Mental Models of Safety - Do Managers and Employees See Eye to Eye

www.elsevier.com/locate/jsr www.nsc.org

Journal of Safety Research 34 (2003) 143–156

Mental models of safety: do managers and employees see eye to eye?

Gregory E. Prussiaa,*, Karen A. Brownb, P. Geoff Willisc

aSeattle University, 900 Broadway, Seattle, WA 98122, USAbUniversity of Washington, Bothell, 18115 Campus Way Northeast, Bothell, WA 98011-8246, USA

cUniversity of Central Oklahoma, 100 North University Drive, Edmond, OK 73024, USA

Received 12 April 2002; accepted 30 September 2002

Abstract

Problem: Disagreements between managers and employees about the causes of accidents and unsafe work behaviors can lead to serious

workplace conflicts and distract organizations from the important work of establishing positive safety climate and reducing the incidence of

accidents. Method and Results: In this study, the authors examine a model for predicting safe work behaviors and establish the model’s

consistency across managers and employees in a steel plant setting. Using the model previously described by Brown, Willis, and Prussia

(2000), the authors found that when variables influencing safety are considered within a framework of safe work behaviors, managers and

employees share a similar mental model. The study then contrasts employees’ and managers’ specific attributional perceptions. Findings from

these more fine-grained analyses suggest the two groups differ in several respects about individual constructs. Most notable were contrasts in

attributions based on their perceptions of safety climate. When perceived climate is poor, managers believe employees are responsible and

employees believe managers are responsible for workplace safety. However, as perceived safety climate improves, managers and employees

converge in their perceptions of who is responsible for safety. Impact on Industry: It can be concluded from this study that in a highly

interdependent work environment, such as a steel mill, where high system reliability is essential and members possess substantial experience

working together, managers and employees will share general mental models about the factors that contribute to unsafe behaviors, and,

ultimately, to workplace accidents. It is possible that organizations not as tightly coupled as steel mills can use such organizations as

benchmarks, seeking ways to create a shared understanding of factors that contribute to a safe work environment. Part of this improvement

effort should focus on advancing organizational safety climate. As climate improves, managers and employees are likely to agree more about

the causes of safe/unsafe behaviors and workplace accidents, ultimately increasing their ability to work in unison to prevent accidents and to

respond appropriately when they do occur. Finally, the survey items included in this study may be useful to organizations wishing to conduct

self-assessments.

D 2003 National Safety Council and Elsevier Science Ltd. All rights reserved.

Keywords: Attributions; Accidents; Mental models; Safe behaviors; Safety climate

1. Problem Although the most successful safety programs involve

Industrial workplace safety requires multilevel support

and cooperation. Top-level managers must establish a pos-

itive safety climate, supervisors must demonstrate caring

attitudes and good examples, maintenance people need to

keep equipment operating safely, and operators must estab-

lish sustainable safe work habits (DeJoy, 1994; Thompson,

Hilton, & Witt, 1998). Moreover, everyone must be in-

volved in removing safety hazards and engage in post-

incident assessments (Hofmann & Stetzer, 1998).

0022-4375/03/$ - see front matter D 2003 National Safety Council and Elsevier

doi:10.1016/S0022-4375(03)00011-2

* Corresponding author. Tel.: +1-206-296-2514.

E-mail address: [email protected] (G.E. Prussia).

multilevel cooperation, years of history and traditional

barriers across organizational lines can make it difficult.

The human tendency for blame-casting, which has been

described in the context of attribution theory (Brown, 1984;

DeJoy, 1994; Mitchell & Wood, 1980), can lead to non-

productive finger-pointing and the deterioration of cross-

level relationships. For example, in postaccident reflection,

the manager may focus on an employee’s carelessness in

using equipment. In contrast, the injured employee is likely

to identify poor equipment function or another factor

external to him- or herself as the cause. If we examine this

phenomenon in more depth, we are likely to find that even if

carelessness is implicated, the behavior is likely to have

been the outcome of the climate created by the organization

Science Ltd. All rights reserved.

G.E. Prussia et al. / Journal of Safety Research 34 (2003) 143–156144

(DeJoy, 1994). Attributions about safety have important

implications for action: DeJoy (1994) has noted, ‘‘actions

to manage safety derive more from attributions than from

actual causes’’ (p. 3).

Many of the conflicts associated with safety arise

because of differences in perception. As in the fable of

the blind men and the elephant, everyone has a tendency

to see only part of the problem. One way to consider

potential similarities/differences between managers and

employees is to examine shared mental models, an organiza-

tional phenomenon that has been studied by a number of

researchers (e.g., Klimoski & Mohammed, 1994). This

study examines safety-related perceptual/attitudinal differ-

ences between managers and operating-level employees

using a shared mental model framework. It builds on a

previous study (Brown, Willis, & Prussia, 2000) that fo-

cused on the factors underlying employees’ propensities to

work safely or unsafely, but adds a managerial dimension

for comparison.

Stemming from the researchers’ interest in mental mod-

els and cross-level differences, two general research ques-

tions are addressed. First, do managers and employees share

similar perceptions with regard to factors that lead to safe or

unsafe work behaviors? Second, if they do share a general

mental model, are there subtleties within construct relation-

ships that differ across the two groups? In terms of this

second question, the authors were interested, in particular, in

attributional effects related to safety climate.

Previous research has examined safety largely from the

employees’ experience or perspective (e.g., Brown et al.,

2000; Mottel, Long, & Morrison, 1995). However, we may

gain a more complete understanding of workplace safety by

comparing managers’ and workers’ perceptions regarding

workplace safety. If workers and managers hold different

perceptions about the intertwining causal factors that lead to

unsafe behaviors and other causes of accidents, then the

discrepancy can create a chasm between actions and the

perceived need for actions.

1.1. Employee/manager safety perceptions

Anecdotal remarks from Vukmir’s (1999) anthology of

steel worker interviews point out manager-versus-employee

differences regarding responsibility for safety. For example,

a retiree asked to comment about plant accidents recalled:

‘‘. . .When a guy got hurt he had to go to the hospital. It

happened, that happened to me. No it was human error,

or. . .well, down there. . .the responsibility rests with the

injured employee’’ (p. 232). This and other narratives from

Vukmir’s book, The Mill, add human faces and emotion to

the accident-related tensions between managers and work-

ers. However, it should be noted that Vukmir’s analysis was

based on historical reflections from the 1930s to the 1970s.

Many safety professionals would say that much has

changed since then, but are the changes real and measur-

able? More specifically, do managers and employees see

eye to eye in a very dangerous work setting where disasters

still happen?

