Beruflich Dokumente
Kultur Dokumente
Peter Göstl
Risk Profile
Contingent Analysis
of Management
Control Systems
Evidence from the Mechanical
Engineering Industry
Unternehmensführung & Controlling
Herausgegeben von
Univ.-Prof. Dr. Dr. habil. Wolfgang Becker Prof. Dr. Patrick Ulrich
Lehrstuhl für Betriebswirtschaftslehre, Professur für Unternehmensführung
insbes. Unternehmensführung & und -kontrolle
Controlling Hochschule Aalen – Technik und
Otto-Friedrich-Universität Bamberg Wirtschaft
Bamberg, Deutschland Aalen, Deutschland
Springer Gabler
© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2020
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
This Springer Gabler imprint is published by the registered company Springer Fachmedien Wiesbaden
GmbH part of Springer Nature.
The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany
Preface
cific risk type exposures by choosing one of two LOC MCS configura-
tions, i.e. a tight configuration in the case of a high exposure and a less
tight configuration in the case of a low exposure. This means that enter-
prises with a high (low) exposure to the three risk types have a tight (less
tight) configuration of their 4 lever-of-control management control sys-
tems composed of the 4 levers in the form of the beliefs, the boundary,
the diagnostic control and the interactive control systems. Furthermore,
companies enjoy positive performance benefits if the implemented LOC
MCS configuration fits to the tightness configuration implied by the enter-
prise’s specific exposure to the three risk types.
In summary: Peter Göstl’s doctoral thesis examines a highly relevant
practical problem, which is addressed with the help of a very solid re-
search foundation. Accordingly, its contribution is two-fold: Firstly, to pro-
fessionals who want to know how to configure their management control
systems in the light of the risk type exposures they face. Secondly, it
contributes to the scientific community by providing additional knowledge
that can be used for further research.
Finally, I want to express my thankfulness to all the commitment, effort,
fruitful discussions and interesting insights Peter delivered during the
years of his doctoral studies. It is a very great pleasure having such peo-
ple to work with. Thank you Peter!
1 Introduction ........................................................................................ 1
1.1 Research questions and design ................................................ 3
1.2 Structure and outline ................................................................. 7
2 Literature Review ............................................................................. 11
2.1 Introduction to management control ........................................ 11
2.1.1 Control problem in management ..................................... 12
2.1.2 Definitions and evolution of management control ........... 14
2.1.3 Uncertainty, risk and management control ...................... 17
2.1.3.1 Risk management and management control............... 19
2.1.3.2 Kaplan & Mikes’ conceptualization of risk types.......... 24
2.1.4 Overlaps with other fields in management literature ....... 27
2.1.4.1 Cybernetics and management control ......................... 28
2.1.4.2 Agency theory and management control ..................... 30
2.1.4.3 Organizational theory and management control ......... 34
2.1.4.4 Contingency theory and management control............. 36
2.2 Design of management control systems ................................. 41
2.2.1 Control systems and package view in
management control ........................................................ 41
2.2.2 Conceptualizations and evolution of MCS frameworks ... 46
2.2.3 Simons’ levers of control framework ............................... 54
2.2.3.1 Beliefs systems ............................................................ 58
2.2.3.2 Boundary systems ....................................................... 59
2.2.3.3 Diagnostic control systems .......................................... 61
2.2.3.4 Interactive control systems .......................................... 63
2.2.3.5 Interrelationship of the levers of control ...................... 65
2.2.3.6 Criticism of the LOC framework .................................. 68
X Table of contents
1
There are also other views in management control research. For a discussion of defini-
tions and evolution of management control see Section 2.1.2.
1995b; Merchant & Van der Stede, 2012). Moreover, taking a broader
package approach of MCS “may facilitate the development of better theo-
ry of how to design a range of controls to support organisational objec-
tives, control activities, and drive organisational performance” (Malmi &
Brown, 2008, p. 288). In regard to design and use of MCS, the contin-
gency-based approach is most prominent (Otley, 2016) which aims to
identify “specific aspects of an accounting system which are associated
with certain defined circumstances and demonstrate an appropriate
matching” (Otley, 1980, p. 413). In general, it is assumed that managers
adapt the organizational control package of their organization (Otley,
1980) to the existence, extent, and changes in contingencies (Chenhall,
2003), in order to attain fit and, as a consequence, enhanced organiza-
tional performance (Donaldson, 2001).
Conclusively, one of the most fundamental contingent factors in man-
agement accounting and control research is uncertainty, the importance
of which is widely acknowledged in literature (Chapman, 1997; Merchant
& Otley, 2007) and which “has gained by far the widest attention in the
area of management accounting” (Otley, 2016, p. 50). However, while
“organizations are centrally, even definitionally, concerned with the man-
agement of uncertainty and the coordination of resources to create forms
of order for identifying risk and making decisions” (Power, 2007, p. 98), it
is argued in literature that “many organizational control activities take
place as if uncertainty barely exists” (Otley, 2014, p. 95). Therefore, the
notion of uncertainty has been addressed increasingly in recent man-
agement accounting and control research (e.g., Bredmar, 2015; Otley,
2
2012, 2014, 2016; Otley & Soin, 2014). As an example, Malmi &
Granlund (2009) argue that the propositions developed by Chenhall
3
(2003) through reviewing the respective research findings in regard to
uncertainty and MCS are too simplistic and “need to advance from this
stage to be able to argue that as uncertainty increases, certain forms of
2
An extensive overview of the status quo is provided by a set of 17 essays on how uncer-
tainty affects the design and operation of MCS (Otley & Soin, 2014).
3
Malmi & Granlund (2009) refer to Chenhall’s proposition that the “more uncertain the
external environment the more open and externally focused the MCS” (Chenhall, 2003,
p. 138).
1.1. Research questions and design 3
MCS used in a certain way would provide better decision making support”
(Malmi & Granlund, 2009, p. 603). In addition, it is well known that when
“uncertainty is organized it becomes a ‘risk’ to be managed” (Power,
2007, p. 6). Still, while recent literature argues that the concepts of risk
and risk management have moved to an issue of management control
(e.g., Otley & Soin, 2014; Soin & Collier, 2013), the area has received
little attention and there is “relatively little understanding about the (com-
plex) interrelation between risk, risk management and management ac-
counting and control practices” (Soin & Collier, 2013, p. 84).
In the light of the above, some interesting and promising approaches
have been developed in literature to facilitate improved insight into the
relationship between management control and risk. A particularly striking
approach has been presented by Mikes & Kaplan (2014, 2015), who
stress the importance of qualitative distinction among types of risks and
argue that “the inherent nature and controllability of the different types of
risk the organization faces” (Mikes & Kaplan, 2015, p. 40) demand certain
4
control models (Simons, 1995b) for these risks to be managed effective-
ly. Thus, a contingency perspective is proposed, and it is called for re-
search to conceptualize and empirically test fit between types of risks and
control models applied.
Overall, the assessment of scarce insight into the intertwinement of man-
agement control and uncertainty in general and the both obvious and
impressive approach by Kaplan & Mikes (2012) in detail raised my curios-
ity and made me want to shed some light on this unresolved but crucial
venture to control uncertainty and contribute to the contingency-based
literature on management control systems.
4
In their article introducing their newly developed classification of risk, Kaplan & Mikes
(2012, p. 53) explicitly refer to the levers of control framework of Simons (1995a) for fur-
ther information on control models.
4 1. Introduction
5
In specifying the research questions, suggestions by Libby et al. (2002) are applied who
argue that “researchers should frame their theories at the least specific level” (Libby et
al., 2002, p. 796) in order to discriminate between existing theories and maximize contri-
butions to the theoretic disciplines. Furthermore, Libby’s et al. (2002) conclusion on Kin-
ney’s (1986) recommendations is considered which states that a research question
should address “the relation between two or more concepts, can be stated clearly and
unambiguously as a question, implies the possibility of empirical testing, and is important
to the researcher and others” (Libby et al., 2002, p. 794).
1.1. Research questions and design 5
RQ2: Packages of
of MCS
MCS
Link 1 Link 3
Risk profile MCS Firm and industry
design and use variables
FIT
Link 2
Performance
Chapter 1
Introduction
Chapter 2
Literature Review
Chapter 3
Theory Development and Hypotheses
Development of a
Development of Theoretical model
risk-based MCS
propositions (3.2) (3.3)
framework (3.1)
Chapter 4
Methods
Variable measurement
Data set and collection Research framework
and data analysis
(4.1 and 4.2) (4.5)
(4.3 and 4.4)
Chapter 5
Results
Chapter 6
Conclusions
tail, as well as a presentation of drivers of MCS design and use are ad-
dressed. Taking into account that the concept of uncertainty in manage-
ment control is the major underlying theme in this dissertation, special
focus will be delivered to this aspect.
Chapter 3 draws on the theoretical and empirical findings of contingency-
based research in management control theory and presents further con-
ceptual developments of Simons’ levers of control framework (Simons,
1995b) as well as testable propositions in order to examine the interrela-
tionship between MCS design and use, risk profile, and organizational
performance. Thus, theory is developed and the theoretical model illus-
trated by further specifying the conceptual level of the PVF depicted in
Figure 1.
Chapter 4 builds on the theoretical model (see Figure 21) comprising the
conceptual level of the predictive validity framework (PVF) and amends
the framework by the operational level (Libby et al., 2002). Thus, the
methods of this study are introduced and the choices and processes re-
garding data set, data collection, variable measurement, and data analy-
sis justified and discussed. Particularly, the operationalization process by
which constructs are translated into operational variables is addressed
(Bisbe et al., 2007; Libby et al., 2002) and the research framework com-
pleted (see Figure 32).
Chapter 5 presents the results of the empirical analysis and provides
evidence for and discussion of the hypotheses. Firstly, risk profile contin-
gent design and use of MCS will be examined in order to reveal associa-
tions between types of risks and specific control systems. Secondly, risk
profile contingent configurations of MCS will be addressed, to gain insight
into packages of MCS in practice. Thirdly, a risk profile contingent per-
formance analysis will be conducted, to provide an answer to the final
research question.
Chapter 6 summarizes the findings of this dissertation and presents the
contributions to the research field of management accounting and control.
