Académique Documents
Professionnel Documents
Culture Documents
net/publication/247727663
CITATIONS READS
277 10,510
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by K. Michele Kacmar on 27 August 2014.
Direct all correspondence to: K. Michele Kacmar, Department of Management. Florida State University, College
of Business, Tallahassee, FL 32306-I 110.
627
628 K. MICHELE KACMAR AND DAWN S. CARLSON
The perceptions individuals hold about the political nature of their work
environment influence the way they do their jobs. These perceptions affect how
employees feel about their company, boss, and co-workers, and they impact the
productivity, satisfaction, and intent to turnover of the workers (Ferris & Kacmar,
1992). Individuals’ perceptions about politics in the organization also determine
how political the environment will be. If employees perceive that others get ahead
by acting politically, these individuals will be more likely to engage in political
behaviors themselves (Ferris, Fedor, Chachere, & Pondy, 1989). Therefore, orga-
nizational culture is influenced by the degree of political activity found in an orga-
nization and how the employees in that organization react to these activities.
Although perceptions of organizational politics play an important role in the
organization, relatively little is known about this process (Ferris, Russ & Fandt,
1989). One reason for a lack of knowledge in this area is that no established scale
existed to measure perceptions of organizational politics. However, this void was
recently filled by Kacmar and Ferris (1991) who developed and performed an
initial validation of a measure entitled the Perception of Politics Scale (POPS).
This twelve-item scale was designed to assess the degree to which respondents
view their work environments as political. Armed with this information, managers
can better understand how employees perceive and react to politics in the organi-
zational environment around them.
While an initial validation of POPS was conducted during the development
of the scale, further examination of the dimensionality and construct validity of
POPS was performed by Nye and Witt (1993). However, before researchers inter-
ested in the area of organizational politics accept this scale for widespread use,
further validation is needed. Hence, the purpose of the present study is to provide
subsequent empirical validation of this scale.
spread support (Cropanzano, Kacmar & Bozeman, 1995; Drory & Romm, 1988,
1990; Ferris, Russ & Fandt, 1989; Porter et al., 1981). Virtually every article writ-
ten in the area includes some reference to the fact that the concept is difficult to
define. In 1990, Drory and Romm wrote an entire article on the definition of orga-
nizational politics. However, after reviewing and integrating the various defini-
tions of organizational politics that had been presented in the literature, they too
avoided offering a definition for the term.
When examining the definitions that have been offered in the literature
(Allen, Madison, Porter, Renwick & Mayes, 1979; Farrell & Peterson, 1982;
Ferris et al., 1993; Ferris & Judge, 1991; Ferris, Russ & Fandt, 1989; Kumar &
Ghadially, 1989; Porter et al., 1981), several commonalities emerge. One
theme is the fact that political activities are a means of exercising social influ-
ence. A second notion is that political behaviors are designed to promote or
protect one’s own self-interests. Finally, the notion that at least two parties
must be included and that these two parties have the potential to possess diver-
gent interests is either explicit or implicit in many definitions. Combining these
perspectives into one general definition allows one to view organizational poli-
tics as social influence attempts directed at those who can provide rewards that
will help promote or protect the self-interests of the actor (Cropanzano et al.,
1995).
Armed with a working definition of organizational politics, we are now
ready to explore past research in this domain. In the sections that follow, theoreti-
cal and empirical efforts in the area of organizational politics will be reviewed
briefly. The review is organized around the three factors generated by Kacmar
and Ferris (1991): general political behavior, which includes the behaviors of
individuals who act in a self-serving manner to obtain valued outcomes; go along
to get ahead, which consists of a lack of action by individuals (e.g., remain
silent) in order to secure valued outcomes; and pay and promotion policies,
which involves the organization behaving politically through the policies it
enacts.
General Political Behavior
It has been suggested that political behavior in organizations will increase
when rules and regulations are not available to govern actions (Drory & Romm,
1990; Fandt & Ferris, 1990; Ferris, Fedor, Chachere & Pondy, 1989; Ferris &
King, 1991; Ferris, Russ & Fandt, 1989; Kacmar & Ferris, 1993; Madison et al.,
1980; Tushman, 1977). In the absence of specific rules and policies for guidance,
individuals have few clues as to acceptable behavior, and therefore, develop their
own. When left to their own, individuals often develop rules that are self-serving
and better the position of the rule maker. Individuals who are more adept at deal-
ing with uncertain situations and persons who impose their own rules on others
are more likely to have their rules adopted.
Another process impacted by uncertainty is decision making. Decision
making under uncertainty has been found to be susceptible to political influence
(Drory & Ron-m, 1990). When the information needed to make an informed deci-
sion is lacking or ambiguous, decision makers rely upon their own interpretations
of the data. Multiple translations of the same information can result in ineffective
decisions that may appear political to those not directly involved in the decision
making process (Cropanzano et al., 1995).
Scarcity of valued resources (e.g., transfers, raises, office space, budgets)
generates competition. Several researchers have suggested that jockeying for a
position that will allow one to receive a valued resource is quintessential political
behavior (Drory & Romm, 1990; Farrell & Peterson, 1982; Kumar & Ghadially,
1989). This implies that organizations with limited resources will have political
environments. Since most organizations will have limited resources in at least one
area, political activities may occur in virtually any organization.
Examining exactly why resources are scant can help to predict who the target
of the political activities will be, as well as how heated the contest may become.
Any individual who has control over critical resources that cannot be secured
elsewhere will be a probable target of political influence tactics (Frost, 1987).
Further, the attractiveness and immediate benefit of the resource also will factor
into the decision to engage in political activities (Drory & Romm, 1990). In some
cases, a scarce resource, such as the organization’s tickets to a sporting event, may
only be valued by a few individuals, and hence, the actions engaged in to secure
this resource may not be as competitive as those used to secure a scare resource
valued by all, such as a raise or a promotion.
Go Along to Get Ahead
Conflict is consistently related to organizational politics in the literature
(Drory & Romm, 1988; Frost, 1987; Gandz & Murray, 1980; Mintzberg, 1985;
Porter et al., 1981; Tushman, 1977). The essence of this connection is that politi-
cal behavior is self-serving, and thus, has the potential to threaten the self-inter-
ests of others. When a threat is followed by retaliation, conflict arises (Porter et
al., 1981). According to Drory and Ron-n-n (1990), the existence of conflict is a
necessary underlying element of organizational politics. Further, the actual influ-
ence attempts themselves are an indication of the potential state of conflict that
exists between the two parties.