Although first-level workers find themselves at the front

lines of production and are involved in the majority of

industrial accidents, it is likely that many root causes are

outside their control (Brown et al., 2000). Management is

responsible for creating a system within which employees

can operate safely. Unfortunately, although management

may perceive the workplace and rules governing it as

benevolent, employees may hold a contrary notion. Pre-

vious research has shown that differing and discordant

beliefs concerning the workplace lead to dysfunctional

effects on quality (Howard & Foster, 1999), corporate

culture (Shadur, 1999), teamwork (Kirkman & Shapiro,

2000), customer service (Zerbi & Dobni, 1998), perceived

fairness (Niehoff & Moorman, 1996), computer monitoring

(George, 1996), organizational commitment (McElroy &

Morrow, 1995), and personnel management (Toulson &

Smith, 1994). These consequences highlight the need for a

better understanding of the existence and extent of shared

perceptions—shared mental models—within an industrial

safety framework.

1.2. Shared mental models

The notion of shared mental models (or collective mind)

has received considerable attention in management and

organizational psychology literature (e.g., Klimoski &

Mohammed, 1994; Rousse & Morris, 1986; Weick &

Roberts, 1993). Holyoak (1984) defined a mental model

as a ‘‘. . .psychological representation of the environment

and its expected behavior.’’ Rousse and Morris (1986)

further noted that if a group shares a mental model, it serves

as the basis for future event prediction and choice regarding

courses of action. Such diagnoses and decisions are all

fundamental to the safety process in any organization. For

example, Weick and Roberts (1993) examined aircraft

carrier processes and demonstrated that groups with shared

models perform more effectively than those without a

‘‘collective mind.’’ Shared mental models are the result of

selection, training, and experience, and they are more likely

to exist under conditions of cohesiveness and membership

stability. Furthermore, they are particularly important in

environments requiring nearly continuous operating reliabil-

ity: ‘‘Organizations concerned with reliability enact mental

processes that are more fully-developed than those found in

organizations concerned with efficiency’’ (Weick & Rob-

erts, 1993, p. 357).

In the current research, the authors place the mental

model concept into the context of organizations that may

not meet the high-reliability standard of aircraft carriers, but

by their nature, have at least some of these characteristics—

steel mills. Perrow (1984) described steel plants as tightly

coupled, fast-paced systems where even a small error in one

place has ripple effects that can lead to disastrous outcomes

throughout. Thus, safety is a critical issue in steel plants, and

G.E. Prussia et al. / Journal of Safety Research 34 (2003) 143–156 145

shared mental models within groups may be indicative of

effective or ineffective behavior patterns. By exploring the

existence and influence of these models, one may better

understand the choices individuals and groups make about

safe work practices.

This research is built on a mental model first described

by Brown et al. (2000). It depicts a set of constructs that

predict employees’ self-reported propensities to work safely

or unsafely. The model is, no doubt, part of a larger mental

model, given that these ‘‘collective mind’’ networks are

normally very complex and multilayered (Weick & Roberts,

1993). Brown et al. refer to this particular mental model as a

sociotechnical model of factors predicting safe work behav-

iors. It is described briefly in the following section and

serves as the foundation for determining whether and to

what degree employees and managers share mental models

pertaining to safe work behavior.

1.3. A sociotechnical model of safe work behaviors

The model was developed using a balanced set of

indicators that comprise a sociotechnical model of causal

predictors of safe work behavior. Sociotechnical systems

(STS) represent a broad area of organizational study that

examines the interactions, synergies, and disconnects

between the social or human factors and the technical work

factors, such as layout, process design, equipment, informa-

tion, and so forth. STS concepts were first introduced by

Trist and Bamforth (1951), who have been followed by a

veritable army of disciples and others interested in identify-

ing principles that will optimize the human–system match

(e.g., Cherns, 1976; Huber & Brown, 1991).

Fig. 1. Preliminary sociotechnical mod

The sociotechnical safety model, shown in Fig. 1, was

empirically tested on employees of a specialty steel mill and

served to confirm the reasonableness of a holistic approach

to safe workplace behavior antecedents. The model and

constructs are discussed in detail in Brown et al. (2000).

Results demonstrated that employee decisions about safe

behavior are driven by a mix of social factors, technical

factors, individual attributes, and organizational conditions.

Items used to measure each of these constructs are presented

in Table 1, and short definitions of the constructs together

with a description of their interrelationships follow.

Safety hazards: Tangible factors in the work environment

that may pose risks for possible injuries and

accidents.

Safety climate: Perceptions of the role of safety within the

organization and whether or not safety is promoted

and emphasized.

Pressure for expediency over safety: Perceptions that unsafe

behaviors result from inappropriate supervisory and

organizational pressures.

Cavalier attitude: The tendency for an employee to feel

that he or she can ignore safety procedures without

incurring an accident or injury.

Safety efficacy: The individual’s self-perceived confidence

in his or her ability to work safely.

Safe work behaviors: The frequency with which an

employee reports that he or she follows safety

procedures.

Safe work behaviors represent the criterion variable in

the model (see Fig. 1). They are important because of their

el predicting safe work behavior.

able 1

onstruct scale items, means, and standard deviations (item format for

anagerial subjects)

onstruct Scale items Summed item

means (S.D.)

ack hazards Employees stand for long periods. 7.09 (3.31)

Work surface heights are not correct

for employees.

6.92 (3.58)

Employees work in physically

awkward positions.

7.72 (3.76)

Employees lift objects that are too

heavy.

8.31 (3.57)

Employees lift objects that are too

bulky or large.

7.32 (3.59)

Employees work on elevated

surfaces or walkways.

6.78 (3.77)

Employees work in confined

spaces.

5.93 (3.64)

dustrial

hygiene

Employees work near electrical

current.

8.75 (4.10)

hazards Employees work near open flames. 7.23 (3.97)

Employees are exposed to toxic

chemicals.

7.49 (4.03)

Employees are exposed to infectious

agents.

3.74 (2.71)

Employees are exposed to toxic gas. 5.64 (3.91)

Employees are exposed to radiation. 3.55 (2.63)

itting hazards Employees sit in the same place for

long periods.

6.67 (2.91)

Employee chairs are uncomfortable. 5.89 (3.54)

Employee chairs are not sturdy. 5.07 (3.48)

quipment

unavailability

Safety equipment is not within

easy reach.

4.70 (2.85)

hazards Safety equipment does not work

very well.

4.44 (2.61)

Tools or equipment do not work the

way they should.

6.74 (3.09)

Employees do not have the correct

tool for the job.

5.81 (3.01)

quipment

handling

Employees work with equipment

that vibrates a lot.

6.72 (3.64)

hazards Employees work with tools that are

awkward to hold.

6.26 (3.42)

Tools/equipment are too cold to

hold comfortably.

3.51 (1.81)

Tools/equipment are too hot to

hold comfortably.

5.72 (3.53)

anagerial

safety climate

Top management believes work place

safety and health are very important.

5.97 (1.47)

The union/company safety committee

is effective in improving workplace

safety and health.

5.69 (1.30)

The company is concerned about

the safety and health of employees

when they are away from work.

5.08 (1.66)

Overall, this is a safe place to work. 5.58 (1.26)

ressure Employees take safety shortcuts

when they feel pressure to work fast.

4.05 (1.60)

It is difficult for employees to do

their tasks while following all of

the safety rules.

3.24 (1.66)

Employees are encouraged to take

shortcuts in safety procedures.

