Sie sind auf Seite 1von 24

Building Theories from Case Study Research

Kathleen M. Eisenhardt

The Academy of Management Review, Vol. 14, No. 4. (Oct., 1989), pp. 532-550.

Stable URL:

The Academy of Management Review is currently published by Academy of Management.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact
Tue Jul 31 21:07:56 2007
F Academy ofManagement Review, 1989, Vol 14, No 4, 532-550

Building Theories from Case

Study Research

Stanford University

This paper describes the process of inducting theory using case stud-
ies-from specifying the research questions to reaching closure.
Some features of the process, such a s problem definition a n d con-
struct validation, a r e similar to hypothesis-testing research. Others,
such as within-case analysis a n d replication logic, a r e unique to the
inductive, case-oriented process. Overall, the process described here
is highly iterative a n d tightly linked to data. This research approach is
especially appropriate in new topic areas. The resultant theory is
often novel, testable, a n d empirically valid. Finally, framebreaking
insights, the tests of good theory fe.g.,parsimony, logical coherence),
a n d convincing grounding in the evidence a r e the key criteria for
evaluating this type of research.

Development of theory is a central activity in lack of clarity about the process of actually
organizational research. Traditionally, authors building theory from cases, especially regard-
have developed theory by combining observa- ing the central inductive process a n d the role of
tions from previous literature, common sense, literature. Glaser a n d Strauss (1967) a n d more
a n d experience. However, the tie to actual data recently Strauss (1987) have outlined pieces of
has often been tenuous (Perrow, 1986; Pfeffer, the process, but theirs is a prescribed formula,
1982).Yet, a s Glaser a n d Strauss (1967)argue, it a n d new ideas have emerged from methodolo-
is the intimate connection with empirical reality gists (e.g., Yin, 1984; Miles & Huberman, 1984)
that permits the development of a testable, rel- a n d researchers conducting this type of re-
evant, a n d valid theory. search (e.g., Gersick, 1988; Harris & Sutton,
This paper describes building theories from 1986; Eisenhardt & Bourgeois, 1988). Also, it ap-
case studies. Several aspects of this process are pears that no one has explicitly examined when
discussed in the literature. For example, Glaser this theory-building approach is likely to b e
a n d Strauss ( 1 967) detailed a comparative fruitful a n d what its strengths a n d weaknesses
method for developing grounded theory, Yin may be.
(1981, 1984) described the design of case study This paper attempts to make two contributions
research, a n d Miles a n d Huberman (1984)codi- to the literature. The first is a roadmap for build-
fied a series of procedures for analyzing quali- ing theories from case study research. This
tative data. However, confusion surrounds the roadmap synthesizes previous work on qualita-
distinctions among qualitative data, inductive tive methods (e.g., Miles & Huberman, 1984),the
logic, a n d case study research. Also, there is a design of case study research (e.g., Yin, 1981,
1984),and grounded theory building (e.g., Gla- search than has existed in the past. This frame-
ser & Strauss, 1967) and extends that work in work is summarized in Table 1.
areas such a s a priori specification of constructs, The second contribution is positioning theory
triangulation of multiple investigators, within- building from case studies into the larger context
case a n d cross-case analyses, and the role of of social science research. For example, the pa-
existing literature. The result is a more nearly per explores strengths and weaknesses of theory
complete roadmap for executing this type of re- building from case studies, situations in which it

Table 1
Process of Building Theory from Case Study Research

Step Activity Reason

Gettlng Started Definition of research question Focuses efforts
Posslbly a prlorl constructs Provldes better grounding of construct
Nelther theory nor hypotheses Retains theoretical flexibility
Selecting Cases Specified population Constrains extraneous varlation and
sharpens external validlty
Theoretical, not random, sampling Focuses efforts on theoretically useful
cases-i.e., those that replicate or extend
theory by filllng conceptual categories
Craftlng Instruments Multiple data collection methods Strengthens grounding of theory by
and Protocols triangulation of evldence
Qualitative and quantitative data combined Synerglstlc vlew of evldence
Multiple lnvestlgators Fosters divergent perspectives and
strengthens grounding
Entering the Fleld Overlap data collection and analysis, Speeds analyses and reveals helpful
Including field notes adjustments to data collection
Flexible and opportunistic data collection Allows investigators to take advantage of
methods emergent themes and unique case
Anaiyzlng Data Withln-case analysls Gains famlllarlty with data and preliminary
theory generation
Cross-case pattern search using divergent Forces investigators to look beyond lnitlal
techniques impressions and see evidence thru
multiple lenses
Shaping Hypotheses Iterative tabulation of evldence for each Sharpens construct definition, validity, and
construct measurability
Repllcatlon, not sampling, loglc across Confirms, extends, and sharpens theory
Search evldence for "why" behind Bullds Internal valldity
Enfolding Literature Comparison with conflicting literature Builds internal valldity, raises theoretical
level, and sharpens construct definitions
Comparison with similar literature Sharpens generallzability, improves
construct definition, and ralses theoretical
Reachlng Closure Theoretical saturation when possible Ends process when marglnal Improvement
becomes small
is a n attractive research approach, a n d some Eisenhardt, 1988)developed cross-case analysis
guidelines for evaluating this type of research. techniques.
Finally, the work of others such a s Van Maa-
Background nen (1988) on ethnography, Jick (1979) on trian-
gulation of data types, a n d Mintzberg (1979) on
Several pieces of the process of building the- direct research has provided additional pieces
ory from case study research have appeared in for a framework of building theory from case
the literature. One is the work on grounded the- study research.
ory building by Glaser and Strauss (1967) and, As a result, many pieces of the theory-
more recently, Strauss (1987). These authors building process are evident in the literature.
have detailed their comparative method for de- Nevertheless, at the same time, there is substan-
veloping grounded theory. The method relies on tial confusion about how to combine them,
continuous comparison of data a n d theory be- when to conduct this type of study, a n d how to
ginning with data collection. It emphasizes both evaluate it.
the emergence of theoretical categories solely
from evidence and a n incremental approach to The Case Study Approach
case selection and data gathering.
More recently, Yin (1981, 1984) has described The case study is a research strategy which
the design of case study research. He has de- focuses on understanding the dynamics present
fined the case study a s a research strategy, de- within single settings. Examples of case study
veloped a typology of case study designs, a n d research include Selznick's (1949) description of
described the replication logic which is essential TVA, Allison's (1971) study of the Cuban missile
to multiple case analysis. His approach also crisis, a n d Pettigrew's (1973) research on deci-
stresses bringing the concerns of validity a n d sion making at a British retailer. Case studies
reliability in experimental research design to the can involve either single or multiple cases, and
design of case study research. numerous levels of analysis (Yin, 1984). For ex-
Miles a n d Huberman (1984) have outlined ample, Harris and Sutton (1986)studied 8 dying
specific techniques for analyzing qualitative organizations, Bettenhausen a n d Murnighan
data. Their ideas include a variety of devices (1986) focused on the emergence of norms in 19
such a s tabular displays and graphs to manage laboratory groups, a n d Leonard-Barton (1988)
a n d present qualitative data, without destroying tracked the progress of 10 innovation projects.
the meaning of the data through intensive cod- Moreover, case studies can employ a n embed-
ing. ded design, that is, multiple levels of analysis
A number of active researchers also have un- within a single study (Yin, 1984). For example,
dertaken their own variations a n d additions to the Warwick study of competitiveness and stra-
the earlier methodological work (e.g., Gersick, tegic change within major U.K.corporations is
1988; Leonard-Barton, 1988; Harris & Sutton, conducted at two levels of analysis: industry a n d
1986). Many of these authors acknowledge a firm (Pettigrew, 1988), a n d the Mintzberg a n d
debt to previous work, but they have also devel- Waters (1982) study of Steinberg's grocery em-
oped their own "homegrown" techniques for pire examines multiple strategic changes within
building theory from cases. For example, Sutton a single firm.
and Callahan (1987)pioneered a clever use of a Case studies typically combine data collection
resident devil's advocate, the Warwick group methods such a s archives, interviews, question-
(Pettigrew, 1988) added triangulation of investi- naires, and observations. The evidence may be
gators, a n d my colleague and I (Bourgeois & qualitative ( e . g . , words), quantitative ( e . g . ,
numbers), or both. For example, Sutton a n d qualitative evidence from interviews and obser-
Callahan (1987) rely exclusively on qualitative vations.
data in their study of bankruptcy in Silicon Val- Finally, case studies can be used to accom-
ley, Mintzberg and McHugh (1985)use qualita- plish various aims: to provide description (Kid-
tive data supplemented by frequency counts in der, 1982),test theory (Pinfield, 1986; Anderson,
their work on the National Film Board of Can- 1983), or generate theory (e.g., Gersick, 1988;
a d a , a n d Eisenhardt and Bourgeois (1988)com- Harris & Sutton, 1986).The interest here is in this
bine quantitative data from questionnaires with last aim, theory generation from case study ev-

