Sage Publications, Inc., Johnson Graduate School of Management, Cornell University Administrative Science Quarterly

Alternative Forms of Fit in Contingency Theory
Author(s): Robert Drazin and Andrew H. Van de Ven

Source: Administrative Science Quarterly, Vol. 30, No. 4 (Dec., 1985), pp. 514-539
Published by: Sage Publications, Inc. on behalf of the Johnson Graduate School of
Management, Cornell University
Stable URL: https://www.jstor.org/stable/2392695
Accessed: 06-02-2019 23:07 UTC
REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2392695?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Sage Publications, Inc., Johnson Graduate School of Management, Cornell University

are collaborating with JSTOR to digitize, preserve and extend access to Administrative
Science Quarterly
This content downloaded from 152.118.24.10 on Wed, 06 Feb 2019 23:07:40 UTC
All use subject to https://about.jstor.org/terms
Alternative Forms of Fit This paper examines the selection, interaction, and sys-
in Contingency Theory tems approaches to fit in structural contingency theory.
These are empirically examined as related to a task-
contingency theory of work-unit design in 629 employment
Robert Drazin security units in California and Wisconsin. Evidence was
and found to support the selection and systems approaches in
Andrew H. Van de Ven these data but not the interaction approach. The general-
izability of these findings is discussed in terms of using
alternative approaches to fit to explain context-structure-
performance relationships in contingency theory.
Structural contingency theory has dominated the study of

organizational design and performance during the past twenty
years. However, despite its favorable status, contingency
theory is continually being called into question because of its
apparent inability to resolve persistent theoretical and empiri-
cal problems. The recent commentaries on contingency theory
(Schoonhoven, 1981; Mohr, 1982; Tosi and Slocum, 1984; Fry
and Schellenberg, 1984; Van de Ven and Drazin, 1985) all
suggest that basic changes in theory and methodology are
needed. Ironically, management researchers have recently
proposed theories that are, at their core, even more complex
and unresolved systems of contingency propositions; for ex-
ample, the McKinsey 7-S framework (Pascale and Athos,
1981), Theory Z (Ouchi, 1981), the eight characteristics that fit
together in excellent companies (Peters and Waterman, 1982),
and expansions of Leavitt's diamond model for designing
innovative organizations and for organizing the stages of
growth of new ventures (Galbraith, 1982).
All these models share in common an underlying premise that
context and structure must somehow fit together if the orga-
nization is to perform well. Despite the critical role that this
concept of fit plays, few studies have carefully examined its
implications (Schoonhoven, 1981; Fry and Schellenberg, 1984;
Van de Ven and Drazin, 1985). Instead, it appears that our
concepts of fit are drawn from a general and often implicit pool
of domain assumptions and methodological conventions.
As Dubin (1976) stated, every theory is a contingency theory,

because for a proposition or "law of interaction" to hold,
assumptions must be made about starting premises, bound-
aries, and system states. Boundary conditions specify the
ranges over which a relationship is expected to hold, and
system states specify the temporal period and other condi-
tions under which the relationships hypothesized by a theory
are expected to occur.
A contingency theory differs from other theories in the specific

(? 1985 by Cornell University. form of the propositions. The distinction between congruent
0001-8392/85/3004-051 4/$1 .00. and contingent propositions made by Fry and Schellenberg
(1984) clarifies this difference. In a congruent proposition a
simple unconditional association is hypothesized to exist
Support for this research was provided in among variables in the model; for example, the greater the
part by the Wisconsin Job Service Division
task uncertainty, the more complex the structure. A contingent
of the Department of Industry, Labor and
Human Relations, the California Employ- proposition is more complex, because a conditional association
ment Department, and by the Program on of two or more independent variables with a dependent out-
Organizational Effectiveness of the Office
of Naval Research under the contract num-
come is hypothesized and directly subjected to an empirical
ber NOOO1 4-S4-K-001 6. test; for example, task uncertainty interacts with structural
514/Administrative Science Quarterly, 30 (1985): 514-539
complexity to affect performance. Central to a structural con-
tingency theory is the proposition that the structure and pro-
cess of an organization must fit its context (characteristics of
the organization's culture, environment, technology, size, or
task), if it is to survive or be effective. In Dubin's terms, the
"law of interaction" in a contingency theory is that organiza-
tional performance depends on the fit between organization
context and structure and process - given that normal
assumptions hold about the premises, boundaries, and system
states derived from the theory.
The key concept in a contingent proposition is fit, and the

definition of fit that is adopted is central to the development of
the theory, to the collection of data, and to the statistical
analysis of the proposition. Van de Ven and Drazin (1985)
indicated that in the development of contingency theory, at
least three different conceptual approaches to fit have
emerged - the selection, interaction, and systems
approaches (Table 1) - and each significantly alters the es-
sential meaning of a contingency theory and the expected em-
Table 1
Interpretation of Fit in the Selection, Interaction, and Systems Approaches to Structural Contingency Theory
Views, definitions,
and test methods Selection Interaction Systems
Initial Views
Definition Assumption: Fit is Bivariate interaction: Consistency analysis:

assumed premise under- Fit is the interaction Fit is the internal
lying a congruence of pairs of organiza- consistency of multiple
between context and tional context- contingencies and
structure. structure factors; it multiple structural
affects performance. characteristics; it
affects performance
characteristics.
Test methods Correlation or Context-structure Deviations from

regression interaction terms in ideal-type designs
coefficients of MANOVA or regression should result in
context (e.g., equations on lower performance.
environment, performance The source of the
technology, or size) should be significant. deviation (in
on structure (e.g., consistency)
configuration, originates in
formalization, conflicting
centralization) should contingencies.
be significant.
Current-Future Views
Definition Macro selection: Fit at Residual analysis: Fit Equifinality: Fit is

micro-level is by is conformance to a a feasible set of
natural or managerial linear relationship of equally effective,
selection at macro- context and design. Low internally consistent
level of organizations. performance is the patterns of organiza-
result of deviations tional context and
from this relationship. structure.
Test methods Variables subject to Residuals of context- Relationship among

universal switching structure relations latent context,
rules should be highly regressed on structure, and
correlated with context. performance performance constructs
Particularistic should be significant. should be significant,
variables should show while observed
lower correlations. manifest charac-
teristics need not be.
515/ASQ, December 1985
Forms of Fit
pirical results. These three different approaches to fit are

presented and then examined empirically in this paper. We
believe that they clarify much of the confusion in the literature
on structural contingency theory and provide alternative direc-
tions to further the development of contingency theories in
general.
SELECTION, INTERACTION, AND SYSTEMS

APPROACHES TO FIT
Selection Approach
Many early structural contingency theories were in fact con-

gruence theories because they simply hypothesized that orga-
nizational context (whether environment, technology, or size)
was related to structure (centralization, formalization, complex-
ity) without examining whether this context-structure rela-
tionship affected performance. For example, using a variety of
technology dimensions, many researchers have hypothesized
and found strong relationships between technology and struc-
ture (1) at the organization level (Perrow, 1967; Hage and
Aiken, 1969; Freeman, 1973; Dewar and Hage, 1978), (2) at
the work-unit level (Hall, 1962; Fullan, 1970; Van de Ven and
Delbecq, 1974; Tushman, 1977; Marsh and Mannari, 1981),
and (3) across levels of organizational analyses (Comstock and
Scott, 1977; Nightingale and Toulouse, 1977; Pierce, Dunham,
and Blackburn, 1979; Fry, 1982). Many of these studies had an
implicit feedback logic underlying the reason for the associa-
tion between context and structure. However, none of these
studies discussed or presented evidence on the effect of the
congruence between technology and structure on organiza-
tional perfomance.
It is unclear whether to conclude that this research did not

address contingency theory or to conclude that contingency
theory operated as an untested assumption underlying this
organization context-structure research. For example, most
technology researchers in the 1960s and 1970s used a con-
tingency theory logic similar to that of Woodward (1965) and
Perrow (1967), but they simply did not test for the link with
performance - either because they did not collect measures
of performance or because they were not interested in this key
part of the theory.
Recently, however, natural selection and managerial selection

perspectives have surfaced and provide some justification for
viewing fit as a basic assumption underlying congruence prop-
ositions between organizational context and structure and
process. In the natural selection argument, fit is the result of an
evolutionary process of adaptation that ensures that only the
best-performing organizations survive (Hannan and Freeman,
1977; Aldrich, 1979; -Comstock and Schroger, 1979; McKel-
vey, 1982). An equilibrium between environment and organiza-
tion is assumed to exist, at least over long periods of time, and
only context-structure relationships need to be examined to
assess fit (Fennell, 1980), because an identity, or isomorphic
relationship, between context and structure, is presumed to
exist for the surviving organizations (DiMaggio and
Powell, 1983).
The managerial selection argument extends this approach and

takes into account macro- and micro-levels of organization
design (Van de Ven and Drazin, 1985). Most organizations (or
subunits) are constrained in choosing or adopting the structural
patterns that reflect their particular circumstances. No matter
what level of organization is examined, there is usually a more
macro-level that imposes, at least in part, uniform practices
and prescriptions on the more micro-level (DiMaggio and
Powell, 1983). For example, government legislative bodies
regulate industries, industries have codes that constrain
businesses, and organizations have policies that impose
uniformities on departments, divisions, and work units.
Macro-rules tend to be imposed on micro-units in two ways:

(1) uniformly without regard for the contexts of subunits to
which they apply, and (2) situationally, through a set of switch-
ing rules that take contextual factors into consideration.
Switching rules are more interesting to contingency theorists,
because they affect the fit between structure and context the
most. They function as guidelines or prescriptions for manag-
ers, enabling them to adjust structure to new contingencies.
Organizations limit the discretion of subunits by adopting a set

of switching rules, or contingency programs, that prescribe
different designs for different types of subunits. For example,
routine production units in an organization are normally struc-
tured in a systematized mode, service units in a discretionary
mode, and R&D units in a developmental mode (Van de Ven
and Delbecq, 1974). Structure and process variables that are
not prescribed at the macro-level are left to the particularistic
control of the subunit. Only these variables should interact
with context to explain variations in performance.
Future developments of the selection approach to fit in con-

tingency theories may yield promising results if multiple levels
of organizational analysis are taken into account. This requires
bracketing into two groups structure and process variables that
are (1) established at the macro-level and (2) particularistic at
the micro-level. For the first grouping of variables, fit is ana-
lyzed as a congruence relationship between context and struc-
ture and process; for the second group, fit might be analyzed
as a contingency relationship, using the interaction approach.
Interaction Approach
A second interpretation of fit is that it is an interaction effect of

the context and structure of an organization on performance
much like the classic studies of the interaction of sun, rain, and
soil nutrients on crop yields (Van de Ven, 1979). The focus here
is not so much on understanding the congruence between
context and structure as in the selection approach, but rather
on explaining variations in organizational performance from the
interaction of organizational structure and context. For exam-
ple, Figure 1 shows a typical interaction hypothesis of environ-
mental heterogeneity and structural complexity on organiza-
tional performance. This interaction hypothesis is based on
Ashby's (1956) concept of requisite variety, in which organiza-
tional adaptability is enhanced when the degree of complexity
present in the environment is reflected in the structure of the
organization.
Mixed results have been obtained for this common and popu-
lar approach to fit. Correlational studies have shown that the
relationships between structure and context are stronger for
Forms of Fit
x
a)
E High performance
0
U
Low performance
E
(n
Homogeneous Heterogeneous
ORGANIZATIONAL ENVIRONMENT
Figure 1. Example of typical interaction hypothesis of environmental

heterogeneity and structural complexity on performance.
higher performing organizations than for lower performing

organizations, but often the differences are small and not
significant (Negandi and Reimann, 1972; Child, 1974; Khand-
walla, 1974; Van de Ven and Ferry, 1980). In studies of
interaction effects (Mohr, 1971; Pennings, 1975; Tush-
man, 1977, 1978, 1979; Van deVen and Drazin, 1978;
Schoonhoven, 1981), only the Tushman and Schoonhoven
studies provided support for the interaction hypothesis.
These mixed results may be due to many methodological

problems of researchers attempting to model interactions
from field survey data. Correlations among structure and con-
text make it difficult to decompose and assess the effects of
interactions versus the effects of intercorrelations (Green,
1978). Classification errors often arise from procedures that
dichotomize or polychotomize variables that have been mea-
sured on a continuous basis for the purpose of creating ANO-
VA classes (Pierce, Dunham, and Blackburn, 1979). Significant
interaction terms may result solely from the scale of measure-
ment of the dependent variable (Green, 1978). Also, as
Schoonhoven (1981 ) pointed out, many researchers have not
appropriately operationalized their concepts of fit. In particular,
multiplicative interaction terms in regression analyses limit the
form of the interaction only to acceleration and deceleration
effects, which researchers have not specifically hypothesized
in their concept of fit. Multiplicative interactions are usually
correlated with the variables from which they are developed,
causing multicollinearity problems in the analysis (Green,
1978; Schoonhoven, 1981; Fry and Slocum, 1984).
Several researchers have proposed a deviation-score approach

for examining the interaction form of fit in contingency theory
(Ferry, 1979; Dewar and Werbel, 1979; J. Miller, 1981; Fry and
Slocum, 1984). Rather than looking for classical interaction
effects, proponents of this approach have analyzed the im-
pact of deviations in structure from an ideal context-structure
model, in which fit is defined as adherence to a linear relation-
ship between dimensions of context and structure. A lack of fit
results from a deviation from that relationship (Alexander,
1964). This approach is consistent with an interaction
approach; that is, only certain designs are expected to give
high performance in a given context, and departures from such
designs are expected to result in lower performance. The
deviation-score approach and the interaction approach are
similar only to the extent that they attempt to model the same
underlying bivariate fit. Statistically, however, they are quite
different. The interaction approach deals with acceleration and
deceleration effects formally equivalent to the catalytic type
found in chemistry. The deviation-score approach relies on the
calculation of a matching variable and is the bivariate equiva-
lent of the multivariate systems approach.1
Figure 2 displays this form of analysis graphically. Organization

A, being further away from the ideal linear context-structure
relationship than Organization B, is expected to have lower
performance. This form of fit is examined statistically by corre-
lating the absolute values of context-structure residuals with
performance.
Systems Approach
Studies that adopt the selection and interaction definitions of

fit tend to focus on how single contextual factors affect single
structural characteristics and how these pairs of context and
structure factors interact to explain performance. This reduc-
tionism treats the anatomy of an organization as being decom-
posable into elements that can be examined independently.
The knowledge gained from each element can then be aggre-
gated to understand the whole organizational system.
Recently, a systems approach to contingency theory has

emerged, reacting against such reductionism. Advocates of
this approach (D. Miller, 1981; Van de Ven and Drazin, 1985)
assert that the understanding of context-structure perform-
ance relationships can only advance by addressing simul-
taneously the many contingencies, structural alternatives, and
performance criteria that must be considered holistically to
1
understand organization design. Unlike the selection and in-
teraction approaches to fit, the systems approach consists of
We are indebted to an anonymous ASQ
reviewer for pointing out this statistical several novel alternative methods characterizing the patterns
difference. of interdependencies present in organizations.
Forms of Fit
(a) c
I
-J
Z wL
<-
,x,) Organization B
0 t Organization A
oI /
? - --?--Amount of deviation
0
Low ORGANIZATIONAL High
CONTEXT
(b) m
w Organizan Bn A
Z |
P cc
0 |
E
i)
Low High
PERFORMANCE g
Figure 2. (a) Deviation of organizations A and B from context structure

relationship; (b) Expected relationship between deviation scores (abso-
lute values) and performance.
The systems approach emphasizes the need to adopt multi-

variate analysis to examine patterns of consistency among
dimensions of organizational context, structure, and perform-
ance (D. Miller, 1981). Most recently, the systems approach
has begun to incorporate the general systems theory concept
of equifinality by interpreting fit as feasible sets of equally
effective alternative designs, with each design internally con-
sistent in its structural pattern and with each set matched to a
configuration of contingencies facing the organization. How-
ever, because analytical procedures for examining equifinality
in organizational design remain to be developed (Van de Ven
and Drazin, 1985), only the pattern-analysis approach is dis-
cussed and examined in this paper.
Organizations operate in contexts of multiple and often con-
flicting contingencies, and theorists have had an ongoing de-
bate about whether organization structure and process should
be matched to the environment, size, or technology of the
organization (Ford and Slocum, 1977). But, as Child (1977: 175)
questioned, "What happens when a configuration of different
contingencies is found, each having distinctive implications for
organizational design?" Bivariate analysis of a given contextual
factor with a structural characteristic cannot address this ques-
tion. The organizational implications of each contingency are
unlikely to be the same and are often in conflict with each
other. As a result, trade-off decisions begin to emerge, and
attempts to respond to multiple and conflicting contingencies
are likely to create internal inconsistencies in the structural
patterns of organizations. To address these problems, a pat-
tern analysis is needed for the interactions of multiple contin-
gencies and structural patterns on organizational performance.
For example, Child (1977: 175), addressing the design di-

lemma of a large organization facing a variable environment,
asked: "Should it set a limit on its internal formalization in
order to remain adaptable, or should it allow this to rise as a
means of coping administratively with the internal complexity
that tends to accompany large scale?" Child, in his study of
manufacturing firms (1 975) and airlines (1 977), found that
high-performing organizations had structures that were inter-
nally consistent, while the low-performing organizations were
inconsistent. He maintained that the inconsistent organizations
adopted structures that attempted to respond to multiple
contingencies, whereas the consistent organizations adopted
structures matched to a single contingency.
Similarly, Khandwalla (1 973) showed that internal consistency

among structural variables - defined as the gestalt of the
organization -was positively related to organizational per-
formance. The systems frameworks of various authors (Alex-
ander, 1964; Gerwin, 1976; Galbraith, 1977; NadlerandTush-
man, 1980; Van de Ven and Ferry, 1980) all hypothesized that
consistency among organizational design characteristics led to
performance. However, they did not develop analytical proce-
dures to examine their hypotheses empirically. In the systems
approach, fit results in a pattern of structure and process that
matches the contextual setting and is internally consistent.
A system analysis approach to fit is graphically presented in