1.66 (1.08)

afety efficacy I am skilled at helping employees

to avoid the dangers of work place

hazards.

5.05 (1.24)

Table 1 (continued )

Construct Scale items Summed item

means (S.D.)

Safety efficacy I am very active in removing

workplace safety and health

hazards.

4.94 (1.45)

I am confident in my ability to

remove work place safety and

health hazards.

5.25 (1.21)

Cavalier attitude

toward safety

procedures

Employees feel they can do the

job safely without following

safety procedures.

3.89 (1.75)

Employees ignore some safety

procedures if they are trying to

save time.

4.08 (1.73)

Employees feel that the safety

procedures are not necessary.

3.43 (1.68)

Safe work

behavior

What percent of the employees

in your area follow all of the

safety procedures for the jobs

that they do?

53.36 (28.47)


T

C

m

C

B

In

S

E

E

M

P

S

link with workplace accidents (Mottel et al., 1995;

Thompson et al., 1998). They have gained increasing

attention in recent years with the safety community’s

growing interest in behaviorally based safety programs.

This variable was selected in lieu of accidents because of

difficulties with frequency of occurrence, flaws in self-

reporting of accidents, and the proactive nature of

addressing behaviors rather than reacting to accidents.

Thompson et al. (1998) provide further arguments for

behavior rather than accidents as a useful criterion

variable.

The constructs on the left side of Fig. 1 (i.e., hazards,

climate, pressure) are predominantly system-level attrib-

utes. These system factors are predicted to influence

person-level factors (cavalier attitude and safety efficacy)

and only indirectly affect safe work behavior. In contrast,

the person-level constructs are expected to directly influ-

ence ultimate safe work behaviors. Brown et al. (2000)

tested this sociotechnical safety behavior model using data

from a survey completed by 551 operating-level employees

in a steel mill. By testing several other models with

competing perspectives, they found this particular model

to be the best-fitting relationship configuration. However,

they did not examine whether employees and managers

shared this mental model.

1.4. Hypotheses

The purpose of the present study is to extend previous

research examining the extent to which managers and

employees see eye to eye on important safety issues. It

was expected that managers and employees would share a

mental model about general safety relationships, the

sociotechnical model in Fig. 1, for two reasons: (a) shared

mental models are particularly important in environments


requiring nearly continuous operating reliability (Weick &

Roberts, 1993); (b) as discussed previously, groups that

have worked together over time and who share commit-

ment to organizational goals tend to exhibit less variance

in their mental models of system functions, behaviors,

and casual sequences. The focal organization in the

current research had just implemented a company-wide

behaviorally based safety program in the preceding 3-year

period. (Although the authors have witnessed situations in

which behaviorally based programs inappropriately shift

responsibility for safety onto the shoulders of employees,

the system implemented in this organization was based on

a strong sense of managerial responsibility and

employee–manager partnership.) Additionally, turnover

in the plant was almost nonexistent; most employees

had spent their entire careers with the company. Con-

sequently, managers and employees within this context

were expected to share general beliefs regarding important

safety relationships.

Although managers and employees possibly share an

overall mental model, they are more likely to differ

regarding specific safety characteristics and the strength

of their relationships. These differences may be the result

of unique attributions made by managers and employees.

Researchers have previously documented the tendency for

an observer to blame a person when an unfortunate event

occurs even when such blame is inappropriate (Brown,

1984; Mitchell & Wood, 1980). This blame-casting bias is

labeled the fundamental attribution error. Furthermore, the

tendency to externalize problems or take undue credit for

success is termed the self-serving bias. These attribution

biases may explain why managers and employees some-

times view safety issues differently. For example, DeJoy

(1986, 1994) noted that when employees perceive expe-

diency-oriented pressures, they are more likely to cut

corners. However, it is doubtful that managers share

employees’ perceptions about either the existence of such

pressures or their possible antecedents. Indeed, managers

might make self-serving attributions to avoid blame asso-

ciated with accidents resulting from pressure-induced

shortcuts. Based on these tendencies, managers and

employees were expected to differ in their perceptions of

the individual constructs depicted in the sociotechnical

model.

Finally, while managers might attempt to foster a safe

work environment, specific employee perceptions regarding

the influence of safety climate on important safety outcomes

may differ from those maintained by managers because of

differences in attributions. Managers have great control over

safety climate (Hofmann & Stetzer, 1996; Perrow, 1984;

Vaughan, 1996), but whether they exercise that control and

whether it ultimately influences safe behavior and percep-

tions of responsibility varies among organizations. How-

ever, given the tendency for blame casting within a safety

context (e.g., DeJoy, 1994) together with the inclination

toward self-serving biases, we expect employees and man-

agers to differ in their mental models concerning important

safety climate effects.

In sum, expected that managers and employees in a steel

plant would share a general mental model regarding the

factors that predict employee safe behavior. However, based

on the authors’ understanding of attribution theory and

safety climate, authors predicted that a more fine-grained

analysis would reveal differences between the two groups.

The hypotheses were tested using a field study, as described

in the following.

2. Method and results

2.1. Sample

Data were collected from a specialty steel company in

the southeastern United States using a survey instrument

and supporting observational study. The organization had

engaged in a comprehensive effort to improve safety

climate and behavior during the 5 years before the study.

The safety program was comprehensive and involved em-

ployee input.

Survey participants included operating-level employees

as well as managers or supervisors. For the present study,

two separate data sets were used: a manager only data set

and a combined data set (employees and managers). The

combined data set included some of the variables used in

the Brown et al. (2000) study, but the current data and

analysis are unique because (a) they include managers, (b)

they allow for an examination of manager-specific data

as well as comparisons between managers and employees,

and, (c) they include variables not considered in previous

research.

Survey items were developed based on extensive inter-

views in several plants across several industries, including

the facilities where the survey was administered for this

study. The survey was pilot-tested with both operating-level

personnel and management-level personnel from both of the

plants where the study was to be conducted. Survey intro-

duction to managers, conducted by one of the authors and

embellished with supporting commentary from top manage-

ment, took place in the company auditorium. A total of 121

managers out of 190 responded to the survey for a response

rate of 64%. Five hundred and fifty-one out of 800 operat-

ing-level employees from 19 departments returned usable

surveys resulting in a response rate of 69%. Respondents

represented a cross-section of departments and shifts and

included those who supervised or conducted work in the

areas of equipment operation, maintenance, quality control,

and others.

Given the two objectives of this research, the data were

considered in two stages. In Stage 1, the structural modeling

analyses was used to examine the extent to which the

sociotechnical model that was tested previously on em-

ployee data yielded similar results with a different popula-


tion—managers. This analysis would indicate whether man-

agers and employees perceive a similar set of important

safety relationships when considered simultaneously. Next,

in Stage 2, the analysis was narrowed to specific variable

relationships. This more fine-grained analysis examined

differences between employees and managers regarding

perceptions of important safety outcomes. The influence of

climate on the perceptions of safety outcomes was also a

focus of the second analysis.

