Forms of contingency fit in management accounting research—a critical review

Description
Contingency literature in the accounting control area has for a long time been criticized for being fragmentary and
contradictory as a result of methodological limitations. A review of 10 articles in the strategy-MAS area adds to this
picture by showing that many different forms of fit have been used, and that very few researchers fully acknowledge the
difficulties of relating these forms to each other.

Forms of contingency ?t in management accounting
research—a critical review
Jonas Gerdin
a,
*, Jan Greve
b
a
Department of Business Administration (ESI), O
¨
rebro University, S-701 82 O
¨
rebo, Sweden
b
Jo¨nko¨ping International Business School, Jo¨nko¨ping University, PO Box 1026, SE-551 11 Jo¨nko¨ping, Sweden
Abstract
Contingency literature in the accounting control area has for a long time been criticized for being fragmentary and
contradictory as a result of methodological limitations. A review of 10 articles in the strategy-MAS area adds to this
picture by showing that many di?erent forms of ?t have been used, and that very few researchers fully acknowledge the
di?culties of relating these forms to each other. As a result, some researchers claim that their ?ndings are contradictory
when this is not necessarily the case, while others incorrectly argue that their results are strongly supported by previous
studies.
#2002 Elsevier Ltd. All rights reserved.
Introduction
Contingency-based research in the accounting
control ?eld has a long tradition (Chapman, 1997;
Chenhall, 2003). The continuous stream of
empirical articles signals the importance and
vitality of this research area. However, the state of
the art of this research ?eld has been questioned.
It has been argued that various factors such as
disparate de?nitions of variables, insu?cient data
and underspeci?ed models have resulted in a frag-
mentary and contradictory theory (Dent, 1990;
Fisher, 1995; Galunic & Eisenhardt, 1994; Lang-
?eld-Smith, 1997; Otley, 1980). We argue that, in
addition to those shortcomings, attention should
also be paid to the way the concept of ?t has been
applied. Our argument is that this stream of
research does not di?er from contingency research
in general, where many di?erent forms of ?t have
been used and where researchers have not always
been aware of the implications of their choice on
theory building and testing (see for example,
Schoonhoven, 1981; Venkatraman, 1989). In
addition, since some of the conceptualizations of
?t used seem not to be comparable (Drazin & Van
de Ven, 1985; Govindarajan, 1988), apparently
contradictory or supportive results may have to be
reinterpreted. One purpose of this paper is to add
to the limited knowledge in this area by examining
which theoretical forms of ?t have been used in
the strategy-MAS literature. A second purpose is
to review critically whether comparisons made
between ?ndings that are based on di?erent forms
of ?t are valid. The strategy-MAS area was selec-
ted since it represents a stream of research in
which various forms of ?t have been used.
The paper ?rst provides a classi?catory frame-
work in which di?erent forms of contingency ?t are
outlined. The extent to which the di?erent forms
can be related to each other is also discussed. Next,
0361-3682/02/$ - see front matter # 2002 Elsevier Ltd. All rights reserved.
doi:10.1016/S0361-3682(02)00096-X
Accounting, Organizations and Society 29 (2004) 303–326
www.elsevier.com/locate/aos
* Corresponding author.
E-mail addresses: [email protected] (J. Gerdin), jan.
[email protected] (J. Greve).
10 works in the strategy-MAS literature with var-
ious approaches to ?t are brie?y described and
classi?ed according to the framework. Next, refer-
ences made in the literature are analysed in terms of
whether the various forms of ?t used allow com-
parisons to be made. Finally, the paper is summar-
ized and some conclusions are presented.
A classi?catory framework for mapping di?erent
forms of contingency ?t used in strategy-MAS
research
The framework used to classify the di?erent
forms of contingency ?t we have found in the
strategy-MAS literature has a hierarchical struc-
ture (see Fig. 1). At the top level, the two forms of
?t represent two con?icting paradigms. At lower
levels, the dichotomies represent alternative ways
of modelling or measuring ?t. Examples of statis-
tical techniques that have been used within each
theoretical form of ?t are depicted at the bottom.
While the distinctions made on the four levels
have been discussed previously in the literature,
they have not, to our knowledge, been combined
into one framework.
A ?rst distinction is made between a Cartesian
approach and a Con?guration approach. This
division originates from a debate between advo-
cates of a traditional ‘‘structural contingency the-
ory’’ and their critics. The traditionalists argue that
?t between context and structure is a continuum that
allows frequent, small movements by organizations
from one state of ?t to another (see e.g. Donaldson,
1996). By contrast, analysts advocating a Con?gur-
ation approach hold that there are only a few states
of ?t between context and structure, with organiza-
tions having to make ‘‘quantum jumps’’ from one
state of ?t to another (Meyer, Tsui, & Hinings, 1993;
Miller & Friesen, 1984; Mintzberg, 1983).
The next distinction derives from Drazin and
Van de Ven’s (1985) division between a Congruence
Fig. 1. A classi?catory framework for mapping di?erent forms of contingency ?t used in strategy-MAS research.
304 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
approach and a Contingency approach. The for-
mer assumes that only the best-performing orga-
nizations survive and therefore can be observed.
Hence, the research task explores the nature of
context-structure relationships without examining
whether they a?ect performance. With a Con-
tingency approach, it is assumed that organiza-
tions may have varying degrees of ?t. Thus, the
researcher must show that a higher degree of ?t is
associated with higher performance. Next, we
make a distinction between moderation and media-
tion
1
(Luft & Shields, 2003; Shields & Shields,
1998), two commonly used models in the strategy-
MAS area. The former speci?es that the e?ects of
an independent variable on a dependent variable
are a function of a moderating variable, while the
latter speci?es the existence of a signi?cant inter-
vening mechanism between an independent vari-
able and the dependent variable (Venkatraman,
1989). Finally, the concept of ?t is divided into two
groups based on whether it depicts the strength of a
relationship between variables, or its form (see e.g.
Hartmann & Moers, 1999; Venkatraman, 1989).
Cartesian vs. Con?guration approach
The fundamental di?erence between the Carte-
sian and Con?guration forms of ?t is in their
dominant modes of enquiry (Meyer et al., 1993).
Cartesian research is characterized by reduction-
ism while Con?gurational research takes a holistic
view. The two approaches lead to divergent opi-
nions about what constitutes ?t and how ?t is
attained (see Table 1).
With a Cartesian approach, the focus will be on
how single contextual factors a?ect single struc-
tural attributes and how these context-structure
pairs a?ect performance (Drazin & Van de Ven,
1985).
2
It is assumed that a limited number of
factors o?er general explanations of organiza-
tional structure. Hence, Donaldson (1996, p. 11)
asserts that the approach ‘‘generalizes across
many di?erent types of organizations and national
settings’’. Furthermore, according to Donaldson
(1996, p. 8), contextual as well as structural factors
are de?ned as continuous variables and ‘‘?ts
between them are also continua, there being many
points of ?t’’ (see Table A in Fig. 2). The numer-
ous points of ?t allow the ?rm continually to
adapt structure as a response also to minor chan-
ges in context.
The holistic view, held by the Con?guration
school, opposes partial analyses of context and
structure variables. Relationships can only be
understood if many contextual and structural
variables are analysed simultaneously (Drazin &
Van de Ven, 1985). Hypothetically, an almost
in?nite number of combinations seem possible.
However, according to theory, there is a rather
limited set of system states (con?gurations) to
which most ?rms can be assigned (Miller & Frie-
sen, 1984). Hence, the research task is to ‘‘identify
the feasible set of organizational structures and
processes that are e?ective for di?erent context
con?gurations and to understand which patterns
of organizational structure and process are intern-
ally consistent and inconsistent’’ (Drazin & Van de
Ven, 1985, pp. 521–522). Table B in Fig. 2 illus-
trates the Con?guration idea of ?t as system states.
Furthermore, it is argued that piecemeal changes of
structure are avoided since they ‘‘will often destroy
the complementarities among many elements of
con?guration’’ (Miller, 1986, p. 263). However,
when the cost of being out of step becomes large
enough, organizations make quantum jumps,
Table 1
Cartesian and Con?guration understanding of ?t
Cartesian approach Con?guration approach
Variables Few Many
Relations Continuous, general
across contexts
System states, context-
speci?c
Change Continuous, incremental Episodic, quantum jumps
1
For the sake of simplicity, the mediation category also
includes bivariate models, i.e. where it is assumed that a uni-
directional causal relationship exists between two variables.
For example, in many early Congruence studies, ?t was con-
ceptualized as a statistically signi?cant correlation coe?cient
between the context and structure.
2
It should be noted that the Cartesian approach does not
reject the use of multivariate models (see e.g. Donaldson, 1996).
Nevertheless, most studies in the strategy-MAS area have
applied statistical models implying bivariate or sharply cir-
cumscribed multivariate relationships. This will be evident
when we describe some of the empirical research undertaken
(see Classifying strategy-MAS studies).
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 305
thereby substituting one system state for another
(Meyer et al., 1993; Miller, 1982).
It is beyond the purpose of this paper to address
the question of whether the Cartesian or the Con-
?guration approach has the best predictive valid-
ity.