Some individuals may desire to avoid conflict, and therefore, not resist
others’ influence attempts. While this may appear to be a nonpolitical act, it
can actually be considered a form of political behavior. It has been suggested
that the distinction between political and nonpolitical behavior in organizations
can be made on the basis of intent (Drory & Romm, 1990). That is, if a behav-
ior is enacted specifically for the purposes of advancing one’s own self-inter-
ests, then the individual is acting politically (Frost, 1987). Individuals who
“don’t rock the boat” are not viewed as a threatening opponent by those who
are acting politically. Hence, the nonthreatening individual may be welcomed
into the “in-group” and received valued outcomes simply for not interfering
with a politically acting individual’s or group’s agenda. Lack of action, or
going along to get ahead, can be a reasonable and profitable approach to take
in order to advance one’s own self-interests when working in a political envi-
ronment .
discriminate validity of POPS because several of the POPS factors from phase
one seemed to parallel the concepts tapped by the JDI subscales.
The 40 items from POPS and the four JDI subscales were subjected to a prin-
cipal components analysis with an orthogonal varimax rotation which yielded five
factors, one for each JDI subscale and one for the politics items. However, as
expected, several of the POPS items (i.e., 13) loaded on one of the JDI factors.
Items that loaded only on POPS and not on a JDI factor were retained. The
retained items were subjected to a principal components factor analysis with a
varimax rotation resulting in three factors. Finally, classical test theory was
applied to the remaining three factors to further reduce them as described above.
The final results for the second phase produced a twelve-item, three-factor scale.
This final scale was further examined by Nye and Witt (1993). Specifically,
these authors examined the dimensionality and construct validity of POPS. With
respect to the dimensionality of POPS, Nye and Witt used both exploratory and
confirmatory factor analysis. First, they used principal components analysis with
a varimax rotation and selected factors with an eigenvalue over 1.0. They also
used structural equation modeling (i.e., LISREL) to compare a three-factor solu-
tion to a one-factor solution. The exploratory factor analysis results (eigenvalues
greater than 1) indicated that a one-factor solution was appropriate for the sample
used. The confirmatory factor analyses results, however, were not as clear. The
three-factor model produced a better model-to-data fit (GFI=.87 versus .89,
NFI=.93 versus .90), but had less parsimony (.74 versus .72). However, the high
factor correlations (-.85, -.94, and .91) provide additional evidence in support of
a one-factor solution.
Nye and Witt (1993) also examined the construct validity of POPS by
comparing it to the Survey of Perceived Organizational Support (SPOS; Eisen-
berger, Huntington, Hutchison & Sowa, 1986). Given that SPOS was designed
to measure the degree to which individuals viewed their organizations as
concerned about their well-being and appreciative of their efforts, a negative
relationship between SPOS and POPS was expected and found. Specifically,
results from these analyses indicated POPS and SPOS were strongly and
inversely related (-.85). Further, each of these scales produced significant but
oppositely signed correlations with other job related measures (e.g., job satis-
faction-POPS = -.62, SPOS = .68; commitment-POPS = -.58, SPOS = .59).
While these results indicate some conceptual overlap between POPS and SPOS,
there also are distinctions. The main difference between these two scales is their
focus. Each item in SPOS asks the respondent to comment on how the “organi-
zation” treats him or her while POPS more explicitly delineates the group to be
evaluated (e.g., influential group, supervisors, people in my department). If
raters focus on the same referent group when completing SPOS and POPS, the
conceptual overlap could be great. However, if top management rather than
departmental colleagues and immediate supervisors are the target of individu-
als’ SPOS ratings, there could be virtually no conceptual overlap.
Several limitations of the Nye and Witt (1993) examination should be noted.
First, the wording of two of the POPS items had to be changed in order for the
survey to be approved and distributed. The original item “There is no place for
Item 7
w-&o Alo
to Get
Ahead
Y
Figure 1. Model of Perceptions of Politics Scale
yes-men around here; good ideas are desired even when it means disagreeing with
supervisors.” (reverse coded) was changed to read “It is safer to agree with
managers than to say what you think is right.” Also, the original item “Since I
have worked in this department, I have never seen pay and promotion policies
applied politically.” (reverse coded) was modified to read “Pay and promotion
decisions are consistent with policies.” Hence, the integrity of the scale was
violated. Second, analysis of individual items was not performed. That is, POPS
and SPOS items were not factor analyzed together as was done with the JDI in the
initial validation of the POPS. An in-depth examination of the individual scale
items might have produced more useful results.
Since the validation research to date on POPS has not fully established the
dimensionality and psychometric properties of the scale, additional work is
needed. To address this need, the present study used a structural equation model-
ing approach (i.e., LISREL) to examine, validate, and modify POPS.
scales were introduced at this point to examine the convergent and discriminant
validity of POPS.
Study 3 - Augmenting POPS
In the third study, additional items were developed to augment the reduced
set of items. The development of the new items was based upon the rationale
described by Kacmar and Ferris (1991) when they originally developed items for
POPS and the theory outlined earlier. Content adequacy, content analysis, and
exploratory factor analyses were performed for each of these new items to deter-
mine which to keep. Those that remained were then examined via structural equa-
tion modeling using a new dataset collected specifically for this purpose.
STUDY 1: DIMENSIONALITY
Methods
Sample and Data Collection
The sample in Study 1 consisted of 749 responses (64% response rate) from an
attitude survey for a large state agency in which POPS was one of many variables
included. The survey was mailed via interoffice mail to all members of the agency.
A cover letter written by the director of the agency was included with each survey.
The cover letter introduced the project and stressed the importance of participation
in the project. An envelope addressed to the researchers also was enclosed so that
the respondents could return the surveys directly to the researchers.
Measure
Perceptions of politics. The Kacmar and Ferris (1991) 12-item Perception
of Politics Scale was administered to respondents from the state agency sample.
The internal reliability estimate for this sample was .87.
Analyses
To assess the dimensionality of POPS, structural equation modeling analysis
using LISREL 8 (Joreskog & Sorbom, 1993) was applied to the data from the
state agency sample to compare the fit of the three-factor model shown in Figure 1
to a one-factor model in which all 12 items were linked to one general political
factor. A X2 difference test was used to compare the two models. The best-fitting
model was then examined more closely for overall model fit using the available
LISREL indicators (e.g., GFI, AGFI, NFI, RFI).