2.2. Stage 1: Structural modeling analyses

Initial structural modeling procedures involved the man-

ager-only data set to assess fit with the previously estab-

lished sociotechnical model. If the model fit, this would

suggest that managers and employees share a similar mental

model concerning the measures and overall constellation of

relationships leading to safe work behavior. Subsequent

analyses in Stage 1 used the combined data set and focused

on the equivalency of modeled relationships across manager

and employee subgroups. If relationships were equivalent,

this would provide added evidence regarding the degree to

which both groups share a mental model when considering

safety relationships simultaneously. To answer these ques-

tions, we applied structural modeling procedures to the

model presented in Fig. 1.

2.2.1. Measures

All the constructs specified in the model (and their

associated indicators) were previously derived and tested

using the employee-only data set (Brown et al., 2000).

Similar latent constructs were created and indicators from

the manager data set were pinpointed to enable model testing

and facilitate group comparisons. Using latent variables

allowed the authors to avoid problems associated with

measurement error (James, Mulaik, & Brett, 1982). Items

included in both employee and manager surveys were

virtually identical, although items in the manager surveys

indicated how the managers felt these factors affected

employees. For example, an employee item, such as ‘‘Tools

that I need are difficult to reach,’’ was modified for the

manager survey to ‘‘Tools that employees need are difficult

to reach.’’ All measures in the employee and manager

surveys reflect this subtle difference unless otherwise indi-

cated. The following measures were derived from the man-

ager-only data set.

2.2.1.1. Safety hazards. Twenty-four items represented

workplace hazards in the plant. Managers were asked to

rate the extent to which they considered each item to be a

serious hazard for employees and the frequency with which

it occurred. Responses for the hazard seriousness items

ranged from 1 (not at all) to 7 (to a great extent), and

responses on the corresponding frequency items ranged

from 1 (never) to 7 (very often). An importance score was

calculated for each hazard item by summing its seriousness

rating with its frequency rating. This approach reduces

problems associated with magnifying error through multi-

plication of two items (Schmidt, 1973). Just the same,

separate factor analyses were run on the hazard variables

using both multiplicative and additive models. Using a scree

plot and an eigenvalue of 1 as a cutoff, results showed that

the components of each eigenvector were identical and the

largest value in the additive analysis was the same as the

largest value in the multiplicative analysis. Thus, no differ-

ences were found based on the approach used. The additive

model was used heeding Schmidt’s (1973) warnings about

stability and theoretical soundness. Because of the summing

procedure, the possible score for each item ranged from 0 (if

it was not applicable) to 14 (if it received the highest

possible rating of 7 for each of the item subcomponents).

The higher the rating, the more serious and more present the

hazard.

To be consistent with the Brown et al. (2000) measures

from employee data, the same five hazard factors were drawn

from the manager-only data to indicate safety hazards.

Specifically, these five factors included back hazards,

hygiene hazards, sitting hazards, equipment unavailability

hazards, and equipment handling hazards (see Table 1 for

items included within each category). Cronbach’s alpha for

the scales represented by each of the five factors ranged from

.76 to .87. In a further move toward data reduction, impor-

tance scores were averaged for individual items within each

factor to conserve degrees of freedom in model estimation

and for the sake of parsimony (James et al., 1982).

2.2.1.2. Safety climate. Four items representing percep-

tions of upper management’s influence on workplace safety

served as indicators of the Safety Climate construct. An

example is, ‘‘Top management believes workplace safety

and health are very important.’’ Coefficient alpha for the

items was estimated at .88. Although these items were

previously included as an aggregated scale indicator of

safety climate (see Brown et al., 2000), in this study, each

item was used as a separate indicator because the manager

survey did not include other items representing ‘‘super-

visory’’ safety climate. All items were rated on 7-point

Likert-type scales ranging from 1 (strongly disagree) to 7

(strongly agree). These items are similar to those Zohar

(1980) included in his examination of industrial safety

climate; however, the individual rather than the organization

was used as the unit of analysis.

The authors focused on the individual for three reasons.

First, observations at the study site led the authors to believe

there were safety climate differences across departments.

Second, the model originally proposed and tested by Brown

et al. (2000) emphasized the effects of employees’ cognitive

interpretations on the person-based outcomes highlighted in

the model. Safety climate was measured at the individual

level in that study to be consistent with other measures in

the model. Third, according to Klimoski and Mohammed

(1994), ‘‘mental models and other cognitive constructs have


traditionally been considered at the individual-level of

analysis’’ (p. 406). Examining mental model differences

between just two groups almost necessitates an individual-

level focus.

2.2.1.3. Pressure. Pressure was assessed to value expe-

diency over safety with three separate item indicators. For

example, managers rated the extent to which ‘‘Employees

take safety shortcuts when they feel pressure to work fast.’’

Responses ranged from 1 (never) to 7 (very often). Coef-

ficient alpha for the three-item scale was only .58 (it had

been .69 for similar items in the employee survey), but each

item was used as a separate indicator so the authors could

be consistent with the previous measurement model and

avoid relying on a single indicator for the pressure con-

struct.

2.2.1.4. Cavalier attitude. Three items measured the

extent to which the managers believed employees held what

was termed a ‘‘cavalier’’ attitude toward safety procedures.

For example, managers assessed the degree to which

‘‘Employees feel that they can work safely without follow-

ing safety procedures.’’ Again, responses ranged from 1

(never) to 7 (very often). Coefficient alpha for this three-

item scale was estimated at .87. Similar to the way in which

the pressure construct was handled, three items were used to

serve as unique indicators of the cavalier construct.

2.2.1.5. Safety efficacy. Based on Bandura’s (1986) rec-

ommendation regarding the measurement of efficacy per-

ceptions, three items were used that assessed safety efficacy

strength. An example item was, ‘‘I am confident in my

ability to remove workplace safety and health hazards.’’

Responses ranged from 1 (strongly disagree) to 7 (strongly

agree). These items were different from employee items in

that they did not capture managers’ impressions of employ-

ees, but rather managers’ beliefs about their own abilities/

behaviours. Coefficient alpha was .85, but for modeling

purposes, the authors used each item as a separate indicator

of the safety efficacy construct.

2.2.1.6. Safe work behavior. Although the employee data

included two items to assess safe work behavior, the man-

agers assessed only one of the two items. Thus, the authors

used a single item as an indicator of this criterion variable.

The item asked managers: ‘‘About what percent of the

employees in your area (or in the plant if you do not

supervise operating personnel) follow all of the safety

procedures for the jobs that they do?’’ Possible responses

ranged from 0% to 100%, in 10% increments. Because the

construct was assessed using a single indicator, the research-

ers corrected for measurement error using the formula

provided by Williams and Hazer (1986). Specifically, the

square root of the reliability estimate of the indicator (esti-

mated at .85) was used to fix the measurement parameter.

Furthermore, the error variance was fixed to 1 minus the

reliability multiplied by the item variance. This procedure is

common in covariance structure analysis (e.g., Farkas &

Tetrick, 1989; Prussia & Kinicki, 1996; Wayne & Ferris,

1990), and the resulting parameter estimates are accurate and

unbiased (see Netemeyer, Johnston, & Burton, 1990).