3
However, it is important to stress that results
(empirical or theoretical) based on one school of
thought should not be validated by comparisons
with those of the other. In fact, even if applied to
the same empirical data, the two approaches may
yield very di?erent results. To illustrate, consider a
case where high-performing Prospectors (Miles &
Snow, 1978) bene?t from high values on the three
MAS attributes ‘‘results monitoring’’, ‘‘report fre-
quency’’ and ‘‘external scanning’’. Assume further
that deviation from this optimal MAS con?gur-
ation is negatively correlated with performance. In
Table A of Fig. 3, the MAS pro?les of three com-
panies (A, B and C) have been plotted. Company
A performs best since its MAS ?ts perfectly with
the strategy, and company B performs better than
company C (since the total distance from the ideal
con?guration is less for B than for C).
Table B of Fig. 3 depicts the Cartesian
approach, where it is assumed that the impact of
each MAS element on performance can be
examined independently. The correlations in the
?rst and second charts correspond with theory; i.e.
‘‘results monitoring’’ and ‘‘report frequency’’,
respectively, are positively related to performance.
However, the expectation that ‘‘external scan-
ning’’ is positively associated with performance is
not veri?ed. It is company C that distorts the pic-
ture. Notwithstanding that C has the optimal
score for ‘‘external scanning’’, its poor match on
the other two attributes results in the lowest per-
formance over the sample. Accordingly, Con?gur-
ation researchers would conclude that ‘‘external
scanning’’ plays an important role among pro-
spectors while Cartesian researchers would draw
the opposite conclusion.
Although this example shows that the two
approaches may occasionally yield corresponding
results,
4
results based on one school of thought
should not be validated by means of the other.
After all, if theory predicts that performance is the
result of consistency between multiple contingent
and structural characteristics, it is not adequate to
investigate the e?ect on the performance of indivi-
dual parts. It is also hard to reconcile the idea of ?t
as a continuous line in a multidimensional space
with the idea of ?t as a limited number of discrete
system states. In our view, the Cartesian and the
Con?guration school therefore represent competing
rather than complementary approaches to ?t.
Fig. 2. Example of typical relationships between context and structure, according to the Cartesian (Table A) and Con?guration
(Table B) approaches, respectively.
3
It should be noted though that the Cartesian approach
o?er a higher degree of speci?city in the functional forms of ?t
compared to the Con?guration approach. With the latter
approach, certain variables are said to ?t together without
describing a precise form. We are indebted to one anonymous
reviewer for pointing this out. The di?erence is re?ected in the
statistical methods used within the two approaches (cf. classi?-
catory scheme in Fig. 1).
4
In fact, corresponding results require that deviations from
the ideal pro?le are proportional, i.e. depart equally on all attri-
butes that constitute the con?guration (such as Firm B in the
example earlier).
306 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
Congruence vs. Contingency approaches
In its simplest form, contingency theory suggests
that an organization’s structure is contingent upon
contextual factors such as environment, strategy
and size. According to Drazin and Van de Ven
(1985), this form of contingency theory represents
a Congruence approach insofar as it is assumed
that structure depends on context without any
examination of whether this relationship a?ects
performance. There is no need to test the link
with performance, since it is (implicitly) assumed
that ?t is the result of a natural selection process
that ensures that only the best-performing orga-
nizations survive to be observed at any point of
time. Accordingly, the research task is to identify
contextual variables that a?ect structure, and to
explore the nature of these context–structure
relationships.
The absence of performance in the models has
been considered as a ?aw because ‘‘signalling
survival of the ?ttest is too crude a proxy for
performance’’ (Pennings, 1992, p. 274). As a
result, contingent models have been developed
where ‘‘a conditional association of two or more
independent variables with a dependent outcome
is hypothesized’’ (Drazin & Van de Ven, 1985, p.
514).
5
In the Contingency approach, ?t is thus
understood as a positive impact on performance
due to certain combinations of context and struc-
ture. Accordingly, it is assumed that high-per-
forming as well as low-performing ?rms do exist as
a result of more or less successful combinations of
context and structure. The research task is then to
explain these variations in performance in terms of
interaction e?ects between context and structure.
The Congruence and the Contingency approach
may be regarded as two irreconcilable ideas about
?t. However, as Drazin and Van de Ven (1985)
pointed out, they may also be understood as two
states of subunit discretion that may exist within
organizations. Their argument is that management
normally limit the discretion of subunits by estab-
lishing ‘‘switching rules’’ that take contextual fac-
tors into consideration when controlling certain
Fig. 3. Comparison between results based on the Con?guration approach (Table A) and the Cartesian approach (Table B).
5
As will be shown in Classifying strategy-MAS studies,
more recent strategy-MAS studies based on the Congruence
approach actually use contingent models (see e.g. Bouwens &
Abernethy, 2000; Perera, Harrison, & Poole, 1997). The fun-
damental di?erence compared to the Contingency approach is,
however, that in the latter approach the researcher is not pri-
marily interested in examining how variables interact to explain
MAS design, but in showing that some combinations are rela-
ted to higher performance compared to others.
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 307
structural dimensions. As an example, Drazin and
Van de Ven argue that organizations tend to
structure routine production units in a ‘‘system-
ized mode’’, whereas R&D units are structured in
a ‘‘developmental mode’’. Since these switching
rules are imposed on the subunits, no variety in
terms of organizational design or performance in a
certain context should be expected. Accordingly,
?t should be analysed as a Congruence relationship
between these particular variables. However, some
structural characteristics are di?cult or impossible
to control through the development of switching
rules. Accordingly, these structural dimensions
should show a broader range of variance between
units, re?ecting the particularistic control of the
subunit. Drazin and Van de Ven (1985, p. 517)
conclude: ‘‘Only these variables should interact
with context to explain variations in perfor-
mance’’. The proper way of analysing ?t between
these variables is thus the Contingency approach.
Regardless of whether Congruence and Con-
tingency are understood as opposing views on ?t
or merely as a matter of variation in subunit dis-
cretion, they do represent two very di?erent forms
of ?t. As a result, it is far from certain that high
congruence between context and structure also
would indicate ?t when performance is included in
the research model. In fact, the opposite would be
more probable. As an illustration, consider a sur-
vey study examining the relation between business
strategy and MAS design. Strategy is a univariate
variable ranging from a pure Defender strategy to
a pure Prospector strategy (Miles & Snow, 1978).
MAS design is measured with respect to ‘‘scope’’
ranging from narrow to broad (Chenhall & Mor-
ris, 1986). Furthermore, assume that Prospectors
are best supported by a broad scope MAS while a
narrow MAS best serves Defenders. From a Con-
gruence angle of view, one would hypothesize that
the higher the commitment to a Prospector’s
strategy, the higher the reliance on broad scope
MAS information. Table A in Fig. 4 depicts a sit-
uation where ?rms in general have adapted their
MAS to strategy in the way theory predicts. Conse-
quently, there is no reason to suspect any signi?cant
variations in performance due to mis?t between
MAS design and strategy. Tables B and C show the
expected performance level across di?erent levels of
MAS scope
6
. Of course, there may be some varia-
tion in performance in reality. The crux of the mat-
ter is, however, that there is no way to predict such
variations from the information given in Table A,
since the underlying theory (implicitly) presumes
that only the best-performing organizations survive.
Now, assume that no association exists between
strategy and MAS design (see Table D in Fig. 4).
In other words, there is no ?t in terms of Con-
gruence. Interestingly enough, however, the
absence of association may actually increase the
probability of ?nding the Contingency form of ?t.
If theory holds (i.e. Prospectors and Defenders
bene?t from broad and narrow scope MAS infor-
mation, respectively), we would expect to ?nd high-
performing ?rms along a diagonal from bottom-left
to upper-right in Table D. Departures from the
optimal designs would result in lower performance.
(The indices denote the level of organizational
e?ectiveness, where low numbers represent high
performance, and high indices denote the opposite.)
Tables E and F illustrate the expected relationship
between MAS scope and organizational perfor-
mance for Defenders and Prospectors, respectively.
Since the relationship between MAS design and
performance di?ers between the two strategies, a
Contingency form of ?t is prevalent.
This example illustrates that, if applied to the
same set of variables, there is no reason to expect
the two forms of ?t to coincide.
7
On the contrary,
6
For illustrative purposes, the ?rms are divided into two
subgroups representing Defenders (D), and Prospectors (P),
respectively.
7
Although a Cartesian approach has been used in the example
presented, the distinction between Congruence and Contingency
forms of ?t is relevant also when a Con?guration approach is
adopted. For example, Miller and Friesen (1984) applied a Con-
gruence approach when they argued that owing to the natural
selection mechanism, most companies could be categorized into a
relatively small number of groups where the ?rms in each group
had similar structures and processes. By contrast, Drazin and Van
de Ven (1985) regarded the Con?guration approach (they called it
a Systems approach) as having a Contingency form of ?t. That is,
many organizational designs exist, but only some of them are
high-performers, in that they have a structure and process that
match the contextual setting and are internally consistent.
Accordingly, ?t was conceptualized as the degree of adherence to
the ideal designs (i.e. the high-performers). The fundamental
assumption is that the more the organization deviates from the
Ideal type, the more negative the e?ect on performance.