Finally, a discussion of potential limitations of this study and key avenues
for future research conclude this dissertation.
2 Literature Review
uncertainty and risk is addressed and the way in which it overlaps with
other fields in management literature is presented.
6
For example, Rathe noted fifty-seven connotations of the term ‘control’ (Rathe, 1960, p.
32).
7
Otley & Berry stated that in the beginning, the most common idea of control had been
the “domination of one individual or group by another through the exercise of power” (Ot-
ley & Berry, 1980, p. 231). On the other hand, the focus of the term control on monitoring
and regulating of activities is affected by the original French term meaning “inspection”
and in line with connotations in several European languages (Otley & Berry, 1980, p.
231-232).
2.1. Introduction to management control 13
8
According to Chenhall (2003), the “terms management accounting (MA), management
accounting systems (MAS), management control systems (MCS), and organizational
controls (OC) are sometimes used interchangeably. MA refers to a collection of practices
such as budgeting or product costing, while MAS refers to the systematic use of MA to
achieve some goal. MCS is a broader term that encompasses MAS and also includes
other controls such as personal or clan controls. OC is sometimes used to refer to con-
trols built into activities and processes such as statistical quality control, just-in-time
management.” (Chenhall, 2003, p. 129)
16 2. Literature Review
9
Otley & Soin argue that production functions of organizations could be “buffered from
external change and allowed to develop well-programmed routines that led to considera-
ble operating efficiency” (Otley & Soin, 2014, p. 6).
10
For example, the organizations’ reactions to external changes caused by competitive
pressure leads to increased levels of uncertainty in the future (Otley & Soin, 2014, p. 7).
What is more, Otley argues that organizations’ failure to attempt to control all aspects of
the value chain, the consequential decline of overall holding organizations, and the con-
centration on smaller and more focused organizations increased exposure to uncertainty
(Otley, 2016).
18 2. Literature Review
to coping with uncertainty (Tessier, 2014; see also Tessier & Otley,
2012b). This change in the connotations of control is illustrated by Otley
& Soin: “Control can no longer be seen as a guarantee that predicted
outcomes will occur. Rather, control is a process which may be helpful in
guiding an organization through the stormy seas of its environment, and
assisting constant adaptation they may even encompass the change of
overall goals as some of the original goals become unrealistic” (Otley &
Soin, 2014, p. 8). Overall, control under conditions of uncertainty is one of
the abiding themes becoming apparent (Otley, 2012, 2016).
In general, uncertainty refers to a state where the information required
and the information possessed by an organization to perform a task dif-
fers from each other (Galbraith, 1977, p. 36-37). More precisely, Duncan
(1972) recognized similarities in the way in which the concept was de-
fined in literature and identified three common components of uncertainty:
“(1) the lack of information regarding the environmental factors associat-
ed with a given decision-making situation, (2) not knowing the outcome of
a specific decision in terms of how much the organization would lose if
the decision were incorrect; and (3) inability to assign probabilities with
any degree of confidence with regard to how environmental factors are
going to affect the success or failure of the decision unit in performing its
function” (Duncan, 1972, p. 318). In other words, uncertainty is “a lack of
ability to predict what the future will hold” (Otley, 2014, p. 83), encom-
passing the inability to predict future events as well as likely consequenc-
es (Otley, 2014). Conclusively, Otley (2014) argues that uncertainty is
rather related to individuals than to the external world, since it is not re-
garded as an objective phenomenon. In fact, in a completely determined
world with a future which is entirely predictable, one could still be uncer-
tain about what was going to happen next if the information with which to
understand or model the world was insufficient (Otley, 2014). More pre-
cisely, the adequacy of individuals’ (or groups’) predictive models to fore-
cast future events and their likely consequences determine the state of
2.1. Introduction to management control 19
uncertainty one is facing (Otley & Berry, 1980), leading to different types
11
and degrees of uncertainty for each individual (or group) (Otley, 2014).
In contrast to uncertainty, risk in a strict sense of the word occurs if the
range of future possibilities and their probabilities for manifestation are
known (Knight, 1921). However, it is evident that such circumstances are
rather subject to artificially closed systems with determined outcomes and
12
thus possess limited applicability in practice (Otley, 2014). Therefore,
Power’s argument is coherent: “Knightian uncertainties become risks
when they enter into management systems for their identification, as-
sessment and mitigation. […] When uncertainty is organized it becomes a
‘risk’ to be managed” (Power, 2007, p. 5-6). In general, it can be argued
that “[r]isk management processes represent one attempt to try to codify
the impact of many different types of uncertainty on major business activi-
ties” (Otley, 2014, p. 95).
11
Otley (2014) compares uncertainty to a rainbow, which appears to be a clearly perceived
phenomenon for all observers but which is actually perceived differently by each individ-
ual depending on one’s own position and movement.
12
Casinos or lotteries are practical examples where techniques of risk analysis can be
applied, although unexpected events can occur even in these situations (e.g., the casino
going bankrupt and being unable to deliver expected winnings) (Otley, 2014).
13
In their review of current developments in management control research, Speklé & Kruis
explicitly include the literature on risk management in their sample of papers to pick up
trends that constitute contemporary research in the field (Speklé & Kruis, 2014).
20 2. Literature Review
al., 2010; Bhimani, 2009; Collier & Berry, 2002; Merchant & Otley, 2007;
Mikes, 2009, 2011; Otley, 2012; Otley & Soin, 2014; Soin & Collier, 2013;
Soin et al., 2014; Speklé & Kruis, 2014; Woods, 2009). While there are
different views on the level and direction of incorporation of both research
fields, Merchant & Otley conclude that seeing “risk management as just
one (important) aspect of overall control systems design” (Merchant &
Otley, 2007, p. 787) is the most compelling view. Similarly, Soin & Collier
state that “risk management has moved away from being seen from the
finance silo’s view of value at risk and derivatives, and the accounting
silo’s view of disclosure in financial reports to a central concern with
management control” (Soin & Collier, 2013, p. 84).
Generally, Knight’s (1921) work on risk, uncertainty and profit is often
regarded as the starting point for risk management in literature (e.g., Soin
et al., 2014). In fact, there have been much earlier contributions to ex-
plaining concerns about risk, for example Bernoulli’s (1738/1954) utility
theory which added significant insights to risk theorizing and dominated
economics for decades (Harris, 2014). Nevertheless, in the last century,
key milestones in economic theory for ‘risk analysis’ (Harris, 2014) have
been set, for example Keynes’ (1921) treatise of probability calculation for
future events and its limitation when derived by similar events in the past;
von Neumann & Morgenstern’s (1953) development of game theory de-
lineating economically rational decision-takers which act on expected
outcomes arrived from measured risk preferences; Markowitz’ (1952)
coining of portfolio theory showing that investors can reduce risk by in-
vesting in diversified assets; and Black & Scholes’ (1973) contributions
on option pricing theory which deals with investing in options for delayed
decisions to reduce uncertainty for expected outcomes (Harris, 2014).
Although risk management reaches back to the beginning of manage-
14
ment science at the start of last century (Soin et al., 2014), the rise of
the research field can be attributed to the mid-1990s (Arena et al., 2010,
p. 660). The reason for this development is the increasingly uncertain
context in which organizations are usually being steered and thus the
14
For a comprehensive overview of the history of risk beginning in ancient times and the
history of risk measurement see Bernstein (1996a, 1996b) and McGoun (1995), respec-
tively.
2.1. Introduction to management control 21
15
Douglas states that: “It is more plausible that uncertainty is normal and the whole idea of
certainty an illusion. […] If we recognize more uncertainty now, it will be because of
things that have happened to the institutional underpinning of our beliefs” (Douglas,
2001, p. 145).
16
Beck refers to individualization meaning “the variation and differentiation of lifestyles and
forms of life, opposing the thinking behind the traditional categories of large-group socie-
ties - which is to say, classes, estates, and social stratification” (Beck, 1992, p. 88).
22 2. Literature Review
17
For example, the BP Deepwater Horizon oil spill in the Gulf of Mexico which resulted in
“environmental, economic and reputational losses, but also highlights the consequences
of poor risk management practices” (Soin & Collier, 2013, p. 83).
18
The Committee of Sponsoring Organizations define ERM as “a process, effected by an
entity’s board of directors, management and other personnel, applied in strategy setting
and across the enterprise, designed to identify potential events that may affect the entity,
and manage risk to be within its risk appetite, to provide reasonable assurance regarding
the achievement of entity objectives” (COSO, 2004, p. 2).
2.1. Introduction to management control 23
Internal Environment
Subsidiary
Objective Setting
Business Unit
Event Identification
Division
Entity-Level
Risk Assessment
Risk Response
Control Activities
Monitoring
cipline, with proven, unambiguous concepts and tools that need only
regulations and compliance to be put into widespread practice. We disa-
gree. We believe that risk management approaches are largely unproven
and still emerging“ (Mikes & Kaplan, 2014, p. 3). What is more, Mikes &
Kaplan argue that it is crucial for further development of ERM practice to
“find fit between contingent factors and firms’ ERM practice, and to estab-
lish propositions of fit that will result in desired outcomes” (Mikes &
19
Kaplan, 2014, p. 10). Furthermore, while aiming at building a contin-
gency theory of ERM Kaplan & Mikes propose that “risk management will
be most effective when it matches the inherent nature and controllability
of the different types of risk the organization faces” (Mikes & Kaplan,
2015, p. 40).
21
For example, in drilling several miles below the surface of the Gulf of Mexico, BP ac-
cepted high risks with the hopes of extracting high values of oil and gas (Kaplan &
Mikes, 2012, p. 51).
26 2. Literature Review
Mikes, 2012, p. 51). Kaplan & Mikes state that external risks comprise
22
events that managers can neither predict nor influence, for example
natural and political disasters and major macroeconomic shifts. In addi-
tion, these risks are “closely entwined with the firm’s strategic choices
and are therefore related to strategy execution risk” (Mikes & Kaplan,
2015, p. 40). Therefore, sources of external risks are also competitive
risks which include emergence of disruptive technologies and radical
strategic moves by industry players such as customers, suppliers, and
competitors. As external risks are outside the organization’s control, the
control models to manage external risks should focus on identification of
risk events, assessment of potential impact, and formulation of mitigation
strategies determining the organization’s resilience in case of materializa-
tion of risk events. In doing so, through using experience, intuition, and
imagination a process of risk envisionment is crucial in order to envision
plausible future disaster scenarios. Typical approaches regarding exter-
23 24 25
nal risks are tail-risk stress tests , scenario planning , and war-gaming
(Kaplan, 2014; Kaplan & Mikes, 2012; Mikes & Kaplan, 2014, 2015).