Table 1. Correlation Matrix for Individual POPS Items*: State Agency Sample N = 749
Ite m Mean SD I 2 3 4 5 6 7 8 9 IO 11 12
I 2.6 1.2 -
2 2.6 1.1 .57 -
3 2.1 1.1 .40 .47 -
4 2.9 1.1 .60 .57 .39 _
5 2.6 1.2 .3X .44 SO .42 -
6 2.7 1.2 .51 .59 .44 .60 .46 -
I 2.7 1.1 .24 .38 .36 .31 .4X .33 -
8 2.6 1.2 .24 .35 .40 .34 .51 .33 .59 -
9 2.9 1.1 .32 .41 .42 .35 .41 .41 .39 .39 -
10 2.9 1.2 .29 .3X .3X .33 .40 .3X .37 .3X .65 -
11 2.3 1.3 .20 .24 .1X .1X .26 .14 .34 .33 .25 .22 -
12 2.6 1.1 .22 .29 .33 .21 .31 .22 .43 .39 .2x .2x .54 -
RMSEA = .lO; l-factor: X2(54) = 919.55, p = .OO, NFI = .76, CFI = .77,
PNFI = .56, RMSEA = .14), however, the overall fit was modest, most likely indi-
cating model misspecification. Further, a X2 difference test between these two
models (426 with 3 degrees of freedom) was significant (p < .Ol). Finally, the
correlations between the three factors were not excessively large (.74, .44, .62).
All of these results suggest that the three-factor model best fit the data, so further
examination of this model was conducted.
LISREL Results
Modification indices. Discriminant validity of the factors can be deter-
mined by examining the modification indices of Lambdax. According to
Medsker, Williams and Holohan (1994), values less than 4 are acceptable, while
values higher than 4 indicate that the items are loading on multiple factors and
that the error terms might be correlated. Results from the present data indicated
that half (12 of 24) exceeded the recommended cutoff.
Item loadings/lambdas. The standardized parameter estimates, often
referred to as the Lambdas, are presented in the middle of Figure 2. In order to
determine the significance of the loadings, t-values are examined because they are
independent of units of measurement (Joreskog & Sorbom, 1993). The t-values,
which ranged from 18.36 to 24.24, were all significant at the p < .Ol level. This
indicates that each item is significantly related to its specified construct.
Squared multiple correlations. The squared multiple correlations (SMCs)
provide information about the reliability of the items as well as the extent to
which they measure what they purport to measure. Specifically, SMC values indi-
cate the percent of variance accounted for by each item in the factor. These values
will always be less than the composite reliabilities. While composite reliabilities
.63
.60
.80
‘62
.7-l
.75
.68
.73
.90
.44
should be greater than .70, it is not possible to even suggest a rule of thumb cutoff
for SMCs (Bagozzi & Yi, 1988).
For the current sample, the SMCs ranged from .39 to .66 (mean = .49;
median = .48; 9 > = .45) with a composite reliability of .87. Since this value was
above the suggested cutoff of .70, the scale had acceptable reliability.
Chi-square. The chi-square test is a measure of overall fit of the model
to the data. The goal when using LISREL is to fail to reject the null hypothesis
that the model fit the data (i.e., Sigmadata = Sigma,,&. In this case, the chi-
square value was statistically significant (X2(51) = 489.76, p c .OOO),indicating
that the model does not fit the data. It is important to note, however, that when
the dataset used with this test is large, it is virtually impossible to fail to reject
the null hypothesis.
Goodness-of-fit. The goodness-of-fit index (GFI) is another indication of
how well the model fits the data. The current data produced a value of .89
which indicates that the model does a reasonable job of fitting the data.
However, to adjust for model parsimony, an adjusted goodness-of-fit index
(AGFI) is calculated which should exceed .90 (Medsker et al., 1994). The AGFI
value for the current sample dropped to .84, which placed it out of the accept-
able range.
Normed fit index. To examine the proportion of total variance accounted
for by a model, the normed fit index (NFI) is used (Medsker et al., 1994). An
acceptable value for the NFI is .90 (Medsker et al., 1994). The NFI for the three-
factor model was .87, indicating that this fit statistic falls just below the border of
acceptability.
Comparative fit index. The comparative fit index (CFI) is similar to the
NFI, except that it overcomes the difficulties associated with sample size
(Medsker et al., 1994). The CFI for the present model was .88, which falls just
below the acceptable level of .90 (Mulaik et al., 1989).
Parsimony fit index. The parsimony fit index (PFI) reflects the amount of
covariance explained by a model when its number of parameters is taken into
account (Renn & Vandenberg, 1995). That is, if the same amount of construct
covariance is explained by two models, the less complex of the two models will
have a higher PFI value. PFI values over .60 are considered acceptable (Mulaik et
al., 1989). A value of .67 was found for the present data, indicating adequate
parsimony for the model.
Summary. While it is clear that the three-factor model is superior to the
one-factor model, the detailed assessment of the overall fit of the three-factor
model demonstrated that it had shortcomings. Therefore, further adjustments to
the model needed to be made. One approach that can be taken to increase the
overall fit of a confirmatory factor analysis model is to remove items that do not
load significantly on the intended factor as well as those items that load on
multiple factors. To determine which, if any, POPS items fell into one or the
other of these categories, each individual item of POPS must be examined.
Study 2 was undertaken to ascertain each item’s contribution to the overall
model fit.
Methods
Overview
A variety of analyses were performed in study 2. First, content adequacy data
(Schreisheim et al., 1993) were collected for each of the twelve items in POPS.
This required respondents to indicate how closely each POPS item represented a
definition of each of the three factors of POPS. To determine which items did not
relate to the factor to which they were intended to measure, an exploratory factor
analysis of the content adequacy data was conducted. In addition, the mean values
for each item across the three factors also were calculated using the content
adequacy data. As a third decision criteria, exploratory factor analyses on the four
POPS datasets were used to ascertain which items to retain and which to discard.
Next, a multiple groups analysis using structural equation modeling was applied
to the remaining items using a series of four different datasets to determine the
overall fit of the model and validity of the remaining items. Finally, additional
scales were introduced to examine the convergent and discriminant validity of the
reduced POPS scale. Each of these steps is examined in greater detail below.
