Making Things Happen: Reciprocal Relationships Between Work
Characteristics and Personal Initiative in a Four-Wave Longitudinal
Structural Equation Model
University of Giessen
University of Amsterdam
University of Giessen and Aston University
The authors used the frameworks of reciprocal determinism and occupational socialization to study the
effects of work characteristics (consisting of control and complexity of work) on personal initiative
(PI)—mediated by control orientation (a 2nd-order factor consisting of control aspiration, perceived
opportunity for control, and self-efficacy) and the reciprocal effects of PI on changes in work charac-
teristics. They applied structural equation modeling to a longitudinal study with 4 measurement waves
(N ? 268) in a transitional economy: East Germany. Results confirm the model plus 1 additional,
nonhypothesized effect. Work characteristics had a synchronous effect on PI via control orientation (full
mediation). There were also effects of control orientation and of PI on later changes in work character-
istics: As predicted, PI functioned as partial mediator, changing work characteristics in the long term
(reciprocal effect); unexpectedly, there was a 2nd reciprocal effect of an additional lagged partial
mediation of control orientation on later work characteristics.
Keywords: personal initiative, job characteristics, reciprocal relationship, transition economy, self-
An important question in philosophy and the social sciences has
been whether people are determined by their work (Marxism) or
whether people can actively shape their environment (cf. Schopen-
hauer’s, 1819/1998, primacy of the will). We use two concepts—
personal initiative (PI) and reciprocal determinism—to understand
and empirically look at this issue.
A great deal of theory and research within organizational be-
havior and industrial and organizational psychology suggests that
work characteristics influence individual attitudes and behaviors.
Within this literature, work characteristics are conceptualized and
studied as exogenous variables, determining, in turn, individuals’
adjustment to their work. People’s motivation is affected by work
characteristics (Hackman & Oldham, 1976), people are socialized
by the work characteristics (occupational socialization; Frese,
1982) and by management (organizational socialization; Van
Maanen, 1976), and they are trained to do the job tasks (Latham,
1989). Thus, work characteristics are conceptualized to be outside
the employees’ influence.
Countering this conceptualization is a growing literature exam-
ining the active side of people’s behaviors at work. For example,
Morrison (1993) argued that “socialization is a process affected
not only by organizational initiatives, but also by newcomer ini-
tiatives” (p. 173). Ashford and Tsui (1991) and Morrison (1993)
have studied concepts such as active feedback seeking. Further-
more, Hacker (1973), Ilgen and Hollenbeck (1991), and Staw and
Boettger (1990) have been concerned with employees’ task revi-
sions. In addition, Organ (1988) has developed the concept of
organizational citizenship behavior.
We think that this more active conceptualization of employees is
beneficial and appropriate but that it has not gone far enough.
Theorists and research by and large have not systematically ex-
amined the ways employees may actively change their work char-
acteristics (Wrzesniewski & Dutton, 2001). Thus, work character-
istics are often conceptualized as extraneous variables, even in the
studies and theories highlighted above. For example, active feed-
back seeking implies that people seek feedback to understand the
work characteristics and the organization better but not how to
change the work characteristics and the organization.
By further developing the concept of PI, we would like to
contribute to the understanding of how people can actively affect
Michael Frese, Department of Psychology, University of Giessen, Gies-
sen, Germany; Harry Garst, University of Amsterdam, Amsterdam, the
Netherlands; Doris Fay, Department of Psychology, University of Giessen,
and Aston Business School, Aston University, Birmingham, England.
Other members of the project team have been Sabine Hilligloh, Thomas
Wagner, Jeannette Zempel, and Christa Speier. The project was supported
by Deutsche Forschungsgemeinschaft Grant Fr 638/6-5 and by the Pro-
grammagroep for Work and Organizational Psychology, University of
For very helpful criticism, we thank Katherine Klein (who helped at
several crucial stages of writing this article), Frank Landy, David Hof-
mann, Elizabeth Morrison, Andreas Utsch, and Dieter Zapf.
Correspondence concerning this article should be addressed to Michael
Frese, Department of Psychology, University of Giessen, Otto-Behaghel-
Strasse 10F, D-35394, Giessen, Germany. E-mail: michael.frese@psychol
Journal of Applied Psychology
2007, Vol. 92, No. 4, 1084–1102
Copyright 2007 by the American Psychological Association
0021-9010/07/$12.00 DOI: 10.1037/0021-9010.92.4.1084
their work characteristics. People show PI when they engage in
self-starting and proactive behaviors that overcome barriers on the
way toward a goal (Frese, Kring, Soose, & Zempel, 1996). Given
the nature of work in the 21st century, PI is likely to become
increasingly important (Frese & Fay, 2001), because (a) compa-
nies are moving from stable structures to change-oriented organi-
zations (Lawler, 1992), (b) these changes bring new responsibili-
ties to rank-and-file workers (Wall, Cordery, & Clegg, 2002), (c)
people who just react to obvious situational cues or who only
follow orders will be unable to actively carry changes forward
(Frese & Fay, 2001), and (d) organizations are placing more
responsibility on the individual for career management, including
training and development (Hall, 1996; London & Mone, 1999).
Theoretically, the PI concept is needed to understand how people
can change the situation in which they work and how they deter-
mine changes in work, in processes, in products, and in society.
Even though we emphasize people’s active approaches as driv-
ers of change, we do not ignore that these active approaches are
themselves driven by other factors. Reciprocal determinism (Ban-
dura, 1997), which argues that “people are both producers and
products of social systems” (p. 6), integrates both lines of thought.
Despite its theoretical influence, to our knowledge, there has been
little systematic or longitudinal examination of this concept in
work settings (Vancouver, 1997). Thus, on the basis of a longitu-
dinal field study, we attempt to further develop the concept of
reciprocal determinism and to provide a more complete picture of
the development of PI as a result of work characteristics. Our
design is based on a six-wave longitudinal study in East Germany;
four of these waves are used to test our ideas.
Thus, we attempt to contribute to the literature by testing recip-
rocal determinism in the field and by introducing PI into this
model. To examine alternative models, we test our hypotheses
with data from a longitudinal study with four measurement points.
The study on PI was carried out in East Germany because a large
amount of change in workplaces occurred there after reunification,
and this makes it easier to look at reciprocal effects. In the
following, we introduce the concept of PI and then develop in
more detail the theoretical model underlying our study.
Core Concepts and Theoretical Model
Figure 1 displays our theoretical model. We assume (a) that
work characteristics change control orientation and (b) that there is
a reciprocal path from PI to changes in work characteristics. This
implies two mediation effects: Work characteristics should change
PI via the mediator control orientation, and control orientation
should lead to changes in work characteristics via the mediator PI.
This also means that the process is energized by three drivers—the
work characteristics, control orientation, and PI (see Figure 1).
The Concept of PI
PI refers to behaviors, mainly directed toward work and orga-
nizational issues, that are characterized by the following aspects
(Frese & Fay, 2001): self-starting, proactive, and persistent in
overcoming barriers. The opposite of PI is a reactive approach in
which one is told what to do, in which the “here and now”
determines the actions (no proactivity), and in which a person
gives up when barriers and difficulties arise (Hacker, 1992).
Self-starting implies that the behavior is regulated by goals
developed without external pressure, role requirements, instruc-
tion, or “obvious” action. Thus, PI is the pursuit of self-set goals,
in contrast to assigned goals. An example of a person with PI is a
blue-collar worker who attempts to fix a broken machine even
though this is not part of his or her job description. Frequently,
initiative deals with subproblems of an assigned task or with issues
that are not obviously related to the task. Blue-collar workers may
perform additional checks on the quality of material or of prior
work. For example, in one study, we observed that the task of
drilling a hole in an automobile could damage cables located
below the drilling surface. In such a case, the worker may think of
the danger of drilling too deeply and tell others about it. PI
sometimes implies that a person takes charge of an idea that has
been around for a while but that has not led to action before. A
secretary who buys bottled water for a guest speaker shows ini-
tiative in this sense, even if this is a small matter. Managers are
often required to show initiative. In this case, we can still speak of
self-starting if a manager does not just follow the example of many
other managers and use obvious “initiatives” that have been sug-
gested by several others in his or her area of interest but self-starts
an action that is not an obvious choice.
Proactivity means to have a long-term focus and not to wait
until one must respond to a demand. A long-term focus at work
enables the individual to consider things to come (new demands,
new or reoccurring problems, and emerging opportunities) and to
do something about them now. Thus, the person anticipates prob-
lems and opportunities and prepares to deal with the problems and
to take advantage of opportunities. An example is a secretary in a
university department who books travel tickets for her boss. Her
formal task is to phone the travel agency the university uses.
Perhaps she is not satisfied with the service, finds the discount
unattractive, and therefore decides to find out whether one can get
better deals somewhere else. In this case, the secretary acts in a
proactive manner because she anticipates having to take care of
travel arrangements in the future. This example also illustrates that
PI can lead to changes in the environment.
Persistence is frequently necessary; PI usually implies that a
process, a procedure, or a task is added or modified, and these
changes often involve setbacks and difficulties. For example,
people affected by the changes may not like having to adapt to
something new and being forced to abandon their routines. This
requires that the persons who are taking initiative display persis-
tence in overcoming barriers to get past technical problems and to
overcome other people’s resistance and inertia. Sometimes, per-
sistence also has to be shown toward supervisors who do not like
their subordinates going beyond the boundaries of their job.
Theoretically, the three aspects of PI—self-starting, proactive-
ness, and persistence—reinforce each other. A proactive stance is
associated with the development of self-started goals, because a
proactive orientation toward the future leads more probably to the
development of goals that go beyond what one is expected to do.
Self-started goals are related to being persistent in overcoming
barriers because of the changes inherent in their implementation.
Overcoming barriers can also contribute to self-starting goals,
because unusual solutions to overcome barriers often require a
self-start. Finally, self-starting implies that one looks at potential
future issues; therefore, there is a higher degree of proactivity, and
higher proactivity, in turn, is related to being more self-starting,
because one wants to exploit future opportunities that others do not
yet see. Thus, there is a tendency for these three aspects of PI to
co-occur (Frese, Fay, Hilburger, Leng, & Tag, 1997).
In principle, PI can be directed against the long-term interests of
the organization or against one’s own long-term interests (e.g., to
be self-starting in illegal substance abuse), but we conceptualize PI
to be aimed at producing, on average, long-term positive or at least
neutral outcomes for the individual and/or for the company. Re-
search has shown that PI is positively linked to important out-
comes. For example, prior individual PI has been shown to be
related to obtaining a new job after becoming unemployed (Frese
et al., 1997), PI is associated with entrepreneurial success in small
business owners (Fay & Frese, 2001) and with performance in
employees (Thompson, 2005), and organizational-level PI (as or-
ganizational climate) predicts increasing profitability of firms
(Baer & Frese, 2003).
The Effects of Work Characteristics on PI
We propose that two aspects of work characteristics—control and
complexity at work—influence PI (see Figure 1). Control at work
implies having an influence on the sequence, time frame, and content
of one’s work goals; on one’s work strategies; on feedback; and on
working conditions (Frese, 1989). Complexity has been defined by
the number of elements that need to be considered (Wood, 1986)—a
large number of elements implies that the work provides many op-
tions for decision making. Control and complexity at work are often
combined into one factor (e.g., Karasek & Theorell, 1990), because,
conceptually, both characteristics refer to decision possibilities. Con-
trol is trivial if exerted in a job with little complexity, because
decisions then refer to unimportant issues only. Empirical correlations
between control and complexity are high (e.g., in one study, r ? .42,
measured on the level of job incumbents, and r ? .70 for observers’
ratings; Semmer, 1982).
