Changing Organizational Designs and Performance Frontiers

Abstract

This paper develops and tests a multilevel organizational contingency theory for designing headquarters—subsidiary relations. We use frontier analysis to overcome problems that have hampered advancements in organizational contingency theory in general and headquarters—subsidiary relationships in particular. Based on a longitudinal study of a large medical group practice of 32 local community clinics, we compute the relative distance of clinics from a best-performance frontier, determine what proportions of changes in clinic performance are due to factors that are endogenous or exogenous to the clinics, and examine the organizational factors that may explain these performance changes. We find that uniform headquarters policies have differing effects on the performance of subsidiary units, benefiting some and hindering others through no fault of their own. We also find significant performance volatility with different types of unit designs, suggesting the need to examine the risks of changing organization designs.

Full text

Organization Scienc
e
Articles in Advance, pp. 1–22
issn 1047-7039 eissn 1526-5455 http://dx.doi.org/10.1287/orsc.1110.0694
© 2011 INFORMS
Changing Organizational Designs and Performance Frontiers
Andrew H. Van de Ven
Carlson School of Management, University of Minnesota, Minneapolis, Minnesota 55455, avandeve@umn.edu
Ricky Leung
School of Medicine, University of Missouri, Columbia, Missouri 65212, rleung@missouri.edu
John P. Bechara
Olin Business School, Washington University in St. Louis, St. Louis, Missouri 63130, johnbechara@gmail.com
Kangyong Sun
Graduate School of International Corporate Strategy, Hitotsubashi University, 101-8439 Tokyo, Japan, sunxx136@umn.edu
This paper develops and tests a multilevel organizational contingency theory for designing headquarters–subsidiary
relations. We use frontier analysis to overcome problems that have hampered advancements in organizational contin-
gency theory in general and headquarters–subsidiary relationships in particular. Based on a longitudinal study of a large
medical group practice of 32 local community clinics, we compute the relative distance of clinics from a best-performance
frontier, determine what proportions of changes in clinic performance are due to factors that are endogenous or exogenous
to the clinics, and examine the organizational factors that may explain these performance changes. We find that uniform
headquarters policies have differing effects on the performance of subsidiary units, benefiting some and hindering others
through no fault of their own. We also find significant performance volatility with different types of unit designs, suggesting
the need to examine the risks of changing organization designs.
Key words : organization design; adaptation; frontier analysis
History: Published online in Articles in Advance.
Introduction
Striking a balance between corporate-wide policies and
subsidiary unit autonomy is an ongoing challenge of
multisite organizations that have many geographically
dispersed units, such as manufacturing plants, service
outlets, and retail stores (Child 2005, Dooms and van
Oijen 2008). On the one hand, organization-wide poli-
cies and procedures are necessary to achieve economies
of scale and scope, and they create reliable and branded
products from all of the organization’s stores, factories,
or service centers regardless of their location. On the
other hand, each dispersed unit requires some autonomy
and flexibility so that it is able to respond to its particular
environmental task demands.
Most of the research on headquarters–subsidiary rela-
tions has been in multibusiness and multinational firms
and has focused on the mechanisms used by head-
quarters to control their subsidiaries and the mecha-
nisms subsidiaries use to manage their fit within their
local environment (Birkinshaw 2008). This research has
mainly been concerned with the degree to which cor-
porate control mechanisms should be tailored to allow
subsidiaries to maintain the requisite autonomy to man-
age their relationship with the local institutional condi-
tions and the degree to which such control mechanisms
should be standardized to take advantage of economies
of scale and scope (Ghoshal and Nohria 1989, Nohria
and Ghoshal 1994). Researchers have also examined
the variations of such control mechanisms across sub-
sidiaries (e.g., Dooms and van Oijen 2008).
Although this research has uncovered mechanisms of
corporate control over subsidiaries and has shown how
subsidiary performance is contingent on the fit between
corporate control and the local environment, it has not
examined the multilevel effects of macro-headquarter
policies and microunit efforts on subsidiary perfor-
mance. We argue and show that an understanding of
the design of headquarters–subsidiary relations requires
a systematic way of distinguishing between the exoge-
nous parts of subsidiary performance that are due to
macro-corporate policies and the endogenous parts that
are due to the micromanagement efforts of individual
subsidiaries. Without unpacking the corporate and sub-
sidiary components of subsidiary performance, we have
no evidence for assessing the relative contributions and
relationships among macro and micro levels of organiza-
tions. In addition, organization design researchers have
not discriminated between the relative effects of corpo-
rate and subsidiary control mechanisms on endogenous
and exogenous unit performance. Hence, our research
question focuses on how corporate (macro) and sub-
sidiary (micro) organizational control mechanisms affect
subsidiary performance.
1
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Published online ahead of print October 19, 2011

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
2Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
Probably the most commonly used theoretical frame-
work to examine headquarters–subsidiary relations is
contingency theory (Donaldson 2001, Doz and Prahalad
1991). In this context, the basic contingency theory
proposition is that organizational performance increases
when subsidiary units are designed to differentially fit
their local environmental demands subject to macro-
organizational constraints of the headquarters (Drazin
and Van de Ven 1985, Meyer et al. 1993, Nohria and
Ghoshal 1994, Birkinshaw 2008). Donaldson (2001) and
Child (2005) review the research on this proposition
and suggest that it represents the most widely accepted
view of our current state of knowledge on designing
headquarters–subsidiary relationships. Although intu-
itively straightforward, this abstract proposition glosses
over three thorny problems that limit advancing organi-
zational contingency theory in general and understand-
ing headquarters–subsidiary relationships in particular.
First, the contingency theory proposition assumes that
organization design is a strategic choice (Child 1972,
Donaldson 2001) that can be made in a purposeful and
rational manner by top managers. In reality, organiza-
tion design reflects some joint deliberate decisions and
some emergent actions, especially when multiple stake-
holders with conflicting interests are involved (Pfeffer
and Salancik 1978). Designing headquarters–subsidiary
relations involves some centralized coordination and
some local autonomy, which can lead to unanticipated
conflicts between central administration and local units
(Blau 1964, Astley and Van de Ven 1983). In addition,
most organizations face multiple and often conflicting
environmental demands, structural arrangements, and
performance criteria at macro- and micro-organizational
levels (Thompson 1967, Lewin and Minton 1986).
Achieving fitness with local environmental demands and
with corporate strategies often conflict (Child 1975,
Khandwalla 1973), and it forces making trade-offs
between purposeful and emergent goals at micro- and
macro-organizational levels (Miller 1993, Sinha and Van
de Ven 2005). These complexities make it difficult
to specify in concrete terms the relationships among
abstract notions of organization environment, configura-
tion, and performance in contingency theory.
To move beyond the limits of armchair theorizing,
we propose taking an empirical approach using meth-
ods of frontier analysis to advance our understanding
of designing headquarters–subsidiary relationships. This
approach provides a systematic way to assess the relative
overall performance of organizational subsidiaries fac-
ing comparable resource and environmental constraints
and to unpack the trade-offs between micro- and macro-
organizational factors that might have a differential
impact on subsidiary unit performance.
Second, the contingency theory proposition treats cor-
porate policy as a uniform constraint on all subsidiary
units when corporate policy should, in fact, be treated
as a variable that differentially affects subunits. As a
constraint, it is generally assumed that headquarters’
policies are implemented in a uniform way in all sub-
sidiaries. Yet, as the concept of “affordance” (Gibson
1979, Norman 1988) suggests, any rule or policy is
open to many action possibilities depending on what
one makes of it. These action possibilities depend not
only on the literal policy prescriptions but also on the
actor’s goals, plans, values, beliefs, and past experi-
ences. The same constraint (or policy) may or may
not limit a subsidiary. It has been shown that macro
policies can also enable productive pathways for sub-
sidiary units to perform tasks (Nelson and Sampat 2001).
In other words, macro-organizational rules and policies
both enable and constrain behavior (Adler and Borys
1996). If that is the case, then macro-organizational poli-
cies may increase or decrease micro-organizational per-
formance variations, depending on how the policies are
interpreted and implemented. Thus, Nelson (2008, p. 8)
observes, “Over the years empirical studies have consis-
tently shown large differences in productivity between
establishments of the same corporation producing the
same things and using the same production machinery
(perhaps the best of these studies remains the old one
by Pratten 1976).”
These different connections between headquarters’
policies and subsidiary units’ behaviors have not
been studied because it is very difficult to get
reliable evidence for assessing the relative perfor-
mance of subsidiary units and for determining what
parts of subsidiary units’ performance are due to
subsidiary-specific (or endogenous) factors within the
control of organizational units and what parts can
be attributed to headquarters-specific (or exogenous)
macro-organizational policies. Using frontier analysis,
we disentangle the endogenous subsidiary-specific com-
ponents and the exogenous headquarters-specific com-
ponents of a subsidiary unit’s performance. This makes
it possible to examine how macro-organizational poli-
cies that impose uniformities across dispersed subsidiary
units may have differential impacts on subsidiary perfor-
mance. The same policies may advantage some units but
disadvantage others in unknown and unintended ways.
Third, the concept of fit in contingency theory has
been criticized for being too static and unable to cap-
ture the changing processes of organizational adapta-
tion within an organization’s local environments and
within other organizational levels (Anderson et al. 1999).
Responding to these criticisms, Donaldson (2001) points
out that even some of the classic contingency theo-
ries made the point that organizations adapt over time
by moving from misfit to fit and that what constitutes
a fit changes as the contingencies change. Our lon-
gitudinal research provides an empirical way to go
beyond this comparative-statics analysis of fit and mis-
fit. Conceptually, it requires a shift in thinking of
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 3
organization–environment relations from one of fit to
one of adaptation. This shift is central to the orga-
nizational complexity perspective of Levinthal (1997),
McKelvey (1999), and Siggelkow (2001) and their dis-
cussions of organizational adaptation on a metaphori-
cal changing landscape over time. In the results section,
we show that uniform headquarters’ policies shift the
best-performance frontier differently for various types
of subsidiary units on this metaphorical landscape. The
image of actors “dancing” across a fitness landscape over
time is apropos, as actors adapt to each others’ steps
as well as to moving frontiers. Through this process,
some improve and others fall behind, reflecting “Red
Queen” dynamics (Barnett and Sorenson 2002, Derfus
et al. 2008).
