Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Academy of Management is collaborating with JSTOR to digitize, preserve and extend access to The Academy
of Management Journal.
http://www.jstor.org
JULIAN BIRKINSHAW
London Business School
209
FIGURE 1
Relationships Predicted
Performance
Unit
Management Alignment
Performance
Social Context Adaptability
seemingly contradictory tensions. For example, in ment. It also avoids the coordination problems be
the manufacturing literature flexibility and cost tween subunits that were identified above. A fur
efficiency are no longer thought of as contradictory ther implication is that although ambidexterity is a
(de Meyer, Nakane, Miller, & Ferdows, 1989; Mac characteristic of a business unit as a whole, itman
Duffie, 1995). Achieving both may involve task ifests itself in the specific actions of individuals
within a single business unit, whereby, the organization. In their day-to-day
partitioning throughout
for example, one group adopts an "organic" struc work, individuals often face choices as to how they
ture while another takes on a "mechanistic" struc to
should spend their time?should they continue
ture (Adler, Goldoftas, & Levine, 1999; Hedlund & focus on an existing customer account to meet
& Leifer,
Ridderstrale, 1997; McDonough 1983). quota, or should they nurture a new customer who
Another structural solution put forth is temporal has a slightly different need? In business units that
separation, a system in which an entire unit fo are aligned or adaptive, individuals are given clear
cuses on one set of tasks one day, then on a differ and receive incentives
instructions only for those
ent set of tasks the next (Adler et al., 1999; Duncan, activities that support either alignment or adapta
1976; McDonough & Leifer, 1983). Both these ap tion. But in a contextually ambidextrous unit, the
proaches allow the competing demands for adapt context is dynamic and flexible to allow
enough
ability and alignment to be met within a single use own as
individuals to their judgment to how
business unit (which removes many of the coordi their time between
they divide alignment-oriented
nation costs noted above), but they still rely on and adaptation-oriented and both are
activities,
structural solutions that require unit managers to
valued and rewarded. In short, the systems that are
judge how best to divide up work groups and/or
at the business-unit level encourage
developed
periods of time to meet those different needs. Thus, ambidextrous behavior that is both aligned and
to date, scholars have focused on structural ambi
to adaptable.
dexterity, developing structural mechanisms
Although this is the first study to develop the
cope with the competing demands faced by the
concept of contextual ambidexterity, the logic used
organization for alignment and adaptability.
here is consistent with that of several earlier stud
In this article, we develop a somewhat different
ies. Adler and colleagues (1999) referred to two
perspective, focusing instead on contextual ambi
specific mechanisms for reconciling the inherent
We concur with previous authors on the
dexterity. tension between and flexibility that rely
idea that ambidexterity is an efficiency
organization's capac on individual own
employees to make their choices:
ity to simultaneously achieve alignment and adapt
a single business (1)meta-routines for systematizing the creative pro
ability within unit, but we suggest
cess and (2) job enrichment schemes that enable
that it is best achieved not through structural, task,
workers to become more innovative and flexible in
or temporal separation, but by building a business
their routine tasks. Likewise, Hedlund and Ridder
unit context that encourages individuals to make
strale (1997) discussed the role of "renaissance
their own judgments as to how best divide their
time between the conflicting demands for align company men" in international firms, people who
con are simultaneously responsible for exploitation
ment and adaptability. Thus, in other words,
can be viewed as ameta-level and creation-oriented activities. In both cases,
textual ambidexterity
these studies point to the need for a behavioral
capacity (for alignment and adaptability) that per
meates all functions and levels in a unit, rather orientation toward dual
capacities, rather than a
than as a "dual structure" separation of those And
(Duncan, 1976) in which higher-level capacities.
the two demands are kept more importantly, they also imply that organiza
separate. Essentially,
contextual is a multidimensional tions must build systems and processes that facili
ambidexterity
with and adaptability each tate the behaviors.
construct, alignment
a but nonsubsti To summarize, we have defined contextual am
constituting separate, interrelated,
tutable element. When contextual bidexterity as an interplay of system capacities?
ambidexterity
in a unit can for alignment and simulta
has been achieved, every individual adaptability?that
deliver value to existing customers in his or her neously permeate an entire business unit, and we
own functional area, but at the same time every have noted that our interest lies in understanding
individual is on the lookout for changes in the task both the antecedent conditions that give rise to
environment, and acts This is poten contextual ambidexterity and the consequences of
accordingly.
tially a more sustainable model than structural sep ambidexterity for business-unit performance. Al
aration because it facilitates the adaptation of an though there is no lack of research concerned with
entire business unit, not just the separate units or reconciling internal organization tensions, no prior
functions responsible for new business develop study has explicitly focused on the measurement,
antecedents, and consequences of organizational terity through the means we discuss below is less
contextual ambidexterity as defined here. expensive than more traditional structural solu
tions because the costs of controlling and supervis
ing employees are much reduced. Thus, our first
Ambidexterity and Performance
hypothesis is:
When conceptualized in the way described
we all the more Hypothesis 1. The higher the level of ambidex
above, ambidexterity is, argue,
to sustainable terity in a business unit, the higher the level of
conducive performance. Alignment
activities are geared toward improving perfor performance.
mance in the short term. Adaptability activities are
well as the set of management practices and behav who possess and are seen to possess required capa
iors that exemplify and reinforce those basic prin bilities contribute to the establishment of trust.
ciples" (1990: 2). Climate has been described as a Ghoshal and Bartlett (1994) conceptualized these
presentation of organizational stimuli or environ four attributes?discipline, and
stretch, support,
mental characteristics presumed to affect individ trust?as interdependent. An
organization, they ar
ual behavior and attitudes (Lewin, Lippitt, &White, gued, needs to foster discipline and stretch to en
1939). Researchers have
subsequently distin courage individuals to push for ambitious goals,
guished between organizational climate and psy but it also needs support and trust to ensure that
chological climate, which consists of individual this a cooperative
within environment.
