Professional Documents
Culture Documents
Testing A Revised Measure of PSM
Testing A Revised Measure of PSM
net/publication/249234533
CITATIONS READS
138 1,239
1 author:
Sangmook Kim
Seoul National University of Science and Technology
52 PUBLICATIONS 3,141 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Sangmook Kim on 12 October 2017.
ABSTRACT
INTRODUCTION
Academics have long explored whether there are special motives for public service that
differ from those for the private sector (Frederickson and Hart 1985; Mosher 1968; Perry
and Porter 1982). Public service motivation (PSM) provides a useful basis for understand-
ing public employee motivation. According to Perry and Wise (1990, 368), PSM is defined
as ‘‘an individual’s predisposition to respond to motives grounded primarily or uniquely in
public institutions and organizations.’’ Even though the definitions of PSM itself vary
slightly by authors (Brewer and Selden 1998; Rainey and Steinbauer 1999; Vandenabeele,
Scheepers, and Hondeghem 2006), ‘‘its definition has a common focus on motives and
action in the public domain that are intended to do good for others and shape the well-being
of society’’ (Perry and Hondeghem 2008, 3).
The author is grateful for valuable comments on earlier drafts from James L. Perry, David H. Coursey,
and anonymous JPART reviewers. Special thanks to Professor Seung Hyun Kim and Professor Hyok-Joo Rhee
at the Department of Public Administration, Seoul National University of Technology, for their kind support
and encouragement. Address correspondence to the author at smook@snut.ac.kr.
doi:10.1093/jopart/muq048
Advance Access publication on August 12, 2010
ª The Author 2010. Published by Oxford University Press on behalf of the Journal of Public Administration Research
and Theory, Inc. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
522 Journal of Public Administration Research and Theory
PSM has rational, norm-based, and affective motives (Perry and Wise 1990). Rational
motives are grounded in individual utility maximization, norm-based motives are grounded
in a desire to pursue the common good and further the public interest, and affective motives
are grounded in human emotion. A variety of rational, norm-based, and affective motives
appear to be primarily or exclusively associated with public service. Rational motives in-
clude participation in the process of policy formulation, commitment to a public program
because of personal identification, and advocacy for a special or private interest. Norm-
based motives include a desire to serve the public interest, loyalty to duty and to the gov-
ernment as a whole, and social equity. Affective motives include patriotism of benevolence
and commitment to a program from a genuine conviction about its social importance.
Perry (1996) identified a multidimensional scale to measure PSM, which has four di-
mensions: attraction to policy making (APM), commitment to public interest (CPI), com-
passion (COM), and self-sacrifice (SS). Since then, many studies conducted in the United
States and in other countries have used the dimensions and items of Perry’s (1996) measure,
and improvement has been made in the PSM scale. Close attention to the measurement of
it also may tap cynicism or negative affect toward politics. Thus the addition of positively
worded items to the APM sub-scale would be desirable.’’ A revised measurement scale
needs to be developed. In a previous study, new items were composed to measure
APM, and the revised 12-item PSM scale was tested (Kim 2009a). However, additional
testing with different samples is still desirable, as is evaluation of new items. The first
research issue is to validate the revised APM items with another sample.
The second issue concerns the dimensions of PSM. Perry’s (1996) exploratory anal-
ysis provided support for both a three- and a four-dimension PSM construct. Four dimen-
sions were empirically distinguished, but the dimensions of CPI and SS were highly
correlated (r 5 .89). Even though a differential x2 test indicates that the four-dimension
model is superior to the three-dimension model, combining CPI and SS, there is relatively
little difference between the two. Among 19 studies using Perry’s (1996) measurement
scale for PSM, eight studies analyzed four dimensions but seven studies used three dimen-
sions. Even in the same country, both the three-dimensional solution and the four dimen-
sions of PSM are reported (see table 1). For example, Vandenabeele (2008a) found that the
1 The choice of measurement model would depend on the generality or specificity of one’s theoretical interest
(MacKenzie, Podsakoff, and Jarvis 2005). A typical example of formative measurement model is socioeconomic status
(SES), which is formed as a combination of education, income, occupation, and residence. If any one of these
indicators increases, SES would increase; conversely, if a person’s SES increases, this would not necessarily be
accompanied by an increase in all four indicators (Diamantopoulos and Winklhofer 2001). Other examples may be job
performance, role stress, procedural justice, transformational leadership, job perceptions, human development index,
and quality-of-life index (Diamantopoulos and Winklhofer 2001; Edwards 2001; MacKenzie, Podsakoff, and Jarvis
2005). A typical example of the reflective model is the Organizational Commitment Questionnaire (Mowday, Steers,
and Porter 1979). Variation in the level of organizational commitment leads to variation in its indicators, but not vice
versa. Other examples may be personality, attitude, general work values, LMX, work withdrawal, and psychological
climate (Diamantopoulos and Siguaw 2006; Edwards 2001). However, some constructs can be regarded not only as
a reflective but also as a formative model. For example, job satisfaction is treated as a unidimensional construct when it
is measured by indicators that each describe satisfaction with the job as a whole. In this case, the indicators of job
satisfaction are reflective, as treated in this study. On the other hand, job satisfaction can also be viewed as a composite
of satisfaction with specific job facets, such as pay, promotion, supervisor, coworkers, and the work itself. In that case,
the dimensions of job satisfaction are formative rather than reflective (Kim 2005b; Ironson et al. 1989; Williams,
Edwards, and Vandenberg 2003).