Sample and Data Collection
Content adequacy. A total of 102 upper level undergraduate students in
the College of Business at a large southern university were asked to complete a
content adequacy analysis of the original 12 items of POPS. Specifically, each
respondent was asked to determine the degree to which each item of POPS repre-
sented a factor definition. This required the judges to rate an item three times,
each time comparing it to a different factor definition. For example, a respondent
read the following definition for the “Go Along to Get Ahead” factor:
These are actions used to maintain the current way of thinking or the
way things are done in an organization in order to get ahead (e.g., not
making your opinion known to get along with important others).
and then rated how closely each of the 12 items from POPS fit this factor defini-
tion. This process was repeated after each new factor definition was presented.
The scale used was a 5-point Likert type scale with definitely not representative
(1) and definitely representative (5) as the anchors. The gender mix of the sample
included 63 (62%) males and 39 (38%) females. The average age was 23.5 years.
POPS data. A total of four separate samples were used to test the model-
to-data fit for the refined POPS scale (i.e., after the content adequacy and explor-
atory factor analyses were conducted). Each of these samples used POPS as a
variable in a larger data collection project. All of the data were collected via mail-
out surveys that were returned directly to the researchers. The first sample was
from an attitude survey for an electric cooperative from which a total of 466 (94%
response rate) responses were included. The next sample came from a survey
about current policies, perceptions, and attitudes of Human Resource profession-
als in a two state area. A total of 58 1 (39% response rate) responses were included
from this survey. The third sample represented 220 (44% response rate) nonfac-
ulty employees at a small no~heaste~ university who responded to an attitude
survey. In the fourth sample, 320 (64% response rate) responses were generated
from full-time employees in the private sector specifically for POPS validation
purposes. A correlation matrix for each of the four samples for each of the twelve
individual POPS items is available from the first author.
Measures
P~~ce~~on~ of politics. The Kacmar and Ferris (1991) 12-item perception
of politics scale was administered to respondents from each sample. The internal
reliability estimate for each sample was as follows: cooperative = X3, human
resource professionals = .86, university employees = .89, and validation study
respondents = .88.
The validation study respondents also were asked to respond to a variety of
scales that were thought to be distinct from but related to POPS as a means of
assessing the discriminant validity of POPS. These scales were used in the second
set of analyses in which the reduced set of items was examined across a variety of
samples. Each of these scales is described below.
F&h in people. The faith in people scale (Rosenberg, 1957) was designed
to measure one’s degree of confidence in the trustwo~hiness, honesty, and good-
ness in mankind. A reverse scored sample item is “If you don’t watch yourself,
people will take advantage of you.” In our study, individuals with high scores
represented people who were willing to use unscrupulous means to get ahead and
believed in the superior efficacy of “contacts” over ability. Thus, these individuals
were likely to believe that politicai behavior can and should be used to get ahead
in life. The Cronbach alpha for this scale was 68.
AGenation. The alienation via rejection scale (Streuning & Richardson,
1965) measures the emotional distance and purposelessness people experience
when dealing with others. A sample item is “Most people don’t realize how much
their lives are controlled by plots hatched in secret by others.” Thus, individuals
high in alienation believe that they cannot trust the people around them. This
would be consistent with POPS in that someone who saw activities as political
also would think that he or she has to beware of people who have an influence on
his or her life. The internal consistency estimate for this scale was .81.
An alienation scale developed by Dean (1961) also was used to examine the
alienation concept. This scale taps the powerlessness and normlessness that an
individual feels in his or her life. A sample item is “Sometimes I have the feeling
that other people are using me.” Thus, individuals scoring high on this scale
would feel they have little power over their own lives and view their lives as
controlled by others. This concept is very similar to perceptions of political
behavior in that an individual who views his or her work environment as political
would feel that he or she is negatively affected by the political actions of others.
This scale produced an internal reliability estimate of 64.
Cynicism. A cynicism subscale from the New F (Authorit~an) Scale
(Webster, Sanford & Freeman, 1955) was used to tap the degree of cynicism or
skepticism an individual has toward people in the world. An example item is “I
don’t blame anyone for trying to grab all he can get in this world.” Hence, this
scale measures ways in which political behavior is enacted in a social context. As
with POPS, the higher the cynicism score, the more likely the respondent will
view others’ behaviors as political. The Cronbach alpha for this scale was .67.
Ah&m. The altruism subscale from the Philosophy of Human Nature
Scale (Wrightsman, 1964) was used to measure the extent of unselfishness,
sincere sympathy, and concern one has for others. Wrightsman has shown both
reliability and validity of this measure. A sample item (reverse scored) is “People
pretend to care more about one another than they really do.” The scale was scored
such that a high score indicated selfishness and lack of concern for others. The
overall internal reliability estimate for this scale was .64.
Trust. Trust, a second subscale from the Philosophy of Human Nature
Scale (Wrightsman, 1964), was used to measure the expectancies people have
about the way other people generally behave. A sample item that is reverse scored
is “Most people would tell a lie if they could gain by it.” The scale was scored
such that a high score indicated low overall favorability toward human nature,
which would be consistent with POPS. This scale produced an internal consis-
tency estimate of .8 1.
Social Attitude. Campbell’s (1966) social attitude scale assesses how posi-
tive a view one has about humankind. In this scale, a positively viewed item was
paired with a negatively viewed item and the respondents were asked to indicate
their agreement with one of the paired items. A sample item is “The golden rule is
still the best rule to live by” versus “Nice guys finish last.” Similar to POPS, the
items were coded such that a high score represented a negative view of mankind.
The Cronbach alpha for this scale was .67.
Self-activity inventory. The self-activity scale is used as a general measure
of self- concept adjustment (Worchel, 1958). An example of one of the items used
is “I am a person who plays up to others in order to advance his/her position.”
This scale measures specific behaviors an individual will enact to adapt to his or
her environment. Those who attempt to fit into their environment should be more
political in nature, and thus be more aware of political behavior around them.
Therefore, an individual who scores high on the self-activity inventory also
should score high on POPS. This scale produced an internal consistency estimate
of .78.
Analyses
Data generated from the content adequacy analysis were examined using
exploratory factor analysis (principal components with an oblimin rotation) to
determine if the content adequacy ratings for the items loaded on the expected
dimensions. In addition, exploratory factor analyses (principal components with
an oblimin rotation) were performed on each of the four POPS datasets to confirm
these results. Finally, the mean rating for each item on each of the factors using
the content adequacy data was calculated and used as an additional decision tool.