308 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
?t conceptualized as Congruence implies that there
is little room for alternative solutions. Accord-
ingly, we should not expect to ?nd a Contingency
form of ?t since it requires that less e?ective designs
also exist (if not, it would be impossible to show
that deviations from the ideal design are related to
lower performance). In fact, the prerequisites under
which the two forms of ?t can be expected to concur
are very speci?c. First, a ‘‘su?cient’’ number of the
?rms must have adapted their MAS to strategy (if
not, ?t in terms of Congruence cannot be identi?ed).
Second, in at least one strategy group, deviations
from the optimal MAS design must be ‘‘su?ciently
large’’ to cause variations in performance (if not, no
Contingency ?t can be found).
Moderation vs. mediation models
Cartesian-type relationships between variables
can be modelled in many di?erent ways. In the
strategy-MAS literature, the two most frequently
used variants are the moderation approach and the
mediation approach.
With a moderation approach, it is assumed that
the impact of an independent variable on the
dependent variable is contingent on the level of a
third variable, the so-called moderator (see Table A
in Fig. 5). That is, the underlying theory speci?es
that the third variable moderates the e?ect that
the independent variable has on the dependent
variable (Luft & Shields, 2003; Venkatraman,
1989). For example, di?erent business strategies
employed by ?rms may moderate the degree by
which a certain MAS design a?ects performance.
In the literature, a common way to test whether
a variable has a moderating e?ect is to use a
moderated regression analysis (MRA) (Hartmann
& Moers, 1999; Venkatraman, 1989). A typical
format is:
Y ¼
0
þ
1
X
1
þ
2
X
2
þ
3
X
1
X
2
þ"
where Y is a dependent variable, X
1
is an inde-
pendent variable, X
2
is a moderator, and X
1
X
2
is
the interaction e?ect (i.e. the moderating e?ect
that X
2
has on the relationship between X
1
and
Fig. 4. Relationship between a Congruence (Table A–C) and a Contingency (Table D–F) form of ?t.
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 309
Y).
8
Fit is prevalent when the interaction coe?-
cient b
3
di?ers signi?cantly from zero.
Notwithstanding that the moderator is modelled
to be a cause of the dependent variable in the
regression model above, an important assumption
underlying moderation forms of ?t is that the
moderator has ‘‘nonsigni?cant, bivariate relation-
ships with both the independent and dependent vari-
ables’’ (Shields & Shields, 1998, p. 51, emphasis
added). That is, the moderator is not theoretically
related with either the dependent or the indepen-
dent variable. If this prerequisite is not ful?lled,
the moderation form of ?t does not give an accu-
rate picture of the ‘‘true’’ relationships between
variables. In these cases, alternative models are
needed.
An alternative type of model found in the strat-
egy-MAS area is the mediation model (see Table B
in Fig. 5). In contrast to a moderation model, it
allows that some variables (depicted as X
2
in
Table B), apart from being contributors Y, may
also be dependent on other variables (X
1
). Fit
exists when the impact of X
1
(e.g. strategy) on Y
(e.g. performance) operates through X
2
(e.g. MAS
created by strategy). In path analysis—a com-
monly used statistical technique for testing media-
tion forms of ?t—?t is depicted as a statistically
signi?cant indirect e?ect.
The moderation and mediation types of models
illustrated above represent two di?erent theoretical
forms of ?t.
9
Both models may be valid, but in a
particular situation, only one model can give the
true picture. To illustrate, consider a situation
where MAS design (X
2
) is unrelated to the strategy
employed (X
1
). In this case, the mediation model is
non-valid (and accordingly, if used anyway, no ?t
would be indicated). However, in the same sit-
uation, the moderation model may very well indi-
cate ?t (since strategy may moderate the e?ect that
MAS design has on performance). The same
argument applies to the opposite situation, i.e.
where strategy really a?ects MAS design. Since
MAS design (by de?nition) cannot be depicted as
a moderator, the moderation model no longer
gives an accurate description of the situation. Only
the mediation model can be applied in order to
?nd out whether ?t exists or not.
In conclusion, since the moderation and mediation
forms of ?t have fundamentally di?erent theoretical
meanings, results based on one of the models can-
not be validated with results obtained from the
other. In other words, the MAS cannot con-
currently play both the role of a moderator (and
thus be independent of strategy) and the role of a
mediating variable (and thus be dependent on
strategy).
Strength vs. form
Within the Cartesian moderation approach, rela-
tionships between variables have been analysed in
terms of their strength and form (Hartmann &
Moers, 1999; Venkatraman, 1989). The two alter-
natives refer to di?erent statistical methods, but
Fig. 5. Moderation (Table A) and mediation (Table B) forms of ?t.
8
Note that an alternative theoretical interpretation of the
interaction term is possible. In addition to amplifying the e?ect
of X
1
on Y, X
2
may have an e?ect of its own—a so-called
independent-variable interaction model (Luft & Shields, 2003).
That is, if X
1
is nonexistent, X
2
would still have an impact on Y
(as opposed to the moderation model where a nonsigni?cant
relationship between X
2
and Y is postulated).
9
In order to simplify, only three-variable models are
contrasted.
310 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
they also represent di?erent theoretical meanings
of ?t. For the sake of simplicity, only examples
drawn from the Contingency approach are used to
illustrate the di?erences. However, the arguments
are equally valid in a Congruence setting.
When researchers, using a moderation model,
claim that the predictive ability of MAS on per-
formance di?ers across di?erent strategies, this
proposition re?ects the strength of the modera-
tion. The hypothesis is often tested by means of
subgroup correlation analysis. In each strategic
subgroup, elements of the MAS are correlated
against performance. The hypothesis of ?t is sup-
ported when statistically signi?cant di?erences
exist in the value of the correlation coe?cient
across the strategic groups. For example, consider
a case where the correlation between the use of
broad scope information and organizational per-
formance is signi?cantly higher among Pro-
spectors (represented by a ‘‘narrow’’ cloud in
Table A of Fig. 6) compared to Defenders (repre-
sented by a ‘‘wide’’ cloud in Table A).
When researchers argue that the impact of MAS
on performance di?ers across di?erent strategies,
they are making a statement about the form of the
moderation. Moderated regression analysis is a
commonly used statistical technique for testing
form type hypotheses (see Moderation vs. media-
tion models earlier). Fit is prevalent when statisti-
cally signi?cant di?erences in slope exist between
strategic groups. For example, it may be found that
the reliance on broad scope MAS information has
a signi?cantly more positive e?ect on performance
among Prospector type ?rms compared to Defen-
der type ?rms (see Table B of Fig. 6).
Similarities in terms of moderation form and
analysis method (subgroup analysis) may give the
impression that it is possible to ‘‘translate’’ results
from a strength setting to a form setting and vice
versa. However, we argue that the two forms of ?t
may yield consistent results, but there is no reason
to expect that they should. To illustrate, consider
the four situations depicted in Fig. 7. A compar-
ison between the scatter diagrams in Table A and
B indicates a di?erence in strength. That is, the
correlation between MAS design and performance
is signi?cantly higher in Table B (a ‘‘narrow’’
cloud) compared to Table A (a ‘‘wide’’ cloud).
This means that there is a di?erence in the pre-
dictive power of MAS design on performance
between subgroups. However, and this is impor-
tant, the diagrams also show that a certain change
in MAS design does not necessarily have a greater
impact on performance in high-correlation groups
compared to low-correlation groups (as illustrated
by the equal slopes of the appropriate regression
lines for the two subgroups). Furthermore, a
strength form of ?t may also exist when the impact
of MAS design on performance is negligible (cf.
the ‘‘?at’’ slopes in Tables A and B, respectively).
A comparison between Tables B and D illus-
trates the opposite situation. In this case, the sta-
tistically di?erent slopes indicate that the e?ect of
MAS design on performance is di?erent for dif-
ferent subgroups (i.e. form). However, no di?er-
ence in strength exists, since MAS design appears
Fig. 6. A strength (Table A) and a form (Table B) variant of moderation ?t.
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 311
to predict performance in both subgroups equally
well.
10
Finally, a comparison between the scatter
diagrams in Tables A and D shows a situation
involving a combination of strength and form
moderations since both the slope of the regression
line and the correlation between MAS design and
performance are di?erent for the two subgroups.
Overall, Fig. 7 shows that the strength and form
types of moderation ?t may yield consistent results
(see comparison 5), but there is no reason to expect
that they should (as illustrated by comparisons 1,
2, 3 and 4). Hence, results based on a strength
variant of ?t should normally not be related to
those based on a form variant. One exception
exists, though. Possible di?erent signs of the
correlation coe?cients in a strength analysis
include some information about the form of the
relationship. However, this is a very crude
measure since the sign only signals whether the
relationship is positive or negative. Again, the
regression lines may be almost ?at, i.e. the e?ect of
MAS on performance is practically nil in both
subgroups.
To summarize, many di?erent forms of ?t have
been used in the strategy-MAS literature. Some
conceptualizations are ‘‘mutually exclusive’’ inas-
much as their theoretical meanings are so di?erent
that results based on one formmust not be related to
results of the other (cf. the Cartesian and Con?gur-
ation forms of ?t). Other conceptualizations may be
complementary—?t in one form reduces the prob-
ability of ?t in another form(cf. the Congruence and
Fig. 7. Moderation as strength and form may yield consistent results (case 5), or not (cases 1–4).