524 Journal of Public Administration Research and Theory
Table 1
Previous Studies Using Perry’s (1996) Measurement Scale for PSM
Four-Dimension Model Three-Dimension Model or Others
Lee (2005) 1) Korea; public and Scott and Pandey (2005) 1) USA; managers in
private employees state health and
human service
agencies
2) 24 items; 24 items 2) 11 items; 11 items
3) APM, CPI, COM, SS; 3) APM, CPI, COM
no report on a
Camilleri (2006) 1) Malta; public officials DeHart-Davis, 1) USA; managers in
Marlowe, and Pandey state health and
(2006) human service
agencies
2) 24 items; 24 items 2) 10 items; 10 items
3) APM (.21), CPI (.63), 3) APM (.72), CPI (.68),
Table 1 (continued)
Previous Studies Using Perry’s (1996) Measurement Scale for PSM
Four-Dimension Model Three-Dimension Model or Others
increases the overall magnitude of PSM, without necessarily affecting the rest of the di-
mensions, PSM should be defined as formative. Different parameter estimates and conclu-
sions can be drawn depending on which measure is used for analysis (Law and Wong 1999;
MacCallum and Browne 1993).
This issue owes not so much to the PSM scale as it was originally constructed but to
advances in measurement theory and methods. Coursey et al. (2008a, 88) wrote, ‘‘The PSM
construct is reflective, not formative, and the studies to date have not tested such a formative
specification.’’ On the other hand, Wright (2008, 85) said, ‘‘Researchers should consider
operationalizing this four-dimension conceptualization as first-order reflective and
526 Journal of Public Administration Research and Theory
second-order formative.’’ Coursey et al. (2008b) examined whether PSM is best repre-
sented as a formative or reflective construct but they exclude APM and used only the three
remaining dimensions. The previous studies have raised this issue but the empirical test
with the four dimensions of PSM has been found wanting. The third research issue is
to discuss a reflective model (the dimensions of PSM function as specific manifestations
of PSM) and a formative model of PSM (the dimensions of PSM are conceived as specific
components, which together collectively constitute PSM) and to examine theoretically and
empirically which model is more desirable in second-order factor specifications. This is
a critical question not only to PSM but also to many other public management constructs
(Coursey and Pandey 2007b).2
The next section of this article will review the previous studies on the dimensions and
measures of PSM. Then both a reflective and a formative model are discussed, and both
models are applied to the PSM construct. We then consider the research sample (n 5 2,497)
as well as the measurement of the variables before discussing the applied methods. The
final section consists of an analysis of the survey results, followed by a discussion and
2 Coursey and Pandey (2007b) used second-order confirmatory factor analysis to test two specifications for red tape,
one as a formative index and the other as a reflective scale.
Kim Reflective versus Formative Specification 527
minimally acceptable Cronbach’s alphas. Castaing (2006) used only the CPI dimension
because other dimensions were not representative of the specific public service ethos of
the French context.
However, there is no discussion whether we can accept the findings of the empirical
tests using only two or three dimensions as those of PSM. If we regard PSM as a reflective
construct, it is possible to analyze PSM with only two or three dimensions because omitting
dimensions does not change the essential nature of PSM. Otherwise, it cannot be acceptable
because omitting a dimension may change the meaning of PSM as a formative construct. The
consequences of dropping a dimension are potentially quite serious in a formative model.
Shorter scales are generally preferred in studies so that respondents’ workload is re-
duced. Abridged versions of Perry’s (1996) scale were used in 10 studies, whereas the full
version was tested in eight studies. Unless the shorter version is a valid and reliable measure
of the construct that the longer scale measures (DeVellis 1991), shortening the scale, could
threaten the integrity of the overall measurement of PSM. Moreover, 7 out of 10 studies
using the short forms of the Perry (1996) scale used two or three dimensions of PSM, and 2
variation in its indicators. Such indicators are termed ‘‘reflective’’ because they represent
reflections, or manifestations, of a construct. In some situations, indicators are viewed as
causes of constructs. Such indicators are termed ‘‘formative’’, meaning that the construct is
formed or induced by its indicators (Edwards and Bagozzi 2000).