The next section of this paper proposes a multi-
level organizational contingency theory of headquarters–
subsidiary relations. We describe the method of fron-
tier analysis that was used to empirically assess the
relative performance of different types of subsidiary
units, and we determine what parts of unit performance
are attributable to the efforts of individual units and
what parts are due to macro-organizational policies. Our
empirical findings are based on a longitudinal study of
a sample of 32 local community clinics that are all
part of a large medical group practice. With longitudi-
nal data, we construct a changing adaptiveness landscape
for these organizational units and determine what pro-
portions of changes in clinic performance are due to
endogenous or exogenous factors of the clinics. We iden-
tify several organizational characteristics that differen-
tially influence these endogenous and exogenous sources
of clinic performance. We conclude with a discussion of
how this research advances organizational contingency
theory in general and headquarters–subsidiary relation-
ships in particular.
Our paper makes at least four contributions. First and
foremost, we advance a multilevel organizational con-
tingency theory that takes into account the different
impact of headquarters’ policies and subsidiary manage-
rial efforts on subsidiary performance. Second, we show
that frontier analysis provides a systematic way to dis-
criminate endogenous from exogenous components of
subsidiary performance, which provides a first step in
examining headquarters–subsidiary relations. Whereas
many studies have examined headquarters–subsidiary
relations (Ghoshal and Nohria 1989, Nohria and Ghoshal
1994), we distinguish between two types of performance
sources that might provide a more accurate assessment
of subsidiary performance, especially as subsidiaries are
differentially affected by headquarters’ policies. Third,
we show that macro- and micro-organizational control
mechanisms have different performance effects. Endoge-
nous subsidiary performance is a function of micro-
subsidiary control mechanisms (autonomy and interac-
tions), whereas exogenous performance is a function
of macro-corporate control mechanisms (standardization
and integration). Our results show that corporate control
mechanisms (standardization and integration) that are
expected to have a consistent effect on subsidiary perfor-
mance have differential effects on exogenous subsidiary
performance and unexpected effects on endogenous sub-
sidiary performance. Fourth, using frontier analysis, we
provide a new way of examining dynamic fit, or adapta-
tion, by identifying the best-performing subsidiaries and
assessing their relative distance from other subsidiaries
subject to the same input and environmental conditions.
This provides a more cogent test of contingency the-
ory’s arguments and avoids the pitfalls of assessing per-
formance based on mean-level regression-based models.
Our empirical findings also suggest that an important
area for future study is in the examination of the differ-
ent levels of risk associated with changing organizational
designs.
Conceptual Framework
Headquarters–Subsidiary Relations
The study of headquarters–subsidiary relationships has
a long and rich history in organization and manage-
ment literature dealing with the division of labor, depart-
mentation, and control (Filley et al. 1976, Galbraith
1977, Walker and Lorsch 1968). Work in organizations
can be divided in a number of ways (e.g., by geogra-
phy, product, function, or goal) and assigned to sub-
sidiaries, departments, divisions, or other units. Filley
et al. (1976, p. 360) note that early writers focused
on the efficiencies of alternative forms of departmenta-
tion and the degrees of headquarters’ control over sub-
sidiary units. Little consideration was given to unique
conditions facing individual subsidiary units and their
relative influence on headquarters–subsidiary relations.
During the past 30 years, much has been learned from
studies of headquarters–subsidiary relations, mostly in
multinational corporations (MNCs) (see reviews by
Egelhoff 1988, Gupta and Govindarajan 1991, Prahalad
and Doz 1987, Ghoshal and Nohria 1989, Birkinshaw
2008). Most MNC researchers adopted an organizational
contingency theory to examine headquarters–subsidiary
relations. In this context, the core contingency theory
proposition is that organization performance increases
when subsidiary units are designed to differentially fit
their local environmental demands subject to corporate-
wide constraints of the headquarters (Donaldson 2001,
Child 2005).
Implicit in this proposition is a multilevel view of sub-
sidiary performance. It states that the performance of
a subsidiary is a function, in part, on the behavior of
managers and employees within the subsidiary and, in
part, on the effectiveness of macro-organizational poli-
cies that subsidiaries are required to implement. Deter-
mining the relative contributions of macro-organizational
policies and micro-organizational behavior on subsidiary
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
4Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
performance is an important first step for empirically
understanding headquarters–subsidiary relations. A basic
assumption in designing headquarters–subsidiary rela-
tions is that these macro and micro contributions are
complementary, or positively correlated. In other words,
macro-organizational policies are intended to support the
performance of most or all subsidiaries, and the man-
agers of subsidiaries are expected to adapt and imple-
ment these macro policies in constructive ways. We test
this assumption in our first hypothesis.
Hypothesis 1. There is a positive correlation
between the components of subsidiary performance that
are attributable to endogenous micro-subsidiary efforts
and to exogenous macro-organizational policies.
In addition to examining the relative contributions
of micro and macro levels of organization, a con-
tingency theory of headquarters–subsidiary relations
requires the study of how organizational units adapt
to their local environments. In a study of 54 MNCs,
Nohria and Ghoshal (1994) find support for this proposi-
tion. Headquarters–subsidiary relations that were differ-
entiated to fit their local contexts were associated with
higher performance of the MNC as a whole. They mea-
sured two dimensions of subsidiary context (the com-
plexity of local environment and the amount of sub-
sidiary resources) and three dimensions of structural
control by headquarters over its subsidiaries through
centralization, formalization, and normative integration
(an informal cultural dimension of shared values). They
found that in high-performing MNCs, greater environ-
mental complexity was associated with lower formaliza-
tion, moderate centralization, and higher shared values.
In addition, subsidiary resources were negatively related
to centralization, and shared norms were positively
associated with the formalization of the headquarters–
subsidiary relations.
Dooms and van Oijen (2008) use variables similar to
Nohria and Ghoshal (1994) to examine the relative bal-
ance between subsidiary autonomy (tailored control) and
headquarters-standardized control. They distinguished
between formal and informal dimensions of tailored con-
trol (subsidiary autonomy and communications, respec-
tively) and formal and informal dimensions of corpo-
rate standardized control (formalization and integration
or shared values, respectively). No measures of organi-
zation performance were reported. Based on a study of
100 subsidiaries of 23 corporations, they found that firms
adopt a balanced combination of subsidiary autonomy
and corporate standardization. Dooms and van Oijen
(2008, p. 250) report that the findings were more com-
plicated than expected. They found no corporate effect
on subsidiary autonomy but found that integration, for-
malization, and communication are subject to corpo-
rate influences. However, these corporate effects did not
diminish the large variations that were still explained by
the subsidiary level.
Building on these studies, we identify four dimen-
sions for studying headquarters–subsidiary relations:
subsidiary autonomy, subsidiary interactions, corpo-
rate standardization, and headquarters–subsidiary inte-
gration. These four dimensions are related, as illus-
trated in Table 1, and moderated by environmental
complexity. Subsidiary autonomy (i.e., decentralization
of decision making) and corporate standardization (i.e.,
formalization of policies and procedures) reflect for-
mal structural mechanisms of controlling activities at
the subsidiary and headquarters levels of organization.
Subsidiary interactions (interdependence and communi-
cations with headquarters) and headquarters–subsidiary
integration (shared norms) refer to informal control
mechanisms. These informal dimensions reflect the view
that organizational culture (Martin 1992) and interac-
tions (Homans 1961) minimize divergent interests and
can serve as substitutes for structural controls (Durkheim
1964, Gouldner 1954). Following Nohria and Ghoshal
(1994) and Dooms and van Oijen (2008), we expect
these formal and informal control mechanisms to com-
pensate and complement each other.
Donaldson (2001) points out that these dimensions
also reflect different levels of organizational differen-
tiation and integration. Unlike Lawrence and Lorsch
(1967), who view differentiation and integration at a
single organizational level, Nohria and Ghoshal (1994)
distinguish them at micro and macro levels of organi-
zation. Differentiation focuses on the subsidiary level
and refers to the structural autonomy and interac-
tions of each subsidiary unit with headquarters. Inte-
gration applies to the corporate level and refers to the
control achieved by headquarters through standardized
policies and procedures, as well as cultural views of
headquarters–subsidiary integration. Subsidiary differen-
tiation and headquarters integration need not be cor-
related: “Differentiation here is the difference between
subsidiaries that may or may not need to be inte-
grated at the corporate level” (Donaldson 2001, p. 85).
Thus, integration also refers to corporate policies and
strategies that apply to all subunits regardless of their
differentiation.
These dimensions of micro-organizational differenti-
ation and macro-organizational integration have impor-
tant implications for designing headquarters–subsidiary
relations. Dooms and van Oijen (2008, p. 245) point
out that unit autonomy and communications are “tailor-
made” to each subsidiary and differentiated through-
out the organization. In contrast, corporate-wide factors,
such as standardization and culture, involve controls that
apply uniformly across organizational subsidiaries. This
leads us to expect, as Table 1 illustrates, that subsidiary
autonomy and interactions most directly influence the
component of subsidiary performance that is endogenous
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 5
Table 1 Formal and Informal Dimensions of Headquarters–Subsidiary Relations
Formal dimensions Informal dimensions Environmental complexity Expected performance effects
Unit differentiation Autonomy Interactions Endogenous
—Decentralization —Interdependence —That part of unit performance
—Communications that is due to unit’s own action
Corporate integration Standardization Integration Exogenous
—Standardization —Shared norms —That part of unit performance
—Culture that is due to corporate policies
or within the control of subsidiaries, whereas corporate
standardization and headquarters–subsidiary integration
directly predict the exogenous component of subsidiary
performance that is attributable to macro-organizational
structure. Specifically, we hypothesize the following.
Hypothesis 2. Micro-subsidiary control mechanisms
of autonomy and interactions have a positive effect on
endogenous performance and no effect on exogenous
performance.
Hypothesis 3. Macro-corporate control mechanisms
of standardization and integration have a positive effect
on exogenous performance and no effect on endogenous
performance.
Finally, we expect that the complexity of the local
environment served by the subsidiary will moder-
ate these relationships. Studies of organizational con-
tingency theory show that environmental complexity
increases the need for unit autonomy to adapt flexibly
to heterogeneous demands (see the review by Donald-
son 2001) and that uniform corporate policies tend to
be too coarse-grained for adapting to diverse, particu-
lar needs (Nohria and Ghoshal 1994, Sinha and Van de
Ven 2005). Conversely, as Thompson’s (1967) concept
of coordination by programming emphasizes, organiza-
tional rationality and efficiency increase when uniform
rules, policies, and procedures can be applied to stable
and homogeneous situations. Hence, the less complex a
subsidiary’s environment, the less the need for exercis-
ing autonomy and the greater the performance benefits
of standardized policies and procedures. Therefore, we
propose the following.
