This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
 Author's personal copy

Journal of Business Research 67 (2014) 403406

Contents lists available at ScienceDirect

Journal of Business Research

Construct measurement in management research: The importance of match between

levels of theory and measurement
Bo Bernhard Nielsen
Copenhagen Business School, Department of Strategic Management and Globalization, Kilen 14, 2nd oor, 2000 Frederiksberg, Denmark
University of Technology Sydney, School of Marketing, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Far too often do management scholars resort to crude and often inappropriate measures of fundamental con-
Received 1 October 2012 structs in their research; an approach which calls in question the interpretation and validity of their ndings.
Received in revised form 1 December 2012 Scholars often legitimize poor choices in measurement with a lack of availability of better measures and/or
Accepted 1 December 2012
that they are simply following existing research in adopting previously published measures without critically
Available online 5 January 2013
assessing the validity, appropriateness, and applicability of such measures in terms of the focal study. Motivated
Keywords:
by a recent dialog in Journal of Business Research, this research note raises important questions about the use
Construct measurement of proxies in management research and argues for greater care in operationalizing constructs with particular
Management research attention to matching levels of theory and measurement.
JV relatedness 2012 Elsevier Inc. All rights reserved.
Multilevel research
SIC code
Alliance termination

1. Introduction
Scholars seeking to investigate important research questions
and develop new theories often face challenges with respect to the
measurement of the constructs that are pivotal to their frameworks.
With expectation to test empirically novel theoretical frameworks,
researchers employ a variety of data collection approaches that may
compromise construct validity. One frequently used approach in-
volves approximation, which enables the researcher to investigate
important research questions at the expense of precise measures.
It is often argued that approximation is reasonable given the data
constraints and that issues of lack of precise measures should not pre-
vent us from investigating important research questions (e.g., Cui
& Kumar, 2012b). In this commentary, I argue that data limitations
should prevent us from analyzing research questions (testing theory)
for which we do not have appropriate data or measures; rather this
limitation than running the risk of providing evidence of key rela-
tionships that may not hold up to closer empirical scrutiny.
Cui and Kumar (2012a) follow previous research and use a 2-digit
SIC code as proxy for JV relatedness in order to investigate to what
extent alliance termination differs among related and unrelated
joint ventures. Essentially, their approach is a contingency approach
The author thanks Jacob Lyngsie, Copenhagen Business School, and Stewart Miller,
University of Texas at San Antonio for their valuable comments and suggestions. Any
and all errors are the sole responsibility of the author.
E-mail address: bn.smg@cbs.dk.
which species a set of factors at different levels that may inuence
the termination rate of JVs. Chief among these variables is JV related-
ness and they advance the argument that the impact of various factors
(at multiple theoretical levels) on termination vary between related
and unrelated JVs. The underlying theoretical logic is that the evolu-
tion (and thus probability of termination) of related and unrelated
JVs is likely to differ because of important differences in the motives
of the rms to form such JVs.
As a commentary (Nielsen, 2012) to Cui and Kumar (2012a) notes,
however, industry-level SIC codes say very little about the underlying
motives behind alliance formation, nor do they approximate well
rm-level resources and capabilities. Yet, the authors argue in their
rejoinder (Cui & Kumar, 2012b) to the commentary that SIC codes
provide a good proxy for rm resources and capabilities because
businesses in the same industry tend to have similar assets, opera-
tions, as well as similar intangible resources. However, using a
2-digit industry level SIC code to proxy rm intangible resources
and capabilities is highly problematic as such industry overlap says
absolutely nothing about rm level operations, efciencies, proce-
dures, knowledge, experience, let alone assets or motivation behind
JV formation. Moreover, within industries, there is likely to be vari-
ance between rms in terms of resources, capabilities and strategic
motives for JV formation. This example, which is by no means unique
in management research (see Boyd, Gove, & Hitt, 2005), raises a num-
ber of fundamental questions about measurement in management
research, the importance of matching the levels of theory and mea-
surement, and the utility of previously established measures.