The notion that control and complexity are important work char-
acteristics follows from occupational socialization theory1
(Frese, 1982; Kohn & Schooler, 1978) and is empirically supported
(Spector, 1986). Control and complexity have been shown to be
related to ill health (Karasek & Theorell, 1990), intellectual flexibility
(Kohn & Schooler, 1978), and work motivation (Hackman & Old-
model of Hackman and Oldham (1975, 1976), as demonstrated by
their strongest relationship with the overall job motivation potential
(Hackman & Oldham, 1975; Wall, Clegg, & Jackson, 1978).
High levels of work characteristics (i.e., control and complexity)
should enhance PI, because these increase the sense of responsi-
bility for the whole job (Hackman & Oldham, 1976) and are
associated with a broader and more proactive role orientation
(Parker, Wall, & Jackson, 1997). This enhances PI by stimulating
proactive thinking, self-starting approaches, and efforts to over-
come barriers. High levels of work characteristics also contribute
to more knowledge of job-relevant dimensions (Holman & Wall,
2002). Knowing one’s job permits one to see opportunities for PI
and provides the skills to intervene. The success of autonomous
work groups depends on people developing an active approach to
work. Much of the job redesign performed to introduce autono-
mous work groups is therefore focused on increasing control and
complexity (Wall et al., 2002). We similarly suggest that work
characteristics affect PI; however, this relationship works via the
mediator control orientation (Figure 1).
The Mediating Role of Control Orientation
We define control orientation as a belief that one is in control of
relevant and important issues at work and that it pays off to have such
control. This is in agreement with other self-regulation concepts
(DeShon & Gillespie, 2005; Heckhausen & Schulz, 1995), which talk
about (a) the desire to exercise control at work (control aspiration;
Rothbaum, Weisz, & Snyder, 1982), (b) the expectation to have such
control (perceived opportunity for control; Rotter, 1972), and (c) the
confidence in one’s ability to exercise control effectively (self-
efficacy; Bandura, 1997). Thus, control orientation is composed of
control aspiration, opportunity for control, and self-efficacy. Control
orientation is conceptualized to function similarly to critical psycho-
logical states (Hackman & Oldham, 1976) that also mediate between
work characteristics and outcomes.
Work characteristics should have an effect on control orienta-
tion. In particular, control aspirations are reduced by lack of
control, as suggested by the helplessness model (Seligman, 1975).
Lack of options and thwarted control lead to helplessness, which
produces negative motivational consequences because the organ-
ism stops trying to control the environment when it does not expect
any positive outcomes (Heckhausen & Schulz, 1995; Rothbaum et
al., 1982; Seligman, 1975). Abramson, Seligman, and Teasdale
(1978) have shown that helplessness can be broadly generalized.
The helplessness effect appears even if there are short-lived op-
posite effects as well, such as the reactance effect: Wortman and
Brehm (1975) combined reactance and helplessness theories by
showing that in the short term, lack of control and options in-
creases aspiration for control, as reactance theory suggests. How-
ever, if attempts to increase control and options get thwarted over
a long period of time, learned helplessness develops—thus, in the
long run, reduced control aspirations result.
Perceived opportunity for control implies that the work envi-
ronment allows people to control certain outcomes and decisions
that lead to these outcomes. People tend to generalize from past
experiences; if they have high control and complexity at work,
they tend to predict that future relevant work characteristics will
also be controllable (Abramson et al., 1978; Rotter, 1972). Thus, a
construct of perceived opportunity for control in the work envi-
1Although there is a great deal of overlap, occupational socialization
can be distinguished from organizational socialization. There are three
interfaces between the organization and the individual: colleagues, man-
agers, and work characteristics (including rules and procedures). The latter
constitute the substance of occupational socialization (Frese, 1982), and
managers and colleagues are the locus of organizational socialization.
FRESE, GARST, AND FAY
Self-efficacy—the belief that one is able to perform a certain
action effectively—is central for Bandura’s (1997) concept of
reciprocal determinism. Self-efficacy increases as a result of high
control and complexity at work, because these characteristics
provide mastery experiences (Bandura, 1997). Mastery experi-
ences at work exist if one controls complex tasks—if a person is in
control of a noncomplex task, mastery is trivial, and therefore no
self-efficacy can develop (self-efficacy has only been measured in
areas in which the skill component is important; therefore, there is
an inherent implication that self-efficacy refers to mastery experi-
ences in cognitively complex or in emotionally difficult task
environments). On average, these mastery experiences at work
should be positive, because we assume that most companies do not
provide a high degree of control and complexity at work to
employees who are not able to produce desired results.
Control aspiration, perceived opportunity for control, and self-
efficacy at work have a common core and are therefore related to
each other empirically and theoretically—we call the common
core control orientation. All three variables are motivational, with
a coherent theme that refers to expectations of being in control
over relevant issues at work; this includes control aspiration,
because the expectation of noncontrol leads to a reduction of
aspirations to control (Seligman, 1975). The common idea among
people with high control orientation is that they are in control of
relevant and important issues in their work situation and that it
pays off to have such control. In contrast, people with a low
control orientation believe that they cannot master the relevant
parts of their work situation. This common core appears because
there is some redundancy between outcome control (perceived
opportunity for control) and action control (self-efficacy) and
between aspiration for control and the belief that one has control.
However, we do not discount that there are unique parts to each
one of these three constructs. Thus, self-efficacy, perceived op-
portunities for control, and control aspirations can produce unique
and important predictions. In this article, we concentrate, however,
on the common substrate of the three aspects of control orientation.
In our model, control orientation is a critical psychological state
(Hackman & Oldham, 1976) that should affect PI behavior. People
with high control orientation are likely to (a) persevere when prob-
lems arise and search for opportunities to take actions to ameliorate
problems (Bandura, 1997); (b) have higher hopes for success and
therefore take a long-term perspective in goal setting and planning,
which leads to more proactive approaches (Heckhausen & Schulz,
1995); and (c) actively search for information (Ashford & Tsui,
1991), which leads to better knowledge of where to show initiative.
of work characteristics (control and complexity) on active behavior
(Karasek & Theorell, 1990; Spector, 1986).
Reciprocal Influence: Effects of PI on Work
Thus, work characteristics affect PI via the mediator control orien-
tation. In keeping with the reciprocal model, we hypothesize, in
addition, that PI increases work characteristics—that is, enhances
control and complexity (see Figure 1). Two mechanisms are likely to
be influential: First, people with high PI may generate some added
complexity and control in their given job. The tasks of a job are not
completely fixed once and for all because of emergent elements in a
job (Ilgen & Hollenbeck, 1991), and role making can occur as a result
of supervisor–member interactions (Graen & Scandura, 1987). For
example, if a person develops initiatives to improve productivity, his
or her work characteristics are changed and control and complexity
are increased; superiors may give high-PI employees more responsi-
bilities, which translates into more complex and controllable work
likely to look for and make use of opportunities for getting more
challenging jobs and for increasing their career success (Seibert,
Kraimer, & Crant, 2001). People with higher PI should also be more
successful in finding those jobs because recruiters will more likely
hire such people for challenging jobs (Frese et al., 1997) that include
tasks with high control and complexity.
One of the few studies that looked at reciprocal influences between
work and person characteristics was Kohn and Schooler’s (1978)
10-year longitudinal study of the reciprocal effects of complexity of
intellectual flexibility had a long-term effect on complexity of work
and that complexity had a concurrent effect on intellectual flexibility.
Our theory builds on this but takes a different focus: We are interested
in the question of what drives the observed changes in work charac-
teristics. Intellectual flexibility per se does not change work charac-
teristics. We think that PI may be a missing link in Kohn and
Schooler’s model. Intellectual flexibility affects PI (Fay & Frese,
2001), and PI may change work characteristics.
Effects of Control Orientation on Work Characteristics
via the Mediator PI
The reciprocal influence of PI on work characteristics discussed
above also implies that PI is a mediator of the relationship between
control orientation and work characteristics (see Figure 1). Desir-
ing and expecting control increases PI, and this, in turn, causes
work characteristics to be higher (higher control and complexity).
If people expect control, if they aspire for control, and if they know
themselves to be competent, they influence their work character-
istics to suit them better (Wrzesniewski & Dutton, 2001) and
increase complexity and control at work.
Design and Setting of the Study
In spite of frequent calls for more longitudinal studies, they are
still the exception rather than the rule. We designed a longitudinal
study consisting of six waves2for the following reasons: (a) We
2This project produced nine separate studies (including a longitudinal
study with six waves, spanning 5 years) and had three objectives: first, to
provide a psychohistorical account of the changes in East Germany after
reunification; second, to examine stress and well-being; and third, to study
PI and its development. Prior publications on PI using data of the longi-
tudinal study have investigated the validity of PI (Fay & Frese, 2001; Frese
et al., 1997) and control aspiration measures (Frese et al., 1994), PI
differences between East and West Germany (Frese et al., 1996), relation-
ships between conservatism and PI (Fay & Frese, 2000), the function of
self-efficacy for the development of PI (Speier & Frese, 1997), and work
stressors and PI (Fay & Sonnentag, 2002). The current article constitutes
the major publication of our reciprocal determinism model (which, for
methodological reasons, does not include the first two waves).
are interested in causal effects, (b) more than three waves help
reduce identification problems in structural equation modeling
(Finkel, 1995), (c) they also allow replication of the effects over
time, and (d) such a longitudinal design makes it possible to test
reciprocal (and therefore complex) models. At this time, we are not
aware of any field studies on reciprocal determinism that meet
these methodological requirements. We restricted the analysis to
four waves (Time 3 [T3] to T6) because one of the relevant PI
variables—qualitative and quantitative initiative—was first intro-
duced at Wave 3.
Ideally, research on the effects of work characteristics should
have a natural zero point—for example, a given day when all
participants start a new job. The study was conducted in East
Germany, which had such a natural zero point in 1990 (the starting
date of East Germany’s transition from socialism to capitalism was
reunification with West Germany in October 1990). People expe-
rienced drastic changes at work: Nearly every company introduced
new technology and new organizational structures, and new man-
agement was often introduced as well. Layoffs were numerous,
and people had to find new jobs, whereas unemployment was
practically nonexistent before 1990. This situation of revolutionary
job change offers us an excellent situation for examining recipro-
cal effects. Thus, East Germany may be a good, albeit radical,
example of how global competition and technological and organi-
zational innovations change the nature of today’s jobs (Bridges,
1995). Likewise, this situation illustrates how—after the demise of
the traditional career in Western societies—people are required to
develop their career proactively (Hall, 1996).
We used a stratified random sample procedure to aim for a
representative sample of the working population of Dresden (a
large city in the southern part of East Germany with roughly
500,000 inhabitants). We drew a random lot to select grid squares
of a map of Dresden. For each square, we selected every second
street that crossed the left side of these grid squares. In each street,
we entered every third (apartment) house; if it was an apartment
house with six or fewer parties, we talked to every third party; if
it was an apartment house with more than six parties, we talked to
each fourth party. In each party, we asked those who were between
18 and 65 years old and who were employed for at least 19 hr per
week to participate in the study (there was practically no unem-
ployment at T1 in socialist East Germany). Confidentiality was
assured. We recontacted the sample five times, ultimately collect-
ing six waves of data between July 1990 and September 1995.
In Wave 1 (T1; July 1990), 463 people participated (a 67%
response rate for the interview). This sample was representative of
the Dresden population for the relevant parameters (tested against
census data, e.g., for age, social class, male/female percentage at
work). At Wave 2 (T2; November and December 1990, right after
reunification), we reinterviewed the participants from T1 and also
selected 202 additional people by using the same sampling proce-
dure as for T1. We added additional people at T2 to ascertain
whether repeated study participation had an influence on partici-
pants’ responses; finding no initiative difference between the re-
peaters and the first timers, we did not seek additional research
participants at subsequent waves. We call the resulting potential
sample at T2 the full sample, with N ? 665. Attrition of 8.9% of
the participants recruited at T1, however, led to an actual sample
size of 624 at T2.