Figure 3. For purposes of illustration, only one ideal type and
two underlying dimensions of structure are shown, but the
patterning involved could be easily extended to multiple ideal
types or higher dimensionalities. Three hypothetical (A, B, C)
organizations are plotted around the ideal type. In the systems
approach, the more an organization deviates from the ideal
type, the lower the expected performance. In Figure 3, the
performance ordering is A, B, C, with Organization C having
the lowest performance.
In summary, the systems approach maintains that two basic

choices confront the organizational designer: (1) to select the
organizational pattern of structure and process that matches
the set of contingencies facing the firm, and (2) to develop
structures and processes that are internally consistent. The
tasks for theorists and researchers adopting the systems
Forms of Fit
Contours of Decreasing Performance

N
0
0
z
ul
LU
2)
0
F-
o) Co oetPromigOgnzto
C) B
0
IT =Ideal Type
A =Highest Performing Organization
B =Moderately Performing Organization
0 C =Lowest Performing Organization
Low High
ORGANIZATIONAL STRUCTURE -DIMENSION 1
Figure 3. A graphic representation of the systems approach to fit: an ideal

type organization and three organizations of different performance.
definition of fit are to identify the feasible set of organizational

structures and processes that are effective for different con-
text configurations and to understand which patterns of organi-
zational structure and process are internally consistent and
inconsistent.
Unique and Complementary Information
The three forms of fit presented in this paper are not mutually
exclusive and can provide both unique and complementary
information on the fit in a researcher's data. For example, the
selection approach is useful for determining important context-
structure relationships. When several contextual factors are
correlated with the structural variables, it is possible that
conflicting contingencies are present (Child, 1 975). In this
case, more complex systems tests for internal consistency,
using the pattern approach, may be called for. Alternatively, a
single contextual variable, strongly related to many organiza-
tional structure and process variables, indicates that ANOVA
might not detect the effects of mismatches between context
and structure on performance, and a deviation-score approach
may be more appropriate.
The selection approach to fit might also be combined with the

interaction approach by categorizing structure and process
variables into two groups, those variables that are subject to
macro-switching rules and those that are more particularistic
and, hence, variable. Fit would be interpreted in two ways.
First, as congruence, or isomorphism between those structure
and process variables that are highly correlated with context,
and, second, as an interaction form of fit for the particularistic
variables.
A comparison of the results of the interaction and systems
approaches to fit can also be illuminating. The interaction
approach assumes that a disaggregated analysis of pairs of
context-structure variables on performance is possible. It may
be that such reductionism cannot detect effects of fit that are
present at a holistic or gestalt level (D. Miller, 1 981; Van de
Ven and Drazin, 1985). Whenever the contingency theory in
question is based, even remotely, on structural types, then
interaction results should be compared with systems results. If
the interaction results are not significant, but the systems
results are, then it can be reasonably concluded that fit does
not occur at the level of any individual variable alone but rather
at the level of deviation from an overall pattern of several
variables (Van de Ven and Drazin, 1985). By relying on the
interaction approach alone one might erroneously conclude
that contingency theory is not relevant (Pennings, 1 975).
If the interaction approach does detect fit, but only among

certain pairs of context-structure relationships, such findings
would indicate that those context-structure boundaries are
more salient predictors of performance than others (Khand-
walla, 1973; D. Miller, 1981). Such findings would be of great
practical utility, implying that limited resources should be allo-
cated to the most critical context-structure relationships. An
interaction approach can therefore supplement and further
specify the findings of the more general systems approach
(D. Miller, 1981).
Examining multiple approaches to fit in contingency studies

and relating these findings to unique sample characteristics
can help in the development of mid-range theories of fit. The
forms of fit that hold at the work-unit or job-design level may
be radically different from those found at the industry or
population level. Similarly the nature of fit may be dependent
on the size and maturity of the organizations under study
(Aldrich, 1979) or the rate of change experienced by the
organizations (D. Miller, 1 981). By relating the pattern of
context-structure-performance relationships to the unique
characteristics of their sample, researchers can develop mid-
range hypotheses about the nature of fit appropriate to their
organizations. Then, by conducting crucial experiments (Stinch-
combe, 1968) based on these a priori grounds, they can
compare types of fit and extend our knowledge of contingency
theory.
TASK-CONTINGENCY THEORY OF WORK-UNIT DESIGN
In this paper, empirical tests of the three approaches to fit are

illustrated by focusing on a task-contingency theory of work-
unit design and the associated data base formed to test that
theory. The common data base allows one to compare unique
and complementary information in the selection, interaction,
and systems approaches to fit in one contingency theory.
Moreover, an examination of these multiple forms of fit pro-
vides for a better understanding of the nature of fit in work
units than would be possible using only one approach.
The three approaches to fit are compared by examining the

task contingency model of work-unit design developed by Van
de Ven and his colleagues (Van de Ven and Delbecq, 1974; Van
de Ven, Delbecq, and Koenig, 1976; Van de Ven, 1 976a,
Forms of Fit
1 976b; Van de Ven and Drazin, 1978). This model has been
extended and incorporated as a core part of the larger Organi-
zational Assessment (OA) framework and instruments (Van de
Ven and Ferry, 1980; Ferry, 1983). The OA research program
aims to develop a conceptual framework and related measure-
ment instruments for assessing the performance of jobs, work
groups, interorganizational relationships, and organizations on
the basis of how they are organized and the environments in
which they operate. At the center of the OA research effort is a
contingency theory of job, work-unit, and organizational de-
sign. Here we focus only on the OA task-contingency theory of
work-unit design. A work-unit is defined as the smallest collec-
tive group in the organization; it consists of a supervisor and all
personnel who report to that supervisor.
The OA task-contingency theory proposes that high-
performing units that undertake work at low, medium, and
high levels of task difficulty and task variability will adopt,
respectively, systematized, discretionary, and developmental
modes of structure and process. Here mode means a logically
coherent pattern of structure and process matched to a level of
task uncertainty. The structural elements of these modes are
defined in terms of: (1) specialization, the number of different
work activities performed by a unit; (2) standardization, the
procedures and pacing rules that are followed in task perform-
ance; (3) discretion, the amount of work-related decision mak-
ing that the supervisor and employees exercise; and (4) per-
sonnel expertise, the skil-ks.required of personnel to operate
the program. Process is defined as the coordination mecha-
nisms used by unit personnel who execute the program.
Coordination consists of the frequency of oral and written
communications, as well as the methods used to resolve
conflict among unit personnel.
Table 2 shows the underlying pattern of structure and process

dimensions that distinguish the systematized, discretionary,
and developmental modes. The systematized mode is a pro-
gram for efficiently organizing and managing repetitive tasks
that are generally well understood. Work roles are specialized,
highly codified, and standardized so that members with lower
expertise, who do not exercise much discretion, can perform
them effectively. Supervisors deal with problems and excep-
tions, and minimal coordination is required among unit mem-
bers. The frequency of conflict is low because of low inter-
dependence among unit members and resolution of conflict by
appeal to authority or rules. Departures from this mode of
operation that allow for greater employee discretion, less
standardization, or greater interchangeability are expected to
cause unnecessary and inefficient repetition of tasks, thereby
reducing efficiency and increasing frustration and
dissatisfaction.
The discretionary mode is a program for managing tasks that

recur periodically but exhibit a sufficient number of variations
and exceptions to require different methods, procedures, and
adjustments for effective handling. The discretionary mode
generally consists of a repertoire of alternative methods for
dealing with tasks, problems, and issues. Guidelines are avail-
able to employees for choosing among these methods; that is,
work is only-partly codified and requires a greater level of
expertise to accommodate the necessary decision making and
Table 2
Hypothesized Systematized, Discretionary, and Developmental Modes in Task-Contingency Model of Work-

Unit Design*
Task Uncertainty (difficulty and variability)