22
Kaplan (2014) notes that managers often “don’t know they don’t know” about external
risk events which corresponds to Donald Rumsfeld’s “unknown unknowns”, i.e. things
that we did not know that we did not know (Otley, 2012).
23
Stress-testing is used to assess major changes in showing how an event affects situa-
tions (Kaplan & Mikes, 2012).
24
Scenario planning refers to long-range analysis for “defining the plausible boundaries of
future states of the world” (Kaplan & Mikes, 2012, p. 58). As an example, participants se-
lect driving forces and assess the long-term impact on organizations and its strategies
(Kaplan & Mikes, 2012).
25
War gaming assesses “a firm’s vulnerability to disruptive technologies or changes in
competitors’ strategies” (Kaplan & Mikes, 2012, p. 58) through devising plausible strate-
gies and examining potential competitors’ strategies aiming at attacking the organiza-
tion’s strategy (Kaplan & Mikes, 2012).
2.1. Introduction to management control 27
26
The term Cybernetics was derived from the Greek word for steersman (Wiener, 1948, p.
11).
27
Relating to feedback activities Wiener states that “when we desire a motion to follow a
given pattern the difference between this pattern and the actually performed motion is
used as a new input to cause the part regulated to move in such a way as to bring its
motion closer to that given by the pattern” (Wiener, 1948, p. 6-7). Moreover, an often-
cited example of a simple cybernetic model in a machine setting is that of a thermostat.
Such a device typically includes a correcting feedback loop in measuring the actual tem-
perature, comparing this temperature to the desired one, and setting corrective actions
through switching on a heater if necessary (e.g., Lawler & Rhode, 1976, p. 40-41; Mer-
chant & Otley, 2007, p. 786).
28
In the early 1970s, Koontz & Bradspies applied to concept of feedforward to managerial
control problems (Koontz & Bradspies, 1972). Moreover, Otley & Berry emphasize the
importance of anticipatory controls, i.e. feedforward controls, as aspects of organization-
al control (Otley & Berry, 1980, p. 232).
2.1. Introduction to management control 29
Standard
Feedback
29
Otley & Berry derived these conditions from Tocher’s (1970, 1976) general control
framework rooted in cybernetic theory (Otley & Berry, 1980, p. 236).
30
Beer compared the control mechanism in an organization with the human nervous sys-
tem and argued that both needed to be adopted at various levels (Beer, 1972).
30 2. Literature Review
31
Ackoff defines that: „A system is a set of interrelated elements. Thus a system is an
entity which is composed of at least two elements and a relation that holds between
each of its elements and at least one other element in the set. Each of a system's ele-
ments is connected to every other element, directly or indirectly. Furthermore, no subset
of elements is unrelated to any other subset” (Ackoff, 1971, p. 662).
32
A comprehensive review of agency theory related to literature on MCS is provided by
Baiman (1990). In addition, Shields (1997) reviews studies in MCS research which have,
among others, employed agency theory.
2.1. Introduction to management control 31
33
Jensen & Meckling (1976) explain that by legal fiction it is meant that organizations being
artificial constructs under law are treated as individuals (Jensen & Meckling, 1976, p.
310).
34
The problem of asymmetric or incomplete information in favor of the agent leads to so
called ‘hidden action’ and/or ‘hidden information’ problems (e.g., Bergen et al., 1992).
32 2. Literature Review
averse to risk than the principal, as agents are less able to diversify their
35
employment than principals their investments (Eisenhardt, 1989).
In general, the primary concern of agency theory is to develop control
arrangements to encounter a conflict of interest between principal and
agent in order to maximize the principal’s objective function (e.g., Eisen-
hardt, 1989). This is done by determining the most efficient contract gov-
erning the relationship between the parties and taking into account as-
sumptions about people, organizations, and information. Specifically, it is
to be decided to what extent a behavior-oriented contract or an outcome-
oriented contract is more efficient in a specific setting, dependent on goal
conflict, information asymmetry, risk preferences, outcome uncertainty
and measurability, task programmability and length of agency relationship
(Eisenhardt, 1989).
As argued by Eisenhardt (1985, 1989), in the special case of complete
information, the behavior of the agent is observable and thus a behavior-
based contract optimal, as both parties know the agent’s actions. This is
not the case if the principal is unaware of the agent’s behavior and there-
fore cannot assess if the agent behaved in the principal’s best interest.
Relying on behavior-based contracts and rewarding the agent for an
agreed behavior without being able to monitor the agent’s actions may
result in the agent to shirk. To overcome this dilemma, it is essential for
the principal to purchase information through surveillance mechanisms
and reward appropriate behavior. Alternatively, the principal rewards the
agent based on outcomes, which is a surrogate measure for behavior.
However, Eisenhardt (1985, 1989) states that while outcome-based con-
tracts encourage efforts by the agent, this scheme also shifts risk from
the principal to the agent and therefore comes at the price of rewarding or
penalizing for outcomes which are at least partially outside of the agent’s
control. According to agency theory the optimal strategy for control is thus
a function of the costs of information systems which are required to rely
35
Whereas the simple model assumes a more risk averse agent than principal, it is evident
that individuals vary considerably in their preferences and attitudes towards risk (e.g.,
Eisenhardt, 1989; MacCrimmon & Wehrung, 1986). Therefore, there are several exten-
sions to this model which relax this assumption. For example, a wealthy agent may be
less risk averse which makes it more attractive to pass risk from the principal to the
agent in using outcome-based contracts (Eisenhardt, 1989).
2.1. Introduction to management control 33
36
As first-best solutions can only be achieved by absence of agency costs and therefore
perfect information of all parties, agency theory evidently deals with second-best solu-
tions (Bromwich, 2007, p. 152).
34 2. Literature Review
Task Programmability
Perfect Imperfect
Behavior or Outcome
High
Measurability
Behavior Socialization
Low
37
For example, conflicting findings of Hopwood (1972) and Otley (1978) regarding the
budget-based superior evaluation style, job-related tension, and managerial performance
have led to a stimulation of the development of contingency theory of management ac-
counting (Brownell & Dunk, 1991, p. 693; Otley, 2003, p. 313).
38 2. Literature Review
38
Other concepts with similar connotation refer to unpredictability, non-routineness, dyna-
mism, task uncertainty, etc.
2.1. Introduction to management control 39
Contingent variables
Intervening variables
Other factors
Organisational effectiveness
(measured in relation to objectives)
39
For a review of contingency studies in management accounting and control, see e.g.,
Chapman (1997), Chenhall (2003, 2006), Fisher (1995) and Otley (1980, 2016).
40 2. Literature Review
More organic
More mechanistic
40
Formal controls may also be derived from organizational culture, for example a formally-
worded mission statement which reflects organizational values and beliefs (Langfield-
Smith, 2007).
44 2. Literature Review
41
Otley first used the term package to address separate parts of an overall MCS arguing
that he “felt unhappy with using the term ‘system’ because it seemed to imply a designed
system of well-coordinated parts, and many overall MCSs did not seem to possess this
property. I used the term ‘package’ to imply a set of pieces that were put together without
ensuring that they were fully coordinated” (Otley, 2016, p. 53).
2.2. Design of management control systems 45
(Malmi & Brown, 2008). Finally, it can be argued that the package view of
MCS has led to the development of various MCS frameworks (e.g., Mer-
chant & Van der Stede, 2012; Otley, 1999; Ouchi, 1979; Simons, 1995)
which are discussed in the subsequent sections below.
Norm of Reciprocity
Bureaucracy Rules
Legitimate Authority
Norm of Reciprocity
Clan Legitimate Authority Traditions
Shared Values, Beliefs
42
In his transaction cost approach, Williamson (1981) explains the existence of firms and
argues that, in specific circumstances, firms can be more efficient as markets which re-
sults in lower transaction costs.
48 2. Literature Review
43
A norm of reciprocity is a social agreement which ensures, with regard to market mech-
anisms, that a party attempting to cheat in a market transaction “will be punished by all
members of the social system, not only by the victim and his or her partners” (Ouchi,
1979, p. 838). Similarly, in regard to the other mechanisms, the norm of reciprocity as-
certains that “in exchange for pay, an employee gives up autonomy in certain areas to
his organizational superiors, thus permitting them to direct his work activities and to mon-
itor his performance” (Ouchi, 1979, p. 838).
2.2. Design of management control systems 49
that “[r]eal organizations will each contain some features of each of the
modes of control” (Ouchi, 1979, p. 840).
This approach by Ouchi (1979) was further developed by Merchant
(1985b) and Merchant & Van der Stede (2012), who noted that all con-
trols are behavioral controls and refined the terminology through renam-
ing the types of controls as action, results, and personnel/cultural controls
(Merchant & Otley, 2007). In its conceptualization, MC primarily has an
internal focus aiming at influencing employees’ behavior in order to en-
sure that firm’s objectives are not harmed but supported. In contrast to
this, strategic control processes have an external focus dealing with how
firms can establish a competitive advantage and compete in its industry
(Merchant & Van der Stede, 2012). Specifically, Merchant & Van der
Stede (2012) identified three causes for the needs of MCS which can
occur simultaneously and in any combination, i.e. lack of direction, moti-
vational problems, and personal limitations, and which are encountered
by different types of control (see Figure 11). While a lack of direction oc-
curs if employees simply do not know the organizational objectives and
how to contribute to them, motivational problems exist due to individuals’
self-interestedness and colliding individual and organizational objectives.