As previously mentioned, our analyses are based on Waves 3 to
6: At Wave 3 (T3; September 1991), 543 individuals participated
(representing a response rate of 81.6% of the full sample); at Wave
4 (T4; September 1992), 506 participants responded (76.1% of the
full sample); at Wave 5 (T5; September 1993), 478 participants
responded (71.9% response rate); and at Wave 6 (T6; September
1995), a total of 489 responses were received (73.5% response
rate). (The sample size was higher at T6 than at T5 because we
made an extra effort to get responses from participants who had
moved away from Dresden.) Experimental mortality did not
change the makeup of the sample. There were no significant
differences in PI between those who dropped out from T1 to T3
and those who participated in each of the waves during this period.
As described below, it was necessary to select those who had a
job into our longitudinal structural models (n ? 268). The demo-
graphic characteristics of this sample were as follows: The mean
age of participants in 1991 (T3) was 39.1 years (SD ? 9.7), and
54% of the participants were male. With regard to the different
levels of school education that were distinguished in East Ger-
many, 61% had graduated from school after 8 or 10 years, and
37.5% had obtained the highest possible school degree (which was
the entry requirement for university). The remaining 1.5% had left
school without graduating or with an exceptional certificate. A
university degree was obtained by 30.1% of all participants.
The sample consisted of 31.4% unskilled, semiskilled, and
skilled blue-collar workers; 23.6% lower level white-collar work-
ers, such as lower professionals and administrative workers; and
42.0% higher professionals and managers. Thirty-nine percent of
the sample was employed in the public sector (e.g., hospitals,
education, and public administration); the remainder worked in the
manufacturing industry (22.6%); the building industry (7.2%); the
trade, hotel, and catering industry (6.0%); and other industries
(finance, utility, transportation, etc.). Forty percent had been em-
ployed by their organization for 3 years or less, 28.0% for 3 to 10
years, and 31.8% for 10 years or longer.
Treatment of Missing Cases
The economic changes in East Germany are conducive to study-
ing the implications of our model. However, they also produce a
greater number of true missing values as a result of frequent
periods of unemployment, sabbaticals, educational years, and so
on. For example, of the 471 participants who had a job at T2, 57
did not work at T3. Participants without work could not respond to
the work-related items. Therefore, we based our analyses only on
those participants who were always employed (or self-employed).
Our analyses were based on the four waves from T3 to T6, which
resulted in a sample of 268 participants. To estimate missing data
of the nonwork-related items, we estimated the covariance matri-
ces with the expectation-maximalization algorithm, using the com-
puter program NORMS (Schafer, 1997).
Treatment of Time
In general, the timing of effects due to working conditions is an
issue that is complicated and far from being resolved theoretically
or empirically (Mitchell & James, 2001). To our knowledge,
FRESE, GARST, AND FAY
theory development on the time frame in which the effects of
working conditions on orientation and, in turn, on behaviors unfold
is too embryonic to allow the development of theory-based hy-
potheses. Therefore, we did not develop an a priori hypothesis with
regard to timing of the effects of control and complexity; instead,
we explore models with different time lags.
Regarding the reciprocal path of the model—effects of PI on
work characteristics—previous research and theoretical thinking
indicate that the processes need a considerable amount of time to
unfold. It takes some time to change jobs and to change work
characteristics. Empirically, Kohn and Schooler (1978) found a
lagged selection effect, with a time lag of 10 years in the United
States. In a different area, Wilk, Desmarais, and Sackett (1995)
established that people in the United States gravitated to jobs
commensurate with their ability within a 5-year period. We there-
fore test whether PI at a given time affects working conditions 4
years later (this is the longest possible time lag in our analysis).
Even though our lag is somewhat shorter than what the cited
research suggests, effects might have unfolded in a slightly shorter
time period because of the high rate of change in East Germany
We used behavioral and structured interviews, self-report sur-
veys, and interviewer ratings to measure the constructs in our
model. The interviewers were psychology and business students in
master’s degree programs. Fifteen to 19 interviewers were in-
volved in each of the four waves reported in this article. They
received 2 days of training in interviewing and coding. The train-
ing consisted of a standard interviewer training (e.g., how to
approach participants, how to take protocols, professional issues,
how to ask questions) and training on the different areas of the
interview (e.g., on PI, activities of unemployed people). This
included observation and discussion of role-playing scenarios per-
formed by the trainers, role-play interviews with observation and
coaching by the trainers, and practice in protocol taking. Coding of
the transcribed information was practiced, with detailed descrip-
tions of the categories of the coding system; training was con-
cluded after successful calibration of raters. Nine interviewers
were involved in several waves; this allowed experienced inter-
viewers to supervise newly trained interviewers and to accompany
them on their first interviews.
Structured interviews were used to measure PI. Participants’
answers were written down by the interviewers in a short form that
was later typed and used as the basis for a numerical coding system
applied by the interviewer and by a second coder; the second coder
was drawn from the same pool of trained interviewers. The coding
system either was factual (e.g., participant is unemployed or
not—a dichotomous variable) or involved some kind of judgment
(e.g., the extent to which a certain answer constituted initiative on
a 5-point scale). Exemplar anchor points were provided for judg-
ment items. After the interview, the participants were given sur-
veys to complete (interviewers picked them up 1 or 2 weeks later).
The surveys included measures of work characteristics (control
and complexity) and of control aspiration, perceived opportunity
for control, and self-efficacy.
We tested the factor structure of the scales with longitudinal
confirmatory factor analyses to confirm measurement equivalence
and unidimensionality, first for the individual scales and then for
the second-order scales. All measures were in German. The mea-
sures (with information on their source), sample items, and—if
applicable—validity studies are presented in Table 1. An English
translation of the scales can be provided on request.
Interview Measures of PI
We measured PI with the structured interview and with the
interviewer evaluation. We used three measures—interviewer
evaluation, qualitative and quantitative initiative at work, and a
situational interview (Frese et al., 1996, 1997). Because they are
based either on behavior shown in the interview or on the inter-
viewers’ judgments, they constitute a separate source from the
questionnaire responses used for the independent and control
orientation variables (alphas and sample items are described in
Qualitative and quantitative initiative.
four questions about activities that can represent initiative at work
(i.e., whether a respondent had presented suggestions, talked to the
supervisor about a work problem, attempted to determine why
work problems existed, or changed a work procedure). The inter-
viewer probed into the nature of the activity reported to ensure its
self-starting and proactive nature (i.e., to make sure it was PI). On
the basis of the protocols, the activities that qualified as PI were
rated according to their level of quantitative initiative and quali-
tative initiative. Quantitative initiative reflects the degree to which
the activity required additional energy (e.g., working longer hours
to finish an important task although nobody required it), and
qualitative initiative relates to the degree to which the problem
addressed and the goal or strategy used went beyond what was
expected from a person in that particular job (e.g., a blue-collar
worker looking into a complicated production problem and sug-
gesting a general solution to it or dealing with a problem in such
a way that it would not appear again). Qualitative and quantitative
initiative were both rated on a 5-point scale (1 ? very little PI
shown; 5 ? very high PI shown). This resulted in eight items: four
qualitative initiative items based on the activities reported with
regard to the four questions asked, and four quantitative initiative
items. The respective qualitative and quantitative initiative items
that were based on the response to the same question (e.g., activ-
ities reported regarding suggestions presented) were highly related.
Therefore, the two parallel items were combined into a so-called
item parcel (Marsh, Hau, Balla, & Grayson, 1998). This resulted in
four item parcels. Interrater agreement values at T3 were .88, .83,
.85, and .91 for the four items.
This scale is based on the Situational
Interview (Latham & Saari, 1984) and consists of two subscales—
Overcoming Barriers and Active Approach. Overcoming Barriers
captures a participant’s initiative and persistence in overcoming
obstacles. Interviewers confronted the participants with four fic-
tional problem situations, both at and outside work (e.g., unem-
ployment compensation is reduced), and asked them what they
would do. After the participant suggested a way to deal with this
problem (representing the first barrier), the interviewer then pre-
sented a reason why this solution would not work out, thus creating
a new barrier. This procedure continued until the third barrier was
presented. Then the respondents were asked whether they could
think of additional solutions. These were written down and later
The interviewers asked
counted as if they had been replies to barriers. Each solution was
counted as one barrier overcome if the solution was, in principle,
feasible; was likely to have the desired effect; and did not present
a small variant of a previous solution. Each barrier was counted
without further weighting. We coded the number of barriers a
respondent had overcome in the following way: 1 ? no barrier
overcome, 2 ? one barrier overcome, 3 ? two barriers overcome,
and so on through 6 ? five or more barriers overcome. Interrater
agreement values at T3 for barriers overcome were .78, .82, .80,
and .81, and for the sum of the four items, r ? .86.
To avoid potential testing effects due to participants recalling
the problem situations, we changed the problem situations across
the waves. Different problems were used at T3, T4, and T5; T3
problems were repeated at T6. At T3 and T6, the problems were as
follows: your unemployment compensation is reduced; you are
thrown out of your apartment; your job is terminated; and you
want to take some classes for continuous education. At T4, the
following problems were used: in your apartment something needs
to be repaired but you cannot find a company to do that; you give
advice to a friend who is unable to find a preschool for his or her
child to attend; you want to start a firm and you need a loan; and
you give advice to somebody who wants to open a shop but has not
found a suitable location for it. At T5, the problems were as
follows: your machine breaks down; you are supposed to get
supplies from another department but you do not get them; you
make a suggestion for improving work to your supervisor but he or
she does not react; and a colleague always works sloppily.
The Active Approach subscale captures the degree of proactiv-
ity shown by the respondent in overcoming the barriers. The raters
coded the respondents’ answers to each problem situation on a
5-point scale as to the degree to which they delegated the problem
to someone else, such as the supervisor (1 ? active), or personally
strived to solve the problem (5 ? passive, reverse coded). Because
Overcoming Barriers and Active Approach were highly correlated,
the two parallel items were combined into four item parcels, which
were aggregated into the Situational Interview scale; the average
cross-sectional intercorrelation of Overcoming Barriers and Active
Approach was .52.
To use the interviewers as an addi-
tional source of information, we asked them to fill out a brief
questionnaire about the participant (interviewer evaluation) imme-
diately following each interview. The interviewers evaluated the
respondent’s initiative with three semantic differential scales with
the following end points: 1 ? s/he behaves actively to 5 ? s/he
behaves passively, 1 ? s/he is goal-oriented to 5 ? s/he gets easily
diverted from goal, and 1 ? s/he is motivated to act to 5 ? s/he
would rather not act (all reverse coded). Interviewers were trained
to use this measure. Because the interviewers knew the participants
well after interviewing them for about 70 min, their ratings are a
valuable additional source for evaluating the participants’ PI.
These ratings were designed to capture the interviewers’ subjective
perceptions of the participant during the whole interview. Hence,
interrater reliability could not be calculated for these ratings;
however, the test–retest correlations were appreciable even though
there were largely different interviewers across the waves (the
average of one-wave test–retest correlations was .51). The mean
intercorrelations of the three PI constructs were between .38
Description of Scales and Psychometric Properties
VariableS/I Sample item
No. items and
alphas for T3, T4,
I Rating on semantic differentials based on behaviors in entire interview: “behaves
actively. . .passively”; “goal-oriented. . .easily gets diverted from goal”
3: .88, .89, .87, .86 Frese et al. (1996,
Fay & Frese (2001)
PI: Qualitative and
I On the basis of reports about four areas at work (e.g., had respondent presented
improvement suggestion? talked to the supervisor about a work problem?),
interviewers rated the degree of quantitative initiative (effort required) and
qualitative initiative (degree to which goal or strategy went beyond what was
expected in a particular job).