Organizational Low (L) Medium (M) High (H)
characteristics (Systematized) (Discretionary) (Developmental)
Unit structure
Unit specialization H M L
Unit standardization H M L
Personnel expertise L M H
Supervisory discretion H M L
Employee discretion L M H
Unit Processes
Oral communication L M H
Written communication L M H
Frequency of conflict L M H
Conflict resolution by
Avoidance and smoothing H M L

Authority H M L
Confrontation L M H
Performancet
Job satisfaction H/L H/L H/L

Unit efficiency H/L H/L H/L
*Adapted from Van de Ven, 1 976a.

tHigh based on pattern described; low based on other pattern.
information processing. As the number and difficulty of excep-

tions increases, more information flows between members of
the unit and more interdependence develops. Unit members
exchange ideas, problems, and solutions laterally in the course
of dealing with the greater uncertainty. Levels of conflict and
disagreement are higher, and mutual adjustment becomes
more important in resolving them. The codification of the
systematized mode would be ineffective in achieving goals in
the discretionary mode; the nature of the work requires discre-
tion and flexibility to adequately accommodate task variations.
However, too much flexibility would reduce performance. The
essence of the discretionary mode is the diagnosis and cate-
gorization of problems into known treatment and resolution
alternatives; only occasionally are true invention and develop-
ment necessary.
The developmental mode is a program for handling tasks,

problems, or issues that are sufficiently difficult to require
extensive search, evaluation, and judgment. Developmental
structure and process are characterized by low levels of stan-
dardization and specialization, group decision making and prob-
lem solving, high employee discretion, and high levels of
interdependence and communication. Whereas a discretionary
program provides procedures, rules, and norms, a develop-
mental program tends to provide only broad and difficult goals,
and much effort is expended in developing unique strategies
for achieving these goals.
Unit efficiency (output per person) and the average level of job
satisfaction are hypothesized in the model presented here to
be contingent upon the fit between the level of task uncertain-
Forms of Fit
ty faced by the unit and the internal pattern or mode of

structure and process the unit adopts. The selection, interac-
tion, and systems approaches to fit are all appropriate methods
for assessing the nature of fit relationships implied in this
model. Each approach yields different information and is
appropriate for testing certain relationships expected in this
model.
Sample and Measurement Procedures

Data to test this contingency theory were obtained from 629
employment security units in 60 offices located in California
and Wisconsin in 1975 and 1978. These units administered the
Department of Labor's Job Services, Unemployment Insur-
ance, Workman's Compensation, and Work Incentive pro-
grams at the local level. The following basic unit types were
studied in the survey:
Intake and claims processing. Received, registered, and pro-

cessed claims for unemployment compensation (UC).
Adjudication: Investigated, documented, and resolved dis-
puted UC claims.
Placement: Matched unemployed individuals to job openings.
Counselling and rehabilitation: Advised clients in training for
career objectives.
Work incentives: Provided intensive job services and employ-
ment development programs for individuals on welfare.
General services: Handled all other client and staff-related
work.
Management and clerical: Provided support, including super-

visory and secretarial services.
Detailed descriptions of each unit's work are available in Van
de Ven and Ferry (1 980).
With the exception of unit efficiency, all the dimensions in Table

2 were measured with the Organization Assessment Instru-
ment (OAI), as developed and evaluated by Van de Ven and
Ferry (1980). Questionnaires were completed by all unit mem-
bers and supervisors during business hours after a member of
the OA research team explained the purpose and use of the
study. The data reported here are at the unit level and were
derived from the responses of the unit supervisor and the
average of all responses of the unit personnel reporting to that
supervisor, equally weighted. This aggregation procedure is
justified theoretically, because a work unit is defined as con-
sisting of two hierarchically related positions, a supervisor and
all employees reporting to that supervisor. When the empirical
implications of this approach versus a simple averaging of the
scores of all unit personnel were examined, they showed that
mean scores and correlations among all variables were the
same for both procedures (Van de Ven and Ferry, 1980).
Measures of efficiency were obtained from organizational per-
formance records for each unit and consisted of the amount of
output produced per full-time equivalent position. Measures of
unit size, office size, administrative intensity, and levels were
obtained from organizational charts developed for each com-
munity office. Due to space limitations, readers are referred to
Van de Ven and Ferry (1980) for details on questionnaire items
Table 3
Correlations among Unit Context, Structure, Process, and Performance Variables (N = 629)
Coeff.
X SD Alpha 1 2 3 4 5 6
Unit context
1. Task uncertainty 2.24 .54 .81
2. Office size 4.20 1.64 NAt .055
3. Unit size 7.32 4.05 NA -.088 .011
4. Administrative intensity .23 .15 NA -.033 .439i .020
5. No. levels from top 4.67 1.26 NA -.2750 .362i .040 .2210
Unit structure
6. Unit specialization* 3.12 :95 .85 .121 Ad -.012 -.113 -.016 -.1780
7. Unit standardization* 3.45 .73 .80 -.4680 -.3570 .085 -.122 .351 Ac -.1880
8. Personnel expertise* 2.96 .49 .40 .4670 .010 -.120 -.062 -.1900 -.096
9. Supervisory discretion 2.98 .68 .81 -.096 -.027 .067 -.043 .074 -.0876
10. Employee discretion 3.52 .74 .84 .194i -.042 -.1176 -.123 -.2950 .157
Unit process
11. Written communication* 1.81 .45 .68 .3000 .064 .079 -.1 23 -.029 .095
12. Oral communication 2.29 .52 .69 .3340 -.2030 -.047 -.1 28 .219000 -.1290
13. Frequency of conflict 2.11 .83 - .13500 -.073 .1740 -.023 -.1580 .058
14. Conflict resolution by:
a. Avoidance and smoothing 2.29 .76 - -.033 -.034 .099 .022 -.029 -.043
b. Confrontation 3.36 .95 - .057 -.045 -.1 326 -.1 27 -.004 -.008
c. Authority 2.73 .88 - -.080 .006 .047 -.072 .115 -.017
Unit performance
15. Unit efficiency 4.88 .92 NA -.023 -.2430 .095 -.2060 -.113 .007
16. Job satisfaction 4.99 1.00 .78 -.043 -.137 .011 -.3610Ac .062 .051
*Design characteristics prescribed at macro-level.

tp < .05, p < .01p , ilp < .001.
NA = Not applicable.
and the psychometric properties of the instrument. However,

where relevant, reliabilities are reported in Table 3.
In the past, contingency studies have been criticized for lack of

variation in the data, especially in the contingent variables
(Pfeffer, 1982). To ensure that the data in this study showed
adequate variation to test the task-contingency theory, median
splits were performed on all variables, and the resultant mean
differences were compared using t-tests.2 Means for all vari-
ables (including task uncertainty) were significantly different at
the p < .001 level. Task uncertainty scores ranged from a low
of 1.09 to a high of 4.1, covering most of the five-point range of
the component items of the task uncertainty scale. Means and
standard deviations for all variables are shown in Table 3.
Selection Approach
The basic hypothesis in a natural selection approach to fit in the

OA task-contingency theory is that task uncertainty should be a
strong predictor of work-unit structure and process. As D.
Miller (1981: 1 0) has pointed out, natural selection is a power-
ful Darwinistic force, which "imposes order on organizational
forms and limits their variety and number." Forms or patterns
that are dysfunctional are likely to be selected against, while
more functional patterns will be propagated. Performance is
notably absent in this hypothesis, because the selection
approach assumes that structural forms must be adaptive to
the environment, or the organizational unit will be selected out
of existence. Under a natural selection view of fit, task uncer-
tainty should be correlated strongly with all the structure and
process variables of Table 2.
2
The authors thank an anonymous ASQ re- The selection process in a managerial selection approach is
viewer for suggesting this procedure. somewhat different. Relationships are presumed to exist be-
Forms of Fit
7 8 9 10 11 12 13 14a 14b 14c 15
-.2480
.1860 -.063
-.1620 .072 .335o
-.078.2150 .0970 .079

.011 .291o .021 .069 .3150
-.099 .050 .074 .039 -.004 .037
-.017 .000 .041 .095 -.081 -.029 .442o