For example, transformational leadership can be modeled as having formative indi-
cators (MacKenzie, Podsakoff, and Jarvis 2005). It is conceptualized as being a function of
charisma, idealized influence, inspirational leadership, intellectual stimulation, and indi-
vidualized consideration (Bass 1985). These forms of leader behavior are conceptually
distinct, likely to have different antecedents and/or consequences and are not interchange-
able. On the other hand, leader-member exchange (LMX) can be an example of a reflective
model. LMX theory suggests that leaders do not use the same style in dealing with all
subordinates but rather develop a different type of relationship with each subordinate
(Liden and Maslyn 1998). LMX is an overall latent variable with various dimensions such
as affect, loyalty, contribution, and professional respect, and these dimensions are expected
to be caused by LMX. Changes in LMX are manifested in changes in all the dimensions.
3 Formally, if h is a latent variable and x1, x2, . . ., xn are a set of observable indicators, the reflective specification
implies that
xi 5 lih 1 ei,
where li is the expected effect of h on xi and ei is the measurement error for the ith indicator (i 5 1, 2, . . ., n). It is
assumed that COV(h, ei) 5 0, and COV(ei, ej) 5 0, for i 6¼ j and E(ei) 5 0. In contrast, the formative specification
implies that
h 5 g1x1 1 g2x2 1 . . . 1 gnxn 1 z,
where gi is the expected effect of xi on h and z is an error term, with COV(xi, z) 5 0 and E(z) 5 0 (Diamantopoulos and
Siguaw, 2006). The error term (z) in a formative measurement model represents the impact of all remaining causes
other than those represented by the indicators included in the model. The more comprehensive the set of formative
indicators specified for the construct, the smaller the influence of the error term and the more valid the construct
(Diamantopoulos, Riefler, and Roth 2008).
Kim Reflective versus Formative Specification 529
specific component variables (Bollen 1989; Bollen and Lennox 1991; Edwards and Bagozzi
2000). A construct should be modeled as having formative indicators if the following con-
ditions prevail: (a) the indicators are viewed as defining characteristics of the construct,
(b) changes in the indicators are expected to cause changes in the construct, (c) changes
in the construct are not expected to cause changes in the indicators, (d) the indicators do not
necessarily share a common theme, (e) eliminating an indicator may alter the conceptual
domain of the construct, (f) a change in the value of one of the indicators is not necessarily
expected to be associated with a change in all the other indicators, and (g) the indicators are
not expected to have the same antecedents and consequences. On the other hand, a construct
should be modeled as having reflective indicators if the opposite is true (Jarvis, MacKenzie,
and Podsakoff 2003, 203).
Reflective indicators are essentially interchangeable—and therefore, while adding or
removing indicators may affect reliability, it does not change the essential nature of the
underlying construct (Diamantopoulos and Winklhofer, 2001). In contrast, ‘‘omitting an
indicator is omitting a part of the construct’’ in a formative model (Bollen and Lennox
4 Each dimensions such as APM, CPI, COM, and SS have several reflective indicators based on the following criteria
(Jarvis, MacKenzie, and Podsakoff 2003): the relative homogeneity and interchangeability of indicators pertaining to
each dimension, the high degree of covariation among indicators of each dimension, and the expectation that the
indicators of each dimension are likely to be affected by the same antecedents and have the same consequences.
530 Journal of Public Administration Research and Theory
flow from PSM and its dimensions, PSM is termed superordinate, meaning that PSM
is a general entity that is manifested or reflected by the specific dimensions that serve
as its indicators. Its dimensions function as specific manifestations of PSM. When the re-
lationships flow from the dimensions to PSM, PSM is called aggregate, meaning that PSM
is a composite of its dimensions. The dimensions are conceived as specific components
which collectively constitute PSM (Edwards 2001; Williams, Edwards, and Vandenberg
2003).5
Even though the nature and direction of relationships between PSM and its dimensions
should be clearly specified by theory, and PSM is defined as a superordinate or an aggregate
construct prior to examining substantive relationships between PSM and related constructs,
there was no explicit consideration at the construct definition stage because at that time
multidimensional constructs were scarcely used in this field (Perry 1996; Perry and Wise
1990). Thus, it is still challenging to determine whether to specify PSM as a superordinate
or an aggregate construct. That is, we need to know whether PSM causes the first-order
dimension such as APM, CPI, COM, or SS (reflective model) or all first-order dimensions
5 PSM can be modeled as reflective or formative in second-order factor specifications (Edwards 2001; Jarvis,
MacKenzie, and Podsakoff 2003). The reflective model treats PSM as a superordinate construct that has reflective
dimensions and the dimensions themselves have reflective indicators (reflective first order, reflective second order);
and the formative model treats PSM as an aggregate construct that has formative dimensions and the dimensions
themselves have reflective indicators (reflective first order, formative second order). In the reflective model, the
meaning generally does not alter when dropping a dimension, whereas in the formative model it is necessary to include
all first-order dimensions that form PSM because dropping one may alter the meaning of PSM. Formally, the second-
order measurement model of PSM can be represented as follows:
1) Reflective model: APM 5 l1PSM 1 e1, CPI 5 l2PSM 1 e2, COM 5 l3PSM 1 e3, SS 5 l4PSM 1 e4
2) Formative model: PSM 5 g1APM 1 g2CPI 1 g3COM 1 g4SS 1 z
Kim Reflective versus Formative Specification 531
characteristics and theoretical backgrounds. The dimensions of PSM are not interchange-
able, and they are not expected to have the same antecedents and consequences.6 Thus, the
dimensions of PSM are not reflective but formative.