Overcoming Barriers: Rating of persistence in dealing with four fictional
problem situations (e.g., a colleague always did his or her work sloppily);
Active Approach: Ratings on proactivity shown in dealing with each of the
problems. (The two parallel ratings were always combined into one parcel.)
“Can you determine how you do your work?”
8: .76, .78, .84, .75 Frese et al. (1996,
Fay & Frese (2001)
Control at work
I 4: .77, .81, .81, .82 Frese et al. (1996,
Fay & Frese (2001)
S 3: .77, .82, .81, .83 Frese et al. (1996),
S “Do you receive tasks that are extraordinary and particularly difficult?” 4: .78, .80, .73, .77 Frese et al. (1996),
Control aspirationsS “I would rather be told exactly what I have to do. Then I make fewer mistakes.”
Perceived influence on work conditions, climate, and work council decisions:
“Personally, my chances of influencing things at the work place are. . .”
“Together with others, my chances of influencing. . .”
“I judge my abilities to be high”
7: .87, .88, .88, .90 Frese et al. (1994)
S6: .76, .75, .71, .74 Frese (2003)
S6: .72, .67, .76, .70 Speier & Frese
A 5-point response format was used throughout. S ? survey; I ? Interview; T ? time; PI ? personal initiative.
FRESE, GARST, AND FAY
Previous studies have provided evidence for the convergent and
discriminant validity of the interview-based PI measure. It con-
verges with a questionnaire-based rating of PI provided by one’s
life partner (Frese et al., 1997) and by fellow students (Fay &
Frese, 2001), and it is different from organizational citizenship
behavior (Fay, 1998). The nomological net of PI implies that PI
requires abilities and skills and is motivated by person variables
and environmental factors (Fay & Frese, 2001; Frese & Fay,
2001). For example, PI is related to general mental ability and job
qualification (Fay & Frese, 2001). PI is also motivated by change
orientations; individuals who show PI should be open to changes
and ready to bear the uncertainty associated with them, because PI
implies that one brings about changes. Accordingly, PI is posi-
tively related to openness to change (Fay & Frese, 2001) and
negatively related to psychological conservatism, which is work-
ing against change (Fay & Frese, 2000). Stressful working condi-
tions require change; we found work stressors to spur PI (Fay &
Sonnentag, 2001). The nomological net also implies that PI in-
volves behaviors that benefit the individuals who show it and the
environment they are working in. Higher levels of PI are associ-
ated with finding a job faster when one becomes unemployed
(Frese et al., 1997) and with students’ better grades (Fay & Frese,
2001). Several studies on small-scale businesses showed that the
owners’ PI was related to their company’s success (an overview is
given in Fay & Frese, 2001) and survival (Zempel, 1999).
The three PI measures (interviewer evaluation, qualitative and
quantitative initiative at work, and the Situational Interview) were
included in a second-order construct, because a second-order con-
struct captures the essence of what defines PI behaviors (i.e.,
self-starting, proactive, persistence) and is methodologically well
balanced, as the first-order constructs were based on different
methods. The data suggested this to be an acceptable approach
because the first-order constructs were well correlated (cross-
sectional intercorrelations, on average, were .41) and the second-
order construct model had a good fit with the data (as shown later).
Unless otherwise stated, survey scales used a 5-point response
format ranging from 1 (not at all true ) to 5 (completely true),
which has been shown to be equidistant (Rohrmann, 1978). The
scale values were divided by the number of items (the scale means,
standard deviations, and alphas are presented in Table 2).
Work characteristics: Control and complexity at work.
measure control and complexity at work, we used two well-
validated German scales (Semmer, 1982; Zapf, 1993; also reported
in Frese et al., 1996). Complexity and control can be measured
well by surveys because both variables show strong relationships
between job incumbents’ self-reports and other people’s judg-
ments (Spector, 1992). We combined control and complexity into
a second-order model for theoretical reasons discussed in the
introduction. We modeled work characteristics with a causal indi-
cator model. This is in contrast to the more frequently used effect
indicator model. The effect indicator model assumes that each item
is an indicator of the underlying construct. Thus, a latent common
construct determines the observed variables, which means that a
change in one issue of control—for example, control over timing
of rest periods—is related to an equivalent change of another issue
of control—for example, control over selecting one’s work meth-
ods. This effect indicator model has been criticized, for instance,
by Cohen, Cohen, Teresi, Marchi, and Velez (1990), who argued
that in cases such as ours, one should not develop a latent construct
to determine the observed variables. An alternative is to conceive
of the items of the work characteristics measures control and
complexity as the causes; thus, the construct is a compound of the
items (Bollen & Lennox, 1991; Edwards & Bagozzi, 2000). In this
case, work characteristics are composite variables plus a distur-
bance term (MacCallum & Browne, 1993). In such a causal
indicator model, a change in one variable is not necessarily ac-
companied by an equivalent change in the other ones. The latent
variable is then only an abstraction of control in the sense that each
specific instance of control, when they are added together, leads to
overall greater control at work. Therefore, the work characteristics
variables were not fitted with a confirmatory factor analysis (Bol-
len & Lennox, 1991; MacCallum & Browne, 1993; Spector & Jex,
1998, p. 357).
However, specifying the work characteristics items as causal
indicators led to identification problems in our models. A condi-
tion for identification of a model including causal indicators is that
each composite variable has at least two emitting paths to other
constructs, which are mutually independent (MacCallum &
Browne, 1993). Thus, from each composite variable, two paths
should go outward to variables, and these two variables should be
independent of each other. If they are interrelated, the model is not
identified. Because identification problems prevented us from
weighting the work characteristic items individually, we used an
equally weighted summation of the two variables control and
complexity (cf. McDonald, 1996). This procedure helped to reduce
the number of variables in the model and, thus, to keep an adequate
ratio of sample size to the number of estimated parameters (Bentler
& Chou, 1987; Jackson, 2003).
Control orientation (control aspiration, perceived opportunity
for control, and self-efficacy).
three established measures. We used a seven-item scale to measure
control aspiration (Frese, 1984). Previous research showed that
attitudes toward job control are best assessed when they include
the potential negative consequences of control (e.g., “I would
rather be told exactly what I have to do; then I make fewer
mistakes”; Frese, 1984). For the purpose of naming and scoring all
mediators in the same direction, we reversed the original scoring
and called it control aspiration. Prior validity studies (Frese, Erbe-
Heinbokel, Grefe, Rybowiak, & Weike, 1994) showed that this
scale was related to wanting control and accepting responsibilities.
People with a low degree of control aspiration also had negative
attitudes toward errors, evaded complex work, did not like
changes, and were bitter about changes at work.
The perceived opportunity for control scale has been developed
in prior studies, starting with qualitative studies, several pilot
studies (with up to 100 participants), and then two cross-sectional
and two longitudinal studies (Frese, 2003), and is used in Germany
(e.g., by Bu ¨ssing, 1999). The measure consists of six items. We
assessed both perceived individual and collective opportunities for
control because many facets of work (e.g., climate in the group)
can only be influenced through cooperation with others. Respon-
dents were asked to rate the level of their influence in three target
areas twice: first their influence as an individual, and second in
cooperation with colleagues. The items were as follows: “As an
individual, my level of influence (1) on things at my work place in
Control orientation consists of
general is . . .,” “. . . (2) on the climate in my department is . . .,”
“. . . (3) on decisions made by the work council is . . . .” (Work
councils are mandated by law in Germany.) Then respondents
rated the three target areas again, with respect to levels of influence
with others: “In collaboration with my colleagues, my level of
influence on . . . .” We used a four-scale answer format that was
pretested and found to produce adequate variance: very little, little,
middle, rather high.
In contrast to control at work, which relates directly to how one
does the work itself, perceived opportunity for control asks for a
more generalized appraisal of control over the work environment.
It is therefore correlated with control at work (average of cross-
sectional correlations of perceived opportunity for control with
control at work was .36; see Table 2) and with complexity (average
of cross-sectional correlations with complexity at work was .28).
We assessed self-efficacy at work with a six-
item scale (Speier & Frese, 1997). Example items are “When I am
confronted with a new task, I am often afraid of not being able to
handle it” (reverse coded) and “If I want to achieve something, I
can overcome setbacks without giving up my goal.” The scale
correlated .53 with generalized self-efficacy (a scale developed by
Schwarzer, Baessler, Kwiatek, Schroeder, & Zhang, 1997), with
work-related self-esteem (r ? .52), and with optimism (r ? .38; in
all cases, p ? .01; cf. Speier & Frese, 1997). We modeled control
aspiration, perceived opportunity for control, and self-efficacy as
one latent variable; the appropriateness of this procedure was
tested with confirmatory factor analysis (see the next section).
Confirmatory Factor Analysis
We used confirmatory factor analyses to test for measurement
equivalence of our scales across time and for unidimensionality.
Table 3 provides the fit indexes of the longitudinal LISREL
measurement models, tested separately for free loadings and with
the loadings restricted to equal factor loadings over time.3All of
the fit indexes of the first-order factor models were very good,
indicated by root-mean-square error of approximation (RMSEA)
values lower than .06 and comparative fit index (CFI) values
higher than .95. There were no significant differences on the
chi-square tests between free and equal factor loadings for the
first-order control orientation variables: perceived opportunity for
control (after two free loadings were allowed), self-efficacy, and
control aspiration. Furthermore, the Akaike (1987) information
criterion (AIC) values for the more restricted and thus more
parsimonious equal factor loading models were lower. This means
that the factor structure was equal across time, and we can there-
fore assume measurement invariance across time. Control orien-
tation consisted of perceived opportunity for control, self-efficacy,
3The first-order factor models were based on five measurement waves
(T2–T6), except for qualitiative and quantitative initiative, which was
added to the study at T3 and is therefore only available from T3 to T6. The
sample sizes for the models were different (see Table 3) because work-
related measures were only collected from people who were employed at
Means, Standard Deviations, and Correlations for Survey Measures
123456789 1011 12 131415 1617 181920 21 2223 24
1. Control at work T3
2. Complexity at work
3. POC T3
4. Self-efficacy T3
5. Control aspiration T3 3.93 0.64 .42 .29 .24 .30 .87
6. Personal initiative T3 2.85 0.44 .30 .35 .28 .24 .36 —
7. Control at work T43.60 0.84 .55 .23 .25 .17 .37 .22 .82
8. Complexity at work
T43.50 0.76 .31 .56 .20 .25 .28 .38 .43 .72
9. POC T4 2.83 0.58 .23 .25 .55 .24 .22 .26 .29 .30 .75
10. Self-efficacy T43.51 0.48 .18 .19 .27 .73 .36 .22 .30 .24 .34 .67
11. Control aspiration T4 3.93 0.67 .36 .24 .33 .26 .67 .38 .50 .31 .30 .36 .88
12. Personal initiative T4 2.84 0.49 .29 .31 .27 .20 .31 .72 .33 .41 .32 .28 .42 —
13. Control at work T5 3.57 0.83 .49 .29 .26 .23 .38 .38 .68 .40 .29 .30 .42 .45 .81
14. Complexity at work
T5 3.51 0.70 .23 .52 .20 .16 .26 .29 .27 .66 .26 .22 .28 .30 .35 .65
15. POC T5 2.84 0.57 .29 .21 .50 .26 .25 .31 .28 .25 .59 .24 .28 .28 .42 .22 .71
16. Self-efficacy T53.50 0.55 .26 .25 .26 .64 .39 .30 .24 .32 .29 .75 .34 .31 .36 .26 .32 .76
17. Control aspiration T5 3.97 0.65 .34 .25 .27 .26 .68 .37 .43 .28 .30 .40 .75 .41 .47 .33 .29 .43 .88
18. Personal initiative T5 2.39 0.39 .25 .28 .31 .13 .36 .78 .28 .38 .32 .21 .44 .69 .41 .35 .33 .28 .46 —
19. Control at work T63.64 0.88 .45 .32 .23 .20 .37 .38 .47 .33 .24 .26 .30 .36 .60 .30 .40 .30 .38 .35 .83
20. Complexity at work
T63.55 0.74 .22 .48 .20 .15 .28 .37 .19 .50 .21 .18 .25 .37 .25 .59 .30 .22 .29 .38 .45 .69
21. POC T62.87 0.57 .25 .23 .53 .31 .25 .29 .28 .27 .55 .33 .29 .29 .31 .25 .59 .34 .28 .32 .39 .34 .74
22. Self-efficacy T63.53 0.51 .15 .15 .20 .66 .25 .19 .24 .29 .24 .75 .25 .25 .29 .24 .21 .71 .32 .19 .29 .19 .31 .70
23. Control aspiration T6 4.01 0.70 .29 .26 .24 .21 .67 .38 .43 .39 .34 .34 .71 .38 .43 .34 .31 .36 .74 .49 .45 .37 .37 .35 .90
24. Personal initiative T6 2.45 0.44 .31 .32 .28 .18 .34 .80 .27 .42 .29 .18 .38 .67 .41 .34 .35 .31 .38 .79 .42 .41 .34 .24 .47 —
3.58 0.80 .77
3.49 0.70 .44 .66
2.80 0.57 .35 .25 .76
3.47 0.51 .15 .16 .30 .72
Note. N ? 286. All correlations are significant at p ? .05; correlations above .16 are significant at p ? .01. Cronbach’s alphas are presented on the
diagonal. Personal initiative ? aggregated raw score. T ? Time; POC ? perceived opportunity for control.