0148-2963/$ see front matter 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jbusres.2012.12.020
 Author's personal copy
404 B.B. Nielsen / Journal of Business Research 67 (2014) 403406
2. Construct measurement in management research
Unobserved constructs (such as capabilities or managerial processes)
lie at the core of management phenomena, which puts a premium on
the researcher's ability to develop sound strategies for operationalizing
and testing constructs that are unobservable (Godfrey & Hill, 1995).
More fundamentally, I suggest that management scholars should strive
harder to ensure construct validity of their measures and seek to
acquire direct measures rather than relying on poor proxies when
examining complex latent constructs. Proxies for variables are often
seemingly selected without concern for their reliability or validity
and future researchers must employ measurement strategies, such as
cross-validation, in order to establish that a model's constructs are
appropriately measured.
The use of dichotomous variables is an extreme example of how
simple measures are used to proxy complex relationships in manage-
ment research. For instance, prior experience with the partner is
often coded as a dummy based on any type of previous relationships
between parties and theorized to reect relational quality between
partners. Yet, such binary variables are imperfect measures of the
nature (type of transaction) and quality (positive or negative) of in-
teraction between JV partners and thus fail to capture the underlying
theoretical construct. Thus, using dichotomous variables as crude
representations of more complex relations allow for little variance
which likely biases results.
When constructs are complex and multi-faceted, multi-dimensional
construct measurement is typically warranted in order to improve
validity and t with theory. A single measure provides no structure
on which to evaluate construct validity. That is not to say that single
measures cannot be valid, however, one must establish such validity
by showing that results obtained from a single measures would be the
same if other measures within the domain were used, or show that
indeed a single measure captures the underlying construct better than
a host of ambiguous items (for an example of trust in alliances, see
Nielsen & Nielsen, 2009). Notwithstanding, combining several mea-
sures provides greater construct validity and, importantly, improves
generalizability relative to using a single measure.
Construct measurement is of pivotal importance if we are to
advance management research and scholars must (at a minimum)
demonstrate that (1) measures employed plausibly capture the theo-
retical constructs and (2) theoretical and empirical levels of analysis
for the proposed construct match (Lawrence, 1997).
3. Levels of theory and measurement
Researchers need to be much more watchful of potential mis-
matches between theory and measurement when operationalizing
difcult-to-measure management phenomena; levels-of-analysis
ambiguity may seriously misrepresent the relationships a researcher
would have found if data had been collected and analyzed at the
same level as the theory (Klein, Dansereau, & Hall, 1994; Lawrence,
1997). Level of theory refers to the focal unit or target (e.g., rm or
alliance) that a researcher aims to explain; it is the level to which
generalizations are made (Rousseau, 1985: 4). The focal unit, in turn,
determines the appropriate level associated with the key constructs of
interests in a study (e.g., JV relatedness is at the JV level). Explicit atten-
tion must be paid to the level of theory (and measurement) because
neglecting to do so increases the risks of falsely attributing effects at
one level (e.g., industry level) to another (e.g., rm or JV level) and
thus committing fallacies of the wrong level (Rousseau, 1985: 5).
Related, level of measurement refers to the actual source of the data
from which inferences are made; in other words the entities from
which the data are drawn or to which the data is attached (e.g., rms,
dyads and industries). The level of measurement should correspond to
the level of constructs specied in the focal unit in order to increase
the variability predicted by the theory.
For instance, if the theory species within-group homogeneity
(e.g., rms within industries are similar in terms of motives for
forming JVs), data collection should be conducted at the rm level in
order to assess whether such homogeneity exists. However, theories
like RBV assert that rms differ in terms of resources and capabilities,
