Accounting, Organizations and Society 38 (2013) 407–419

Contents lists available at ScienceDirect

Accounting, Organizations and Society

journal homepage: www.elsevier.com/locate/aos

Management control as a system or a package? Conceptual

and empirical issues
Isabella Grabner, Frank Moers ⇑
Maastricht University School of Business and Economics, European Centre for Corporate Engagement (ECCE), The Netherlands

a b s t r a c t

Despite the trend in investigating combinations of MC practices that form packages or sys-
tems, there is ambiguity about what is meant by a ‘‘control package’’ or ‘‘control system’’.
In this paper, we address questions related to MC as a package and as a system, both from a
conceptual and empirical perspective. The purpose of our study is to clarify a number of
issues, so as to guide future research in this area. We do not claim to have all the answers.
Rather we make a number of assumptions explicit to show when it is important to address
multiple MC practices simultaneously, why this is the case, and what the empirical impli-
cations are. Our research directly relates to the systems approach to contingency theory
and the literature’s struggle to clearly define its most fundamental concept, i.e., ‘‘internal
consistency’’. Our analysis provides a formal definition of ‘‘internal consistency’’ and shows
how it can be used to conceptualize MC systems, thereby giving contingency theorists the
necessary ingredient to move forward.
Ó 2013 Elsevier Ltd. All rights reserved.

Introduction independently of other MC practices and it is this indepen-

dence assumption that has been heavily criticized. As a re-
Contingency theory has a long tradition in management sponse, theorists have put forward the systems approach
accounting research. Historically Management Control to contingency theory and empiricists have begun to exam-
(MC) was studied, in the main, by examining single MC ine combinations of MC practices that form packages or
practices related to budgets, performance measures, incen- systems.
tives, etc.1 As is typically the case with any significant Despite the similarity of these responses, they have
stream of research, critical reviews of this contingency liter- developed into two literature streams that are quite inde-
ature have emerged over the last decades (e.g., Chapman, pendent of each other, leading to the current status of the
1997; Chenhall, 2003; Fisher, 1995; Hartmann & Moers, broader MC literature in which there is ambiguity about
1999; Otley, 1980). One of the major criticisms that has been what is meant by a ‘‘package’’ or ‘‘system’’. In this paper,
put forward relates to the literature’s reductionist approach. we address questions related to MC as a package and as a
Reductionism assumes that MC practices can be examined system, both from a conceptual and empirical perspective.
The purpose of our paper is to clarify a number of issues, so
as to guide future research in this area. We do not claim to
⇑ Corresponding author. Address: Maastricht University School of have all the answers. Rather we make a number of assump-
Business and Economics, Department of Accounting & Information tions explicit to show when it is important to address
Management, P.O. Box 616, 6200 MD Maastricht, The Netherlands. Tel.:
multiple MC practices simultaneously, why this is the case,
+31 433883753.
E-mail addresses: [email protected] (I. Grabner), and what the empirical implications are.
[email protected] (F. Moers). In their seminal paper, Drazin and Van de Ven (1985, p.
1
Examples of MC practices that are related to budgets, performance 521) state that the systems approach to contingency
measures, and incentives are the interactive use of financial budgets, the theory maintains that organizational design involves two
use of subjective performance evaluation, and the extent of variable pay,
basic choices: (1) selecting MC practices that match the
respectively.

0361-3682/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.aos.2013.09.002
408 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419

set of contingencies facing the organization, and (2) assur- systems approach. Given the importance of interdepen-
ing that the MC practices are internally consistent. We ar- dence for understanding ‘‘internally consistent’’, we pro-
gue that the problem with this approach is its lack of a vide a formal definition of interdependence between MC
definition of ‘‘internally consistent’’. To address this prob- practices and the two types of interdependencies that ex-
lem, we first separate two views of MC, i.e., MC as a pack- ist, i.e., complements and substitutes. In general terms,
age and MC as a system, based on whether the MC interdependence implies that the value of one MC practice
practices are interdependent, and we provide two distinct depends on the use of another MC practice, and vice versa.
working definitions for these views. In particular, we pro- It thus not only provides a formalization of when and why
pose that MC practices form a system if the MC practices the reductionist approach can be problematic, but also
are interdependent and the design choices take these what ‘‘internally consistent’’ captures.
interdependencies into account. In contrast, MC as a pack- Although our use of complementarity theory to define
age represents the complete set of control practices in ‘‘internally consistent’’ is new, the theory and insights are
place, regardless of whether the MC practices are interde- obviously not new and have been used in previous MC
pendent and/or the design choices take interdependencies studies. However, an analysis of these previous studies
into account. This implies that the MC package can be com- shows that there is no consistent interpretation of interde-
posed of a set of MC systems and/or of a set of independent pendence, and the terms complements and substitutes
MC practices addressing unrelated control problems. Based have been used for conceptualizing relations between con-
on these definitions, we subsequently point out that, if the trol practices other than interdependence as well.2 Thus,
purpose of a study is to examine relationships among MC despite the researchers’ awareness of the literature on com-
practices, then the view of MC as a package is ill-suited plementarity theory, these results further underscore the
as a starting point. To examine interdependencies it is nec- need for creating a clear understanding of what is meant
essary to take a MC as a system perspective. This also im- by interdependence and when to use the terms ‘‘comple-
plies that to investigate MC as a package – and aiming to go ments’’ and ‘‘substitutes’’.
beyond mere descriptions of observed MC clusters – it is When we move from the theory to empirics we show
crucial to first understand the various MC systems within that the conceptual issues naturally extend to the empiri-
the package. cal setting. We revisit the general theoretical framework
The conclusion that MC as a system is the correct per- to show the empirical consequences of varying some of
spective to examine multiple MC practices raises the fun- the underlying assumptions. We further discuss two differ-
damental question of what theory is useful from this ent empirical strategies aimed at examining interdepen-
perspective in order to put structure on what is meant by dencies among MC practices, as well as the cost and
‘‘internally consistent’’. We propose that complementarity benefits of these strategies. The difference between the
theory is suitable to address issues related to MC as a sys- methods resembles the difference between a contingency
tem because the theory explicitly addresses how a decision approach to fit, i.e., examining the performance effects of
maker tries to maximize ‘‘performance’’ by (simulta- a combination of a MC practice and context, and a congru-
neously) deciding on multiple choice variables. This obser- ence approach to fit, i.e., examining the association be-
vation per se is not new. In their paper on the disciplinary tween a MC practice and context. In this case, the
constraints on the advancement of knowledge, Merchant, ‘‘contingency’’ approach examines the performance effects
Van der Stede, and Zheng (2003) hint at similarities be- of a combination of MC practices, while the ‘‘congruence’’
tween the concepts of fit and complementarity and pro- approach examines the association between MC practices.
pose potential synergies of the economics and the More importantly, we show that, at the conceptual level,
accounting literatures when it comes to explaining the both approaches rely on the same type of ‘‘systems fit’’.
concept of fit. As Milgrom and Roberts (1995, p. 180) ar- We end our paper with providing explicit guidelines for fu-
gue: ‘‘the ideas of complementarities [. . .] give substance ture research that aims to examine the design of MC
to previously elusive notions such as ‘fit’ or ‘systems ef- systems.
fects’’’. However, it is important to note at this point that
‘‘fit’’ in contingency theory relates to how MC practices
fit with context, while ‘‘fit’’ in complementarity theory re- 2
We analyzed the line of argumentation in MC studies that rely on the
lates to how MC practices fit with each other. As such, terms complements and substitutes published during the period 2000–
complementarity theory does not give more substance to 2012 in the following seven accounting journals (in alphabetical order):
Accounting, Organizations and Society, Contemporary Accounting Research,
the common view of contingency fit. Rather, complemen- Journal of Accounting and Economics, Journal of Accounting Research, Journal
tarity theory gives more substance to what a system is of Management Accounting Research, Management Accounting Research, and
and thus also to how a system fits with context, both of The Accounting Review. Besides the conceptualization of a reciprocal
which are the core components of the systems fit to con- relationship between control practices proposed by Milgrom and Roberts
(1995), the terms have also been used in arguments related to (1) a
tingency theory.
unidirectional relation between MC practices that are sequential control
Using complementarity theory, we provide a simple choices (e.g., Indejejikian and Matêjka, 2012), (2) a relation between a MC
formal analysis of MC design to illustrate the decision mak- practice and an element of the control environment that is not a choice
ing process when confronted with having to decide on (e.g., Dekker, 2004), and (3) a holistic view on MC in which no relation
multiple MC practices. We use a general framework that between MC practices is taken into account, i.e., MC as a package (e.g.,
Widener, 2004). While we do not argue that these alternative views are
allows us to vary some of the underlying assumptions incorrect by themselves, they have nothing to do with interdependence of
and illustrate the consequences of different theoretical MC practices and MC systems. We elaborate on this issue in Section
perspectives, i.e., the reductionist approach and the ‘Implications and guidelines for future research’.
 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419 409