All constructs and indicators were specified such that

analogous measures existed for both manager and employee

data sets. This facilitated the eventual creation of the

combined data set.

2.2.2. Analyses

Initial Stage 1 analyses focused on the manager-only

data. Before examining structural relationships in the model,

the authors wanted to see if the measures made sense and

whether the various constructs were theoretically independ-

ent. Thus, a confirmatory factor analysis of the proposed

measurement model was run initially. This established a

baseline measurement model. Next, the researchers eval-

uated the theoretical independence of the proposed con-

structs by comparing the baseline measurement model to a

collapsed model that specifies perfect correlation among all

latent variables. Structural model relationships were exam-

ined after measurement model propriety was established.

Subsequent Stage 1 analyses involved the combined data

set on which the authors made multiple group comparisons

using a procedure in EQS (Bentler, 1995). More specifi-

cally, both data sets were included in a model in which the

structural paths were freely estimated. This model was then

compared to a model in which the structural paths were

constrained to be equivalent across subgroups. Model chi-

square values were then compared to determine whether the

models were significantly different. If the constrained model

did not significantly differ from the freely estimated model,

it would suggest that employees and managers view impor-

tant safety relationships similarly.

Covariance structure analysis (Bentler, 1995; James et

al., 1982) was used to examine the proposed model initially

with the manager data only and then with the combined data

set. CFI and IFI values of .90 and greater indicate adequate

model fit (Bollen, 1989). PFI values of .60 and greater are

suggested as a rule of thumb criterion for model retention

(Williams & Podsakoff, 1989). RMSEAvalues of .08 or less

indicate reasonable error of approximation, and values of

.05 or less indicate a close fit (Browne & Cudek, 1992).

Individual model paths were evaluated for significance, and

the sequential chi-square difference test (James et al., 1982)

was used to evaluate model comparisons. All models were

tested using procedures in Bentler’s (1995) EQS program.

2.2.3. Measurement model results

Means, standard deviations, and indicator correlations

are provided in Table 2. The results of the confirmatory

factor analysis appear in Fig. 2. The figure highlights the

factor loadings of the indicators associated with each of the

six latent constructs included in this model. The measure-

ment model fit the data well, v2(138) = 190.17, p < .05, and

Table 2

Descriptive statistics and interrelationships among indicators

Indicator Mean S.D. Correlations

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

1. Back hazards 7.19 2.68 –

2. Industrial

hygiene hazards

6.13 2.62 .69** –

3. Sitting hazards 5.88 2.7 .52** .37** –

4. Equipment

unavailability

hazards

5.42 2.23 .69** .62** .50** –

5. Equipment

handling hazards

5.56 2.45 .77** .61** .49** .77** –

6. Top management

safety belief

5.97 1.47 � .26* � .21* � .10 � .37** � .23** –

7. Union/Company

safety

effectiveness

5.69 1.3 � .08 � .19* .01 � .35** � .14 .64** –

8. Company

concern about

safety

5.08 1.66 � .19* � .28** � .06 � .37** � .24** .59** .63** –

9. Overall safe

place to work

5.58 1.26 � .23* � .29** � .08 � .36** � .33** .65** .67** .70** –

10. Pressure for

shortcuts

4.05 1.6 .30** .15 .19* .25** .17 � .12 � .11 � .09 � .14 –

11. Cumbersome

rules

3.24 1.66 .34** .27** .04 .20* .14 � .13 � .12 � .06 � .15 .29** –

12. Supervising

encouragement

1.66 1.08 .25** .20* .22* .31** .19* � .29** � .19* � .12 � .06 .29** .34** –

13. Skill at

avoiding dangers

5.05 1.24 � .02 � .01 � .02 � .18* � .11 .31** .39** .29** .46** � .04 � .27** � .02 –

14. Active in

removing hazards

4.94 1.45 � .14 � .12 � .14 � .24** � .17 .35** .46** .42** .47** � .13 � .24** � .20* .63* –

15. Confident in

removing hazards

5.25 1.21 � .14 � .11 � .10 � .20* � .11 .33** .41** .30** .46** � .08 � .27** � .15 .67** .65** –

16. Work safely

without rules

3.89 1.75 .36** .26** .17 � .23** .21* � .18* � .22* � .21* � .24** .39** .22* .16 � .08 � .11 .00 –

17. Ignore rules

to save time

4.08 1.73 .48** .41** .32** � .37** .39** � .20* � .17 � .15 � .24** .57** .32** .33** � .03 .09 .02 .76** –

18. Safety

procedures not

necessary

3.43 1.68 .34** .16 .17 � .26** .20* � .19* � .14 � .18* � .22* .35** .23** .28** � .07 � .18 � .04 .71** .59** –

19. Percent of time

rules followed

53.36 28.47 � .11 � .05 .03 � .04 � .05 .23* .24* .15 .23* � .06 � .00 � .03 .06 .08 .10 � .34** � .31** � .27** –

*P<0.5.

**P<0.1.

G.E.Prussia

etal./JournalofSafety

Resea

rch34(2003)143–156

150

Fig. 2. Baseline measurement model results.


all standardized factor loadings were significant (M=.79).

Furthermore, the model in which proposed constructs were

collapsed, v2(153) = 876.05, p < .05, was a significantly

worse fit to the data when compared to the baseline

measurement model as seen from the chi-square difference

test, v2(15) = 685.88, p < .05. There are two implications of

these results. First, when compared to previous findings

(i.e., Brown et al., 2000), these results suggest that managers

and employees interpret the measures similarly; thus, they

can be considered separately or together. Second, the results

provide evidence regarding the discriminant validity of the

specified constructs.

2.2.4. Structural model results

Initial structural model tests show that the model fit the

data well when the manager data alone was used (see Fig.

3). Results from this model show that managers believe the

presence of safety hazards can lead to increased pressure on

employees to value expediency over safety (.46). Ironically,

this suggests that the presence of hazards may lead to a

reduced emphasis on safety, perhaps due to an impression

that safety is a low organizational priority. In contrast

however, managers feel that a strong safety climate can

diminish perceptions of pressure (� .31). Results further

show that managers believe system factors affect safe

behaviors indirectly through person factors. Specifically,

pressure is positively related to cavalier attitude (.72) and

negatively related to safety efficacy (� .35). Thus, manag-

ers feel that when there is more pressure to ignore safety

rules and guidelines, employees may develop cavalier

attitudes. Furthermore, as they see this pressure rising,

managers have less confidence in their own abilities to

work safely. Finally, managers believe that a cavalier

attitude, a person-level factor, results in less safe behavior

(� .24). The structural paths along with overall model fit

are very similar to results derived from the employee-only

data (see Brown et al., 2000).