and that this heterogeneity determines variation in rm strategy
and performance even within the same industry. Such rm-level
heterogeneity, in turn, is likely to inuence the motivational intend un-
derlying JV formation and, as a result, the probability of JV termination.
In other words, measuring JV-rm relatedness at the industry level
assumes no rm heterogeneity within the same 2-digit SIC code,
yet such heterogeneity is clearly possible (indeed likely according
to theory) as not all rms with similar 2-digit SIC codes are similar
(homogenous) in terms of resources, capabilities or motivational in-
tend. Testing such theories is best accomplished by (a) using measures
that (like the theory) highlight the position of each individual rm
relative to the JV and by (b) maximizing within-group variability
(Nielsen, 2010).
Before higher-level (e.g., industry) data are used to measure
lower-level (e.g., JV or rm) phenomena, psychometric evidence of
the suitability of using the data in this way must be demonstrated. Dif-
ferent types of validity evidence exist depending on the nature of the
constructs in question (see Chan, 1998; Chen, Mathieu, and Bliese,
2004). More generally, however, researchers need to be explicit about
how data collected at one level of analysis is related to constructs at a
lower/higher level of analysis (see also Klein & Kozlowski, 2000). Here
it should be noted, that simply arguing that more appropriate measures
were difcult to obtain and that this reects the constraints that a re-
searcher typically faces when conducting research (Cui & Kumar,
2012b) is a poor proxy for rigorous and valid research. If indeed theory
(and data) warrants multilevel treatment, then researchers need to ac-
count for this in adequate ways both theoretically and empirically.
Building multilevel theories and measuring and testing them according-
ly is difcult but necessary if one wishes to answer challenging and in-
teresting research questions. Conceptualizing multilevel yet building
and testing a single-level model without paying due attention to the
levels of theory and measurement in the study design may mislead
the reader and lead to erroneous conclusions due to level-related con-
founds (Klein & Kozlowski, 2000; Nielsen, 2010).
4. The utility of existing measures
Moreover, as is often the practice, it can be dangerous to follow
existing research in their operationalization of measures since legiti-
mizing the use of a potentially poor measure by referencing its previ-
ous use in the literature does not ensure its appropriateness or indeed
validity. Poor measurement oftentimes appears to be the result of
relying on previously published work, regardless of the quality and
purpose of that measurement approach, and may be responsible for
the persistence and pervasive reliance on poor measures in a particu-
lar eld (Boyd et al., 2005). To be sure, adopting existing measures can
be very valuable; however, doing so without assessing the applicabil-
ity and meaning of such measures may lead to poor match between
theory and measurement. Meaningfulness depends upon context
and thus researchers must take extra steps to ensure validity when
utilizing existing measures in different contextual settings. Hence,
although it is normally thought to lend credibility and legitimacy to a
study that measures (and particularly proxies) have been established
previously in the literature, we, as researchers, have to be more critical
of the way we adopt and apply such measures in our work.
First of all, be careful in adopting existing measures at face value
without evaluating to what extent the original study was within the
same domain theoretically and empirically (e.g., levels of theory and
measurement). For instance, while a number of scholars have used
industry similarity measured by SIC codes to infer business related-
ness, the fact remains that industry overlap (in terms of 2 or even
 Author's personal copy
B.B. Nielsen / Journal of Business Research 67 (2014) 403406
4-digit SIC codes) says very little about underlying motivation behind
JV formation as well as the complexity of managing such JVs as a func-
tion of rm-level resources and capabilities. In order to more accu-
rately capture the actual inuence of a particular unobserved
theoretical construct, one must strive to directly measure it. For in-
stance, in the case of JV relatedness, studies have developed direct
measures of complementarity, operationalized as competence sim-
ilarity (e.g., Nielsen, 2007) or resource overlap (e.g., Hill &
Hellriegel, 1994; Suarez & Garca-Canal, 2003). The question here