Our paper contributes to the literature in several ways. control processes that are complementary to the AIS’’. This
The first contribution is relatively straightforward. If exam- statement indicates that the usefulness of one MC practice
ining multiple MC practices at the same time is relevant depends on what other MC practices are in use. Thus, if an
and important, which we believe to be the case, then organization does not take the interdependence among MC
understanding when it is important to address multiple practices into account, it will forego some benefits or incur
MC practices simultaneously, why this is the case, and some unnecessary costs. Alternatively, if an organization
what the empirical implications are, is fundamental. We does take it into account, then the choice of one MC prac-
not only provide answers to these questions, but also pro- tice is related to the choice of other MC practices. Related
vide guidelines for future research in this area. Second, to this latter case, Chenhall (2003, p. 131) raises the con-
while the systems approach has clearly identified the prob- cern that:
lem with the reductionist approach, it has itself been
‘‘. . .if specific accounting controls are systematically
unsuccessful in addressing the problem. Drazin and Van
linked with other organizational controls, studies that
de Ven (1985, pp. 521–522) state that
exclude or do not control for these elements within
‘‘the tasks for theorists and researchers adopting the the research method may report spurious findings’’.
systems definition of fit are [. . .] to understand which
This statement indicates that whether or not MC prac-
patterns of organizational structure and process are
tices are related to each other does not only have concep-
internally consistent and inconsistent’’.
tual implications but also empirical implications. In the
context of the contingency literature discussed by Chenhall
Given that the literature has made no attempt to explic- (2003), it specifically implies that an observed relationship
itly define ‘‘internal consistency’’, progress has been slow between a contextual variable and one MC practice can be
at best. Management control research, and contingency re- spurious due to the contextual variable being related to a
search in particular, has more than other areas been highly second MC practice that is related to the first. We elaborate
self-reflective with ongoing debate and commentary on on such issues later in the paper in more detail.
how to improve the field (e.g., Bisbe, Batista-Foguet, & The papers by Otley (1980) and Chenhall (2003) are just
Chenhall, 2007; Chenhall, 2003; Chenhall & Moers, 2007; two examples where a hint at a system is made in terms of
Fisher, 1995; Hartmann & Moers, 1999; Luft & Shields, relationships among MC practices. Both quotes emphasize
2003; Otley, 1980; Van der Stede, Young, & Chen, 2005). the importance of accounting for potential interdepen-
Our study follows in this tradition. In particular, our anal- dence between MC practices for management control to
ysis provides a formal definition of ‘‘internal consistency’’ be most effective. We therefore propose that MC practices
and shows how it can be used to conceptualize MC sys- form a system if these practices are interdependent and
tems, thereby giving contingency theorists the necessary the design choice has taken the interdependencies into
ingredient to move forward. account.
The remainder of this paper is structured as follows. In A somewhat different view seen in the literature is the
Section ‘What is a MC system or package?’, we discuss two conceptualization of MC as a package. In a recent paper,
views of MC, i.e., MC as a package and MC as a system and Malmi and Brown (2008) directly confront the question
provide two distinct working definitions for these views. In of what is conceptually meant by a MC package by provid-
Section ‘Formal analysis of MC design’, we present our gen- ing a ‘‘conceptual typology’’. In essence, Malmi and Brown
eral theoretical framework and provide a formal analysis of (2008) assume that there is a MC package because we ob-
MC design, while in Section ‘From theory to empirics’ we serve, in real-life, that organizations have multiple MC
translate this framework to the empirical setting. In Sec- practices in place at the same time. They state that: ‘‘As a
tion ‘Implications and guidelines for future research’, we general conception, a management control systems (MCS)
provide explicit guidelines for future research and we con- package is a collection or set of controls and control sys-
clude in Section ‘Conclusion’. tems.’’ (Malmi & Brown, 2008, Footnote 1). Their typology
therefore provides a (non-exhaustive) list of MC practices
What is a MC system or package? that can be included in a package, each of them grouped
in a particular category, to guide future research. This per-
When examining MC as a system or as a package, the spective on management control does not rely on a partic-
first essential question is what is meant by these terms. ular theory that accounts for possible interdependencies
Numerous hints at this have been made in the accounting between control choices because the aim is to provide a
literature, albeit lacking clarity and consistency. From our more holistic view of the MC practices in place, not inter-
reading of the literature, these terms have often been used dependence among them.
interchangeably. However, some authors suggest signifi- The difference between a system of control choices and
cant differences between the two concepts. In this section, the notion of a control package is best described by Ferreira
we develop two working definitions of MC as a system ver- and Otley (2009, p. 276), who state that
sus a package based on the type of relation that exists be-
‘‘. . . it should be noted that it is not assumed that an
tween MC practices.
extant PMS will be coherent. Otley (1980) discussed
The existence and importance of MC interdependencies
control ‘packages’ rather than control ‘systems’ because
has been raised multiple times in the literature. In his sem-
he had found that they tended to be composed of sets of
inal paper, Otley (1980, p. 421) states ‘‘. . .what constitutes
loosely coupled elements. These were often designed
an appropriate AIS will be influenced [. . .] by the other
410 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419