Table 2
Recent Examples of lnductive Case Study Research*

Description Research Data

Study of Cases Problem Sources Investigators Output
Burgelman (1983) 6 mternal cor- Management of Archives Single mvestigator Process model
porate ventures new ventures Intervlews linhng multiple
in 1 malor Some observation organizational
corporation levels
Mintzberg & 1 Natlonal Film Formulation of Archives Research team Strategy-mahng
McHugh (1985) Board of Can- strategy in a n Some ~ntervlews themes, "grass
a d a , 1939- 1975, adhocracy roots" model of
wlth 6 perlods strategy forma-
Harris & Sutton 8 diverse organl- Parting cere- Interviews Research team Conceptual
( 1986) zations monies during Archives framework
organizational about the
death functions of
parting cere-
monies for
Eisenhardt & 8 microcomputer Strategic decision Interviews Research team Mid-range theory
Bourgeois (1988) firms making in high Questionnaires Tandem inter- linhng power,
velocity environ- Archives views politics, a n d
ments Some observation firm perform-
Gersick (1988) 8 project groups Group develop- Observation Single investigator Punctuated
wlth deadlines ment in project- Some interviews equilibrium
teams model of group
Leonard-Barton 10 technical ~ n n o - Internal technol- Interviews Single investigator Process model
( 1988) vations ogy transfer Experiment
Pettigrew (1988) 1 high performing Strategic change Interviews Research teams In progress
& 1 low per- & competi- Archives
forming flrm in tiveness Some observation
each of 4
Examples were chosen from recent organizational writlng to provide illustrations of the possible range of theory building
from case studies.
idence. Table 2 summarizes some recent re- to the decision process, there were strong, trian-
search using theory building from case studies. gulated measures on which to ground the emer-
gent theory.
Although early identification of the research
Building Theory from Case
question a n d possible constructs is helpful, it is
Study Research
equally important to recognize that both a r e ten-
tative in this type of research. No construct is
Getting Started
guaranteed a place in the resultant theory, no
An initial definition of the research question, matter how well it is measured. Also, the re-
in at least broad terms, is important in building search question may shift during the research.
theory from case studies. Mintzberg (1979, p. At the extreme, some researchers (e.g., Gersick,
585) noted: "No matter how small our sample or 1988; Bettenhausen & Murnighan, 1986) have
what our interest, we have always tried to go converted theory-testing research into theory-
into organizations with a well-defined focus-to building research by taking advantage of seren-
collect specific kinds of data systematically." The dipitous findings. In these studies, the research
rationale for defining the research question is focus emerged after the data collection had be-
the same a s it is in hypothesis-testing research. gun. As Bettenhausen a n d Murnighan (1986, p.
Without a research focus, it is easy to become 352) wrote: ". . . we observed the outcomes of a n
overwhelmed by the volume of data. For exam- experiment on group decision making and co-
ple, Pettigrew a n d colleagues (1988) defined alition formation. Our observations of the
their research question in terms of strategic groups indicated that the unique character of
change and competitiveness within large British each of the groups seemed to overwhelm our
corporations, a n d Leonard-Barton (1988) fo- other manipulations." These authors proceeded
cused on technical innovation of feasible tech- to switch their research focus to a theory-
nologies. Such definition of a research question building study of group norms.
within a broad topic permitted these investiga- Finally and most importantly, theory-building
tors to specify the kind of organization to be ap- research is begun a s close a s possible to the
proached, and, once there, the kind of data to be ideal of no theory under consideration a n d no
gathered. hypotheses to test. Admittedly, it is impossible to
A priori specification of constructs can also achieve this ideal of a clean theoretical slate.
help to s h a p e the initial design of theory- Nonetheless, attempting to approach this ideal
building research. Although this type of specifi- is important because preordained theoretical
cation is not common in theory-building studies perspectives or propositions may bias and limit
to date, it is valuable because it permits re- the findings. Thus, investigators should formu-
searchers to measure constructs more accu- late a research problem a n d possibly specify
rately. If these constructs prove important a s the some potentially important variables, with some
study progresses, then researchers have a reference to extant literature. However, they
firmer empirical grounding for the emergent should avoid thinking about specific relation-
theory. For example, in a study of strategic de- ships between variables a n d theories a s much
cision making in top management teams, Bour- a s possible, especially at the outset of the pro-
geois and Eisenhardt (1988) identified several cess.
potentially important constructs (e.g., conflict,
power) from the literature on decision making. Selecting Cases
These constructs were explicitly measured in the
interview protocol a n d questionnaires. When Selection of cases is a n important aspect of
several of these constructs did emerge a s related building theory from case studies. As in hypoth-
esis-testing research, the concept of a popula- Several studies illustrate theoretical Sam-
tion is crucial, because the population defines pling. Harris a n d Sutton (19861, for example,
the set of entities from which the research Sam- were interested in the parting ceremonies of
ple is to be drawn. Also, selection of a n appro- dying organizations. In order to build a model
priate population controls extraneous variation applicable across organization types, these re-
a n d helps to define the limits for generalizing the searchers purposefully selected diverse organi-
findings. zations from a population of dying organiza-
The Warwick study of strategic change and tions. They chose eight organizations, filling
competitiveness illustrates these ideas (Petti- each of four categories: private, dependent; pri-
grew, 1988). In this study, the researchers se- vate, independent; public, dependent; a n d pub-
lected cases from a population of large British lic, independent. The sample was not random,
corporations in four market sectors. The selec- but reflected the selection of specific cases to ex-
tion of four specific markets allowed the re- tend the theory to a broad range of organiza-
searchers to control environmental variation, tions. Multiple cases within each category al-
while the focus on large corporations con- lowed findings to be replicated within catego-
strained variation due to size differences among ries. Gersick (1988)followed a similar strategy of
the firms. Thus, specification of this population diverse sampling in order to enhance the gen-
reduced extraneous variation a n d clarified the eralizability of her model of group develop-
domain of the findings a s large corporations op- ment. In the Warwick study (Pettigrew, 1988),
erating in specific types of environments. the investigators also followed a deliberate,
However, the sampling of cases from the cho- theoretical sampling plan. Within each of four
sen population is unusual when building theory markets, they chose polar types: one case of
from case studies. Such research relies on theo- clearly successful firm performance a n d one un-
retical sampling (i.e., cases are chosen for theo- successful case. This sampling plan was de-
retical, not statistical, reasons, Glaser & Strauss, signed to build theories of success and failure.
1967).The cases may be chosen to replicate pre- Finally, the Eisenhardt a n d Bourgeois ( 1988)
vious cases or extend emergent theory, or they study of the politics of strategic decision making
may be chosen to fill theoretical categories and illustrates theoretical sampling during the
provide examples of polar types. While the course of research. A theory linking the central-
cases may be chosen randomly, random selec- ization of power to the use of politics in top man-
tion is neither necessary, nor even preferable. agement teams was built and then extended to
As Pettigrew (1988) noted, given the limited consider the effects of changing team composi-
number of cases which can usually be studied, tion by adding two cases, in which the executive
it makes sense to choose cases such a s extreme teams changed, to the first six, in which there
situations and polar types in which the process was no change. This tactic allowed the initial
of interest is "transparently observable." Thus, framework to be extended to include dynamic
the goal of theoretical sampling is to choose effects of changing team composition.
cases which are likely to replicate or extend the
Crafting Instruments and Protocols
emergent theory. In contrast, traditional, within-
experiment hypothesis-testing studies rely on Theory-building researchers typically com-
statistical sampling, in which researchers ran- bine multiple data collection methods. While in-
domly select the sample from the population. In terviews, observations, a n d archival sources
this type of study, the goal of the sampling pro- are particularly common, inductive researchers
cess is to obtain accurate statistical evidence on are not confined to these choices. Some investi-
the distributions of variables within the popula- gators employ only some of these data collection
tion. methods (e.g., Gersick, 1988, used only obser-
vations for the first half of her study), or they may the findings. Convergent perceptions a d d to the
add others (e.g., Bettenhausen & Murnighan, empirical grounding of the hypotheses, while
1986, used quantitative laboratory data). The ra- conflicting perceptions keep the group from pre-
tionale is the same a s in hypothesis-testing re- mature closure. Thus, the use of more investiga-
search. That is, the triangulation made possible tors builds confidence in the findings a n d in-
by multiple data collection methods provides creases the likelihood of surprising findings.
stronger substantiation of constructs a n d hy- One strategy for employing multiple investi-
potheses. gators is to make the visits to case study sites in
Of special note is the combining of qualitative teams (e.g., Pettigrew, 1988). This allows the
with quantitative evidence. Although the terms case to b e viewed from the different perspectives
qualitative and case study are often used inter- of multiple observers. A variation on this tactic is
changeably (e.g., Yin, 1981), case study re- to give individuals on the team unique roles,
search can involve qualitative data only, quan- which increases the chances that investigators
titative only, or both (Yin, 1984). Moreover, the will view case evidence in divergent ways. For
combination of data types can be highly syner- example, interviews can be conducted by two
gistic. Quantitative evidence can indicate rela- person teams, with one researcher handling the
tionships which may not be salient to the re- interview questions, while the other records
searcher. It also can keep researchers from notes and observations (e.g.,Eisenhardt & Bour-
being carried away by vivid, but false, impres- geois, 1988).The interviewer has the perspective
sions in qualitative data, and it can bolster find- of personal interaction with the informant, while
ings when it corroborates those findings from the notetaker retains a different, more distant
qualitative evidence. The qualitative data are view. Another tactic is to create multiple re-
useful for understanding the rationale or theory search teams, with teams being assigned to
underlying relationships revealed in the quan- cover some case sites, but not others (e.g., Pet-
titative data or may suggest directly theory tigrew, 1988). The rationale behind this tactic is
which can then be strengthened by quantitative that investigators who have not met the infor-
support (Jick, 1979). Mintzberg (1979) described mants a n d have not become immersed in case
this synergy a s follows: details may bring a very different and possibly
more objective eye to the evidence. An extreme
For while systematic data create the foundation form of this tactic is to keep some member or
for our theories, it is the anecdotal data that en- members of the research team out of the field
able us to do the building. Theory building
seems to require rich description, the richness altogether by exclusively assigning to them the
that comes from anecdote. We uncover all kinds role of resident devil's advocate (e.g., Sutton &
of relationships in our hard data, but it is only Callahan, 1987).
through the use of this soft data that we are able
to explain them. (p. 587) Entering the Field
Also, of special note is the use of multiple in- A striking feature of research to build theory
vestigators. Multiple investigators have two key from case studies is the frequent overlap of data
advantages. First, they enhance the creative analysis with data collection. For example, Gla-
potential of the study. Team members often ser a n d Strauss (1967)argue for joint collection,
have complementary insights which a d d to the coding, a n d analysis of data. While many re-
richness of the data, and their different perspec- searchers do not achieve this degree of overlap,
tives increase the likelihood of capitalizing on most maintain some overlap.
any novel insights which may be in the data. Field notes, a running commentary to oneself
Second, the convergence of observations from and/or research team, are a n important means
multiple investigators enhances confidence in of accomplishing this overlap. As described by
Van Maanen (1988), field notes are a n ongoing given situation. In other situations adjustments
stream-of-consciousness commentary about can include the addition of data sources in se-
what is happening in the research, involving lected cases. For example, Sutton a n d Callahan
both observation a n d analysis-preferably sep- (1987) added observational evidence for one
arated from one another. case when the opportunity to attend creditors'
One key to useful field notes is to write down meetings arose, a n d Burgelman (1983) added
whatever impressions occur, that is, to react interviews with individuals whose importance
rather than to sift out what may seem important, became clear during data collection. Leonard-
because it is often difficult to know what will a n d Barton (1988) went even further by adding sev-
will not be useful in the future. A second key to eral experiments to probe her emergent theory
successful field notes is to push thinking in these in a study of the implementation of technical in-
notes by asking questions such a s "What a m I novations.
learning?" a n d "How does this case differ from These alterations create a n important ques-
the last?" For example, Burgelman (1983) kept tion: Is it legitimate to alter a n d even a d d data
extensive idea booklets to record his ongoing collection methods during a study? For theory-
thoughts in a study of internal corporate ventur- building research, the answer is "yes," because