Each individual’s PSM needs to be analyzed and evaluated by all dimensions of PSM.
That is, the dimensions of PSM combine to produce PSM. It can be defined as a linear sum
of its dimensions, and so a formative composite in the form of indexes is conceptually
appropriate. So PSM needs to be defined as an aggregate construct. It is necessary to include
all first-order dimensions that form PSM because omitting one may alter the meaning of
PSM.7 This consideration of the relationships between PSM and its dimensions shows that
PSM is formative in nature. The next step is to analyze whether the formative model has
more desirable statistical properties than the reflective model because when constructing
a measure, it needs to reconcile the theory-driven conceptualization of the measure with the
desired statistical properties of the items comprising the measure as revealed by empirical
testing (Diamantopoulos and Siguaw 2006). The two models are identical except for the
direction of causality between PSM and its dimensions.
6 Perry (1997) found some significant differences in the influence of independent variables on the four different
dimensions of PSM. Taylor (2007) confirmed that when the multiple dimensions of PSM are analyzed simultaneously,
certain dimensions are found to be more important than others in influencing work outcomes. Moynihan and Pandey
(2007a) reported that red tape is negatively and significantly related to APM, but not to CPI, whereas reform orientation
and hierarchical authority is positively related to CPI, but not to APM.
7 That is, PSM 5 g1APM 1 g2CPI 1 g3COM 1 g4SS 1 z 6¼ g1APM 1 g2CPI 1 g3COM 1 z (see footnote 5).
8 This survey was conducted as a part of the Survey on Firefighters’ Working Conditions supported by Gyonggi
Provincial Firefighting and Disaster Headquarters in Korea. Because the firefighters were very concerned about their
working conditions, they actively participated in the survey process.
9 The population for this study comprised all firefighters in central and local governments. The number of the
population is 30,158, in which men were 94.9% and women 5.1% (National Emergency Management Agency
[NEMA], Korea, 2008). Most (45.2%) were in their 30s, 33.4% were in their 40s, and 10.6% in their 20s. Those in the
largest group (46.7%) had worked for more than 10 years and fewer than 20 years, and 39.1% had worked for fewer
than 10 years. Those at the managerial levels were 16.7%.
532 Journal of Public Administration Research and Theory
Table 2
Background of Respondents (n 5 2,497)
Variables Characteristics Respondents (%)
Sex Male 93.1
Female 6.9
Age 20s 14.1
30s 47.0
40s 29.7
50s 9.2
Length of service (years) 0–5 33.7
5–10 15.2
10–15 22.5
15–20 15.5
20–25 8.2
251 4.9
Education High school diploma or under 31.9
Measure
All the indicators described below allowed responses on a five-point Likert-type scale (1 5
strong disagreement, 5 5 strong agreement). PSM was measured using 12 items: 3 items for
each of the dimensions of APM, CPI, COM, and SS. Perry (1996) developed a list of 24
items measuring 4 dimensions of PSM, but the APM items in Perry’s scale may not be
appropriate to represent the rational base of PSM in the Korean context (Kim 2009b). Ac-
cording to Perry and Wise (1990), the rational base is understood as that individuals may be
drawn to government or to pursue particular courses of action within government because
of a belief that their choices will facilitate the interests of special groups and that one motive
prevalent in pluralistic societies is an individual’s advocacy for special interests. However,
the original APM items do not ask whether the respondents are more attracted to partic-
ipation in the process of policy formulation or in advocating specific public programs.
Thus, the 3 items of APM were replaced with more representative ones, and the 12-item
scale for the 4 factors was proposed. The test results provided support for convergent val-
idity as well as discriminant validity of the four-factor model, and the reliability coefficients
of all subscales were good (Kim 2009a). Among the items, eight are exactly same as those
originally specified by Perry (1996), whereas one in the COM dimension replaces a reverse-
scored item of Perry’s (1996) study. Table 3 sets out the 12 items in list form.
A formative measurement model, in isolation, is under-identified and cannot be es-
timated (Bollen 1989). Consequently, in this study, two reflectively measured constructs,
job satisfaction, and organizational commitment, are added in empirical analysis as
Kim Reflective versus Formative Specification 533
outcome variables for solving a problem of underidentification.10 Job satisfaction was mea-
sured with three items. The sample items are ‘‘My job provides a chance to do challenging
and interesting work’’ and ‘‘I feel good about my job, the kind of work I do.’’ The reliability
coefficient was .783. Organizational commitment was measured with five items from the
Organizational Commitment Questionnaire (Mowday, Steers, and Porter 1979). The sam-
ple questions are ‘‘I talk up this organization to my friends as a great organization to work
for’’ and ‘‘I find that my values and the organization’s values are very similar.’’ The
reliability coefficient was .865.