and implemented by different people, in different parts and derivations simple, but nevertheless explain them in
of an organization, at different times’’. detail.
Assume that a decision maker, say a top executive of an
From this it becomes apparent that the notion of MC as organization, is confronted with the following decision
a package cannot be used interchangeably with the notion (maximization) problem:
of MC as a system, as the latter assumes conscious deci-
Max pðx; z; pÞ ¼ Max Vðx; zÞ  pT x ð1Þ
sions on the design of interdependent controls. In line with x x
Malmi and Brown (2008), we propose to view the MC
where pðx; z; pÞ represents the net-benefits to the decision
package of a given organization as containing the complete
maker, which solely for convenience we label ‘‘profit’’ in
set of MC practices in place, thus mirroring the organiza-
the remainder of the text; Vðx; zÞ is the pay-off (produc-
tion’s control environment. This implies that the MC pack-
tion) function relevant to the decision maker; x is a vector
age can be composed of a set of MC systems and/or of a set
of MC (choice) practices; z is a vector of contextual (prede-
of independent MC practices addressing unrelated control
termined) variables; and p is a vector of the ‘‘cost per unit’’
problems.3
of x, which represents the separate costs of adopting each
Using the proposed definitions of MC system and MC
MC practice.5
package, it also becomes apparent that once the aim is to
Thus, the decision maker aims to maximize p by mak-
investigate interdependencies between MC practices, the
ing decisions about x, which are under his control, given
MC as a package perspective is ill-suited to start addressing
z and p, which are predetermined. For now we interpret
such issues. To examine interdependencies it is necessary
z as a vector of organization characteristics, but this vector
to take a MC as a system perspective. Furthermore, this
can contain any variable that is predetermined from the
perspective matches the maintained assumption of the
perspective of making decisions about x. Maximization of
systems approach to contingency theory that organiza-
the objective function specified in Eq. (1) results in the
tional design choices involve selecting multiple MC prac-
following:
tices and assuring that these are internally consistent.
To address issues related to MC as a system, i.e., inter- V x ðx ; zÞ ¼ p ð2Þ
dependence/internal consistency, we rely on complemen-
tarity theory because this theory explicitly addresses how where the subscript refers to the (first) derivative of VðÞ to
a decision maker tries to maximize ‘‘performance’’ by x and the superscript ‘‘*’’ reflects the optimal decision. Eq.
simultaneously deciding on multiple choice variables. This (2) says nothing more than that the decision maker will
choice follows Merchant et al.’s (2003) call to exploit make design choices such that the marginal revenues
synergies between the economics and accounting litera- equal marginal costs, thus choosing a level of a particular
tures when it comes to explaining the concept of fit. In MC practice where the marginal returns of that practice
the next section, we therefore build on complementarity equal the marginal adoption cost of that practice.
theory to provide insights into the decision making process Instead of characterizing the optimal MC design choices
when confronted with having to decide on multiple MC in a more detailed fashion, we discuss two design ap-
variables.4 proaches, each of which makes different assumptions
about Eq. (1). The two design approaches that we formally
analyze are the two different approaches to contingency
Formal analysis of MC design
theory, i.e., the reductionist approach and the systems ap-
proach. Of importance to the subsequent section is the fol-
We provide a simple formal analysis of MC design to
lowing general characterization:
illustrate several issues related to the choice of multiple
MC practices. We provide a general framework that allows @p
us to illustrate different theoretical perspectives by making ¼ V xi ðx; zÞ  pi ð3Þ
@xi
specific assumptions within this framework. By doing this,
we make it explicit what assumptions some theories or Eq. (3) reflects the impact of a change in MC practice xi
perspectives rely on and to what extent these assumptions on profit and shows that the value of this MC practice is
are incompatible. We keep all our formal characterizations potentially affected by all MC practices, including itself
(non-linearity), by the contextual variables, and the sepa-
3
rate adoption costs. In the following scenarios, we assume
It is possible that a set of interdependent MC practices are designed
taking these interdependencies into account, and thus form a MC system,
away non-linearities of xi itself, as it is not pertinent to our
but that at a later stage MC practices are added without taking any discussion. The notation we use for a vector that contains
interdependencies into account. The MC system then gets embedded all MC practices except for xi is xi . Eq. (3) further shows
within a broader set of MC practices that are a MC package. If the latter that any interdependence between MC practices is cap-
practices get redesigned to ‘‘connect’’ with the former, the broader set of
tured by the pay-off function VðÞ. While the costs of
MC practices move back into a system configuration. This trial and error
approach to design is likely to occur in practice. We thank an anonymous adopting a particular practice clearly affect adoption, these
reviewer for pointing this out. Such a trial and error approach to design costs are separate and do not affect productivity. Therefore,
further has an impact on the researcher’s choice of empirical strategy, the adoption costs of MC practices do not play a role in
which we discuss in Section ‘From theory to empirics’. determining whether these practices are interdependent.
4
We note that our goal is not to provide a comprehensive discussion of
complementarity theory, but rather to apply the insights from this theory
5
to the issue of MC as a system. We particularly refer the reader to Milgrom Adoption should be interpreted broadly here, in the sense that it not
and Roberts (1990, 1995) for more details about complementarity theory. only captures presence (vs. absence), but also higher (vs. lower) values.
 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419 411

The reductionist approach to design which implies that the choice of a MC practice depends on
contextual factors and separate adoption costs, as well as
The reductionist approach assumes that the anatomy of the use of other MC practices.
management control within the organization can be
decomposed into individual MC practices that can be Conceptual implications of different approaches
examined in isolation. Formally, this implies that the
reductionist approach assumes that Eq. (1) is characterized The previous two approaches cover two typical views
in a way such that the following holds: on MC design that form the implicit or explicit basis for
developing hypotheses about this design. Given the trend
@2p in examining multiple MC practices within one study, it
¼0 ð4Þ
@xi @xj–i is important to understand whether or not this is critical
That is, it assumes that the value of any MC practice for theory testing purposes or just a ‘‘courtesy’’ of the re-
does not depend on the use of any of the other MC prac- searcher to broaden the paper. A comparison between
tices. Given the contingency nature, it further assumes that the assumptions underlying the reductionist approach
and the systems approach reveals the (obvious) criterion.
@2p That is, it depends on the researcher’s assumption about
–0 8i; 9k ð5Þ
@xi @zk
@2p
i.e., the value of each MC practice is affected by at least one ð9Þ
@xi @xj–i
contextual factor. Thus, the value of each MC practice only
depends on the contextual variables and the separate This cross-partial captures, in one equation, the key
adoption costs, i.e., criticism on the reductionist approach. That is, the reduc-
@p tionist approach assumes that MC practices are indepen-
¼ V xi ðzÞ  pi ð6Þ dent, in which case the cross-partial is zero, while the
@xi
systems approach addresses this ‘‘problem’’ by assuming
The reductionist approach thus assumes that the bene- that there is interdependence, in which case the cross-par-
fits of each individual MC practice (and thus the choice) are tial is non-zero.
only affected by contextual factors. Eq. (6) further implies Despite the fundamental nature of Eq. (9), this issue is
that the reductionist approach allows each MC practice only positioned as the ‘‘second choice’’ of the systems ap-
to be affected by multiple contextual variables, and the proach. Although the ordering of the choices might not re-
same contextual variable can affect the choice of multiple flect an intended ordering of importance, subsequent
MC practices. That is, even though historically empirical research using the systems approach has focused on exam-
research following the reductionist approach has exam- ining how different clusters of MC practices are associated
ined only one MC practice and one contextual factor at a with contextual factors (e.g., Chenhall & Langfield-Smith,
time, this is not a restriction imposed by the approach. 1998; Gerdin, 2005; Moores & Yuen, 2001). It has thus ad-
dressed the first question of the systems approach of how
The systems approach to design the selection of (clusters of) MC practices is associated
with (clusters of) contextual factors, but it has not explic-
The systems approach maintains that the decision ma- itly addressed the second question of why these MC prac-
ker needs to make two design choices: (1) selecting MC tices should be adopted together, i.e., why they are
practices that match the set of contingencies facing the ‘‘internally consistent’’. Given that it is critical to take into
organization, and (2) assuring that the MC practices are account multiple MC practices only when these MC prac-
internally consistent (e.g., Drazin & Van de Ven, 1985). In tices are interdependent, the first question that researchers
terms of our framework, this formally implies that the sys- need to address is whether or not this is the case. To pro-
tems approach assumes that Eq. (1) is characterized in a vide guidance, we conclude this section with summarizing
way such that the following holds: what the formal definition of interdependence is and what
types of interdependencies can exist.
@2p @2p
– 0 8i; 9k and – 0 8i; 9j ð7Þ
@xi @zk @xi @xj–i
@2p @2V
That is, it assumes that the value of each MC practice is
Definition of interdependence : ¼ –0
@xi @xj–i @xi @xj–i
affected by at least one contextual factor – the first part/first
ð10Þ
design choice – as well as that the value of each MC practice
depends on the use of at least one other MC practice (and That is, the value of one MC practice depends on the use
vice versa) – the second part/second design choice. Note that of (some) other MC practice, which is captured by the pay-
the second part is required because ‘‘internal consistency’’ off (production) function. Note that interdependence is
cannot have any substance when MC practiczes are inde- symmetric, i.e., if the value of xi depends on xj–i, then the
pendent. As a result, the value of a MC practice, according value of xj–i also depends on xi. Further note that interde-
to the systems approach, is characterized by: pendence is a pairwise relationship. The sign of Eq. (10)
@p subsequently indicates what type of interdependence
¼ V xi ðxi ; zÞ  pi ð8Þ exists.
@xi
412 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419