-.024 -.004 -.027 .116 .089 .162o -.2930 -.4360
-.086 -.162o .080 -.052 .036 .071 -.065 -.110 .3480
.013 .038 -.021 .113 .047 .117 -.071 -.118 .065 .064
.181o -,021 .038 .089 .006 .120o -.3050 -.2860 .339o .2970 .214o
tween work units and the macro-organizations in which they

are embedded. Management, through staff units, is expected
to establish switching rules that control certain structural
dimensions of different types of subunits. In this study, staff
units in the headquarters office of the Employment Security
Agencies, as well as the state-level Civil Service Departments,
exerted strong influence over the structural characteristics of
specialization, expertise, standardization, and written com-
munications at the work-unit level. The level of a work unit's
specialization and expertise was partly controlled through spe-
cific job descriptions and the civil service requirements (educa-
tion, experience, etc.) associated with those descriptions.
Standardization was also governed by switching rules imposed
at the macro-level. Staff units developed and disseminated
clerical and computer procedures, which were codified and
documented in unit operations manuals. These same rules
also set forth requirements for the number and degree of
written communications related to documenting actions taken
on clients and for periodic management information reports.
In the managerial selection approach, other structure and

process characteristics, such as oral communications, level of
conflict, conflict-resolution style, and employee or supervisory
discretion are difficult if not impossible to control through the
development of switching rules. These parameters should
show a broader range of variance within unit type, reflecting
the more particularistic style of unit leaders and personnel.
Therefore, in managerial selection a strong correlation should
exist only between task uncertainty and those structure and
process variables capable of being programmed at the
macro-level.
Table 3 presents a correlation matrix among the unit context,
structure, process, and performance variables. The variables
subject to macro-organizational switching rules are designated
with an asterisk. The significant correlations with task uncer-
tainty support the basic congruency hypothesis in OA theory.
As task uncertainty increases, unit structure and process
change to match this uncertainty. Specialization, personnel
expertise, and employee discretion increase, while standard-
ization and supervisory discretion decrease. Some aspects of
unit process are also related to the level of task uncertainty.
Written and oral communications increase with higher levels of
task uncertainty as does the frequency of conflict, again, in
accordance with OA theory. Only the style of conflict resolu-
tion is not related to task uncertainty.
Other contextual factors are correlated with unit structure and

process as well, but not as strongly as task uncertainty. In
particular, the size of a unit and the number of levels that it is
removed from the top have a number of significant effects on
the unit process dimensions many of them in the opposite
direction of the effect of task uncertainty.
A review of the correlations between task uncertainty and unit

structure and process allows comparison of the natural and
managerial selection hypotheses. Task uncertainty is signifi-
cantly correlated with all the unit structure and process vari-
ables except the three styles of conflict resolution (which have
small but significant correlations with other contextual factors).
These findings support the natural selection hypothesis.
However, Table 3 shows large differences in the sizes of the
correlations. Three of the four dimensions hypothesized in the
managerial selection model as subject to macro-organizational
switching rules (unit standardization, personnel expertise, and
written communications) are strongly correlated with task
uncertainty. Unit specialization, while significant, has a sub-
stantially lower correlation with task uncertainty. The correla-
tions of task uncertainty with the other unit structure and
process variables are substanially lower than these four. Only
oral communications is an exception. Overall, although the
evidence provides some support for both natural and manage-
rial selection theories of forms of fit in the OA task-contingency
theory, more support is shown for the managerial selection
perspective.
Interaction Approach Examined with ANOVA and

Deviation Scores
Although the OA task-contingency theory of work-unit design

is a theory of modes of behavior, it can also be thought of as a
set of independent mini-theories of task-structure-process-
performance relationships. This approach requires disaggre-
gating the modal characteristics of the OA theory into its
3 component structure and process variables and then analyzing
We are indebted to Michael Tushman for the effects of the interactions of each of these variables with
pointing out that certain contingency task uncertainty on performance. The advantage of this
theories either focus theoretically only on
one variable or measure organizations in
approach is that it provides accurate and useful details about
such a way that gestalt characteristics are individual structure and process variables (D. Miller, 1981). Its
captured in one variable. In such cases primary disadvantage is its implied reductionism. The reduc-
there is no reductionism, and this dis-
advantage of the interaction approach is tionist approach may not capture the very gestalt character of
eliminated. organization that the theory implies.3
Forms of Fit
The most common approach to the interaction test of fit

consists of a series of two-way analyses of variance (or regres-
sions) with task uncertainty, individual unit structure and pro-
cess variables, and interactions of task uncertainty with these
dimensions, as the independent variables, and unit perform-
ance (efficiency and satisfaction) as the dependent variable. To
conduct this test, task uncertainty was trichotomized into
roughly equal categories representing low, medium, and high
levels of task uncertainty. The eleven unit structure and pro-
cess variables were dichotomized into low and high levels,
based on frequency counts. Twenty-two separate ANOVAs
were conducted, eleven each for unit efficiency and job satis-
faction as dependent performance variables. Several alterna-
tives were explored to ensure that the interaction approach
was given an adequate testing, including using several
polychotomizing schemes, and treating the data continuously,
with multiplicative interaction terms. In all cases the results
were essentially the same as those shown in Table 4. The
Schoonhoven (1981) procedure would not be appropriate to
explore here because only 4 out of 22 multiplicative interac-
tions were significant.
Table 4 shows the results of the ANOVA tests for job satisfac-
tion and unit efficiency. An examination of the interaction
effects shows only one significant interaction effect (conflict
resolution by authority x task uncertainty) that explains average
unit satisfaction.
Table 4
Analysis of Variance of Task Uncertainty, Unit Structure and Process, and Interaction Effect for Efficiency
(N = 230) and Job Satisfaction (N = 473).
Organizational Task Uncertainty Structure and Process Interaction Effect

characteristics F p F p F P
Unit structure
Unit specialization .31 .733 2.59 .109 1.67 .189

1.85 .158 .45 .500 2.23 .180
Unit standardization .31 .734 1.00 .318 1.80 .168
1.90 .151 12.50 .001 2.13 .121
Personnel expertise .30 .738 .11 .735 .20 .819
1.85 .157 4.47 .035 .91 .402
Supervisory discretion .31 .736 .04 .843 1.02 .363
1.86 .157 4.66 .031 .95 .387
Employee discretion .31 .735 .40 .525 1.66 .192
1.86 .158 1.48 .225 2.55 .079
Unit process
Written communication .31 .736 .84 .361 .54 .583

1.84 .159 .04 .841 1.84 .159
Oral communication .31 .736 .84 .361 .54 .583
1.88 .154 8.67 .003 1.55 .212
Frequency of conflict .30 .736 1.49 .224 .51 .604
1.85 .160 28.40 .001 .45 .630
Conflict resolution by:
Avoidance and
smoothing .30 .738 .47 .495 .09 .910
1.95 .144 29.30 .001 .40 .667
Conf rontation .31 .737 .46 .496 .58 .560
2.01 .135 45.73 .001 .14 .865
Authority .31 .738 .47 .495 .09 .910
1.99 .137 34.21 .001 3.90 .021
A second approach to testing the interaction form of fit in
contingency theory is to compute deviations of residual scores
from a regression line (Ferry, 1979; Dewar and Werbel, 1979;
J. Miller, 1981; Fry and Slocum, 1984). A two-step procedure
was followed to conduct this deviation-score test. First, devia-
tion scores were constructed by regressing each unit structure
and process dimension separately on task uncertainty. Re-
siduals were calculated from the best-fitting least-squares
lines. The absolute values of these residuals were used as
deviation scores. The second step of the analysis was the
actual test of fit. The eleven deviation scores developed were
separately regressed on efficiency and satisfaction. If the
correlations of the deviation scores with efficiency and satis-
faction were significant and negative (the greater the deviation,
the lower the performance) these data were taken as evidence
of fit.
The results of the unit structure and process and task uncer-
tainty regressions used to create the deviation scores are
shown in Table 5. Because of the low correlations reported
earlier for certain structure and process dimensions with task
uncertainty, some beta values are quite close to zero, indicat-
ing that deviation scores should be interpreted as roughly
equivalent to dispersion around the mean for these variables.
The results of the actual tests of fit using the deviation scores
calculated from the above regressions are also shown in
Table 5.
Table 5
Regression Analysis of Unit Structure and Process on Task Uncertainty to Develop Deviation Scores and
Correlations of These Scores with Job Satisfaction and Unit Efficiency
Regression analysis (N = 471) Correlation of Deviation Scores*

Job Unit
Organizational satisfaction efficiency
characteristics Intercept Beta F P (N = 471) (N = 230)
Unit structure
Unit specialization 5.496 -.176 5.01 .0260 .042 -.020

Unit standardization 4.714 -.561 98.01 .0001 -.035 -.053
Personnel expertise 1.754 .731 145.50 .0001 .010 -.050
Supervisory discretion 3.176 -.086 2.29 .1310 .005 -.173g
Employee discretion 3.022 .214 11.78 .0240 -.030 -.040
Unit process
Written communication .955 .351 47.94 .0001 .070 .054

Oral communication 1.706 .475 56.98 .0001 .106 .052
Frequency of conflict 1.589 .236 9.16 .0026 -.078 -.033
Avoidance and smoothing 2.439 -.067 .87 .3510 -.114 -.089
Confrontation 3.112 .114 1.64 .2010 -.033 -.082
Authority 3.042 -.146 3.14 .0770 -.078 -.033
*p < .05, *p < .01.