Analysis
The statistical analysis applies structural equation modeling (SEM) using Amos 17.0 with
the asymptotically distribution-free (ADF) estimation method.11 After testing the four-
dimension measurement model of PSM, I conducted a series of CFAs to compare with
three-factor models. Then I compared a formative measurement model with a reflective
one. The only difference between the two models is the direction between PSM and its
10 In order to get the necessary conditions for the identification of formative indicator constructs,
(1) the scale of measurement for the latent construct is established by constraining a path from
one of the construct’s indicators to be equal to 1 or by constraining the residual error variance for
the construct to be equal to 1 and
(2) to resolve the indeterminacy associated with the construct level error term, a formative
construct emits paths to
c) one reflective indicator and one latent construct with reflective indicators (Jarvis, MacKenzie,
and Podsakoff 2003; MacCallum and Browne 1993; MacKenzie, Podsakoff, and Jarvis 2005).
11 The discrete, noncontinuous distributions are not suitable for standard maximum likelihood estimation (MLE) in
confirmatory factor analysis. Applying standard MLE in such cases produces significant estimation problems, such as
inflation of chi-square fit statistics and biased underestimation of parameters and SEs. One preferred approach for
Likert-type items is to apply weighted least squares (WLS) or robust WLS (see Coursey and Pandey 2007b, footnote 2).
WLS performs adequately with large sample sizes. WLS is known as the ADF estimation method when used with
a correct asymptotic covariance matrix (Flora and Curran 2004).
12 The composite reliability estimates the extent to which a set of latent construct indicators share in their
measurement of a construct, whereas Cronbach’s alpha measures how well a set of variables or items measures a single,
unidimensional latent construct. Composite reliability 5 (sum of standardized loading)2/{(sum of standardized
loading)2 1 sum of indicator measurement error}. Indicator measurement error 5 1 2 the square of each standardized
loading.
13 AVE indicates the amount of variance captured by the construct in relation to the variance due to measurement
error, and AVEs above 0.50 are treated as indications of convergent validity. AVE 5 (sum of squared standardized
loading)/(sum of squared standardized loading 1 sum of indicator measurement error).
534 Journal of Public Administration Research and Theory
Table 3
Descriptive and Measurement Statistics (n 5 2,497)
APM
PSM1: I am interested in making public programs that 3.57 (0.812) .834 .658
are beneficial for my country or the community I
belong to.
PSM2: Sharing my views on public policies with 3.51 (0.827) .845 .714
others is attractive to me.
PSM3: Seeing people get benefits from the public 3.79 (0.782) .811 .695
program I have been deeply involved in brings me
a great deal of satisfaction.
Commitment to the public interest
PSM4: I consider public service my civic duty. 4.02 (0.743) .858 .737
PSM5: Meaningful public service is very important 4.00 (0.741) .882 .778
to me.
around the correlation estimate between two factors included 1.00 (Anderson and Gerbing
1988). The methods traditionally used for assessing construct reliability and validity are not
appropriate for a formative model because the direction of causality is posited to flow from
the dimensions to PSM (Jarvis, MacKenzie, and Podsakoff 2003). Instead, indicator
validity (Bollen 1989) and criterion validity or predictive validity (Diamantopoulos and
Siguaw 2006) will be analyzed.
To get the necessary conditions for the identification of formative indicator constructs
(Jarvis, MacKenzie, and Podsakoff 2003; MacCallum and Browne 1993; MacKenzie,
Kim Reflective versus Formative Specification 535
Podsakoff, and Jarvis 2005), the residual error variance for PSM is constrained to unity and
two unrelated latent constructs with reflective indicators are added. The two output con-
structs are job satisfaction and organizational commitment. Theoretical relationships can be
postulated to exist between PSM and job satisfaction (Kim 2005a, 2006; Liu, Tang, and Zhu
2008; Naff and Crum 1999; Park and Rainey 2008; Steijn 2008; Taylor 2008; Wright and
Pandey 2008) and between PSM and organizational commitment (Castaing 2006; Kim
2005a, 2006).