ing. These ideas can be cross-case compari- investigators a r e trying to understand each case
sons, hunches about relationships, anecdotes, individually a n d in a s much depth a s is feasible.
a n d informal observations. Team meetings, in The goal is not to produce summary statistics
which investigators share their thoughts a n d about a set of observations. Thus, if a new data
emergent ideas, a r e also useful devices for over- collection opportunity arises or if a new line of
lapping data collection a n d analysis. thinking emerges during the research, it makes
Overlapping data analysis with data collec- sense to take advantage by altering data collec-
tion not only gives the researcher a head start in tion, if such a n alteration is likely to better
analysis but, more importantly, allows re- ground the theory or to provide new theoretical
searchers to take advantage of flexible data col- insight. This flexibility is not a license to be un-
lection. Indeed, a key feature of theory-building systematic. Rather, this flexibility is controlled
case research is the freedom to make adjust- opportunism in which researchers take advan-
ments during the data collection process. These tage of the uniqueness of a specific case a n d the
adjustments c a n be the addition of cases to emergence of new themes to improve resultant
probe particular themes which emerge. Gersick theory.
(1988), for example, added several cases to her
original set of student teams in order to more Analyzing Within-Case Data
closely observe transition point behaviors
among project teams. These transition point be- Analyzing data is the heart of building theory
haviors had unexpectedly proved interesting, from case studies, but it is both the most difficult
a n d Gersick added cases in order to focus more a n d the least codified part of the process. Since
closely on the transition period. published studies generally describe research
Additional adjustments can be made to data sites a n d data collection methods, but give little
collection instruments, such a s the addition of space to discussion of analysis, a huge chasm
questions to a n interview protocol or questions to often separates data from conclusions. As Miles
a questionnaire (e.g., Harris & Sutton, 1986). a n d Huberman (1984, p. 16) wrote: "One cannot
These adjustments allow the researcher to ordinarily follow how a researcher got from 3600
probe emergent themes or to take advantage of pages of field notes to the final conclusions,
special opportunities which may be present in a sprinkled wth vivid quotes though they may be."
However, several key features of analysis can familiarity with each case which, in turn, accel-
be identified. erates cross-case comparison.
One key step is within-case analysis. The im- Searching for Cross-Case Patterns
portance of within-case analysis is driven by
one of the realities of case study research: a Coupled with within-case analysis is cross-
staggering volume of data. As Pettigrew (1988) case search for patterns. The tactics here are
described, there is a n ever-present danger of driven by the reality that people a r e notoriously
"death by data asphyxiation." For example, poor processors of information. They leap to con-
Mintzberg a n d McHugh (1985) examined over clusions based on limited data (Kahneman &
2500 movies in their study of strategy making at Tversky, 1973),they a r e overly influenced by the
the National Film Board of Canada-and that vividness (Nisbett & Ross, 1980) or by more elite
was only part of their evidence. The volume of respondents (Miles & Huberman, 1984), they ig-
data is all the more daunting because the re- nore basic statistical properties (Kahneman &
search problem is often open-ended. Within- Tversky, 1973), or they sometimes inadvertently
case analysis can help investigators cope with drop disconfirming evidence (Nisbett & Ross,
this deluge of data. 1980). The danger is that investigators reach
Within-case analysis typically involves de- premature a n d even false conclusions a s a re-
tailed case study write-ups for each site. These sult of these information-processing biases.
write-ups are often simply pure descriptions, but Thus, the key to good cross-case comparison is
they a r e central to the generation of insight (Ger- counteracting these tendencies by looking at the
sick, 1988; Pettigrew, 1988) because they help data in many divergent ways.
researchers to cope early in the analysis process One tactic is to select categories or dimen-
with the often enormous volume of data. How- sions, a n d then to look for within-group similar-
ever, there is no standard format for such anal- ities coupled with intergroup differences. Di-
ysis. Quinn (1980) developed teaching cases for mensions can be suggested by the research
each of the firms in his study of strategic decision problem or by existing literature, or the re-
making in six major corporations a s a prelude to searcher can simply choose some dimensions.
his theoretical work. Mintzberg a n d McHugh For example, in a study of strategic decision
(1985) compiled a 383-page case history of the making, Bourgeois and Eisenhardt ( 1988) sifted
National Film Board of Canada. These authors cases into various categories including founder-
coupled narrative description with extensive use run vs. professional management, high vs. low
of longitudinal graphs tracking revenue, film performance, first vs. second generation prod-
sponsorship, staffing, film subjects, a n d so on. uct, a n d large vs. small size. Some categories
Gersick (1988) prepared transcripts of team such a s size a n d product generation revealed
meetings. Leonard-Barton ( 1988) used tabular no clear patterns, but others such a s perfor-
displays a n d graphs of information about each mance led to important patterns of within-group
case. Abbott (1988) suggested using sequence similarity a n d across-group differences. An ex-
analysis to organize longitudinal data. In fact, tension of this tactic is to use a 2 x 2 or other cell
there a r e probably a s many approaches a s re- design to compare several categories at once, or
searchers. However, the overall idea is to be- to move to a continuous measurement scale
come intimately familiar with each case a s a which permits graphing.
stand-alone entity. This process allows the A second tactic is to select pairs of cases a n d
unique patterns of each case to emerge before then to list the similarities a n d differences be-
investigators push to generalize patterns across tween each pair. This tactic forces researchers
cases. In addition, it gives investigators a rich to look for the subtle similarities a n d differences
between cases. The juxtaposition of seemingly a close fit with the data. Also, cross-case search-
similar cases by a researcher looking for differ- ing tactics enhance the probability that the in-
ences can break simplistic frames. In the same vestigators will capture the novel findings which
way, the search for similarity in a seemingly dif- may exist in the data.
ferent pair also can lead to more sophisticated
Shaping Hypotheses
understanding. The result of these forced com-
parisons can be new categories and concepts From the within-site analysis plus various
which the investigators did not anticipate. For cross-site tactics a n d overall impressions, tenta-
example, Eisenhardt a n d Bourgeois (1988)found tive themes, concepts, a n d possibly even rela-
that CEO power differences dominated initial tionships between variables begin to emerge.
impressions across firms. However, this paired The next step of this highly iterative process is to
comparison process led the researchers to see compare systematically the emergent frame
that the speed of the decision process was with the evidence from each case in order to
equally important (Eisenhardt, in press). Fi- assess how well or poorly it fits with case data.
nally, a n extension of this tactic is to group cases The central idea is that researchers constantly
into threes or fours for comparison. compare theory a n d data-iterating toward a
A third strategy is to divide the data by data theory which closely fits the data. A close fit is
source. For example, one researcher combs ob- important to building good theory because it
servational data, while another reviews inter- takes advantage of the new insights possible
views, a n d still another works with question- from the data a n d yields a n empirically valid
naire evidence. This tactic was used in the sep- theory.
aration of the analyses of qualitative a n d One step in shaping hypotheses is the sharp-
quantitative data in a study of strategic decision ening of constructs. This is a two-part process
making (Bourgeois & Eisenhardt, 1988; Eisen- involving (1) refining the definition of the con-
hardt & Bourgeois, 1988). This tactic exploits the struct a n d (2) building evidence which measures
unique insights possible from different types of the construct in each case. This occurs through
data collection. When a pattern from one data constant comparison between data a n d con-
source is corroborated by the evidence from an- structs so that accumulating evidence from di-
other, t h e finding is stronger a n d better verse sources converges on a single, well-
grounded. When evidence conflicts, the re- defined construct. For example, in their study of
searcher can sometimes reconcile the evidence stigma management in bankruptcy, Sutton a n d
through deeper probing of the meaning of the Callahan ( 1987)developed constructs which de-
differences. At other times, this conflict exposes scribed the reaction of customers a n d other par-
a spurious or random pattern, or biased think- ties to the declaration of bankruptcy by the focal
ing in the analysis. A variation of this tactic is to firms. The iterative process involved data from
split the data into groups of cases, focusing on multiple sources: initial semi-structured tele-
one group of cases initially, while later focusing phone conversations; interviews with key infor-
on the remaining cases. Gersick (1988)used this mants including the firm's president, other
tactic in separating the analyses of the student executives, a major creditor, a n d a lawyer; U.S.
group cases from her other cases. Bankruptcy Court records; observation of a
Overall, the idea behind these cross-case creditors' meeting; a n d secondary source mate-
searching tactics is to force investigators to go rial including newspaper a n d magazine articles
beyond initial impressions, especially through and firm correspondence. The authors iterated
the use of structured and diverse lenses on the between constructs a n d these data. They even-
data. These tactics improve the likelihood of ac- tually developed definitions a n d measures for
curate a n d reliable theory, that is, a theory with several constructs: disengagement, bargaining
for a more favorable exchange relationship, ined for each case, not for the aggregate cases.
denigration via rumor, a n d reduction in the Thus, the underlying logic is replication, that is,
quality of participation. the logic of treating a series of cases a s a series
This process is similar to developing a single of experiments with each case serving to con-
construct measure from multiple indicators in firm or disconfirm the hypotheses (Yin, 1984).
hypothesis-testing research. That is, researchers Each case is analogous to a n experiment, a n d
use multiple sources of evidence to build con- multiple cases are analogous to multiple exper-
struct measures, which define the construct and iments. This contrasts with the sampling logic of
distinguish it from other constructs. In effect, the traditional, within-experiment, hypothesis-
researcher is attempting to establish construct testing research in which the aggregate rela-
validity. The difference is that the construct, its tionships across the data points are tested using
definition, a n d measurement often emerge from summary statistics such a s F values (Yin, 1984).
the analysis process itself, rather than being In replication logic, cases which confirm
specified a priori. A second difference is that no emergent relationships enhance confidence in
technique like factor analysis is available to col- the validity of the relationships. Cases which
lapse multiple indicators into a single construct disconfirm the relationships often can provide
measure. The reasons a r e that the indicators a n opportunity to refine a n d extend the theory.
may vary across cases (i.e., not all cases may For example, in the study of the politics of stra-
have all measures), a n d qualitative evidence tegic decision making, Eisenhardt a n d Bour-
(which is common in theory-building research) geois (1988) found a case which did not fit with
is difficult to collapse. Thus, many researchers the proposition that political coalitions have sta-
rely on tables which summarize a n d tabulate ble memberships. Further examination of this
the evidence underlying the construct (Miles & disconfirming case indicated that the executive
Huberman, 1984; Sutton & Callahan, 1987). For team in this case had been newly formed at the
example, Table 3 is a tabular display of the ev- time of the study. This observation plus replica-
idence grounding the CEO power construct tion in another case led to a refinement in the
used by Eisenhardt a n d Bourgeois (1988),which emergent theory to indicate that increasing sta-
included qualitative personality descriptions, bilization of coalitions occurs over time.
quantitative scores from questionnaires, a n d At this point, the qualitative data are particu-
quotation examples. The reasons for defining larly useful for understanding why or why not
a n d building evidence for a construct apply in emergent relationships hold. When a relation-
theory-building research just a s they do in tra- ship is supported, the qualitative data often pro-
ditional, hypothesis-testing work. That is, care- vide a good understanding of the dynamics un-
ful construction of construct definitions and evi- derlying the relationship, that is, the "why" of
dence produces the sharply defined, measur- what is happening. This is crucial to the estab-
able constructs which are necessary for strong lishment of internal validity. Just a s in hypothe-
theory. sis-testing research a n apparent relationship
A second step in shaping hypotheses is veri- may simply be a spurious correlation or may
fying that the emergent relationships between reflect the impact of some third variable on each
constructs fit with the evidence in each case. of the other two. Therefore, it is important to dis-
Sometimes a relationship is confirmed by the cover the underlying theoretical reasons for why
case evidence, while other times it is revised, the relationship exists. This helps to establish the
disconfirmed, or thrown out for insufficient evi- internal validity of the findings. For example, in
dence. This verification process is similar to that her study of project groups, Gersick (1988) iden-
in traditional hypothesis testing research. The tified a midpoint transition in the lives of most
key difference is that each hypothesis is exam- project groups. She then used extensive quali-
Table 3
Example of Tabulated Evidence for a Power Centralization Construct*