@2p @2V Given this specification, interdependence is reflected by

Definition of complements : ¼ >0 ð11Þ
@xi @xj–i @xi @xj–i 2
@ p @2V
¼ ¼ bx1 x2 ð17Þ
MC practices are complements when the benefits of one @x1 @x2 @x1 @x2
MC practice increase with the use of (some) other MC prac-
The optimal decisions resulting from Eq. (16), i.e., the
tice (and vice versa).6
optimal x1 and x2, are characterized by
@2p @2V
Definition of substitutes : ¼ <0 ð12Þ bx1 bx1 x2  bz1 x1 bz x
@xi @xj–i @xi @xj–i x1 ¼ þ x þ z1 þ 2 1 z2 ð18Þ
p1 p1 2 p1 p1
MC practices are substitutes when the benefits of one
MC practice decrease with the use of (some) other MC bx2 bx1 x2  bz1 x2 bz x
practice (and vice versa).
x2 ¼ þ x þ z1 þ 2 2 z2 ð19Þ
p2 p2 1 p2 p2
These equations show that, if bx1 x2 is not equal to zero,
From theory to empirics
then the MC practices affect each other and the effects
are identical in direction (sign), though not necessarily in
In the previous section we discussed the core concep-
magnitude (as this depends on the separate adoption costs
tual difference between two design approaches, focusing
of each MC practice). We can now use the above equations
on interdependence. We now turn to the empirics and
to illustrate the empirical implications and (potential)
show that this difference naturally has empirical conse-
problems, in terms of Type I and/or Type II errors, of using
quences. We further discuss different empirical strategies
either a reductionist approach or a systems approach.
aimed at examining interdependence.
We first analyze a setting in which the assumption of
the reductionist approach is valid, i.e., bx1 x2 ¼ 0. In that
Empirical implications
case, Eqs. (18) and (19) become
To make things more tangible, we now position our dis- bx1 bz1 x1 bz x
cussion in an empirical context by putting more structure x1 ¼ þ z1 þ 2 1 z2 ð20Þ
p1 p1 p1
on our general theoretical framework. Remember that
our general framework is built on Eq. (1), i.e., bx2 bz1 x2 bz x
x2 ¼ þ z1 þ 2 2 z2 ð21Þ
Max pðx; z; pÞ ¼ Max Vðx; zÞ  p x T
ð13Þ p2 p2 p2
x x
That is, if the MC practices are independent, then the
We impose specific functions for the pay-off function
‘‘regression coefficients’’ related to the design of x1 (x2) re-
Vðx; zÞ and the cost function pT x. In particular, we assume
flect cost-benefit trade-offs associated only with the use of
that the decision maker is confronted with making deci-
x1 (x2). In this case, a study that empirically examines, for
sions about two MC practices, x1 and x2, given two contex-
example, x1 in isolation is as informative about the design
tual factors, z1 and z2, and that the pay-off function has the
of x1 as a study that examines both x1 and x2. The anatomy
following form:
of management control within the organization can thus
Vðx; zÞ ¼ b0 þ bx1 x1 þ bx2 x2 þ bx1 x2 x1 x2 þ bz1 x1 z1 x1 be decomposed into individual MC practices that can be
examined in isolation. This expands the point made in
þ bz2 x1 z2 x1 þ bz1 x2 z1 x2 þ bz2 x2 z2 x2 ð14Þ
Section ‘The reductionist approach to design’ that, if the
where the subscripts on the coefficients refer to the ‘‘vari- reductionist approach is valid, it is neither conceptually
able’’ to which it relates. Eq. (14) captures the pay-off effects nor empirically necessary to examine multiple MC
of different combinations of x1 and x2, and the degree to practices.
which the pay-off of x1 (x2) depends on z1 and z2. We assume The implication that it is ‘‘not necessary’’ to examine
that the cost function has the following form: multiple MC practices does not imply that one cannot do
p1 2 p2 2 this when the reductions approach is valid. Problems,
pT x ¼ x þ x ð15Þ however, occur when researchers rely on a systems ap-
2 1 2 2
proach in a setting in which the reductionist approach is
The decision maker thus maximizes the following valid. Even when bx1 x2 ¼ 0, the empirical results associated
objective function with multiple MC practices can be misinterpreted as pro-
Max pðx; z; pÞ ¼ Max b0 þ bx1 x1 þ bx2 x2 þ bx1 x2 x1 x2 viding evidence of interdependence. For example, assume
x x ;x1 2 that bz1 x1 > 0, bz2 x1 > 0, bz1 x2 < 0, and bz2 x2 < 0, i.e., assume
þ bz1 x1 z1 x1 þ bz2 x1 z2 x1 þ bz1 x2 z1 x2 that the value of x1 (x2) increases (decreases) with higher
p p  values of z1 and z2. In that case, a number of empirical anal-
þ bz2 x2 z2 x2  1 x21 þ 2 x22 ð16Þ yses will show the following results: (1) the cross-sectional
2 2
correlation qðx1 ; x2 Þ is negative, (2) there are clusters of
High/Low and Low/High x1–x2 combinations, and (3) in
6
explaining the MC practices, the regression coefficients
Strictly speaking, in complementarity theory, complements are defined
by a non-negative cross-partial rather than a strictly positive one. We use a
on the contextual factors have opposite signs for the two
stricter requirement here to emphasize the difference between indepen- MC practices. All these results might be interpreted as
dence and interdependence. evidence of a substitution effect. It should be clear,
 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419 413