*Absolute values of task-structure residuals.
Of the 22 correlations, only four are significant at the .05 level.

Deviations for oral communications are positively correlated
with satisfaction a result that is hard to interpret, given the
expectation of a negative correlation. Three other correlations
are significant and negative: conflict resolution by avoidance
and smoothing with job satisfaction, supervisory decision mak-
Forms of Fit
ing with unit efficiency, and conflict resolution by authority

with unit efficiency. However, the correlations are weak, the
highest one being only - .18. Since only 4 of the 22 possible
relationships are significant, it is probable that they are due to
chance alone (Hays, 1973, 1976). To be certain that these
results were not caused by the choice of the base-line model, a
second deviation-score procedure was tried. Here the base
line was calculated using the 45 highest performing units that
were chosen for the systems analysis. None of these 22
deviation scores correlated significantly with performance.
The results obtained using the ANOVA and deviation-score

approaches to fit are discouraging to supporters of the interac-
tion approach and in the past have led some researchers
(Pennings, 1975) to question the overall relevance of structural
contingency theory. However, since this form of fit is only one
of the several that exist for contingency-theory analysis,
perhaps it is the interaction approach, rather than contingency
theory itself, that should be questioned.
Systems Approach
Conceptually, the systems approach is similar to deviation-

score analysis. The major difference is that deviation is not
measured from a single linear equation line, but rather as a
distance from a profile described as a point in an eleven-
dimension structure and process space. The deviation scores
in the interaction approach analyzed the fit between task
uncertainty and each of the unit structure and process charac-
teristics, one dimension at a time. This systems analysis
focused on differences in pattern profiles and accounted for all
eleven variables as a set. A three-step procedure was used to
analyze the systems approach to fit in this data base.
This theory, like most contingency theories, expresses the

contingent relationships ordinally, not in ratio or interval scales.
For example, standardization is presumed to be high for the
systematized mode and low for the developmental mode. To
test the pattern approach, empirical ideal types representing
the three modes of the task-contingency theory were required.
Empirical profiles were therefore generated for the 45 highest
performing units, based on the efficiency measure, under
conditions of low, medium, and high task uncertainty (15 units
for each level of uncertainty). The mean scores of these 45
units on the 11 structure and process variables were consid-
ered as empirically derived ideal types, representing systema-
tized, discretionary, and developmental modes. These ideal
types were tested using ANOVA and MANOVA to determine if
the profiles actually differed. A comparison was also made
between these results and the theory shown in Table 2 to
determine if the derived values matched the predicted ordinal
relationships.
The results of the first step of the pattern analysis procedure

are shown in Table 6, which shows the unit structure and
process profiles of the 45 highest efficiency units under condi-
tions of low, medium, and high task uncertainty. The Fcolumn
shows the results of one-way ANOVAs to determine if the
means of the profiles on each dimension were different. Eight
of the eleven structure and process variables showed signifi-
cant differences at the .10 level. An overall MANOVA, using all
eleven variables, was also significant (F = 2.94; p <.0004).
Furthermore, the results of orthogonal planned comparisons
(Hays, 1973), to assess which means differed, revealed that
the mean scores of the structure and process variables of
standardization, supervisory discretion, and written and oral
communication differed between low, medium, and high task-
uncertainty levels. Where differences were significant, the
patterns of scores matched very closely the predicted patterns
of Table 2. These profiles, then, appear to represent the
systematized, discretionary, and developmental modes of the
OA task-contingency theory.
Table 6
Profiles of Mean Unit Structure and Process Scores for High Efficient, Low, Medium, and High Task-Uncertainty
Units*
Orthogonal
Organizational Task Uncertainty ANOVA planned
characteristics Low(N= 15) Medium (N 15) High (N= 15) F P comparisonst
Unit structure
Unit specialization -.078 -.154 .257 7.22 .002 M,H

Unit standardization .445 .085 -.477 12.95 .001 L,M,H
Personnel expertise -.215 -.132 .343 3.99 .026 M,H
Supervisory discretion .026 -.210 -.283 2.52 .093 L,M,H
Employee discretion -.157 -.057 .201 1.94 .156 M,H
Unit process
Written communication -.337 .048 .214 4.02 .025 L,M,H

Oral communication -.275 -.002 .228 3.01 .060 L,M,H
Frequency of conflict -.141 -.101 .243 1.01 .375
Avoidance and smoothing -.150 .044 .067 .29 .751
Confrontation .248 -.101 -.075 .11 .898
Authority .399 -.252 -.005 3.36 .049 L,M
*Based on standardized scores.

tSignificant (p < .10) differences in mean values; based on orthogonal planned comparisons, for low, medium, and high
task-uncertainty units.
In the second step, differences between these ideal patterns

and the patterns of the remaining units were calculated using a
Eucledian distance metric. The resultant distance calculations
are between a focal unit and its respective ideal type, accord-
ing to the focal unit's level of task uncertainty. The distance
measure is calculated as follows:
DIST = / E (Xis - Xjs)2,
where Xis is the score of the ideal unit on the sth structure or
process dimension and where Xjs is the score of the jth focal
unit on the sth dimension.
Based on the distances calculated for all units in the sample, a

third step actually tested the pattern approach to contingency
theory. The calculated distance measure was correlated with
the two performance measures of satisfaction and efficiency.
Fit, or perhaps more appropriately, misfit, would be demon-
strated if the distance score was negatively correlated with the
performance measures. The greater the distance from the
respective ideal type, the lower the hypothesized
performance.
Forms of Fit
The results of this analysis are shown in Table 7. As predicted,

both unit efficiency and job satisfaction are negatively corre-
lated with a unit's distance from its ideal-type profile. Effi-
ciency correlated - .241 (p <.001) with overall distance (for all
units except the original high-performing units used to calcu-
late the ideal types), and satisfaction correlated - .1 27 (p <.01)
with overall distance. Table 7 also shows the component
correlations between distance and performance within low,
medium, and high task-uncertainty levels. Low and high uncer-
tainty units showed the greatest correlation between distance
and efficiency (r = - .308 for low task-uncertainty units and
r = - .320 for high-uncertainty units), while the distance-
efficiency correlation was not significant for medium task-
uncertainty units. For the satisfaction-dependent variable, dis-
tance was only significant for the low task-uncertainty units
(r = -.1 94).
Table 7
Correlations of Distance Measure with Unit Efficiency and Job Satisfac-

tion, Excluding High-Performance Units
Unit Efficiency Job Satisfaction

Distance N N
All units -.241g 185 -.1270 438

Task uncertainty units
Low -.3080 54 -.1940 137
Medium -.093 76 -.091 173
High -.320* 55 -.101 128
*Not significant; *p < .02; *0p < .001.
These results show significant support for the systems

approach to fit in the OA task-contingency theory of work-unit
modes. Departures from the ideal systematized, discretionary,
and developmental modes at each level of task uncertainty
were found to influence significantly unit efficiency and
satisfaction.
DISCUSSION AND IMPLICATIONS
It is clear, first, that managerial selection, operating through

macro-organizational switching rules that are contingent on
task uncertainty, has a significant influence on the structural
characteristics of subunits. For those variables subject to pre-
scription at a higher level in the organization, significant correla-
tions were found between context and structure and process.
However, certain process characteristics of subunits appear to
be less influenced by these macro-organizational switching
rules and tend to reflect the particularistic style and discretion
of unit personnel. With the exception of Comstock and Scott
(1977), these findings and their consequences have been
overlooked in many studies of organizational subunits. Consist-
ent with their findings, the results obtained here emphasize
that the structure and process choices for a particular organiza-
tional level are constrained and limited by design criteria im-
posed from macro-organizational levels. These findings not
only support a managerial selection or congruence view of fit
but also have important implications for understanding the
other patterns of fit found in analyzing this contingency theory.
Second, no empirical evidence was obtained to substantiate
the interaction approach to fit in the OA task-contingency
model. These results were somewhat anticipated because of
previous related analyses (Van de Ven and Drazin, 1978; Ferry,
1979; Van de Ven and Ferry, 1980). One explanation for this
finding is that the empirical support present for the selection or
congruence approach to fit in this data base implies that little
variance exists for unit structure within levels of task uncer-
tainty. The probability of detecting significant interactions of
task uncertainty and structure on unit performance using
ANOVA is therefore substantially reduced.
Furthermore, the deviation-score approach to fit, designed to

overcome some of the limitations of the interaction approach,
also failed to yield significant results. One explanation for this
finding may lie in the difficulties associated with choosing the
base-line context-structure relationship (Dewar and Werbel,
1979) from which residuals are calculated. If the regression
equation chosen does not adequately represent high-
performing units, then deviations from that equation will not be
meaningful. However, using a high-performance holdout sam-
ple to establish the base-line model did not improve the
results.
As discussed, the OA task-contingency model is essentially a