Decision Power Power Dominated Decision

Finn Description Score Distance" Functions Styleb Examplesc

First Strong 9.6 3.5 Mkt, R&D, Ops, Authoritarian Geoff (Chairman) is THE
Volatile Fin decision maker. He runs the
Dogmatic whole show. (VP, Marketing)
Alpha Impatient 9.6 3.8 Mkt, R&D, Ops, Authoritarian Thou shalt not hire wlo
Parental Fin Presidential approval. Thou
Tunes You Out shalt not promote wlo
Presidential approval. Thou
shalt not explore new
markets wlo Presidential
approval. (VP, Operations)
Cowboy Strong 9.1 3.1 Mkt, R&D, Fin Authoritarian The tone of meetings would
Power Boss Consensus change depending upon
Master whether he was in the room.
Strategist If he'd leave the room.
discussion would spread out,
go off the wall. It got back
on focus when he came
back. (Director of Marketing)
Neutron Organized 9.1 2.3 Mkt, Ops, Fin Authoritarian If there is a decision to make, I
Analytic will make it. (President)
Omicron Easy Going 8.4 1.2 Fin Consensus Bill (prior CEO) was a
Easy to Work suppressor of ideas. Jim is
With more open. (VP, Mfg.)
Promise People- 8.9 1.3 Ops, Fin Consensus (My philosophy is) to make
Oriented quick decisions involving a s
Pragmatic many people a s possible.
Forefront Aggressive 8.3 1.2 None Consensus Art depends on picking good
Team people and letting them
Player operate. (VP, Sales)
Zap Consensus- 7.5 0.3 Fin Consultative It's very open. We're successful
Style most of the time in building
People- consensus. (VP, Engineering)
" Difference between CEO power score and score of next most powerful executive.
Authoritarian-Decisions made either by CEO alone or in consultation with only one person.
Consultative-Decisions made by CEO in consultation with either most of or all of the team.
Consensus-Decisions made by entire team in a group format.
" Individual in parentheses is the source of the quotation.
' Taken from Eisenhardt & Bourgeois, 1988.