however, that such an interpretation represents a Type I is typically negatively related to the strength of the other
error, as any cross-sectional relationship between x1 and method (Aral, Brynjolfsson, & Wy, 2012), it further pro-
x2 is solely driven by their joint contextual determinants. vides guidance as to which method is preferred and when.
In empirical terms, the results are driven by a correlated
omitted variable problem and in particular by not ade-
quately controlling for contextual factors. Estimating pay-off (production) function
This correlated omitted variable problem is obviously
not restricted to relying on a systems approach in a valid If the focus is on the pay-off function VðÞ, then an at-
reductionist setting. The opposite holds as well. For exam- tempt is made to directly estimate bx1 x2 by designing an
ple, assume now that the systems approach to design is va- empirical equation that mimics the pay-off function. Typi-
lid, i.e., bx1 x2 –0, but that the researcher uses a reductionist cally the regression equation will look something like
approach and examines x1 in isolation. From Eqs. (18) and
(19) it should be clear that, given that x2 is affected by z1 Performance ¼ b0 þ b1 x1 þ b2 x2 þ b3 x1 x2 þ b4 z1 x1
and z2, and bx1 x2 –0, omitting x2 from Eq. (18) will bias
the coefficients bz1 x1 and bz2 x1 (see Chenhall & Moers, þ b5 z2 x1 þ b6 z1 x2 þ b7 z2 x2 þ e ð22Þ
2007 for a detailed discussion of the omitted variable prob-
lem). This can lead to the extreme case where the value of where b3 is the estimate of bx1 x2 . Essentially, the equation
x1 is not affected by z1, i.e., the real bz1 x1 ¼ 0, but the empir- examines whether or not the MC practices are more or less
ical analysis reveals a significant association between x1 productive when adopted in tandem. The equation is fur-
and z1. Thus, in a cross-section there will be an association ther consistent with a contingency approach to fit in which
between x1 and z1 despite the fact that z1 is not a relevant performance is explained by different pairs of design and
contextual factor for x1 per se (there is only an indirect ef- context.
fect via the interdependence between x1 and x2). This for- To apply this method, a number of issues have to be
malizes Chenhall’s (2003) point mentioned earlier that, if explicitly considered. First, the measure of ‘‘Performance’’
two MC practices are ‘‘systematically linked’’, i.e., interde- on the left-hand-side (LHS) of the equation should be di-
pendent, then not controlling for this interdependence can rectly related to the decision process. The decision maker
trigger spurious findings (Type I error). decides on MC practices from the perspective of maximiz-
The problem of ‘‘finding things that are not there’’, i.e., ing performance related to a particular management con-
Type I errors, is not the only problem associated with mak- trol problem. More to the point, the measure of
ing incorrect assumptions and/or designing incorrect performance should be directly related to the pay-off func-
empirical tests. For example, Type II errorsh can occur when i tion VðÞ instead of the profit function pðÞ because interde-
the sign of bx1 x2 is opposite to the sign of bzi x1  bzi x2 , i.e., pendence is captured by VðÞ and using a broader measure
when contextual factors have opposite (identical) direc- of performance adds (potentially systematic) error in a
tional effects on MC practices that are complements (sub- cross-sectional analysis. As a result, the typical measures
stitutes).
  In that case, the cross-sectional correlation of performance used in management control studies such
q x1 ; x2 will be biased towards zero and thus underesti- as subunit performance seem to be ill-suited when esti-
mate the ‘‘true’’ interdependence captured by bx1 x2 . This mating Eq. (22).
indicates that not adequately controlling for contextual Second, there is a problem when using this method to
factors is as problematic as not adequately controlling for examine more than two MC practices. That is, if more than
interdependencies. two MC practices are examined, then the number of two-
way interactions increases and, depending on the theory,
Empirical strategies might require three-way interactions (see e.g., Aral et al.,
2012). Although this is not a hurdle for any statistical/
Now that we know what the empirical consequences of econometrics software, it does trigger the problem of inter-
different (theoretical) assumptions are and how some preting higher-order interactions (see e.g., Hartmann &
empirical observations are not informative for theory-test- Moers, 1999).
ing purposes, the next question is obviously: what empir- Last but not least, given that the purpose is to examine
ical strategy will be informative about bx1 x2 and can thus be what combinations of x1 and x2 are most productive, all
employed to test for interdependence? In essence, there combinations should in principle be empirically observed
are two ways to proceed; one is to focus on the pay-off for b3 to be an estimate of bx1 x2 . This implies that, in testing
function VðÞ in Eq. (1) and one is to focus on the choices for interdependence, the method’s underlying assumption
that result from the maximization of the profit function is that some organizations do not make decisions based on
pðÞ. We briefly discuss each of the two methods and refer this interdependence, resulting in suboptimal control
the reader to Arora and Gambardella (1990), Arora (1996) choices (and thus also cross-sectional variation in perfor-
and Athey and Stern (1998) for more detailed discussions. mance). The ‘‘pay-off function’’ (‘‘contingency’’) approach
As will become clear, the two methods resemble Drazin is therefore most suited (1) when it is more likely that
and Van de Ven’s (1985) distinction between a ‘‘contin- (some) decision makers have a lack of understanding of
gency’’ approach to fit, i.e., examining the outcome of a the interdependence and are thus still experimenting with
combination of MC practices, and a ‘‘congruence’’ approach particular MC practices, and/or (2) when an argument can
to fit, i.e., examining an association between MC practices. be made that (some) decision makers have a lack of incen-
More importantly, given that the strength of each method tive to take the interdependencies into account.
414 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419

Estimating demand functions basically each other’s mirror image, which implies that, if
one method becomes more suited to examine a certain set-
If the focus is on the choices made by the decision maker ting, then the other method becomes less suited to exam-
(revealed preferences), then the empirical test is aimed at ine that setting. Despite this difference, it is important to
examining whether or not the MC practices are adopted note that both methods, and thus also the contingency
jointly, which implies that these choices covary. Our discus- and congruence approach to fit, result from the exact same
sion in Section ‘Empirical implications’ shows that simply theoretical framework and thus do not differ in their con-
interpreting the cross-sectional correlation qðx1 ; x2 Þ is not ceptualization of fit. As a result, applying this framework to
informative about interdependencies between MC practices empirically examine interdependencies does not require
because this correlation might be biased upwards or down- the assumption that managers actually make optimal deci-
wards due to joint determinants. Fortunately, the observa- sions. What is required is a theory that predicts ‘‘optimal’’
tion that this correlation is potentially biased due to such decisions and thus also predicts the consequences of man-
contextual factors also hints at a solution. In particular, agers deviating from these decisions.
Arora and Gambardella (1990) formally show that, if choice
variables are interdependent, then these variables will
Implications and guidelines for future research
covary conditional on a set of ‘‘firm characteristics’’. In our
set-up this implies that, if bx1 x2 is positive (negative), then
The assumptions of MC interdependence carry some
the conditional covariance E½ðx1  Eðx1 jzÞÞ  ðx2  Eðx2 jzÞÞjz
important implications for future research on management
will also be positive (negative).
control system design. In this section, we put forward
To apply this method, we need an estimate of Eðxi jzÞ
some important questions that researchers aiming to
and thus need to estimate the demand function xi ðzÞ. That
investigate MC interdependencies need to address upfront,
is, we need to regress each MC practice on a set of contex-
and discuss potential pitfalls if not taken into
tual factors, which reflects the ‘‘demand’’ for each of the
consideration.
MC practices that is driven by the contextual factors. The
residuals of these regressions are estimates of
ðxi  Eðxi jzÞÞ and the pair-wise correlation between the Question 1: What is the control problem I want to address?
residuals of two MC practices thus an estimate of (the sign
of) bx1 x2 .7 Given that a set of controls implemented always reflects
As with estimating the pay-off function, estimating de- the underlying control problem it tries to resolve, it is crucial
mand functions requires some careful considerations. First to first define the control problem that should be addressed
and foremost is the choice of contextual variables on which by a particular combination of controls. Control mecha-
to condition the covariance. In a setting with only two rel- nisms can be complements in order to solve a control
evant MC practices, the covariance needs to be conditioned problem that is prevalent in one context, and substitutes
on all joint determinants, i.e., all contextual variables that to deal with another control problem in a different context.
affect the value of both MC practices. If more than two
MC practices are (theoretically) relevant, the covariance Question 2: What are the specific MC practices I want to
between two choices needs to be conditioned on all con- analyze?
textual variables that are observable, irrespective of
whether they are joint determinants for a particular pair Not only the control problem needs to be specified in
of MC practices. Unobserved heterogeneity obviously re- detail, also the MC practices have to be clearly defined. In
mains a potential concern here (see for a detailed discus- particular, the practices cannot be defined at a too aggre-
sion of this, Athey & Stern, 1998). gate level, since this reduces the inferences that can be
A second consideration is the optimality of choices. The drawn. For example, studies that examine whether ‘‘the re-
method does not assume that just any preferences are re- sults control system’’ is used diagnostically and interac-
vealed but that the revealed preferences are (more or less) tively at the same time are not informative about
consistent with the optimal decisions that follow from whether the same and/or different practices within this
maximizing the profit function. This is identical to the ‘‘system’’ are used in complementary ways. Finding a posi-
assumption underlying the congruence approach to fit. tive relationship between the interactive and diagnostic
The ‘‘demand function’’ (‘‘congruence’’) approach is there- use of results controls might be interpreted as evidence
fore best applied to problems where the interdependence of complementarity. However, such an interpretation con-
between MC practices is well-understood and/or there is tradicts Simons’ seminal field research on which the levers
no reason to believe that decision makers have incentives of control (LOC) framework is based because it ignores the
to ignore the interdependencies. individual practices that make up the control system. In his
In sum, the pay-off function (contingency) approach description of Johnson & Johnson, it becomes apparent that
and the demand function (congruence) approach are two in order to be able to use budgets interactively, as part of
distinct ways to examine interdependencies. The assump- the results control system, and elicit local (private) infor-
tions about optimality underlying each method are mation from middle management, it is essential that these
budgets are not used to reward these managers at the
same time, since otherwise they would have an incentive
7
For applications in the accounting literature, see Evans et al. (2010) and to withhold their information or bias it (Simons, 1987).
Indjejikian and Matêjka (2012). In other words, the benefit of using budgets interactively
 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419 415

