed.),Advances in Experimental Social Psychology,vol.25.Academic Press, San Diego,CA,pp.1–65.

Schwartz,S.H.,Sagie,G.,2000.Value consensus and importance–a cross-national study.Journal of Cross-Cultural Psychology31,465–497.

Simard,M.,Marchand,A.,1996.A multilevel analysis of organisational factors related to the taking of safety initiatives by work groups.Safety Science21(2),113–129.

Sorensen,J.N.,2002.Safety culture:a survey of the state-of-the-art.Reliability Engineering and System Safety 76,1–204.

Thompson,R.C.,Hilton,T.F.,Witt,L.A.,1998.Where the safety rubber meets the shopfloor:a confirmatory model of management influence on workplace safety.Journal of Safety Research29(1),15–24. Vredenburgh,A.G.,2002.Organizational safety:which management practices are most effective in reducing employee injury rates?Journal of Safety Research33,259–276.

Walker,A.,Hutton,D.M.,2005.Applying psychological contracts to workplace safety.Paper presented at the 6th Industrial and Organisational Psychology Conference,Gold Coast,Queensland,Australia,30June–3 July.

Williamson,A.M.,Feyer,A.-M.,Cairns,D.,Biancotti,D.,1997.The development of a measure of safety climate: the role of safety perceptions and attitudes.Safety Science25,15–27.

Zohar,D.,2002.The effects of leadership dimensions,safety climate,and assigned priorities on minor injuries in work groups.Journal of Organizational Behaviour23,75–92.