FRESE, GARST, AND FAY
and control aspiration, with all three showing similar loadings
(standardized loadings from .43 to .66).
Measurement equivalence testing was more difficult for the
three PI constructs. The Situational Interview asked different ques-
tions at different times (therefore we cannot assume complete
measurement invariance), and there was only one instance of
interview questions being repeated (the same items were used at
T3 and T6). Insofar as we used the same items, the results suggest
that there was measurement equivalence (see Table 3). For the
nonrepeated items, the factor loadings were different. For qualita-
tive and quantitative initiative, a model with equal factor loadings
yielded a lower AIC value, but the chi-square difference test was
not significant at our criterion of p ? .01. Thus, we can assume
measurement equivalence as well. For the interviewer evaluation
of PI, the equal loadings model had a worse fit than the free
loading model (significant difference). This is not surprising, given
that the interviewer evaluation was based on the interviewers’
interpretations and that different interviewers were used at differ-
ent waves. However, a partial measurement invariance found in
these data in a longitudinal study is sufficient (Byrne, Shavelson,
& Muthe ´n, 1989; Pentzt & Chou, 1994).
Next, for all the first-order constructs, the summated scores
were calculated and used as indicators for the second-order longi-
tudinal factor models for control orientation and PI. These models
fitted well, with CFI values higher than .96 and RMSEA values
lower than .06. Models with equal factor loadings did not fit
significantly worse, which produces evidence for measurement
invariance. Thus, for both PI and control orientation, the second-
order models were well supported by the data.
Although our theoretical model is very straightforward, we had
an enormously complex array of potentially analyzable models,
Goodness-of-Fit Measures of LISREL Longitudinal Measurement Models
First-order longitudinal factor models
Perceived opportunity for control
Factor loadings free
Equal factor loadings
Two factor loadings free
Factor loadings free
Equal factor loadings
Factor loadings free
Equal factor loadings
Situational Interview (PI)
Factor loadings free
equal T2 ? T5 T3 ? T6
Qualitative and quantitative initiative
Factor loadings free
Equal factor loadings
Interviewer evaluation (PI)
Factor loadings free
Equal factor loadings
56 .042 208.21.989247
Second-order longitudinal factor models
Factor loadings free
Equal factor loadings
Factor loadings free
Equal factor loadings
T ? Time.
*p ? .01.
RMSEA ? root-mean-square error of approximation; AIC ? Akaike information criterion; CFI ? comparative fit index; PI ? personal initiative;
with four different measurement points, two levels of variables
(first-order constructs, second-order constructs), and several dif-
ferent causal time lags. Therefore, we made certain decisions to
reduce the number of potential models. As pointed out earlier, we
had no a priori hypotheses about the time frame in which the
effects of working conditions on control and, in turn, on PI
develop. We therefore tested different models with synchronous
and lagged effects (see Figure 2, Models I-A to I-D). In contrast,
there is research suggesting that it takes several years for the effect
of PI on working conditions to unfold (see Figure 2, Model
II-A-R). In the following, we describe the models in more detail.
The baseline stability model assumes that there are no rela-
tionships among the variables except stabilities. It is used as a
baseline model to test further structural causal models. The next
models are all socialization models with substantive paths be-
tween the constructs. The fully synchronous socialization model
(Model I-A) is a longitudinal model in which work character-
istics have an impact on the mediating latent construct control
orientation, which, in turn, affects PI. It is fully synchronous
because all the causal paths are assumed to work concurrently.
In this model and in the following models, the previous values
of the dependent variables are controlled, so that we predict
residual changes (Finkel, 1995).
Next, models with a mixture of lagged and synchronous effects
are fitted. The mixed synchronous–lagged socialization model
(Model I-B) tests a lagged effect of work characteristics on control
orientation and a synchronous effect of control orientation on PI.
The mixed lagged–synchronous socialization model (Model I-C)
interchanges the synchronous and lagged effects. The fully lagged
socialization model (Model I-D) specifies 1-year time lags from
work characteristics to control orientation and from control orien-
tation to PI (an exception is T5–T6, which represents a 2-year time
We then tested a mediation model called the socialization plus
direct effects of work characteristics model (Model II-A-M1). It
has a direct path added from work characteristics to PI and there-
fore examines whether control orientation is a full mediator in this
relationship. If this model fits significantly better than the best
Level I model, then control orientation is not a full but, at best, a
displays control orientation, and the bottom shows work characteristics. The circles in each model indicate, from
left to right, Time 3 to Time 6.
Different structural models. The first row of each model shows personal initiative (PI), the middle
FRESE, GARST, AND FAY
We then tested a reciprocal model—the socialization plus re-
ciprocal PI effect model (Model II-A-R), which tests the lagged
reciprocal effect of PI on work characteristics. We hypothesized
that PI had a slow effect on work characteristics. Therefore, we
calculated a model with a 4-year lag (note that there was a 2-year
lag between T5 and T6). Finally, we tested a mediation effect by
forcing the effects of work characteristics on control orientation to
be zero—the nonsocialization model (Model II-A-R-M2).
Statistical Analysis Method
All the models were tested with LISREL (Versions 8.54 and
8.72), via the two-step approach of Anderson and Gerbing (1988),
with a measurement model fitted first. Our models are complex not
only because they are longitudinal but also because they test for
mediation. The use of structural equation modeling provides re-
searchers with a good strategy to test for mediation (Brown, 1997)
because it uses a simultaneous estimate of the complete model and
deals with measurement error and nonrecursive parts of the model
as well. Model fit was assessed with RMSEA, CFI, chi-square
difference test for comparing nested models, and the AIC to
compare nonnested models (Hu & Bentler, 1999). RMSEA values
lower than .06 indicate good model fit, and CFI values higher than
.95 are desirable (Hu & Bentler, 1999).
Table 2 displays the intercorrelations, means, and standard
deviations of the observed variables. There was little change over
time in the means for control and complexity at work (work
characteristics) as well as for control aspiration, perceived oppor-
tunity for control, and self-efficacy (control orientation), whereas
there was a slight decrease in PI means over time; the PI standard
deviations were rather stable. Stabilities tended to be moderately
high for work characteristics (one-wave stabilities were between
.55 and .68; i.e., people tended to stay in the same type of job) and
for perceived opportunity for control (from .55 to .59); they were
higher for self-efficacy (.71 to .75), control aspiration (.67 to .75),
and PI (.69 to .79). Table 2 shows that all prerequisites for
mediation effects were met for all waves (Baron & Kenny, 1986).
There were sizable intercorrelations among work characteristics;
the mediator variables control aspiration, perceived opportunity for
control, and self-efficacy (control orientation); and PI.
Table 4 displays the fit indexes for the structural models. The
maximum model imposes (in contrast to all models depicted in
Figure 2) no constraints on the relationships among the latent
variables. It therefore fitted the data very well and can be used as
a best fit comparison model. The baseline model did not fit very
well in comparison with the maximum model. The fit of the
baseline model improved clearly when we allowed autoregressive
paths from T3 PI to T5 and T6 PI. This may indicate that there are
some state fluctuations, so that not only the immediately preceding
PI score was predictive of later PI but also the T3 PI score (Kenny
& Campbell, 1989). This is not surprising in a historically volatile
situation such as the one in East Germany, in which T3 was the last
year of some stability. The T4 score of PI could be more strongly
influenced by the profound changes in comparison with later
waves; hence, in later waves, people showed their typical behavior
pattern (as presented in T3) to a greater extent.
The modified baseline stability model’s fit indexes improved
when the hypothesized substantial paths among the constructs
were specified. All of the Level I (socialization) models had
adequate fit indexes, and all but one were significantly better than
the modified baseline model (see the chi-square difference tests in
Table 4). Models that differ only in time lags but otherwise
hypothesize identical structural relationships very rarely show
substantial fit differences. Given this, the fully synchronous so-
cialization model (Model I-A) appears to be the best because it
consistently showed the highest fit indexes; furthermore, AIC—the
best indicator for comparing nonnested models—showed the clear-
est differences from the other Level I models. Model I-A is a full
mediation model: Control orientation completely mediated the
effects of work characteristics on PI. Therefore, we performed a
mediation test by specifying a model that also allowed a direct path
from work characteristics to PI—the socialization plus direct ef-
fects of work characteristics model (Model II-A-M1). This model
was not significantly better than the fully synchronous socializa-
tion model (Model I-A), a finding that suggests the more parsi-
monious fully synchronous socialization model (Model I-A) as the
better model (Bollen, 1989).