Both of these aspects are referred to as a lack of goal congruence in
management accounting and control literature (Merchant & Van der
Stede, 2012). On the other hand, personal limitations appear if employ-
ees are unfit to perform well despite of being aware of objectives and
being highly motivated (Merchant, 1985b; Merchant & Van der Stede,
2012). Merchant & Van der Stede (2012) point out that some limitations
are person-specific, for example lack of training, experience, knowledge,
or aptitude. In addition, there are also aspects suggested by psychology-
based research which concern potentially all individuals, for example the
tendency to overestimate the likelihood of common or recently occurred
events compared to rare or not recently occurred events, respectively
(Merchant & Van der Stede, 2012).
50 2. Literature Review
Control problems
Results controls
Results accountability x x
Action controls
Behavioral constraints x
Preaction reviews x x x
Action accountability x x x
Redundancy x
Personnel/cultural controls
Selection and placement x x x
Training x x
Provision of necessary resources x
Creation of a strong organizational culture x x
Group-based rewards x x
The first type of controls in Merchant’s (1985b) and Merchant & Van der
Stede’s (2012) framework is results controls, which involves “rewarding
individuals […] for accomplishing particular results or outcomes” (Mer-
chant, 1985b, p. 17) and “influenc[ing] actions or decisions because they
cause employees to be concerned about the consequences of their ac-
tions or decision” (Merchant & Van der Stede, 2012, p. 30). Therefore,
44
results controls empower employees, as the organization does not dic-
tate which actions not to take and this is particularly necessary for the
successful realization of decentralized organizations. While pay-for-
performance is the most prominent example for this type of control, there
are also other rewards such as autonomy, promotions, job security, and
recognition. In general, implementation of results controls involves the
following steps: “(1) defining the dimension(s) on which results are de-
sired; (2) measuring performance in the chosen dimensions; (3) setting
performance targets for employees to attain for each of the measures;
44
The employee empowerment approach has become a major trend since the 1990s, see
for example Seibert, Silver & Randolph (2004).
2.2. Design of management control systems 51
and (4) providing rewards for targets attainment to encourage the behav-
iors that will lead to the desired results” (Merchant & Van de Stede, 2012,
p. 33). In addition, for feasibility of results controls there are three pre-
conditions which should all be present: organizations must know and
communicate desired results, employees must be able to affect the re-
sults, and finally, controllable results must be effectively measurable
(Merchant, 1985b; Merchant & Van der Stede, 2012). Thus, the control-
lability principle is central to responsibility accounting which recognizes
that “results measures are useful only to the extent that they provide in-
formation about the desirability of the actions or decisions that were tak-
en” (Merchant & Van der Stede, 2012, p. 37) and are not effective in situ-
ations where significant uncontrollable influences affect the measures
(Merchant & Van der Stede, 2012). In regards to effective measurability,
Merchant & Van der Stede (2012) state that to “evoke the right behavior,
in addition to being congruent and controllable, results measures should
be precise, objective, timely, and understandable. And even when a
measure has all of the above qualities, it should also be cost efficient”
(Merchant & Van der Stede, 2012, p. 38).
Secondly, action controls involve “ensuring that employees perform (do
not perform) certain actions known to be beneficial (harmful) to the organ-
ization” (Merchant & Van der Stede, 2012, p. 81). Merchant & Van der
Stede (2012) point out that, as the employees’ actions are the focus of
control, these are the most direct management controls which can take
the forms of behavioral constraints, preaction reviews, action accountabil-
45
ity, and redundancy. While behavioral controls can be applied physically
or administratively and aim at preventing employees from taking undesir-
46
able actions, preaction reviews are demanded for approval of proposed
actions. Action accountability, on the other hand, involves defining ac-
ceptable (unacceptable) actions, communicating and observing these
45
Behavioral controls include physical controls and administrative controls. While typical
examples of physical constraints include computer passwords or fingerprint readers,
administrative controls include, among others, restrictions of decision-making authorities
or separation of duties, a basic requirement of internal control (Merchant & Van der
Stede, 2012).
46
For example, reviewing planned actions and budgets at higher levels of an organization
are a form of common preaction review (Merchant & Van der Stede, 2012).
52 2. Literature Review
47
For example, harmful side effects comprise of behavioral displacement, gamesmanship,
operating delays, or negative attitudes (Merchant & Van der Stede, 2012).
48
Net benefits of controls are the benefits of implementing a control minus the direct and
indirect costs of a control caused (Merchant & Van der Stede, 2012).
54 2. Literature Review
1. “What are the key objectives that are central to the organization’s
overall future success, and how does it go about evaluating its
achievement for each of these objectives?
2. What strategies and plans has the organization adopted and what are
the processes and activities that it has decided will be required for it to
successfully implement these? How does it assess and measure the
performance of these activities?
3. What level of performance does the organization need to achieve in
each of the areas defined in the above two questions, and how does it
go about setting appropriate performance targets for them?
[…]
4. What rewards will managers (and other employees) gain by achiev-
ing these performance targets (or, conversely, what penalties will they
suffer by failing to achieve them)?
5. What are the information flows (feedback and feed-forward loops) that
are necessary to enable the organization to learn from its experience
and to adapt its current behaviour in the light of that experience?” (Ot-
ley, 1999, p. 365-366)
Otley’s framework was further developed by Ferreira & Otley (2009) who
extended the framework to 12 questions to provide “a useful checklist of
the major aspects of control systems design that may need to be included
in a comprehensive analysis of such systems” (Merchant & Otley, 2007,
p. 789-790).
To this day, one of the most influencing frameworks for MCS (Otley,
2016, p. 55) remains that developed by Simons (1995b). As Simons’
levers of control framework is applied in the empirical part of this study, it
is presented in more detail in the following section.
49
This so-called command-and-control rhetoric reaches back to the 1960s and implicates
MC with terms such as standardization and efficiency, keeping things on track without
surprises, results according to plans, and top-down strategy setting etc. (Simons, 1995b,
p. 3).
50
Since Simons’ (1995b) LOC framework there has been considerable research on the
tension between control and innovation. For further thoughts on formalized MCS and
creativity see Adler & Chen (2011) and Pfister (2014).
51
Simons refers to a unique set of opportunities which an organization can identify as
opportunity space, and argues that, in fact, organizations face unlimited opportunity (Si-
mons, 1995b, p. 14-16). This aspect is put in contrast to the study of economics, which is
mainly concerned with choices among restricted sets of opportunities (Arrow, 1974a;
Simons, 1995b).
52
Mintzberg argues that managers have to act as figureheads, leaders, liaisons, monitors,
information disseminators, spokesmen, entrepreneurs, disturbance handlers, resource
56 2. Literature Review
allocators, and negotiators and thus must allocate their restricted attention to these mul-
tiple roles (Mintzberg, 1973, p. 167-169; Simons, 1995b, p. 17)
53
For example, at the beginning of twentieth century, Taylor (1911/1919) perceived work-
ers to be solely motivated by financial incentives. Thus, in pursuit of efficiency through
piece rate incentives, it was aimed at improving repetitive tasks and ensuring that behav-
ior complied with prescribed procedures (Taylor, 1911/1919). Some decades later, social
aspects in performance implications were elaborated and motivations for human beings
extended by aspects of self-actualization (Maslow, 1954). Later again, economists
viewed individuals as self-interested effort-averse agents who aim at maximizing their
utility function (e.g., Jensen & Meckling, 1976).
2.2. Design of management control systems 57
four levers of control, which are (1) beliefs systems, (2) boundary sys-
tems, (3) diagnostic control systems, and (4) interactive control systems
(Simons, 1995b). These levers are applied in order to control four key
constructs which have to be analyzed for successful strategy implemen-
tation: (1) core values, (2) risks to be avoided, (3) critical performance
variables, and (4) strategic uncertainties. The following figure provides an
overview of the key aspects of Simons’ framework:
Beliefs Boundary
Systems Systems
Core Risks to
Values Be Avoided
Business
Strategy
Critical
Strategic
Performance
Uncertainties
Variables
Interactive Diagnostic
Control Systems Control System
54
Simons defines internal controls as controls that are “designed to safeguard assets from
misappropriation and ensure reliable accounting records and information systems” (Si-
mons, 1995b, p. 84). Internal controls differ from boundary systems as they are not con-
cerned which risks to be avoided but rather specify detailed procedures for information
handling (Simons, 1995b, p. 84-85). While internal controls are the most basic building
block for effective control, they are not considered a lever of control as these systems
“contain little information of importance to managers” (Simons, 1994, p. 170).
55
Simons uses terms of Chinese philosophy and compares the four levers of control with
the yin and yang. Whereas boundary systems and diagnostic control systems create
constraints and represent yin, i.e. negative controls, beliefs systems and interactive con-
trol systems create inspirational forces and represent yang, i.e. positive controls (Si-
mons, 1995b, p. 7-8).
2.2. Design of management control systems 59
values amongst employees” (Otley, 2003, p. 317). In doing so, they have
to be broad enough to provide inspirational appeal to all organizational
participants and use symbolic information which is of special importance
in modern decentralized organizations (Simons, 1995b). As is generally
well understood by successful leaders, symbolism is essential in generat-
ing inspiration and belief (Feldman & March, 1981; Westley & Mintzberg,
1989). As a result, managers attempt to direct organizations by “(1) as-
serting uniqueness, (2) providing prestige to group membership, and (3)
using formal beliefs as symbols of what the organization represents” (Si-
mons, 1995b, p. 37). This results in increased commitment and positively
influences the perceived distinctiveness and stability of organizations
(Ashfort & Mael, 1989). As committed individuals are more willing to
make an effort to reach organizational goals, organizational inertia is
challenged (Mundy, 2010, p. 501) and the level of performance usually
increases (Locke et al., 1988).
56
Simons mentions the Ten Commandments from the Old Testament to be an excellent
example for a boundary system (Simons, 1995b, p. 41).
60 2. Literature Review
and uncertainty (e.g., Leblebici & Salancik, 1982; Simons, 1995b), per-
formance pressure (e.g., Rich et al., 1990) as well as prior experience
with discrete incidents (e.g., Andrews, 1989) is argued to have an influ-
ence on business conduct implementation. In addition, Schwartz (2001)
reports evidence that business code of conducts in fact influence behav-
ior of employees and finds a number of reasons why codes are or are not
57
complied with. However, while these systems limit the freedom in indi-
vidual creativity, they can also be liberating in protecting against pressure
from superiors to engage in inappropriate behavior (e.g., Brenner & Mo-
lander, 1977).