happens
interpretive perceptions (James & Jones, 1974; Organization in other
context, words, can be con
Klein & Koslowski, 2000). Importantly, climate re ceptualized in terms
of "the yin and yang of con
searchers have come to view organization climate tinuous self-renewal" (Ghoshal & Bartlett, 1997:
as an objective higher-level phenomenon. Al 151): a balance between a pair of hard elements
though individual perceptions of an organization (discipline stretch) and a pair of soft elements
and
climate may be used to assess it, if these percep (support and
trust). Too much emphasis on disci
tions are homogenous, they can be aggregated to pline and stretch creates burnout and disillusion
represent climate as a property of an organization ment among employees, but too much emphasis on
(Klein & Koslowski, 2000). Our broad notion of support and trust creates a "country club" atmo
context no work gets done.
organization encompasses these elements; sphere in which
it reflects a combination of the structural context, Ghoshal and Bartlett (1994) did not argue explic
culture, and climate of a business unit and is con itly that these contextual features will develop the
sidered an objective, higher-level attribute of the capacity for contextual ambidexterity. Rather, they
unit as a whole. described discipline, stretch, support, and trust as
This view coincides with that
put forth by engendering individual-level behaviors that result
Ghoshal and Bartlett (1994: 95), who defined or in initiative, cooperation, and learning. But accord
ganizational context in terms of four behavior ing to Ghoshal and Bartlett, individuals take these
framing attributes: discipline, stretch, support, and actions of their own volition. A context does not
trust. These attributes are created and reinforced by dictate specific types of action; rather, it creates a
a variety of micro- and macro-level actions taken by supportive environment that inspires an individual
managers in a business unit. Discipline induces to do "whatever it takes" to deliver results. Thus,
members to voluntarily strive to meet all expecta we extend their framework arguing that when
by
tions generated by their explicit or implicit com a supportive organization context is created, indi
mitments. Establishment of clear standards of per viduals engage in both exploitation-oriented ac
formance and behavior, a system of open, candid, tions (geared toward alignment) and exploration
and rapid feedback, and consistency in the appli oriented actions (geared toward adaptability), and
cation of sanctions contribute to the establishment this results in contextual ambidexterity, which sub
of discipline. Stretch is an attribute of context that sequently enhances performance.
induces members to voluntarily strive for more, Evidence from the qualitative interviews we con
rather than less, ambitious objectives. Establish ducted with members of the business units in our
ment of a shared ambition, the development of a sample (described below) support these ideas. For
collective identity, and the ability to give personal example, in three of the most highly ambidextrous
meaning to the way in which individuals contrib business units, there was evidence of a context rich
ute to the overall purpose of an organization con in support and trust for many years, which over the
tribute to the establishment of stretch. Support in last five years had been complemented with a
duces members to lend assistance and countenance greater focus on discipline and stretch. Until 1990,
to others. Mechanisms that allow actors to access employees had viewed the company in question as
the resources available to other actors, freedom of a comfortable and secure place to work, with an
initiative at lower levels, and senior functionaries informal atmosphere. Over the last decade, a num
giving priority to providing guidance and help ber ofchanges were brought about, primarily
rather than to exercising authority contribute to the through top-down initiatives around cost reduction
establishment of stretch. Finally, trust is an at and quality, and through greater focus on key stra
tribute of context that induces members to rely on tegic objectives and personal commitment to those
the commitments of each other. Fairness and equity objectives. One respondent commented that this
in a business unit's decision processes, involve business unit was run as a "commando-type organ
ment of individuals in decisions and activities af ization?appraisal and evaluation interviews are
fecting them, and staffing positions with people run in a pyramidal form and compensation is [now]
geared towards short-term objectives." Most of Likewise, more support cannot substitute for a lack
these changes were instituted through a new exec of discipline. Thus, ambidexterity is achieved
utive team that gave people more structure, which when all four of the elements characterize a busi
led to a focus on new products and new opportu ness unit:
nities as a means of delivering on the more ambi
tious goals. The net result was that the imposition Hypothesis 2. The more that a business-unit
of greater discipline, and more top-down direction,
context is characterized by an interaction of
whereas before the stretch, discipline, support, and trust, the
generated greater adaptability,
unit had been in a relatively aimless higher the level of ambidexterity.
evolving
i
fashion.
As a second example, in a business unit of a
Mediation Effects
North American oil and gas firm, the organization
context was clearly very balanced, and itworked in Finally, we argue that contextual
ambidexterity
an autonomous or bottom-up manner. Their "loose mediates the between
relationship the four at
tight" model ensured that stretch and discipline tributes of organization context and subsequent
were built into performance targets. Trust and sup business-unit performance. That is, the attributes of
port emerged in a subtle way. For example, trust context influence performance through the devel
emerged through tangible examples of individuals opment of ambidexterity. When ambidexterity has
not for well-intentioned business not been developed (that is, when an organization
being punished
failures. Support manifested itself in the use of IT has not developed the simultaneous capacities for
systems to increase knowledge of what was hap alignment and adaptability), the context character
pening in other parts of the business, and various istics, in and of themselves, may or may not influ
forums and councils for cooperating and sharing ence performance. The reason for hypothesizing a
best practices. But for the most part, support mediating effect is that contextual ambidexterity is
through the enlightened seen as a meta-capability that is developed gradu
emerged spontaneously,
self-interest of individuals who knew they could ally over time through the interaction of the var
not get things done on their own. ious features of an organization context. As both
In the (and poorly Ghoshal and Bartlett (1994) and Adler and coau
low-ambidexterity perform
in contrast, there was evidence thors (1999) showed, the development of this sort
ing) business units,
that the organization contexts were weak or incom of capability takes many years. Stated slightly dif
plete. For example, in several business units there ferently, it would be wrong to suggest that a com
was evidence of inconsistent messages from top pany could simply institute the four attributes of
management (which undermined trust), and a organization context and expect them to deliver
sense that the business lacked the ambition or focus superior performance. Rather, the four attributes
needed to generate stretch. As one manager said, shape individual and collective behaviors that in
"There is no overarching vision; each division de turn shape business-unit capacity for contextual
vises its own vision and objectives." In several ambidexterity, and it is the ambidexterity that leads
other business units, there was evidence of a lack of to superior performance.