tative data to understand the cognitive and mo- building research involves measuring con-
tivational reasons why such abrupt a n d pre- structs a n d verifying relationships. These pro-
cisely timed transitions occur. cesses a r e similar to traditional hypothesis-
Overall, s h a p i n g hypotheses in theory- testing research. However, these processes are
more judgmental in theory-building research other conflicts allowed these researchers to es-
because researchers cannot apply statistical tablish the unique features of strategy making in
tests such a s a n F statistic. The research team a n "adhocracy" in relief against "machine
must judge the strength a n d consistency of rela- bureaucracies" a n d "entrepreneurial firms."
tionships within and across cases and also fully The result was a sharper theory of strategy for-
display the evidence a n d procedures when the mation in all three types of organizations.
findings are published, so that readers may ap- Similarly, in a study of politics, Eisenhardt
ply their own standards. a n d Bourgeois (1988) contrasted the finding that
centralized power leads to politics with the pre-
Enfolding Literature vious finding that decentralized power creates
An essential feature of theory building is com- politics. These conflicting findings forced the
parison of the emergent concepts, theory, or hy- probing of both the evidence a n d conflicting re-
potheses with the extant literature. This involves search to discover the underlying reasons for
asking what is this similar to, what does it con- the conflict. An underlying similarity in the ap-
tradict, and why. A key to this process is to con- parently dissimilar situations was found. That is,
sider a broad range of literature. both power extremes create a climate of frustra-
Examining literature which conflicts with the tion, which leads to a n emphasis on self-interest
emergent theory is important for two reasons. and ultimately politics. In these extreme situa-
First, if researchers ignore conflicting findings, tions, the "structure of the game" becomes a n
then confidence in the findings is reduced. For interpersonal competition among the execu-
example, readers may assume that the results tives. In contrast, the research showed that a n
are incorrect (a challenge to internal validity), or intermediate power distribution fosters a sense
if correct, are idiosyncratic to the specific cases of personal efficacy among executives a n d ulti-
of the study (a challenge to generalizability). mately collaboration, not politics, for the good of
Second a n d perhaps more importantly, conflict- the entire group. This reconciliation integrated
ing literature represents a n opportunity. The the conflicting findings into a single theoretical
juxtaposition of conflicting results forces re- perspective, and raised the theoretical level a n d
searchers into a more creative, framebreaking generalizability of the results.
mode of thinking than they might otherwise be Literature discussing similar findings is impor-
able to achieve. The result can be deeper insight tant a s well because it ties together underlying
into both the emergent theory and the conflict- similarities in phenomena normally not associ-
ing literature, a s well a s sharpening of the limits ated with each other. The result is often a theory
to generalizability of the focal research. For ex- with stronger internal validity, wider generaliz-
ample, in their study of strategy making at the ability, and higher conceptual level. For exam-
National Film Board of Canada, Mintzberg and ple, in her study of technological innovation in a
McHugh (1985) noted conflicts between their major computer corporation, Leonard-Barton
findings for this highly creative organization (1988) related her findings on the mutual adap-
a n d prior results at Volkswagenwerk and other tation of technology a n d the host organization to
sites. In the earlier studies, they observed differ- similar findings in the education literature. In so
ences in the patterns of strategic c h a n g e doing, Leonard-Barton strengthened the confi-
whereby periods of convergence were long and dence that her findings were valid a n d gener-
periods of divergence were short a n d very alizable because others had similar findings in a
abrupt. In contrast, the National Film Board ex- very different context. Also, the tie to mutual ad-
hibited a pattern of regular cycles of conver- aptation processes in the education setting
gence a n d divergence, coupled with a long- sharpened a n d enriched the conceptual level of
term trend toward greater diversity. This a n d the study.
Similarly, Gersick (1988)linked the sharp mid- such a s time and money to dictate when case
point transition in project group development to collection ends. In fact, it is not uncommon for
the more general punctuated equilibrium phe- researchers to plan the number of cases in ad-
nomenon, to the literature on the adult midlife v a n c e . For e x a m p l e , the Warwick g r o u p
transition, a n d to strategic transitions within or- planned their study of strategic change a n d
ganizations. This linkage with a variety of liter- competitiveness in British firms to include eight
ature in other contexts raises the readers' confi- firms (Pettigrew, 1988). This kind of planning
dence that Gersick had observed a valid phe- may be necessary because of the availability of
nomenon within her small number of project resources and because time constraints force re-
teams. It also allowed her to elevate the concep- searchers to develop cases in parallel. Finally,
tual level of her findings to the more fundamen- while there is no ideal number of cases, a num-
tal level of punctuated equilibrium, and strength- ber between 4 a n d 10 cases usually works well.
en their likely generalizability to other project With fewer than 4 cases, it is often difficult to
teams. Finally, Burgelman ( 1983) strengthened generate theory with much complexity, and its
the theoretical scope and validity of his work by empirical grounding is likely to b e unconvinc-
tying his results on the process of new venture ing, unless the case has several mini-cases
development in a large corporation to the selec- within it, a s did the Mintzberg a n d McHugh
tion arguments of population ecology. The result study of the National Film Board of Canada.
again was a higher conceptual level for his find- With more than 10 cases, it quickly becomes dif-
ings and enhanced confidence in their validity. ficult to cope with the complexity and volume of
Overall, tying the emergent theory to existing the data.
literature enhances the internal validity, gener- In the second closure issue, when to stop iter-
alizability, and theoretical level of theory build- ating between theory and data, again, satura-
ing from case study research. While linking re- tion is the key idea. That is, the iteration process
sults to the literature is important in most re- stops when the incremental improvement to the-
search, it is particularly crucial in theory- ory is minimal. The final product of building the-
building research because the findings often ory from case studies may be concepts (e.g., the
rest on a very limited number of cases. In this Mintzberg a n d Waters, 1982, deliberate a n d
situation, any further corroboration of internal emergent strategies), a conceptual framework
validity or generalizability is a n important im- ( e . g . , Harris & Sutton's, 1986, framework of
provement. bankruptcy), or propositions or possibly mid-
range theory (e.g., Eisenhardt a n d Bourgeois's,
Reaching Closure
1988, midrange theory of politics in high velocity
Two issues are important in reaching closure: environments). On the downside, the final prod-
when to stop adding cases, a n d when to stop uct may be disappointing. The research may
iterating between theory a n d data. In the first, simply replicate prior theory, or there may be no
ideally, researchers should stop adding cases clear patterns within the data. The steps for
when theoretical saturation is reached. (Theo- building theory from case studies are summa-
retical saturation is simply the point at which rized in Table 1.
incremental learning is minimal because the re-
searchers are observing phenomena seen be- Comparison with Other Literature
fore, Glaser and Strauss, 1967.)This idea is quite
similar to ending the revision of a manuscript The process described here has similarities
when the incremental improvement in its qual- with the work of others. For example, I have
ity is minimal. In practice, theoretical saturation drawn upon the ideas of theoretical sampling,
often combines with pragmatic considerations theoretical saturation, and overlapped coding,
data collection, a n d analysis from Glaser a n d Discussion
Strauss (1967). Also, the notions of case study
design, replication logic, and concern for inter- The process of building theory from case study
nal validity have been incorporated from Yin research is a strikingly iterative one. While a n
(1984). The tools of tabular display of evidence investigator may focus on one part of the process
from Miles and Huberman (1984) were particu- at a time, the process itself involves constant it-
larly helpful in the discussion of building evi- eration backward a n d forward beween steps.
dence for constructs. For example, a n investigator may move from
However, the process described here has im- cross-case comparison, back to redefinition of
portant differencesfrom previous work. First, it the research question, and out to the field to
is focused on theory building from cases. In con- gather evidence on a n additional case. Also,
trast, with the exception of Glaser and Strauss the process is alive with tension between diver-
(1967),previous work was centered on other top- gence into new ways of understanding the data
ics such a s qualitative data analysis (e.g.,Miles, a n d convergence onto a single theoretical
1979; Miles & Huberman, 1984; Kirk & Miller, framework. For example, the process involves
1986), case study design (Yin, 1981, 1984; Mc- the use of multiple investigators a n d multiple
Clintock et al., 1979), a n d ethnography (Van data collection methods a s well a s a variety of
Maanen, 1988). To a large extent, Glaser and cross-case searching tactics. Each of these tac-
Strauss (1967) focused on defending building tics involves viewing evidence from diverse per-
theory from cases, rather than on how actually spectives. However, the process also involves
to do it. Thus, while these previous writings pro- converging on construct definitions, measures,
vide pieces of the process, they do not provide and a framework for structuring the findings. Fi-
(nor do they intend to provide) a framework for nally, the process described here is intimately
theory building from cases a s developed here. tied with empirical evidence.
Second, the process described here contrib- Strengths of Theory Building from Cases