On the other hand, Simons (1995b) argues that strategic boundaries aim
at supporting organizational strategies and guide opportunity-seeking
behavior in limiting search activities. For this purpose, it is usually speci-
fied which business opportunities not to pursue in order to avoid a waste
58
of resources such as financial capital or management attention. Fur-
thermore, strategic boundaries can be used to limit opportunity sets
59
which may lead to exit existing businesses. For both business conduct
and strategic boundaries, the means of enforcement are sanctions which
must imply credible threats to be effective (Simons, 1995b).
Nevertheless, boundary systems are regarded a prerequisite for organi-
zational freedom and flexibility as they enable delegation of decision
60
making and actually facilitate entrepreneurial behavior (Simons, 1995b).
57
Schwartz (2001) identified five reasons for employees’ non-compliance with the business
code of conduct: “(a) self-interest (i.e., greed, being a star, financial distress, avoid har-
assment); (b) dissatisfaction (i.e., with one's job or level of reimbursement); (c) environ-
ment (i.e., peer pressure, supervisors' behaviour, opportunity); (d) company's best inter-
est; and (e) ignorance (i.e., never aware, didn't perceive, forgot)” (Schwartz, 2001, p.
253-254). On the other hand, employees tend to comply with the code due to: “(a) per-
sonal values; (b) fear of discipline; and (c) a feeling of loyalty to the company” (Schwartz,
2001, p. 254).
58
Checklists and asset acquisition systems (capital budgeting systems setting a hurdle
rate for profitability) are instruments to set minimum boundaries (Simons, 1995b, p. 49-
51).
59
For example, John Welch, Jr., chairman of General Electric, stated repeatedly that GE
would exit any business in which GE is not able to achieve a number 1 or 2 position on
the world market (Simons, 1995b, p. 49).
60
Simons compares boundary systems with brakes in a car and suggests understanding
their primary function as possibility to go fast instead of slowing down (Simons, 1995b, p.
41).
2.2. Design of management control systems 61
61
According to Lawler & Rhode (1976, p. 148; see also Cammann & Nadler, 1976), diag-
nostic control measures should be independently verifiable (i.e., objective), capturing all
relevant actions (i.e., inclusive), and reflecting the efforts of individuals being measured
(i.e., responsive).
62 2. Literature Review
62
mostly regarded as single-loop learning systems (Argyris & Schön,
1978; Simons, 1995b).
Output variables which are usually controlled by diagnostic control sys-
tems are critical performance variables that have an influence if intended
strategies are implemented successfully (Simons, 1995b). A well-known
instrument for identifying and analyzing critical performance variables is
Kaplan and Norton’s concept of the balanced scorecard, which suggests
using diagnostic measures in regard to financial as well as non-financial
63
perspectives simultaneously to reach business goals (Kaplan & Norton,
1992). These critical performance variables are not overseen constantly
by the management, but rather attention is focused on three steps: (1)
setting and negotiating goals, (2) receiving updates and exception re-
ports, and (3) following up on significant exceptions (Simons, 1995b, p.
70-71). Therefore, diagnostic control systems allow management-by-
exception leaving constant activities to staff experts and enable effective
allocation of management attention by triggering managerial corrective
action in case of identified variances (Simons, 1995b).
Simons (1995b) argues that another key function of diagnostic control
systems is linking accomplishment of preset goals with extrinsic rewards
to increase the motivation of organizational participants. In doing so, spe-
cial attention must be given to designing these systems carefully, in order
to avoid dysfunctional behavior such as engaging in actions that influence
64
the measure but do not further organizational goal achievement or
building slack into targets (Simons, 1995b).
62
Argyris & Schön note that single-loop learning describes “a single feed-back loop which
connects detected outcomes of action to organizational strategies and assumptions
which are modified so as to keep organizational performance within the range set by or-
ganizational norms. The norms themselves […] remain unchanged” (Argyris & Schön,
1978, p. 18-19).
63
Kaplan & Norton (1992) introduce the financial, customer, internal business, and innova-
tion and learning perspective to track implementation of business strategy.
64
This so-called gaming the system means that individuals maximize rewards without
considering effects on organizational goals. For example, a customer service operation
of a credit card company which measures and rewards the number of calls per day could
tempt to minimize talk time per call and thus lead to dissatisfied customers and less
deals (Simons, 1995b, p. 82).
2.2. Design of management control systems 63
65
Argyris & Schön note that double-loop learning describes “a double feedback loop which
connects the detection of error not only to strategies and assumptions for effective per-
formance but to the very norms which define effective performance” (Argyris & Schön,
1978, p. 22).
64 2. Literature Review
66
Other properties are identified as ‘intensive use by top management’, ‘intensive use by
operating managers’, ‘face-to-face challenge and debate’, and ‘focus on strategic uncer-
tainties’ (Bisbe et al., 2007, p. 795-798).
2.2. Design of management control systems 65
1 “To be used interactively, the control system must require the refore-
casting of future states based on revised current information.
2 To be used interactively, the information contained in a control system
must be simple to understand.
3 To be used interactively, a control system must be used not only by
senior managers but also by managers at multiple levels of the organ-
ization.
4 To be used interactively, a control system must trigger revised action
plans.
5 To be used interactively, a control system must collect and generate
information that relates to the effects of strategic uncertainties on the
strategy of the business.” (Simons, 1995b, p. 108-109)
67
Mintzberg (1978) distinguishes various types of strategy, i.e. intended strategies, emer-
gent strategies, and realized strategies, as well as deliberate and unrealized strategies.
Intended strategies are “(a) explicit, (b) developed consciously and purposefully, and (c)
made in advance of the specific decisions to which it applies” (Mintzberg, 1978, p. 935).
They represent a plan. Realized strategies, on the other hand, are defined as “pattern[s]
in a stream of decisions” (Mintzberg, 1978, p. 935) and refer to the decisional behavior of
an organization which manifests in a consistent sequence of decisions. Finally, emergent
strategies refer to “realized strategies that were never intended […] because no strategy
was intended at the outset or […] those that were got displaced along the way”
(Mintzberg, 1978, p. 945).
2.2. Design of management control systems 67
Beliefs Systems
Boundary Systems
Target
Intended Realized
Process Outputs
Strategy Strategy
Business Strategic
Strategy Uncertainties
Unrealized
Strategy
Emergent Strategy
As shown in Figure 13, all four control systems work together to frame
the strategic domain and provide an effective control environment. The
dynamic tension between the levers of control implies that “effective or-
ganizations must achieve simultaneously high degrees of learning and
high degrees of control” (Simons, 1995b, p. 158). This is supported by
Widener who found the control systems in the LOC framework to be in-
terdependent and complementary (Widener, 2007). Similarly, having
conducted within-case analyses and cross-case comparisons, Bruining et
68 2. Literature Review
al. concluded that the control systems in the LOC framework do not re-
68
place but rather complement each other (Bruining et al., 2004).
68
More specifically, Bruining et al. found evidence that control systems that stimulate
opportunity-seeking and learning extend and are in balance with traditional control sys-
tems (i.e., diagnostic control systems) (Bruining et al., 2004).
69
Tessier & Otley reported almost 790 citations in literature in 2011 (Tessier & Otley,
2012a, p. 172).
70
Simons identified the beliefs systems, boundary systems, and feedback and perfor-
mance measurement systems as design attributes of MCS. Consequently, while diag-
nostic and interactive control systems must be designed as well, they represent attention
patterns of the design of the feedback and measurement systems (Simons, 1995b, p.
180).
2.2. Design of management control systems 69
2007; Bisbe & Malagueno, 2009; Naranjo-Gil & Hartmann, 2007). Fur-
thermore, a strong strategic focus and the resulting implications for con-
trol systems, as well as remarks on balances between positive and nega-
tive controls have been regarded as useful (Ferreira & Otley, 2009, p.
266) and have “contributed to a broadening of the role of MCS” (Berry et
al., 2009, p. 6). Similarly, Otley concludes that the “value of Simons’ work
is that it sets out a more complete picture of the whole gamut of a wide
range of possible controls, and the particular mix that has been chosen
by an organization” (Otley, 2003, p. 317).
In terms of weaknesses, it is argued that the LOC framework lacks con-
71
sideration of socio-ideological, i.e. informal controls which were found to
72
be an important aspect in control systems (Collier, 2005). In addition,
Simons’ conceptualization of MCS strongly focuses on the top level of
management (Simons, 1994) which constitutes a limitation in application
of the framework as the operation of control systems in levels of lower
hierarchies is not considered (Ferreira & Otley, 2009). Furthermore, im-
portant concepts of the LOC framework are criticized of having a diffuse
73
(e.g., core values) or ambiguous (e.g., interactive control systems )
meaning leading to operationalization issues due to subjective interpreta-
tion (Bisbe et al., 2007; Ferreira & Otley, 2009). Finally, it is debatable if
Simons’ framework is applicable to universal settings since beliefs and
boundary systems usually refer to the whole organization but are beyond
the domain of control of divisions or subsidiaries (Ferreira & Otley, 2009,
p. 266).
71
Although Collier argues that beliefs systems implicitly comprise, at least to some extent,
informal controls (Collier, 2005), Simons explicitly excludes informal control processes
according to his definition for MCS (see Section 2.1.2) (Simons, 1995b, p. 5).
72
In his 10-year longitudinal field study, Collier found that informal controls such as sociali-
zation, group norms, and culture tend to be even more important than formal controls
(Collier, 2005).
73
Bisbe et al. argue that interactive control systems is an ambiguous construct in the
research of management accounting and control systems and call for precise conceptual
specification by production of specified meaning and specification of the epistemic rela-
tionship (Bisbe et al., 2007; see Section 4.3.1). Ferreira & Otley, on the other hand,
demonstrate that interactive controls comprise of two distinct components, i.e. interactive
use of controls and strategic validity controls (Ferreira & Otley, 2009, p. 266; see also
Footnote 92).