followthrough when management systems were A useful analogy to contextual ambidexterity is
used, so it was hard to create discipline. For exam the market orientation construct in the field of mar
ple, there had been a number of new initiatives, keting (Deshpande, 1999; Jaworski & Kohli, 1993;
which, according to one manager "had lost ac Narver & Slater, 1990). Scholars have developed
countability and steam" within less than a year. reliable and valid measures of market orientation,
systems providing training, feedback, and and they have shown that market orientation is
Support
information across the functions could be identi associated with superior performance. But the pro
fied in both of these low-ambidexterity units, but cess of developing market orientation in a firm is
were insufficient on their own to develop an recognized to be complex, time-consuming, and
they
effective context. causally Research has identified some
organizational ambiguous.
In view of the previous literature and this evi of the necessary systems and techniques needed
dence, we argue that stretch, support, (such as high-quality market intelligence), but
discipline,
and trust are interdependent, fea these techniques do not have a direct effect on
complementary
tures of organization context that are nonsubstitut performance; rather, they contribute to the overall
able, and therefore all four must be present in order market orientation of a firm, which then leads to
for a business unit to become ambidextrous, and performance.
subsequently, to perform well. In other words, The mediating effect of contextual ambidexterity
more stretch cannot substitute for a lack of trust. also occurs because the attributes of context them
selves can create and amplify internal tensions if commented, "Moving at this high rate of speed
they do not contribute to the simultaneous capaci makes it impossible to maintain formal processes.
ties for alignment and adaptability that comprise Instead a lot of people are making unilateral deci
ambidexterity. For example, the more that manag sions." Alignment, in turn, was achieved through
ers focus on discipline, the less they are seen to be clear objectives, goal-setting programs, and incen
supportive of risk taking and the less trusting is the tive systems that supported adaptability. There
climate, inhibiting learning (Edmondson, 2001). In were several executive forums in which senior
a new product development context, Leonard-Bar managers came together to ensure their strategy
ton (1992) found that inconsistencies between core was consistent. "Employees in all lines of business
capabilities and innovation demands frequently have a clear idea of the company's objectives,"
led to more intense use of extant strengths, result observed one manager. It was the simultaneous
ing in teams repeatedly missing opportunities for achievement of alignment and adaptability through
creative breakthroughs. Experimentation reduced the development of systems that fostered the
the risk of testing new frames and practices, but stretch, discipline, support, and trust that enabled
facilitating experimentation required that systems this business unit to perform.
and processes in the organization develop a sup We also the interview from
analyzed transcripts
portive and trusting context. Hatch and Ehrlich business units that were the least ambidextrous
(1993) detailed similar struggles. To fulfill increas (and the most poorly performing) for clues regard
ing demands for profit, managers sought to protect ing the mediating effects of ambidexterity. One
corporate assets from being squandered, yet imple common theme to emerge was that the business
menting security mechanisms contradicted their units were suffering from inconsistent manage
messages of trust and cooperation. O'Conner (1995) ment. They had
typically embraced popular man
described related behaviors that occurred during agement practices such as incentive-based pay or
the implementation of participative management. goal setting, but only did so for a while before
Top management called for midlevel managers' in moving on to other initiatives. As a result, even
volvement, while at the same time setting strict though the ratings for certain contextual attributes
limits on their discretion. Such mixed messages were high, they had not been adopted in combina
increased the resistance and alienation of the tion with other attributes, and they were not left in
midlevel managers as they questioned their new place long enough for the capacity for
higher-order
roles. ambidexterity to emerge. Other business units had
A potential strategy for addressing these possible clearly failed to adapt quickly to changing market
contradictions between discipline and stretch on conditions, and to some extent were living off their
the one hand, and support and trust on the other, is successes of ten years earlier. Thus, reflecting the
the development of contextual ambidexterity at the arguments above and evidence from our inter
business-unit level. In support of this proposition, views, our third hypothesis is:
Denison and colleagues (1995) found that effective
leaders displayed behavioral Hypothesis 3. Ambidexterity mediates the re
complex repertoires
that simultaneously fostered consistency, lationship between context?as captured by
stability,
and as well as and won the interaction of discipline, stretch, support
control, passion, courage,
der. Finally, Lewis argued that in the end, manag and trust?and business-unit performance.
ing tensions "denotes not compromise between
flexibility and control, but awareness of their si
. .
METHODS
multaneity. emphasizing the coexistence of au
thority and democracy, discipline and empower Previous studies in this genre have typically ei
ment, and formalization and discretion" (2000: ther adopted a single-case-study or
methodology,
770). they have relied on single informants to answer
The qualitative interviews with members of busi questions on behalf of an entire organization, but
ness units in our sample corroborated this evidence both have obvious limitations. Our ap
approaches
and argument. One highly ambidextrous business proach, in contrast, was to ask a large sample of
unit was responsible for consulting services in a individuals to rate their business units on both
U.S. software firm. Here the emphasis was on context and ambidexterity, and to then aggregate
adaptability?the unit's "remarkable ability to turn their responses to create unit-level measures. This
on a dime." This was achieved,
according to re procedure was undertaken in 41 business units,
spondents, through "hiring very smart people," set each of which had distinct contexts. This sample
ting aggressive but not unrealistic targets, and was sufficient to allow statistical analysis at the
avoiding too much formalization. As one manager business-unit level.