Subsequent tests involving the combined data set indi-

cated that the two subgroups did not differ when the model

paths were constrained to be equal. Specifically, the model

in which the paths were constrained to be equivalent,

v2(299) = 623.56, p < .05, was not a significantly worse fit

compared to the model in which the paths were freely

estimated, v2(292) = 615.17, p < .05. This was demonstrated

using the sequential chi-square difference test, v2(7) = 8.39,p>.10. Together with the results from the initial structural

model analyses, these results demonstrate that the socio-

technical model fits for both employee and manager sub-

groups. More specifically, employees and managers share a

general mental model when multiple relationships are con-

sidered simultaneously.

2.3. Stage 2: Specific variable relationships

Following Stage 1 analyses, the authors wanted to

examine specific differences between employees and man-

agers regarding safety constructs. To do this, the authors

first compared employee and manager perceptions regarding

the constructs specified in Stage 1 analyses. The researchers

next focused on perceived safety climate differences. Spe-

cifically, the authors wanted to determine if the relationship

between climate and important safety outcomes differed

between employees and managers. The authors focused on

perceived climate differences because managers have great

control over the existence and ultimate effects of a safe

climate (Hofmann & Stetzer, 1996; Perrow, 1984; Vaughan,

1996) and climate affects attitudes and behaviors (Hoffmann

& Stetzer, 1996; Zohar, 1980). All Stage 2 analyses used the

combined data set.

2.3.1. Measures

Indicators of Stage 1 constructs were averaged to create

separate aggregated variables to compare employee–man-

ager perceptions of the safety constructs in the model. For

example, the three items used as indicators of climate in

Stage 1 were averaged to create a single climate variable for

Stage 2 analyses. Similar procedures were used to create

aggregated hazard, pressure, efficacy, and cavalier attitude

Fig. 3. Structural model results.


variables. Thus, five aggregated variables were created. The

single item safe work behavior measure was the same as the

one used in Stage 1.

In addition, a measure not used in Stage 1 represented a

second important safety outcome—responsibility for safety.

This item asked all respondents, ‘‘Who is responsible for

workplace health and safety?’’ Respondents then indicated,

using a maximum of 100 points, the percentage of respon-

sibility they attributed to employees (M = 39.90 for the

combined sample). Finally, a variable labeled position

reflected whether the respondent was an employee (0) or a

manager (1).

2.3.2. Analyses

Initial analyses evaluated differences between managers

and employees on the aggregated constructs. T tests using

pooled variance estimates were conducted to examine group

differences. Next moderated regression (Neter, Wasserman,

& Kutner, 1989) was used to examine differences regarding

the influence of safety climate on important safety out-

comes. Specifically, two regression equations tested the

extent to which the influence of climate on safety outcomes

is dependent on whether the respondent is an employee or

manager. In the first equation, safe work behavior was

regressed on safety climate, position, and a climate/position

interaction term. In the second equation, responsibility for

safety was regressed on safety climate, position, and a

climate/position interaction term. The aggregated climate

measure was used in both equations.

2.3.3. Results

T tests revealed significant differences ( p < .05) between

managers and employees on four of the aggregated con-

structs as well as the safe work behavior variable. Relative to

employees, managers perceived greater levels of two of the

three system-level constructs. Managers perceived signifi-

cantly stronger safety climate (M = 5.58) and believed

employees were significantly more likely to react to pressure

(M = 2.98), in contrast to employees’ beliefs about themselves

(M = 4.99 and M = 2.48, respectively). No significant differ-

ences were found for hazard perceptions between managers

(M = 5.51) and employees (M = 5.66), suggesting the two

groups interpret the physical work environment similarly.

For the person-level constructs, managers were significantly

less confident (safety efficacy) than employees in their ability

to work safely (M = 5.07), but believed that employees were

significantly more cavalier (M = 3.80) than employees

believed themselves to be (M= 5.36,M = 2.08, respectively).

Finally, employees estimated the frequency with which they

engage in safe work behaviours to be much higher than what

managers attributed them (respectivemeanswere 82.15% and

53.36%). These results suggest that although employees and

managers share a mental model of the factors affecting safe

employee behaviors, they also maintain unique perspectives

(i.e., lack a shared mental model) when more fine-grained

comparisons on attributions are made.

Results for both regression equations are displayed in

Table 3, and interaction plots are given in Figs. 4 and 5. The

regression of safe work behavior on safety climate and

Fig. 5. Interaction plot of safety climate predicting percent employee

responsibility for safety.

Table 3

Regressions on important safety outcomes (N = 672)

Independent variables Dependent variables

Safe work

behavior

Responsibility

for safety

Safety climate (b1) .12* .10*

Position (b2) � .90* .56*

Safety Climate� Position (b3) .40* � .47*

F 61.97* 4.72*

R2 .28 .03

*p< .05.


position, together with the climate–position interaction

term, revealed a statistically significant model explaining

28% of the variance in safe work behavior. The main effects

of safety climate and position were both significant as was

the interaction term. The interaction plot appears as Fig. 4

and demonstrates that levels of safety climate have a differ-

ent impact on front-line workers than on managers. Under

conditions of poor climate, managers and employees appear

to disagree about the extent to which employees engage in

safe work behaviors. However, as safety climate improves,

managers are more likely to report higher percentages of

safe worker behavior. While the same holds true for work-

ers, the effect is noticeably diminished. A stronger safety

climate apparently influences the degree of convergence or

shared perception between employees and managers.

The prediction of responsibility for safety based on safety

climate, position, and the climate/position interaction term

also yielded statistically significant results for all three terms

in the model. The model’s F statistic is significant, but the

R2 term suggests that the predictor variables do not capture

the majority of the variance in responsibility. Just the same,

the interaction plot (Fig. 5) illustrates the significance of the

climate/position interaction term. Under poor climate con-

ditions, managers and employees differ in their attributions

for responsibility, but the two groups converge in their

Fig. 4. Interaction plot of safety climate predicting employee percent safe

behaviour.

opinions as safety climate improves. It appears that manag-

ers may be less likely to make fundamental attribution errors

when the safety climate is strong.

3. Discussion/limitations

This study addressed two research questions: First, do

managers and employees share similar perceptions with

regard to factors that lead operators to engage in safe or

unsafe work behaviors? Second, if they do share mental

models, are there subtleties within construct relationships

and beyond these models that differ across the two groups?

Within the second question, specific queries about (a)

whether the role of safety climate and its interaction with

position on perceptions about safe work behaviors were

addressed and (b) the role of safety climate as an influence

on manager versus employee perceptions of safety respon-

sibility.

Regarding the first question, results demonstrate that

managers and employees, in this particular setting, share

an embedded mental model about the factors that influence

safe behavior decisions. The authors expected that the two

would agree about how things work based on the authors’

understanding of the tendency for mental model sharing in

tightly coupled systems (Weick & Roberts, 1993), and on

the fact that there was little or no turnover in the organ-

ization, allowing for years of shared experience. In addition,

the authors believed that the strong safety climate derived

from the recently implemented safety programs and

observed through ethnographic study would contribute to

an overall shared mental model between the two groups.

The fact that managers and employees shared this overall

mental model provides testimony for the value of well-

implemented safety programs.