Using Model I-A as a starting point, we tested the reciprocal
model, the socialization plus reciprocal PI effect model (Model
II-A-R). This model had adequate absolute goodness-of-fit in-
dexes, but the modification indexes indicated that there were
additional lagged paths from control orientation to work charac-
teristics. Therefore, we added an additional model: the socializa-
tion plus reciprocal PI and control orientation effects model
(Model II-A-R2; see Figure 3), which tested whether there were
lagged paths from control orientation to work characteristics. This
model had good fit indexes, and it was also significantly better
than Model I-A, the fully synchronous socialization model, ??2(4,
N ? 268) ? 58.51, p ? .000. It was significantly better than Model
II-A-R, ??2(3, N ? 268) ? 44.04 p ? .000. Moreover, this model
had an AIC fit that was even better than that for the maximum
model; thus, its fit to the data was excellent. The long-term
reciprocal effect of PI—covering a span of 4 years—was signifi-
cant (all models with shorter time lags had worse fit indexes). The
effect of prior work characteristics on later work characteristics
appeared because of the stability between the two waves of work
characteristics but also because of the mediation via control ori-
entation and the lagged effects of PI on work characteristics. To
examine whether partial mediation existed, we tested the media-
tion effect by forcing the effects of work characteristics on control
orientation to be zero—the nonsocialization model (Model II-A-
R-M2). The nonsocialization model was significantly worse than
the mediating model, socialization plus reciprocal effects of con-
trol orientation (Model II-A-R2; see Table 4), thus confirming a
The Best Fitting Structural Model: Socialization Plus
Reciprocal PI and Control Orientation Effects Model
The socialization plus reciprocal PI and control orientation
effects model (Model II-A-R2), shown in Figure 3, demonstrates
that the hypothesized paths were significant and that they were
regular across time. Work characteristics had significant effects on
control orientation in each case (standardized path coefficients of
.18 and above), as suggested by our model. Furthermore, the
effects of control orientation on PI were significant in all three
cases, with betas between .21 and .34. There was one long-term
significant reciprocal effect of PI on work characteristics, with a
path of .18. This effect size was similar to the work socialization
effects (the latter paths were around .22). Finally, there were
additional, nonexpected, sizable reciprocal 1-year time-lagged
paths from control orientation on work characteristics (.33 and
above), suggesting an effect of control orientation on changes in
The stability of work characteristics between T3 and T4 was
lower than the stability between T4 and T5. This coincides well
with the informal observations that workplace changes were most
dramatic in the 2nd year after German reunification (between T3
and T4) and then leveled off 2 years later. The stability between T5
and T6 was also lower than the one between T4 and T5, which is
due to the time lag of 2 years (in contrast to all other time lags of
Our results for the reciprocal PI effects on work characteristics
show the hypothesized long-term effect. This is not surprising,
because the effects of rare behaviors such as PI do not play out
quickly. Moreover, it takes some time for employees to convince
peers and supervisors around them that their initiatives are worth
pursuing and that they should get a greater degree of control and
complexity (or to change to a job with more control and complex-
ity). On an exploratory basis, we also modeled shorter term effects
of 1 and 2 years; they were, however, not significant. This sug-
gested a test of the whole model from a long-term perspective. We
therefore calculated the socialization plus reciprocal PI and con-
trol orientation effects—long-term model (Model III-A-R2–long-
term; see Table 4), a model with only T3 and T6 data, to look at
the effects as they unfolded over the long term (4 years in our
study). As Table 4 shows, this model had very good fit indexes.
Figure 4 shows that in the long term, the effect of control orien-
tation on work characteristics (.31) became more similar to the
effect of PI on work characteristics (.20) than was the case in the
short term (Figure 3). Moreover, the stabilities were, of course,
reduced when we observed paths in the long term, and the sub-
stantive paths increased in size. PI had a stability of .60, control
orientation had a stability of .50, and work characteristics had a
relatively low stability of .24. Apparently, there was quite a lot of
change in work characteristics during the 4 years of our study,
which were, to a large extent, determined by control orientation
and PI. The path from work characteristics to control orientation
was substantial (.41), as was the path from control orientation to PI
The reciprocal effects found here imply that people with high
control orientation and high initiative will eventually move to
more responsible jobs with more control and complexity or create
these kinds of jobs for themselves by changing the job content.
This finding speaks for reciprocal determinism, in which both
socialization effects and effects of PI and control orientation on
work characteristics can be observed.
Descriptive and Qualitative Results on the Long-Term
Effect of PI
Some descriptive results and qualitative impressions may help
to interpret the effects of PI on work characteristics. For this, we
Goodness-of-Fit Measures for Structural Models
Baseline stability model
Difference: Baseline stability model and maximum model
Modified baseline stability model
Fully synchronous socialization
Difference: Modified baseline stability model and I-A
Mixed synchronous-lagged socialization
Difference: Modified baseline stability model and I-B
Mixed lagged-synchronous socialization
Difference: Modified baseline stability model and I-C
Fully lagged socialization
Difference: Modified baseline stability model and I-D
Mediation test: Socialization plus direct effects of work
Difference: I-A and II-A-MI
Socialization plus reciprocal PI effect model
Difference: I-A and II-A-R
Socialization plus reciprocal PI and control orientation
effects model (see Figure 3)
Difference: I-A-R and II-A-R2
Difference: I-A and II-A-R2
Mediation test: Nonsocialization model
Difference: II-A-R-M2 and II-A-R2
Socialization plus reciprocal PI and control orientation
effects model—long term (T3–T6) (see Figure 4)
I-B 357 .053 723.70.968
I-D358 .054735.53 .966
II-A-MI 353.048 678.58.974
II-A-R2 352.042 627.05 .982
II-A-R-M2359 .052718.43 .967
Note. N ? 268 for all models. RMSEA ? root-mean-square error of approximation; AIC ? Akaike information criterion; CFI ? comparative fit index;
Difference ? chi-square difference test; I ? socialization models with various time lags; II ? best Model I plus other effects; III ? II-A-R2 as long-term
model (Time 3-Time 6); PI ? personal initiative.
FRESE, GARST, AND FAY
differentiated four extreme groups (10–12 participants each) using
data from T3 and T6: groups showing (a) high–high or (b) low–
low PI at both time periods, one group with (c) a substantial
decrease (high–low) in PI over time, and one group with (d) a
substantial increase (low–high). Using residualized scores of work
characteristics at T6 (with T3 work characteristics held constant)
illustrates the finding from the structural equation analysis that PI
helped to change work characteristics. The group that had always
been low in PI decreased dramatically in work characteristics over
time (M ? ?.55 residualized scores), whereas the group that had
high scores on PI both at T3 and at T6 increased in work charac-
teristics (M ? .33); the downward PI group (M ? .13) and the
upward PI group (M ? .10) were in the middle, F(3, 42) ? 3.75,
p ? .018.
Examples based on the interviews with the participants further
illustrate the relevance of the reciprocal model for PI. Both the
group members with low PI and those with high PI at both
measurement waves did not tend to change their company. How,
then, did the members of the high–high PI group increase their
control and complexity? It appears that this group took initiative in
skill enhancement—individuals used and even created learning
opportunities whenever they could. For example, one supervisor of
an operations planning group started learning English, although it
meant that he had to do that on the weekend. He did not have an
immediate use for the language but thought that in the future he
might need it (in East Germany, high school students did not learn
English but instead learned Russian). In the long run, this skill
enabled him to get involved in tasks of higher control and com-
plexity. In contrast, the always-low PI group was not interested in
continuing education. A security guard for the city said, “I would
go to some course if I were sent.” With skills becoming outdated,
loss in control and complexity in this group was a result of being
assigned increasingly simpler tasks.
The members of the downward-PI group were quite hetero-
geneous: Two participants had just started a new job at T3 and
were quite enthusiastic at this time. They had many ideas about
changes, and, apparently, the reduction of PI at T6 was just an
adaptation to the job. Many other members of this group used
uncontrollable work demands as a reason for not having devel-
oped PI at T6 (e.g., “I do not want to participate in continuing
education; I am glad if I am able to deal with my work right
now”). This suggests that an increase of feelings of uncontrol-
work characteristics change model. Autocorrelations between unique item factors are not shown. All freely
estimated factor loadings were significant. Si ? Situational Interview (Overcoming Barriers and Active
Approach subscales); Qi ? qualitative and quantitative initiative at work; Ie ? interviewer evaluation; T ?
Time; poc ? perceived opportunity for control; s-e ? self-efficacy; asp ? control aspiration. *p ? .05.
Paths and explained variance of the structural equation model of the reciprocal socialization plus
lable overload, low self-efficacy, and low control aspirations
was related to lower PI.
Similarly, the members of the upward-PI group did not fall into
one simple pattern. Some had just started a new job at T6, and this
fresh perspective might have helped them to detect things that
needed improvement. Other participants were still in their old job
at T6 but had received new responsibilities because of higher
business volume. This piqued their PI, although it had not yet
translated into a noticeable increase in control and complexity.
One member of this group had external reasons to show little PI at
T3: This person had worked only a few hours at T3 and expected
that the job would soon be eliminated. After the threat of losing the
job was removed, this person increased PI at work. This qualitative
description suggests that people did not necessarily change their
job (and even less their company) to increase or decrease their PI;
furthermore, it demonstrates that people can change the particulars
of their work characteristics within a given job.
Our model has fared quite well (see Figures 3 and 4). First, work
characteristics (control and complexity) affected control orienta-
tion (the common core of control aspiration, perceived opportunity
for control, and self-efficacy); second, control orientation had a
significant effect on PI; third, there were reciprocal relationships
from PI to work characteristics; and, fourth, control orientation
mediated the effects of work characteristics on PI.
The results seem at first glance to confirm a Marxist point of
view (people are determined by work) and the notion of social-
ization through work. However, this notion of socialization
through work needs to be refined: Work characteristics cannot
directly influence behavior; instead, this process is mediated by
control orientation as a critical psychological state. The effect of
work characteristics on one facet of control orientation—self-
efficacy—was also found by Parker (1998).
Conversely, the PI and control orientation effects on work
characteristics seem to confirm the worldview of Schopenhauer
(1819/1998). This shows that the seemingly opposing worldviews
of Marx and Schopenhauer seem to both be correct. Theoretically,
the two views have been integrated in Bandura’s (1997) notion of
reciprocal determinism, and our study provides an empirical un-
derpinning for this popular yet rarely studied notion. Furthermore,
our results are consistent with Bandura’s (1997) argument that
control orientation and personal initiative effects model—long term (includes only Time 3 [T3] and T6).
Autocorrelations between unique item factors are not shown. All freely estimated factor loadings were
significant. Si ? Situational Interview (Overcoming Barriers and Active Approach subscales); Qi ? qualitative
and quantitative initiative at work; Ie ? interviewer evaluation; poc ? perceived opportunity for control; s-e ?
self-efficacy; asp ? control aspiration. *p ? .05.
Paths and explained variance of the structural equation model of the socialization plus reciprocal
FRESE, GARST, AND FAY
reciprocal determinism works via self-efficacy, as self-efficacy
was part of the latent factor control orientation. In addition, the
results suggest an extension of Bandura’s model. Although a high
level of control orientation is important for the development of
work characteristics, our results suggest that PI has an additional
and independent effect on control orientation.
Our study also has produced unexpected findings. We had
originally hypothesized that PI would fully mediate the path from
control orientation to later work characteristics. This was not the
case; PI was only a partial mediator, as indicated by the direct
lagged effects from control orientation to work characteristics. One
possible interpretation is based on an effect of control orientation
on delegation behavior: Supervisors delegate challenging tasks to
those employees in whom they have confidence. This confidence
is not just created by past performance, as in past PI (Bauer &
Green, 1996), but may also be shaped by the impressions the
supervisor develops on the basis of employees’ statements of
control orientation. Individuals with high levels of control orien-
tation are likely to create an impression of high reliability and
competence, which makes them recipients of positive delegation
(Bauer & Green, 1996), producing higher work characteristics.