Results from all of these analyses were used to determine which of the original
POPS items should be retained. Next, structural equation modeling was applied to
covariance matrices of the remaining items for each of the four datasets to explore
POPS 1 ,488 -.407 ,010 -.654 ,356 -.024 3.57 3.45 2.75
POPS2 ,623 -.114 ,113 -.685 .333 -.I31 4.00 3.60 2.79
POPS3 ,765 ,134 ,085 -.621 -.200 ,422 3.65 2.70 3.49
POPS4 ,758 ,008 -.012 -.838 p.203 -.I17 4.29 2.28 2.51
POPS5 ,885 ,116 -.054 .005 ,257 ,789 3.03 3.06 3.92
POPS6 -.072 -.894 ,046 -.151 ,806 -.073 3.39 4.34 2.54
POPS7 ,111 -.792 ,080 .074 -.065 .791 2.90 2.64 4.00
POPS8 ,471 -.354 -.013 -.200 -.092 ,715 3.06 2.52 3.74
POPS9 ,593 -.052 ,032 ,065 ,865 -.044 2.66 4.25 2.42
POPS10 ,555 -.262 p.034 .044 ,887 ,100 2.97 4.23 2.78
POPS 11 -.044 ,074 ,917 ,122 -.006 ,783 2.76 2.67 3.98
POPS12 ,084 -.134 .741 ,059 -.080 ,810 2.80 2.39 3.90
Eigenvalues 5.25 .94 1.16 1.77 2.51 3.71
Factor Correlations Fl->F2 Fl->F3 F2->F3 Fl->F2 Fl->F3 F2->F3
-.511 ,328 -.283 -.187 -.009 -.lll
644 K. MICHELE KACMAR AND DAWN S. CARLSON
the overall fit of the reduced model. Specifically, a reduced model was examined
by applying LISREL 8’s multiple group analysis technique to the four datasets.
Finally, the discriminant and convergent validity of the reduced POPS model was
examined using LISREL 8.
item 124
analysis also was applied to the four POPS datasets just discussed. Results for
each of these analyses are presented in Table 2. Also presented in Table 2 are the
mean values for each item from the content adequacy test across the three factors.
The six decision tools shown in Table 2 were used to judge which of the original
twelve POPS items should be retained.
In order for an item to be retained, it had to load on its intended factor in a
minimum of three of the five factor analyses. Further, the mean for the item had to
be largest on the factor it was designed to measure, and this value had to be signif-
icantly different than the mean for the other two factors. Implementing these rules
resulted in items 2,4, 9, 10, 11, and 12 being retained and items 1, 3, 5,6,7, and 8
being deleted.
Fit Statistics:
x2 12 35 32 21
df 6 6 6 6
GFI .99 .98 .99 .99
AGFI .98 .95 .95 .96
PGFI .28 .28 .28 .28
NFI .99 .96 .97 .98
PNFI .40 .38 .39 .39
CFI 1.oo .96 .98 .98
IF1 1 .oo .96 .98 .98
RF1 .98 .89 .93 .94
RMSEA ,036 ,079 ,075 ,068
Path Loadings: *
Item 1 -> Factor 1 1.02 .94 .I3 .86
Item 2 -> Factor 1 .64 .88 .69 .92
Item 3 -> Factor 2 .90 .99 .93 .80
Item 4 -> Factor 2 .73 .96 .83 .&O
Item 5 -> Factor 3 .80 .70 .70 .69
Item 6 -> Factor 3 .96 .59 .65 .71
Factor Correlations:
Factor 1 - Factor 2 .59 .72 .81 .80
Factor 1 - Factor 3 .56 .60 .55 .76
Factor 2 - Factor 3 .64 .54 .61 .69
LISREL Results
Figure 3 depicts the structural model of POPS after removing the items that
did not perform as expected in the content adequacy analysis. The model shown in
Figure 3 was analyzed using the multiple group feature of LISREL 8. Specifi-
cally, each of the remaining four datasets was used to explore the data-to-model
fit of the reduced model.
The first analysis run was a four-group comparison that required the path
loadings, factor correlations, and the error variances for each dataset to be equiva-
lent. The X2 for the comparison between the datasets was significant with a value
of 284.88 and 69 degrees of freedom (p < .OOO).However, the comparative fit
index for the datasets was acceptable (CFI = .90) as was the IFI (.90), and the GFI
(.94). These results suggest that the four different datasets map well to the model
with respect to the factor loadings, factor correlations, and error variances, indi-
cating that the model is generalizable to a variety of datasets. This model was
rerun allowing the path loadings and factor correlations to be re-estimated for
each dataset. The fit statistics for this model were stronger (GFI = .97, CFI = .93,
IF1 = .93). Table 3 presents the specific fit statistics for each of the four datasets
and the resulting path loadings and correlations for the freed model.
Convergent and Discriminant Validity
A scale has convergent validity if it is associated positively with other
measures of the same construct or other theoretically relevant constructs. That is,
different measures of the same construct should converge. A scale has discrimi-
nant validity if it is able to discriminate itself from measures of conceptually simi-
lar constructs (Kerlinger, 1986). That is, constructs similar to one another should
be correlated, just not too highly. A number of different measures that purport to
measure constructs that are conceptually similar to POPS were used in order to
establish the convergent and discriminant validity of the POPS. Specifically, the
validation sample respondents were simultaneously administered POPS and eight
other scales that measured theoretically similar constructs to POPS: faith in
people scale (Rosenberg, 1957) alienation via rejection scale (Streuning & Rich-
ardson, 1955) alienation (Dean, 1961), cynicism (Webster et al., 1955) altruism
and trust (Wrightsman, 1964) social attitude (Campbell, 1966), and a self-activity
inventory (Worchel, 1958).
To examine the convergent and discriminate validity of POPS, the process
described by Anderson and Gerbing (1988) was followed. Anderson and Gerbing
suggested developing and testing a measurement model with all of the variables
of interest included. This model should include multiple indicators that are linked
directly to the variable of interest and to no other variables. These multiple indica-
tors can be individual scale items or composites of these items. The only require-
ment is that each set of indicators be unidimensional. Therefore, exploratory
factor analyses were run on the eight scales used for validity purposes to create
subscales where necessary to insure unidimensionality.