Analyses designed to answer the second question and its

subsets demonstrate that, in spite of an embedded mental

model about systematic human and technical relationships


managers and employees disagree to some extent on their

perceptions of most of the safety constructs that were

measured. The one exception was safety hazards; the two

groups agreed about the presence of hazards. Hazards,

mostly identified as physical conditions, are less subject to

interpretation than other, more socially oriented variables.

Thus, the pattern of agreement across hazard construct sets

made sense. On the other hand, differences (e.g., managers

generally viewed employees as more cavalier about safe-

ty than employees believed themselves to be) are best

explained by the naı̈ve attributional biases so pervasive in

human nature. Beyond the simple contrasts, regression

analyses suggest that perceived safety climate differentially

influences manager and employee perceptions of the percent

of the time that employees engage in safe or unsafe work

behaviors. As perceived climate improves, managers believe

there will be fewer unsafe behaviors. Employees reported

safer behavior when climate improved, but the relationship

was not as strong, suggesting that managers may over-

estimate the beneficial effects of climate on employees—

perhaps an example of self-serving bias.

There also were differences in the way managers and

employees attribute responsibility for safety, and these

differences were influenced by safety climate. In the parts

of the plant where people felt safety climate was poor,

managers and employees disagreed significantly in their

attribution assessments (managers believed employees were

responsible and vice versa). In areas where climate was

better, they tended to make similar attributions for respon-

sibility. This suggests that the establishment of a better

climate diminishes the effects of the fundamental attribution

error. Similar to the prediction of safe work behavior, as

safety climate improves, employee and manager perceptions

apparently converge and the two groups share perspectives.

In sum, shared mental models, such as the sociotechnical

model of workplace safety, allow employees and managers

to operate within a common paradigm. However, although

general relationships among a constellation of constructs

may be shared, managers and employees differ significantly

regarding specific perceptions as well as attributions result-

ing from safety climate effects. Just the same, organizations

can operate more effectively when employees and managers

share mental models pertaining to safety. Furthermore, a

well-run safety program may be a precursor to mental model

sharing within and between groups.

DeJoy (1996) noted that the safety climate could serve as

a guiding paradigm for cause and effect within the context

of workplace safety. The authors believe that to establish a

multilevel shared mental model around safety, managers

should establish and improve safety climate. Based on

observations from the focal plant in this research, managers

may take a number of actions to engender a positive safety

climate. Although experience in many organizations dem-

onstrates that the potential list of initiatives is much longer,

four factors seem to stand out about the safety program at

this research site: (a) Focus on behaviors—the company in

this study ended its emphasis on recordable accident sta-

tistics, switching instead to an emphasis on safe behaviors

that included daily safety meetings. However, they made

clear their understanding that management held significant

responsibility for safety. (b) Involve employees—employees

were brought into the decision process for workplace safety.

(c) Use symbols—one day, a safety manager handed out a

silver dollar to every employee who was wearing a seat belt

while driving through the gate at the end of the day. (d)

Remove hazards—managers showed a sincere concern

about hazards and took great efforts to remove them.

Industrial safety scholars have advocated these activities

for several years. However, this research goes beyond

anecdotal reports to demonstrate that safety climate has a

real and measurable effect on safety behaviors, and that it is

critical to creating a shared mental model of safety between

employees and managers.

Although this study adds to the understanding of shared

mental models regarding safe work behaviors, the authors

acknowledge three specific limitations of this research. First,

the data were collected from a single firm in a single

industry. Thus, these results may be generalizable only to

other firms within the steel industry that have implemented

safety-oriented programs. Clearly, future research should

attempt to replicate these findings in other environments.

A second limitation involves the use of same-source data;

common method bias can lead to spurious relations among

modeled variables. However, the study tried to mitigate this

potential problem by (a) including multiple groups in these

analyses, (b) measuring the indicators with different types of

response categories, and (c) using observationally based

validity checks. Furthermore, the authors tested a collapsed

measurement model and found support for the multidimen-

sional nature of these constructs. The poor fit of the

collapsed model provides evidence that these results are

not due to a common method factor.

Finally, the authors recognize that correlation does not

imply causation, and these data limit the inferences that can

be made. However, these results indicate that the causal

relations depicted in the sociotechnical model cannot be

rejected. Just the same, the authors hope future research

would replicate these findings using methods from which

causal inferences are more appropriate.

4. Impact on industry

In the aftermath of an industrial accident or ‘‘near miss,’’

there is a natural tendency for employees to blame the

system (including management) and managers to blame

employees. Managers must be aware of these biases when

diagnosing and responding to accidents. In spite of these

differences, it appears that in a tightly coupled industrial

setting, such as a steel plant, managers and employees may

share causal mental models, perhaps out of necessity in

keeping a complex system running effectively. Managers in


more decoupled systems may wish to use tightly coupled

and necessarily high-reliability systems as benchmarks

when they try to increase agreement and mental model

sharing across organizational levels. When organizations

improve safety climate, they can move employees and

managers closer together in their perspectives about safety

responsibility. Given the importance of attributions about

responsibility in shaping postaccident response, an organ-

ization and its employees can benefit from such a conver-

gence in perspectives. The survey items included in this

study provide research insights, but they also may be useful

to those who wish to conduct safety self-assessments within

their organizations. For example, a manager may wish to

measure safety climate and/or perceived safety hazards

before and after a major safety improvement initiative.

In sum, the authors believe that examining the existence

and extent of shared mental models can lead to improve-

ments in organizational safety. This research represents a

modest investigation of this pursuit in a defined population.

Future research should further examine the importance of

shared mental models and determine their application across

safety contexts.

References

Bandura, A. (1986). Social foundations of thought and action. Englewood

Cliffs, NJ: Prentice-Hall.

Bentler, P. M. (1995). EQS structural equations program manual. Encino,

CA: Multivariate Software.

Bollen, K. A. (1989). Structural equations with latent variables. New York:

Wiley.

Brown, K. A. (1984). Explaining group poor performance: an attributional

analysis. Academy of Management Review, 9, 54–63.

Brown, K. A., Willis, G., & Prussia, G. (2000). Predicting safe employee

behavior in the steel industry: development and test of a sociotechnical

model. Journal of Operations Management, 18, 445–465.

Browne, M. W., & Cudek, R. (1992). Alternative ways of assessing model

fit. Sociological Methods and Research, 21, 230–258.

Cherns, A. B. (1976). The principles of sociotechnical design. Human

Relations, 9(8), 783–792.

DeJoy, D. M. (1986, December). A behavioral –diagnostic tools for self-

protective behaviors in the work environment. Professional Safety,

26–30.

DeJoy, D. M. (1994). Managing safety in the workplace: an attribution

theory analysis and model. Journal of Safety Research, 25, 3–17.

DeJoy, D. M. (1996). Theoretical models of health behavior and workplace

self-protection. Journal of Safety Research, 27, 61–72.

Farkas, A. J., & Tetrick, L. E. (1989). A three-wave longitudinal analysis of

the causal ordering of satisfaction and commitment on turnover deci-

sions. Journal of Applied Psychology, 74, 855–868.