10
Note that this conclusion holds also in situations when
MAS design and performance are signi?cantly positively cor-
related in one subgroup (say, r=0.5) and to the same extent
negatively correlated in another subgroup (r=À0.5). The
argument is that the sign of the correlation coe?cient does not
contain information on the predictive power (i.e. MAS design
predict performance equally well in both subgroups). See also
Hartmann and Moers (1999, pp. 297–298).
312 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
Contingency approaches). Finally, there are forms
of ?t that may result in corresponding results.
However, there is no reason to expect that they
should (cf. the strength and form variants of
moderation ?t).
Classifying strategy-MAS studies
In this section, 10 important strategy-MAS
papers will be described and classi?ed into the
framework of di?erent forms of ?t developed
earlier. The list is not exhaustive. Rather, these
studies were selected to highlight the great number
of conceptualizations of ?t used in the literature.
The studies are dealt with in chronological order.
Khandwalla (1972)
The main purpose of Khandwalla’s study was
‘‘to outline some of the competitive conditions
under which sophisticated management controls
are more extensively utilized and those under
which they are less extensively utilized’’ (Khand-
walla, 1972, p. 275). Based on a correlation ana-
lysis, he found that competition increases the use
of sophisticated control systems. Furthermore,
when analysing subcategories of competition, he
found that the control variables in general were
strongly correlated with product competition, but
only weakly with price competition. Hence, he
concluded that product competition has a larger
e?ect on the use of control systems than has price
competition.
In our view, Khandwalla’s study clearly repre-
sents a Cartesian approach, in that ?t is depicted
as a continuum (conceptualized as correlation
between variables), and in that the MAS is exam-
ined part by part. The absence of a performance
variable implies a Congruence variant of the Car-
tesian approach. That is, it is assumed that only
high-performing ?rms survive to be studied, and
the research task is thereby reduced to exploring
what form the relations between strategy and
MAS take. The subgroup correlation analysis
used and the concluding discussion indicates a
moderation form depicting the strength of the
relationship between strategy and MAS.
Miles and Snow (1978)
The pioneering book by Miles and Snow forms
the basis for many subsequent studies. It not only
provided MAS analysts with a framework for
conceptualizing di?erent strategies, it also con-
tributed signi?cantly to our understanding of the
relationship between strategy and MAS. Four
strategic types were identi?ed. The Defender type
has a narrow product market domain, a cost e?-
cient technology and a specialized and formalized
structure. The MAS has an e?ciency focus with
detailed and close control. By contrast, the Pro-
spector type continually develops new markets/
products by emphasizing ?exibility in its technol-
ogy and structure. Its MAS is result-oriented at
lower organizational levels. The third strategic
type, Analysers, is a unique combination of the
Defender and Prospector types. As a result, it has
a dual organization and MAS: centralized and
functional, with an e?ciency focus for the stable
sphere, and an organic structure and e?ectiveness
focus for the dynamic sphere. Finally, the Reactor
type is characterized by inconsistency in the way it
responds to change in its environment, which
implies lower performance and instability.
Although Miles and Snow themselves did not use
the term Con?guration approach, their writings
clearly indicate that their framework meets the
requirements. First, they argue that organizations
need internal consistency between multiple con-
tingencies and multiple structural characteristics if they
are to perform well. For example, they conclude that:
. . .the Defender aligns itself with a particular
portion of the overall environment and man-
ages the internal interdependencies created by
its form of alignment. This adjustment pro-
cess produces a unique con?guration of
domain, technology, structure, and process.
(Miles & Snow, 1978, p. 47)
Second, Miles and Snow claim that, at least
three e?ective organizational types represent
stable and discrete system states (equilibriums);
Three of these strategic types—which we
label the Defender, the Analyser, and the
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 313
Prospector— are ‘‘stable’’ forms of organiza-
tions. That is, if management chooses to pur-
sue one of these strategies, and designs the
organization accordingly, then the organiza-
tion may be an e?ective competitor in its
particular industry over a considerable period
of time. On the other hand, if management
does not choose to pursue one of these
‘‘pure’’ strategies, then the organization will
be slow to respond to opportunities and is
likely to be an ine?ective performer in its
industry. We shall call these latter organiza-
tions Reactors and argue that they are essen-
tially ‘‘unstable’’. (Miles & Snow, 1978, p. 14)
The presence of the fourth group of organiza-
tions, Reactors, is interesting insofar as it indicates
that deviations from the three e?ective types will
result in lower organizational performance.
Hence, if management aims at changing the orga-
nization’s strategic focus, it should prefer a quan-
tum jump between strategic types rather than
frequent small movements, since intermediate
forms are less e?ective. The Reactor type is also
interesting because it indicates that lower-per-
forming organizations exist in reality, and thus
can be the subject of investigation. Thus, in con-
clusion, the work of Miles and Snow should be
classi?ed as a Contingency form of the Con?gur-
ation approach (cf. Doty, Glick, & Huber, 1993;
Zahra & Pearce, 1990).
Simons (1987)
11
The major purpose of Simons’ study was to test
the hypothesis that control system attributes di?er
between Prospectors and Defenders. The hypoth-
esis was supported by means of a logit model
12
where strategy was modelled as the dependent
variable, and 10 management control factors and
industry dynamism (i.e. dynamism in the technical
and economic environment) were independent
variables. It was also found that predictability was
improved when the ?rms were divided into sub-
groups based on their size.
In a second step, pro?t performance data were
correlated with each control system attribute for
Prospectors and Defenders, respectively. The
results were summarized in the following way:
High performing Prospector ?rms seem to
attach a great deal of importance to forecast
data in control systems, setting tight budget
goals, and monitoring outputs carefully. For
Prospectors, cost control is reduced. In addi-
tion, large ?rms appear to emphasize frequent
reporting and the use of uniform control sys-
tems, which are modi?ed when necessary.
Defenders, particularly large ?rms, appear to
use their control systems less intensively. In
fact, negative relationships were noted
between performance and attributes such as
tight budget goals and output monitoring.
Defenders emphasized bonus remuneration
based on the achievement of budget targets
and tended to have little change in their con-
trol systems. (Simons, 1987, p. 370)
Simons pointed out that high-performing Pro-
spectors’ reliance on budget control was in accor-
dance with Khandwalla’s (1972) observation that
?rms competing with product development use
elaborated budgeting techniques. However, the
lack of a positive correlation between Defenders’
cost control and performance contradicted the
results of Miles and Snow (1978), who found that
the Defender strategy calls for detailed cost infor-
mation. (See also Dent (1990) and Lang?eld-
Smith (1997) for a discussion about this contra-
dictory result).
Simons used two di?erent forms of ?t in his
study. The test of the ?rst hypothesis suggests a
Congruence form of the Con?guration approach.
The argument is that the logit model examines
whether the overall pattern of control system
attributes di?ers between two strategic types.
However, the model does not test whether adher-
ence to the ideal pro?les results in higher perfor-
mance, which implies a Congruence form of ?t.
By contrast, the correlation analyses where each
control variable is examined independently and
11
Data collection and analysis was based on a questionnaire
and semi-structured interviews. In our description, we will
focus on the quantitative parts of the analysis.
12
Somewhat simpli?ed, a logit model may be characterized
as a regression model with a binary dependent variable.
314 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
performance is included in the research model
implies a Contingency variant of the Cartesian
approach. That is, it is assumed that also less
e?ective organizations exist to be observed, and
the research task is to show that deviations from
the ideal design are related to lower performance.
The use of subgroup correlation analyses suggests
that this part of the analysis should be classi?ed as
a moderation form of ?t depicting the strength of
the relationships.
Govindarajan (1988)
Govindarajan examined the relations between
strategy and several administrative control
mechanisms used by chief executives to control
Strategic Business Units (SBUs). Porter’s (1980,
1985) strategic types (‘‘Low-cost’’ and ‘‘Di?er-
entiation’’) were used to classify SBUs with
respect to strategy. The administrative mechan-
isms were ‘‘budget evaluation style’’, ‘‘decen-
tralization’’ and ‘‘locus of control’’. In three of
the hypotheses, an interaction e?ect of competi-
tive strategy and of each of the control variables
on performance was expected. Moderated regres-
sion analyses showed that ‘‘high managerial
internal locus of control and low emphasis on
meeting a budget are associated with high per-
formance in SBU’s employing a strategy of dif-
ferentiation’’ (Govindarajan, 1988, p. 843).
However, no support was found for the hypothe-
sized interaction e?ect of strategy and decen-
tralization on performance.
Based on the argument of Drazin and Van de
Ven (1985), the three bivariate interaction-type
propositions were complemented with a fourth
hypothesis stating that an appropriate match
among all three control variables with strategy
would be associated with high SBU e?ectiveness.
In order to test this proposition, the Euclidean
distance between each SBUs actual score on the
three control variables and the scores of their
(theoretically derived) ideal type (Low-cost or
Di?erentiation) was measured. Next, the corre-
lation between the deviation score and perfor-
mance was calculated. The coe?cient was
signi?cantly negative, thereby providing support
for the fourth hypothesis.
The ?rst three hypotheses assessed ?t in terms of
interaction e?ects between strategy and each
administrative mechanism on performance. This
suggests a Contingency form of the Cartesian
approach. A moderation variant of ?t was used
where the form of the relationship between vari-
ables (and not the strength) was measured. By
contrast, the fourth hypothesis examined the
impact of the administrative mechanisms, taken as
a set, and SBU strategy on SBU e?ectiveness.