decreases with using budgets diagnostically (and vice ver- problem to be addressed.8 The observation that MC prac-
sa), i.e., they are substitutes. Furthermore, the benefit of tices can be affected by and/or can affect trust clearly raises
using budgets interactively increases with using subjectiv- important conceptual and empirical issues. However, the
ity in rewards as part of the results control system (and implication here is that complementarity theory and the no-
vice versa), i.e., they are complements. Examining MC tion of complements and substitutes, as defined by Milgrom
practices at a too aggregated level, as is typically done in and Roberts (1990, 1995), are not applicable to these prob-
LOC studies, results in overlooking these subtleties in MC lems, and should thus be avoided.
design.
Question 4: How and why do the returns of one MC practice
Question 3: Does increasing (decreasing) one MC practice not depend on the use of other MC practices and vice versa?
prohibit increasing (decreasing) any other MC practice, i.e.,
are they really choices? After having identified the control problem and specific
MC practices, and established that these practices are real
Investigating whether MC practices are complements choices, the most fundamental question is raised. That is, if
(substitutes) in solving a particular control problem re- research explicitly aims at investigating MC interdepen-
quires that the designer of the MCS can freely decide about dencies, a crucial aspect is to clearly present a theory as
the adoption of a particular MC practice. Some MC prac- to why and how the benefits of using one MC practice
tices, however, are regulated by law, in which case the are related to the use of another, and vice versa. One exam-
decision maker has no option other than to adopt these ple in the literature is Evans, Kim, Nagarajan, and Patro
practices (for example, standardization of procedures in (2010), who build their arguments regarding the interrela-
hospitals). In that case, these practices are exogenously tion of financial and non-financial measures on the eco-
imposed and thus predetermined variables (in terms of nomic definition of complementarity. As Evans et al.
our model, this implies that these MC practices are not part (2010, p. 38) argue:
of x, but part of z).
‘‘Paying physicians based solely on their financial pro-
A somewhat different example of elements of the con-
ductivity, however, may discourage physicians from
trol environment that are not true choices can be found
participating in important non-revenue-generating
in the literature on the control of inter-organizational
activities, such as quality management or cost control
relationships. This literature is among other things con-
(Latham Consulting Group, 1990). Further, when pro-
cerned with the relationship between trust and control
ductivity control is not accompanied with proper con-
(see for example Dekker, 2004, 2008; Emsley & Kidon,
trol procedures for cost containment, it could lead to
2007; Tomkins, 2001). As Dekker (2004, p. 33) argues:
overuse of resources with no increase in quality. The
‘‘A much-debated question in the IOR literature is the professional health care literature expresses similar
relationship between formal control mechanisms and concerns (Latham Consulting Group, 1990; Rodeghero,
trust; i.e. whether (goodwill) trust is a substitute or a 2001). Rodeghero (2001, 156) observes that ‘‘only when
complement for formal control mechanisms in the effective resource management and quality assurance
management of appropriation concerns. [. . .] A substi- programs are in place can productivity-based pay plans
tutive relationship suggests that trust and formal con- ensure the appropriate rewards for the right practice
trol are inversely related; more trust results in less behaviors.’’ These arguments suggest that the benefits
use of formal control mechanisms and vice versa.’’ from using productivity-based performance measures
will be larger when they are used in conjunction with
Although a particular MC practice might be associated
nonfinancial performance measures than when they
with trust, the decision maker cannot directly decide on
are used in isolation.’’
trust, i.e., trust itself is not a real choice. For example, if
contingencies change the optimal level of formal control, Another example is Abernethy, Bouwens, and Van Lent
managers can neither ‘‘abandon’’ the trust of their employ- (2004, p. 546), who investigate the interdependence be-
ees nor ‘‘adopt’’ it. The only thing that might happen is that tween decentralization and the use of performance mea-
the control choice subsequently impacts employees’ trust. sures. In particular, they state that
The following citation from Dekker (2004, pp. 33–34) con-
‘‘. . .the two choices are tightly coupled. Jensen (2001)
firms that it is this association that is actually being argued
argues that corporate management will only decentral-
for rather than interdependent choices:
ize when it can implement a performance measurement
‘‘When, for instance, a firm has a reputation for being system that captures decision rights allocated to divi-
trustworthy, its partners may choose to use less formal sion managers. DSMs are designed to summarize the
control mechanisms compared to a firm with a less decisions over which a divisional manager has responsi-
favorable reputation. [. . .] Extensive use of formal con- bility. They provide corporate management with the
trol suggests a lack of belief in one’s goodwill or compe- possibility to assess the contribution of the division to
tence and therefore results in a damaging effect on firm value and to determine if divisional managers are
relational trust’’.
From a research perspective this implies that variables
such as trust are not part of the MC choices x, but rather 8
For example, it could be the objective of the decision maker to simply
part of the predetermined factors z or part of the control maximize trust.
416 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419

using decision rights optimally (Solomons, 1965; Zim- evaluation (Kaplan, 1983). The MCS is a system com-
merman, 1997).’’ posed of complementary components (Milgrom & Rob-
erts, 1995; Otley, 1994). This study includes personnel
Other papers investigate multiple controls but explicitly
controls, traditional results controls, and nontraditional
argue that there is a sequence in the choices of these con-
results controls [. . .]. Thus, this study provides a broader
trols. For example, Indejejikian and Matějka (2012) argue
investigation into the design of the MCS’’.
that accounting and operational centralization are comple-
ments, but also argue that the level of operational central-
Although this argumentation suggests interdependen-
ization determines the degree of accounting centralization.
cies between complementary MC practices, the hypothe-
Specifically, Indejejikian and Matějka (2012, p. 275) state
sis development solely focuses on the effects of a set of
that
contextual variables on each MC practice separately.
‘‘operational decentralization is the most fundamental Conceptually, such an argumentation matches the
choice among the decision variables considered in this assumption underlying the reductionist approach, which
study and is determined only by exogenous characteris- implies that (1) complementarity theory and interdepen-
tics of a BU’s environment. That is, we assume that dence are not applicable, and (2) it is a ‘‘courtesy’’, not a
operational decentralization is predetermined at the necessity to investigate multiple control practices at the
time a firm considers the extent of accounting same time.
decentralization’’. In sum, if control practices are considered to be deter-
mined sequentially or independently of each other, com-
In a similar vein, Widener (2007, p. 761) posits that the
plementarity theory is not suitable to address these
four levers of control are complementary, but states that
problems. Also in these cases, the use of the terminology
the belief system is the starting point. She, for example,
proposed by Milgrom and Roberts (1990, 1995) is mislead-
states
ing, and should be avoided. If control practices are, how-
‘‘once the firm’s strategy has been clarified and commu- ever, assumed to be interdependent, it is crucial to
nicated through the mission and vision contained in its develop a theory as to why and how the benefits of using
beliefs system, top managers will realize where poten- one MC practice are related to the use of another, and vice
tial threats and opportunities may reside and can versa.
implement an interactive system’’.
Question 5: What causes the optimal system solution to move
Both these quotes unambiguously indicate that the as-
around, i.e., what causes the optimal MC system to vary in a
sumed relationship between two (or more) MC practices
cross-section and/or over time?
is one of unidirectionality. The assumption about a unidi-
rectional relationship between MC choices, however, hides
While interdependence implies that optimal MC
an even more critical assumption. For example, while oper-
practices covary, an important question is whether such
ational decentralization is a real choice and affects the re-
behavior is observable in an empirical setting. Since, if
turns of another real choice (accounting decentralization),
there is only one optimal solution applicable to all firms
it is assumed that the returns to operational decentraliza-
all the time, the empiricists is stuck with observing only
tion do not depend on accounting decentralization and/
one particular system. The researcher therefore needs to
or that the decision maker cannot respond to any
address why there is variation in firms’ adoption of inter-
(inter)dependence. The assumption that the returns of
dependent practices in a cross-section and/or over time.
one MC choice depend on another MC choice but not vice
We discuss three different causes of variation in the opti-
versa is not only inconsistent with complementarity, it is
mal MC system, which are related to the cost function
also impossible once both are considered as choices.9 The
and the pay-off function.
critical, hidden assumption is thus that a real choice is trea-
ted as not being a real choice, in which case we are back at
Question 5a: Is there variation in the separate adoption costs
Question 3. The main difference with the discussion related
of the MC practices?
to Question 3 is that now the researcher has the difficult task
of substantiating why a MC practice that was once a choice
If two MC practices are complements and the sepa-
cannot be revisited when other MC practices change. In sum,
rate adoption costs of these practices are zero, then the
once unidirectionality is conceptually assumed, it is neither
optimal decision for all firms is to jointly adopt both
possible to apply complementarity theory nor relevant to
practices and there is thus no variation in MC systems.
empirically test for interdependence.
For there to be variation in MC systems, there should
Finally, another set of papers refer to MC as interrelated
be variation in the adoption costs of at least one MC
systems, but does not explicitly take interdependencies
practice across firms at one point in time (cross-sectional
between controls into account. For example, Widener
analysis) and/or over time (longitudinal analysis). That is,
(2004, p. 308) states that
the separate adoption costs of at least one MC practice
‘‘the purpose of the MCS is to provide information that should be higher for some firms than for others and/or
is useful in decision-making, planning, control, and should change over time. Although it is not necessary
to measure these costs, it is important to provide a plau-
sible argument for why there would be variation in these
@2 p @2 p @2 p 2
9
If @x1 @x2 –0; then @x2 @x1 –0; since @x1 @x2
p .
¼ @x@2 @x1
costs.
 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419 417