For model fit comparison, goodness-of-fit index (GFI) and the root mean square error of
approximation (RMSEA) are used because the Monte Carlo simulations showed that
RMSEA and GFI among various goodness-of-fit indices are better for detecting measurement
model misspecification (Jarvis, MacKenzie, and Podsakoff 2003; MacKenzie, Podsakoff,
and Jarvis 2005).14 The model achieves an acceptable fit to the data when GFI equals or
exceeds 0.90, and RMSEA values fall below 0.05 (Hu and Bentler 1999). In general, the
larger the value of GFI and the smaller the value of RMSEA, the better the fit of the model
(Bollen 1989). The proper specification of the direction of causality between PSM and
RESULTS
The four-correlated factor model with 12 items of PSM was tested using CFA, which hy-
pothesized a priori that (a) responses to the 12-item PSM scale could be explained by four
factors; (b) each item would have a nonzero loading on the PSM factor it was designed to
measure and zero loadings on all other factors; (c) the four factors, consistent with the
theory, would be correlated; and (d) measurement error terms would be uncorrelated
(Byrne 2001). The resulting CFA for the sample (n 5 2,497) shows that it had an acceptable
fit to the data, x2 (degrees of freedom [df] 5 48) 5 334.6, p , .001; GFI 5 0.931,
RMSEA 5 0.049. As table 3 shows, both Cronbach’s coefficient alphas and the composite
reliability of the set of reflective indicators for each dimension of PSM exceed .80. The
resulting factor structure shows a clean four-factor structure with all items loading signif-
icantly onto their a priori dimension (p , .001), and the standardized factor loadings rang-
ing from 0.700 to 0.845. In each factor, AVE exceeds 0.50. The results provide support for
14 The chi-square is also able to detect the measurement model misspecification, but it needs to be discounted because
lower values of chi-square indicate a better fit and should be nonsignificant, but for large sample sizes, this statistic may
lead to rejection of a model with a good fit (Jarvis, MacKenzie, and Podsakoff 2003).
536 Journal of Public Administration Research and Theory
Table 4
Model Fit Indices for Different Dimension Scales
Model x2, (df), p value GFI RMSEA
(1) Three-factor model #1 (APM, CPI 1 COM, SS) 481.6, (51), .000 0.901 0.058 [0.053, 0.063]
(2) Three-factor model #2 (APM 1 SS, CPI, COM) 521.2, (51), .000 0.892 0.061 [0.056, 0.066]
(3) Three-factor model #3 (APM, CPI 1 SS, COM) 678.9, (51), .000 0.860 0.070 [0.066, 0.075]
(4) Three-factor model #4 (APM, CPI, COM 1 SS) 714.3, (51), .000 0.852 0.072 [0.068, 0.077]
(5) Four-factor model 334.6, (48), .000 0.931 0.049 [0.044, 0.054]
Note: Dx 2(3) between model (5) and any three-factor models is significant at p , .001.
convergent validity. The correlation estimates between the two factors range from .586
to .835, and the confidence intervals (62 standard errors [SEs]) around the correlation
estimates between the two factors do not include 1.00. The result provides support for
discriminant validity.
In the APM dimension of the four-correlated factor model, all factor loadings of the
Figure 1
A Reflective Model of PSM with JS and QC
Figure 2
A Formative Model of PSM with JS and QC
Naff and Crum 1999; Park and Rainey 2008; Steijn 2008; Taylor 2008; Wright and Pandey
2008), ‘‘job satisfaction appears primarily to be a function of an individual’s unique wants
and expectations’’ (Pandey and Stazyk 2008, 112). For firefighters, these wants and expect-
ations may be linked to a desire to help individuals and to contribute to society through
protecting life and property from fires and providing relevant services to communities. Fire-
fighters may choose their occupation to realize these desires, and so they are committed to
the honorable profession and the organization that impose the role on them. They can re-
alize their wants and expectations through working as firefighters in fire stations and doing
the job, fighting fires and taking care of communities (Lee and Olshfski 2002). Thus, PSM
is an important individual predisposition, which explains the job satisfaction and organi-
zational commitment of firefighters (Castaing 2006).
The model fit indices show that the formative model (x2 5 432.3, df 5 149, p , .001,
GFI 5 0.945, RMSEA 5 0.028) fits the data better than the reflective model (x2 5 800.5,
df 5 155, p , .001, GFI 5 0.898, RMSEA 5 0.041). Respecification of dimensions from
reflective to formative results in an improved model fit (DGFI 5 0.047, DRMSEA 5
15 The degree of the effects of common method variance can be evaluated by calculating each indicator’s variances
substantively explained by the principal construct and by the method (Liang, Saraf, Hu, and Xue 2007; Williams, Cote,
and Buckley 1989). In the formative model after controlling the effects of common method variance, the average
substantively explained variance of the indicators is 0.585, whereas the average method-based variance is 0.040. In
addition, most method factor loadings are not significant. In the reflective model after controlling the effects of
common method variance, the average proportion of variance attributed to the measurement factors is higher than the
average proportion of variance attributed to the method factor (0.520 vs. 0.120). These provide some evidence that
substantive relationships, and not merely common method bias, are likely responsible for the observed findings.
16 One may argue that the reflective model may be a better comparable fit because every path is highly significant for
the reflective specification, whereas there is one insignificant path under the formative model. However, the
standardized path estimates in the reflective model cannot be directly compared with those in the formative model
because the directions of paths are different between the two models (see footnotes 3 and 5).