Strengths and Limitations
Our results are based on a unique study—a longitudinal design
with four waves with various data sources. It allowed us to
estimate different time lags and models with reciprocal paths
without running into identification problems and to essentially
replicate the findings within a single study. The longitudinal de-
sign overcomes some of the problems of common method vari-
ance, or unmeasured third variables. Because earlier levels of the
variables are held constant, constant sources of common method
variance (e.g., negative affectivity, response biases, personality
effects) are also held constant and can be controlled to a certain
extent (Zapf, Dormann, & Frese, 1996). Of course, our longitudi-
nal study cannot rule out the existence of unknown and changing
Although the participants were the source of all data, an impor-
tant feature of our study is our use of multiple perspectives
(participants and interviewers/coders) and multiple modes of data
collection to reduce percept–percept biases: survey responses,
interview responses, objective performance during the interview,
and interviewer evaluations. The variable overcoming barriers
(which measures one part of PI) is particularly interesting because
it is essentially a measure of respondents’ performance during the
interview (how many barriers was the participant able to over-
come?). Because the coders were trained and had a common
anchor point across participants, we avoided the problem of dif-
ferential anchor points that besets survey research. In the inter-
view, we asked the participants whether they had shown certain
behaviors—for example, whether they had developed an idea and
implemented it. Because interviewers probed the answers, the
coding procedure could isolate those behaviors that met our defi-
nition of PI (e.g., past PI behaviors). It was the coders who
decided, after substantial probing, whether a behavior constituted
PI, not the participant. Therefore, our interview may lead to Type
II errors (i.e., not finding PI where it exists), but it reduces Type I
errors (i.e., assuming PI to be present when it is not). Additionally,
relatively high stabilities for PI existed, even though in most cases
different interviewers conducted the interviews at different time
points. This indicates that our interviewer training was successful
in keeping coding errors to a minimum.
One limitation of our study is that we do not have objective
measures of work characteristics. Theoretical reasoning and em-
pirical data support our assumption, however, that behavior re-
quirements (e.g., complexity) can be described as relatively unbi-
ased; there is a certain kind of objectivity to the task situation
(Wood, 1986). The empirical literature reports substantial corre-
lations between job incumbents’ perceptions of work characteris-
tics and external observations (cf. Spector, 1992). Moreover, LIS-
REL analyses held prior perceptions of work characteristics
constant. Therefore, persistent tendencies to over- or underrate
work characteristics were controlled for to a certain extent. How-
ever, the possibility does exist that situational influences changed
the perception of work characteristics at any one time. Even so,
this is not likely to be the major factor that produced the pattern of
results, because there was stationarity of the items across time,
suggesting no change in their meaning.
Many of the paths were synchronous, and synchronous paths
cannot be interpreted unequivocally: They do not necessarily im-
ply an immediate effect (e.g., the effects of work characteristics on
control orientation). Their interpretation depends on the time frame
of the waves: If the time between two waves is 1 year, synchronous
means that the effect unfolds in 1 year or less. As Dwyer (1983)
pointed out, “the effects that are modeled as synchronous are
actually cross-lagged effects for which the appropriate lag is much
shorter than the period between waves of observation” (p. 397).
Thus, a conservative interpretation of our synchronous results is
that the effect times are smaller than one measurement lag.
At first glance, the stabilities far outweigh the paths between the
different constructs in Figure 3. Does this mean that the paths are
trivial because they are so small? We argue that this is not the case.
First, even small relationships have practical importance—the
paths, which were .28 on average (excluding stabilities) in our
final model, were higher than, for example, the relationship be-
tween alcohol and aggressive behavior (Meyer et al., 2001). Sec-
ond, our design increases stabilities and decreases the correlates
between variables because the model partitions the full 4 years into
smaller pieces. Stabilities are higher if time for change is short.
Therefore, the reanalysis in Figure 4 is important, as it shows
lower stabilities and, most often, higher substantive paths. If time
periods are longer, stabilities may decrease and paths between the
variables may increase.
Our argument that East Germany was in a situation of revo-
lutionary job change during the course of this study might raise
the question of whether our findings would generalize to the
more stable market economies in Western Europe and in the
United States. However, the relationships in our model are
relatively regular across time, which suggests that they would
also hold (albeit maybe not as strongly and more slowly) if the
change situation were not quite so radical. Evidence for this is
found in the similar cross-sectional intercorrelations in East and
West Germany (Frese et al., 1996). Moreover, Western econo-
mies are becoming increasingly like East Germany because of
accelerating job changes in today’s Western economies
Directions for Future Research and Practical
Our results suggest future research in the area of change pro-
cesses. High PI and control orientation lead to increased work
characteristics. We suggest two processes to be operative: (a)
changing work characteristics in current jobs by altering the
boundaries of one’s tasks or job and by adding or modifying
elements (and maybe eliminating others; cf. the concept of job
crafting; Wrzesniewski & Dutton, 2001), and (b) changing jobs
and companies and getting jobs with higher control and complex-
ity. Unfortunately, our study design and the situation in East
Germany did not allow us to unravel these two processes, but we
think it would be worthwhile to examine these processes in more
Future studies should examine contingency factors. There may
also be negative effects. PI should be useful for people with high
cognitive ability, knowledge, and skills. PI may also depend on job
design; job design that is mechanistic, Tayloristic, and oriented
toward simplification may not profit from PI, and in those jobs PI
may even have a negative effect on performance (Morgeson &
Campion, 2002; Wall et al., 2002). In a more general sense,
expectations of success and failure of PI and their effects on
showing PI, as well as the factors that shape individuals’ valence
of PI, have to be empirically studied (Vroom, 1964). PI may not
always be appreciated (at least in the long run) by coworkers and
supervisors. People who show a high degree of PI may be per-
ceived as being tiring and strenuous. Each initiative rocks the boat
and makes changes. Because people tend not to like changes, they
often greet initiatives with skepticism, as the literature on organi-
zational change has shown (e.g., Begley, 1998). However, in many
situations, PI should produce positive effects at work and on the
way a company works (Baer & Frese, 2003).
Our results have important practical implications. Because many
companies are moving from stable structures to change-oriented
organizations, managers should want to increase PI so that em-
ployees support change processes effectively (Baer & Frese,
2003). Managers may have to break the vicious cycle of con-
strained work characteristics, lack of PI, and low control orienta-
tion. Probably the best strategy is to simultaneously increase work
characteristics (control and complexity) and support the develop-
ment of control orientation. Training can be used to increase
control orientation by improving self-regulation (Frayne &
Latham, 1987; Neck & Manz, 1996). A complementary approach
is to select staff on the basis of past PI behavior.
Our results support a pluralistic approach to encouraging initia-
tive. There are various entry points, or drivers, to change the cycles
described (work characteristics, control orientation, and PI behav-
ior)—because all of the paths feed on each other, the end result
may be rather similar. The reciprocal model suggests, however,
that organizations can produce more powerful changes if the
different drivers point in the same direction. Some companies that
introduce new production initiatives (e.g., quality circles or lean
production) tell employees to be more daring, although they keep
the traditional assembly line intact and therefore do not increase
control and complexity at work. Thus, work itself is not changed,
but people are encouraged to show initiative. This strategy may be
effective to a certain extent but will prove to be limited (Lawler,
1992). People who take more initiative may leave the job to find
other work with more control and complexity. Others may not
show any initiative because they do not have enough mastery
experiences in their current job. Therefore, to get the strongest
effect, combining several drivers into a general, integrated ap-
proach may be best.
Abramson, L. Y., Seligman, M. E. P., & Teasdale, J. D. (1978). Learned
helplessness in humans: Critique and reformulations. Journal of Abnor-
mal Psychology, 87, 49–74.
Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52, 317–332.
Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in
practice: A review and recommended two step approach. Psychological
Bulletin, 103, 411–423.
Ashford, S. J., & Tsui, A. S. (1991). Self-regulation for managerial effec-
tiveness: The role of active feedback seeking. Academy of Management
Journal, 34, 251–280.
Baer, M., & Frese, M. (2003). Innovation is not enough: Climates for
initiative and psychological safety, process innovations and firm perfor-
mance. Journal of Organizational Behavior, 24, 45–68.
Bandura, A. (1997). Self-efficacy: The exercise of control. New York:
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable
distinction in social psychological research: Conceptual, strategic, and
statistical considerations. Journal of Personality and Social Psychology,
Bauer, T. N., & Green, S. G. (1996). Development of leader-member
exchange: A longitudinal test. Academy of Management Journal, 39,
Begley, T. M. (1998). Coping strategies as predictors of employee distress
and turnover after an organizational consolidation: A longitudinal anal-
ysis. Journal of Occupational and Organizational Psychology, 71, 305–
Bentler, P. M., & Chou, C. (1987). Practical issues in structural modeling.
Sociological Methods and Research, 16, 78–117.
Bollen, K. A. (1989). Structural equations with latent variables. New
Bollen, K., & Lennox, R. (1991). Conventional wisdom on measurement:
A structural equation perspective. Psychological Bulletin, 110, 305–314.
Bridges, W. (1995). Jobshift. London: Allen & Unwin.
Brown, R. L. (1997). Assessing specific mediational effects in complex
theoretical models. Structural Equation Modeling, 4, 142–156.
Bu ¨ssing, A. (1999). Can control at work and social support moderate
psychological consequences of job insecurity? Results from a quasi-
experimental study in the steel industry. European Journal of Work and
Organizational Psychology, 8, 219–242.
Byrne, B. M., Shavelson, R. J., & Muthe ´n, B. (1989). Testing for the
equivalence of factor covariance and mean structures: The issue of
partial measurement invariance. Psychological Bulletin, 105, 456–466.
Cohen, P., Cohen, J., Teresi, J., Marchi, M., & Velez, C. N. (1990).
Problems in the measurement of latent variables in structural equations
causal models. Applied Psychological Measurement, 14, 183–196.
DeShon, R. D., & Gillespie, Z. J. (2005). A motivated action theory
account of goal orientation. Journal of Applied Psychology, 90, 1096–
Dwyer, J. H. (1983). Statistical models for the social and behavioral
sciences. New York: Oxford University Press.
Edwards, J. R., & Bagozzi, R. P. (2000). On the nature and direction of
relationships between constructs and measures. Psychological Methods,
Fay, D. (1998). Personal initiative: Construct validation of a new concept
of performance at work. Unpublished doctoral dissertation, Universiteit
FRESE, GARST, AND FAY
Fay, D., & Frese, M. (2000). Conservatives’ approach to work: Less
prepared for future work demands? Journal of Applied Social Psychol-
ogy, 30, 171–195.
Fay, D., & Frese, M. (2001). The concept of personal initiative (PI): An
overview of validity studies. Human Performance, 14, 97–124.
Fay, D., & Sonnentag, S. (2002). Rethinking the effects of stressors: A
longitudinal study on personal initiative. Journal of Occupational
Health Psychology, 7, 221–234.
Finkel, S. E. (1995). Causal analysis with panel data. Thousand Oaks, CA:
Frayne, C., & Latham, G. (1987). Application of social learning theory to
employee self-management of attendance. Journal of Applied Psychol-
ogy, 72, 387–392.
Frese, M. (1982). Occupational socialization and psychological develop-
ment: An underemphasized research perspective in industrial psychol-
ogy. Journal of Occupational Psychology, 55, 209–224.
Frese, M. (1984). Do workers want control at work or don’t they: Some
results on denial and adjustment. Berlin: Institut fu ¨r Humanwissenschaft
in Arbeit und Ausbildung der Technischen Universita ¨t.
Frese, M. (1989). Theoretical models of control and health. In S. L. Sauter,
J. J. Hurrel Jr., & C. L. Cooper (Eds.), Job control and worker health
(pp. 107–128). Chichester, England: Wiley.
Frese, M. (2003). Zur Konstruktvalidita ¨t von Kontrollwahrnehmung und
Kontrollbewusstsein am Arbeitsplatz: Gu ¨tekriterien von Kontrollerwar-
tung, Kontrollhoffnung, Alternativen auf dem Arbeitsmarkt und Kontrol-
lablehnung [Construct validity of control cognitions at work: Reliability
and validity of control expectancy, hope for control, alternatives on the
labor market, and control rejection]. Unpublished manuscript, Univer-
sity of Giessen, Giessen, Germany.
Frese, M., Erbe-Heinbokel, M., Grefe, J., Rybowiak, V., & Weike, A.