Question 5b: What are the exogenous variables that affect the tions certain practices are interdependent than determin-
marginal returns of individual MC practices? That is, what are ing whether they are interdependent per se.
the practice-specific contextual factors?
Question 6: What assumptions about optimality are made?
The need for control variables is relevant in any type of
research, and not a particularity of research on MC interde- As discussed before, the contingency and congruence
pendencies. However, this point cannot be made too often, approach are two distinct ways to examine interdependen-
since it is obviously crucial to control for the (joint) deter- cies. In both cases, the researcher needs a theory that pre-
minants of MC choices. Ignoring (joint) determinants in the dicts interdependence and thus also predicts the
analysis of interdependence significantly reduces the infer- consequences of managers deviating from the ‘‘optimal
ences that can be drawn about MC design. Such studies decisions’’ triggered by this interdependence. The decision
cannot draw inferences about interdependence, but only of which approach to follow depends on the assumptions
about ‘‘clustering’’ of controls or a loosely-coupled control about optimality that the researcher makes. If it can be rea-
package. Joint movement (i.e., correlation) can be induced sonably assumed that managers are still experimenting
by contextual factors that affect all control choices. For with the use of multiple MC practices and are thus likely
example, let’s say that task uncertainty affects both the to not make optimal decisions, the contingency approach
use of selection procedures and the use of non-financial is suitable, i.e., estimating pay-off functions. If, however,
performance measures. This implies that an increase in the researcher aims at investigating the interdependence
uncertainty will lead to higher use of both types of control. of established control practices whose interrelation is well
In a cross-section, we would observe a positive correlation, understood, following a congruence approach is most
indicating joint movement of the control choices. This cor- appropriate, i.e., estimating demand functions.
relation, however, does not necessarily mean that the two A possible extension regarding the assumption of opti-
MC practices are complements (see also the discussion in mality is to assume multiple decision makers. Assume for
Section ‘Empirical implications’). example that two MC practices x1 and x2 are complements
from an organization performance perspective but that the
decision rights with respect to each of these practices re-
Question 5c: What are the exogenous variables that affect the side with two different managers. If these managers do
marginal returns of combinations of MC practices? That is, not coordinate their decisions, potentially because of the
what are the system-specific contextual factors? incentives they are provided with, they will ignore the
complementarity when making their decisions. The result
So far we have characterized the contextual factors z as of this is that, what should be a system of interdependent
impacting the value of individual MC practices. There is, MC practices is potentially observed as a package of inde-
however, no reason to believe that such a ‘‘contingency pendent MC practices. More importantly, the theory that
theory’’ type of assumption is only applicable to individual predicts the complementarity between these two MC prac-
MC practices. On the contrary, it seems at least as plausible tices also predicts the negative consequences of this lack of
that contextual factors also affect the degree of interdepen- coordination. An opportunity for future research is there-
dence among MC practices. This implies that some MC fore to identify under what conditions the assumption of
practices might be, say, complements in one context but one decision maker is important but unlikely to hold, and
independent in another context, potentially because these examine the consequences of ignoring interdependencies
contexts trigger different control problems. Formally, this (see e.g., Siggelkow, 2002).
implies that Eq. (10) should be rewritten into In summary, the following check-list needs to be com-
pleted before examining issues related to MC
@2p @2V interdependencies:
¼ ¼ V xi xj–i ðzÞ ð23Þ
@xi @xj–i @xi @xj–i
1. What is the control problem I want to analyze?
in which case the vector z captures system-specific contex- 2. What are the specific control practices I want to investi-
tual factors. gate, and can I order each MC practice in a meaningful
An opportunity for future research is therefore to ex- way?
tend the view of contextual factors being only practice- 3. Does increasing (decreasing) one MC practice not
specific to (some of) these factors being system-specific. prohibit increasing (decreasing) any other MC practice,
What we are not saying is that we now should blindly as- i.e., are they really choices?
sume that contextual factors are system-specific. Clearly 4. How and why do the returns of one MC practice depend
there should be a theoretical argument for why a specific on the use of other MC practices and vice versa, i.e., what
contextual factor affects the degree of interdependence does the pay-off/production function look like and why?
among multiple MC practices. What we are saying is that 5. What causes the optimal MC system to vary in a cross-
such a theoretical perspective might better explain the section and/or over time?
‘‘combinations’’ of MC practices we observe and/or recon- a. Is there variation in the separate adoption costs of
cile some ‘‘inconsistent’’ findings related to relationships the MC practices?
between a contextual factor and an individual MC practice. b. What are the practice-specific contextual factors?
In addition, Cassiman and Veugelers (2006) suggest that it c. What are the system-specific contextual factors?
might be more important to understand under what condi- 6. What assumptions about optimality do I make?
418 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419