540 Journal of Public Administration Research and Theory
Table 5
Test Results for Formative and Reflective Models (n 5 2,497)
Formative Model Reflective Model
Path Standardized Estimate SMC (R2) Standardized Estimate SMC (R2)
Models without controlling for the effects of common method variance
PSM / JS .671*** .450 .629*** .396
z Value for differences in R2: 2.32*
PSM / OC .987*** .973 .851*** .724
z Value for differences in R2: 31.28***
APM / PSM .132*** .423
CPI / PSM .244***
COM / PSM 2.038
SS / PSM .366***
PSM / APM .821*** .674
PSM / CPI .883*** .779
PSM / COM .785*** .616
DISCUSSION
Specification of the measurement model is a critical decision that needs to be made on the
basis of conceptual criteria (MacKenzie, Podsakoff, and Jarvis 2005). Measurement model
misspecification can bias structural parameter estimates and result in errors of inference.
The theoretical consideration of the relationships between PSM and its dimensions shows
that PSM is an aggregate construct. The empirical testing also shows that the formative
model has more desirable statistical properties than the reflective model. First, the fit in-
dices show that the formative model provides a better fit than the reflective model. The poor
fit experienced in covariance structure models may indicate that the wrong type of mea-
surement model has been applied (Diamantopoulos and Winklhofer 2001). Second, the
formative model significantly outperforms reflective measures in terms of predictive
Kim Reflective versus Formative Specification 541
validity. ‘‘While reliable and valid measurement provides a necessary foundation upon
which to test causal claims, causal tests (predictive validity) can provide some of the stron-
gest evidence that the existence of a construct is being properly measured’’ (Wright 2008,
80). This study provides empirical evidence in support of a second-order, formative ap-
proach to PSM. Therefore, both theoretical and empirical considerations suggest that
the formative model is more plausible than the reflective one. Thus, it is more reasonable
to define PSM as a formative construct, as Wright (2008, 85) said: ‘‘Researchers should
consider operationalizing this four-dimension conceptualization as first-order reflective
and second-order formative.’’
Although it is appropriate to measure PSM as a formative second-order factor spec-
ification, a problem is the dimension of COM. Table 5 shows a negative and insignificant
effect of COM on PSM. PSM has rational, norm-based, and affective motives (Perry and
Wise 1990). Affective motives, grounded in human emotion, are commitment to a program
from a genuine conviction about its social importance and the patriotism of benevolence.
The dimension of COM should coincide with affective motives because each dimension
(Hair et al. 2006, 140), and ‘‘a well-developed summated rating scale can have good re-
liability and validity’’ (Spector 1992, 2). When developing the measure of PSM using SEM
approaches, once items have been selected following dimensionality and reliability tests
and dimensions have been confirmed, it is practical to combine them to generate overall
measures of PSM (Diamantopoulos and Siguaw 2006). Overall indexing PSM by measur-
ing all the dimensions of PSM and simply summing the averages of each dimension can
make it easier to measure PSM in practice.
CONCLUSION
The purposes of this study were to confirm the revised APM items that were developed
for replacing the original items that had little face validity as indicators of a rational base
of PSM, to compare a four-factor model with three-factor models because the dimensionality
of PSM is still controversial, and to analyze whether PSM is formative or reflective in
nature. Using survey responses from 2,497 firefighters and a revised 12-item measure of
analyzing the full dimensions of PSM. The third contribution is the raising of the question
about the measurement and fit of the COM dimension. It will help to make researchers of
PSM aware of the difficulties of measuring the concept and offers an alternative lens for
conceptualizing and measuring PSM.
This study has several limitations. First, this study used exactly the same set of items
and dimensions for comparing reflective and formative specification. It was assumed that
the only difference resulting from applying the formative versus reflective approach relates
to the direction of the relationship between PSM and its dimensions (Diamantopoulos and
Siguaw 2006). However, items and dimensions might be developed differently if PSM is
defined as an aggregate construct at the construct definition stage. It could not compara-
tively test both a formative model and a reflective model that are intentionally and inde-
pendently developed from the concept of PSM. Second, this study used a shortened and
adapted version of Perry’s (1996) instrument instead of the original one because the revised
12-item scale was shown as a valid and reliable measure in the previous study and respond-
ents’ workload is reduced with shorter scale (Kim 2009a). The development of a measure-
REFERENCES
Anderson, James C., and David D. Gerbing. 1988. Structural equation modeling in practice: A review and
recommended two-step approach. Psychological Bulletin 103:411–23.
Bass, Bernard M. 1985. Leadership and performance beyond expectations. New York: Free Press.
Bollen, Kenneth A. 1989. Structural equations with latent variables. New York: Wiley.
Bollen, Kenneth A., and Richard D. Lennox. 1991. Conventional wisdom on measurement: A structural
equation perspective. Psychological Bulletin 110:305–14.