Since the overall fit for this model was reasonable (GFI = .88, PGFI = .74,
PNFI = .70, CFI = .88), both discriminant and convergent validity could be
assessed. Convergent validity is present when each indicator’s estimated path
coefficient for its assigned variable is significant. Since all of the path coefficients
in the measurement model were significant, convergent validity was achieved.
Discriminant validity can be assessed by examining the confidence intervals
for the factor correlations. This required that confidence intervals for the correla-
tions between POPS and faith in people, alienation, alienation via rejection, cyni-
cism, altruism, trust, social attitude, and self-activity inventory be calculated. In
each case, the confidence intervals showed that the correlations were significantly
different from 1.O. The magnitude of the range of correlations was from .24 to .59,
further showing that different constructs were being measured. Thus, the results
suggested, both statistically and practically, that the concepts considered here are
distinct from POPS.
Summary
Study 2 provided many significant and important findings. First, the
content adequacy analysis and exploratory factor results showed that several of
the original POPS items were not functioning as originally intended. Several
items cross loaded or did not clearly relate to the factor that they were intended
to measure. These results suggested that the original scale should be reduced to
six items (i.e., 2, 4, 9, 10, 11, and 12) with two items remaining on each of the
three factors.
Next, the generalizability of the new factor structure for POPS was examined
via a multiple group LISREL 8 analysis. Results from this test indicated that the
new factor structure did indeed map to the four different datasets used in the anal-
yses. Individual fit statistics for each of the datasets were extremely strong. These
results suggested that the remaining items were stable across a variety of datasets
representing a variety of samples and that the new refined POPS model fit all of
the datasets well.
Finally, the convergent and discriminant validity of the reduced scale was
examined. With respect to convergent validity, all of the retained items in POPS
loaded significantly on the factor that they were intended to measure. To assess
discriminant validity, POPS was compared to a variety of scales that were
suggested to represent a construct similar in nature to the underlying construct
measured by POPS. All of these analyses indicated that while POPS was similar
to these scales, it was distinct (i.e., it showed discriminant validity).
Method
Sample and Data Collection
Content adequacy. The same respondents who performed the content
adequacy analysis for the original items in Study 2 were used to perform the
content adequacy analysis on the new items that were developed to augment the
existing POPS scale. The same procedure was followed in that respondents were
required to rate each item three times, once for each factor definition. As previ-
ously explained, the ratings made determined the degree to which the item
measured the definition of the factor being considered.
ro-- ---mchw~Q-a-N-
113-m-3
Results from these four analyses were used to decide which new items to
include and exclude. In order to pass the factor analysis tests, an item had to load
at .40 or higher on its intended factor and less than or equal to .35 on all others.
With respect to the mean decision criteria, the mean for an item had to be greater
than or equal to 4.00 on the intended factor and less than 3.5 on all others to pass.
Finally 10 judges (71%) had to place the item on its intended factor to meet this
decision criteria. Applying these four decisions criteria, 9 of the 14 new items
were retained (i.e., 1, 2, 3, 4, 5, 6, 8, 9, 10). When these 9 items were combined
with the original 6, the final scale was composed of 15 items (shown in Table 5)
that represent three factors.
To examine the dimensionality of the modified POPS scale, a structural
equation modeling approach was applied to a variancekovariance matrix of the
POPS data created by including the items shown in Table 5. Specifically, the
three-factor model shown in Figure 4, as well as a one-factor model (i.e., all
fifteen items loading on one factor), were examined using LISREL 8. Results for
these analyses indicated that the fit statistics for the three-factor model were better
than the fit statistics for the one factor model on all indicators (three-factor:
RMSEA = .076, GFI = .91, AGFI = .87, NFI = .86, NNFI = .87, PNFI = .72, CFI
.67
.67
.74
.61
40
.29
.61
.66
.70
I
Item 9
.97
.91
.68
.27
.25
= .91, IF1 = .91, RF1 = 84; l-factor: RMSEA = .13, GFI = .79, AGFI = .72, NFI
= .70, NNFI = .70, PNFI = .60, CFI = .74, IF1 = .74, RFI = .66). Additionally, a
X2 difference test between these two models (276.55 with 3 degrees of freedom)
was significant @ < .OOO). Finally, two of the three correlations between the
factors were moderate (.54, .71, and 1.OO). All of these results suggest that the
three-factor model best fit the data, so further examination of this model was
conducted.
Modification indices. An examination of the modification indices of
Lambdax can be performed to examine the discriminant validity of the factors.
According to Medsker et al. (1994), values less than 4 are acceptable, while
values higher than 4 indicate that the items are loading on multiple factors and
that the error terms might be correlated. Results from the present data indicated
that only 8 of the 30 values (27%) exceeded the recommended cutoff.
Item loadings/lambdas. The completely standardized parameter estimates,
often referred to as the Lambdas, are presented in the middle of Figure 4. In order
to determine the significance of the loadings, t-values were examined because
they are independent of units of measurement (Joreskog & S&born, 1993). The t-
values, which ranged from 3.02 to 17.89, were all significant at the p < .Ol level.
This indicates that each item is significantly related to its specified construct.
Squared multiple correlations. The squared multiple correlations (SMCs)
provide information about the reliability of the items, as well as the extent to
which they measure what they purport to measure. Specifically, SMC values indi-
cate the percent of variance accounted for by each item in the factor. These values
will always be less than the composite reliabilities. While composite reliabilities
should be greater than .70, it is not possible to even suggest a rule of thumb cutoff
for SMCs (Bagozzi & Yi, 1988).
For the current sample, the SMCs ranged from .03 to .75 (mean = .41;
median = .34; 6 > = .45) with a composite reliability of .87 for the overall scale.
Since this value was above the suggested cutoff of .70, the scale had acceptable
reliability.
Chi-square. The chi-square test is a measure of overall fit of the model
to the data. The goal when using LISREL is to fail to reject the null hypothesis
that the model fit the data (i.e., Sigmadata = Sigma,,del). In this case, the chi-
square value was statistically significant (X2(87) = 237.29, p = .OO) indicating
that the model does not fit the data. However, when using large samples, it is
normally impossible to fail to reject the null hypothesis. One way in which the
size of the sample can be taken into consideration is by dividing the X2 value by
the degrees of freedom. Values less than 5 indicate that the model does fit the
data (Wheaton, Muthen, Alwin & Summers, 1977). For the present sample, the
X2/df ratio was 2.73 indicating good model-to-data fit.