Accordingly, the Contingency form of the Con-
?guration approach was also used.
Abernethy and Guthrie (1994)
The hypothesis tested by Abernethy and
Guthrie stated that ‘‘broad scope information will
have a more positive e?ect on performance in
prospector-type ?rms than in defender-type ?rms’’
(p. 56). It was tested and supported by means of a
moderated regression analysis.
As a complement, Abernethy and Guthrie
assessed the match between strategy, MAS, and
performance by means of a subgroup correlation
analysis. According to the authors, the statistically
signi?cant di?erences in correlation coe?cients
between strategic groups provided further support
for the hypothesis stated earlier.
Both the verbal and statistical analysis clearly
indicates that a Contingency form of the Cartesian
approach was used. Furthermore, since the model
speci?es that the impact of MAS design on per-
formance is contingent on SBU strategy,
13
it can
be classi?ed as a moderation variant of ?t. The
moderating e?ect was analysed in terms of both
‘‘form’’ (cf. the moderated regression analysis) and
strength (cf. the subgroup correlation analysis).
Abernethy and Lillis (1995)
Abernethy and Lillis formulated three research
questions: (1) Do ?rms committed to manufactur-
ing ?exibility use performance measures that de-
emphasize e?ciency measures? (2) Do they increase
their reliance on integrative liaison devices? (3) Is
13
Actually, this interpretation is somewhat uncertain, since
Fig. 1 models MAS design as the moderating variable.
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 315
performance enhanced when the manufacturing
performance measurement system or structural
arrangements are adapted to manufacturing ?ex-
ibility? In order to answer questions (1) and (2),
bivariate correlation analysis was used. As expected,
?exibility correlated negatively with the use of e?-
ciency-based performance measures, and positively
with the use of integrative liaison devices. The third
question was tackled by ?rst dividing the sample into
two subgroups: ?exible ?rms and non-?exible ?rms.
After that, subgroup correlation coe?cients between
each management control variable and performance
were estimated. For e?ciency-based measures, there
was a signi?cant di?erence between ?exible and non-
?exible ?rms. A di?erence was also found with
respect to the ‘‘integrative liaison device’’ element
(correlation was higher in the ?exible group). How-
ever, it was not statistically signi?cant.
In a complementary analysis (presented in a
footnote and in Appendix C), the authors assessed
whether the match between ?exibility and each
control element in?uenced performance by ?tting
two moderated regression equations. As expected,
the interaction e?ect between ?exibility and e?-
ciency-based measures was negative and sig-
ni?cant. Surprisingly enough, the same was also
true with respect to the interaction e?ect of ?ex-
ibility and integrative liaison devices on perfor-
mance. However, this latter result was not
statistically signi?cant.
In our view, a Cartesian approach was chosen,
and questions (1) and (2) clearly indicate a Con-
gruence form of ?t, depicting the strength of the
relationships. The third question suggests a Con-
tingency form of ?t where ?exibility exercises a
moderating e?ect on performance. The moderating
e?ect was analysed in terms of both strength (cf.
the subgroup correlation analyses) and form (cf.
the moderated regression analyses).
Perera et al. (1997)
The study of Abernethy and Lillis (1995) was
followed up and extended by Perera et al. (1997).
Two hypotheses were tested: (1) increasing custo-
mer focus in manufacturing strategy is associated
with increasing use of non-?nancial measures
(NF), and (2) increasing use of NF is associated
with enhanced performance in customer-focused
?rms. Customer focus was broken up into
Advanced Management Practices (AMP) and
Advanced Manufacturing Technology (AMT). A
bivariate correlation analysis showed strong posi-
tive correlations between NF and AMP and
AMT, respectively (Hypothesis 1). Additionally, a
regression equation was ?tted to data where the
NF was modelled as the dependent variable, while
AMP, AMT and the interaction term
AMPÂAMT were depicted as independent vari-
ables. The signi?cantly positive interaction coe?-
cient was interpreted as additional support for the
?rst hypothesis, and was also in accordance with
the results of Abernethy and Lillis (1995). The
second hypothesis was tested by means of a
regression equation where performance was the
dependent variable. Independent variables were
AMT, AMP, NF, and second- and third-order
interaction e?ects. The third-order interaction
e?ect (between AMT, AMP and NF) was non-
signi?cant and the second hypothesis was rejected.
Both hypotheses imply a Cartesian form of ?t.
The ?rst hypothesis indicates a Congruence form
of ?t and it is analysed with respect both to
strength (correlation analysis) and form (regression
analysis). The second hypothesis models a Con-
tingency situation where the form of the moderat-
ing e?ect was tested.
Chong and Chong (1997)
Chong and Chong presented a model where the
MAS scope was modelled as an intervening vari-
able between strategy and performance (Hypoth-
esis 1), and between perceived environmental
uncertainty (PEU) and performance (Hypothesis
2). A path analytic technique was used to test the
two hypotheses. It was demonstrated that the
direct e?ects of strategy and PEU on performance
were non-signi?cant, while the indirect e?ects (via
MAS) were signi?cant. Accordingly, it was con-
cluded that broad scope MAS information is an
important antecedent of SBU performance. This
?nding was considered to be in accordance with
that of Abernethy and Guthrie (1994).
To conclude, the Cartesian approach was used,
and since the dependent variable was performance,
316 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
this study must be classi?ed as a Contingency var-
iant. The reliance on a path analytic technique
implies a mediation (as opposed to moderation)
form of ?t.
Chenhall and Lang?eld-Smith (1998)
In this article, the authors investigated ‘‘the way
in which management accounting practices com-
bine with management techniques, under various
strategic priorities, to enhance performance’’
(Chenhall & Lang?eld-Smith, 1998, p. 243). It was
expected that higher performing ?rms employing a
di?erentiation strategy would bene?t from certain
management techniques and management
accounting practices (Hypothesis 1) while higher
performing ?rms employing a low price strategy
would bene?t from other techniques and practices
(Hypothesis 2). The authors adopt a systems view
since ‘‘it was expected that various combinations
of strategies, management techniques and man-
agement accounting practices would combine in
mutually supportive ways to enhance organiza-
tional performance’’ (ibid, p. 251). A cluster ana-
lysis was performed and six clusters emerged. They
were ranked according to performance. Three of
the clusters had characteristics that were consistent
with an emphasis on product di?erentiation. Since
performance was positively related with most of
the management techniques and accounting prac-
tices that were expected in this type of ?rm, it was
argued that the overall results provided support for
Hypothesis 1. Two of the clusters exhibited char-
acteristics consistent with the price competition
strategy. Some support was provided for Hypoth-
esis 2. However, it was also concluded that the
hypotheses were too limited in some respects. Some
techniques and practices employed by ?rms with a
di?erentiating strategy were also important among
well-performing ?rms with a low-price strategy.
The systems approach chosen in this study
implies a Con?guration form of ?t. The compar-
isons between the empirically grounded clusters
and the two ideal strategic types (see Hypotheses 1
and 2) were quite tentative, but nevertheless imply
a pro?le deviation analysis. In other words, this
study represents a Contingency form of the Con-
?guration approach.
Bouwens and Abernethy (2000)
Bouwens and Abernethy formulated two
hypotheses: (1) There is a positive relation
between customization and interdependence, and
(2) there is a positive indirect relation between
customization and the MAS dimensions (i) scope,
(ii) integration, (iii) aggregation, and (iv) time-
liness, acting through departmental inter-
dependence. The ?rst hypothesis was tested and
supported by means of a bivariate correlation
analysis. In order to assess the direct and indirect
relations (Hypothesis 2), a path model was devel-
oped. Since the MAS consisted of four separate
dimensions, four path coe?cients were com-
puted—one for each dimension. The impact of
strategy on MAS scope proved to be negligible—
and this was true both for direct and indirect
e?ects. The authors pointed out that this result
contradicted the ?ndings of earlier research,
including Abernethy and Guthrie (1994). The
other three path coe?cients of the indirect e?ects
(via interdependence) were generally higher than
the coe?cients of direct e?ects. Accordingly, the
authors concluded that ‘‘managers recognize the
importance of receiving more sophisticated infor-
mation to manage the interdependencies that stem
from the pursuit of customization’’ (Bouwens &
Abernethy, 2000, p. 234).
In conclusion, a Cartesian approach was chosen,
and the fact that the path analysis did not include
performance implies that a Congruence variant of
the mediation form of ?t was applied.
Analysis of research evidence
In Table 2, the 10 strategy-MAS papers have
been arranged into the classi?catory framework
developed earlier. The arrows signify that refer-
ences have been made between ?ndings, including
comparisons made within the scope of one paper.
For example, Simons (1987) relates his ?ndings
based on the Con?guration/Congruence form of ?t
to his own ?ndings based on the Cartesian/Con-
tingency form of ?t, which in turn are related to the
?ndings of Khandwalla (1972) and Miles and Snow
(1978), respectively. All interrelated references
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 317
constitute a ‘‘cluster’’, which is identi?ed by a
capital letter (A–D). The clusters form the basis
for the following discussion, and we start by ana-
lysing the three comparisons made within Cluster
A.