is not so much about the transitive nature of the construct as it is a
basic issue of mismatch between theory and measurement in the
rst place. In the context of the Cui and Kumar (2012a) study of JV
termination such mismatch is particularly important given that the
differences in the probability of JV termination is theorized to be
contingent upon the relatedness of the JV to the (focal) parent rm,
where relatedness is thought to relate directly to the motive behind
the JV as well as rm tangible and intangible resources. Clearly, a
crude industry-level measure of overlap constitutes a mismatch
with the underlying theory.
5. Conclusion
Theories have two equally important components; (1) the mea-
surement component that dictates what constructs are to be mea-
sured and how, and (2) the structural component that describes the
properties of the resulting measures in terms of how constructs inter-
relate. Unfortunately, management research seems to be predomi-
nantly preoccupied with the latter often at the expense of the
former. It is strongly recommended that scholars pay more attention
to measurement issues in general, and construct validity, appropri-
ateness of measures across contexts, as well as the matching of levels
of measurement to theory in particular. Indeed, structural equation
modeling techniques, which are often used in management research
when dealing with latent constructs, advocate a two-stage procedure,
where the researchers rst must establish a good measurement
model (via conrmatory factor analysis) before moving on to testing
relationships in the structural regression model (Anderson & Gerbing,
1988). The strong focus on the measurement instrument (and mea-
surement error) is a particular strength associated with structural
equation modeling which should be used more often before moving
on to hypothesis testing via regression.
As management advances as a scientic eld, we need to increase
the criteria for validity of measures; for instance, while it was tradi-
tionally (up until the 1990s) thought appropriate to mix levels of the-
ory and measurement and to use simple proxies for certain difcult-
to-measure variables, such as R&D intensity as proxy for absorptive
capacity, we now must ask a higher standard of management scholars
in their attempt to capture these often latent constructs.
While one can certainly appreciate the difculty associated with
obtaining and validating critical measures of management phenome-
na, we cannot as a eld be satised with the use of poor proxies;
rather, management scholars must recognize the crucial importance
of construct operationalization and measurement as a determinant
of the scientic validity of their study. Just as convenience sampling
is unacceptable for a number of reasons, so too are poor approxima-
tions due to the lack of data availability or legitimized through prior
use. This does not, however, free us from the inherent dangers of
equating measurability of a construct with its relevance in explanation
and treating as important only that which is readily accessible to mea-
surement (Hayek, 1989). Yet, as noted by Lawrence (1997), the use
of proxies moves the researcher farther and farther away, both em-
pirically and theoretically, from the actual mechanisms underlying
observed relationships.
Management as a science invariably involves functional relations
among measured variables, and thus management research can prog-
ress no faster than the measurement of its key variables. Future
405
management research must raise the bar of measurement and avoid
falling into the trap of applying existing proxies for critical constructs.
Instead, researchers must pay particular attention to consistency
between the levels of theory and measurement and appropriateness
of operationalization of important research constructs in order to
increase validity and interpretability of results.
In practice, this entails that researchers should increase the con-
cern for reliability and construct validity of their measures in order
to improve the quality of measurement. One way of doing so is to
use multiple rather than single indicators for specic constructs;
in particular complex and unobserved constructs. The use of multi-
ple items or indicators allows the evaluation of the reliability
of the measure. While use of single item indicators (proxies) is a
common practice with archival data, such research is not exempt
from the requirements of validation and reliability. The use of
proxies implicitly emphasizes measurement reliability over con-
struct validity yet such choices must be explicitly stated and, more