70 2. Literature Review
Contingent Factor
LEVEL 1
Outcome Variable
is not addressed
Control System Mechanism
Contingent Factor
LEVEL 2
Outcome Variable(s)
Contingent Factor
LEVEL 3
Outcome Variable(s)
Control System Mechanisms i
(i = 1 to n)
Contingent Factor j
(j = 1 to m) LEVEL 4
Outcome Variable(s)
Control System Mechanisms i
(i = 1 to n)
74
Drazin & Van de Ven exemplarily mention government legislative bodies that regulate
industries, industries that have constraining codes for businesses, and organizations that
have policies affecting divisions and departments (Drazin & Van de Ven, 1985, p. 517).
74 2. Literature Review
tion fit studies (e.g., Abernethy & Lillis, 1995; Brownell, 1982a; Go-
vindarajan & Gupta, 1985; Ittner et al., 2002) is that organizations move
towards optimal combinations but these adaptions in fact constitute a
dynamic process as context and structure change over time (Chenhall &
Chapman, 2006). For example, Chenhall & Chapman (2006) argue that if
a company changes its strategy, the MCS must be adjusted in order to
reflect the new situation, which takes some time to accomplish. There-
fore, through moving in and out of equilibrium, poor performance can be
expected if there is a current misfit between MCS and context (Chenhall
& Chapman, 2006). The following figure shows an exemplary interaction
fit between environmental heterogeneity and structural complexity and
the corresponding effect on performance:
Complex
ORGANIZATIONAL STRUCTURE
High performance
Simple
Low performance
Homogeneous Heterogeneous
ORGANIZATIONAL ENVIRONMENT
As Drazin & Van de Ven point out, the “focus here is not so much on
understanding the congruence between context and structure as in the
selection approach, but rather on explaining variations in organizational
performance from the interaction of organizational structure and context”
2.3. Contingency-based studies in management control 75
(Drazin & Van de Ven, 1985, p. 517). Furthermore, they introduce differ-
ent types of interaction fit and stress its implication on operationalization
issues and application of statistical methods (Drazin & Van de Ven, 1985,
p. 518-519; see also Schoonhoven, 1981). Drazin & Van de Ven explain
mixed results in various interaction fit studies, among others, with meth-
odological issues due to a lack of appropriate operationalization of inter-
action fit for which different approaches exist, i.e. the deviation-score
approach and the multiplicative interaction approach (Drazin & Van de
Ven, 1985). When considering deviation-score approaches, the “impact
of deviations in structure from an ideal context-structure model, in which
fit is defined as adherence to a linear relationship between dimensions of
context and structure” (Drazin & Van de Ven, 1985, p. 519) is analyzed,
multiplicative interaction approaches usually “limit the form of the interac-
tion only to acceleration and deceleration effects, which researchers have
not specifically hypothesized in their concept of fit” (Drazin & Van de Ven,
1985, p. 519). Similarly, Donaldson divided this type of contingency fit
into matching and multiplicative interaction fit (Donaldson, 2001). While
the latter has been widely used in respective contingency studies, re-
search in MCS literature applying concepts of matching fit is rare (Chen-
hall & Chapman, 2006). In matching fit approaches, “optimal combina-
tions between aspects of a contextual variable and particular dimensions
of an MCS” (Chenhall & Chapman, 2006, p. 40) are assumed to exist.
More precisely, there is a unique state of fit regarding each level of con-
textual variable and level of MCS variable which leads to maximal per-
formance or, if there is a deviation in either direction, to an according
deficiency in output (Chenhall & Chapman, 2006). As the performance
outcome on any point of the fit line is the same, Donaldson used the term
‘isoperformance’ to describe this kind of relationship (Donaldson, 2001).
Furthermore, the idea of matching fit can model and test conceptualiza-
tions of curvilinear relationships between MCS design and context
(Chenhall & Chapman, 2006). For example, Chenhall & Chapman state
that it can be argued that, while conditions of environmental uncertainty
demand more organic forms of control systems for reasons of flexibility, it
is possible that high levels of uncertainty create a need for more mecha-
nistic controls to ensure survivability of the organization (Chenhall &
76 2. Literature Review
Chapman, 2006, 40-41; for organic and mechanistic controls see Section
2.2.1).
Multiplicative interaction fit, on the other hand, presumes that “higher
(lower) values of context require higher (lower) values of MCS to achieve
higher (lower) performance” (Chenhall & Chapman, 2006, p. 45). For
example, multiplicative interaction fit models tested whether more flexibil-
ity in budget use led to better performance in case of high environmental
uncertainty as compared to less flexibility in budget use (Chenhall &
Chapman, 2006). While in monotonic multiplicative interaction relation-
ships more (less) of an MCS variable leads to higher (lower) performance
given any level of context variable (though the improvement might be
relative), in non-monotonic relationships the effect of an interaction on
performance might be positive for higher levels of an independent varia-
ble but negative at lower levels (Chenhall & Chapman, 2006; Luft &
Shields, 2003). Although such approaches are theoretically less precise
in general, multiplicative interaction fit models are used in most studies of
MCS contingency studies (Chenhall & Chapman, 2006). In terms of sta-
tistical methods, Euclidean distance is used for matching fit models and
multiple regression analysis is used for multiplicative interaction fit mod-
els to test propositions in regard to MCS design and organizational con-
text (Chenhall & Chapman, 2006).
Finally, the systems fit approach (e.g., Chenhall & Langfield-Smith, 1998;
Gerdin, 2005; Govindarajan, 1988; Selto et al., 1995) as a third type of fit
presumes that “the understanding of context-structure performance rela-
tionships can only advance by addressing simultaneously the many con-
tingencies, structural alternatives, and performance criteria that must be
considered holistically to understand organization design” (Drazin & Van
de Ven, 1985, p. 519). The following figure conceptually shows the sys-
tems fit approach with several states of configurations leading to high
performance:
2.3. Contingency-based studies in management control 77
High
IT A
B
IT = Ideal Type
A = Highest Performing Organization
Low
Low High
ORGANIZATIONAL STRUCTURE - DIMENSION 1
2006) as well as profile deviation analysis (Gerdin & Greve, 2004), are
the main statistical methods to testing systems fit approaches.
Another classificatory framework for mapping different forms of contin-
gency fit in literature is provided by Gerdin & Greve (2004). The frame-
work, shown in the figure below, exhibits a hierarchical structure and
comprises four levels representing conflicting paradigms at the top level
and dichotomous differentiation of modeling or measuring at the lower
levels.
Forms of Fit
Cartesian Configuration
Figure 17: Gerdin & Greve’s classificatory framework for different forms of contingency fit
(Source: Gerdin & Greve, 2004, p. 304)
At the top level of Gerdin & Greve’s (2004) classificatory framework, the
distinction between a Cartesian approach and a Configuration approach
is presented. The Cartesian approach is characterized by reductionism
and advocated by traditionalists of a structural contingency theory argu-
ing that “fit between context and structure is a continuum that allows fre-
2.3. Contingency-based studies in management control 79
75
Gerdin & Greve point out that “fit conceptualized as Congruence implies that there is
little room for alternative solutions. Accordingly, we should not expect to find a Contin-
gency form of fit since it requires that less effective designs also exist (if not, it would be
impossible to show that deviations from the ideal design are related to lower perfor-
mance)” (Gerdin & Greve, 2004, p. 309).
80 2. Literature Review
The third level of the classificatory framework further subdivides the Car-
tesian-type relationships and differentiates in types of fit following a Mod-
eration approach and following a Mediation approach (Gerdin & Greve,
2004; Luft & Shields, 2003; Shields & Shields, 1998). In moderation
models, 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” (Gerdin & Greve, 2004, p. 309) and that the moderator
has ‘‘nonsignificant, bivariate relationships with both the independent and
dependent variables’’ (Shields & Shields, 1998, p. 51), theoretically not
being significantly related to either of them. A mediation model, on the
other hand, “specifies the existence of a significant intervening mecha-
nism between an independent variable and the dependent variable”
(Gerdin & Greve, 2004, p. 305; see also Venkatraman, 1989) and thus fit
exists if the impact of one variable (e.g., strategy) on another variable
(e.g., performance) operates through a third one (e.g., MCS) (Gerdin &
Greve, 2004).
Finally, theoretical meanings of fit can be represented in terms of the
strength and form of the relationships between variables. The strength of
a relationship between variables using a moderation model is reflected if,
for example, the predictive ability of a variable such as MCS design on
performance differs across different strategies. In such a case, the corre-
lation between MCS design and organizational performance may be
higher among organizations following a specific type of strategy com-
pared to others. On the other hand, the form of a relationship between
variables is reflected, for example, if the impact of MCS design on per-
formance differs across different strategies (Gerdin & Greve, 2004). That
is, the reliance on specific MCS design has a significantly more positive
effect, i.e. a significantly higher slope when modeling the relationship, on
performance among organizations following a specific type of strategy
compared to others. Gerdin & Greve conclude that “the strength and form
types of moderation fit may yield consistent results […], but there is no
reason to expect that they should” (Gerdin & Greve, 2004, p. 312).
To summarize, it is apparent that variations in the use of vocabulary to
describe different types of fit have led to a lack of clarity and ongoing
debates on this topic (e.g., Gerdin & Greve, 2004). As Chenhall & Chap-
2.3. Contingency-based studies in management control 81
man point out, the use of same words especially in relation to different
forms of fit have caused confusion in this area (Chenhall & Chapman,
2006, p. 37). However, as Drazin & Van de Ven point out, the “definition
of fit that is adopted is central to the development of the theory, to the
collection of data, and to the statistical analysis of the proposition” (Drazin
& Van de Ven, 1985, p. 515).
76
Nevertheless, Otley points out that there is still a debate on “whether an internal variable
which can be affected by the firm itself should be treated as a contingent variable or as a
dependent variable, most notably concerning organizational structure and strategy” (Ot-
ley, 2016, p. 48).
82 2. Literature Review
portant to note that most empirical tests conducted in these studies are
tests of association and not causality. While sometimes referred to as
‘drivers’ affecting the emergence of MCS, Davila accurately points out in
regard to his own contributions that the “use of this term is related to the
evolutionary perspective and the causal association that these models
adopt” (Davila, 2005, p. 226) and thus results should be interpreted with
caution. While there is a multitude of contextual variables mentioned in
literature which are supposed to influence MCS, this section focuses on
the relationships between MCS and uncertainty, strategy, organizational
size, company age, and ownership structure as essential factors applied
in this empirical study.