(1994). “Mir ist es lieber, wenn ich genau gesagt bekomme, was ich tun
mu?”—Probleme der Akzeptanz von Verantwortung und Hand-
lungsspielraum in Ost und West [“I prefer to be told exactly what to
do”—Acceptance of responsibility and job discretion in East and West
Germany]. Zeitschrift fu ¨r Arbeits- und Organisationspsychologie, 38,
Frese, M., & Fay, D. (2001). Personal initiative (PI): An active perfor-
mance concept for work in the 21st century. In B. M. Staw & R. M.
Sutton (Eds.), Research in organizational behavior (Vol. 23, pp. 133–
187). Amsterdam: Elsevier Science.
Frese, M., Fay, D., Hilburger, T., Leng, K., & Tag, A. (1997). The concept
of personal initiative: Operationalization, reliability and validity in two
German samples. Journal of Occupational and Organizational Psychol-
ogy, 70, 139–161.
Frese, M., Kring, W., Soose, A., & Zempel, J. (1996). Personal initiative at
work: Differences between East and West Germany. Academy of Man-
agement Journal, 39, 37–63.
Graen, G. B., & Scandura, T. A. (1987). Toward a psychology of dyadic
organizing. Research in Organizational Behavior, 9, 175–208.
Hacker, W. (1973). Allgemeine Arbeits-und Ingenieurpsychologie [General
work and engineering psychology]. Berlin: VEB Deutscher Verlag der
Hacker, W. (1992). Expertenko ¨nnen. Erkennen und Vermitteln [Expertise –
cognition and communication]. Go ¨ttingen: Hogrefe.
Hackman, J. R., & Oldham, G. R. (1975). Development of the Job Diag-
nostic Survey. Journal of Applied Psychology, 60, 259–270.
Hackman, J. R., & Oldham, G. R. (1976). Motivation through the design of
work: Test of a theory. Organizational Behavior and Human Perfor-
mance, 16, 250–279.
Hall, D. T. (1996). Protean careers of the 21st century. Academy of
Management Executive, 10(4), 8–16.
Heckhausen, J., & Schulz, R. (1995). A life-span theory of control. Psy-
chological Review, 102, 284–304.
Holman, D. J., & Wall, T. D. (2002). Work characteristics, learning-related
outcomes, and strain: A test of competing direct effects, mediated, and
moderated models. Journal of Occupational Health Psychology, 7, 283–
Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance
structure analysis: Conventional criteria versus new alternatives. Struc-
tural Equation Modeling, 6, 1–55.
Ilgen, D. R., & Hollenbeck, J. R. (1991). The structure of work: Job design
and roles. In M. D. Dunnette & L. M. Hough (Eds.), Handbook of
industrial and organizational psychology (pp. 165–208). Palo Alto, CA:
Consulting Psychologists Press.
Jackson, D. L. (2003). Revisiting sample size and number of parameter
estimates: Some support for the N:q hypothesis. Structural Equation
Modeling, 10, 128–141.
Karasek, R. A., & Theorell, T. (1990). Healthy work: Stress, productivity,
and the reconstruction of working life. New York: Basic Books.
Kenny, D. A., & Campbell, D. T. (1989). On the measurement of stability
in over-time data. Journal of Personality, 57, 445–481.
Kohn, M. L., & Schooler, C. (1978). The reciprocal effects of the substan-
tive complexity of work and intellectual flexibility: A longitudinal
assessment. American Journal of Sociology, 84, 24–52.
Latham, G. P. (1989). Behavioral approaches to the training and learning
process. In I. Goldstein (Ed.), Training and development in work orga-
nizations: Frontiers of industrial and organizational psychology (pp.
256–295). San Francisco: Jossey-Bass.
Latham, G. P., & Saari, L. M. (1984). Do people do what they say? Further
studies on the Situational Interview. Journal of Applied Psychology, 69,
Lawler, E. E., III. (1992). The ultimate advantage: Creating the high
involvement organization. San Francisco: Jossey-Bass.
London, M., & Mone, E. M. (1999). Continuous learning. In D. R. Ilgen &
E. D. Pulakos (Eds.), The changing nature of performance (pp. 119–
153). San Francisco: Jossey-Bass.
MacCallum, R. C., & Browne, M. C. (1993). The use of causal indicators
in covariance structure models: Some practical issues. Psychological
Bulletin, 114, 533–541.
Marsh, H. W., Hau, K., Balla, J. R., & Grayson, D. (1998). Is more ever too
much? The number of indicators per factor in confirmatory factor
analysis. Multivariate Behavior Behavioral Research, 33, 181–220.
McDonald, R. P. (1996). Path analysis with composite variables. Multi-
variate Behavioral Research, 31, 239–270.
Meyer, G. J., Finn, S. E., Eyde, L. D., Kay, G. G., Moreland, K. L., Dies,
R. R., et al. (2001). Psychological testing and psychological assessment:
A review of evidence and issues. American Psychologist, 56, 128–165.
Mitchell, T., & James, L. R. (2001). Building better theory: Time and the
specification of when things happen. Academy of Management Review,
Morgeson, F. P., & Campion, M. C. (2002). Minimizing tradeoffs when
redesigning work: Evidence from a longitudinal quasi-experiment. Per-
sonnel Psychology, 55, 589–612.
Morrison, E. W. (1993). Longitudinal study of the effects of information
seeking on newcomer socialization. Journal of Applied Psychology, 78,
Neck, C. P., & Manz, C. C. (1996). Thought self-leadership: The impact of
mental strategies training on employee cognition, behavior, and affect.
Journal of Organizational Behavior, 17, 445–467.
Organ, D. W. (1988). Organizational citizenship behavior: The good
soldier syndrome. Lexington, MA: Lexington Books.
Parker, S. K. (1998). Enhancing role breadth self-efficacy. Journal of
Applied Psychology, 83, 835–852.
Parker, S. K., Wall, T. D., & Jackson, P. R. (1997). “That’s not my job”:
Developing flexible employee work orientations. Academy of Manage-
ment Journal, 40, 899–929.
Pentzt, M. A., & Chou, C. P. (1994). Measurement invariance in longitu-
dinal research assuming change from development and intervention. Download full-text
Journal of Consulting and Clinical Psychology, 62, 450–462.
Rohrmann, B. (1978). Empirische Studien zur Entwicklung von Ant-
wortskalen fu ¨r die sozialwissenschaftliche Forschung [Empirical studies
for the development of answer scales in social science research].
Zeitschrift fu ¨r Sozialpsychologie, 9, 222–245.
Rothbaum, F. W., Weisz, J. R., & Snyder, S. S. (1982). Changing the world
and changing the self: A two-process model of perceived control. Jour-
nal of Personality and Social Psychology, 42, 5–37.
Rotter, J. B. (1972). Generalized expectancies for internal versus external
control of reinforcement. In J. B. Rotter, J. E. Chance, & E. J. Phares
(Eds.), Applications of a social learning theory of personality (pp.
260–294). New York: Holt.
Schafer, J. L. (1997). Analysis of incomplete multivariate data. London:
Chapman & Hall.
Schopenhauer, A. (1998). Die Welt als Wille und Vorstellung [The world
as will and idea]. Munich, Germany: Deutscher Taschenbuch Verlag.
(Original work published 1819)
Schwarzer, R., Baessler, J., Kwiatek, P., Schroeder, K., & Zhang, J. X.
(1997). The assessment of optimistic self-beliefs: Comparison of the
German, Spanish, and Chinese versions of the General Self-Efficacy
Scale. Applied Psychology: An International Review, 46, 69–88.
Seibert, S. E., Kraimer, M. L., & Crant, J. M. (2001). What do proactive
people do? A longitudinal model linking proactive personality and
career success. Personnel Psychology, 54, 845–874.
Seligman, M. (1975). Helplessness: On depression, development and
death. San Francisco: Freeman.
Semmer, N. (1982). Stress at work, stress in private life and psychological
well-being. In W. Bachmann & I. Udris (Eds.), Mental load and stress
in activity: European approaches (pp. 42–55). Amsterdam: Elsevier.
Spector, P. E. (1986). Perceived control by employees: A meta-analysis of
studies concerning autonomy and participation at work. Human Rela-
tions, 11, 1005–1016.
Spector, P. E. (1992). A consideration of the validity and meaning of
self-report measures of job conditions. In C. L. Cooper & I. T. Robertson
(Eds.), International review of industrial and organizational psychology
(pp. 123–151). Chichester, England: Wiley.
Spector, P. E., & Jex, S. M. (1998). Development of four self-report
measures of job stressors and strain: Interpersonal Conflicts at Work
Scale, Organizational Constraints Scale, Quantitative Workload Inven-
tory, and Physical Symptoms Inventory. Journal of Occupational Health
Psychology, 3, 356–367.
Speier, C., & Frese, M. (1997). Generalized self-efficacy as a mediator and
moderator between control and complexity at work and personal initia-
tive: A longitudinal field study in East Germany. Human Performance,
Staw, B. M., & Boettger, R. D. (1990). Task revision: A neglected form of
work performance. Academy of Management Journal, 33, 534–559.
Thompson, J. A. (2005). Proactive personality and job performance: A
social capital perspective. Journal of Applied Psychology, 90, 1011–
Vancouver, J. B. (1997). The application of HLM to the analysis of the
dynamic interaction of environment, person and behavior. Journal of
Management, 23, 795–818.
Van Maanen, J. (1976). Breaking in: Socialization to work. In R. Dubin
(Ed.), Handbook of work, organization, and society (pp. 67–130). Chi-
cago: Rand McNally.
Vroom, V. H. (1964). Work and motivation. New York: Wiley.
Wall, T. D., Clegg, C. W., & Jackson, P. R. (1978). An evaluation of the
job characteristics model. Journal of Occupational Psychology, 51,
Wall, T. D., Cordery, J. L., & Clegg, C. (2002). Empowerment, perfor-
mance, and operational uncertainty: A theoretical integration. Applied
Psychology: An International Review, 51, 146–169.
Wilk, S. L., Desmarais, L. B., & Sackett, P. R. (1995). Gravitation to jobs
commensurate with ability: Longitudinal and cross-sectional tests. Jour-
nal of Applied Psychology, 80, 79–85.
Wood, R. E. (1986). Task complexity: Definition of the construct. Orga-
nizational Behavior and Human Decision Processes, 37, 60–82.
Wortman, C. B., & Brehm, J. W. (1975). Responses to uncontrollable
outcomes: An integration of reactance theory and the learned helpless-
ness model. Advances in Experimental Social Psychology, 8, 277–336.
Wrzesniewski, A., & Dutton, J. E. (2001). Crafting a job: Revisioning
employees as active crafters of their work. Academy of Management
Review, 26, 179–201.
Zapf, D. (1993). Stress-oriented job analysis of computerized office work.
European Work and Organizational Psychologist, 3, 85–100.
Zapf, D., Dormann, C., & Frese, M. (1996). Longitudinal studies in
organizational stress research: A review of the literature with reference
to methodological issues. Journal of Occupational Health Psychology,
Zempel, J. (1999). Selbstaendigkeit in den neuen Bundeslaendern: Praedik-
toren, Erfolgsfaktoren und Folgen - Ergebnisse einer Laengsschnit-
tuntersuchung [self-employment in East Germany: Predictors, success
factors, and outcomes]. In K. Moser, B. Batinic & J. Zempel (Eds.),
Unternehmerisch erfolgreiches Handeln [Entrepreurially successful ac-
tions] (pp. 69–91). Goettingen, Germany: Verlag fuer Angewandte
Received May 18, 2004
Revision received September 22, 2006
Accepted October 23, 2006 ?
FRESE, GARST, AND FAY