theory of organizational modes. A systems approach to fit may
be a more appropriate form of analysis for this type of theory.
Here, fit is explained by a departure from a multivariate pattern
of unit context and structure and process not by the
departures of isolated pairs of unit context and structure and
process variables. For example, a given variable, such as
standardization, may have a perfect match with a unit's level of
task uncertainty, yet overall performance for that unit may be
low because other variables not included in the analysis may
be inconsistently matched with task uncertainty. Pairwise
analysis may not be capable of detecting overall patterns of
internal consistency among unit context and structure and
process.
Support for the systems approach to fit was found in these

data. Inconsistencies in a unit's structure and process, arising
from departures from ideal-type systematized, discretionary,
and developmental modes, were significantly related to per-
formance. By viewing the OA task-contingency model as a
theory of organizational modes and adopting a systems
approach to fit, it was shown that fit is a significant predictor of
unit performance.
Overall, these empirical findings suggest that explaining the

performance of organizational units requires a more sophisti-
cated approach to contingency theory than earlier efforts have
used. A contingency model for the subunits in this sample
appears to require that fit is the joint product of managerial
selection and departures from an ideal multivariate pattern. No
evidence was found to support the mainstream view of con-
tingency theorists that fit is the simple interaction between
isolated pairs of unit-context and structure and process dimen-
sions on performance. Using multiple approaches to the eval-
uation of fit in this data base revealed that both congruent and
contingency forms of fit were operating. This result is impor-
tant because it replicates (although using different procedures)
Forms of Fit
the universal and contingency findings of Dewar and Werbel

(1979) and Fry and Slocum (1984). An exploration of the
interrelationships between selection (congruent) and con-
tingency approaches to fit is an important direction for
contingency-theory researchers to follow.
We believe that the evaluation of multiple approaches to fit in

the OA task-contingency theory provides an example of knowl-
edge accumulation that contingency-theory researchers
should follow. By documenting such results and accumulating
knowledge across and between organizational levels and
populations, researchers can make significant advances in
mid-range theory. If future subunit studies replicate the find-
ings on the alternative approaches to fit reported here, macro-
micro relationships may be more readily understood. If a series
of studies at an industry level of analysis or for professional
rather than bureaucratic subunits shows a different pattern of
findings, then some systematic relationships between types
(or levels) of organizations may become evident. Knowing that
forms of fit differ across conditions will be useful and may
help to clear up inconsistent contingency-theory findings. Re-
porting tests of only one form of fit leaves more questions
unanswered than resolved.
These research findings have a number of broader implications

for general contingency-theory research. First, contingency
studies should be designed to permit comparative evaluation
of several forms of fit. The resulting complementary informa-
tion can lead to more comprehensive descriptions of context-
structure-performance relationships than a single approach to
fit alone. By examining multiple approaches to fit in contingen-
cy studies and relating these findings to unique sample charac-
teristics, one can develop mid-range theories of fit. In particu-
lar, researchers should attempt to explore and resolve the
relationships and interdependencies among congruency
(selection) and contingency (interaction and systems) forms
of fit.
Second, contingency-theory researchers should be encour-

aged to further develop systems approaches to fit. Pattern
analysis, as presented in this paper, is only one of several
alternatives available to examine the gestalt characteristics of
organizations. For example, it is common in both the strategy
and organization literatures to examine the pattern of intercor-
relations among variables (environment, strategy, structure) by
dividing the sample into low- and high-performing groups. The
high-performing group is expected to reveal relationships clos-
er to a hypothesized model than the low-performing group
(Van de Ven and Ferry, 1 980).4
Furthermore, researchers could examine the effects of multi-
ple contextual elements on fit. In this paper, the multivariate
nature of the data was limited to only structure and process
variables, while context was treated as a single variable.
However, as Child (1977) has pointed out, multiple contextual
factors can have conflicting implications on design. Indeed, in
the analysis of these data, size correlated with several structur-
al variables in a direction opposite to that of task uncertainty.
4 Under these circumstances researchers might investigate how
We would like to thank Richard Daft for large organizations in uncertain environments or small orga-
pointing out this approach to us. nizations in certain environments address this apparent design
dilemma. Adopting a systems approach seems uniquely prom-
ising in addressing these types of research questions.
Finally, these concepts of fit may be applied not only to

structural contingency theory but to contingency theories in
general. Fit is a concept of broad utility that is increasingly
important in a wide range of organizational theories. Re-
searchers interested in job design, leadership, or strategy-
structure relationships have all at one time postulated that
organizational performance is a function of the fit or match
between two or more factors. Each of these management
disciplines could potentially benefit from a more explicit ex-
amination of fit in their area.
REFERENCES
Aldrich, Howard Dewar, Robert, and James Werbel Fry, Louis W.