Goodness-of-fit. The goodness-of-fit index (GFI) is another indication of
how well the model fit, the data. The current data produced a GFI value of .91,
which indicates that the model fits the data. However, to adjust for model parsi-
mony, an adjusted goodness-of-fit index (AGFI) is calculated which should
exceed .90 (Medsker et al., 1994). The value for the current sample was .87,
which is just below the acceptable range.
General Discussion
In an effort to further validate POPS, a three-study investigation of the
dimensionality, reliability, and validity of POPS was undertaken. In the first
study, a comparison was made of the original three-factor model to a one-factor
model. Results suggested that while the three-factor model produced a better
model-to-data fit than the one-factor model, the fit of the three-factor model could
be improved. Model fit can be improved by adding or deleting paths. Given the
format of the model tested, adding paths meant adding new items to the scale,
while deleting paths meant deleting items from the scale. Both of these possibili-
ties were investigated in Study 2 and Study 3.
In Study 2, each item of the original scale was examined for its contribution
to the overall scale. Six items that either cross loaded or did not load on the
intended factor were removed. The reduced scale was then examined across four
different samples to assess its stability. Results indicated good fit for the refined
scale across each dataset as well as within each. In addition, the new six-item
scale demonstrated both convergent and discriminant validity.
The final step was to create new items to add to the scale. In Study 3, four-
teen new items were generated and examined for inclusion. Applying four deci-
sion criteria to the new items indicated that 9 items should be retained. The
resulting fifteen item scale was then examined for overall dimensionality by
comparing it to a one-factor model as was done in Study 1. Results indicated that
the three-factor model fit substantially better than the one-factor model. Further,
the X2 difference test was significant, indicating that a three-factor model was
more appropriate than a one-factor model. The three-factor model was then exam-
ined for overall model fit. Results suggested that the model-to-data fit was strong.
As with any empirical study, there are limitations which need to be
mentioned. For example, using students as content adequacy raters may be
considered a limitation. It is important to note, however, that content adequacy
indicated that factor 1 was unique and should be included as a separate factor.
However, future research efforts should be undertaken to add additional items to
this factor in an effort to distinguish it even more from factor 2.
Given that the lack of interest in organizational politics research may be
attributed to the difficulty early researchers had defining, quantifying, and
measuring this elusive phenomenon, the final scale resulting from this work is an
important contribution to the organizational politics literature. Only when consen-
sus is reached about what organizational politics is and how it should be measured
will the field be advanced. The development and refinement of POPS is offered as
a step toward reaching a meeting of the scholarly minds.
APPENDIX
1. When if comes to pay raises and promotion decisions, policies are irrelevant.
2. Agreeing with powerful others is the best alternative in this organization.
4. Promotions around here are not valued much because how they are deter-
mined is so political.
5. I have seen changes made here that only serve the purposes of a few individu-
als, not the whole work unit or department.
6. Sometimes it is easier to remain quiet than to fight the system.
7. Favoritism, rather than merit, determines who gets good raises and promo-
tions around here.
8. Telling others what they want to hear is sometimes better than telling the
truth.
9. It is safer to think what you are told than to make up your own mind.
10. Inconsistent with organizational policies, promotions in this organization
generally do not go to top performers.
11. None of the raises I have received are consistent with the policies on how
raises should be determined.
12. This organization is not known for its fair pay and promotion policies.
13. Rewards such as pay raises and promotions do not go to those who work hard.
14. The stated pay and promotion polices have nothing to do with how pay raises
and promotions are determined.
References
Allen, R.W., Madison, D.L., Porter, L.W., Renwick, P.A. & Mayes, B.T. (1979). Orgamzational polltics: Tactics
and characteristics of its actors. California Mana,qement Review. 22: 77-83.
Anderson, J.C. & Gerbing, D.W. (1988). Structural equation modeling in practice: A review and recommended
two-step approach. Psychdogical Bulkfin, IOj: 41 1423.
Bagozzi, R.P. & Yi, Y. (1988). On evaluation of structural equation models. Jolrrnal ofAcademy ofMnrketing
Science, 16: 74-94.
Campbell. D. (1966). Unpublished paper, Department of Psychology. Northwestern University.
Cropanzano, R.S.. Kacmar. K.M. & Bozeman, D.P. (1995). Organizational polmcs, justice, and support: Their
differences and similarities (pp.I-IX), in R.S. Cropanzano & K.M. Kacmar (Eds.). OrX”ni~ntionnIpolitics.
Justice and support: Managing socicll climate nr work. Westport, CT: Quorum Books.
Dean. D. (1961). Alienation: Its meaning and measurement. American Socio&icrrl ReiYe~, 25: 753- 758.
Drory, A. & Romm, T. (1988).Politics in organization and its perception in the organization. Ory+vzi,-utionnl Stud-
ies. 9: 16.5~179.
Drory, A. & Romm, T. (I 990). The definition of organizational politics: A review. Humcv~ Relmior~. 43: 1 133%
1154.
Eisenberger. R., Huntington. R., Hutchison. S. & Sowa. D. (I 986). Perceived organizational support. Journal oj
Applied Psychology, 71: 500-507.
Fandt. P.M. & Ferris, G.R. (1990). The management of information and impressions: When employees behave
opportumstically. Orgcmiz,ationnl Behm?or and Human Decision Processes, 45: 130-l 5X.
Farrell, D. & Peterson, J.C. (1982). Patterns of political behavior in organizations. Acuderny of Mnnc~gement
Rev&v, 45: 403-412.
Ferris, G.R., Brand, J.F.. Brand, S., Rowland, K.M., Gilmore. D.C.. King. T.R., Kacmar, K.M. & Burton. C.A.
(1993). Politics and control in organizations (pp. X3-1 I I). in E.J. Lawler. B. Markovsky, J. O’Brien, K.
Heimer (Edu.), Adwmres in groupprocesses (vol. 10). Greenwich, CT: JAI Press.
Ferris, G.R. & Buckley, M.R. (1990). Performance evaluation in high technology firms: Process and politics (pp.
243-263). in L.R. Gamer-Mejia & M.E. Lawless (Eds.). Organizational issues in high technology manage-
ment. Greenwich, CT: JAI Press.
Ferris. G.R.. Fedor. D., Chachere. J.G. & Pondy. L. (1989). Myths and politics in organizational contexts. Gmup
& Orgrrni:ntionnl Srudies. 14: 8X-103.