79
For a discussion of the issue of increasing uncertainty during recent years see Section
2.1.3.
84 2. Literature Review
“The more uncertain the external environment, the more open and exter-
nally focused the MCS.
The more hostile and turbulent the external environment, the greater the
reliance on formal controls and emphasis on traditional budgets.
Where MCS focused on tight financial controls are used, in uncertain
external environments, they will be used together with an emphasis on
flexible, interpersonal interactions.” (Chenhall, 2007, p. 173)
80
Interestingly, another aspect of the environment, environmental hostility, has been asso-
ciated with strong reliance on formal controls (Chenhall, 2007; Otley, 2016), e.g., strict
emphasis on meeting budgets (Otley, 1978). As uncertainty and hostility often occur
simultaneously but produce different impacts for MCS design and use, the question of
how this tension should be managed is still open (Chenhall, 2007; Otley, 2016).
81
Note that there is a fundamental difference between uncertainty and risk (Knight, 1929;
see Section 2.1.3). Nevertheless, there are various interferences in literature, e.g.,
Kaplan & Mikes’ (2012) categorization of different types of risks which comprise uncon-
trollable, i.e. uncertain, events as one facet of the conceptualization.
2.3. Contingency-based studies in management control 85
82
Widener (2007) differentiates between operational uncertainties, competitive uncertain-
ties, and technological uncertainties on the one hand and operations risk and competi-
tive risk on the other hand. In her study, she found that one facet of strategic uncertainty,
i.e. competitive uncertainty, is also associated with interactive control systems (Widener,
2007).
86 2. Literature Review
83
In addition, Porter (1980) also introduced a third generic strategy, i.e. strategy of focus,
which describes organizations focusing on specific segments of a market which is poorly
served but which has special needs.
2.3. Contingency-based studies in management control 87
84 N(N−1)
Davila argues that the number of potential interactions in a company is with N
2
being the number of employees and thus efficiency of informal management rapidly de-
creases and coordination and control costs increases with size, resulting in formalized
control mechanisms in order to regain management efficiency (Davila, 2005, p. 226).
90 2. Literature Review
informal control systems to be more effective than formal MCS and thus
the age of the company may not be relevant (Davila, 2005).
85
Otley (2016) states that many “of the performance measurements which are highly
valued and of significant impact are non-financial measurements such as production
process measures, defect rates, cycle time and customer service measures; or qualita-
tive measurement, such as customers’ perception, attitude of employees towards jobs,
and product innovation” (Otley, 2016, p. 48).
86
In their study Coase et al. (1992) show that there are a variety of measures used in
practice and that these differences seem to be connected with country-specific prefer-
ences in regard to financial stability (Coase et al., 1992; Otley, 2016).
87
As can be seen from the subsequent brief summaries of the studies, most of them place
an emphasis on the relatedness between MCS and strategy. However, this focus of the
2.3. Contingency-based studies in management control 93
selected studies is rather coincidental and reflects the prominent role of strategy in con-
tingency-based MCS research.
94 2. Literature Review
Beliefs
System
Strategic
Uncertainties Boundary Attention
System
Performance
Diagnostic
Strategic Controls
Risks Learning
Interactive
Controls
choice of initial MCS reflects the firms’ strategy, and that a match be-
tween MCS and strategy would result in superior performance. For char-
acterization of the firm’s strategy, the generic strategies of cost leader-
ship and/or differentiation based on Porter’s (1980, 1985) conceptualiza-
tion were used.
¾ Cost controls
1. Cost MCS
¾ Quality controls
Strategy
Low cost / ¾ Marketing Databases
2. Revenue MCS
differentiation Multinomial ¾ Sales Productivity
Logit
Performance:
¾ Initial MCS usefulness
¾ Business Performance
addition to Basic MCS, Sandino found evidence for three types of MCS,
i.e. Cost MCS, Revenue MCS, and Risk MCS, which are contingent on
specific purposes such as focusing on enhancing operating efficiencies,
on firm growth and customer responsiveness, and on reducing risks and
protecting integrity, respectively. In order to yield a model of fit, a multi-
nomial logit analysis is conducted which predicts the category of initial
MCS chosen. By using deviations from the model’s predictions, Sandino
found that firms with a better fit between initial MCS and strategy per-
formed better than others in terms of perceived performance, usefulness
of MCS, and actual performance. To summarize, it can be stated that
Sandino’s empirical investigation not only provided evidence on contin-
gency-based MCS design and implications for performance but that it
also presented an interesting approach with regard to the methodology of
modelling and conducting analyses of fit, essentially influencing the ap-
proach of the present study.
88
In his extensive review of the literature on the contingency theory of management ac-
counting from 1980 to 2014, Otley (2016) found that the amount of publications on this
topic over the period is steadily increasing.
89
The central themes of Otley’s suggestions of promising areas for future research on MC
are strategy and structure, innovation and diversity, sustainability, power and control,
culture, and more sophisticated research methods (Otley, 2003).
100 2. Literature Review
ation to the risk profile of an organization. Finally, the last set of hypothe-
ses, i.e. the performance hypotheses, examines the implications for per-
formance through matching risk profile and configuration of MCS.
Thirdly, in order to illustrate the research questions and conceptual rela-
tionships between the key variables, the theoretical model of this study
will be presented. A predictive validity framework (PVF) (Libby et al.,
2002) will be applied as generic framework, later to be promoted from the
conceptual to the operational level in Chapter 4. While following a contin-
gency approach through applying Otley’s (1980) minimum contingency
framework, an overview of the main variables, i.e. explaining variables,
explained variables, other potentially influential variables, and perfor-
mance variables, will be provided.
PANEL 1
attributes
Design Beliefs Boundary
systems systems
Attention
patterns
Interactive Diagnostic
controls controls
PANEL 2
Formal controls
that frame the strategic domain
attributes
Design
Beliefs Boundary
systems systems
Attention
patterns
Interactive Diagnostic
controls controls
Use of controls
which determines style of usage
Figure 20: Extending Simons’ LOC framework to develop a risk-based MCS framework
(continued)
106 3. Theory Development and Hypotheses
PANEL 3
Formall controls
F t l
that frame the strategic domain
Design attributes
Beliefs Boundary
systems systems
Attention patterns
Interactive Diagnostic
controls controls
Use of controls
which determines style of usage
Figure 20: Extending Simons’ LOC framework to develop a risk-based MCS framework
(continued)
3.1. Development of a risk-based MCS framework by extension of the LOC framework 107
PANEL 4
Formal controls
that frame the strategic domain
Design attributes
Beliefs Boundary
systems systems
Attention patterns
Interactive Diagnostic
controls controls
Use of controls
which determines style of usage
Figure 20: Extending Simons’ LOC framework to develop a risk-based MCS framework
(Source: own illustration)91
91
The presentation of the evolutionary process of developing the extended risk-based
MCS framework in Figure 20 was inspired by Tessier & Otley’s conceptual development
of Simons’ LOC framework (Tessier & Otley, 2012a, p. 183).
108 3. Theory Development and Hypotheses
The starting point for extending Simons’ LOC framework and putting for-
ward a holistic risk-based MCS framework are the levers of control and
its distinguishing features, i.e. the distinction in design attributes of formal
controls and attention patterns of the feedback and measurement sys-
tems, as illustrated in Panel 1 of Figure 20 above (Simons, 1995b, p.
180). As pointed out by Ferreira & Otley the framework “offers a broad
perspective of the control system by looking at the range of controls em-
ployed and how they are used by companies” (Ferreira & Otley, 2009, p.
92
266). Thus, while the concepts of diagnostic and interactive controls
have been discussed intensively in literature (Ferreira & Otley, 2009;
Tessier & Otley, 2012a) and have been described as ambiguous con-
93
cepts (Bisbe et al., 2007), the intention of Simons’ (1995b, 2000) con-
ceptualization seems clear and serves as a point of origin for further con-
94
siderations. Furthermore, as design attributes and attention patterns
already represent distinct concepts within the LOC framework, it seems
92
Ferreira & Otley (2009) point out that: “Simons conflates the intensive use of information
by managers with the identification of an inadequate strategy“ (Ferreira & Otley, 2009, p.
274). Thus, it is their argument that linking these concepts together creates confusion
and they make the suggestions to divide Simons’ (1995b) concept into interactive use of
controls which addresses the intensity of managers’ use of controls, and strategic validity
controls which monitor the adequacy of the strategy (Ferreira & Otley, 2009). Tessier &
Otley (2012a) follow this approach and relate the interactive use of controls to the first
three components identified by Bisbe et al. (2007; see Footnote 93) and the strategic va-
lidity controls to the component which addresses the focus on strategic uncertainty
(Tessier & Otley, 2012a). Thus, the term interactive controls is applied in their revised
framework and it is pointed out that their said revised framework “does not consider di-
agnostic and interactive controls as control systems in their own right, but rather as a
description of how control systems are used” (Tessier & Otley, 2012a, p. 178). As the ex-
tended risk-based MCS framework in this study builds on this distinctive feature and is
thus aimed in the same direction, the terminology of interactive (diagnostic) controls in-
stead of the original interactive (diagnostic) control systems is adopted here.
93
Bisbe et al. (2007) attempted to provide a distinct definition of interactive control systems
based on Simons‘ (1995b) conceptualization and proposed five components: intensive
use by superiors, intensive use by subordinates, face-to-face communication, focus on
strategic uncertainty, and non-invasive management style.
94
Alternative considerations would have been to choose the distinction in positive and
negative controls as suggested by Simons (1995b) as a starting point for implementing
risk-based dimensions of MCS, i.e. the distinction in beliefs systems and interactive con-
trol systems on the one hand and boundary systems and diagnostic control systems on
the other (see also Section 2.2.3). This approach seemed inappropriate as the risk-
based approach being aimed for should consider risk as being neither good nor bad but
rather defined as the “effect of uncertainty on objectives” (ISO, 2009a, p. 1).