utes new ideas. For example, the process in-
cludes a priori specification of constructs, popu- One strength of theory building from cases is
lation specification, flexible instrumentation, its likelihood of generating novel theory. Cre-
multiple investigators, cross-case analysis tac- ative insight often arises from the juxtaposition of
tics, a n d several uses of literature. Their inclu- contradictory or paradoxical evidence (Cam-
sion plus their illustration using examples from eron & Quinn, 1988).As Bartunek (1988)argued,
research studies a n d comparison with tradi- the process of reconciling these contradictions
tional, hypothesis-testing research synthesizes, forces individuals to reframe perceptions into a
extends, a n d adds depth to existing views of the- new gestalt. Building theory from case studies
ory-building research. centers directly on this kind of juxtaposition. That
Third, particularly in comparison with Strauss is, attempts to reconcile evidence across cases,
(1987) and Van Maanen (19881, the process de- types of data, and different investigators, a n d
scribed here adopts a positivist view of research. between cases and literature increase the like-
That is, the process is directed toward the devel- lihood of creative reframing into a new theoret-
opment of testable hypotheses and theory which ical vision. Although a myth surrounding theory
are generalizable across settings. In contrast, building from case studies is that the process is
authors like Strauss and Van Maanen are more limited by investigators' preconceptions, in fact,
concerned that a rich, complex description of just the opposite is true. This constant juxtaposi-
the specific cases under study evolve and they tion of conflicting realities tends to "unfreeze"
appear less concerned with development of thinking, a n d so the process has the potential to
generalizable theory. generate theory with less researcher bias than
theory built from incremental studies or arm- staggering volume of rich data, there is a temp-
chair, axiomatic deduction. tation to build theory which tries to capture ev-
A second strength is that the emergent theory erything. The result can be theory which is very
is likely to be testable with constructs that can be rich in detail, but lacks the simplicity of overall
readily measured and hypotheses that can be perspective. Theorists working from case data
proven false. Measurable constructs are likely can lose their sense of proportion a s they con-
because they have already been measured dur- front vivid, voluminous data. Since they lack
ing the theory-building process. The resulting quantitative gauges such a s regression results
hypotheses a r e likely to b e verifiable for the or observations across multiple studies, they
same reason. That is, they have already under- may be unable to assess which are the most
gone repeated verification during the theory- important relationships and which are simply
building process. In contrast, theory which is idiosyncratic to a particular case.
generated apart from direct evidence may have Another weakness is that building theory from
testability problems. For example, population cases may result in narrow a n d idiosyncratic
ecology researchers borrowed the niche con- theory. Case study theory building is a bottom
cept from biology. This construct has proven dif- up approach such that the specifics of data pro-
ficult to operationalize for many organizational duce the generalizations of theory. The risks are
researchers, other than its originators. One rea- that the theory describes a very idiosyncratic
son may be its obscure definition, which ham- phenomenon or that the theorist is unable to
pers measurability: ". . . that area in constraint raise the level of generality of the theory. In-
space (the space whose dimensions are levels of deed, many of the grounded case studies men-
resources, etc.) in which the population outcom- tioned earlier resulted in modest theories. For
petes all other local populations" (Hannan & example, Gersick (1988) developed a model of
Freeman, 1977, p. 947). One might ask: How do group development for teams with project dead-
you measure a n area in constraint space? lines, Eisenhardt a n d Bourgeois (1988) devel-
A third strength is that the resultant theory is oped a mid-range theory of politics in high ve-
likely to be empirically valid. The likelihood of locity environments, and Burgelman ( 1983) pro-
valid theory is high because the theory-building posed a model of new product ventures in large
process is so intimately tied with evidence that it corporations. Such theories are likely to be test-
is very likely that the resultant theory will be able, novel, a n d empirically valid, but they do
consistent with empirical observation. In well- lack the sweep of theories like resource depen-
executed theory-building research, investiga- dence, population ecology, and transaction cost.
tors answer to the data from the beginning of the They are essentially theories about specific phe-
research. This closeness can lead to a n intimate nomena. To their credit, many of these theorists
sense of things-"how they feel, smell, seem" tie into broader theoretical issues such a s adap-
(Mintzberg, 1979). This intimate interaction with tation, punctuated equilibrium, a n d bounded
actual evidence often produces theory which rationality, but ultimately they are not theories
closely mirrors reality. about organization in any grand sense. Perhaps
"grand" theory requires multiple studies-an
Weaknesses of Theory Building from Cases
accumulation of both theory-building and the-
However, some characteristics that lead to ory-testing empirical studies.
strengths in theory building from case studies
also lead to weaknesses. For example, the in-
tensive use of empirical evidence can yield the- When is it appropriate to conduct theory-
ory which is overly complex. A hallmark of good building case study research? In normal sci-
theory is parsimony, but given the typically ence, theory is developed through incremental
empirical testing a n d extension (Kuhn, 1970). Second, the assessment of theory-building re-
Thus, the theory-building process relies on past search also depends upon empirical issues:
literature and empirical observation or experi- strength of method and the evidence grounding
ence a s well a s on the insight of the theorist to the theory. Have the investigators followed a
build incrementally more powerful theories. careful analytical procedure? Does the evidence
However, there are times when little is known support the theory? Have the investigators ruled
about a phenomenon, current perspectives out rival explanations? Just a s in other empirical
seem inadequate because they have little em- research, investigators should provide informa-
pirical substantiation, or they conflict with each tion on the sample, data collection procedures,
other or common sense. Or, sometimes, seren- and analysis. Also, they should display enough
dipitous findings in a theory-testing study sug- evidence for each construct to allow readers to
gest the need for a new perspective. In these make their own assessment of the fit with theory.
situations, theory building from case study re- While there are no concise measures such a s
search is particularly appropriate because the- correlation coefficients or F values, nonetheless
ory building from case studies does not rely on thorough reporting of information should give
previous literature or prior empirical evidence. confidence that the theory is valid. Overall, a s in
Also, the conflict inherent in the process is likely hypothesis testing, a strong theory-building
to generate the kind of novel theory which is study has a good, although not necessarily per-
desirable when extant theory seems inade- fect, fit with the data.
quate. For example, Van d e Ven and Poole (in Finally, strong theory-building research should
press) have argued that such a n approach is result in new insights. Theory building which
especially useful for studying the new area of simply replicates past theory is, at best, a mod-
longitudinal change processes. In sum, building est contribution. Replication is appropriate in
theory from case study research is most appro- theory-testing research, but in theory-building
priate in the early stages of research on a topic research, the goal is new theory. Thus, a strong
or to provide freshness in perspective to a n al- theory-building study presents new, perhaps
ready researched topic. framebreaking, insights.
How should theory-building research using Conclusions
case studies be evaluated? To begin, there is no
generally accepted set of guidelines for the as- The purpose of this article is to describe the
sessment of this type of research. However, sev- process of theory building from case studies.
eral criteria seem appropriate. Assessment The process, outlined in Table 1, has features
turns on whether the concepts, framework, or which range from selection of the research
propositions that emerge from the process are question to issues in reaching closure. Several
"good theory." After all, the point of the process conclusions emerge.
is to develop or at least begin to develop theory. Theory developed from case study research is
Pfeffer (1982) suggested that good theory is par- likely to have important strengths like novelty,
simonious, testable, and logically coherent, and testability, and empirical validity, which arise
these criteria seem appropriate here. Thus, a from the intimate linkage with empirical evi-
strong theory-building study yields good theory dence. Second, given the strengths of this the-
(that is, parsimonious, testable, and logically co- ory-building approach a n d its independence
herent theory) which emerges at the end, not from prior literature or past empirical observa-
beginning, of the study. tion, it is particularly well-suited to new research
areas or research areas for which existing the- concept development (e.g., parsimony, testabil-
ory seems inadequate. This type of work is ity, logical coherence) and are grounded in con-
highly complementary to incremental theory vincing evidence.
building from normal science research. The Most empirical studies lead from theory to
former is useful in early stages of research on a data. Yet, the accumulation of knowledge in-
topic or when a fresh perspective is needed, volves a continual cycling between theory and
while the latter is useful in later stages of knowl- data. Perhaps this article will stimulate some re-
edge. Finally, several guidelines for assessing searchers to complete the cycle by conducting
the quality of theory building from case studies research that goes in the less common direction
have been suggested. Strong studies are those from data to theory, a n d equally important, per-
which present interesting or framebreaking haps it will help others become informed con-
theories which meet the tests of good theory or sumers of the results.