Simons (1987)
Simons’ article consists of two separate analyses
that are undertaken on the same data set. In Part
1, he concluded that the overall MAS design dif-
fers between Prospectors and Defenders. In Part 2,
he found that in many cases the correlations
between MAS elements and performance di?er
between the strategic types.
In our view, Simons assumed that there is a
connection between the two parts (see Connection
A1 in Table 2) when he argued that if ‘‘Defender
and Prospector strategies require di?erent control
systems, then the correlations between control
system attributes and economic performance
should be di?erent for Defender ?rms and Pro-
spector ?rms’’ (p. 360). In other words, a sig-
ni?cant Congruence relationship implies that a
Contingency relationship should exist. However,
as was argued before (see Congruence vs. Con-
tingency approaches), there is no reason to believe
that the results based on the two forms of ?t
should concur. The argument is as follows. Part 1
of the study (i.e. the logit model) indicates that
?rms have adapted their MAS to their strategic
commitment, i.e. MAS designs are quite homo-
geneous within each strategy group. Accordingly,
it is less likely that di?erences in MAS design can
Table 2
Summary of results
a
318 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
explain variations in the performance of ?rms (cf.
Part 2). Our expectations about the circumstances
under which the two forms of ?t will concur would
thus be formulated in the opposite direction to
those of Simons.
That is, the stronger the association found when
testing for Congruence forms of ?t, the less the
association between MAS and performance
should be expected. Hence, the corresponding
results between the two results in Simons’ study
only imply that a ‘‘su?cient’’ number of ?rms
have adapted their control structure to strategy at
the same time as deviations from the optimal
design of control elements are ‘‘su?ciently’’ large
to cause variations in performance. This argument
also applies to the seemingly corresponding results
in Simons (1987) and Khandwalla (1972) (Con-
nection A2).
By linking the two parts of the study together
(A1), Simons actually crosses two borders in the
classi?catory scheme. Apart from relating the
Congruence and Contingency forms of ?t to each
other, he also compares results based on a Con-
?guration form of ?t (Part 1) with those based on a
Cartesian form (Part 2). Furthermore, this latter
border is crossed once again when Simons com-
pares his ?ndings from Part 2 with those of Miles
and Snow (1978) (see connection A3 in Table 2).
The pitfalls associated with this course of action
are best illustrated by an example.
Based on Miles and Snow (1978), Simons
expected a positive correlation between cost con-
trol and ?rm performance among ?rms pursuing a
Defender strategy. However, no correlation was
found, and Simons considered the result as con-
tradictory. In our view, however, there may be no
contradiction at all. The argument is that Simons
applied a Cartesian approach (where the research
task is to identify the particular contingent factor or
factors to which each particular MAS element needs
to ?t if the organization is to performwell), while the
work of Miles and Snow is based on a Con?guration
approach (where it is assumed that performance is
the result of the degree of consistency among all
parts of an organization). Accordingly, a Defender
that has an appropriate match among most dimen-
sions of the ?rm’s total control package may be
reasonably high-performing, even when its ‘‘score’’
on the cost control dimension deviates sub-
stantially from the ideal Defender con?guration.
Likewise, some Defenders may have an optimal
‘‘score’’ on the cost control variable and yet their
performance is low because cost control is not in
harmony with other parts of the control package
(see Fig. 2 where company C is low-performing
despite a high value on the ‘‘External scanning’’
variable). In conclusion, the Cartesian mode of
enquiry used by Simons is not suitable for identi-
fying ‘‘system e?ects’’ on performance that are
implied in the Con?guration approach.
Govindarajan (1988)
Govindarajan used two Contingency forms of ?t
in his study: a moderation variant of the Cartesian
form, and a Con?guration form (see Connection 3
in Table 2). As has been discussed in depth in the
former section, ?ndings based on a Cartesian form
of ?t may not be consistent with those based on a
Con?guration form. However, in contrast to many
other researchers who have used several forms of
?t in one and the same study, Govindarajan
recognized this fact and explored the possible
complementarities between them.
A comparison of the results of the bivariate
and systems approaches yielded several useful
insights that could not have been tapped if I
had placed sole reliance on either one of the
approaches. First, the systems results indicate
that all three mechanisms of concern are cri-
tical in strategy implementation. Sole reliance
on bivariate interaction might have led to the
incorrect conclusion that the decentralization-
strategy ?t has no implications for perfor-
mance. Second, the bivariate interaction did
detect ?t for two administrative mechanisms,
thereby suggesting that budget evaluative
style and locus of control are more salient
predictors of performance than decentraliza-
tion. Thus corporate executives might focus
?rst on matching budget evaluative style and
locus of control to strategy. Such an insight
would not have been available from a sole
reliance on the systems approach. Finally,
using the systems approach, I assigned equal
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 319
weights to deviations to all the administrative
mechanisms. Using the bivariate results might
have suggested that deviations along budget
evaluative style and locus of control be
assigned higher weights. (Govindarajan,
1988, p. 843)
Govindarajan’s study is interesting in that it
explicitly explores how di?erent approaches to ?t
can be related to each other. However, his argu-
ments that bivariate analyses can be used to assign
higher weights to certain dimensions in the pro?le
deviation analysis should, in our opinion, be
questioned. First, Govindarajan does not recog-
nize (at least not explicitly) many of the di?culties
that arise when results based on moderated
regression analyses are related to those based on
pro?le deviation analyses (see the Appendix for a
discussion on this topic).
Second, and even more importantly, Govindar-
ajan’s interpretation of the ?ndings represents a
Cartesian (i.e. reductionistic) view of con?gur-
ations. That is, the e?ect of di?erent dimensions
on performance is regarded as ‘‘a sum of the
parts’’ although, from a holistic perspective, the
e?ect is determined by interrelationships among
all dimensions that constitute the con?guration.
Accordingly, some ?rms pursuing di?erentiation
may be highly decentralized and yet low-perform-
ing if they fail to combine decentralization with
appropriate values of the two remaining control
variables. Other di?erentiators may have a less
decentralized structure, but owing to an overall
adaptation of the structure to strategy, perfor-
mance may still be rather high. In both these
situations, a bivariate analysis would identify
decentralization as a bad predictor of perfor-
mance. However, from a Con?guration point of
view it is possible that in a particular situation (i.e.
with given values on contingency and structure
variables), a certain change in decentralization
would a?ect the overall performance more than an
equivalent alteration of any other control variable.
In another situation, the impact on performance
of a corresponding change in decentralization may
be negligible. From a Con?guration perspective, it
is therefore not possible to assign weights to indi-
vidual variables, because the impact that variables
have on performance is a function of all relations
between contingency and structure variables in the
particular situation.
Abernethy and Guthrie (1994)
Table 2 shows that Abernethy and Guthrie
make the same type of comparisons between
strength and form variants of moderation ?t as
Abernethy and Lillis (1995) (Connections C1 and
D2). However, since more data is presented in the
Abernethy and Lillis paper, the discussion of
whether this type of comparison is valid will be
dealt with later, where the Abernethy and Lillis
paper is discussed.
Chong and Chong (1997)
Chong and Chong found ‘‘the impact of broad
scope MAS information acts as an intervening
construct between SBU strategy and SBU perfor-
mance’’ (p. 270). According to them, this ?nding
was in accordance with the Abernethy and
Guthrie (1994) study (see Connection C2 in
Table 2). However, in our view, it is di?cult the-
oretically to relate the two forms of ?t to each
other, as there is no reason to believe that MAS
should be related to strategy based on Abernethy
and Guthrie’s results. On the contrary, the mod-
eration form of ?t used by Abernethy and Guthrie
is based on the assumption that the moderator (in
this case, strategy) is not theoretically related to
either the independent variable (MAS) or the
dependent variable (performance), but a?ects the
relationship between them (Luft & Shields, 2003;
Hartmann & Moers, 1999; Shields & Shields,
1998). Or, put in the opposite way, if a causal
relationship between strategy and MAS is hypo-
thesized, the moderated regression analysis does
not depict the expected relationship between vari-
ables, and, accordingly, another model (e.g. a path
model) should have been used.
Bouwens and Abernethy (2000)
The only explicit reference to other strategy-
MAS studies made in Bouwens and Abernethy is
that to Abernethy and Guthrie (1994), as depicted
320 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
in Connection C3 in Table 2. Using their own
words, Bouwens and Abernethy claim that ‘‘the
scope dimension of a ?rm’s MAS is not important
for operational decision-making, which is contrary
to the general expectation . . . and ?ndings of ear-
lier research’’ (p. 234). Two objections may be
raised against this seemingly contradictory result.
First, the implicit assumption that strategy and
MAS are unrelated (as modelled in Abernethy and
Guthrie’s moderation form of ?t) is incompatible
with the assumption that a causal relation exists
between the two variables (as depicted in their
own mediation model). Accordingly, the non-sig-
ni?cant result found in Bouwens and Abernethy
actually is in line with the ?ndings of Abernethy
and Guthrie, since it gives some support to the
implicit assumption that there is no relation
between strategy and MAS.
Second, it is hazardous to relate empirical
results based on Contingency assumptions to those
based on Congruence assumptions. Again, ?t in
one setting suggests non-?t in the other [see
Congruence vs. Contingency approaches and the
earlier section on Simons (1987)]. Hence, the
non-signi?cant results in Bouwens and Aber-
nethy’s study obtains support from the study of
Abernethy and Guthrie, not the converse.