Hypothesis 4. Local environmental complexity en-
hances Hypothesis 2 and dampens Hypothesis 3.
In the next section, we discuss operational procedures
for distinguishing between the endogenous and exoge-
nous parts of subsidiary performance with a method of
frontier analysis.
Methodology
Frontier Analysis
The empirical study of organizational fitness or adap-
tiveness involves two steps: (1) in a sample, identify
the most efficient organizations that best achieve perfor-
mance outcomes subject to their particular resource and
environmental constraints; and (2) analyze their design
configurations with those of less efficient, comparable
organizations facing similar resource and environmen-
tal constraints (Donaldson 2001, Meyer et al. 1993).
The first step entails a constrained maximization prob-
lem of calculating the maximum performance outputs
of organizational units in a sample subject to different
resource and environmental input constraints. Organiza-
tional design factors are not part of this first step. The
second step treats this calculated result from the first step
as the dependent variable whose variance is explained
in terms of a set of organizational design factors using
a standard regression model.
To perform the first step of the analysis, we chose
a method of frontier analysis to avoid a logical prob-
lem with regression-based models used in prior studies
of organizational contingency theory (e.g., Drazin and
Van de Ven 1985, Doty et al. 1993). As Bryce et al.
(2000, p. 511) discuss, regression is designed to explain
variance in average behavior; for example, Y=fX
estimates variations in average outputs, Y, from a set
of independent variables, X. In contrast, frontier anal-
ysis mathematically calculates the outlying ideal-type
organization that maximizes desired performance out-
puts subject to its particular input constraints (Lewin
and Minton 1986). This latter estimate, at least intu-
itively, provides an appropriate measure of the relative
efficiency of organizational units facing comparable con-
straints. Frontier analysis provides a more direct method
of empirically identifying these most adaptive outliers
in a sample of organizations than do regression methods
that examine the distance of residuals from the center of
a least squares line.
Frontier analysis is a method that begins with the
outliers in a sample. It empirically identifies the most
adapted or best-performing units on the outlying fron-
tier in the sample and then provides a way to exam-
ine the relative distance of other units in the sample
from their comparable cohorts on the frontier. Frontier
analysis directly addresses the constrained optimization
problem in contingency theory that is central to con-
figuration and complexity perspectives (Sinha and Van
de Ven 2005). The best-performance frontier consists of
organizational units that maximize desired output criteria
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
6Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
Figure 1 Geometric Portrayal of Frontier with DEA
U1
Performance:
High
Low
Low High
–Net income
–Productivity
–Patient
satisfaction
–Care quality
Inputs: Resources (no. of personnel)
Environmental complexity (patient mix)
U2
U3
U5
U4
subject to input resource and environmental constraints
in comparison with others examined in the sample.
Figure 1 provides a geometric representation of how
Data Envelopment Analysis (DEA) works. The appendix
presents a technical description of DEA calculations.
DEA searches for the weights that optimize outcome
performance measures (the Yaxis) subject to a set
of input factors (on the Xaxis) for organizational
units being investigated. After scores are calculated, as
described in the appendix, a best-performance frontier
can be identified from which other units can be com-
pared. A best-performance frontier refers to the maxi-
mum output that can be attained given a set of input
conditions for a sample of units that use a similar
transformation process to convert inputs into outputs
(Jayanthi et al. 1996). Instead of trying to fit a regres-
sion plane through the center of the data, DEA floats
a piecewise-linear surface to rest on top of the most
extreme observations in a sample of organizational units
(i.e., DEA envelops the observations—hence, its name).
Compared with other frontier estimation methods, the
features of DEA that make it particularly appropriate for
our research objective are that it (1) can handle multi-
ple input and output variables, where each variable may
be measured in different scales; (2) does not require
that functional relationships be specified between the
input and output variables; and (3) allows for a nonlin-
ear shape to the frontier. We also show how longitudinal
observations of a sample of organizations can be ana-
lyzed with DEA to determine, over time, the changes
in performance frontiers and the relative adaptiveness of
organizational units on and off of the shifting frontier.
Exemplary applications of DEA in organizational
studies have been made by Lewin and Minton (1986),
Chilingerian (1995), Cooper et al. (1996), and Johnson
et al. (1996). DEA has also been used in several health-
care settings; it has been used to evaluate the efficiency
of U.S. health maintenance organizations (HMOs) from
1985 to 1994 (Wholey and Bryce 1997), physician effi-
ciency in hospitals (Chilingerian 1995), primary health
care in England (Salinas-Jiménez and Smith 1996), pri-
mary care physicians of a large HMO in the eastern
United States (Chilingerian and Sherman 1996), and
case workers in home health-care services (Johnson et al.
2002). With the exception of the Wholey and Bryce
(1997) study, all others used DEA to examine cross-
sectional data. It is important to note that our objective
and method in using the DEA frontier analysis differs
from these applications. Past studies have used DEA
results as the final deterministic criterion to evaluate
and prescribe interventions for the specific organizations
being investigated. Our objective is to draw inferences of
study findings that go beyond the immediate sample in
order to advance a more general multilevel contingency
theory of designing headquarters–subsidiary relations.
Field Research Setting
This research is part of a larger longitudinal study of
organizational integration in a large midwestern man-
aged health-care system (hereafter, Midwestern) in the
United States. This system emerged as a vertically inte-
grated health-care provider (with 20,000 employees and
$2 billion in revenues) through a merger in 1994 of
15 hospitals, about 50 primary care clinics, a variety
of home care and ancillary services, and several health
insurance plans that cover over approximately one mil-
lion people. From 1994 to 2002, our study tracked the
formation and integration of the system’s medical group
of primary care clinics.
This group practice was founded in 1994 with 20 clin-
ics that were owned by Midwestern’s parent hospitals at
the time of the merger. During its first few years, the
group grew rapidly by acquiring 30 additional primary
care clinics to provide geographical coverage of the
regions served by the group’s sister health plan, whereas
others were acquired to provide primary care and patient
referrals to Midwestern’s hospitals. During this acqui-
sition period, the group experienced rising costs as it
struggled with integrating its new clinics. By the end
of 1996, the group’s management attention shifted from
growth by acquisition to organizational integration of
its now 50 clinics with 450 physicians and more than
3,000 employees. From 1997 to 1999, group top man-
agers focused on introducing standardized clinic operat-
ing procedures, increasing clinic financial performance
through various cost reduction and revenue-generating
initiatives, improving clinical quality and patient satis-
faction as well as strengthening employee morale.
The group developed and implemented many pol-
icy and procedural changes in all of its clinics dur-
ing the study period examined in this report. They
included uniform and consolidated patient billing, sup-
ply purchasing, equipment maintenance, laboratory ser-
vices, and other related administrative functions. These
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 7
changes required clinics to sever some old vendor rela-
tionships and develop new ones; the clinics sometimes
had to let staff go who had provided them with these
functions within the clinics. With extensive physician
input, several clinical groupwide initiatives were also
undertaken, such as clinical care quality improvement,
drug formularies, and risk management programs. Clinic
managers and physicians were charged with develop-
ing new documentation and data collection systems and
modifying previous practices to comply with new clin-
ical quality and risk management standards. To pro-
mote equity in compensation and improve productiv-
ity in all clinics, a uniform compensation system for
all clinic employees was adopted, and a standardized
physician productivity metric was implemented based on
an industry standard of relative value per unit of care
(RVU). Managers in the clinics had to adjust their pre-
vious human resource practices to match new group-
wide standards, and physicians had to adjust to a differ-
ent kind of performance measurement and compensation
system than they had experienced before. Group man-
agers consolidated some previously stand-alone clinics
to decrease operating expenses, requiring clinic staff and
physicians to accommodate to new locations and new
working relationships. The group also adopted some new
operating procedures designed to improve patient sat-
isfaction and care quality, such as a system allowing
same-day scheduling for patients and an automated elec-
tronic medical record. Clinics were charged with assem-
bling teams to implement these new initiatives. Finally,
the group practice negotiated a uniform payment reim-
bursement contract with health insurance companies that
treated all clinics as one provider. This represented a
major change from the previous procedure where indi-
vidual (and more-or-less powerful) clinics negotiated
their own reimbursement contracts with health plans.
These are just a few of a myriad of changes that the
medical group and its clinics implemented between 1997
and 1999.
This field setting provides an ideal opportunity to
examine whether and how these group-level policy
changes are related to the organization and performance
of subsidiary clinics that are serving diverse patients in
local communities. This report examines the two-year
time period (1997 and 1999) when most of the group
integration initiatives mentioned above began and con-
cluded. Of course, we cannot measure the influence of
each of these macro-policy changes on clinics. Instead,
we view the cumulative effects of all macro-group poli-
cies as producing the total change observed in the group
performance frontier for all the clinics during the two-
year study period. We then examine how this group fron-
tier change differentially affects clinic performance.
Data Collection
Data on the clinic variables were collected in the fall
of 1997 and again in the fall of 1999. In each wave,
the data came from three different sources. First, we
were given access to patient satisfaction surveys that
were conducted in 1997 (7,700 in total); the surveys
taken in 1999 were completed by 8,000 patients who
were served by the primary care clinics. The group prac-
tice contracted with an independent health-care survey
organization to conduct the patient satisfaction surveys.
After cleansing the data of any patient identification
information, the vendor provided us with a copy of the
patient satisfaction surveys for each clinic. Second, data
on clinic environmental characteristics and economic
performance were obtained from organizational records.
Third, we designed and conducted questionnaire surveys
and obtained responses from about 1,000 employees
from all clinics in 1997 and 1999; our surveys included
measures of the perceived quality of health care and the
other variables in our model of headquarters–subsidiary
relations in Table 1.
The specific sample of organizational units examined
in this study consists of 32 primary care clinics1that are
owned by the large managed care organization. Given
the relatively small number of clinics for statistical anal-
ysis, we had to restrict the scope of our data analysis
and model by selecting the fewest number of clinic input
and outcome variables in our DEA model.