1 979 Organizations and Environ- 1979 "Universalistic and contingen- 1982 "Technology-structure re-
ments. Englewood Cliffs, NJ: cy predictions of employee search: Three critical issues."
Prentice-Hall. satisfaction and conflict." Academy of Management
Administrative Science Journal, 25 (3): 532-552.
Alexander, Christopher
Quarterly, 24: 426-448.
1964 Notes on the Synthesis of Fry, Louis W., and Debra
Form. Boston: Harvard Uni- DiMaggio, Paul J., and Walter W. Schellenberg
versity Press. Powell 1984 "Congruence, contingency and
1983 "The iron cage revisited: Insti- theory building: An integrative
Ashby, W. Ross
tutional isomorphism and col- perspective." Unpublished
1956 Introduction to Cybernetics.
lective rationality in organiza- manuscript, University of
London: Chapman & Hall.
tional fields." American Washington, Seattle.
Child, John Sociological Review, 48 (April):
Fry, Louis W., and John W.
1974 "Managerial and organization 147-1 60.
Slocum, Jr.
factors associated with com-
Dubin, Robert 1984 "Technology, structure and
pany performance - Part I.-
1976 "Theory building in applied work group effectiveness: A
Journal of Management
areas." In Marvin Dunnette test of a contingency model."
Studies, 11: 175-189.
(ed.), Handbook of Industrial Academy of Management
1975 "Managerial and organization
and Organizational Psycholo- Journal, 27: 221-246.
factors associated with com-
gy: 17-39. Chicago: Rand
pany performance -Part II: A Fullan, M.
McNally.
contingency analysis." Journal 1970 "Industrial technology and
of Management Studies, 12: Fennell, Mary L. worker integration in the orga-
12-27. 1980 "The effects of environmental nization." American Sociologi-
1977 Organization: A Guide to Prob- characteristics on the structure cal Review, 35: 1028-1 039.
lems and Practice. New York: of hospital clusters." Adminis-
Galbraith, Jay R.
Harper & Row. trative Science Quarterly, 25:
1977 Organization Design. Reading,
485-510.
Comstock, Donald E., and L. S. MA: Addison-Wesley.
Schroger Ferry, Diane L. 1982 "Designing the innovating
1979 "Hospital services and com- 1979 "A test of a task contingent organization." Organizational
munity characteristics: The model of unit structure and Dynamics, 10 (3): 5-25.
physician as mediator." Jour- efficiency." Unpublished doc-
Gerwin, Donald
nal of Health and Social Be- toral dissertation, The Wharton
1976 "A systems framework for
havior, 20: 89-97. School, University of
organization structural design."
Pennsylvania.
Comstock, Donald E., and W. In Ralph H. Kilmann, Louis R.
1983 "The Organization Assess-
Richard Scott Pondy, and Donald P. Slevin
ment Instrument: An evalua-
1977 "Technology and the structure (eds.), The Management of
tion of intrinsic validity." Paper
of subunits: Distinguishing in- Organization Design, 1: 89-
presented at 43rd Annual
dividual and workgroup 102.
Meeting of Academy of Man-
effects." Administrative Sci-
agement, Dallas, August. Green, Paul E.
ence Quarterly, 22: 177-202.
1978 Analyzing Multivariate Data.
Ford, Jeffrey D., and John W.
Dewar, Robert, and Jerald Hage Hinsdale, IL: Dryden.
Slocum, Jr.
1978 "Size, technology, complexity
1977 "Size, technology, environ- Hage, Jerald, and Michael Aiken
and structural differentiation:
ment and the structure of orga- 1969 "Routine technology, social
Toward a conceptual syn-
nizations." Academy of Man- structure and organization
thesis." Administrative Sci-
agement Review, 4: 561-575. goals." Administrative Science
ence Quarterly, 23: 111-136.
Quarterly, 14 (September):
Freeman, John H.
366-376.
1973 "Environment, technologyand
the administrative intensity of
organizations." American
Sociological Review, 38: 750-
763.
537!ASQ, December 1985
Forms of Fit
Hall, Richard H. Nadler, David A., and Michael L. Tosi, Henry, and John W.
1962 "Intraorganizational structural Tushman Slocum, Jr.
variation: Application of the 1980 "A model for diagnosing orga- 1984 "Contingency theory: Some
bureaucratic model." Adminis- nizational behavior: Applying a suggested directions." Journal
trative Science Quarterly, 7 congruence perspective." of Management, 10 (1): 9-26.
(December): 295-308. Organizational Dynamics, 9 (2):
Tushman, Michael L.
35-51.
Hannan, Michael T., and John H. 1977 "Special boundary roles in the
Freeman Negandhi, Anant R., and Bernard innovation process." Adminis-
1977 "The population ecology of C. Reimann trative Science Quarterly, 22:
organizations." American Jour- 1972 "A contingency theory of orga- 587-605.
nal of Sociology, 82: 929-964. nization reexamined in the con- 1978 "Technical communication in
text of a developing country." R&D Laboratories: The impact
Hays, William L.
Academy of Management of project work characteris-
1973 Statistics. New York: Holt,
Journal, 19: 137-146. tics." Academy of Manage-
Rinehart and Winston.
ment Journal, 21: 624-645.
1976 Statistics. New York: Holt, Nightingale, Donald V., and Jean-
1979 "Work characteristics and sub-
Rinehart and Winston. Marie Toulouse
unit communication structure:
1977 "Toward a multi-level congru-
Khandwalla, Pradip N. A contingency analysis." Ad-
ence theory of organization."
1973 "Viable and effective organiza- ministrative Science Quarterly,
Administrative Science
tional designs of firms." 24: 82-98.
Quarterly, 22: 264-280.
Academy of Management
Van de Ven, Andrew H.
Journal, 16: 481-495. Ouchi, William G.
1 976a "A framework for organization
1974 "Mass output orientation of 1981 Theory Z. Reading, MA: Addi-
assessment." Academy of
operations technology and son-Wesley.
Management Review, 1 (1):
organizational structure."
Pascale, Richard T., and Anthony 64-78.
G. Athos 1 976b "Equally efficient structural
Quarterly, 19: 74-97.
1981 The Art of Japanese Manage- variations within organiza-
Marsh, Robert M., and Hiroshi ment. New York: Warner tions." In R. H. Kilmann, L. R.
Mannari Books. Pondy, and D. P. Slevin (eds.),
1981 "Technology and size as deter- The Management of Organiza-
Pennings, Johannes M.
minants of the organizational tion Design: Research and
1975 "The relevance of the structural-
structure of Japanese facto- Methodology, 2: 150-170.
contingency model of organiza-
ries." Administrative Science New York: Elsevier, North-
tional effectiveness." Adminis-
Quarterly, 26: 33-57. Holland.
trative Science Quarterly, 20:
1979 Review of Howard Aldrich,
McKelvey, Bill 393-410.
Organizations and Environ-
1982 Organizational Systematics: Perrow, Charles ments. Administrative Science
Taxonomy, Evolution, Classi- 1967 "A framework for the compara- Quarterly, 24: 320-326.
fication, Berkeley, CA: Univer- tive analysis of organizations."
sity of California Press. American Sociological Review,
Van de Ven, Andrew H., and Andr6
L. Delbecq
Miller, Danny 32: 194-208.
1974 "A task contingent model of
1981 "Toward a new contingency
Peters, Thomas J., and Robert H. work-unit structure." Adminis-
approach: The search for orga- Waterman trative Science Quarterly, 192:
nizational gestalts." Journal of 1982 In Search of Excellence. New 183-197.
Management Studies, 18: York: Harper& Row.
1-26. Van de Ven, Andrew H., Andr6 L.
Pfeffer, Jeffrey Delbecq, and Richard Koenig, Jr.
Miller, John P. 1 982 Organizations and Organization 1976 "Determinants of coordination
1981 "Information processing in Theory. Marshfield, MA: modes within organizations."
organizations: An examination Pitman. American Sociological Review,
of the contingent effects of in-
Pierce, Jon L., Randall B. Dunham, 332-338.
formation ambiguity, organiza-
tion structure, communication, and Richard S. Blackburn Van de Ven, Andrew H., and Robert
and decision making on organi- 1979 "Social systems structure, job Drazin
zational conflict and effective- design and growth needs 1978 "Test of a task contingent
ness." Unpublished Ph.D. dis- strength: A test of a congruen- theory of work unit design and
sertation, Northwestern cy model." Academy of Man- performance." Paper presented
U niversity. agement Journal, 22: 223-240. at the Annual Meeting of the
Schoonhoven, Claudia B. Academy of Management, San
Mohr, Lawrence B.
1981 "Problems with contingency Francisco, August.
1971 "Organizational technology
theory: Testing assumptions 1985 "The concept of fit in contin-
and organizational structure."
hidden within the language of gency theory." In Barry M.
contingency theory." Adminis- Staw and L. L. Cummings
Quarterly, 16: 444-459.
trative Science Quarterly, 26: (eds.), Research in Organiza-
1982 Explaining Organization Be-
349-377. tional Behavior, 7: 333-365.
havior. San Francisco: Jossey-
Greenwich, CT: JAI Press.
Bass. Stinchcombe, Arthur L.
1968 Constructing Social Theories.
New York: Harcourt, Brace and
World.
Van de Ven, Andrew H., and Diane Woodward, Joan
L. Ferry 1965 Industrial Organization: Theory
1980 Measuring and Assessing and Practice. London: Oxford
Organizations. New York: University Press.
Wiley.

Sage Publications, Inc., Johnson Graduate School of Management, Cornell University Administrative Science Quarterly

Uploaded by

Copyright:

Available Formats

Sage Publications, Inc., Johnson Graduate School of Management, Cornell University Administrative Science Quarterly

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sage Publications, Inc., Johnson Graduate School of Management, Cornell University Administrative Science Quarterly

Uploaded by

Copyright:

Available Formats

Alternative Forms of Fit in Contingency Theory

Author(s): Robert Drazin and Andrew H. Van de Ven

Sage Publications, Inc., Johnson Graduate School of Management, Cornell University

Structural contingency theory has dominated the study of

As Dubin (1976) stated, every theory is a contingency theory,

A contingency theory differs from other theories in the specific

514/Administrative Science Quarterly, 30 (1985): 514-539

The key concept in a contingent proposition is fit, and the

Definition Assumption: Fit is Bivariate interaction: Consistency analysis:

Test methods Correlation or Context-structure Deviations from

Definition Macro selection: Fit at Residual analysis: Fit Equifinality: Fit is

Test methods Variables subject to Residuals of context- Relationship among

515/ASQ, December 1985

pirical results. These three different approaches to fit are

SELECTION, INTERACTION, AND SYSTEMS

Many early structural contingency theories were in fact con-

It is unclear whether to conclude that this research did not

Recently, however, natural selection and managerial selection

The managerial selection argument extends this approach and

516/ASQ, December 1985

Macro-rules tend to be imposed on micro-units in two ways:

Organizations limit the discretion of subunits by adopting a set

Future developments of the selection approach to fit in con-

A second interpretation of fit is that it is an interaction effect of

517/ASQ, December 1985

Figure 1. Example of typical interaction hypothesis of environmental

higher performing organizations than for lower performing

These mixed results may be due to many methodological

518/ASQ, December 1985

Several researchers have proposed a deviation-score approach

Figure 2 displays this form of analysis graphically. Organization

Studies that adopt the selection and interaction definitions of

Recently, a systems approach to contingency theory has

519/ASQ, December 1985

Low ORGANIZATIONAL High

Figure 2. (a) Deviation of organizations A and B from context structure

The systems approach emphasizes the need to adopt multi-

520/ASQ, December 1985

For example, Child (1977: 175), addressing the design di-

Similarly, Khandwalla (1 973) showed that internal consistency

A system analysis approach to fit is graphically presented in

In summary, the systems approach maintains that two basic

521/ASQ, December 1985

Contours of Decreasing Performance

ORGANIZATIONAL STRUCTURE -DIMENSION 1

Figure 3. A graphic representation of the systems approach to fit: an ideal

definition of fit are to identify the feasible set of organizational

Unique and Complementary Information

The selection approach to fit might also be combined with the

522/ASQ, December 1985

If the interaction approach does detect fit, but only among

Examining multiple approaches to fit in contingency studies

TASK-CONTINGENCY THEORY OF WORK-UNIT DESIGN

In this paper, empirical tests of the three approaches to fit are

The three approaches to fit are compared by examining the

523/ASQ, December 1985

Table 2 shows the underlying pattern of structure and process

The discretionary mode is a program for managing tasks that

524/ASQ, December 1985

p < .05, p < .01.

Not significant; p < .02; *0p < .001.