Conclusion Acknowledgements

The reductionist approach to contingency theory has The content of this paper has benefited from comments
been heavily criticized for valid reasons, but at the same from Chris Chapman, Rob Chenhall, Mike Shields (Editor),
time the systems approach has not been able to deliver two anonymous reviewers, and participants at the 2nd ER-
the pledged ‘‘solutions’’. The predominant criticism regard- MAC Conference in Vienna, the 8th Accounting Research
ing the systems approach is that it has not yet succeeded in Symposium in Madrid, the Research Workshop on ‘‘Perfor-
advancing our knowledge on the configuration of multiple mance Management: From Strategy to Action’’ in Prague,
control practices. The few studies relying on the systems and the 8th EIASM Conference on New Directions in Man-
approach have reacted to the deficiencies of the reduction- agement Accounting in Brussels.
ist approach by trying to depict the entire control environ-
ment of the organization by identifying clusters of control
practices. However, there is a lack of convincing theories as References
to why some control configurations might be observed
and/or be more successful than others. While the reduc- Abernethy, M. A., Bouwens, J., & Van Lent, L. (2004). Determinants of
tionist approach can be criticized for not being ambitious control system design in divisionalized firms. The Accounting Review,
70, 545–570.
enough, the systems approach struggles with being too Aral, S., Brynjolfsson, E., & Wy, L. (2012). Three-way complementarities:
ambitious. Performance pay, human resource analytics, and information
In this paper, we argue that to advance our knowledge technology. Management Science, 58(5), 913–931.
Arora, A. (1996). Testing for complementarities in reduced-form
on management control systems, we need to take one step
regressions: A note. Economic Letters, 50(1), 51–55.
at a time. That is, before we address the entire control envi- Arora, A., & Gambardella, A. (1990). Complementarity and external
ronment of the organization, i.e., the control package, we linkages – The strategies of the large firms in biotechnology. Journal
first have to understand if and how certain control prac- of Industrial Economics, 38(4), 361–379.
Athey, S., & Stern, S. (1998). An empirical framework for testing theories
tices form control systems in order to solve a specific con- about complementarity in organizational design. Working paper,
trol problem. The systems approach to contingency theory NBER.
maintains that the organizational design involves two ba- Bisbe, J., Batista-Foguet, J. M., & Chenhall, R. H. (2007). Defining
management accounting constructs: A methodological note on the
sic choices (Drazin and Van de Ven, p. 521): (1) selecting risks of conceptual misspecification. Accounting, Organizations and
MC practices that match the set of contingencies facing Society, 32, 789–820.
the organization, and (2) assuring that the MC practices Cassiman, B., & Veugelers, R. (2006). In search of complementarity in
innovation strategy: Internal R&D and external knowledge
are internally consistent. So far, studies have focused on acquisition. Management Science, 52(1), 68–82.
the first question, but avoided the second, in our opinion Chapman, C. S. (1997). Reflections on a contingent view of accounting.
even more important one. We argue that a major obstacle Accounting, Organizations and Society, 22(2), 189–205.
Chenhall, R. H. (2003). Management control system design within its
to moving the systems approach forward is the lack of clar- organizational context: Findings from contingency-based research
ity of what ‘‘internally consistent’’ means. and directions for the future. Accounting, Organizations and Society, 28,
To address this problem, we propose that complemen- 127–168.
Chenhall, R. H., & Langfield-Smith, K. (1998). The relationship between
tarity theory is suitable to put structure on what is meant
strategic priorities, management techniques and management
by ‘‘internally consistent’’ because the theory explicitly ad- accounting: An empirical investigation using a systems approach.
dresses how managers try to maximize ‘‘performance’’ by Accounting, Organizations and Society, 23(3), 243–264.
simultaneously deciding on multiple choice variables. Gi- Chenhall, R. H., & Moers, F. (2007). The issue of endogeneity within
theory-based, quantitative management accounting research.
ven the importance of interdependence for understanding European Accounting Review, 16(1), 173–195.
‘‘internally consistent’’, we provide a formal definition of Dekker, H. C. (2004). Control of inter-organizational relationships:
interdependence between MC practices and the two types Evidence on appropriation concerns and coordination requirements.
Accounting, Organizations and Society, 29, 27–49.
of interdependencies, i.e., complements and substitutes. In Dekker, H. C. (2008). Partner selection and governance design in interfirm
general terms, interdependence implies that the value of relationships. Accounting, Organizations and Society, 33, 915–941.
one MC practice depends on the use of another MC prac- Drazin, R., & Van de Ven, A. H. (1985). Alternative forms of contingency fit.
Administrative Science Quarterly, 30, 514–539.
tice, and vice versa. Emsley, D., & Kidon, F. (2007). The relationship between trust and control
With this conceptualization of internal consistency in in international joint ventures: Evidence from the airline industry.
mind, we discuss when it is important to address multiple Contemporary Accounting Research, 24(3), 829–858.
Evans, J. H., Kim, K., Nagarajan, N. J., & Patro, S. (2010). Nonfinancial
MC practices simultaneously, why this is the case, and what performance measures and physician compensation. Journal of
the empirical implications are. We not only provide answers Management Accounting Research, 22, 31–56.
to these questions, but also show that inadequately address- Ferreira, A., & Otley, D. (2009). The design and use of performance
management systems: An extended framework for analysis.
ing these questions severely affects what we can learn. The
Management Accounting Research, 20, 263–282.
specific guidelines that we put forward should help in Fisher, J. (1995). Contingency-based research on management control
overcoming this problem, thereby significantly improving systems: Categorization by level of complexity. Journal of Accounting
future research in this area. For contingency theory to devel- Literature, 14, 24–53.
Gerdin, J. (2005). Management accounting system design in
op further, a bridge between the reductionist and the manufacturing departments: An empirical investigation using a
systems approach needs to be build. By providing a defini- multiple contingencies approach. Accounting, Organizations and
tion of internal consistency, and proposing how it can be Society, 30, 99–126.
Hartmann, F. G. H., & Moers, F. (1999). Testing contingency hypotheses in
used to conceptualize MC systems, we believe to have made budgetary research: An evaluation of the use of moderated regression
one important step in that direction. analysis. Accounting, Organizations and Society, 24, 291–315.
 I. Grabner, F. Moers / Accounting, Organizations and Society 38 (2013) 407–419 419

Indejejikian, R., & Matějka, M. (2012). Accounting decentralization and Otley, D. T. (1980). The contingency theory of management accounting:
performance evaluation of business unit managers. The Accounting Achievements and prognosis. Accounting, Organizations and Society,
Review, 87(1), 261–290. 5(4), 413–428.
Luft, J., & Shields, M. D. (2003). Mapping management accounting: Siggelkow, N. (2002). Misperceiving interactions among complements
Graphics and guidelines for theory-consistent empirical research. and substitutes: Organizational consequences. Management Science,
Accounting, Organizations and Society, 28, 169–249. 48(7), 900–916.
Malmi, T., & Brown, D. A. (2008). Management control systems as a Simons, R. (1987). Codman & Shurtleff, Inc.: Planning and control
package – Opportunities, challenges and research directions. systems: Harvard Business School Case 187-081.
Management Accounting Research, 19, 287–300. Tomkins, C. (2001). Interdependencies, trust and information in
Merchant, K. A., Van der Stede, W. A., & Zheng, L. (2003). Disciplinary relationships, alliances and networks. Accounting, Organizations and
constraints on the advancement of knowledge: The case of Society, 26, 161–191.
organizational incentive systems. Accounting, Organizations and Van der Stede, W. A., Young, S. M., & Chen, C.-H. (2005). Assessing the
Society, 28, 251–286. quality of evidence in empirical management accounting research:
Milgrom, P., & Roberts, J. (1990). The economics of modern The case of survey studies. Accounting, Organizations and Society, 30,
manufacturing: Technology. strategy, and organizationg. American 655–684.
Economic Review, 80(3), 511–528. Widener, S. K. (2004). An empirical investigation of the relationship
Milgrom, P., & Roberts, J. (1995). Complementarities and fit: Strategy, between the use of strategic human capital and the design of the
structure, and organizational change in manufacturing. Journal of management control system. Accounting, Organizations and Society,
Accounting and Economics, 19(2/3), 179–208. 29, 377–399.
Moores, K., & Yuen, S. (2001). Management accounting systems and Widener, S. K. (2007). An empirical analysis of the levers of control.
organizational configurations: A life-cycle perspective. Accounting, Accounting, Organizations and Society, 32, 757–788.
Organizations and Society, 26, 351–389.