Abbott, A. (1988,September) Workshop on sequence meth- Gersick, C. (1988)Tlme and transition In work teams:Toward
ods. Natlonal Science Foundation Conference on Longl- a new model of group development. Academy of Man-
tudinal Research Methods in Organizations, Austin. agement Journal, 31, 9-41.
Alllson, G. (197 1 ) Essence of decision. Boston: Llttle, Brown. Glaser, B., & Strauss, A. (1967) The discovery of grounded
Anderson, P. (1983)Decision making by objection and the theory: Strategies o f qualitative research. London:
Cuban misslle crisis. Administrative Science Quarterly, Wledenfeld and Nicholson.
28, 201-222. Hannan, M., & Freeman, J . ( 1977)The population ecology of
Bartunek, J . (1988)The dynamics of personal and organiza- organizations. American Journal of Sociology, 82, 929-
tional reframing. In R. Quinn & K. Cameron (Eds.),Para- 964.
dox and transformation: Towards a theory of change in
Harris, S., & Sutton, R. (1986)Functions of parting ceremo-
organization and management ( P P . 137-162). Cam-
nies in organlzatlons, Academy o f Management
bridge, MA: Ballinger.
Journal, 29, 5-30.
Bettenhausen, K., & Murnighan, J . K . (1986)The emergence
of norms in competitive decision-mahng groups, Admin- lick, T. (1979)Mixing qualitative and quantltatlve methods:
istrative Science Quarterly, 30, 350-372. Triangulation in action. Administrative Science Quarterly,
24, 602-61 1.
Bourgeois, L., & Eisenhardt, K . (1988)Strategic decision pro-
cesses In high veloclty environments: Four cases In the Kahneman, D . , & T v e r s k ~A.
, (lg73)O n the ~ s ~ c h o l o gofy
microcomputer industry. Manaqement Science, 34, 816- prediction. Psychological Review, 80, 237-251.
Kidder, T . (1982)Soul of a new machlne. New York: Avon.
Burgelman, R. (1983)A process model of internal corporate Kimberly, J . (1988)A review of Walter R. Nord and Sharon
venturing in a major diversifled firm. Administrative Sci- Tucker: Implementing routine and radical innovations.
ence Quarterly, 28, 223-244. Administrative Science Quarterly, 33, 314-316.
Cameron, K . , & Quinn, R. (1988)Organizational paradox Klrk, J . , & Miller, M. (1986)Reliability and validity in quali-
and transformation. In R. Quinn & K. Cameron (Eds.), tative research. Beverly Hllls, CA: Sage.
Paradox and transformation (pp. 1 - 18). Cambridge, MA:
Ballinger. Kuhn, T. (1970)The structure of scientific revolutions (2nd
ed.).Chicago: University of Chicago Press.
Eisenhardt, K . (in press) Mahng fast strategic decisions In
hlgh velocity envlronments. Academy of Management Leonard-Barton, D. (1988)Synergistic deslgn for case stud-
Journal. les: Longitudinal single-site and repl~catedmultiple-site.
Elsenhardt, K . , & Bourgeois, L. J . (1988)Polltics of strategic Paper presented at the National Science Foundation Con-
decislon malung In high velocity envlronments: Toward a ference on Longltudlnal Research Methods In Organlza-
mld-range theory. Academy of Management Journal, 31, tions, Austln.
737-770. McClintock, C., Brannon, D . , & Maynard-Moody, S. (1979)
Applying the logic of sample surveys to qualitative case Pfeffer,J . (1982) Organizations and organization theory.
studies: The case cluster method. Administrative Science Marshfield, MA: Piiman.
Quarterly, 612-629.
Pinfleld, L. (1986)A fleld evaluation of perspectives on orga-
Miles, M. (1979)Qualitative data as an attractive nuisance: nizatlonaldecision malung, ~ d ~ iscience
~ Q~ ~~ ~t ~~ . ~
The problem of analysis. Administrative Science Quar- terly, 31, 365-388,
terlv. 24. 590-601.
Miles, M . , & Huberman, A. M . (1984)Qualitative data anal- Qulnn, J. B. (1980)Strategies for change. Homewood, IL:
ysis. Beverly Hills, CA: Sage Publlcatlons. Dow-JonesIrwin.
Mlntzberg, H. (1979) An emerglng strategy of "dlrect" re- Selznick, P. (1949)TVA and the grass roots. Berkeley, CA:
search. Administrative Science Quarterly, 24, 580-589. University of California Press.
Mintzberg,H . ,8( McHugh,A. Strategy in an Strauss, A. (1987)Qualitative analysis for social scientists.
adhocracy. Administrative Science Quarterly, 30, 160- cambridge, ~ ~ ~cambrldge ~ d :p ress,
l ~ university