importantly, justied. Indeed, future management research should
seek to triangulate between data sources (be they archival or
survey-based) in order to ensure the reliability and validity of the
construct measures used. In addition, constructs must be measured
at the appropriate level in order to ensure conformity of the data
to the level of theory; doing so increases their predictive power
(Nielsen, 2010). Moreover, crude proxies should be avoided unless
strictly necessary (e.g., in early exploratory research) as the value
of a theory is bound by its measurement; good theories work pre-
cisely because of their close connection to empirical observation
(Lawrence, 1997).
Finally, despite the fact that relying on past research for measures
provides opportunities for cross-study comparisons, without proper
validation and assessment of reliability, such practice only serves
to ensure that the measurement problem persists. Essentially, more
attention must be paid to research design and process, construct
validation, and more sophisticated analytical strategies in order to
avoid measurement problems (Bergh, 2001). Validation is an unending
process and we, as management scholars, need to continuously assess
and re-evaluate the usefulness of a particular measure in order to
move management research forward.
References
Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice:
A review and recommended two-step approach. Psychological Bulletin, 103(3),
411423.
Bergh, D. D. (2001). Diversication strategy research at a crossroads. In M. A. Hitt, R. E.
Freeman, & J. S. Harrison (Eds.), Handbook of strategic management. Oxford, UK:
Blackwell.
Boyd, B. K., Gove, S., & Hitt, M. A. (2005). Construct measurement in strategic manage-
ment research: Illusion or reality? Strategic Management Journal, 26, 239257.
Chan, D. (1998). Functional relations among constructs in the same content domain at
different levels of analysis: A typology of composition models. Journal of Applied
Psychology, 83, 234246.
Chen, G., Mathieu, J. E., & Bliese, P. D. (2004). A framework for conducting multilevel
construct validation. In F. J. Dansereau, & F. Yammarino (Eds.), Research in
multi-level issues: The many faces of multi-level issues. (pp, 3. (pp. 273303) Oxford,
UK: Elsevier Science.
Cui, A. S., & Kumar, M. V. S. (2012a). Termination of related and unrelated joint ventures:
A contingency approach. Journal of Business Research, 65(8), 12021208.
Cui, A. S., & Kumar, M. V. S. (2012b). Advancing multilevel alliance research: Rejoinder
to commentary on Cui and Kumar (2012) and future directions. Journal of Business
Research, http://dx.doi.org/10.1016/j.jbusres.2012.05.017.
Godfrey, P. C., & Hill, C. W. L. (1995). The problem of unobservables in strategic
management research. Strategic Management Journal, 16, 519533.
Hayek, F. A. (1989). The pretense of knowledge. American Economic Review, 79(6),
37.
Hill, R. C., & Hellriegel, D. (1994). Critical contingencies in joint venture management:
Some lessons from managers. Organization Science, 5, 594607.
Klein, K. J., Dansereau, F., & Hall, R. J. (1994). Levels issues in theory development, data
collection, and analysis. Academy of Management Review, 19, 195229.
Klein, K. J., & Kozlowski, S. W. J. (2000). From micro to meso: Critical steps in concep-
tualizing and conducting multilevel research. Organizational Research Methods, 3,
211236.
Lawrence, B. S. (1997). The black box of organizational demography. Organization
Science, 8(1), 122.
 Author's personal copy

406 B.B. Nielsen / Journal of Business Research 67 (2014) 403406

Nielsen, B. B. (2007). Determining international strategic alliance performance:
A multidimensional approach. International Business Review, 16(3), 337361.
Nielsen, B. B. (2010). Multilevel issues in strategic alliance research. In T. K. Das (Ed.),
Researching strategic alliances: Emerging issues (pp. 126). Charlotte, NC: Informa-
tion Age Publishing.
Nielsen, B. B. (2012). What determines joint venture termination? A commentary
essay. Journal of Business Research, 65(8), 11091111.
Nielsen, B. B., & Nielsen, S. (2009). Learning and innovation in international strategic
alliances: An empirical test of the role of trust and tacitness. Journal of Management
Studies, 46(6), 10311056.
Rousseau, D. M. (1985). Issues of level on organizational research: Multi-level and
cross-level perspectives. Research in Organizational Behavior, 7, 137.
Suarez, M. V., & Garc'a-Canal, E. (2003). Complementarity and leverage as drivers of stock
market reactions to global alliance formation. Long Range Planning, 36(6), 565578.