DEA Model Variables
Specifying an empirical DEA model of key clinic inputs
and performance outcomes is the first and perhaps most
important step in the research process. Model misspecifi-
cation results in selecting the wrong organizational units
on and off the frontier and irrelevant findings for theory
and practice. To decrease this likelihood, we collabo-
rated with managers and lead clinicians of the medical
group practice to develop an empirical model of clinic
input and outcome variables that the managers used to
evaluate clinic performance. This involved a series of
meetings, including about three with the top managers
of the group practice, another meeting with the district
managers and medical directors, and a review session
with the clinic managers. The DEA model that the man-
agers and researchers jointly developed to assess the per-
formance of clinics is the following:
Maximize: Business Care (clinic productivity and net
income) and Patient Care (patient satisfaction and care
quality) subject to clinic resources (size), and for differ-
ent levels of environmental complexity (patient mix).
These clinic input and performance measures are now
described.
Clinic Input Conditions. Clinic resources (the size or
number of employees) was chosen as the key input fac-
tor in the DEA model for clinics serving a broad and
focused mix of patients (our indicator of environmen-
tal complexity). Clinic size was selected as the major
resource input factor because labor costs account for
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
8Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
approximately 80% of annual clinic operating costs.
Using size as the input variable, the DEA analysis iden-
tifies the relative performance of clinics in terms of
those that accomplish the most with the resources they
have. In addition, studies over the years indicate that
size is perhaps one of the best overall predictors of
organization structure (Tosi and Patt 1967, Kimberly
1976, Cullen and Baker 1984, Bluedorn 1993, Camisón-
Zornoza et al. 2004).
With regard to environmental complexity, clinic man-
agers and physicians emphasized the importance of the
mix of patients served by different clinics. An internal
organizational study found that there were large differ-
ences in demands on clinics that served a broad ver-
sus narrow mix of patients. The study characterized
patients of broad or heterogeneous clinics as “option
seekers” (where 58% of all patients are shoppers, bas-
ing decisions on urgency, convenience, and cost). In
contrast, the patients served in more focused or homo-
geneous clinics were referred to as “relationship seek-
ers” (42% want long-term relationships). Medical group
managers observed that smaller clinics appear more effi-
cient and effective, but that may be because they serve
a more homogeneous set of patients. Larger clinics that
serve a greater proportion of option seekers stay open
longer hours (resulting in higher overhead and staffing
costs), provide urgent care and more diverse services,
and deal with patients who are not easy to keep satisfied.
With the assistance of the group practice managers and
physicians, we classified the environmental complexity
of primary care clinics in this sample into two groups
based on whether they serve a broad or focused mix of
patients.
Clinic Outcome Performance. Medical group man-
agers and physicians selected two composite measures
of clinic performance: patient care and business care. As
Table 2 indicates, patient care is the average of patient
satisfaction and staff courtesy (measured with eight and
four items, respectively, in a patient survey conducted
by an independent research firm) and clinical quality
care (measured with five items in a survey completed by
clinic health-care providers). Business care is the aver-
age of clinic productivity (measured from organizational
records as the number of standardized clinical services
(RVUs) per clinic provider) and clinic net income (mea-
sured as net revenue per clinic provider from organiza-
tional records). Group practice managers confirmed the
relevance of these performance measures. They reported
using these measures in their performance appraisals of
clinics and physicians within the clinics.
Measures of the Clinic–Group Relations Model
As Table 2 shows, we measured four dimensions
of clinic–group relations: clinic autonomy, clinic–
group interactions, group work standardization, and
headquarters–subsidiary integration. These dimensions
were defined and measured as follows.
1. Clinic autonomy: This measure is defined as the
extent to which clinic employees (physicians, nurses,
and staff) perceive that they exercise discretion or influ-
ence decisions about what work to perform, how work
is performed, and the development of work policies
or procedures. Survey questions for measuring these
three decisions were adapted from Van de Ven and
Ferry (1980).
2. Work standardization: As noted above, during the
period of study, group practice management introduced
a variety of policies and procedures that all clinics were
to follow. Our measure of standardization captures the
combined effects of how these group policy changes
influenced the formalized structuring of work in clinics.
As Table 2 indicates, work standardization was mea-
sured with four survey questions dealing with the clarity,
degree, number, and enabling of work rules and proce-
dures. Three of these items come from Van de Ven and
Ferry (1980).
3. Clinic–group interactions: This is a measure of
informal clinic–group relations that is tailored to individ-
ual clinics and not specified by group policies (Dooms
and van Oijen 2008, p. 247). It was measured as the
degree of perceived interdependence and communica-
tions between individual clinics and group headquarters
with the five survey items outline in Table 2.
4. Headquarters–subsidiary integration: Creating an
integrated medical group practice out of many previ-
ously independent clinics was a major objective of group
managers. Nearly biweekly meetings and semiannual
retreats of clinical and managerial employees occurred
during the study period; these featured discussions of
group culture, shared norms, and integrative synergies.
These meetings provided an informal mechanism for
group coordination and integration. The survey included
10 questions, outlined in Table 2, that measure the per-
ceived degree of headquarters–subsidiary integration.
The right column of Table 2 shows the interitem
reliabilities (coefficient ), the intraclass correlations
(ICC(1)) (Bliese 2000), and the within-clinic interrater
reliabilities (rwg(James et al. 1984) for the four orga-
nizational dimensions. The values for the four con-
structs range from 0.68 to 0.86, providing good evidence
of reliabilities among items used for measuring the four
constructs. In addition, the interrater reliability estimates
with rwg among all informants within clinics are clearly
in an acceptable range from 0.73 to 0.94. Moreover, the
ICC(1) values (ranging from 0.42 to 0.46) are significant
and comparable to the median ICC(1) values of aggre-
gated constructs reported in the organizational literature
(see Bliese 2000, Zellmer-Bruhn and Gibson 2006).
Table 3 presents the results of a confirmatory factor
analysis of the 21 survey items that were used to mea-
sure the four organizational dimensions. The table shows
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 9
Table 2 Measurement of Variables
Variable Measures , ICC, rwg
Clinic resources (size) Number of full-time equivalent positions based on organization records
Clinic environmental complexity Mix of patients served by a clinic, coded by group managers as a broad
(heterogeneous) or focused (homogeneous) mix of patients
Clinic productivity The number of RVUs of care per provider. The RVU is an industry
standard established by Medicare to measure the units of patient care
delivered by health-care providers
Clinic net income Net revenue per provider based on organization records
Patient satisfaction as perceived
by patients served by clinic
Mean of eight items in a patient survey conducted by an independent
research firm contracted by the medical group practice:
=084
1. Chance to explain the reason for the visit
2. Provider listened
3. Received answers to questions
4. Trust in provider
5. Treated with respect
6. Involved in treatment decisions
7. Sufficient time with provider
8. Provider courtesy
Patients’ perceptions of clinic
staff courtesy
Mean of four items in a patient survey conducted by an independent
research firm contracted by the medical group practice:
=086
1. Courtesy of appointment maker
2. Courtesy of office staff
3. Courtesy of telephone advice staff
4. Rating of telephone advice received
Patient care quality Mean of five items in an employee survey about the extent of agreement
on following process indicators of quality of care provided to patients:
=074
1. Patients see the same physician
2. Clinicians meet to discuss patient care
3. Patient information is available when needed
4. Follow-up phone calls are made
5. Patients are reminded when they need additional care
Clinic autonomy Mean of three items in an employee survey about the amount of
influence in making decisions about
1. What work is performed
2. How work is performed
3. Work policies and procedures
=080
ICC(1) =043∗
rwg =073
Work standardization Mean of four items in an employee survey about the extent to which =068
1. Work is clearly defined ICC(1) =041∗
2. Work requires following rules and procedures rwg =085
3. Number of rules and procedures to follow
4. Rules make work easier
Clinic–group interactions Mean of five items in an employee survey about =084
1. Clinic’s dependence on the group ICC(1) =046∗
2. Group’s dependence on the clinic rwg =074
3. Frequency of communications with people in the group
4. How often one works with people in the group
5. Frequency of meetings with people in the group
Headquarters–subsidiary
integration
Mean of 10 items in an employee survey about the extent to which the
medical clinics and group
1. Complement each other
2. Consider each other in their actions
3. Share a primary focus on patient care
4. Respect each other’s views and values
5. Go out of their way to help each other
6. Get in each other’s way (reverse scored)
7. Believe that their future is tied to one another
8. Feel they have a better chance of succeeding together
9. Share information that is helpful
10. Have incentives that are aligned
=086
ICC(1) =042∗
rwg =094
∗p<0001.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
10 Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
Table 3 Confirmatory Factor Analysis on Clinic–Group Relations Measures
Headquarters–subsidiary Clinic–group
integration interactions Clinic autonomy Work standardization
1234
Work standardization
Work is clearly defined 024 −013 −027 042
Degree that rules are followed 018 −013 012 083
Number of rules 011 −007 016 083
Rules enable work 036 −007 −011 071
Clinic autonomy
Decide what work to perform 000 020 −088 −010
Decide how to perform work 000 016 −083 −003
Decide on work policies and procedures 000 019 −077 −006
Clinic–group interactions
Clinic depends on group −001 073 −012 −012
Group depends on clinic 000 064 −004 005
Communication frequency −003 085 −020 −014
Work frequency −004 087 −018 −011
Meeting frequency −006 078 −024 −023
Headquarters–subsidiary integration
Complement each other 070 −004 −004 024
Consider each other 080 −009 001 019
Share a primary focus 075 005 −015 021
Respect each other 084 −001 −008 023
Help each other 082 −005 −005 024
Get in each other’s way (reverse scored) −046 020 −007 −031
Future is tied together 068 004 −004 010
Succeed together 075 003 000 013
Share helpful information 078 −009 −005 021
Have aligned incentives 077 −008 −002 021
Eigenvalues 584 351 191 179
Cumulative % common variance 2650 4250 5120 5930
Notes. n=1205 respondents. Extraction method: principal component analysis. Rotation method: oblimin with Kaiser normalization.
the structure matrix produced by a principal components
analysis using an oblimin rotation procedure with Kaiser
normalization. The results show strong evidence of con-
vergent and discriminant validities. All items intended
to measure one of the four organizational constructs (in
bold) converge with high loadings on a single factor and
clearly discriminate by having low loadings on all other
factors. The four factors explain about 60% of the com-
mon variance, and each factor contributes a substantial
percentage to this cumulative common variance.
DEA Performance Analysis
DEA was performed on the sample of clinics measured
in 1997 and then again on data collected in 1999. As
explained in the appendix, DEA computes the efficiency
of clinics (called decision-making units, or DMUs, in
the DEA literature) with the following equation:
Maximize Eu=s
r=1yruOru
m
i=1xiuIiu
(1)
where urepresents the units of DMU; Erepresents per-
formance efficiency; Iand Orepresent all inputs and
outputs, respectively, for each DMU; and xand yrep-
resent the weights assigned to each input and output.