Mlntzberg, H., & Waters, J . (1982)Tracking strategy in an Sutton, R.8 & Callahan, A. (1987)The stigma of bankruptcy:
entrepreneurial firm, ~~~d~~~ o f ~~~~~~~~~t ~ ~ S~olled
~ organizational
~ ~ image
~ andl its management.
, Acad-
25, 465-499. emy of Management Journal, 30, 405-436.
Nisbett, R., & Ross, L. (1980)Human inference: Strategies Van de Ven, A., & Poole, M. S. ! ~ press)
n Methods to develop
and shortcomings of soc~aljudgment. Englewood Cllffs, a grounded theory of lnnovatlon processes in the Mlnne-
N J : Prentice-Hall. sota Innovation Research Program. Organization Sci-
Perrow, C . (1986) Complex organizations (3rd ed.). New ence,
York: Random House. Van Maanen, J. (1988)Tales of the field: On wrlting ethnog-
Pettiqrew, A. (1973)The politics of organizational decision raphy. Chicago: University of Chicago Press.
maklng. London: Tavlstock.
Yin, R. (1981)The case study crisis: Some answers. Admln-
Pettigrew, A. (1988)Longltudinal field research on change: istrativescience Q ~26, 58-65,~ ~ ~ ~ ~
Theory and practice. Paper
. presented
. at the Natlonal Sci-
ence Foundation Conference on Longltudinal Research Yin, R. (1984)Case study research. Beverly Hills, CA: Sage
Methods in Organizations, Austin. Pubhcat~ons.

Kathleen M . Elsenhardt (Ph.D.,Stanford Unlversity)is

Assistant Professor at Stanford University. Correspon-
dence can be sent to her at the Department of Indus-
trial Engineering and Engineering Management, 346
Terman Bullding, Stanford University, Stanford, Cal-
ifornia, 94305.
I appreciate the helpful comments of Paul Adler, Ken-
nith Bettenhausen, Constance Gersick, James Freder-
~ckson,James Jucker, Deborah Myerson, Dorothy Le-
onard-Barton, Robert Sutton, and the partic~pantsin
the Stanford NlMH Colloquium, I also benefitted from
informal conversations with many participants at the
Natlonal Sclence Foundation Conference on Longitu-
dinal Research Methods in Organizations, Austin,

- Page 1 of 4 -

You have printed the following article:

Building Theories from Case Study Research
Kathleen M. Eisenhardt
The Academy of Management Review, Vol. 14, No. 4. (Oct., 1989), pp. 532-550.
Stable URL:

This article references the following linked citations. If you are trying to access articles from an
off-campus location, you may be required to first logon via your library web site to access JSTOR. Please
visit your library's website or contact a librarian to learn about options for remote access to JSTOR.


Decision Making by Objection and the Cuban Missile Crisis

Paul A. Anderson
Administrative Science Quarterly, Vol. 28, No. 2. (Jun., 1983), pp. 201-222.
Stable URL:

The Emergence of Norms in Competitive Decision-Making Groups

Kenneth Bettenhausen; J. Keith Murnighan
Administrative Science Quarterly, Vol. 30, No. 3. (Sep., 1985), pp. 350-372.
Stable URL:

Strategic Decision Processes in High Velocity Environments: Four Cases in the

Microcomputer Industry
L. J. Bourgeois, III; Kathleen M. Eisenhardt
Management Science, Vol. 34, No. 7. (Jul., 1988), pp. 816-835.
Stable URL:

A Process Model of Internal Corporate Venturing in the Diversified Major Firm

Robert A. Burgelman
Administrative Science Quarterly, Vol. 28, No. 2. (Jun., 1983), pp. 223-244.
Stable URL:

- Page 2 of 4 -

Politics of Strategic Decision Making in High-Velocity Environments: Toward a Midrange

Kathleen M. Eisenhardt; L. J. Bourgeois III
The Academy of Management Journal, Vol. 31, No. 4. (Dec., 1988), pp. 737-770.
Stable URL:

Time and Transition in Work Teams: Toward a New Model of Group Development
Connie J. G. Gersick
The Academy of Management Journal, Vol. 31, No. 1. (Mar., 1988), pp. 9-41.
Stable URL:

The Population Ecology of Organizations

Michael T. Hannan; John Freeman
The American Journal of Sociology, Vol. 82, No. 5. (Mar., 1977), pp. 929-964.
Stable URL:

Functions of Parting Ceremonies in Dying Organizations

Stanley G. Harris; Robert I. Sutton
The Academy of Management Journal, Vol. 29, No. 1. (Mar., 1986), pp. 5-30.
Stable URL:

Mixing Qualitative and Quantitative Methods: Triangulation in Action

Todd D. Jick
Administrative Science Quarterly, Vol. 24, No. 4, Qualitative Methodology. (Dec., 1979), pp.
Stable URL:

- Page 3 of 4 -

Review: [Untitled]
Reviewed Work(s):
Implementating Routine and Radical Innovations. by Walter R. Nord; Sharon Tucker
John R. Kimberly
Administrative Science Quarterly, Vol. 33, No. 2. (Jun., 1988), pp. 314-316.
Stable URL:

Applying the Logic of Sample Surveys to Qualitative Case Studies: The Case Cluster Method
Charles C. McClintock; Diane Brannon; Steven Maynard-Moody
Administrative Science Quarterly, Vol. 24, No. 4, Qualitative Methodology. (Dec., 1979), pp.
Stable URL:

Qualitative Data as an Attractive Nuisance: The Problem of Analysis

Matthew B. Miles
Administrative Science Quarterly, Vol. 24, No. 4, Qualitative Methodology. (Dec., 1979), pp.
Stable URL:

An Emerging Strategy of "Direct" Research

Henry Mintzberg
Administrative Science Quarterly, Vol. 24, No. 4, Qualitative Methodology. (Dec., 1979), pp.
Stable URL:

Strategy Formation in an Adhocracy

Henry Mintzberg; Alexandra McHugh
Administrative Science Quarterly, Vol. 30, No. 2. (Jun., 1985), pp. 160-197.
Stable URL:

- Page 4 of 4 -

Tracking Strategy in an Entrepreneurial Firm

Henry Mintzberg; James A. Waters
The Academy of Management Journal, Vol. 25, No. 3. (Sep., 1982), pp. 465-499.
Stable URL:

A Field Evaluation of Perspectives on Organizational Decision Making

Lawrence T. Pinfield
Administrative Science Quarterly, Vol. 31, No. 3. (Sep., 1986), pp. 365-388.
Stable URL:

The Stigma of Bankruptcy: Spoiled Organizational Image and Its Management

Robert I. Sutton; Anita L. Callahan
The Academy of Management Journal, Vol. 30, No. 3. (Sep., 1987), pp. 405-436.
Stable URL:

The Case Study Crisis: Some Answers

Robert K. Yin
Administrative Science Quarterly, Vol. 26, No. 1. (Mar., 1981), pp. 58-65.
Stable URL: