* Academy of Management Review

1999, Voi, 24, No, 4, 691-710.

STRATEGIES FOR THEORIZING FROM
PROCESS DATA
ANN LANGLEY
Universite du Quebec a Montreal
In this article I describe and compare a number of alternative generic strategies for
the analysis oi process data, looking at the consequences oi these strategies ior
emerging theories. I evaluate the strengths and weaknesses of the strategies in
terms oi their capacity to generate theory that is accurate, porsimonious. general,
and useful and suggest that method and theory are inextricably intertwined, that
multiple strategies are oiten advisable, and that no analysis strategy will produce
theory without an uncodiiiable creative leap, however small. Finally, I argue that
there is room in the organizational research literature ior more openness within
the academic community toward a variety of iorms oi coupling between theory and
data.
As change sweeps through industries, organi-
zations, and workgroups, we are seeing a surge
of interest among organizational researchers in
process theory and dynamic phenomena, such
as organizational learning (Cohen & Sproull,
1991), competitive interaction {Illnitch, D'Aveni,
& Lewin, 1996), innovation and change (Van de
Ven & Huber, 1990), and strategic evolution (Bar-
nett & Burgelman, 1996). One group of research-
ers has chosen to address these dynamics by
formulating a priori process theories and testing
them using coarse-grained longitudinal time se-
ries and event-history methods. Another camp
has chosen rather to plunge itself deeply into
the processes themselves, collecting fine-
grained qualitative dataoften, but not always,
in real timeand attempting to extract theory
from the ground up {Bower, 1997; Pettigrew, 1992;
Van de Ven, 1992). The philosophy of this camp
I am indebted to colleagues Louise CStS. Jean-Louis
Denis, and Jean Truax. who worked with me on several
research projects that inspired this article. I also thank col-
leagues, students, and conference participants ior their
helplul comments and discussions and, especially, Chris-
tiane Demers, Richard Dery, Taieb Hafsi. Veronika Kisfalvi.
G6rard Koenig, Gilbert Laporte. Jean Perrien, and Raymond-
Alain Thietart. Bob Sutton and two anonymous reviewers
also encouraged me to improve the articie. An earlier and
less developed rendering of these ideas was published
in French in Management International, under the title
"L'etude des processus strategiques: Defis conceptuels et
analytiques" (Langley, 1997). This article was prepared
while I was on sabbatical leave al l'Ecole des Hautes fitudes
Commerciales de Montreal. I thank the school for their wel-
come and their support.
is that to truly understand how and why events
play out over time, we must examine them di-
rectly (Mintzberg, 1979).
I identify myself as a member of the second
camp, but in no way think the task we have set
ourselves is easy. Process data are messy. Mak-
ing sense of them is a constant challenge. In this
article I examine a number of different strate-
gies for approaching this task. My objective is
not to advocate one strategy or another, or even
to propose radically new strategies (although I
do draw on my own research with colleagues in
delineating some of them), but, rather, to con-
sider the strengths and weaknesses of different
modes of analysis of process data in terms of
their capacity to generate theory that is accu-
rate, parsimonious, general, and useful (Weick,
1979). I further draw attention to the mutual de-
pendence between methods and theories.
I begin by clarifying what I mean by process
theory and process data and how I conceive of
the theory-building task. After presenting the
different analysis strategies, I discuss their var-
ious qualities, place them within an overall
framework, and argue for more openness within
the academic community toward a variety of
forms of coupling between theory and data.
PROCESS DATA AND PROCESS
THEORIZATION: THE CHALLENGE
Process data collected in real organizational
contexts have several characteristics that make
631
692
Academy o/ Management Review October
them difficult to analyze and manipulate.' First,
they deal mainly with sequences of "events":
conceptual entities that researchers are less fa-
miliar with. Second, they often involve multiple
levels and units of analysis whose boundaries
are ambiguous. Third, their temporal embed-
dedness often varies in terms of precision, du-
ration, and relevance. Finally, despite the pri-
mary focus on events, process data tend to be
eclectic, drawing in phenomena such as chang-
ing relationships, thoughts, feelings, and inter-
pretations. I elaborate briefly on these four char-
acteristics below.
Data Composed of Events
Process research is concerned with under-
standing how things evolve over time and why
they evolve in this way (see Van de Ven & Huber,
1990), and process data therefore consist largely
of stories about what happened and who did
what whenthat is, events, activities, and
choices ordered over time. In his classic work on
organization theory, Mohr (1982) makes a clear
distinction between what he calls "variance the-
ory" and "process theory." Figure 1 illustrates
this distinction applied to the problem of ex-
plaining strategic change.
Whereas variance theories provide explana-
tions for phenomena in terms of relationships
among dependent and independent variables
(e.g., more of X and more of Y produce more of Z),
process theories provide explanations in terms
of the sequence of events leading to an outcome
(e.g., do A and then B to get C). Temporal order-
ing and probabilistic interaction between enti-
ties are important here (Mohr, 1982). Understand-
ing patterns in events is thus key to developing
"process" theory.
"Events," however, are quite different entities
from the "variables" that dominate methodology
seminars and that most of us are more used to
manipulating. The analysis of process data, there-
iore, requires a means of conceptualizing events
and of detecting patterns among them. As sug-
gested by Van de Ven and Poole (1995), these pat-
terns may take a variety of different forms, but the
most common pattern found in the literature is the
linear sequence of "phases" that occur over time
to produce a given result (e.g., Burgelman, 1983;
Rogers, 1983). However, the passage from raw data
to synthetic models, whether expressed in terms of
phases or otherwise, is far from simple. Abbott
(1990) and Van de Ven (1992) have presented a
number of techniques for analyzing event se-
quences when events are sharply defined in terms
of the units of analysis they refer to and their
location in time. These provide one strategy for
analysis that I will discuss later. However, raw
process data usually do not come quite so neatly
sliced and packaged.
Data on Multiple Units and Levels of Analysis
with Ambiguous Boundaries
Any researcher who has collected qualitative
process data in organizations has seen how dif-
ficult it is to isolate units of analysis in an un-
ambiguous way. For example, what should or
should not be included in the definition of a
decision-making process? Can researchers al-
ways distinguish (as Eisenhardt, 1989a, did in
her research) between the decision to make a
strategic change and the decision about what
strategy should be adopted?
More complex phenomena, such as strategy
formation or learning, are even harder to isolate.
Process phenomena have a fluid character that
spreads out over both space and time (Pettigrew,
1992). In addition, one of the main reasons for
taking a qualitative process approach is pre-
cisely to take into account the context (Petti-
grew, 1992; Yin, 1994). This leads, inevitably, to
the consideration of multiple levels of analysis
that are sometimes difficult to separate from one
anothermade up of a continuum, rather than a
hierarchy or a clear classification. This further
complicates the sensemaking process.
' Note that the collection oi process data and the design oi
process studies also pose a number of challenges. However,
these are not the main tocus of this article. While recogniz-
ing that, to some extent, data collection and design issues
constrain future options, my concern here is how to deal with
the data, once collected. For more on design and data col-
lection issues, see Eisenhardt (1989b), Leonard-Barton (1990),
Pettigrew (1990), and Yin (1994).
Data of Variable Temporal Embeddedness
When collecting process data, the researcher
attempts to document as completely as possible
the sequence of events pertinent to the pro-
cesses studied. However, unless the process is
highly circumscribed, certain phenomena will
tend to be absent from a systematic list of or-
1999
Langley
693
FIGURE 1
Two Approaches to Explaining Strategic Change"
Variance theory
Explaining strategic change with
a variance model
Attributes of
environment
leadership
decision
processes
performance
Extent of
strategic
change
Process theory
Explaining strategic change with
a process model
Strategy 1
to
events
activities
choices
Strategy 2
tn
" After Mohr (1982).
dered incidents. For example, there are often
gradual background trends that modulate the
progress oi specific events. Also, part of what
interests us may be going on in people's heads
and leave no concrete trace of the exact moment
of its passing.
Despite the apparent temporal precision indi-
cated by the word "event," there are also clearly
different levels of events: an event may include
a bad year, a merger, a decision, a meeting, a
conversation, or a handshake. Finally, particu-
larly in macrolevel studies oi such processes as
strategy making, innovation, and decision mak-
ing, the researcher is often obliged to combine
historical data collected through the analysis of
documents and retrospective interviews with
current data collected in real time. While the
iirst type oi data is sparse and synthetic, focus-
ing on memorable moments and broad trends,
the second is richer and iiner grained. And,
while the first type misses certain useful nu-
ances and details, the second type may require
a certain distancing before it is possible to sep-
arate out what is really significant from what
will be treated as merely noise (Leonard-Barton,
1990). These phenomena are often unavoidable,
but they all render analysis and interpretation
more difficult.
Data That Are Eclectic
In his work Mohr (1982) insists strongly on the
necessity of keeping variance and process the-
ories separate. This requirement is extremely
difficult to satisfy. Perhaps for aesthetic rea-
sons, Mohr (1982) seems to want to artiiicially
separate variables and events, although, in
practice, phenomena oi different kinds are inter-
twined. I would argue that the insistence on
exclusion of variables from process research un-
necessarily limits the variety of theories con-
structed. It may be important to understand the
effect of events on the state oi an entity (a vari-
able) or to identify the effect oi a contextual
variable on the evolution oi events. Process re-
search may also deal with the evolution oi rela-
tionships between people or with the cognitions
and emotions oi individuals as they interpret
and react to events (Isabella, 1990; Peterson,
1998). Thus, although temporal phenomena re-
main one of their distinguishing features, pro-
cess data are not composed only of descriptions
of discrete events. They also incorporate a vari-
ety oi other types oi qualitative and quantitative
iniormation. Again, this makes analysis and in-
terpretation more complex.
A process database, thus, poses considerable
challenges. The sheer volume oi words to be
organized and understood can create a sense of
drowning in a shapeless mass of information
(Pettigrew's, 1990, much-quoted "death by data
asphyxiation"). The complexity and ambiguity
of the data make it difficult to know where to
start. Also, although offering greater potential
for new discovery, the open-ended inductive ap-
proach that most researchers use in process re-
694
Academy of Management Review October
search tends to lead to postponement of the mo-
ment of decision between what is relevant and
what is not, sometimes aggravating these diffi-
culties (Miles & Huberman, 1994).
The complexity of process data is, of course, a
reflection of the complexity of the organizational
phenomena we are attempting to understand.
More and more researchers have been question-
ing simple process models that assume neat
linear progressions of well-defined phases lead-
ing to well-defined outcomes (Schwenk, 1985;
Van de Ven, 1992). Although the linear phase
model still has attractions, process representa-
tions now often show divergences from the main
route, recycling between phases and parallel
tracks (Mintzberg, Raisinghani, & Theoret, 1976;
Nutt, 1984; Schroeder, Van de Ven, Scudder, &
Polley, 1989). Researchers are also increasingly
recognizing that the presence of multilayered
and changing contexts, multidirectional causal-
ities, and feedback loops often disturb steady
progression toward equilibrium. Several schol-
ars have, in fact, argued that chaos theory or
complexity theory may offer the potential for
better understanding organizational processes
(e.g., Stacey, 1995; Thietart & Forgues, 1995).
Thus, it is clear that we need better ways to
model process phenomena. However, research
that concludes simply that "everything is com-
plex" or that "simple normative models do not
work" is limited in its appeal. As Van de Ven
(1992) notes, process theorization needs to go
beyond surface description to penetrate the
logic behind observed temporal progressions
whether simple or complex. I find it difficult to
share the enthusiasm oi some writers for the
application of complexity theory to organization-
al phenomena, precisely because the specific
explanatory mechanisms behind its application
are often not specified. The general but banal
insight that organizational processes involve
opposing forces, nonlinear relationships, and
feedback loops needs fleshing out. One interest-
ing point raised by these theorists, however, is
that the interaction of a relatively small number
of simple deterministic elements may generate
complexity, if they take into account such phe-
nomena. With this, there is hope that relatively
parsimonious theoretical formulations may be
able to make sense of the complexity observed
in process data.
And this is where the central challenge lies:
moving from a shapeless data spaghetti toward
some kind of theoretical understanding that
does not betray the richness, dynamism, and
complexity of the data but that is understand-
able and potentially useful to others. Through-
out the remainder of this article, I examine
seven generic strategies for achieving this. Fol-
lowing Weick (1979), I term these sensemaking
strategies. The word "sensemaking" is used for
two reasons. First, it implies the possibility that
a variety of "senses" or theoretical understand-
ings may legitimately emerge from the same
data. In fact, I argue that different strategies
tend to produce different forms of theory that are
neither intrinsically better nor worse but may
have different strengths and weaknesses. Sec-
ond, it Implies that the closing of the gap be-
tween data and theory can begin at either or
both ends (data or theory) and may often iterate
between them (Orton, 1997). Rigid adherence to
purely deductive or purely inductive strategies
seems unnecessarily stultifying. Indeed, Tsou-
kas (1989) goes further, arguing that while the
data themselves can yield empirical regulari-
ties, abstract conceptualization is required to
imagine the "generative mechanisms" that are
driving them. For him, understanding comes
from a combination of the two.
STRATEGIES FOR SENSEMAKING
The seven strategies for sensemaking de-
scribed in this section were derived from an
in-depth reading of the organization studies and
methods literature and from my own research
experience. I see the strategies as generic ap-
proaches, rather than step-by-step recipes or
techniques. They are not necessarily exhaus-
tive, and they can be used in combination. Each
approach tends to overcome the overwhelming
nature of boundaryless, dynamic, and multi-
level process data by fixing attention on some
anchor point that helps in structuring the mate-
rial but that also determines which elements
will receive less attention. It is because of this
that the strategy used can have an important
impact on the nature of the emerging theory.
Thorngate's (1976) and Weick's (1979) catego-
ries of accuracy, generality, and simp/ici(/ are
used here to consider the theoretical forms
likely to be developed using different strategies.
Some strategies tend to stick closely to the orig-
inal data, whereas others permit greater ab-
straction. Close data fitting reflects what Weick
1999
Langley
695
(1979) calls "accuracy." However, accuracy may
act against generality'another desirable qual-
ity related to the potential range of situations to
which the theory may be applicable. Finally,
simplicity concerns the number of elements
and/or relationships in a theory. It affects the
theory's aesthetic qualities. Simple theories
with good explanatory power may actually be
preferred to complex ones that explain a littie
more: as Daft (1983) suggests, good research is
more like a poem than a novel.
In describing each strategy, I draw on exem-
plars in the organizational literature that ap-
pear to represent the best of what can be
achieved with each approach. In my analysis I
also look at the relative data needs of each
approach both in terms of depth (process detail)
and breadth (number of cases), as well as the
extent to which each strategy deals with each of
the process dat a characteristics mentioned
above. Finally, I show how each strategy tends
to favor different types of process understanding
("senses"). Some strategies seem best adapted
to the detection of patterns in processes,
whereas others penetrate their driving mecha-
nisms. Some are more oriented toward the
meaning of process for the people involved,
whereas some are more concerned with predic-
tion. The discussion is summarized in Table 1.
Narrative Strategy
This strategy involves construction of a de-
tailed story from the raw data. In the area of
strategic management, the classic example of
this style is Chandler's (1964) history of the evo-
lution of American enterprise. The same style
also dominates the work of strategy researchers
who adopt a "contextualist" perspective, nota-
bly Andrew Pettigrew and members of the Cen-
tre for Corporate Strategy and Change (Petti-
grew, 1985, 1990; Pettigrew & Whipp, 1991), but
also others working in this tradition (Dawson,
1994; Johnson, 1987). Descriptive narratives (or
"realistic tales") are also the traditional tool of
ethnographers {Van Maanen, 1988), and they fre-
quently play a key role in studies of cultural
change (Bartunek, 1984).
In fact, almost all process research involves
recourse to this strategy at some point. However,
the narrative can serve different purposes, de-
pending on the objectives of the researcher. For
many it is merely a preliminary step aimed at
preparing a chronology for subsequent analy-
sis essentially, a data organization device that
can also serve as a validation tool (e.g., Eisen-
hardt, 1989b). For "contextualists" it plays a
more substantial role, incorporating an analyti-
cal element;
Our analytical chronologies reach towards the-
ory presentation but are prepared to get on top of
the data, to clarify sequences across levels of
analysis, suggest causal linkages between lev-
els, and establish early analytical themes (Petti-
grew, 1990: 280).
Finally, for others who adopt a constructivist or
naturalistic perspective (Dyer & Wilkins, 1991;
Guba & Lincoln, 1994), the narrative can be the
main product of the research. The aim is to
achieve understanding of organizational phe-
nomena^not through formal propositions but
by providing "vicarious experience" of a real
setting in all its richness and complexity (Lin-
coln & Guba, 1985: 359). For the proponents of
this approach, it is the contextual detail in the
narrative ("thick description") that will allow the
reader to judge the transferability of the ideas to
other situations. Indeed, good research of this
type will often produce a sense of "dejd vu"
among experienced readers. The theorist who
adopts this philosophy tries to avoid excessive
data reduction and to present as completely as
possible the different viewpoints on the process
studied.
This strategy avoids commitment to any spe-
cific anchor point, although because of the struc-
ture of narrative, time tends to play an important
role. Also, because of its focus on contextual
detail, this approach works best for one or a few
cases. Ideally, the variety and richness of the
incidents described and of the linkages between
them should convey a high degree of authentic-
ity that cannot be achieved economically with
large samples (Golden-Biddle & Locke, 1993). In
the hands of an accomplished writer, this sense-
making strategy has the great advantage of re-
producing in all its subtlety the ambiguity that
exists in the situations observed. It avoids the
necessity of clear definitions when boundaries
are not clear, and it easily accommodates vari-
able temporal embeddedness and eclectic data.
The philosophy behind this type of analysis is
well expressed by Van Maanen: "To be determi-
nate, we must be indeterminate" (1995: 139).
In Weick's (1979) terms, accuracy is therefore
expected to be high. However, those who adopt
696
D)
U3
o
a
.a
a
S
< u
n
a
a
en
"o
a
o
li
0 0)
Pi
o
d
0
Q
C/3
.t: o
Academy of Management Review
O.S
w
d g;^
3 2
-eg c
_Dij d
October
ui
dl o
.^:aE
*- d 2.-5
i^-^>:' .2-6
"5,'oiS b't)
P s b d 0
a
o -
r u b
8i i ^
2: Q9S
d -G "
2
0>
o >
Q) ?
a
0)
a
d 0=^ 3 c
'r^-Bsi ^ ^ ^
d tt
D
0) D I" o
a UJ3 u
01
"d f^
U '
CO
C
worn
S a d
CO O U
O
13
O m
li
O d ETJ 3 Cj - m
Xi Si > a) <u X d o U
O
-d
01
;C d 0)
Otaa,
d d
3
o
0)
Si!
g ^ a
s i -^ B
d S 0) a
" P'E E
o
o o a>5
'-' 0) S
M H 3
Sas
3 S c
a ^ 2
> c
U --; H QJ O
"uT3 Zi:
W 1^ , -'
0 ) |
- 3 o .C 75 OJ u)
wU a.s a o
2
1" . '5 O M
S-r^ P-2 >
a
.2 o
Cl
C -7*
^ w t: -d C
3 < S ' " " " "
Q
o d
' 9
5 U
TJ
g -B
d -^
E3 ui
'3 'y g ^ ^
TJ 3 _^ 0)
m a > <U
ID d >- 0 "1
C o O. tn 3
O
u
d S
ii
co
U
u u y
Ul
en
o B
D) C
o d)
in
to d)
tn
cn
0)
u
S 25 3
So. " "'
^^ d s 2 a^
fl) r i " d ID CU^* d)
i S S o g.
o o "2 g^
- d .S in
t3U.E
[fl T-

ID
i
1999
Langley
697
a more traditional research perspective may be
dissatisfied because this approach does not, on
its own, lead to either simple or general theory.
Without denying the usefulness of the narrative
approach for communicating the richness of the
context to readers, most of us expect research to
offer more explicit theoretical interpretations.
When relying on this strategy alone, one may
too easily end up with an idiosyncratic story of
marginal interest to those who were not in-
volved and a rather thin conceptual contribu-
tion. Appealing process research needs to push
beyond authenticity to make readers feel that
they learned something of wider value (Golden-
Biddle & Locke, 1993).
The instrinsic interest of the phenomenon
studied can sometimes offer this valuefor ex-
ample, narratives that dig under the surface of
dramatic events can be very effective, as in
Vaughan's (1996) analysis of the Challenger di-
saster. But, beyond this, the most interesting and
compelling narratives (including Vaughan's)
are not so purely descriptive. They know where
they are going. Like Chandler's (1964) stories of
the invention of the M-form organization, they
have embedded "plots" and "themes" that serve
as sensemaking devices (Woiceshyn, 1997) and
that ultimately become more explicit theories
(e.g., "structure follows strategy"). However, si-
multaneously telling the complete story while
setting the plot is a tall order. Other strategies
can help out here.
Quantification Strategy
At the opposite end of the spectrum from the
narrative strategy is a form of process analysis
that has been most effectively promoted by An-
drew Van de Ven and colleagues of the Minne-
sota Innovation Research Project (Van de Ven &
Poole, 1990). In this approach researchers start
with in-depth process data and then systemati-
cally list and code qualitative incidents accord-
ing to predetermined characteristics, gradually
reducing the complex mass of information to a
set of quantitative time series that can be ana-
lyzed using statistical methods.
For example, in their innovation project. Van
de Ven's team first collected detailed real-time
data and then identified five characteristics or
tracks that could be used to analyze each iden-
tifiable incident (people, ideas, transactions,
context, and results). The incidents and their
corresponding tracks were transformed into a
series of binary codes associated with a specific
date, forming a 0-1 matrix that the authors call a
"bit-map." Each incident corresponded to a line
in the matrix. One column was reserved to indi-
cate positive outcomes (0 or 1) associated with
an incident and another was reserved for nega-
tive outcomes. A third column was used to indi-
cate whether or not there was a change in the
people involved in the innovation, and so on
(Van de Ven & Poole, 1990). Once the coding was
complete, the researchers worked with the bi-
nary data matrix and used statistical methods to
search for patterns and test theoretical explana-
tions. For example, Garud and Van de Ven (1992)
and Van de Ven and Polley (1992) tested a dy-
namic theory of learning during innovation. The
same data were used to examine whether the
sequences reflected random, chaotic, or periodic
processes as the innovation evolved (Cheng &
Van de Ven, 1996).
Similar approaches have been used by Smith,
Grimm, and Gannon (1992) to analyze competi-
tive interactions among airlines and by Ro-
maneili and Tushman (1994) to examine patterns
of change in the microcomputer industry. Unlike
Van de Ven's team, however, these researchers
began with mainly documentary databases con-
sisting of newspaper articles, IOK reports, and
the like.
The advantage of the quantification approach
lies in the systematization of process analysis.
Assuming that the original data are complete
and that the coding of incidents is reliable, de-
scriptive patterns in the sequence of events can
be verified systematically and explicit process
theories can be tested rigorously. Note, however,
that despite the conversion of the data to quan-
titative form, the types of statistical analysis
appropriate to process theorizing are somewhat
different from those used in most variance re-
search.
For example, many process theories are
founded on the idea that there are fundamental
similarities in the patterns of event sequences
across cases. However, traditional techniques
(regression, ANOVA, and so forth) are designed
to explain differences (variance)^not to show
similarities. The sequence methods proposed by
Abbott (1990) include the use of multidimen-
sional scaling, to identify "typical sequences"
across different cases, and of optimal matching
algorithms (like those used in DNA testing), to
698
Academy of Management Review
October
estimate the proximity between sequences and
to develop event-sequence typologies. Sabher-
wal and Robey (1993) adopted this method, for
example, to develop a taxonomy of information-
system implementation processes based on the
detailed coding of 53 event chronologies.
More commonly, however, quantitative re-
searchers examining process phenomena have
used techniques such as event-history analysis,
lagged regression, log-linear models, and dy-
namic simulation. Rather than testing for the
similarity of whole sequence patterns across
cases, these methods are appropriate for exam-
ining the dynamic relationships between events
within a single case or a population. Monge
(1990) provides a detailed overview of the theo-
retical forms that can be considered in this way
and the appropriate statistical techniques for
each. The approach seems particularly useful
for the verification of dynamic theories that in-
clude causal feedback loops. All this supposes,
however, that comparable incidents within the
same case or across similar cases are suffi-
ciently large in number to create enough de-
grees of freedom for the statistical analysis. For
this, "incidents" must be defined to be very ge-
neric in form, with little contextual richness and
variability remaining attached to them.
In contrast with the narrative approach, this
strategy leads more easily to parsimonious the-
oretical conceptualizations (i.e., simplicity). Be-
cause of the generic character of coded events
and the mathematical formulation of the models
testedoften supported by deductive reason-
ingthe theorization is also likely to have
greater potential generality (although replica-
tion is needed to verify this). Yet, to achieve this
result, the approach drastically simplifies the
original data, setting aside certain dimensions
and replacing the ambiguous, rich, and specific
context by precise, thin, and general indicators.
There is little room for variable temporal em-
beddedness or ill-defined boundaries in the
emerging models. Accuracy, thus, is not neces-
sarily the strong suit of such theories, even
though the gap between the dat a and the
emerging model may appear to be more defen-
sible than in certain other strategies, because it
can be assessed and justified rationally by rea-
sonable interrater reliabilities (that take us from
the data to its coded representation) and good
R-squareds (that get us from the coded represen-
tation to the final model).
In fact, although I have taken part in similar
exercises myself and know why we do these
things, there is a certain irony in the idea that
researchers who give themselves the trouble of
collecting rich qualitative data in real organiza-
tions are so uncomfortable with this richness
that they immediately rush to transform it,
through another extremely demanding process,
into a much thinner data set that can be man-
aged in traditional ways. The quantification
strategy will be much more convincing if it is
used in combination with other approaches that
allow contextualization of the abstract data,
adding nuances of interpretation and confirm-
ing the mechanics of the mathematical model
with direct evidence. The articles by Garud and
Van de Ven (1992) and Van de Ven and PoUey
(1992) on learning during innovation are inter-
esting from this viewpoint. However, those who
rely solely on the quantification strategy may
lose critical elements of process understanding
in abstractions so general that the results ob-
tained may be clear but fairly banal.
The two strategies just described lie at the two
ends of a continuum that opposes empirical ac-
curacy and theoretical parsimony. I now present
some more middle-of-the-road approaches.
Alternate Templates Strategy
In this sensemaking strategy the analyst pro-
poses several alternative interpretations of the
same events based on different but internally
coherent sets of a priori theoretical premises. He
or she then assesses the extent to which each
theoretical template contributes to a satisfac-
tory explanation.
The strategy was popularized by Allison
(1971), in his classic study of the decisions made
during the Cuban Missile Crisis. The three ex-
planatory templates used by Allison were a "ra-
tional actor model," in which the United States
and the Soviet Union were viewed as unified
rational actors selecting alternatives to achieve
national objectives; an "organizational process
model," in which decision making was seen as
driven by organizational routines (Cyert &
March, 1963); and a "political model," in which
individuals involved in the crisis were viewed
as pursuing their own personal interests within
a distributed power structure. Allison (1971) pro-
duced three retellings of the story, each drawing
on a different model. He concluded that the last
1999 Langiey
699
two seemed superior to the first, allowing the
explanation of certain events that otherwise ap-
peared mysterious.
This strategy has been used often since then
for the study of decision processes (e.g., Pinfield,
1986; Steinbruner, 1974), perhaps partly because
of Allison's example, but also perhaps because
of the difficulty of developing a unique model of
decision making that simultaneously captures
all of its di mensi ons {Langley, Mintzberg,
Pitcher, Posada, & Saint-Macary, 1995). The
strategy also has attracted adherents among in-
formation systems researchers concerned with
i mplementati on processes (e.g.. Lee, 1989;
Markus, 1983). In strategic management the
work of CoUis (1991) on globalization also re-
flects this approach.
Because this strategy draws theory from out-
side the data, it is essentially deductive. In some
applications predictions of the competing theo-
ries are formally "tested" in a hypothetico-
deductive fashion, with specific predictions be-
ing refuted to reject weaker theories (e.g.,
Markus, 1983). This is similar to Yin's (1994) idea
of "pattern-matching." Often, though, the differ-
ent interpretations are less like true "tests" of
theory and more like alternate complementary
readings that focus on different variables and
levels of analysis and reveal different types of
dynamics. Many broad process theories, such as
political models (Allison, 1971), organizing the-
ory (Weick, 1979), or structuration theory (Gid-
dens, 1984), are alternative modes of sensemak-
ing that are not easily refutable because their
constructs seem adaptable (e.g., in political mod-
els it is usually possible to find personal goals
that make observed action rational). However, a
confrontation among different interpretations can
reveal the contributions and gaps in each.
Although some researchers have counseled
against using single case studies in process re-
search because of the lack of material for repli-
cation and comparison (Eisenhardt, 1989b; Petti-
grew, 1990), this strategy provides a powerful
means of deriving insight from a single rich
case because the different theoretical interpre-
tations provide the base for comparison needed
(Lee, 1989: Yin, 1994). Each interpretation strat-
egy may also force the researcher to collect dif-
ferent types of data, with the more finely
grained theories actually becoming very de-
manding, further revealing the relative contri-
bution of each perspective (Allison, 1971),
Overall, this strategy combines both richness
and theoretical parsimony (simplicity) by de-
composing the problem. Qualitative nuances
are represented through the alternative expla-
nations, and theoretical clarity is maintained by
keeping the different theoretical lenses sepa-
rate (at least in most applications of this ap-
proach). Between them, then, different theoreti-
cal perspecti ves provide overall accuracy,
although each one is inaccurate on its own. Gen-
erality in this approach comes from the use of
deductive theories that have broad application.
However, despite its advantages, the use of
this strategy often leaves the researcher and the
reader puzzled as to how the various theoretical
perspectives can be combined. Almost inevita-
bly, each explanation taken alone is relevant
but insufficient. Yet, any theory that attempted
to integrate the different perspectives would
tend to become unwieldy and aesthetically un-
satisfying. As Allison indicates at the end of his
book:
The developed sciences have little hesitation
about partial models.... The aspiring sciences
tend to demand general theory. In satisfying this
demand, they often force generalization at the
expense of understanding. Refining partial para-
digms, and specifying the classes of actions for
which they are relevant, may be a more fruitful
path to limited theory and propositions than the
route of instant generalization {171; 275).
Grounded Theory Strategy
As noted by several meta-analysts of qualita-
tive research (Larson & Lowendahi, 1995; Locke,
1996), Glaser and Strauss's (1967) The Discovery
of Grounded Theory is one of the most-cited
methods texts in qualitative research articles.
And, yet, there is sometimes limited evidence in
these articles of the systematic theory-building
methods proposed by the authors and subse-
quently refined by Glaser {1978) and Strauss and
Corbin (1990; see Locke, 1996). For many,
"grounded theory" is basically a generic syn-
onym for any kind of inductive theorizing. This is
perhaps not surprising, for the language itself
expresses this idea. However, in this article I use
grounded theory strategy to refer to the more spe-
cific methods described by the original authors.
When followed "by the book," the grounded
theory approach as described most recently by
Strauss and Corbin (1990) incorporates a series
of highly structured steps. It involves the sys-
700
Academy of Management fleview October
tematic comparison of small units of data (inci-
dents) and the gradual construction of a system
of "categories" that describe the phenomena be-
ing observed. The categories may have several
"subcategories," and associated "dimensions"
and "properties," which are gradually elabo-
rated and refined as specific incidents are ex-
amined, systematically coded, and compared.
As the categories are developed, the researcher
deliberately seeks out data that will enable ver-
ification of the properties of emerging category
systems. The analysis should eventually result
in the identification of a small number of "core
categories," which serve to tightly integrate all the
theoretical concepts into a coherent whole firmly
rooted ("grounded") in the original evidence.
At first sight, process data offer many oppor-
tunities for grounded theorizing. Indeed, Glaser
(1978) and Strauss and Corbin (1990) insist on the
necessity of incorporating processes into any
grounded theory study. They note that processes
are categories that have two or more identifi-
able "stages" (Glaser, 1978) and that the most
useful core categories are often expressed as
gerunds (i.e., in process terms). Several
grounded theory process studies in the literature
are faithful to this portrait (e.g., Sutton's, 1987,
model of organizational death as "disbanding"
and "reconnecting" and Gioia & Chittipeddi's,
1991, representation of the initiation of strategic
change as "sensemaking" and "sensegiving").
However, I would argue that the strategy
"makes more sense" for some types of process
data than for others. Generally, it demands a
fairly large number of comparable incidents
that are all richly described. Thus, while one
setting may be sufficient, there should at least
be several distinct processes that can be com-
pared in depth (e.g., as in Burgelman's, 1983,
internal venturing study). Alternatively, the
level of analysis can be dropped away from the
overall site to a more microlevel to explore the
interpretations and emotions of different indi-
viduals or groups living through the same pro-
cesses (e.g., Isabella, 1990; Sutton, 1987). It is
here that the strategy often appears at its most
powerful. However, when the objective is to un-
derstand more macroscopic processes that occur
one at a time over long periods (like strategic
change in a large organization), the processes'
broad sweep seems to fit less well with the
microanalysis of the textbook grounded theory
approach. The data themselves may not have
the density required to support it, and the micro-
focus risks losing the broad pattern of the forest
for the descriptive detail of the trees.
In summary, used alone, this is a strategy that
tends to stay very close to the original data and
is therefore high in accuracy. It starts with em-
pirical details expressed in interview tran-
scripts and field notes and attempts to build a
theoretical structure "bottom up" from this base.
Yet, because of the specialized language, the
logic of the method, and the deliberately hierar-
chical structure of category systems, theories
developed in this way are at the same time very
dense (low to moderate in simplicity) but often
seem to have a similar flavor and general struc-
ture (compare, for example, the exemplary
grounded theory studies of very different phe-
nomena by Gioia, Thomas, Clark, & Chittipeddi,
1994, and Browning, Beyer, & Shetler, 1995). As
its proponents note, firm grounding in the raw
data can also sometimes make it difficult to
move from a "substantive" theory of a specific
phenomenon to more general formal theory
(Glaser & Strauss, 1967).
Visual Mapping Strategy
Process data analysis may involve the manip-
ulation of words (e.g., narrative strategies or
grounded theory), of numbers (quantification), or
of matrix and graphical forms (Miles & Huber-
man, 1994). Such forms have several advantages
over narrative approaches, according to Miles
and Huberman (1994). They allow the presenta-
tion of large quantities of information in rela-
tively little space, and they can be useful tools
for the development and verification of theoret-
ical ideas. Visual graphical representations are
particularly attractive for the analysis of pro-
cess data because they allow the simultaneous
representation of a large number of dimensions,
and they can easily be used to show precedence,
parallel processes, and the passage of time.
For example. Figure 2 is taken from a study of
the process of adoption of new technology in
small manufacturing firms (Langley & Truax,
1994). The drawing presents an event chronology
coded in multiple ways. The form of the boxes
indicates whether the event described repre-
sents a decision (round-cornered rectangles), an
activity (sharp-cornered rectangles), or an event
outside the control of the firm (ovals). The loca-
tion of each box in one of the six horizontal
1999
Langley
701
0
d
u
o
C
702
Academy of Management Review
October
bands indicates the issue domain with which
the event is associated. Certain boxes cross sev-
eral bands, indicating the integrative character
of that event. The arrows leading from each box
to the central band indicate the effect of this
event on the technology adoption process (posi-
tive effect [ + ], negative effect [-] , precipitating
effect [ + + 1, reorienting effect [0]). The thickness
of the horizontal lines linking the boxes indi-
cates the degree of continuity among linked
events. Finally, the horizontal time scale allows
representation of event ordering and parallel
tracks over time and provides a rough indication
of their temporal duration. The drawing is obvi-
ously a summary of what took place in the case,
but the link between it and the qualitative da-
tabase is maintained through the use of short
descriptions of each element in its correspond-
ing box.
This type of drawing obviously is not a "theo-
ry" but an intermediary step between the raw
data and a more abstract conceptualization. To
move toward a more general understanding,
one might, in further analysis, compare several
such representations to look for common se-
quences of events and common progressions in
sources of influence {Langley & Truax, 1994).
One could also proceed to developing more ab-
stract coding to generate local "causal maps"
that would constitute the beginnings of a mid-
dle-range theoretical explanation (as described
by Miles & Huberman, 1994). Finally, one might
compare and integrate several such causal
maps to elaborate a more general theory. Lyles
and Reger (1993) apply an approach like this in
their study of the evolution of managerial influ-
ence in a joint venture over a period of 30 years.
Different forms of process mapping have long
been used by organizations to plan, understand,
and correct their own work processes (in sys-
tems analysis, quality improvement, business
process reengineering, and so forth). Organiza-
tional researchers could perhaps learn from this
example. Meyer (1991) notes how flowcharts of
capital budgeting processes proved useful in
making sense of disparate accounts and in com-
municating with informants to collect further
data. These charts also became the raw mate-
rial for the development of both a more compre-
hensive process model (Meyer, 1984) and a vari-
ance model of innovation adoption (Meyer &
Goes, 1988).
Process mapping also has been a favored
technique for decision researchers, constituting
the foundation of Mintzberg et al.'s (1976) classic
article and of Nutt's (1984) work on decision pro-
cess typologies. In his advocacy of a "grammat-
ical" approach to process analysis, Pentland
(1995) also implicitly suggests the usefulness of
process mapping. He discusses the need to de-
tect the underlying rules (grammar) driving the
ordering of different types of moves (specific
operations) or syntactic elements (groups of
moves serving similar functions) within a repet-
itive process.
Approaches like those described require
many observations of similar processes. This
indicates that the mapping strategy may be
most fruitful as a theory development tool for the
analysis of multiple holistic or embedded cases.
Of course, as a simple presentational method, it
has broader application.
Process mapping allows the preservation of
some dimensions of data ambiguity but ex-
cludes others. For example. Figure 2 does not
force artificial clarity on the identification of the
main unit of analysis, and it conceptualizes
technology adoption as an evolutionary phe-
nomenon that interacts in a dynamic way with
other issues important to the firm {Langley &
Truax, 1994). Yet this representation gives no
room to such factors as power, conflict, and emo-
tion. In part, the range of possibilities for map-
ping depends on the researcher's objectives and
creativity (e.g., see Newman & Robey, 1992, for
ways of representing encounters between actors
on process diagrams). However, graphical forms
may be biased toward the representation of cer-
tain types of information and against others.
Relations of temporal precedence, authority,
and influence between objects or individuals
are quite easily represented. Continuous traces
could even be used to represent the levels of key
variables (e.g., financial performance). However,
emotions and cognitions are less easy to ex-
press in this way, being more difficuit to tempo-
rally pin down.
The graphical strategy, thus, offers a means of
data reduction and synthesis that is less radical
and more flexible than that used in the quanti-
fication strategy (moderate accuracy). However,
unless supported by other methods, the conclu-
sions derived from it can have a rather mechan-
ical quality, dealing more with the surface struc-
ture of activity sequences t han with the
1999
Langley
703
underlying forces driving them. For this reason
its conceptualizations will tend to be of moder-
ate generality. The approach can produce useful
typologies of process components, but attempts
to reach beyond this to deeper generalizations
are often less parsimonious because of the large
number of variations possible and the difficulty
of predicting which ones will occur and why
(moderate simplicity).
Temporal Bracketing Strategy
The time scale along the bottom of Figure 2 is
decomposed into three successive "periods."
These periods do not have any particular theo-
retical significance. They are not "phases" in the
sense of a predictable sequential process but,
simply, a way of structuring the description of
events. If those labels were chosen, it was be-
cause there is a certain continuity in the activi-
ties within each period and there are certain
discontinuities at its frontiers (Langley & Truax,
1994). Many temporal processes can be decom-
posed in this way, at least partly, without pre-
suming any progressive developmental logic.
However, beyond its descriptive utility, this type
of temporal decomposition also offers interest-
ing opportunities for structuring process analy-
sis and sensemaking. Specifically, it permits the
constitution of comparative units of analysis for
the exploration and replication of theoretical
ideas. This can be especially useful if there is
some likelihood that feedback mechanisms, mu-
tual shaping, or multidirectional causality will
be incorporated into the theorization. We see
this strategy at work in the contributions of sev-
eral process researchers (e.g.. Barley, 1986;
Denis, Langley, & Cazale, 1996; Doz, 1996; Dutton
& Dukerich, 1991).
We call this strategy "bracketing" in reference
to Gidden's (1984) structuration theorya clas-
sic example of a perspective involving mutual
shaping. At the heart of structuration theory is
the idea that the actions of individuals are con-
strained by structures (including formal and in-
formal rules and norms) but that these actions
may also serve to reconstitute those structures
over time. Because mutual influences are diffi-
cult to capture simultaneously, it is easier to
analyze the two processes in a sequential fash-
ion by temporarily "bracketing" one of them
(Giddens, 1984). The decomposition of data into
successive adjacent periods enables the explicit
examination of how actions of one period lead to
changes in the context that will affect action in
subsequent periods.
In his study of structuring in two radiology
departments following the acquisition of CT
scanners. Bailey (1986) consciously adopts this
approach. He observed how the initial institu-
tional context of the departments studied af-
fected the pattern of interactions between radi-
ologists and technicians and then how these
patterns evolved and led to changes in the in-
stitutional context. This, in turn, became the
point of departure for another phase of structur-
ing. His detailed process data were analyzed
and compared across successive periods sepa-
rated by discontinuities in the institutional con-
text, producing a compelling account of the role
of technology in the evolution of structure.
In their study of strategic change under am-
biguous authority, Denis et al. (1998) al so
adopted this strategy in order to better under-
stand the mutual linkages between the tactics
used by the members of a management team
and the evolution of leadership roles within it.
These were traced over five periods, separated
by discontinuities in team membership. Denis et
al. (1996) observed that certain types of tactics
favor the creation of a unified team with the
power to successfully promote change. How-
ever, once the team is created, the temptation
and the possibility of using more coercive tac-
tics lead to the fragmentation of the team, even
as the change is solidified. Alternating dynam-
ics, thus, are observed in successive periods.
Again, the "periods" become units of analysis
for replicating the emerging theory. Doz (1996)
used a similar approach to trace patterns in the
cycles of learning and reevaluation associated
with strategic alliance development.
With this strategy, a shapeless mass of pro-
cess data is transformed into a series of more
discrete but connected blocks. Within phases,
the data are used to describe the processes as
fairly stable or linearly evolving patterns. Evi-
dence is also drawn together to examine how
the context affects these processes, and what
the consequences of these processes are on the
future context and other relevant variables of
interest. Discontinuities lead to replication of
the analysis in a new phase.
This sensemaking strategy fits well with a
nonlinear dynamic perspective on organization-
al processes, and it can quite easily handle
704
Academy o/ Management fieview October
eclectic data that include events, variables, in-
terpretations, interactions, feelings, and so on.
Because of its internal replication possibilities,
one or a few cases may be sufficient to produce
useful insights (all studies cited in this section
are based on one or two cases only). However,
temporal decomposition can create certain dis-
tortions. For example, there is no a priori guar-
antee that discontinuities will naturally syn-
chronize themselves to produce unequivocal
periods. Overall, then, accuracy is likely to be
moderate to high, depending on the appropri-
ateness of the temporal decomposition and the
robustness of the analysis to different periodiza-
tions. Conceptualizations emerging from the
process are unlikely to be very simple, although
they stand a better chance of dealing with fun-
damental process drivers than those produced
by certain other strategies. Assuming that they
have been derived inductively, they will also have
moderate generality, until tested on more data.
Synthetic Strategy
One recurring criticism of process theorizing
is that despite its capacity to produce enriched
understanding and explanation, it often lacks
predictive power (Rumelt, 1997; Van de Ven,
1992). With the sensemaking strategy that we
have termed synthetic, the researcher takes the
process as a whole as a unit of analysis and
attempts to construct global measures from the
detailed event dat a to describe it. The re-
searcher then uses these measures to compare
different processes and to identify regularities
that will form the basis of a predictive theory
relating holistic process characteristics to other
variables (e.g., outcomes and contexts). The
work of Eisenhardt and colleagues (1989a,b;
Eisenhardt & Bourgeois, 1988) on decision mak-
ing in high-velocity environments is the obvious
exemplar for this strategy. Others include Meyer
and Goes' (1988) work on technology adoption
and Bryson and Bromiley's (1993) work on new
product planning and implementation.
When this strategy is used, the original pro-
cess data are transformed from stories com-
posed of "events" to "variables" that synthesize
their critical components. The emerging models,
thus, are "variance theories"not "process the-
ories"in Mohr's (1982) words. For example,
Eisenhardt (1989a) compared eight cases of de-
cision making and developed a causal model to
explain decision speed as a function of five pro-
cess constructs: (1) the type of information used,
(2) the pattern of alternatives examined, (3) the
advice process adopted, (4) the conflict-resolu-
tion approach used, and (5) the degree of inte-
gration of decisions. In this case the constructs
were developed through inductive exploration
and coding of case narratives, as well as certain
quantitative indicators. Their linkages to the de-
pendent variable (speed) were verified through
tabular displays and investigation of the mech-
anisms by which the effects were obtained,
drawing on both the data and existing theory.
One interesting aspect of these process vari-
ables is that they are not necessarily the stan-
dard process variables (e.g., use of planning and
rationality) that might have been chosen had
the researcher simply developed a question-
naire-based study (e.g., cf. Dean & Sharfman,
1996). Rather, they incorporate more subtle nu-
ances, including aspects of timing (e.g., simul-
taneity of alternatives, deadline versus leader-
driven conflict resolution, and real-time versus
delayed information), that could only be de-
tected as important through close contact with
real processes. In this way detailed process data
can lead to more meaningful and potentially
more powerful explanatory variables for nomo-
thetic research (see also Nutt, 1993, for another
example).
However, this is not process theory; the com-
plexities of the probabilistic interaction of
events, parallel and alternate tracks, patterns of
mutual shaping over time, and evolving perfor-
mance have been compressed into positions on
a small number of scales that can now be re-
lated to a single overall "success" assessment.
In fact, it is clear that despite major investments
in the collection of process data, synthetic vari-
ance models exert an inexorable attraction. As
soon as researchers become interested in under-
standing the reasons for different outcomes,
they tend to be drawn into formulating the prob-
lem in terms of explanatory variables (see, for
example, even the major longitudinal studies on
strategic change processes by Hinings & Green-
wood, 1988, and by Pettigrew & Whipp, 1991).
Such an approach can generate important con-
clusionsoften richer and more credible ones
than could be obtained from thinner cross-
sectional data, because the causal links are
more explicitly traceable. Nevertheless, as with
the quantification strategy, care must be taken
1999 Langley 705
not to ditch the detailed temporal understand-
ing obtained for its shadow. This means draw-
ing (as the researchers cited do) on the entire
qualitative database to show how and why the
variables identified lead to the consequences
predicted.
In terms of data requirements, the synthetic
strategy requires a clear definition of the bound-
aries of the processes studied and a level of
abstraction high enough to permit the compari-
son of several cases {accuracy will therefore be
moderate at best). It also requires sufficient
cases to allow satisfactory comparison and con-
clusion drawing (e.g., Eisenhardt, 1989a, used 8
cases and Nutt, 1984, included 78 cases). This
tends to correspond to a thinner level of detail in
process tracing for each case than for other
strategies. When the number of cases is moder-
ate, this adds to the need to show strong ground-
ing of the explanatory mechanisms within the
data itself and to connect these to other litera-
ture, in order to make the relationships identi-
fied credible and to enhance external validity
{Eisenhardt, 1989b). Like the quantification strat-
egy, this strategy has the advantage of produc-
ing relatively simple theoretical formulations
that are also moderately generai because they
have been conceived to make sense of data from
a number of cases.
Qualitative/Quantitative Data versus Process/
Variance Analysis: Other Approaches
The description of seven sensemaking strate-
gies for process data is now complete. In my
analysis I assumed that the problem was to con-
struct theory from qualitative "process data" col-
lected in close contact with real contexts. Thus,
I emphasized the large area of overlap between
qualitative data and process theorizing. How-
ever, it is important to note that qualitative data
do not necessarily demand process analysis
and that process theory can be built from quan-
titative data.
The first point should be obvious. Qualitative
data can be used for many purposes that have
little to do with how events are sequenced over
time. For example, they can be used to develop
rich descriptions of meanings, behaviors, and
feelings evoked by workplace issues at one
point in time {e.g., Pratt & Rafaeli, 1997, on or-
ganizational dress). They can be used to under-
stand individuals' mental maps of the elements
in their world (Huff, 1990) and so on. Some but
not all of the seven sensemaking strategies I
have described can be used in these non-
process situations (e.g., grounded theory in the
first example and visual mapping in the sec-
ond), but my discussion does not pretend to deal
with these rather different applications.
The second issue is whether process theory can
be derived from purely quantitative data, such as
archival time series or panel questionnaires. It
can, of course, using similar statistical techniques
to those mentioned under the quantification strat-
egy, but this is not a perspective that I have ex-
plored or favored here. Quantitative time series
constitute rather coarse-grained outcroppings of
events and variables over time: they skim the sur-
face of processes rather than plunge into them
directly. Nevertheless, such methods are rapidly
penetrating the strategy field and contributing
significantly to a more dynamic understanding of
strategic evolution (e.g., Barnett & Burgelman,
1996). As such, they are complementary to the ap-
proaches discussed here. Indeed, as Ven de Ven
and colleagues' work has shown, there is much to
be gained from collecting both quantitative time
series and qualitative stories in the same process
research effort (Brewer & Hunter, 1989).
It is also worth mentioning another quantita-
tive approach to developing process theory that,
at first sight, appears to be even more distant
from real processes because its "data" are en-
tirely artificial. This is computer simulation, of
which the most influential examples are Cyert
and March's (1963) behavioral theory of the firm
and Cohen, March, and Olsen's (1972) "garbage
can" model of organizational choice (but see
also Sastry's, 1997, formalization of the punctu-
ated equilibrium model of change and Lant &
Mezias's, 1992, work on organizational learning).
As Weick (1979) himself noted, these models are
high in simplicity and generality but generally
weak in terms of accuracy. Real data may have
been collected at some time and may have in-
spired the ideas behind the model. But, in most
cases, the model is not linked to specific empir-
ical observations.^
Yet, such models have several advantages.
First, provided their basic assumptions are intu-
^ Note, however, that empirical calibration can be at-
tempted and may add to the credibility of such models. See,
lor example, Hall's (1976) study of the decline and failure of
the Saturday Evening Post.
706
Academy of Management Review October
itively reasonable, the models can be used as
sites for risk-free experimentation. Second, be-
cause they are not constrained by real measure-
ments, they can deal with constructs that would
be unobservable in reality (e.g., managerial en-
ergy in the garbage can model). Third, they may
allow the detection and correction of inconsis-
tencies in existing theoretical frameworks (e.g.,
Sastry, 1997). But, above all, these models are
powerful when they show how a few simple and
plausible mechanisms can generate complex
behavior patterns that we all recognize. Para-
doxically, just like the narrative strategy, which
is, on the contrary, very deeply rooted in real-life
processes, the strength of a simulation comes
from its capacity to create a feeling of "dejd vu,"
making sense of previously impenetrable expe-
rience. (It is surely no accident that the garbage
can model has been popular among academics!)
DISCUSSION AND CONCLUSION
There are constant calls in the scholarly litera-
ture for more in-depth process research that will
enable us to understand organizational phenom-
ena at more than a superficial level. And yet,
when we actually go out and get the data required
to achieve this, we find that the deep understand-
ing we sought does not magically leap out at us.
Process data are notoriously challenging. In this
article I have examined seven generic strategies
for making sense of them (see Table 1). In the
following discussion I review the strategies from a
number of different angles. First, I compare their
positioning according to Weick's (1979) criteria.
Second, I situate them within a more general
framework and examine the ways in which they
can be combined. Third, I consider the roles of
induction, deduction, and inspiration in the theory
development process.
Accuracy, Simplicity, and Generality
All seven strategies have unique strengths. But
all have weaknesses. As Thorngate (1976) and
Weick (1979) indicate, any research strategy de-
mands tradeoffs among accuracy, generality, and
simplicity. In particular, accuracy tends to conflict
with both simplicity and generality, while, at least
in my analysis, simplicity and generality tend to
be more compatible (see Table 1). The approxi-
mate positioning of each strategy with respect to
the dimensions is illustrated in Table 2. For the
sake of contrast, I have also included the computer
simulation approach in this diagram.
This portrait of the different strategies does
not provide an answer to the question "Which
strategy is best?" However, it maps the terrain
and shows that "good" process research can
take a variety of routes. Some strategies favor
accuracy, remaining more deeply rooted in the
raw data (narrative strategy and grounded the-
ory). Others are more reductionist, although they
allow the development and testing of parsimo-
nious theoretical generalizations (quantifica-
tion, synthetic strategy, and simulation).
Overall (see Figure 2), the different strategies
tend to run the length of an "efficient frontier"
that represents the range of tradeoffs between
accuracy and simplicity. From a pragmatic
standpoint, the two extremes (simulation and
narrative) are riskier because of the sacrifices
they require on key dimensions. In addition, the
alternate templates approach is a special case
not positioned within the table. While each in-
dividual template provides simplicity but lim-
ited accuracy, between them multiple templates
can increase overall accuracy while maintain-
ing simplicity and generality, as long as the
temptation to integrate divergent perspectives
is avoided. The idea that multiple templates can
produce better understandings may also be gen-
eralized to the use of multiple strategies, again
provided the combinations are complementary
and provided simplicity is not compromised in
the attempt to achieve integration.
TABLE 2
Sensemaking Strategies and Accuracy,
Simplicity, and Generality'^
Strategy Accuracy Simplicity Generality
High Low Low
Narrative
Grounded theory
Temporal bracketing
Visual mapping
Synthetic strategy
Quantification
Computer simulation
Low High High
" The orderings in this table are approximate; there are
variations among specific applications. In particular, while
accuracy and simplicity are almost always in opposition to
one another, the generality of emerging theories will depend
on other factors, such as the degree and scope of replication
and the source of the conceptual ideas.
1999 Langley
707
Variations, Permutations, and Combinations
One way to explore the potential for combina-
tions of sensemaking strategies and to organize
them within a common framework is to consider
them as falling into three sequentially linked
groups that I term grounding strategies, organiz-
ing strategies, and rep/icafing strategies.
The grounded theory and alternate templates
strategies can be considered as grounding strate-
gies because they suggest two different sources
for concepts that can be used within the context of
other strategies. Grounded theory involves data-
driven categories, whereas the alternate tem-
plates strategy involves (heory-driven constructs.
The two strategies, thus, represent the purist
forms of inductive and deductive reasoning, re-
spectively. Both forms of grounding can contribute
to the construction of narratives and visual maps,
and both strategies can be used as tools in the
comparative analysis of cases (the synthetic strat-
egy) or the comparative analysis of phases (tem-
poral bracketing). Alternate templates also can be
used to test quantitative process models.
The narrative and visual mapping strategies
can be viewed as organizing strategies because,
as described earlier, they are ways of descrip-
tively representing process data in a systematic
organized form. As such, they often, although not
always, constitute the initial rather than final
steps in the sensemaking process. Both narratives
and visual maps can serve as intermediary data-
bases for the identification of phases (temporal
bracketing), events (quantification), and con-
structs (synthetic strategy) and for the formulation
of hypotheses and propositions. Since narratives
are closer to the raw data than visual maps, they
may also precede their development.
Finally, the remaining three strategies (tempo-
ral bracketing, quantification, and synthesis) can
be considered replicating strategies since they
represent different ways of decomposing the data
for the replication of theoretical propositions (by
phase, by event, and by case). These strategies
can draw on almost any or all of the others. Quan-
tified event data may also be aggregated for use
in synthetic case comparisons (Eisenhardt, 1989b)
or for comparative analysis of phases (e.g., see
Barley, 1986). Conversely, phase-by-phase infor-
mation (Garud & Van de Ven, 1992) or case-by-case
information (Cheng & Van de Ven, 1996) may be
incorporated into quantitative models.
This categorization imposes some order on
what may so far have seemed a rather eclectic
typology of sensemaking approachesbut not, I
hope, too much order. The last thing I wish to
advocate is a homogenous recipe for theorizing
from process data that leaves no room for loose
ends or creativity. The choice of strategies is more
than just a case of desired levels of accuracy,
simplicity, and generality and more than just a
case of picking logically linked combinations; it is
also a question of taste, of research objectives, of
the kind of data available, and of imagination.
Moreover, variety contributes to richness. The
seven sensemaking strategies produce seven dif-
ferent senses. Method and theory are closely in-
tertwined. As I have noted, some strategies tend to
focus on the meaning of processes for individu-
alsthat is, the way they are experienced
(grounded theory and narrative strategy). Others
are better equipped for tracing overall temporal
patterns (visual mapping, quantification, and
grounded theory). Some more easily reveal driv-
ing process motors or mechanisms (alternate tem-
plates, temporal bracketing, and quantification),
and some are more useful for predicfion (synthetic
strategy). There are also undoubtedly other strat-
egies with which I am less familiar (e.g., literary or
critical approaches) that could make different
kinds of sense again.
Induction,. Deduction, and Inspiration
Beyond the individual strategies and their bi-
ases, my reading of the literature and my own
experience reinforce the belief that there is a
step in the connecting of data and theory that
escapes any deliberate sensemaking strategy a
researcher might decide to apply. As Mintzberg
(1989) insists, analysis does not produce synthe-
sis. Theory development is a synthetic process.
Whatever strategy is used, there will always be
an uncodifiable step that relies on the insight
and imagination of the researcher (Weick, 1989).
Wolcott (1994) distinguishes clearly between the
two processes of analysis and interpretation. In-
terpretation corresponds to this creative ele-
ment. Clearly, this does not absolve the re-
searcher from the need to test his or her
interpretations systematically. Analysis, thus, is
important to stimulate and verify theoretical
ideas. But, unfortunately for those who seek the
magic bullet, it cannot produce them alone.
708
Academy of Management Review October
This also means that persistent calls for cod-
ification of qualitative methods (Larson & Lo-
wendahl, 1995; Orton, 1997) can reach a point of
diminishing returns, because we just do not
know and cannot tell where that critical insight
came from. Nobody asks quantitative research-
ers to explain how they thought up their concep-
tual frameworks (although Sutton, 1997, sug-
gests that many may have been inspired by
"closet" qualitative research!).
Another way to think about this is that theory
building involves three processes: (1) induction
{data-driven generalization), {2) deduction (theory-
driven hypothesis testing), and {3) inspiration
{driven by creativity and insight). "Inspiration"
may be stimulated by empirical research, by read-
ing, by thought experiments, and by mental exer-
cises {Weick, 1979, 1989), but its roots are often
untraceable. It draws indiscriminately on formal
data, experience, a priori theory, and common
sense. It works when it succeeds in creating new
and plausible connections between all of these
that can be made explicit as theoretical products,
exposed to the scrutiny of others, and verified.
In closing, this brings me to the question of the
nature of the linkage between data and theory in
process research. In theorizing from process data,
we shoiald not have to be shy about mobilizing
both inductive (data-driven) approaches and de-
ductive (theory-driven) approaches iteratively or
simultaneously as inspiration guides us. There is
room not only for building on existing constructs
to develop new relationships (Eisenhardt, 1989b)
but for designing process research that selectively
takes concepts from different theoretical traditions
and adapts them to the data at hand, or takes
ideas from the data and attaches them to theoret-
ical perspectives, enriching those theories as it
goes along. There is also room for developing new
strategies for understanding processes that mix
and match those I have presented here or that take
a new tack entirely. Sensemaking is the objective.
Let us make sense whatever way we can.
REFERENCES
Abbott, A. 1990. A primer on sequence methods. Qrganiza-
tion Science, 1: 375-392.
Allison, G. T. 1971. Essence of decision. Boston: Little, Biov/n.
Barley, S. R. 1986. Technology as an occasion for structuring:
Evidence from observations of CT scanners and the so-
cial order oi radiology departments. Administrative Sci-
ence OuoffeWy, 31: 78-108.
Bamett, W. P., & Burgelman, R. A. 1996. Evolutionary perspec-
tives on strategy. Strategic Management Journal,
17(Special Issue): 5-20.
Bartunek, J. 1984. Changing interpretive schemes and organ-
izational restructuring: The example of a religious order.
Administrative Science Quarteriy, 29: 355-372.
Bower, J. 1997. Strategy process research. Symposium pre-
sented at the annual meeting oi the Academy of Man-
agement, Boston.
Brewer, 1., & Hunter, A. 1989. Multimethod reseatch. Newbury
Park. CA: Sage. |
Browning, L. D., Beyer, J. M., & Shetler, J. C. 1995. Building
cooperation in a competitive industry: SEMATECH and
the semiconductor industry. Academy of Management
/ouma/, 38: 113-151.
Bryson, ]. M., & Bromiley, P. 1993. Critical iadors affecting the
planning and implementation of major products. Strate-
gic Management/oumai, 14: 319-338.
Burgelman, R. A. 1983. A process model oi corporate ventur-
ing. Administrative Science Quarterly. 28: 223-244.
Chandler, A. D. 1964. Strategy and structure. Cambridge, MA:
MIT Press.
Cheng. Y.-T., & Van de Ven, A. H. 1996. Learning the innova-
tion journey: Order out oi chaos. Oi^antzation Science.
7: 593-614. |
Cohen, M. D., March, J. G., & Olsen, J. P. 1972. A garbage can
model of organizational choice. Administrative Science
Quarterly. 17: 1-25.
Cohen, M. D., 8E Sproull, L. S. 1991. Editors' introduction:
Special issue on organizational teaming: Papers in
honor oi James G. March. Organization Science, 2(1).
Collis, D. J. 1991. A resource-based analysis of global com-
petition: The case of the bearings industry. Strategic
Management/ournaJ, I2(Special Issue): 49-68.
Cyert, R. M., & March. J. G. 1963. A behavioral theory oi the
iirm. Englewood Cliffs, NJ: Prentice-Hall.
Dait, R. L. 1993. Learning the craft oi organizational research.
Academy of Management Review. 8: 539-546.
Dawson, P. 1994. Organizational change: A processuaJ ap-
proach. London: Chapman,
Dean. J. W.. & Shariman, M. P. 19%. Does decision process
matter? A study of strategic decision-making efiective-
ness. Academy of Management Journal, 39: 368-396.
Denis, J.-L., Langley, A., & Cazale, L. 1996. Leadership and
strategic change under ambiguity. Organization Stud-
ies, 17: 673-699.
Doz, Y. L, 1996. The evolution of cooperation in strategic
alliances: Initial conditions or teaming processes? Stra-
tegic Management/ournai, 17{Special Issue): 55-83.
Dutton, J. E., & Dukerich, J. M. 1991. Keeping an eye on the
mirror: Image and identity in organizational adaptation.
Academy of Management Journal. 34: 517-554,
Dyer, W. G., & Wilkins. A. 1991. Better stories, not better con-
structs, to generate better theory: A rejoinder to Eisenhardt.
Academy of Management Review, 16: 613-619.
Eisenhardt, K. M. 1989a. Making fast strategic decisions in
1999 LangJey
709
high velocity environments. Academy o/ Management
Journal. 31: 543-576.
Eisenhardt, K. M. 1989b. Building theories trom case study
research. Academy oi Management Review, 14: 532-550.
Eisenhardt, K. M., & Bourgeois. L. J. 1988. Politics of strategic
decision making in high-velocity environments. Acad-
emy of Management Journal. 31: 737-770.
Grorud, R., & Van de Ven, A. H. 1992. An empirical evaluation
oi the internal corporate venturing process. Strategic
Management Journal, 13(SpeciaI Issue): 93-109.
Giddens. A. 1984. The constitution of society. Berkeley, CA:
University of California Press.
Gioia, D. A., & Chittipeddi. K. 1991. Sensemaking and sense-
giving in strategic change initiation. Strategic Manage-
ment Journal. 12: 433-448.
Gioia, D. A., Thomas, I. B., Clark, S. M., 8i Chittipeddi. K. 1994.
Symbolism and strategic change in academia: The dy-
namics of sensemaking and influence. Organization
Science, 5: 363-383.
Glaser, B. G. 1978. Theoretical sensitivity. Mill Valley, CA:
Sociology Press.
Glaser. B. G.. & Strauss, A. L. 1967. The discovery of grounded
theory. Chicago: Aldine.
Golden-Biddie, K., & Locke, K. 1993. Appealing work:
An investigation of how ethnographic texts convince.
Organization Science, 4: 595-816.
Guba, E., & Lincoln, Y. S. 1994. Handbook of qualitative
research. Newbury Park, CA: Sage.
Hall, R. 1976. A system pathology of an organization: The rise
and fall oi the old Saturday Evening Post. Administra-
tive Science Quarferiy, 21: 185-211.
Hinings, C. R., & Greenwood, R. 1988. The dynamics of stra-
tegic change. Oxford, England: Blackwell.
Huff, A. 1990. Mapping strategic thought. Chichester. Eng-
land: Wiley.
Ilinitch, A. Y.. D'Aveni. R. A.. & Lewin, A. Y. 1996. New organ-
izational forms and strategies for managing in hyper-
competitive markets. Organization Science, 7: 211-221.
Isabella, L. A. 1990. Evolving interpretations as change un-
folds: How managers construe key organizational
events. Academy of Management Journal, 33: 7-41.
Johnson. G. 1987. Strategic change and the management
process. Oxford, England: Blackweli.
Langley, A. 1997, L'6tude des procesaus strat6giques: D6iis
conceptuels et analytiques. Management International.
2(1): 37-50.
Langley, A., Mintzberg. H.. Pitcher, P., Posada, E., & Saint-
Macary, J. 1985, Opening up decision making: The view
from the black stool. Organization Science, 6: 260-279.
Langley, A.. & Truax. J. 1994. A process study oi new technol-
ogy adoption in smaller manufacturing firms. Journal of
Management Studies. 31: 619-652.
Lant, T. K., & Mezias. S. J. 1992. An organizational learning
model of convergence and reorientation. Organization
Science, 3: 47-71.
Larson, R., & Ldwendahl, B. 1997. The qualitative side of
management research: A meta-analysis of espoused and
used case-study methodologies. Paper presented at the
annual meeting ol the Academy of Management. Boston.
Lee, A. S. 1989. A scientific methodology lor MIS case studies.
MIS Quarterly, 13: 33-50.
Leonard-Barton. D. 1990. A dual methodology for case studies:
Synergistic use oi a longitudinal single site with repli-
cated multiple sites. Organization Science. 1: 248-266.
Lincoln, Y. S., 8E Guba, E. 1985. Naturalistic enquiry. Beverly
Hills. CA: Sage,
Locke, K. 1996. Rewriting "The Discovery oi Grounded Theory"
after 25 years. Journal of Management Inquiry, 5: 239-245.
Lyles, M. A., & Beger, R. K. 1993. Managing ior autonomy in
joint ventures: A longitudinal study of upward influence.
Journal of Management Studies. 30: 383-404.
Markus. L. 1983. Power, politics and MIS implementation.
Communications of the ACM. 26: 430-444.
Meyer, A. D. 1984. Mingling decision making metaphors.
Academy of Management Review, 9: 6-17.
Meyer. A. D. 1991. Visual data in organizational research.
Organization Science, 2: 218-236.
Meyer, A. D., & Goes, J. B. 1988. Organizational assimilation
of innovations: A multilevel contextual analysis. Acad-
emy of Management Journal. 31: 897-923.
Miles, M. B., 8E Huberman, A. M. 1994. Qua/itative data anai-
ysis. Newbury Park, CA: Sage.
Mintzberg, H. 1979. An emerging strategy of "direct" re-
search. Administrative Science QuarteWy, 24: 580-589.
Mintzberg, H. 1989. Mintzberg on management. New York;
Free Press.
Mintzberg, H.. Raisinghani, D.. & Theoret, A. 1976. The struc-
ture oi unstructured decision processes. Administrative
Science Quarterly. 21: 246-275.
Mohr. L. B. 1982. Expiaining organizationai behavior. San
Francisco: Jossey-Bass.
Monge, P. R. 1990. Theoretical and analytical issues in study-
ing organizational processes. Organization Science. 1:
406-430.
Newman, M., & Robey. D. 1992. A social process model of
user-analyst relationships. MIS Quarterly, 16: 249-259.
Nutt, P. C. 1984. Types ol organizational decision processes.
Administrative Science Quarterly. 29: 414-450,
Nutt, P, C. 1993. The formulation processes and tactics used
in organizational decision making. Organization Sci-
ence, 4: 226-251.
Orton. J, D. 1997. From inductive to iterative grounded theory:
Zipping the gap between process theory and process
data. Scandinavian/ourna/o/Management, 13: 419-438.
Pentland, B. T. 1995. Grammatical models of organizational
processes. Organization Science, 6: 541-556.
Peterson, M. F. 1998. Embedded organizational events: The
units of process in organizational science. Organization
Science, 9: 16-33.
Pettigrew, A. M. 1985. The awakening giant. Oxford, England:
Blackweli.
710
Academy oi Management Review October
Pettigrew, A. M. 1990. Longitudinal lield research on change:
Theory and practice. Organization Science, 1: 267-292.
Pettigrew, A. M. 1992. The character and significance of
strategy process research. Strategic Management Jour-
nal, 13{Special Issue): 5-16.
Pettigrew, A. M,, & Whipp, R. 1991. Managing change tor
competitive success, Oxford, England: Blackwell.
Piniield, L. T. 1986. A field evaluation of perspectives on
organizational decision making. Administrative Science
Quarterly. 31: 365-388.
Pratt, M., & Ralaeli, A. 1997. Organizational dress as a sym-
bol of multilayered social identities. Academy of Man-
agement Journal, 40: 862-898.
Rogers, E. M. 1983. The dUtusion oi innovations. New York:
Free Press.
Romanelli, E., & Tushman, M. 1994. Organizational transfor-
mation as punctuated equilibrium: An empirical test.
Academy o/Management/ournai/ 37: 1141-1166.
Rumelt, R. 1997. Comments during symposium on process
research, presented at the annual meeting of the Acad-
emy of Management, Boston.
Sabherwal, R., & Robey, D. 1993. An empirical taxonomy of
implementation processes based on sequences of
events in information system development. Organiza-
tion Science, 4: 548-576.
Sastry, M. A. 1997. Problems and paradoxes in a model of
punctuated organizational change. Administrative Sci-
ence Quarterly. 42: 237-275.
Schroeder. R. A.. Van de Ven, A. H., Scudder, G. D., & PoUey, D.
1989. The development of innovation ideas. In A. H. Ven
de Ven, H. L. Angle, & M. S. Poole (Eds.), Research on the
management oi innovation: 107-134. New York: Ball-
inger/Harper & Row.
Schwenk, C. 1985. The use of participant recollection in the
modeling ol organizational processes. Academy ot Man-
agement Review. 10: 496-503,
Smith, K. G.. Grimm, C. M., & Gannon, M. J. 1992. Dynamics ot
competitive strategy. Newbury Park, OA: Sage.
Stacey, R. D. 1995. The science of complexity: An alternative
perspective for strategic change processes. Sfrategic
Management/oumaJ, 16: 477-495.
Steinbruner, J. D. 1974. A cybernetic process of decision.
Princeton. NJ: Princeton University Press.
Strauss, A., & Corbin, J. 1990. Basics ot qualitative research,
Newbury Park, CA: Sage.
Sutton, R. 1987. The process of organizational death. Admin-
istrative Science Quarterly, 32: 542-569.
Sutton, R. 1997. The virtues of "closet" qualitative research.
Organization Science, 8: 97-106.
Thietart, R. A., & Forgues. B. 1995. Chaos theory and organi-
zation. Organization Science, 6: 19-31.
Thorngate, W. 1976. Possible limits on a science of social
behavior. In ]. H. Strickland, F, E. Aboud, & K. J. Gergen
(Eds.), Sociai psychology in transition: 121-139. New
York: Plenum.
Tsoukas, H. 1989. The validity of idiographic research expla-
nations. Academy ot Management Review, 14: 551-561.
Van de Ven, A. H. 1992. Suggestions for studying strategy
process: A research note. Strategic Management Jour-
nal. 13(Special Issue): 169-188.
Van de Ven, A. H., & Huber, G. P. 1990. Longitudinal lield
research methods for studying processes oJ organiza-
tional change. Organization Science, 1: 213-219.
Van de Ven, A. H.. & Polley, D. 1992. Learning while innovat-
ing. Organizafion Science, 3: 92-116. |
Van de Ven, A. H., & Poole, M. S. 1990. Methods for studying
innovation development in the Minnesota Innovation
Research Program. Organization Science, 1: 313-335.
Van de Ven, A. H., 8E Poole. M. S. 1995. Explaining develop-
ment and change in organizations. Academy ot Manage-
ment Review. 20: 510-540.
Van Maanen, J. 1988. Tales oi the Held, Chicago: University of
Chicago Press.
Van Maanen, J. 1995. Style as theory. Organization Science,
6: 133-143.
Vaughan, D. 1996. The Challenger launch decision: Risky
technology, culture and deviance at NASA, Chicago:
University of Chicago Press.
Weick, K. 1979. The social psychology oi organizing. Reading,
MA: Addison-Wesley.
Weick, K. 1989. Theory construction as disciplined imagina-
tion. Academy ot Management Review. 14: 516-531.
Woiceshyn, J. 1997. Literary analysis as a metaphor in pro-
cessual research: A story of technological change. Scan-
dinavian Journal ot Management, 13: 457-471.
Wolcott, H. F. 1994. Trans/ormlng qualitative data. Newbury
Park, CA: Sage.
Yin, R. K. 1994. Case study research, Newbury Park, CA: Sage.
Ann Langley is professor of strategy at the Ecole des Sciences de la Gestion, Univer-
site du Quebec a Montreal. She received her Ph.D. from the Ecole des Hautes tudes
Commerciales in Montreal. Her research interests include strategic change processes,
decision making, innovation, and interorganizational relationships, with particular
emphasis on health care organizations.