Our procedures consisted of: (1) interviews with All constructs were measured with multi-item
top executives in ten multinational firms, (2) inter scales. Scores on these measures were means cal
views in 2-7 business units in each firm, (3) a culated across items. We based our survey items on
survey of a stratified random sample of 50-500 previous research and on input from an expert
employees at four hierarchical levels in each busi panel of academics and pretested them on a small
to ensure were
ness unit (identified using a random number gen sample of managers that meanings
erator and rosters), (4) identifying and clear. Using our final sample, we conducted nu
employee
each business unit's key context merous analyses (described below) to verify that
understanding
characteristics through qualitative analysis of inter our measures were sound.
view notes
and quantitative analysis of survey data, Performance. The dependent variable was mea
and (5) feedback sessions in each firm. Here, we sured with four items that required the senior and
report primarily on the quantitative data. The total middle management respondents to reflect on per
number of survey respondents was 4,195 individ formance over the last five years and indicate the
uals from units in the ten multinational
41 business degree to which they agreed with the following
firms. Table 1 provides a breakdown of the sample. (for this and for subsequent measures, item factor
Precautions to avoid same-source bias. We col "loadings" are in brackets): (1) "This business unit
lected the data using a comprehensive survey, and is achieving its full potential" [.76], (2) "People at
all items re my level are satisfied with the level of business
required seven-point Likert-style
sponses. The stratified random sample of respon unit performance" [.84], (3) "This business unit
dents from four levels of each business unit does a good job of satisfying our customers" [.81],
ratings of organization context, ambidex and (4) "This business unit gives me the opportu
provided
and To mitigate the problem of nity and encouragement to do the best work I am
terity, performance.
same-source bias, we used different levels of re capable of" [.84]. Principal component analysis
spondents for the independent variable (organiza demonstrated that all items loaded on a single fac
tion context) and the dependent variables (ambi tor having an eigenvalue of 2.56 and accounting for
and That is, for the 65 percent of the variance. Internal reliability was
dexterity performance).
we those =
independent variables aggregated only high [a .80).
who identified themselves as line Ambidexterity. Although we conceptualized
respondents
and nonmanagement. For the depen ambidexterity as a multidimensional construct
management
dent variables, we aggregated only those respon comprised of the nonsubstitutable combination of
dents who identified themselves as senior and alignment and adaptability (that is, as the multipli
middle management, because our studies in cative interaction of the two capacities), we antici
pilot
dicated these respondents were the best informed pated that post hoc analyses might involve exami
about unit-level outcomes. nation of the capacities independently as well as in
TABLE 1
Characteristics of the Sample Firms
Number of
Business Total Line Middle Senior
Firm Units Respondents Nonmanagement Management Management Management Industry Country
engineering
8 356 71 171 83 25 Food products Canada
9 2,122 1,443 483 149 47 Industrial conglomerate South Korea
10 71 35 21 15 0 Defense France
combination. Therefore, we
began by constructing their people" [.65], (5) "reward or punish based on
separate scales for alignmentand adaptability. We rigorous measurement of business performance
captured alignment by asking senior and middle against goals" [.84], (6) "hold people accountable
managers to indicate the degree to which they for their performance" [.83], and (7) "use their ap
agreed with the following: (1) "The management praisal feedback to improve their performance"
systems in this organization work coherently to [.66]. The seven items loaded on a single factor
support the overall objectives of this organization" having an eigenvalue of 3.77 and accounting for 54
=
[.56], (2) "The management systems in this organi percent of the variance [a .89). To capture social
zation cause us to waste resources on unproductive context, we asked line management and nonman
activities" (reversed) [.85], and (3) "People in this agement respondents to indicate the extent to
organization often end up working at cross-pur which systems encouraged people at their level to:
poses because our management systems give them (1) "devote considerable effort to developing their
conflicting objectives" (reversed) [.85]. All items subordinates" [.69], (2) "give everyone sufficient
loaded on a single factor having an eigenvalue of authority to do their jobs well" [.87], (3) "push
1.75 and accounting for 58 percent of the variance decisions down to the lowest appropriate level"
=
[a .73). We captured adaptability by asking se [.86], (4) "give ready access to information
that
nior and middle management respondents to indi others need" [.72], (5) "work hard to develop the
cate the degree to which they agreed with the capabilities needed to execute our overall strategy/
following: (1) "The management systems in this vision" [.63], (6) "base decisions on facts and anal
organization encourage people to challenge out ysis, not politics" [.76], (7) "treat failure (in a good
moded traditions/practices/sacred cows" [.79], (2) effort) as a learning opportunity, not something to
"The management systems in this organization are be ashamed of" [.68], (8) "are willing and able to
flexible enough to allow us to respond quickly to take prudent risks" [.66], and (9) "set realistic
changes in our markets" [.92], and (3) "The man goals" [.57]. The items loaded on a single factor
agement systems in this organization evolve rap having an eigenvalue of 4.36 and accounting for 55
in to our of variance = In a third
idly response shifts in business priorities" percent the (a .93). step in
[.90]. All items loaded on a single factor having an constructing the measure, we created an interaction
eigenvalue of 2.29 and accounting for 76 percent of term using the multiplicative interaction of the per
= a
the variance [a .80). In third step, we computed formance management context variable and the so
the multiplicative interaction between alignment cial context variable, reflecting our argument that
and adaptability, reflecting our argument that these these should be considered holistically and are
two capacities are nonsubstitutable and interde nonsubstitutable. This variable constitutes our
pendent. measure of organizational context.
-
Organization context We measured organiza Control variables. Finally, we entered nine [n
tion context by developing multi-item scales to rep 1) dummy variables representing the firms in our
resent the dimension of discipline, stretch, sup sample to control for firm-level effects in order to
port, and trust identified by Ghoshal and Bartlett parcel out differences due to business-unit-level
(1994). However, factor analysis revealed that it effects.
was not possible to identify four distinct constructs
using these items. Instead, two factors were identi
fied. One of these factors represents a combination Aggregation
of the items developed for discipline and stretch, Each of the variables in our model represents
and so we refer to this as "performance manage business-unit characteristics, but we utilized indi
ment context." The other factor represents a com viduals as raters of those characteristics. In the
bination of the items developed for support and parlance of multilevel theory (Klein & Koslowski,
trust, so we refer to this factor as "social context" to 2000: 41), our model consists entirely of "shared
represent the content of the items in this construct. unit-level constructs," meaning that we gathered
Specifically, to capture performance manage data from individuals to assess unit-level charac
ment context, we asked line management and non teristics that we presumed to be shared within a
management respondents to indicate the extent to unit and capable of differentiating among units.
which systems encouraged people at their level Conceptually, this makes sense, given that individ
(1) "to set challenging/aggressive goals" [.76], (2) ual employees are most familiar with the extent to
"issue creative challenges to their people, instead which their business unit exhibits certain attributes
of narrowly defining tasks" [.75], (3) "be more fo of an organizational context, as well as ambidexter
cused on getting their job done well than on getting ity and performance. Yet it is critical with such
promoted" [.59], (4) "make a point of stretching aggregated variables to statistically demonstrate
within-unit agreement and between-units differ confirmation that managers were reflecting on per
ences (Ancona & Caldwell, 1992: 655;
George, formance over a five-year as to
period, opposed
1990: 110; Goodman, Ravlin, & Schminke, 1990; on short-term
focusing gains.
Klein & Koslowski, 2000). discriminant (Venkatraman &
Finally, validity
We conducted several analyses to ensure that Grant, was established
1986) through exploratory
such agreement and such differences were present. and confirmatory factor analysis to verify the dis
First, we calculated an interrater agreement score tinctiveness of our constructs using all items from
(rWG;James, Demaree, & Wolf, 1993) for each vari all of the scales. The exploratory factor analysis
able. This measure ranges from 0 ("no agreement")
clearly replicated the intended three-factor struc
to 1 ("complete agreement"). Glick (1985) sug ture (including ambidexterity, context, and perfor
gested .60 as the cutoff for acceptable interrater
mance) to be used in tests of hypotheses. Items
agreement values. Median interrater agreement was
loaded on the intended factors, all of which had
.71 for alignment, .76 for adaptability, .90 for per
formance management, .93 for social context, and eigenvalues greater than one, supporting the three
factor model. Further, exploratory factor analysis
.82 for performance, suggesting adequate agree
ment for aggregation. We also generated intraclass (with and without rotation) did not reveal a single
or general factor that would suggest the presence
correlation coefficients?ICC(l) and ICC(2)?, us
of common method (Brewer, Campbell, & Crano,
ing one-way analysis of variance (ANOVA) on the
1970) or social variance (Thomas &
individual-level data, with unit as the independent desirability
variable and the scale scores as the dependent vari Kilmann, 1975).
ables. Others have also suggested that an indication We employed confirmatory factor analysis using
of convergence within units is an ICC(l) value LISREL 8 (J?reskog & S?rbom, 1996) to compare the
than zero with a proposed three-factor model to an alternative five
greater corresponding significant
ANOVA test statistic [F] (Kenny & LaVoie, 1985). In factor model (including performance management
all cases, the ICC(l) was greater than zero and the F context, social
context, alignment, adaptability,
was significant. The ICC(2) values, which are valu performance), and a seven-factor model (including
able indicators of the reliability of the unit mean, stretch, discipline, support, trust, alignment, adapt
were .92 for alignment, .98 for adaptability, .92 for
ability and performance). Absolute fit indexes for
.90 for social context, the proposed three-factor model
performance management, ranged from ade
and .92 for performance, that the means = =
indicating quate to excellent (#2 909.73, df 274, p < .001,
for the sets of perceptions for each variable were GFI = CFI = IFI = RMSEA = and
.98, .98, .98, .04),
accurate representations of the true score for the these fit indexes were to both the five- and
superior
unit (James, 1982; Lord & Novick, 1968). the seven-factor In addition, the value
models.2
of another comparative informaindex, Akaike's
tion criterion 2000; Hu & Bentler,
Validity Checks (AIC; Boomsma,
1999) was better (that is, smaller) for our three
We conducted an external validity check for our
factor model than for the five- or seven-factor
performance measure. Because objective indicators = =
model (AIC3.factor 1,063.79; 1,688.93;
were not AIC5.factor
of business-unit performance available for =
AIC7.factor 3,127.54). All of these
results indicate
all business units, we conducted the validity check
that our three-factor model provided a better fit to
at the firm level. First, we aggregated subjective
the data than did its plausible rival specifications.
business-unit performance to the firm level by av
senior manager's of all business Further, these results indicate that the three scales
eraging ratings
units within a firm. Then we obtained financial represent concepts that are not only theoretically,
indicators for each firm relative to its but also empirically, distinguishable.
performance
industry group, calculating measures of return on
assets (ROA), return on equity (ROE), and share
holder return over a five-year period for each com
pany and then dividing these performance mea 2
As is typical in the confirmatory factor analysis
sures by the equivalent figures for a group of peer
(Kelloway, 1998), the chi-square associated with our
companies. These relative measures of financial
three-factor model was For a discussion of
significant.
performance were highly correlated with aggre reasons for significant chi-squares apart from real speci
gated measures of subjective performance as rated fication see Boomsma that
errors, (2000). Note, however,
=
by senior managers (r .75, p < .05), lending the RMSEA (.04) did not exceed the .08 cutoff value
strong external validity to the subjective perfor recommended by experts (e.g., Browne & Cudeck, 1993;
mance measure. Further, this analysis provides Hu & Bentler, 1999).
TABLE 2
Descriptive Statistics and Correlations3
a
u = 41 (business units).
b
Organizational context is the multiplicative interaction of performance management and social context. Ambidexterity is the
TABLE 3
Results of Regression Analysis
*
p < .05
< .01
**p
***
p < .001
groups. Under the K-means algorithm (Hartigan, the null hypothesis that all four groups had the
1975; Hartigan & Wong, 1979), the four-group same performance level. Group 4 (highly ambidex
model provided the best fit. Table 4 shows the trous) was the best performing, followed by group 3
alignment and adaptability scores for the four clus (moderately ambidextrous), group 2 (adaptive),
ter centers. Group 1 consisted of 7 "aligned" busi and group 1 (aligned). Using the post hoc S-N-K
ness units, with higher ratings on alignment than (Student-Newman-Keuls) procedure, we estab
adaptability. Group 2 consisted of 11 "adaptive" lished that the differences between each and every
business units, with higher ratings on adaptability group were significant. We performed two addi
than alignment. Group 3 consisted of 18 "moder tional analyses to ensure that this finding was ro
DISCUSSION
In this study, we adopted a multimethod ap
proach to understanding contextual ambidexterity
in organizations, in order to investigate several spe
< cific hypotheses while also generating new insights
2.0,
2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 into the mechanisms and processes at work in the
sample's business units. We found strong evidence
Adaptability
that contextual ambidexterity?the simultaneous
achievement of capacities for aligned and adapt
individuals were likely to have a broader perspec able behavior?mediates the relationship between
tive on issues of alignment, adaptability, and per features of the organizational context that encour
formance than the lower-level employees. How age these behaviors and subsequent performance.
ever, this also created a risk of same-source bias in This study raises important issues for both theory
the analysis, so we reanalyzed the data using the and practice.
senior respondents' of performance
ratings and the First, there does not seem to be a trade-off be
junior respondents' ratings of alignment and adapt tween alignment and adaptability, whereby one is
ability. This analysis produced similar results, with sacrificed for the other. Successful business units
mean performance levels of 4.13 for group 4 (highly were able to simultaneously develop these capaci
ambidextrous), 4.11 for group 3 (moderately ambi ties by aligning themselves around adaptability.
dextrous), 3.88 for group 2 (adaptive), and 3.41 for Importantly, the systems that they used to do this
group 1 (aligned). The ANOVA F-test was signifi were often quite simple?indeed, they often in
cant (F = 5.57, p < .01). Comparing these results to volved less formality, rather than more (O'Toole,
the previous set shows that the rankings of the four 2001). This observation implies a reconceptualiza
groups on performance are identical. tion of ambidexterity. Traditionally, research has
It is important not to overinterpret the meaning focused on what we have called structural ambi
of membership of these groups, because there will dexterity, which involves using structural or archi
always be business units on the margin of one tectural solutions to the problem of achieving
group that could easily become members of other alignment and adaptability. For instance, certain
groups. This becomes an important issue in the business units are designated as responsible for
case of group 1 (aligned), which is dominated by adaptability, and others are designated as respon
business units from a single corporation, and sible for alignment; or perhaps temporal separation
which also ends up being the most poorly perform or task partitioning within a unit serves as a struc
ing group. Accordingly, we conducted some sensi tural means to separately achieve both alignment
tivity analysis, which established that changes to and adaptability. In this study, we have developed
the membership of this group did not affect the the complementary concept of contextual ambidex
results in a significant way. We identified the next terity, whereby an organization context encourages
four business units that were closest to group 1 on individuals to make their own choices as to how
the dimensions of alignment and adaptability and they divide their time between alignment- and
added themto group 1 (one of these came from adaptability-oriented activities. In general, this
group 2, and three came from group 3). We then view supports the recent focus on a paradoxical
reanalyzed the data using this new grouping. approach to management, as opposed to an "either/
Again, similar results emerged, with mean perfor or" focus (Lewis, 2000). Our results indicate that
mance levels of 4.60 for group 4 (highly ambidex achieving ambidexterity through contextual sup
trous), 4.14 for group 3 (moderately ambidextrous), port is possible and does relate positively to per
TABLE 4
Results of Post Hoc Cluster Analysis
Group Ratings
Group 1
3.69 2.75 3.21
Aligned
Firm 6 Furniture products 3.88
Firm 9 Semiconductors 3.17
random access memory (DRAM) 2.87
Dynamic
Static random access memory (SRAM) 3.36
Memory 3.63
production
Systems 2.82
Support 2.80
Group 2
3.49 4.07 11 3.74
Adaptive
Firml Customer operations 3.58
Firm 3 Direct banking 4.07
Firm 4 Energy 3.44
Firm 5 vertical services 4.15
Consulting
Marketing 3.89
Group 3
ambidextrous 4.19 4.22 18 4.10
Moderately
Firm 1 Office documents 3.56
Firm 2 Building construction products 4.14
Wheel loaders and excavation 4.71
Firm 3 Mortgages 4.14
Firm 5 Vertical sales 4.17
Firm 6 Appliances
3.32
Design 3.88
formance. Further, it is not enough to simply create likely a critical capability for many, if not all, firms.
a supportive context. It is when this supportive However, future research might explore whether an
context creates the capacity for ambidexterity that important boundary condition to this finding is the
are realized. level of dynamism in a business environment. In a
performance gains
In our sample, ambidexterity was an asset across highly stable or placid business environment, it
a wide of industries, that it is may be that although a baseline level of adaptabil
variety suggesting
ity is necessary, the business units with the highest because while the Ghoshal and Bartlett (1994)
levels of alignment are the best performers. Future framing suggested a common language around dis
research investigating an even greater array of in stretch, support, and trust, the reality in the
cipline,
dustries, varying even more than those in our sam business units in our sample was that each used its
ple in terms of business environment, would be an own rather idiosyncratic implementation strategy
important extension of our study. to create a performance and social
management
Second, the qualitative data collected during the context conducive to the simultaneous
achieve
research suggests that there are different paths to ment of alignment and adaptability. on
Building
ambidexterity. The ambidextrous business units in the work of Denison and colleagues (1995) and
the automotive in our sample had gradu others, our results of tran
company suggest the importance
ally built adaptability skills on top of their tradi the either/or of performance
scending management
tional model of alignment, whereas the ambidex and social context to develop simultaneous and
trous business unit in the oil and gas company behavioral that foster both
complex responses
focused on adaptability and created alignment and adaptability. A promising extension
alignment
around it, and the business unit in the software of our study would be to more ex
systematically
firm created a blend of the two. So
deliberately amine the behaviors of senior executives in an ef
there is evidence of "equifinality" (multiple paths) fort to understand how create ambidex
they help
in the process of establishing the systems that re
terity.
sult in ambidexterity (Gresov & Drazin, 1997). De In terms of the methods used in this research, it is
on the administrative heritage of a given
pending worth the pros and cons of using a ran
discussing
business, and the values of its leaders, equally dom sample of individual respondents in multiple
valid, but slightly different, organization context
business units. While our approach has advantages
solutions can be created.
over either focusing on a single company or using
The third conceptual development is the impor
single respondents in multiple business units, the
tant role played by senior executives in making an
consequence of aggregating a large number of re
organization context effective and developing am
spondents in each business unit was a limited vari
bidexterity. Inherent to this research project from
ance in our key constructs. Future research should
its beginning was the argument that as key leaders
explore additional means of capturing contextual
in organizations, senior executives play a critical and context
ambidexterity characteristics, perhaps
role?because they put in place systems that allow
through archival means, or broad-scale interviews,
supportive contexts to emerge?that in turn shape or surveys of customers and/or stakeholders. In
individual behaviors (Burgelman, 1983b; Ghoshal
& Bartlett, 1994). Certainly there is some evidence addition, future research should explore more ob
jective indicators of business-unit performance,
for this model (for instance, in the oil and gas
perhaps replicating our validity check at the busi
business units), but there was also strong evidence
ness-unit level to create a peer group comparison of
in several units in both the software and oil and gas
firms that senior executives a more inter
outcomes for business units. Here, such a compar
played as these data were not avail
on recognizing and pro ison was impossible,
ventionist role, focused
new ideas and building for those able for many of the business units in non-U.S.
moting energy
ideas the business. Without this form of firms in our sample, but with a more narrow set of
throughout
intervention, there is a risk that new ideas will fail firms, financial indicators might be available for all
to get the resources business units.
they need to be developed and
that coherence will be compromised. In conclusion, we view the concept of contextual
strategic
the implication for management is ambidexterity as highly promising for understand
Thus, practice
that contextual is likely an important ing the tensions, balances, and equilibrium that
ambidexterity
and desirable that business units can leaders must manage in complex organizational en
capability
and that it can be shaped at least in part vironments. Encouraging a supportive organiza
develop,
leaders' behavior. to estab tional context
that generates simultaneous capaci
through Beyond helping
lish a supportive context, senior executives
likely
ties for alignment and adaptability may be a key
a role
in source of competitive advantage for leaders in the
play fostering ambidexterity, primarily by
and nurturing In light of 21st century. Future research aimed at further de
encouraging adaptability.
our interviews, we would speculate that often this lineation of the underlying features of organiza
is accomplished by simply serving as a good exam tional context that are most critical in developing
ple, modeling the adaptable behavior, and then re contextual ambidexterity will likely go a long way
inforcing it with rewards and recognition. to sustaining business-unit performance in the era
However, it is difficult to be too prescriptive, of dynamic economic environments.
Amit, R., & Schoemaker, P. 1993. Strategic assets and Carlsson, B. 1989. Flexibility and the theory of the firm.
rent. International Journal of Industrial Organization,
organizational Strategic Management Journal,
14: 33-46. 7: 179-203.
Ancona, D. G., & Caldwell, D. F. 1992. Bridging the De Meyer, A., Nakane, J., Miller, J., & Ferdows, K. 1989.
boundary: External activity and performance in or Flexibility: The next competitive battle. The manu
ganizational teams. Administrative Science Quar facturing futures survey. Strategic Management
37: 634-665. Journal, 10: 135-144.
terly,
for action: A guide to over R. D. 1990. culture and
Argyris, C. 1993. Knowledge Denison, Corporate organiza
tional New York: Wiley.
coming barriers to organizational change. San Fran effectiveness.
cisco: Jossey-Bass. Denison, D., Hooijberg, R., & Quinn, R. E. 1995. Paradox
Baron, R. M., & Kenny, D. A. 1986. The moderator-medi and performance: Toward a theory of behavioral
ator variable distinction in social psychological re complexity in managerial leadership. Organization
search: Conceptual, and statistical consid Science, 6: 524-540.
strategic
erations. Journal of Personality and Social R. 1999. Developing a market orientation.
Deshpande,
Psychology, 51: 1173-1182. Thousand Oaks, CA: Sage.
J. 1991. Firm resources and sustained
Barney, competi Drucker, P. 1985. Innovation and entrepreneurship:
tive advantage. Journal of Management, 17(1): 99 Practice and principles. New York: Harper & Row.
120.
Duncan, R. B. 1976. The ambidextrous organization: De
Bartlett, C. A., & Ghoshal, S. 1989. Managing across
signing dual structures for innovation. In R. H. Kil
borders: The transnational solution. Boston: Har
mann, L. R. Pondy, & D. Slevin (Eds.), The manage
vard Business School Press. ment of organization, vol. 1: 167-188. New York:
Boomsma, A. 2000. of covariance North-Holland.
Reporting analyses
structures. Structural equation modeling, 7: 461
Earley, P. C, & Gibson, C. B. 2002. Multinational teams:
483.
A new perspective. Mahwah, NJ: Erlbaum.
Bouchikhi, H. 1998. Living with and building on com
Edmondson, A. 2001. Disrupted routines: Team learning
plexity: A constructivist perspective on organiza
and new technology implementation in hospitals.
tions. 2: 217-232.
Organization, Administrative Science Quarterly, 46: 685-716.
Bower, J. L. 1986. the resource allocation
Managing Ford, & Ford, L. W. 1994. of con
J. D., Logics identity,
process: A study of corporate planning and invest
tradiction, and attraction in change. Academy of
ment. Boston: Harvard Business School Press.
Management Review, 19: 756-795.
Burgelman, R. A. 1983b. A model of the interaction of Goodman, P. S., Ravlin, E. C, & Schminke, M. 1990.
Cummings & B. M. Staw (Eds.), Leadership, partic analysis in social psychology. In D. Gilbert, S. Fiske,
ipation, and group behavior: 323-385. Greenwich, & G. Lindzey (Eds.), The handbook of social psy
CT: JAI Press. chology, vol. 1 (4th ed.): 233-265. Boston: McGraw
& Drazin, R. 1997. functional
Hill.
Gresov, C, Equifinality,
equivalence in organization design. Academy of Klein, B. H. 1984. Prices, wages and business cycles: A
ing algorithm, algorithm. Applied Statistics, 28: Lawrence, P., & Lorsch, J. 1967. and envi
Organization
100-108. ronment: and
Managing differentiation integra
M. tion. Boston: Harvard
Hatch, J., & Ehrlich, S. B. 1993. Spontaneous humour University.
as an indicator of paradox and ambiguity in organ Leonard-Barton, D. 1992. Core and core ri
capabilities
izations. Organization Studies, 14: 505-526. gidities: A paradox in managing new product devel
Hedlund, G, & Ridderstrale, J. 1997. Toward a theory of opment. Strategic Management Journal, 13: 111
the self-renewing MNC. In B. Toyne & D. Nigh (Eds.), 125.
International business: An emerging vision: 329 Lewin, K., Lippitt, R., & White, R. K. 1939. Patterns of
353. Columbia: University of South Carolina Press.
aggressive behavior in experimentally created "so
Hofstadter, D. R. 1979. Godel, Escher, Rach: An eternal cial climates." Journal of Social Psychology, 10:
271-299.
golden braid. New York: Vintage.
Hu, L., & Bender, P. M. 1999. Cutoff criteria for fit in Lewis, M. W. 2000. Exploring paradox: Toward a more
dexes in covariance structure analysis: Conventional comprehensive guide. Academy of Management
criteria versus new alternatives. Structural Review, 25: 760-777.
Equa
tion Modeling, 6: 1-55.
Lord, F., & Novick, M. R. 1968. Statistical theories of
James, L. R. 1982. Aggregation bias in estimates of per mental test sores. Reading, MA: Addison-Wesley.
ceptual agreement. Journal of Applied Psychology, MacDuffie, J. P. 1995. Human resource bundles and man
67: 219-229.
ufacturing performance: Organizational logic and
James, L. R., & Brett, J. M. 1984. Mediators, moderators, flexible production systems in the world auto indus
and tests of mediation. Journal of Applied Psychol try. Industrial and Labor Relations Review, 48:
ogy, 69: 307-321. 197-221.
L. R., Demaree, R. G, & Wolf, G. 1993. An MacKinnon, D. P., & Dwyer, J. H. 1993. me
James, rwg: Estimating
assessment of within group inter-rater agreement.
diated effects in prevention studies. Evaluation Re
Journal of Applied Psychology, 78: 306-339. view, 17(2): 144-158.
James, L. R., & Jones, A. P. 1974. Organizational climate: March, J. G. 1991. Exploration and in organ
exploitation
A review or and research. Rul izational learning. Organization Science, 2: 71-86.
theory Psychological
letin, 81: 1096-1112.
March, J. G, & Simon, H. 1958. New
Organizations.
Jaworski, B. J., & Kohli, A. K. 1993. Market orientation: York: Wiley.
Antecedents and consequences. Journal of Market M. & Zaccaro,
Marks, A., Mathieu, J. E., S. J. 2001. A
ing, 57(7): 53-70.
temporally based framework and taxonomy of team
J?reskog, K. G, & S?rbom, D. 1996. LISREL 8: User's processes. Academy of Management Review, 26:
reference guide (2nd ed.). Chicago: Scientific Soft 356-376.
ware International.
McDonough, E., & Leifer, R. 1983. Using simultaneous
Kelloway, E. K.
1998. Using LISREL for structural structures to cope with uncertainty. Academy of
equation modeling: A researcher's guide. Thou Management Journal, 26: 727-735.
sand Oaks, CA: Sage.
Morgeson, F. P., & Hofmann, D. A. 1999. The structure
Kenny, D. A., & LaVoie, L. 1985. Separating individual and function of collective constructs: Implications
and group effects. Journal of Personality and Social for multilevel research and theory development.
Psychology, 48: 339-348. Academy of Management Review, 24: 249-265.
Narver, J.C, & Slater, S. F. 1990. The effect of a market ability variable in organizational research: An alter
orientation on business profitability. Journal of native explanation for reported findings. Academy
Marketing, 54(4): 20-35. of Management Journal, 18: 741-752.
O'Conner, E. S. 1995. Paradoxes of participation. Textual Tushman, M. L., & O'Reilly, C. A. 1996. Ambidextrous
analysis and organizational change. Organization organizations: Managing evolutionary and revolu
Prahalad, C. K., & Hamel, G. 1990. The core in teams and the impact of culture and gender
tiveness
competencies
of the corporation. Harvard Rusiness on workbehavior. Cristina's research has appeared in
Review, 68(3):
79-90. many journals, and she has collaborated on two recent
E. H. 1985. culture and leader
major books.
Schein, Organizational
ship. San Francisco: Jossey-Bass. Julian Birkinshaw is an associate professor in and the
an chair of the Strategic and International Management
Schneider, K. J. 1990. The paradoxical self: Toward
our nature. New Group at the London Business School, and a Fellow of
understanding of contradictory
the Advanced Institute of Management Research. He re
York: Insight Books.
ceived his Ph.D. from the University ofWestern Ontario.
Sobel, M. E. 1982. Asymptotic confidence intervals for His work has been widely published, and he is the author
indirect effects in structural equation models. In S. or coauthor of seven books.
Leinhart (Ed.), Sociological methodology: 219-312.
San Francisco: Jossey-Bass.