Bollen, Kenneth A., Richard D. Lennox, and Darren L. Dahly. 2009. Practical application of the vanishing
tetrad test for causal indicator measurement models: An example from health-related quality of life.
Statistics in Medicine 28:1524–36.
Brewer, Gene A., and Sally Coleman Selden. 1998. Whistle blowers in the federal civil service: New
evidence of the public service ethic. Journal of Public Administration Research and Theory 8:413–39.
Bright, Leonard. 2008. Does public service motivation really make a difference on the job satisfaction and
turnover intentions of public employees? American Review of Public Administration 38:149–66.
Byrne, Barbara M. 2001. Structural equation modeling with amos: Basic concepts, applications, and
programming. Mahwah, NJ: Erlbaum.
Camilleri, Emanuel. 2006. Towards developing an organizational commitment—Public service moti-
vation model for Maltese public service employees. Public Policy and Administration 21:63–83.
Castaing, Sébastien. 2006. The effects of psychological contract fulfillment and public service motivation on
organizational commitment in the French civil service. Public Policy and Administration 21:84–98.
544 Journal of Public Administration Research and Theory
Choi, Do Lim. 2004. Public service motivation and ethical conduct. International Review of Public
Administration 8:99–106.
Clerkin, Richard M., Sharon R. Paynter, and Jami Kathleen Taylor. 2009. Public service motivation in
undergraduate giving and volunteering decisions. American Review of Public Administration
39:675–98.
Collier, Joel E., and Carol C. Bienstock. 2006. Measuring service quality in E-retailing. Journal of Service
Research 8:260–75.
Coursey, David H., and Sanjay K. Pandey. 2007a. Public service motivation measurement: Testing an
abridged version of Perry’s proposed scale. Administration & Society 39:547–68.
———. 2007b. Content domain, measurement, and validity of the Red Tape concept. American Review of
Public Administration 37:342–61.
Coursey, David H., James L. Perry, Jeffrey L. Brudney, and Laura Littlepage. 2008a. Psychometric
verification of Perry’s public service motivation instrument: Results for volunteer exemplars. Review
of Public Personnel Administration 28:79–90.
Coursey, David H., Jeffrey L. Brudney, James L. Perry, and Laura Littlepage. 2008b. Measurement
questions in public service motivation: Construct formation and nomological distinctiveness and
explanatory power for volunteering activities. Paper presented for the Minnowbrook III Conference,
———. 2006. Public service motivation and organizational citizenship behavior in Korea. International
Journal of Manpower 27:722–40.
———. 2009a. Revising Perry’s measurement scale of public service motivation. American Review of
Public Administration 39:149–63.
———. 2009b. Testing the structure of public service motivation in Korea: A research note. Journal of
Public Administration Research and Theory 19:839–51.
Kline, Rex B. 2004. Principles and practices of structural equation modeling, 2nd ed. New York: Guilford.
Law, Kenneth, and Chi-Sum Wong. 1999. Multidimensional constructs in structural equation analysis: An
illustration using the job perception and job satisfaction constructs. Journal of Management
25:143–60.
Lee, Geunjoo. 2005. PSM and public employees’ work performance. Korean Society and Public
Administration 16:81–104.
Lee, Seok-Hwan, and Dorothy Olshfski. 2002. Employee commitment and firefighters: It’s my job. Public
Administration Review 62:108–14.
Leisink, Peter, and Bram Steijn. 2009. Public service motivation and job performance of public sector
employees in the Netherlands. International Review of Administrative Sciences 75:35–52.
Liang, Huigang, Nilesh Saraf, Qing Hu, and Yajiong Xue. 2007. Assimilation of enterprise systems: The
Perry, James L., Jeffrey L. Brudney, David Coursey, and Laura Littlepage. 2008. What drives morally
committed citizens? A study of the antecedents of public service motivation. Public Administration
Review 68:445–58.
Perry, James L., and Annie Hondeghem. 2008. Editor’s introduction. In Motivation in public management,
ed. James L. Perry and Annie Hondeghem, 1–14. Oxford: Oxford Univ. Press.
Perry, James L., and Lyman W. Porter. 1982. Factors affecting the context for motivation in public
organizations. Academy of Management Review 7:89–98.
Perry, James L., and Lois R. Wise. 1990. The motivational bases of public service. Public Administration
Review 50:367–73.
Podsakoff, Philip M., Scott B. MacKenzie, Jeong-Yeon Lee, and Nathan P. Podsakoff. 2003a. Common
method biases in behavioral research: A critical review of the literature and recommended remedies.
Journal of Applied Psychology 88:879–903.
Podsakoff, Philip M., Scott B. MacKenzie, Nathan P. Podsakoff, and Jeong-Yeon Lee. 2003b. The
mismeasure of man(agement) and its implications for leadership research. Leadership Quarterly
14:615–56.
Rainey, Hal G., and Paula Steinbauer. 1999. Galloping elephants: Developing elements of a theory of
effective government organizations. Journal of Public Administration Research and Theory 9:1–32.