These weights are chosen in such a way that the DEA
efficiency ratio is maximized for each DMU in the inter-
val 01. As Figure 1 illustrates, this frontier bounds
(“envelops”) the remaining data points from above, and
the interior observations of DMUs below the frontier
receive nonnegative scores less than 1 based on their
proximity to the frontier (Bryce et al. 2000, p. 513).
As noted previously, we follow two methodological
procedures to analyze the DEA model and the clinic–
group design model. First, we use the DEA procedure to
identify the organizational units on a best-performance
frontier as discussed in the appendix. Applied to our
longitudinal data, we compute a Malmquist Index (MI)
developed by Färe et al. (1994) to determine what
parts of changes in unit DEA productivity are due to
endogenous efforts of organizational units (often called
“managerial” efficiency) and exogenous shifts in the
environment or macro-organization (also called “policy”
efficiency by Thanassoulis 2001). We subsequently use
these DEA results as the dependent variables to ana-
lyze the clinic–group relations model using a standard
regression-based model and to examine how organiza-
tional units adapt over time in their relative positions
on or off the best-performance frontier; we also exam-
ine how this frontier changes over time to produce Red
Queen dynamics.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 11
Table 4 Correlations Among Clinic Performance Measures
Patient care Business care
Patient Patient Staff Patient Business Productivity RVUs Net revenue
care composite satisfaction courtesy care quality care composite per provider per provider
Patient care
Patient care composite 1
Patient satisfaction 089∗∗ 1
Staff courtesy 068∗∗ 041∗∗ 1
Patient care quality 017∗∗ −005∗061∗1
Business care
Business care composite −010 −004 −010 005 1
RVUs per provider −003 0005 −013 0001 072∗∗ 1
Net revenue per provider −014 −012 −002 015 079∗∗ 018∗∗ 1
∗p<005; ∗∗p<001.
Findings
Descriptive Statistics
Table 4 shows the correlations among clinic performance
measures of business and patient care. As expected, the
two measures of business care are significantly corre-
lated, as are the three indicators of patient care. The
table also shows very low correlations between the mea-
sures of patient and business care. This latter finding
was gratifying to clinic physicians and managers who
feared that clinic cost-cutting efforts were compromis-
ing patient care quality. They indicate that clinic busi-
ness care and patient care are independent; advancing
one outcome does not decrease or increase attaining the
other desired outcome in this sample of clinics.
Table 5 Clinics Serving Broad and Focused Mixes of Patients
Broad Focused Group differences
(heterogeneous) mix (homogeneous) mix
Number of clinics 18 14
Mean Std. dev. Mean Std. dev. Significance
Resources (no. of personnel) 5773 2828 2695 2025 ∗∗
Organization design
Autonomy 316 030 322 045
Standardization 365 018 368 030
Interactions 243 038 258 042
Integration 319 017 306 031
Clinic performance
Business care composite 208 036 195 100
RVU productivity/provider 640393 124583 685558 365032
Net income/provider ($) −3150233 2723939 −4918007 5579270
Patient care composite 148 048 256 076 ∗∗
Patient satisfaction 279 004 285 004 ∗∗
Staff courtesy 387 015 405 015 ∗∗
Perceived care quality 348 020 384 038 ∗∗
DEA performancea
Clinic performance change 093 029 081 035
Group frontier change 103 003 113 001 ∗∗
aRefer to Table 6.
∗∗p<001.
Table 5 presents a comparison of the clinics serving
a broad (heterogeneous) versus focused (homogeneous)
mix of patients. The table shows that the two groups of
clinics are significantly different in terms of size, patient
care, and group frontier change, but not in terms of orga-
nizational design, business care, and clinic DEA per-
formance change over time. (These DEA performance
measures are discussed in the next section.) On aver-
age, clinics with a more complex heterogeneous mix
of patients are larger (staffed with about 58 clinicians)
than clinics serving a less complex and homogeneous
mix of patients (with 27 employees). The more focused
mix of patients in these clinics tend to provide OB/GYN
services for expectant families, treat patients with dia-
betes and other chronic diseases, or serve more acute
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
12 Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
needs of elderly patients. In addition to these kinds of
patients, the broad clinics also serve a more heteroge-
neous array of patients and families seeking less acute
and chronic health-care services such as vaccinations;
medical checkups; the treatment of colds, flus, cuts, and
scrapes; and drug prescriptions. Because they have medi-
cal conditions requiring long and repeated care, a greater
proportion of patients in focused clinics seek a long-
term relationship with their “own” doctor, whereas more
patients of broad clinics seek immediate, competent, and
comprehensive health-care services when it is convenient
for their busy work and family schedules.
Clinic site visits by the researchers revealed that the
more complex broad (compared with focused) clinics
were not only larger but also much busier, with full wait-
ing rooms and a constant buzz of clinicians and patients
going in all directions. Given these different patient
expectations and practices, the findings in Table 5 are
not surprising: patients served by less complex focused
clinics are more satisfied with their care and perceive
clinic staff as more courteous, and staff perceive that
they provide a higher quality of health care than broad
clinics. The insignificant statistical differences between
broad and focused clinics on organizational and busi-
ness care dimensions may be due to the much greater
standard deviations in these dimensions among focused
clinics than among broad clinics. When reporting these
findings, managers of the clinic group practice stated
that the cost structure of focused clinics was lower than
broad clinics. They referred to an internal study by the
health insurance plan (a sister division of the Midwest-
ern system) that found that the per-member per-month
cost of the average patient served in the broad-mix clin-
ics was $153 and was $98 in the focused-mix clinics.
These observed differences between the two types of pri-
mary care clinics provide an ideal natural field setting
to examine how uniform macro-organizational policies
might have differential effects on the performance of
subsidiary clinics depending on their type.
DEA Analyses Results
Following Thanassoulis (2001), we evaluated clinics’
efficiency in two DEA steps. The results of these two
steps, repeated in each time period (1997 and 1999), are
reported in the columns labeled “Step 1” and “Step 2”
in Table 6. Step 1 evaluates each clinic’s efficiency com-
pared to clinics serving the same type of patient mix
(i.e., either broad or focused). Step 2 reports each clinic’s
efficiency compared to all types of clinics in the sam-
ple. In terms of an adaptiveness landscape metaphor,
Step 1 compares the relative performance among clinics
on their own design hill, whereas Step 2 compares the
relative performance among clinics on all (both broad
and focused) hills.
As the Step 1 columns show for both time periods,
each design group has its own best-performing clinics.
Clinics with DEA efficiency ratings of 100% in Table 6
are on the best-performance frontier for this sample,
whereas the scores of clinics with DEA efficiency rat-
ings lower than 100% indicate how far the clinics are off
the frontier (relative to their best-performing peers). In
1997 (Time 1), clinics C23, C24, and C31 were on the
frontier among broad clinics; C17, C32, C41, and C57
were on the frontier among focused clinics in that year.
Two years later (Time 2), clinics C14 and C31 were on
the frontier among broad clinics, and C17, C32, C42,
and C60 were on the frontier among focused clinics in
that year. Only clinic C31 among broad patient care clin-
ics and clinics C17 and C32 remained on the frontier in
both years, whereas other clinics came on and fell off
the frontier relative to their cohorts.
The relative performance efficiency of broad and
focused clinics operating in complex and simple envi-
ronments (respectively) in each year is indicated in the
Step 2 columns of Table 6. The table shows that focused
clinics have higher DEA efficiency scores than the clin-
ics serving a broad mix of patients. In fact, not one
broad clinic has a DEA efficiency score that exceeds the
lowest-performing focused clinic. This is shown in Fig-
ure 2, which plots the clinics based on their Step 2 scores
in 1997 and 1999. As the figure shows, there are dra-
matic DEA performance differences between broad and
focused clinics in both time periods. Equally clear from
the figure is the close clustering among broad clinics in
the low DEA efficiency range and the even closer clus-
tering among focused clinics in the high-performance
range in 1997 and 1999. The correlation between the
DEA efficiency in 1997 and 1999 for all clinics is 0.97
(statistically significant), whereas it is only 0.33 among
clinics within each design type.
As these data suggest, although there is some shifting
in DEA performance among clinic peers of each type,
the relative performance of all clinics remained the same
during the study period. Focused and broad types of
clinics are located on different organization design hills
in our metaphorical landscape, with the focused design
hill much higher in performance than the broad clinic
design hills. In terms of Red Queen competitive dynam-
ics (Barnett and Sorenson 2002, Derfus et al. 2008), the
clinic managers may have run as fast as they could for
two years, and then found their performance relatively
unchanged. As we examine in the next section, this may
be a result of the endogenous efforts and capabilities of
clinic managers, and/or it may be due to exogenous pol-
icy or environmental reasons that are beyond the imme-
diate control of clinic managers.
Findings on Endogenous and Exogenous
Components of Clinic Performance
The last three columns of Table 6 report the values
of the Malmquist Index in terms of changes over time
in each clinic’s productivity that are attributable to the
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 13
Table 6 Frontier Analysis of Clinics, 1997 and 1999
MI
1997 DEA 1999 DEA Total productivity
Macro Within type Between types Within type Between types Clinic change Frontier change change (T)
Clinic ID policy type (Step 1) (Step 2) (Step 1) (Step 2) (C)(F)T =C∗F
C16 Broad 7128 7368 8442 8627 042a002 045
C14 Broad 8718 7078 10000 8608 032 006 041
C15 Broad 6469 7449 8276 8403 018 003 022
C8 Broad 6153 7641 7959 8608 02−001 019
C61 Broad 7643 7911 7962 8510 014 −001 014
C59 Broad 7321 7911 9430 8490 012 −001 012
C31 Broad 10000 6883 10000 8403 0 008 008
C19 Broad 9138 7287 8788 8608 004 004 008
C44 Broad 6727 6948 4138 8403 −01011 −001
C50 Broad 9454 7166 9195 8373 −005 002 −003
C23 Broad 10000 7911 9436 8529 −009 −001 −009
C5 Broad 8679 7911 7587 8373 −018 001 −017
C45 Broad 8464 7911 5172 8403 −02002 −019
C33 Broad 9475 7166 6970 8608 −026 004 −023
C38 Broad 9731 7132 8395 8529 −025 002 −024
C24 Broad 10000 6945 7484 8403 −027 −001 −027
C58 Broad 8170 6883 5154 8490 −061 005 −059
C51 Broad 8472 7013 5455 8608 −065 005 −063
C22 Focused 6586 9841 7237 9925 036 012 052
C52 Focused 5685 9874 7368 10000 026 013 04
C11 Focused 6027 10000 8294 9900 023 012 037
C6 Focused 5685 9870 6858 9972 002 014 016
C32 Focused 10000 10000 10000 10000 0 013 013
C42 Focused 9016 10000 10000 10000 001 012 013
C17 Focused 10000 10000 10000 10000 −006 014 006
C60 Focused 8462 10000 10000 10000 −026 014 −017
C37 Focused 7513 10000 6317 9961 −033 013 −025
C48 Focused 6018 9870 6505 10000 −045 014 −037
C1 Focused 6778 10000 4300 9950 −053 013 −047
C34 Focused 9107 9830 6951 9900 −057 012 −052
C57 Focused 10000 10000 4737 10000 −064 013 −059
C41 Focused 10000 10000 8344 9958 −068 011 −065
Notes. All values in the table represent relative productivity. The first four columns are presented in terms of percentage, within the range
of 0% and 100%, where 100% means maximal efficiency in DEA; other values are productivity relative to 100%. The last three columns are
raw values according to the MI computations but have been subtracted by 1. Thus, a positive value means an increase in efficiency, and
a negative value means a decrease in efficiency; a value of 0 means no change.
aClinics sorted by type and total productivity change (in descending order).
clinic’s endogenous efficiency and its exogenous fron-
tier. The clinic’s endogenous efficiency change consists
of the change in the clinic’s performance with respect
to its peers of the same clinic type, whereas the clinic’s
exogenous efficiency change consists of the change in
the clinic’s performance with respect to all the clin-
ics in the sample. See the appendix for computations.
The correlation between endogenous efficiency change
and exogenous frontier change is −014, which is sta-
tistically insignificant. This finding rejects Hypothesis 1.
In this sample of clinics, contrary to expectations, no
positive relationship exists between the components of
clinic performance that are due to endogenous manage-
rial efforts within the clinics and exogenous macro poli-
cies at the group level. Instead, endogenous and exoge-
nous components of clinic performance are unrelated
and tend to move in a slightly negative direction. This
low correlation may be due to the absence of a relation-
ship between group and clinic performance, or it may
be due to a washout effect of group-level policy changes
benefiting the performance of some clinics and hurting
others.
To examine these possibilities, we plot all clinics
in terms of clinic efficiency change, group frontier
change, and total productivity changes from 1997 to
1999 (i.e., the right three columns of Table 6) on a three-
dimensional (3D) graph (see Figure 3). The graph shows
that all of the more complex clinics serving a broad
patient mix lie to the left of the line separating the clin-
ics on group frontier change, and all of the less com-
plex focused clinics with a homogeneous mix of patients
lie to the right. Figure 3 shows that changes over time
in endogenous clinic performance varied from a 40%
decrease to a 60% increase among both focused and
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
14 Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
Figure 2 Scatterplot of Clinic DEA Efficiency, 1997 and 1999
82
84
86
88
90
92
94
96
98
100
102
104
68 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 100 102
1997 clinic DEA efficiency
1999 clinic DEA efficiency
Broad
Focused
C31
C33
C44
C58
C51C14 C19
C38
C50 C15
C16 C8
C5
C59
C61
C23
C34 C22
C6
C52
C1
C45
C48
broad clinics. With respect to group frontier changes,
however, it is clear that all focused clinics lie along
the positive range (from 10% to 40% increases), and
all broad clinics are in the negative range (from 0% to
20% decreases). These data show that the focused clin-
ics uniformly benefit much more from exogenous group
frontier changes than do their counterparts who serve a
broad mix of patients.
These results might lead managers to decide to change
the design of their clinics from serving a broad mix of
patients to a narrow one. This conclusion, however, does
not take into consideration the performance risks in mak-
ing such design changes. The peaks and troughs in the
changing performance landscape of the clinics in Fig-
ure 3 suggest that the performance volatility or risk of
focused and broad clinics is not the same. Changes in
the sizes and locations of performance peaks and troughs
of clinics in Figure 3 call attention to such risks. The
organizational design of focused clinics reaches a higher
performance peak, but it also has a deeper trough of
performance declines than that of the broad clinics. In
other words, the focused organizational design that has
the greatest total productivity gains also has the highest
total productivity losses over the same time period and
in the same sample of clinics.
The organization design literature has largely ignored
the risks associated with changing organizational
designs. Organizational ecologists, of course, have high-
lighted the liabilities of organization change because of
inertial forces (Stinchcombe 1965; Hannan and Freeman
1977, 1989). Organizational inertia, however, is different
from risk. Whereas inertia focuses on the difficulties or
rigidities of changing organizations, risk deals with the
likelihood of alternative consequences of changing orga-
nizational designs regardless of the difficulties of doing
so. Theories of risk are prominent in finance literature
(e.g., Gollier 2001), and volatility is the commonly used
indicator of risk in this literature. The greater perfor-
mance volatility of focused versus broad clinics shown
in Figure 3 calls for an examination of the risk–return
trade-offs of changing organizational designs. Although
Figure 3 Change in Clinic Performance, Group Frontier, and
Total Productivity
–0.6
–0.4
–0.2
0.0
0.2
0.4
0.6
0.8
–0.4
–0.2
0.0
0.2
0.4
0.6
–0.2
–0.1
0.0
0.1
0.2
0.3
0.4
Total productivity
Group frontier change
Focused clinics Broad clinics
Clinic performanc
e
change
Notes. This 3D graph uses a negative exponential smoothing tech-
nique that applies a Gaussian weight function to the data and a
quadratic fit. The diagonal line distinguishes where the broad and
focused clinics are plotted in the figure.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 15
switching from broad to focused designs may increase
the likelihood of higher total clinic productivity gains,
it comes at the risk of experiencing greater productivity
loses. As a consequence, the return-to-risk ratio of the
broad clinic design may not be inferior to that of the
focused clinic design.
The organizational complexity literature (e.g.,
Levinthal and Warglein 1999, Anderson et al. 1999,
Siggelkow 2001) that models organizational adaptive-
ness on changing landscapes using Kauffman’s NKC
model (Kauffman 1993) also does not address this risk,
for it assumes a base of zero or greater performance
to a design hill, and it does not entertain negative
valleys or troughs in a performance landscape. In effect,
these organizational simulation models right-censure
simulation data to zero or greater performance of alter-
native design hills and thereby forfeit the opportunity
to examine the risk–return trade-offs of alternative
organizational designs (Elton and Gruber 1997).
One aspect of the literature on organization design
that begins to touch on this risk–return trade-off is com-
plementarity theory, which suggests that during organi-
zation change, performance may reflect a steep decline
for several periods and then improve slowly with time
(Milgrom and Roberts 1995, Whittington and Pettigrew
2003). The peaks and troughs in Figure 3 suggest that
this curvilinear relationship may be possible if clinics
change from the broad type to the focused type. How-
ever, as Figure 2 shows, no shifts between broad and
focused clinic designs were observed during the study
period.
Our study suggests (but does not demonstrate) that
a multilevel explanation of the risk–return trade-off is
needed. Significant organizational changes may not be
within the endogenous capabilities of organizational sub-
sidiaries or subunits. Other studies indicate that exoge-
nous interventions are often needed for organizational
units to undertake radical changes (Virany et al. 1992,
Van de Ven et al. 1999). This may explain why the clin-
ics in our sample did not change from serving a broad to
a focused mix of patients. No exogenous macro-policy
decisions and resource investments occurred to make
it feasible for the clinics to undertake this transforma-
tive change. As a consequence, as we will now see, the
broad clinics did the best they could with their endoge-
nous (but limited) resources and capabilities to adapt to
changing conditions in serving their heterogeneous mix
of patients. Focused clinics did the same and performed
better collectively as a result of the performance ben-
efits they received from exogenous group-level frontier
changes.
Findings on the Model of Headquarters–Subsidiary
Relations
Finally, we examine Hypotheses 2–4 in our mul-
tilevel contingency model of headquarters–subsidiary
design. As discussed previously and illustrated in
Table 1, we expected the unit differentiation dimen-
sions (clinic autonomy and clinic–group interactions) to
influence the endogenous component of clinic perfor-
mance that is attributable to clinic managerial efforts
and the group integration dimensions (standardization
and headquarters–subsidiary integration) to influence the
exogenous component of clinic performance that is
attributable to changes in group frontier. Table 7 shows
the results of two regression analyses of these organi-
zational dimensions on the endogenous and exogenous
components of clinic performance changes. The equa-
tions also include a control variable for patient mix
(broad or focused) to capture the influence of environ-
mental complexity that featured prominently in the pre-
ceding DEA analysis.
The first regression equation in Table 7 examines
the effects of unit differentiation dimensions (clinic
autonomy and clinic–group interactions) and group inte-
gration dimensions (standardization and headquarters–
subsidiary integration) on the endogenous change in
clinic performance. Partially supporting Hypothesis 2,
the results suggest that increases in clinic autonomy have
a significantly positive effect on the endogenous change
in clinic performance, whereas group integration dimen-
sions (standardization and headquarters–subsidiary inte-
gration) have no significant effect on the endogenous
change in clinic performance. Contrary to Hypothesis 2,
however, the results suggest that clinic–group interac-
tions have a significantly negative effect on endoge-
nous change in clinic performance. As an indicator of
Table 7 Regression Results on Clinic Performance Change
and Group Frontier Change
Clinic performance Group frontier
endogenous exogenous
change change
Beta P-value Beta P-value
Unit differentiation
dimensions
Clinic autonomy 042 005 002 080
Clinic–group −065 001 −027 002
interactions
Group integration
dimensions
Standardization −040 008 −013 021
Headquarters–015 053 026 003
subsidiary
integration
Local environment
complexity
Patient mix −007 071 096 000
(heterogeneous–
homogeneous)
R2036 085
F-statistic 232 007 2347 000
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
16 Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
informal clinic–group coordination, we expected clinic–
group interactions to be a mechanism for tailoring
headquarters–subsidiary relations to serve the particu-
lar needs of local clinics, as Dooms and van Oijen
(2008) found. Instead, the data suggest that these clinic–
group interactions hampered clinics’ efforts to achieve
the highest performance changes given their resources.
The second regression in Table 7 examines the effects
of unit differentiation dimensions (clinic autonomy and
clinic–group interactions) and group integration dimen-
sions (standardization and headquarters–subsidiary inte-
gration) on the exogenous change in clinic perfor-
mance or group frontier change. Partially supporting
Hypothesis 3, the results suggest that increases in
headquarters–subsidiary integration have a significant,
positive effect on the exogenous change in clinic per-
formance, whereas clinic autonomy has no significant
effect on the exogenous change in clinic performance.
Contrary to Hypothesis 3, the results suggest that clinic–
group interactions have a significant, negative effect on
the exogenous change in clinic performance, whereas
standardization has no effect on the exogenous change
in clinic performance.
Hypothesis 4 predicted that clinic environmental com-
plexity moderates these relationships. Unfortunately, we
do not have sufficient degrees of freedom to add interac-
tion terms to our regression equations to test this hypoth-
esis. As an alternative, we examined the simple cor-
relations for clinics serving a simple (focused) versus
complex (broad) mix of patients. Table 8 shows the cor-
relations among clinic autonomy (a dimension of sub-
sidiary differentiation) and standardization (a dimension
of headquarters integration) with endogenous and exoge-
nous changes in clinic performance. The correlations are
in the directions hypothesized by our multilevel contin-
gency model. As expected, clinic autonomy is positively
correlated with the endogenous change in clinic per-
formance and negatively correlated with the exogenous
change in clinic performance or group frontier change
for complex clinics, but it is uncorrelated for simple clin-
ics. Conversely, group standardization is strongly corre-
lated with both endogenous and exogenous performance
changes for simple clinics, whereas these correlations
are very low for complex clinics.
Table 8 Correlations of Clinic and Group Design and
Performance Measures
Clinic performance Group frontier
endogenous change exogenous change
Complex Simple Complex Simple
(broad) (focused) (broad) (focused)
Clinic autonomy 034 −0002 −034 007
Standardization −013 044 −001 035
These results remind us of the principle of oppo-
site part–whole relationships (Simmel 1955, Dahren-
dorf 1979, Astley and Van de Ven 1983). Many orga-
nization design problems and relationships manifest
themselves in different and contradictory ways at differ-
ent organizational levels. At the micro level, the focus
is on the particularistic needs of patients served by local
community clinics and on the autonomous discretion of
clinicians to decide how best to enable this. This clinic
autonomy does not necessarily advance macro group-
level objectives where the focus is on strategic policies,
structural arrangements, and building an integrative cul-
ture. Moreover, the substantive effects of clinic design
factors can be different for different kinds of clinics,
depending on their structure and patient mix.
Concluding Discussion
This paper advances a multilevel organizational contin-
gency theory that takes into account the different impact
of headquarters policies and subsidiary managerial efforts
on subsidiary performance. Our paper builds on the
contingency theory view of designing headquarters–
subsidiary relations by proposing that organization per-
formance increases when subsidiary units are designed
to differentially fit their local environments subject to
corporate constraints. An important contribution of this
research is that it addresses three thorny problems with
this contingency theory proposition: (1) satisfying mul-
tiple and often conflicting dimensions of organizational
context, design, and performance at micro and macro
levels; (2) disentangling what parts of the performance
of organizational units are attributable to endogenous
and exogenous factors; and (3) examining dynamic pat-
terns of organizational adaptation on changing fitness
landscapes over time. These problems have hampered
advancements in organizational contingency theory in
general and headquarters–subsidiary relationships in par-
ticular. To move beyond the limits of armchair theorizing,
we take an empirical approach, using frontier analysis, to
deal with these challenges.
Based on our longitudinal study of a large medical
group practice of 32 local community clinics, we used
data envelopment analysis to determine the clinics that
best achieve a set of desired performance criteria subject
to their resource constraints and environmental complex-
ity. This provides an analytical way to address the first
problem of identifying comparable organizational units
and determining how well they satisfy multiple and often
conflicting environmental demands and performance cri-
teria. It also provides a direct way to empirically identify
the best-fit ideal-type units that are necessary for test-
ing contingency theory. DEA does this by identifying
the outlying most adaptive or fit units in a sample that
define the contours of a best-performance frontier and
from which the relative distance of other less adaptive
units can be determined.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 17
We show how the DEA method of frontier analysis
provides a systematic way to disentangle endogenous
and exogenous components of unit performance. This
is crucial for addressing the second thorny problem in
designing headquarters–subsidiary relations—difficulties
in assessing fit across micro and macro levels of organi-
zation design. Organization change and performance are
simultaneously subject to macro and micro forces that
may differ in direction and degree. Without a method
of partitioning unit performance into its endogenous and
exogenous components, one cannot study the relative
influence and interactions of units at different levels in
a nested organizational hierarchy. We have shown how
the two-step DEA process and the Malmquist Index pro-
vide ways of partitioning unit adaptiveness at a given
time and performance changes over time into endoge-
nous and exogenous components. This also addresses
the third problem of examining the dynamic patterns of
organizational adaptation on changing fitness landscapes
over time.
This methodology provides the opportunity to make
an important substantive contribution of developing
and examining a multilevel model of headquarters–
subsidiary relations. The model predicted that clinic
autonomy and interactions with the group would most
directly influence the endogenous component of clinic
performance change, whereas group standardization and
headquarters–subsidiary integration would predict the
exogenous component of clinic performance change.
We also hypothesized that local environmental com-
plexity moderates these relationships. The data provided
some support for the model. As expected, clinic auton-
omy was a significant positive predictor of endoge-
nous clinic performance change but was unrelated to
exogenous group frontier change. From a macro view,
headquarters–subsidiary integration or shared values was
a significant positive predictor of exogenous group fron-
tier change but was unrelated to endogenous clinic per-
formance change. Contrary to expectations, however,
standardization was not a significant predictor of group
frontier change, and clinic–group interactions was a
significant, negative predictor of both clinic endogenous
and exogenous performance change.
Overall, however, these findings clearly show that
different organizational dimensions influence different
sources of organizational performance under different
environmental conditions. This finding is important,
because the literature tends to assume that all organiza-
tional design variables must fit together both internally
and externally (Donaldson 2001). For example, configu-
ration theory proposes that organizational structure, sys-
tems, culture, incentives, and strategies must all be inter-
nally coherent and fit environmental demands (Meyer
et al. 1993, Nadler and Tushman 1999). Our research
findings suggest that a more complex model is needed—
one that anticipates how different organizational charac-
teristics are designed to meet different criteria or com-
ponents of organizational performance.
We found that the correlation between the endoge-
nous component of clinic performance, or efficiency
change, and the exogenous component of clinic per-
formance change, or frontier change, was only −014.
Although not statistically significant, this finding calls
into question the commonplace expectation that micro-
and macro-organizational performance are positively
related in a complementary manner rather than being
independent of each other or negatively related such
that one level is sacrificed for the other. We found that
the small correlation was the result of a cancelling-
out effect, where some clinics were helped and oth-
ers hurt by changes in the group performance frontier
over the two-year study period. We observed that per-
formance changes in group frontier only benefited the
focused clinics serving a homogeneous mix of patients
and not the general clinics serving a heterogeneous mix
of patients. The latter gained no performance benefits
from macro-policy frontier changes over time. By con-
trast, focused clinics gained significant productivity ben-
efits from both exogenous frontier and endogenous clinic
changes.
This finding has important implications for organiza-
tional performance appraisals. Because the benefits and
costs of corporate policies on organizational subunits
are seldom known or turn out as intended, they also
tend to be overlooked in performance assessments of
organizational units. This is especially so when it is
difficult to attribute what aspects of observed changes
in the performance of organizational units are due to
macro-organizational policies or unit-specific factors. As
a result, the units benefiting from organization-wide
policies tend to be unfairly rewarded for performance
improvements that are not the result of their own efforts,
whereas other units disadvantaged by uniform policies
tend to be disproportionately reprimanded for perfor-
mance declines produced through no fault of their own.
During the study period, we found that clinics with
a focused design showed significantly higher and lower
performance changes than clinics serving a broad mix
of patients. The organizational designs with the high-
est performance peaks also have the lowest performance
declines. This volatility of performance changes is illus-
trated in Figure 3, from which we drew the inference
that some organization designs are riskier than others
on a changing performance landscape. This volatility of
performance changes calls attention to the risks asso-
ciated with changing organizational designs. We noted
that whereas organizational inertia (i.e., the difficul-
ties of change) has received considerable attention, the
risks (i.e., likely performance consequences) of chang-
ing organizational designs have been largely ignored.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
18 Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
We think an important future direction in studying orga-
nizational change is to incorporate theories of risk in
designing organizations.
Donaldson (1999) has made an effort in this direction.
Drawing on portfolio theory in finance, he proposed that
certain environmental factors (e.g., business cycle) and
internal organizational characteristics (e.g., diversifica-
tion) affect the level of performance risk (performance
fluctuation) and that a greater level of risk makes an
organization more likely to change. Donaldson (1999)
also links risk across organizational levels by suggesting
that a subsidiary’s performance risk, which arises from
local industry characteristics, contributes to the risk of
the overall organization.
Following Donaldson (1999), we think that designing
headquarters–subsidiary relations requires consideration
of the different risks and returns of design changes at dif-
ferent organizational levels. For example, we observed
no instance of any clinic in our sample changing from
serving a broad to a focused mix of patients. In terms
of our fitness landscape metaphor, broad and focused
clinics occupy different design hills and compete with
one another to be the “king or queen of their hill,” but
they did not engage in hill jumping. This last action rep-
resents a major design change that may not be within
the endogenous capabilities of organizational units. Clin-
ics may not have changed their designs because no
exogenous macro-policy decision or resource investment
occurred to make it feasible for the clinics to undertake
transformative change. Unlike many organizations where
headquarters adopt contingent policies that are imple-
mented differently in selective subsidiaries, the medical
group examined here focused on developing and imple-
menting its macro policies uniformly across all of its
clinics.2
The fact that no clinic-contingent policy interventions
were observed, however, does not imply that the medical
group had no effects on its clinics or that the observed
associations are just random perturbations. The latter, of
course, is the null hypotheses that we believe has been
rejected by our study data. The Field Research Setting
section describes numerous headquarters policies and
procedures that the medical group implemented in all of
its clinics during the study period. Although we cannot
measure the effects of individual group interventions on
clinics, we argue that the cumulative effects of all macro-
group interventions produced the changes observed in the
group performance frontier. Our statistical analysis shows
that this group frontier change benefited focused clinics
and unintentionally hurt the performance of broad clin-
ics. Moreover, the regression analyses indicate that the
clinic–group relationships were not totally random per-
turbations. Indeed, our measures of clinic–group relations
explain 36% and 85% of the variation in clinic endoge-
nous and exogenous performance changes, respectively
(shown in Table 7).
Finally, although we believe our study findings gen-
eralize to a theory of headquarters–subsidiary relation-
ships, a major limitation is that they are limited to, and
cannot be generalized beyond, the sample of organi-
zational units observed. Our sample of medical group
clinics may be unique in certain respects (such as the
group’s uniform treatment of clinics and the different
demands on clinics serving a broad versus focused mix
of patients). In addition, our small sample of 32 clinics
of one group practice limited abilities to systematically
examine the organizational design configurations of clin-
ics moving on and off the frontier over time. We think
that a limited generality of research findings is a neces-
sary trade-off for developing a more penetrating under-
standing of organization design in real-world settings.
As other studies in different contexts accumulate, meta-
analysis projects can provide opportunities for examin-
ing the generality of research findings across samples
and contexts of organizations.
Acknowledgments
The authors recognize contributions to this longitudinal
study by prior research associates: David Grazeman, Stuart
Bunderson, Jeffrey Thompson, Frank Schultz, Shawn Lof-
strom, Russel Rogers, Rhonda Engleman, and Jisun Yu. The
authors appreciate the useful comments on earlier drafts of
this paper from Rich Burton, Tim Hoff, Dan Levinthal, Bill
McKelvey, Michael Pratt, Nicolaj Siggelkow, Kingshuk Sinha,
Sharon Topping, and Doug Wholey, as well as participants at
the organization design conference at Harvard Business School
(December 2008), the Health Organization Research Associ-
ation conference in Chapel Hill (June 2009), the Center for
Innovation Research at Tilburg University in the Netherlands
(October 2009), and the Organization Science Winter Confer-
ence in Steamboat Springs, CO (February 2010).
Appendix. Computation of DEA Efficiency
and Malmquist Index
DEA is a nonparametric frontier estimation method that was
developed by Charnes and Cooper (1962), Charnes et al.
(1978, 1994), and Banker et al. (1984, 1989). DEA contrasts
with stochastic frontier analysis (SFA), which is another major
form of frontier analysis. Most importantly, SFA assumes
that some random elements in an efficiency analysis follow
stochastic distributions and can be specified; DEA is determin-
istic and does not make this assumption. Thanassoulis (2001)
provides an informative introduction to DEA.
DEA computes the performance (or efficiency) of the
DMUs (decision-making units) in question with the following
equation:
Maximize Eu=s
r=1yruOru
m
i=1xiuIiu
(2)
where urepresents the units of DMU; Erepresents perfor-
mance; Iand Orepresent all inputs and outputs, respectively,
for each DMU; and xand yrepresent the weights assigned to
each input and output, respectively.
Two constraints are placed on Equation (1) to allow for opti-
mization and comparison purposes: (a) the assigned weights
cannot be negative, and (b) any assigned weight for a specific
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS 19
DMU can be applied across all DMUs so that it will not lead
any DMU to achieve an efficiency ratio greater than 1. These
constraints also avoid infinite results stemming from weight
assignments. Stated formally,
xru ≥0r=1s
yiu ≥0i=1m (3)
and s
r=1yrkOrk
m
i=1xikIik ≤1 for all DMUs. (4)
Given the above constraints, DEA uses a linear program-
ming algorithm to compute xand y(the weights). Specifically,
we adopted a nonparametric, deterministic, output-based,
variable-returns-to-scale specification of the DEA model.3
This DEA model enables us to identify the clinics that max-
imize multiple outcomes (patient and business care) with the
least amount of resource inputs (clinic size). The first step
of this computed DEA score identifies the efficiency of each
clinic relative to its comparable broad or focused cohort. The
second DEA step estimates the relative efficiency all clinics
together across all cohorts.
The Malmquist Index
Best-performance frontiers are not static; they change over
time (Sinha 1996). This implies that the frontier is a moving
target and that clinics move on and off the shifting frontier
over time. Longitudinal analysis of DMUs (e.g., health-care
clinics) involves the use of the Malmquist Index, developed
by Färe et al. (1994). The MI assesses productivity change
by considering the efficiency of a DMU (in comparison with
other DMUs) in two different time periods according to the
following equation:
MI =Efficiency of DMUjat Time2with respect to Time1
Efficiency of DMUjat Time1with respect to Time1
×Efficiency of DMUjat Time2with respect to Time2
Efficiency of DMUjat Time1with respect to Time21/2

This equation computes a geometric mean of the efficiency
change of a particular DMU at two different time points. In the
above formula, the two terms “Efficiency of DMUjat Time1
with respect to Time1” and “Efficiency of DMUjat Time2
with respect to Time2” are relatively straightforward. In our
case, they are simply the DEA scores of a particular clinic in
1997 and 1999.
Methodological requirements of computing the MI assume
constant returns to scale. Although this is technically different
from Steps 1 and 2 (see the DEA Analyses Results section), it
does not affect our analysis substantively. The other two terms
“Efficiency of DMUjat Time2with respect to Time1” and
“Efficiency of DMUjat Time1with respect to Time2” evaluate
the efficiency of a focal clinic in one year with respect to all
clinics (including the focal clinic itself) in another year.
For an example of how the MI is computed, we examine
the productivity change of clinic C1 in our sample. According
to the above equation, we conduct four DEA runs and obtain
the following four measures regarding clinic C1’s efficiency:
Efficiency at 1997 (Time 1) with respect to 1997 (Time 1):
28.50.
Efficiency at 1999 (Time 2) with respect to 1999 (Time 2):
13.50.
Efficiency at 1997 (Time 1) with respect to 1999 (Time 2):
25.10.
Efficiency at 1999 (Time 2) with respect to 1997 (Time 1):
15.10.
The MI for clinic C1 is therefore
1510
2850 ×1350
2510 =053
To facilitate interpretations, we subtract 1 from this num-
ber. Thus, total productivity change for clinic C1 is now
−047. The negative sign indicates that clinic C1 experienced
a decline in total productivity between 1997 and 1999. We
perform this simple transformation for all the clinics in our
sample, and the values are presented in the last column of
Table 6.
Mathematically, MI can be decomposed into two compo-
nents (that which is attributable to the DMU’s endogenous
efficiency change and that attributable to an exogenous fron-
tier policy or environmental change) (Färe et al. 1994). The
first component is the ratio between Efficiency of DMUjat
Time1and Time2:
Efficiency of DMUjat Time2with respect to Time2
Efficiency of DMUjat Time1with respect to Time1
or clinic efficiency change
The second component has a form very similar to the MI
index, but there is an important difference: the denominator
of the first term and the numerator of the second term inter-
change:
Efficiency of DMUjat Time2with respect to Time1
Efficiency of DMUjat Time2with respect to Time2
×Efficiency of DMUjat Time1with respect to Time1
Efficiency of DMUjat Time1with respect to Time21/2
or frontier change
The Malmquist Index is equal to the multiplication
between these terms: MI =clinic efficiency change ×
frontier change. In the case of clinic C1, its endoge-
nous efficiency is 1350/2850 =047. Frontier change is
√1510/1350 ×2850/2510 =113. Thus its MI is 053 =
047 ∗113. As mentioned previously, we subtract 1 from the
original value. Thus, a positive MI value indicates an increase
in productivity, and a negative value indicates a decrease in
productivity. The last three columns of Table 6 report the val-
ues of MI in terms of the total productivity change, the clinic’s
endogenous efficiency change, and the exogenous frontier
change for all clinics.
Endnotes
1Because of missing data and combined organizational
accounting statements for some clinics, some of the clinics in
our initial sample had to be removed from this analysis.
2Including such contingent macro policies increases the
complexity of correctly specifying a DEA model of
headquarters–subsidiary relations. However, the steps in the
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.

Van de Ven et al.: Changing Organizational Designs and Perfor mance Frontiers
20 Organization Science, Articles in Advance, pp. 1–22, © 2011 INFORMS
DEA methodology would be similar to our treatment of broad
and focused clinics in this study.
3There are, of course, other ways to compute unit efficiency.
Bryce et al. (2000) compare three commonly used approaches:
DEA (as described here), stochastic production frontier (see
Aigner et al. 1977), and fixed-effects regression. Based on
data from 585 HMOs operating from 1985 to 1994, Bryce
et al. find that the results from the three methods identify the
same industry trends, and those correlations of individual-unit
efficiency scores from the three methods vary from 0.67 to
0.79. Although these results show high agreement, Bryce et al.
caution that the results are not identical because, indeed, the
different methods are designed for different purposes.
References
Adler, P. S., B. Borys. 1996. Two types of bureaucracy: Enabling and
coercive. Admin. Sci. Quart. 41(1) 61–89.
Aigner, D., C. A. K. Lovell, P. Schmidt. 1977. Formulation and
estimation of stochastic frontier production function models.
J. Econom. 6(1) 21–37.
Anderson, P., A. Meyer, K. Eisenhardt, K. Carley, A. Pettigrew. 1999.
Introduction to the special issue: Applications of complexity the-
ory to organization science. Organ. Sci. 10(3) 233–236.
Astley, W. G., A. H. Van de Ven. 1983. Central perspectives
and debates in organization theory. Admin. Sci. Quart. 28(2)
245–273.
Banker, R. D., A. Charnes, W. W. Cooper. 1984. Some models for
estimating technical and scale inefficiencies in data envelopment
analysis. Management Sci. 30(9) 1078–1092.
Banker, R. D., A. Charnes, W. W. Cooper, J. Swarts, D. A. Thomas.
1989. An introduction to data envelopment analysis with some
of its models and their uses. J. L. Chan, J. M. Patton, eds.
Research in Governmental and Nonprofit Accounting, Vol. 5.
JAI Press, Greenwich, CT, 125–163.
Barnett, W. P., O. Sorenson. 2002. The red queen in organizational
creation and development. Indust. Corporate Change 11(2)
289–325.
Birkinshaw, J. 2008. Strategy and management in MNE subsidiaries.
A. M. Rugman, T. L. Brewer, eds. The Oxford Handbook
of International Business, 2nd ed. Oxford University Press,
Oxford, UK, 380–401.
Blau, P. M. 1964. Exchange and Power in Social Life. John Wiley &
Sons, New York.
Bliese, P. D. 2000. Within-group agreement, non-independence, and
reliability: Implications for data aggregation and analysis. K. J.
Klein, S. W. J. Kozlowski, eds. Multilevel Theory,