Ferris, G.R.. Frink, D.D., Galang. M.C.. Zhou. J.. Kacmar. K.M. & Howard. J.L. (1996). Political work environ-
ments. Hutntrn Relations. 49: 233-266.
Ferris, G.R., Frmk. D.D., Gilmore. D.C. & Kacmar, K.M. (1994). Understandmg politics: Antidote for the
dysfunctional consequences of organizational politics as a stressor. Journcd ofApplied Social P.+~olog~
24: 1204&1220.
Ferris. G.R. & Judge, T.A. (1991). Personnel/human resources management: A political influence perspective.
Joumcrl of Management. 17: 447-488.
Ferris, G.R. & Kacmar, K.M. (I 992). Perceptions of organizational politics. Journal ofManclge,nent. 18: 93-l 16.
Ferris, G.R. & King, T.R. (1991). Politics in human resource decisions: A walk on the dark side. Orgcmi:crtiona/
D~mnics. 20: 59-7 I.
Ferris, G.R.. Russ, G.S. & Fandt. P.M. (1989) Politics in organizations. (pp. 143-170), in R.A. Giacalone & P.
Rosenfeld (Eds.), Impression management in the orgnni:nrron. Hillsdale. NJ: Lawrence Erlbaum.
Frost, P.J. (1987). Power, politics, and influence. In F. Jablin, L. Putnam, K. Roberts, & L. Porter (Eds.), Hnnd-
hook oforgmbtionul communicntion. Beverly Hills, CA: Sage.
Gandz. J. & Murray, V.V. (1980). The experience of workplace politics. Academy ofManagement Journal. 23:
231-25 1.
Humphreys, L.G. & Montanelli, R.G. (1975). An investigation of the parallel analysis criterion for determining
the number of common factors. Multivarime Behavioral Research, IO: 193-205.
Joreskog. K.G. & Sorbom, D. (1993). LISREL 8: Structural equntion modeling with the SIMPLIS co~nmtmd
language. Hillsdale, NJ: SSI.
Kacmar. K.M. & Ferris, G.R. (1991). Perceptions of organizational politics scale (POPS): Development and
construct validation. Educntional and Psychological Measurement, 51: (193-205.
Kacmar, K.M. & Ferris, G.R. (1993). Politics at work: Sharpenmg the focus ofpolitical behavior in organizations.
Business Hori:on.s, 36: 70-74.
Kerlinger, F.N. (1986). Foundations ofbehn~ioral resenrch (3rd ed.). New York: Holt, Rinehart. and Winston.
Kumar, P. & Ghadially, R. (1989). Organizational politics and its effect on members of organizations. Human
Relations. 42: 305-314.
Lewin, K. (1936). Principles oftopoloRicnlp.s~~hol~g~. New York: McGraw-Hill.
Madison, D.L., Allen, R.W., Porter, L.W., Renwick, P.A. & Mayes, B.T. (1980). Organizational politics: An
exploration of managers’ perceptions. Humun Relafions, 33: 79-100.
Mayes, B.T. & Allen, R.W. (1977). Toward a definition of organizational politics. Academ,v ofManagement
Review, 2: 672-678.
Medsker, G.J., Williams, L.J. & Holohan, P.J. (1994). A review of currentpractices for evaluating causal models
in organizational behavior and humanresources managementresearch. J~~urnalofManaRement, 20: 439464.
Mintzberg, H. (1985). The organization as political arena. Journnl ofManagement Studies, 22: 133-154.
Mulaik. S.A., James, L.R.. Van Alstine, .I., Bennet, N., Lind, S. & Stillwell, C.D. (1989). An evaluation of good-
ness-of-fit indices for structural equation models. P.@o/ogicn/ Bullerin, 105: 430445.
Nunnally, J.C. (1967). Psychometric theory. New York: McGraw-Hill.
Nye, L.G. & Witt, L.A. (1993). Dimensionality and construct validity of the perceptions of politics scales (POPS).
Education and Psychological Measurement, 53: 821-829.
Porter, L.W. (1976). Organizations as political animals. Presidenfial address, Division of Industrial- Orgnni:rr-
rionul Psychology, 84th Annual Meeting of the American Psychological Association, Washington, DC.
Porter, L.W., Allen, R.W. &Angle, H.L. (1981). The politics of upward influence in organizations (pp. 109% 149)
in L.L. Cummings & B.M. Staw (Eds.), Research in organizational behavior (vol. 3). Greenwich, CT: JAI
Press.
Renn, R.W. & Vandenberg, R.J. (1995). The critical psychological states: An underrepresented component in job
characteristics model research. Journal ofManagemenr, 21: 279-303.
Rosenberg, M. (1957). Occupations and values. Glencoe, IL: The Free Press.
Schein. V.E. (1977). Individual power and political behaviors in organizations: An inadequately explored reality.
Academy of Management Renew, 2: 64-72.
Schreisheim, C.A., Powers, K.J., Scandura, T.A., Gardiner, C.C. & Lankau, M.J. (1993). Improving construct
measurement in management research: Comments and a quantitative approach for assessmg the theoretical
content adequacy of paper-and-pencil survey-type instruments. Journc~l ofMunagement, 19: 3855417.
Smith, P.C., Kendall, L.M. & Hulin, C.L. (1969). The meamrement c>f sarisfaction innvrk and retirement.
Chicago: Rand McNally.
Streuning, E. & Richardson, A. (1965). A factor analytic exploration of the alienation, anomia, and authoritarian-
ism domain. American Sociological Review, 30: 768-776.
Tushman, ME. (1977). A political approach to organization: A review and rationale. Academy of Manu,gement
Review, 2: 206-2 16.
Webster, H., Sanford, N. & Freeman, M. (1955). A new instrument for studying authoritarianism in personality.
Journul of Psychology, 40: 73-85.
Wheaton, B.B., Muthen, B., Alwin. D.F. & Summers, G.F. (1977). Assessing reliability and stability in panel
models. In D.R. Heise (Ed.). Sociologicul Methodology, San Francisco: Jossey-Bass.
Worchel, P. (1958). Personality factors in the readiness to express aggression, Jorrrnal of Clinical Psychology, I4:
355-359.
Wrightsman. L. (1964). Measurement of philosophies of human nature. Psychologicnl Reports. 14: 743-75 1.