George, J. F. (1996). Computer-based monitoring: common perceptions

and empirical results. MIS Quarterly, 20, 459–480.

Hofmann, D. A., & Stetzer, A. (1996). A cross-level investigation of factors

influencing unsafe behaviors and accidents. Personnel Psychology, 49,

307–339.

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 Journal, 19, 644–657.

Holyoak, K. J. (1984). Mental models in problem solving. In J. R.

Anderson, S. M. Kosslyn (Ed.), Tutorials in learning and memory

( pp. 193–218). New York: W.H. Freeman.

Howard, L. W., & Foster, S. T. (1999). The influence of human resource

practices on empowerment and employee perceptions of management

commitment to quality. Journal of Quality Management, 4, 5–22.

Huber, V. L., & Brown, K. A. (1991). Human resource issues in cellular

manufacturing: a sociotechnical perspective. Journal of Operations

Management, 19, 138–159.

James, L. R., Mulaik, S., & Brett, J. M. (1982). Causal analysis: assump-

tions, models, and data. Beverly Hills, CA: Sage.

Kirkman, B. L., & Shapiro, D. L. (2000). Understanding why team mem-

bers won’t share. Small Group Research, 31, 175–209.

Klimoski, R., & Mohammed, S. (1994). Team mental models: construct or

metaphor? Journal of Management, 20, 403–437.

McElroy, J. C., & Morrow, P. C. (1995). Railroad employee commitment

and work-related attitudes and perceptions. Transportation Journal, 34,

13–24.

Mitchell, T. R., & Wood, R. E. (1980). Supervisors’ responses to subordi-

nate poor performance: a test of an attributional model. Organizational

Behavior and Human Performance, 25, 123–138.

Mottel, W., Long, J., & Morrison, D. (1995). Industrial safety is good

business: the Dupont story. New York: Van Nostrand Reinhold.

Netemeyer, R. G., Johnston, M. W., & Burton, S. (1990). Analysis of role

conflict and role ambiguity in a structural equations framework. Journal

of Applied Psychology, 75, 148–157.

Neter, J., Wasserman, W., & Kutner, M. (1989). Applied linear regression

models (2nd ed.). Homewood, IL: Irwin.

Niehoff, B. P., & Moorman, R. H. (1996). Exploring the relation-

ships between top management behaviors and employee perceptions

of fairness. International Journal of Public Administration, 19,

941–961.

Perrow, C. (1984). Normal accidents. New York: Basic Books.

Prussia, G. E., & Kinicki, A. J. (1996). A motivational investigation of

group effectiveness using social –cognitive theory. Journal of Applied

Psychology, 81, 187–198.

Rousse, W. B., & Morris, N. M. (1986). On looking into the black box:

prospects and limits in the search for mental models. Psychological

Bulletin, 100, 349–363.

Schmidt, F. (1973). Implications of a measurement problem for expectancy

theory research. Organizational Behavior and Human Performance, 10,

243–251.

Shadur, M. A. (1999). The relationship between organizational climate and

employee perceptions of involvement. Group and Organization Man-

agement, 24, 479–503.

Thompson, R. C., Hilton, T. F., & Witt, L. A. (1998). Where the safety

rubber meets the shop floor: a confirmatory model of management

influence on workplace safety. Journal of Safety Research, 29, 15–24.

Toulson, P., & Smith, M. (1994). The relationship between organizational

climate and employee perceptions of personnel management. Public

Personnel Management, 23, 453–468.

Trist, E., & Bamforth, K. W. (1951). Some social and psychological con-

sequences of the longwall method of coal-getting. Human Relations, 4,

6–24.

Vaughan, D. (1996). The challenger launch decision: risky technology,

culture, and deviance at NASA. Chicago: University of Chicago Press.

Vukmir, R. B. (Ed.) (1999). The mill. Lanham, MD: University Press of

America.

Wayne, S. J., & Ferris, G. R. (1990). Influence tactics, affect, and exchange

quality in supervisor– subordinate interactions: a laboratory experiment

and field study. Journal of Applied Psychology, 75, 487–499.

Weick, K. E., & Roberts, K. H. (1993). Collective mind in organizations:

heedful interrelating in flight decks. Administrative Science Quarterly,

38, 357–381.

Williams, L. J., & Hazer, J. T. (1986). Antecedents and consequences of

satisfaction and commitment in turnover models: a reanalysis using

latent variable structural equation methods. Journal of Applied Psychol-

ogy, 71, 219–231.

Williams, L. J., & Podsakoff, P. M. (1989). Longitudinal field methods for

studying reciprocal relationships in organizational behavior research:


toward improved causal analysis. Research in Organizational Behavior,

11, 247–292.

Zerbi, W. J., & Dobni, D. (1998). Promoting employee service behaviour:

the role of perceptions of human resource management. Canadian Jour-

nal of Administrative Sciences, 15, 165–179.

Zohar, D. (1980). Safety climate in industrial organizations: theoretical and

applied implications. Journal of Applied Psychology, 65, 96–101.

Gregory E. Prussia is an associate professor of management in the Albers

School of Business and Economics at Seattle University. He teaches

undergraduate and graduate teambuilding and leadership classes as well

as principles of management classes. He has a BA in economics and an

MBA from California State University, Chico, and a PhD in human

resource management from Arizona State University. His publications

appear in several journals including Academy of Management Journal,

Academy of Management Review, Journal of Applied Psychology, and

Journal of Operations Management. He is a member of the Academy of

Management, the American Psychological Association, and the Decision

Sciences Institute, and serves on the Editorial Board for the Academy of

Management Journal.

Karen Brown is a professor of operations and project management at the

University of Washington, Bothell. She also serves as a visiting professor at

the China Europe International Business School (CEIBS) in Shanghai,

China, and at IESE in Barcelona, Spain. Dr. Brown holds BS, MBA, and

PhD degrees from the University of Washington. She serves as an associate

editor of the Journal of Operations Management and is a vice president of

the Decision Sciences Institute. Her research focuses on sociotechnical

systems and has appeared in Journal of Operations Management, Academy

of Management Journal, Academy of Management Review, Human Rela-

tions, Journal of Applied Psychology, Business Horizons, and other

journals. Her recent work on workplace safety has grown out of her

previous work in the healthcare field.

Geoff Willis is an assistant professor of operations management in the

College of Business Administration at the University of Central Oklahoma.

He holds BS degrees in biomedical engineering and mathematics from

Vanderbilt University and MS and PhD degrees in production and oper-

ations management from Texas Tech University. Prof. Willis has published

articles in Journal of Operations Management, International Journal of

Production Research, Quality Engineering, and Journal of Clinical Engi-

neering. He has conducted research at Texas Instruments, Lubbock’s

University Medical Center, Seattle Public Schools, the City of Seattle,

and Seagate Technology on quality issues related to product development

and process improvement.

Documents

Mental Models of Safety - Do Managers and Employees See Eye to Eye