Abernethy and Lillis (1995)
Like many other studies (e.g. Abernethy &
Guthrie, 1994; Govindarajan, 1988; Simons,
1987), Abernethy and Lillis’ (1995) paper includes
several forms of ?t. Since the possibilities of relat-
ing Congruence forms of ?t to Contingency forms
have already been thoroughly discussed [see the
earlier section on Simons (1987)], Relation D1 in
Table 2 will not be further discussed here from a
Congruence vs. Contingency point of view. How-
ever, we argue that there is yet another reason why
the two forms of ?t should not be compared with
each other. Consider the following argument. In a
?rst step, the authors hypothesized that there
exists a causal relationship between ?exibility and
MAS design: ‘‘the role of accounting in measuring
performance depends on the strategic commit-
ments of the SBU’’ (p. 242). In a second step,
however, the authors adapted a research model
that assumes that ?exibility moderates the e?ect of
MAS design on performance (but is itself theore-
tically independent of both MAS design and per-
formance). Of course, both models cannot be valid
descriptions of the situation, and consequently
consistent results should not be expected.
We now shift our focus and concentrate on the
two moderation forms of ?t used, i.e. strength and
form, respectively (Connection D2 in Table 2). In
their primary analysis, a subgroup correlation
analysis suggested that performance is enhanced
when the MAS is adapted to the ?rm’s strategic
commitment (strength). In a complementary ana-
lysis, Abernethy and Lillis formulated two regres-
sion equations, each including an interaction term
(form). In our view, the two forms of ?t should not
be related to each other since there is no reason to
believe that they should yield consistent results
(see Strength vs. form). The strength analysis
depicts whether the predictive power of MAS on
performance di?ers between subgroups (mani-
fested by signi?cantly di?erent correlation coe?-
cients), while the form analysis examines whether
the impact of MAS on performance di?ers
between subgroups (manifested by signi?cantly
di?erent regression slopes). It is true that the sign
of the correlation coe?cient contains some infor-
mation about form, in that it shows the direction
of the relationship (which Abernethy and Lillis
point out). However, correlation coe?cients are
less reliable as indicators of form since they do not
tell how much changes in strategy a?ect the rela-
tionship between MAS and performance (the
impact may be negligible).
Perera et al. (1997)
Strictly speaking, Perera et al. (1997) carried out
three di?erent statistical tests, which may imply
that three theoretical meanings of ?t are prevalent
in the paper [i.e. (i) the bivariate correlation ana-
lysis implies a strength variant of Congruence ?t,
(ii) the regression model where the dependent
variable is the MAS suggests that the form of a
Congruent relationship is depicted, and (iii) the
regression model where performance is the depen-
dent variable indicates a Contingency variant of ?t
where the form of the association is modelled].
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 321
Connection D3 suggests that a questionable
reference between two forms of ?t (strength and
form) has been made. However, since the corre-
lation analysis is used exclusively to test whether
the associations between variables are in the pre-
dicted direction, it can be considered as a ?rst step
in the analysis of the form of the relationship, later
extended in the regression model. For the same
reason, the reference made to Abernethy and Lillis
(Connection D4 in Table 2) is appropriate since
they too use correlation coe?cients when they
comment on the form of the relation.
The connection between the Congruence and the
Contingency forms of ?t (see Connection D5 in
Table 2) is more dubious. It seems that the authors
expect that the clear relationship between strategy
and MAS (found in the former analysis) should be
associated with an equally clear link to perfor-
mance (i.e. ?rms that have a proper match
between strategy and MAS perform better than
those that do not). Again, if a Congruence test
indicates a strong association between variables
(as predicted by theory), this suggests that most
?rms have an appropriate match. Accordingly, the
chance of ?nding signi?cant interaction e?ects
between strategy and MAS on performance is low,
and vice versa (see the earlier section on Simons
(1987)].
Summary of ?ndings, discussion and conclusions
The purpose of this study was to identify the
di?erent forms of ?t used in the strategy-MAS
area and to explore the extent to which they have
been, and can be, related to each other. Our
review of 10 studies in the area shows that three
studies in the sample were, rightfully, isolated
‘‘islands’’ either because at the date of publication
they could not be related to other strategy-MAS
studies, or because their speci?c form of ?t could
only be related with di?culty to the forms used in
earlier studies (see Chenhall & Lang?eld-Smith,
1998; Khandwalla, 1972; Miles & Snow, 1978). In
the remaining seven articles, either ?ndings were
related to other studies based on alternative forms
of ?t, or several forms of ?t were used within the
scope of the one article.
In two articles, comparisons were made between
a Cartesian and a Con?guration form of ?t
(Govindarajan, 1988; Simons, 1987). In Govin-
darajan (1988), the author discussed the di?erent
meanings of ?t, although we have argued that he
did not fully recognize the problems associated
with comparing the two forms. In our opinion, it
makes no sense to relate results based on a Carte-
sian form of ?t to results based on a Con?guration
form of ?t, since the di?erence between the two
forms is paradigmatic. It stems from the con?ict
between a reductionistic and a holistic under-
standing of organizations and their environment.
We found four articles where comparisons had
been undertaken between a Congruence and a
Contingency form of ?t. Overall, it was presumed
that ?t in one form implies ?t in the other form. In
one article (Simons, 1987), this assumption was
veri?ed, but in two articles (Bouwens & Aber-
nethy, 2000; Perera et al., 1997), it was found that
the two forms of ?t did not coexist. In the fourth
article (Abernethy & Lillis, 1995), the results were
mixed. In our view, comparisons between Con-
gruence and Contingency studies may be valuable.
As Drazin and Van de Ven (1985) pointed out,
some dimensions of organizational design may be
adapted to context as the result of macro-switch-
ing rules imposed by top management, while other
dimensions are the subject of subunit discretion.
However, if applied to the same set of variables, we
should not expect the two forms to coincide. Rather,
the opposite result is consistent and even likely.
In the present study, we found four examples
where comparisons were made between a modera-
tion and a mediation form of ?t (Abernethy &
Lillis, 1995; Bouwens & Abernethy, 2000; Chong
& Chong, 1997; Perera et al., 1997). Since these
forms represent competing ways of modelling
relationships between variables, only one can be
correct in a particular situation. That is, either
strategy and MAS are theoretically related con-
cepts (as depicted in a mediation model), or they
are not (as implied by a moderation model). If
both models give signi?cant results, this should be
interpreted as contradictory, not the converse.
In two of the articles, comparisons were under-
taken between a strength and a form variant of ?t
(Abernethy & Guthrie, 1994; Abernethy & Lillis,
322 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
1995). Strength was conceptualized as a signi?cant
di?erence in correlation coe?cients between MAS
and performance across di?erent strategic types,
while form was depicted as a signi?cant interaction
term in a moderated regression analysis. In our
view, it makes no sense to relate results based on
one form to those of the other. The argument is
twofold. First, the theoretical implications are
very di?erent (see Strength vs. form). Second,
there is no reason to expect that results based on
the two forms of ?t should concur. Again, the sign
of the correlation coe?cient is an indicator of
form in that it depicts the direction of the rela-
tionship. However, it must be remembered that a
signi?cant correlation between variables does not
necessarily imply that the regression slopes di?er
signi?cantly from zero.
The evidence reviewed in this paper shows that
very di?erent conceptualizations of ?t have been
used and that very fewresearchers fully acknowledge
the di?culties of relating di?erent forms to each
other. As a result, some researchers claim that their
results are contradictory when this is not necessarily
the case, while others incorrectly argue that their
results are strongly supported by former studies.
In our view, the results suggest that future the-
ory building and testing in the accounting control
area would bene?t if the following two items were
considered. First, specify whether a Cartesian or a
Con?guration approach has been adopted. Our
argument is that the di?erences between these
approaches are paradigmatic and, accordingly,
references to the existing literature should only be
done within each school of thought. If both
approaches to ?t are included in one and the same
study, the research task should be to explore and
contrast the predictive power of each approach,
rather than to search for complementary informa-
tion. Second, within the scope of each paradigm,
future studies should discuss explicitly whether
and how the particular form of ?t used can be
related to studies that have adopted other forms.
Acknowledgements
We gratefully acknowledge the valuable com-
ments made by the anonymous reviewers and by
participants in workshops held at the Centre for
Empirical Research on Organizational Control
(CEROC) at O
¨
rebo University. Financial support
for this project was provided by O
¨
rebo University
and Jo¨ nko¨ ping International Business School
Appendix. Some potential pitfalls that arise when
results based on moderated regression analyses
are related to those based on pro?le deviation
analyses
The pitfalls are best illustrated by an example.
Assume (in accordance with Govindarajan, 1988)
that the ideal pro?les of control mechanisms for
the low-cost and the di?erentiation strategy sub-
groups can be depicted as illustrated in Fig. A1.
The low-cost group should have the lowest possi-
ble value on each control variable if the organiza-
tion is to perform well, while the di?erentiation
group bene?ts from the highest possible values.
Accordingly, if the theory is valid, an increased
value on any variable will be associated with lower
performance for ?rms competing with low costs,
while the opposite applies to ?rms competing with
di?erentiation. However, in a regression equation
with a multiplicative interaction term, ?t may be
indicated also when both groups bene?t from
higher (lower) values on each variable. The inter-
action term only shows that the reliance on a
Fig. A1. Ideal pro?les of control mechanisms for the low-cost
and di?erentiation subgroups.
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 323
certain control mechanism has a more positive
(negative) e?ect on performance in one of the
groups compared to the other. Graphically, this
e?ect can be depicted as signi?cantly di?erent
slopes of the regression coe?cients for each stra-
tegic group (see Table A in Fig. A2). In conclu-
sion, therefore, when we compare the situations in
Fig. A1 and Table A in Fig. A2, they must be
considered contradictory as far as low-cost ?rms
are concerned.
In order to show that the results based on the
moderated regression analyses correspond with
results based on the pro?le deviation approach,
we need to demonstrate that the interaction
e?ect is non-monotonic (i.e. for higher values of
the moderator, the control mechanism will posi-
tively (negatively) a?ect performance, and for
lower values, it will negatively (positively) a?ect
performance) (see e.g. Schoonhoven, 1981).
14
That is, ?rms competing with low costs should
be positively a?ected by lower values on each
control variable, while ‘‘Di?erentiators’’ are
negatively a?ected by lower values (see Table B
in Fig. A2).
Another potential pitfall that may arise when
results based on moderated regression analyses are
related to results based on pro?le deviation ana-
lyses is that the former only contains information
about changes in the relationship between variables,
while the latter contains information about the
optimal level on variables. Accordingly, as Hart-
mann and Moers (1999, p. 305) pointed out, ‘‘the
proper interpretation of a positive interaction
therefore is not that Y achieves the highest values
for the highest values of X
1
and X
2
, but that for
higher values of X
1
, X
2
there is a more positive
e?ect on Y. Hence, MRA cannot be used to test
expectations about the highest values of X
1
and X
2
for which Y will have the highest value’’. This is
illustrated in Fig. A3. In Table A, Y has the high-
est value when both X
1
and X
2
are high, while
panel B depicts a situation where Y has the highest
value when both X
1
and X
2
are low. However, in
both cases, the interaction e?ects are equal in
terms of direction and size.
Applied to our example, it may thus be the case
that ?rms that combine a low-cost strategy with
high values on each control variable actually have
higher values on performance (as depicted by
Subgroup 1 in panel B in Fig. A3) compared to
?rms that combine a di?erentiation strategy with
high values on each control variable (Subgroup 2).
Of course, this is contradictory to the expected
‘‘Low-cost’’ pro?le suggested in Fig. A1. Never-
theless, in a moderated regression analysis, a sig-
ni?cant non-monotonic interaction e?ect in the
predicted direction would be found.
Fig. A2. Monotonic (Table A) and non-monotonic (Table B) interaction e?ects.
14
Interestingly enough, Govindarajan’s hypotheses are
expressed in non-monotonic terms; e.g. ‘‘for SBUs employing a
strategy of di?erentiation, de-emphasizing budgetary goals
during performance evaluations is likely to be associated with
high SBU e?ectiveness. For SBUs employing a strategy of low
cost, emphasizing budgetary goals during performance eval-
uation is likely to be associated with high SBU e?ectiveness’’
(pp. 832–833). However, he does not explore whether the
interaction e?ects found are non-monotonic in the predicted
direction that, in turn, make his comparisons with the results of
the pro?le deviation analysis unsteady.
324 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326
References
Abernethy, M. A., & Guthrie, C. H. (1994). An empirical
assessment of the ‘‘?t’’ between strategy and management
information system design. Accounting and Finance, Novem-
ber, 49–66.
Abernethy, M. A., & Lillis, A. M. (1995). The impact of man-
ufacturing ?exibility on management control system design.
Accounting, Organizations and Society, 20, 241–258.
Bouwens, J., & Abernethy, M. A. (2000). The consequences of
customization on management accounting system design.
Accounting, Organizations and Society, 25, 221–241.
Chapman, C. S. (1997). Re?ections on a contingent view of
accounting. Accounting, Organizations and Society, 22, 189–
205.
Chenhall, R. H. (2003). Management control systems design
within its organizational context: ?ndings from contingency-
based research and directions for the future. Accounting,
Organizations and Society, 28, 127–168.
Chenhall, R. H., & Lange?eld-Smith, K. (1998). The relation-
ship between strategic priorities, management techniques and
management accounting: an empirical investigation using a
systems approach. Accounting, Organizations and Society, 23,
243–264.
Chenhall, R. H., & Morris, D. (1986). The impact of structure,
environment, and interdependence on the perceived useful-
ness of management accounting systems. The Accounting
Review, 61, 16–36.
Chong, V. K., & Chong, K. M. (1997). Strategic choices,
environmental uncertainty and SBU performance: a note on
the intervening role of management accounting systems.
Accounting and Business Research, 27, 268–276.
Dent, J. F. (1990). Strategy, organization and control: some
possibilities for accounting research. Accounting, Organiza-
tions and Society, 15, 3–25.
Donaldson, L. (1996). For positivist organization theory: prov-
ing the hard core. London: Sage.
Doty, H., Glick, W. H., & Huber, G. P. (1993). Fit, equi?n-
ality, and organizational e?ectiveness: a test of two con?g-
urational theories. Academy of Management Journal, 36,
1196–1250.
Drazin, R., & Van de Ven, A. H. (1985). Alternative forms of
?t in contingency theory. Administrative Science Quarterly,
30, 514–539.
Fisher, J. (1995). Contingency-based research on management
control systems: categorizing by level of complexity. Journal
of Accounting Literature, 14, 24–53.
Galunic, D. C., & Eisenhardt, K. M. (1994). Renewing the
strategy–structure–performance paradigm. Research in
Organizational Behavior, 16, 215–255.
Govindarajan, V. (1988). A contingency approach to strategy
implementation at the business level: integrating adminis-
trative mechanisms with strategy. Academy of Management
Journal, 828–853.
Hartmann, F. G. H., & Moers, F. (1999). Testing contingency
hypotheses in budgetary research: an evaluation of the use of
moderated regression analysis. Accounting, Organizations
and Society, 24, 291–315.
Khandwalla, P. N. (1972). The e?ect of di?erent types of com-
petition on the use of accounting controls. Journal of
Accounting Research, 10, 275–285.
Lang?eld-Smith, K. (1997). Management control systems and
strategy: a critical review. Accounting, Organizations and
Society, 22, 207–232.
Luft, J., & Shields, M. D. (2003). Mapping management
accounting: graphics and guidelines for theory-consistent
empirical research. Accounting, Organizations & Society, 28,
169–249.
Meyer, A. D., Tsui, A. S., & Hinings, C. R. (1993). Con?gura-
tional approaches to organizational analysis. Academy of
Management Journal, 36, 1175–1195.
Miles, R. E., & Snow, C. C. (1978). Organizational strategy,
structure, and process. New York: McGraw-Hill.
Miller, D. (1982). Evolution and revolution: a quantum view of
structural change in organizations. Journal of Management
Studies, 19, 131–151.
Miller, D. (1986). Con?guration of strategy and structure:
toward a synthesis. Strategic Management Journal, 7, 233–249.
Fig. A3. Interaction e?ects contain no information on the value of the dependent variable (a slightly revised version of Hartmann &
Moers, 1999: 306).
J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326 325
Miller, D., & Friesen, H. (1984). Organizations, a quantum view.
Englewood Cli?s: Prentice Hall.
Mintzberg, H. (1983). Structure in ?ves: designing e?ective
organizations. Englewood Cli?s: Prentice-Hall.
Otley, D. T. (1980). The contingency theory of management
accounting: achievement and prognosis. Accounting, Organi-
zations and Society, 4, 413–428.
Pennings, J. (1992). Structural contingency theory: a reapprai-
sal. In B. M. Staw, & L. L. Cummings (Eds.), Research in
organizational behavior, vol. 14. Greenwich CT: JAI Press.
Perera, S., Harrison, G., & Poole, M. (1997). Customer-focused
manufacturing strategy and the use of operations-based non-
?nancial performance measures: a research note. Accounting,
Organizations and Society, 22, 557–572.
Porter, M. E. (1980). Competitive strategy. New York: Free
Press.
Porter, M. E. (1985). Competitive advantage: creating and sus-
taining superior performance. New York: Free Press.
Schoonhoven, C. B. (1981). Problems with contingency theory:
testing assumptions hidden within the language of contingency
theory. Administrative Science Quarterly, 26, 349–377.
Shields, J. F., & Shields, M. D. (1998). Antecedents of partici-
pative budgeting. Accounting, Organizations and Society, 23,
49–76.
Simons, R. (1987). Accounting control systems and business
strategy: an empirical analysis. Accounting, Organizations
and Society, 12, 357–374.
Venkatraman, N. (1989). The concept of ?t in strategy
research: toward verbal and statistical correspondence.
Academy of Management Review, 14, 423–444.
Zahra, S. A., & Pearce, J. A. (1990). Research evidence on the
Miles–snow typology. Journal of Management, 16, 751–768.
326 J. Gerdin, J. Greve / Accounting, Organizations and Society 29 (2004) 303–326

doc_187012645.pdf