3.1. Development of a risk-based MCS framework by extension of the LOC framework 109
95
Schall (1983) suggests defining organizational culture as “a relatively enduring, interde-
pendent symbolic system of values, beliefs, and assumptions evolving from and imper-
fectly shared by interacting organizational members that allows them to explain, coordi-
nate, and evaluate behavior and to ascribe common meanings to stimuli encountered in
the organizational context; these functions are accomplished through the mediation of
implicit and explicit rules that act as cultural warrants” (Schall, 1983, p. 557). Similarly,
Schein (2004) defines organizational culture as “a pattern of shared basic assumptions
that was learned by a group as it solved its problems of external adaptation and internal
integration, that has worked well enough to be considered valid and, therefore, to be
taught to new members as the correct way to perceive, think, and feel in relation to those
problems” (Schein, 2004, p. 17).
96
Specifically, Kaplan & Mikes (2016) identify a three-part-solution for effective risk man-
agement: agreement about the beliefs systems, formulation of the risk appetite of the
company, and monitoring of risk-taking behavior.
3.1. Development of a risk-based MCS framework by extension of the LOC framework 111
97
In identifying the indicators, the deviation from validated indicators of respective scales
was designed to be kept to a minimum. Thus, the emphasis on values in respect of the
scale for beliefs systems and the emphasis on inappropriate behavior in respect of the
scale for boundary systems (see Section 4.3.4) were replaced by the concepts of risk
appetite and risk limits. In addition, the notion of communication of respective concepts
on the one hand and awareness on the other hand was adopted. Finally, in order to de-
sign a reflective scale with four indicators for the construct risk-based formal controls,
two reflective indicators derived from each construct have been applied.
112 3. Theory Development and Hypotheses
98
In identifying the indicators, the deviation from validated indicators of respective scales
was designed to be kept to a minimum. Thus, indicators of the scales for diagnostic con-
trol systems and interactive control systems, respectively, have been amended by the
concept of consideration of risk information (see Section 4.3.4). Finally, in order to de-
sign a reflective scale with four indicators for the construct risk-based use of controls,
two reflective indicators derived from each construct have been applied.
3.2. Development of propositions 113
According to Kaplan & Mikes (2012), the “first line of defense against
preventable risk events is to provide guidelines clarifying the company’s
goals and values” (Kaplan & Mikes, 2012, p. 52). This is necessary be-
cause an organization “cannot anticipate every circumstance or conflict of
interest that an employee might encounter” (Kaplan & Mikes, 2012, p.
52). Therefore, a mission statement is crucial to communicate which in-
terests should dominate and take precedence in any situation (Kaplan &
99
Mikes, 2012). A well-crafted mission statement not only inspires the
workforce and clarifies the fundamental purpose of an organization (Si-
mons, 1995b), but contributes in “guiding people’s behaviors and deci-
sions toward desired norms” (Kaplan & Mikes, 2012, p. 51). Furthermore,
for a mission statement to be effective, it needs to be communicated to
and understood by all employees and articulate the values of an organi-
zation towards all its relevant stakeholders, such as customers, suppliers,
employees, communities, and shareholders (Kaplan & Mikes, 2012).
These core values that are linked to business strategy are the key feature
of beliefs systems which aim to provide inspiration and organizational
direction (Simons, 1995b).
In addition to guiding behavior through inspirational beliefs systems, for-
mal statements of organizations also “help employees avoid violating the
company’s standards and putting its reputation and assets at risk”
(Kaplan & Mikes, 2012, p. 52). Therefore, Kaplan & Mikes (2012) also
indicate a rules-based compliance approach for identifying and managing
preventable risks. For this purpose, boundary systems such as corporate
codes of business conduct “prescribe behaviors relating to conflicts of
interest, antitrust issues, trade secrets and confidential information, brib-
ery, discrimination, and harassment” (Kaplan & Mikes, 2012, p. 53). In
her study, Widener (2007) argues that organizations “use both the beliefs
and boundary systems to manage risk since they help ensure the align-
ment of employee behavior, which minimizes the possibility that the or-
ganization can be harmed” (Widener, 2007, p. 763) and found operations
99
Simons (1995b, p. 35) and Kaplan & Mikes (2012, p. 52) both quote Johnson & John-
son’s credo for an example of a well-crafted mission statement which begins with the fol-
lowing sentence: “We believe our first responsibility is to the doctors, nurses and pa-
tients, to mothers and fathers and all others who use our products and services”.
3.2. Development of propositions 115
100
risks to be associated with the emphasis placed on beliefs systems
(Widener, 2007).
To sum up, Kaplan & Mikes (2012) propose an integrated culture-and-
compliance model as a control model for preventable risks and suggest
the following activities: “Develop mission statement; values and belief
systems; rules and boundary systems; standard operating procedures;
internal controls and internal audit” (Kaplan & Mikes, 2012, p. 55). Thus,
the following hypotheses are proposed:
100
Widener (2007) uses Simons’ (2000) definition of operations risk which is very similar to
Kaplan & Mikes’ (2012) conceptualization of preventable risks. While Widener proposes
that operations risks are associated with beliefs and boundary systems of an organiza-
tions, for the latter significant relationships could not be demonstrated (Widener, 2007).
101
The process of taking action just in case there is a deviation from preset standards is
called management by exception and is a key feature of diagnostic control systems (Si-
mons, 2000).
116 3. Theory Development and Hypotheses
After discussing the design attributes and attention patterns of MCS, the
association between preventable risks and risk-based dimensions of
MCS is argued. Kaplan (2014) proposes that “[c]lear and frequently
communicated Beliefs and Boundary Systems” (Kaplan, 2014, p. 5) play
essential roles in managing preventable risks and that reliance on formal
controls is thus indicated in such circumstances (Kaplan & Mikes, 2012).
Additionally, it is also expected that risk-based formal controls provide an
effective control model for management of preventable risks with regards
to the risk-based dimensions of the extended risk-based MCS framework.
Thus, the following hypothesis is proposed:
H1d: Firms facing preventable risks to a greater extent will focus on risk-
based formal controls more intensively than firms facing preventa-
ble risks to a lower extent.
H2a: Firms facing strategy execution risks to a greater extent will focus
on beliefs systems more intensively than firms facing strategy exe-
cution risks to a lower extent.
es” (Kaplan & Mikes, 2012, p. 55). For example, while Simons (1991)
provided evidence for the association of the use of interactive controls
and product introduction as well as technology risks, Bisbe & Otley
(2004) found evidence that interactive control systems enhance the effect
102
of innovation risks on firm performance. In addition, Abernethy &
Brownell (1999) studied the effect of interactive use of budgets in times of
strategic change and found a positive effect on performance.
With regards to the relation between interactive and diagnostic control
systems, Simons points out that “the information and learning generated
by interactive systems can be embedded in the strategies and goals that
are monitored by diagnostic control systems” (Simons, 2000, p. 303).
Similarly, Chenhall & Morris (1995) argue that structure is necessary for
interactive control systems to be effective. In her study, Widener (2007)
argues that “firms will use both an interactive and diagnostic system, and
the more top managers rely on the interactive control system, the more
they will rely on the diagnostic control system“ (Widener, 2007, p. 762). In
addition, she claims that: “the diagnostic system provides the structure
that enables the interactive system to be effective” (Widener, 2007, p.
782). This conclusion is also consistent with Henri’s (2006a) argument
that “the dynamic tension generated by the joint effect of diagnostic and
interactive use” (Henri, 2006a, p. 548) ensures positive effects of interac-
tive use. Therefore, it is argued that, for interactive control systems to be
effective, diagnostic control systems must already be in place. Thus, the
following hypotheses are proposed:
H2b: Firms facing strategy execution risks to a greater extent will focus
on diagnostic control systems more intensively than firms facing
strategy execution risks to a lower extent.
102
Bisbe & Otley (2004) define product innovation by referring “to the development and
launching of products which are in some respect unique or distinctive from existing prod-
ucts” (Bisbe & Otley, 2004, p. 711). These risks are connected to R&D activities (Aber-
nethy & Brownell, 1997; Bisbe & Otley, 2004) which have been argued to being typical
strategy execution risks (see Section 2.1.3.2).
3.2. Development of propositions 119
H2c: Firms facing strategy execution risks to a greater extent will focus
on interactive control systems more intensively than firms facing
strategy execution risks to a lower extent.
H2d: Firms facing strategy execution risks to a greater extent will focus
on risk-based formal controls more intensively than firms facing
strategy execution risks to a lower extent.
H2e: Firms facing strategy execution risks to a greater extent will focus
on risk-based use of controls more intensively than firms facing
strategy execution risks to a lower extent.
H3a: Firms facing external risks to a greater extent will focus on diag-
nostic control systems more intensively than firms facing external
risks to a lower extent.
H3b: Firms facing external risks to a greater extent will focus on interac-
tive control systems more intensively than firms facing external
risks to a lower extent.
103
Taleb (2007) uses the term black swan as a synonym for a rare event that comes at a
surprise and has a major impact. Taleb provides an analogy to explain the black swan
problem: “Reality is far more vicious than Russian roulette. First, it delivers the fatal bul-
let rather infrequently, like a revolver that would have hundreds, even thousands of
chambers instead of six. After a few dozen tries, one forgets about the existence of a
bullet, under a numbing false sense of security” (Taleb, 2007, p. 28).
3.2. Development of propositions 121
H3c: Firms facing external risks to a greater extent will focus on risk-
based use of controls more intensively than firms facing external
risks to a lower extent.
104
See Malmi & Brown (2008) for a review of the relevant literature.
122 3. Theory Development and Hypotheses
ined Simons’ levers of control and found all four LOC to be interdepend-
ent and complementary. In addition, in line with Simons’ (1995b, 2000)
argument she found that the “result suggests that managers must con-
sider all four control systems when designing their control system” (Wid-
ener, 2007, p. 782) in order to increase effectiveness and thus organiza-
tional performance.
In this study, a contribution to the literature of studying MCS as a pack-
age and adopting a more comprehensive and integrated approach will be
made by examining the association between different types of control
systems to derive those MC practices, i.e. configurations of MCS, which
are put in place. Therefore, this study aims to investigate whether control
systems are not only relevant independently but particularly when ana-
lyzed simultaneously as a package. Based on Simons’ (1995b) theoreti-
cal conceptualization and the results of Widener (2007), it is expected
that the levers of control be found to be complementary and interdepend-
ent rather than substitutes for one another. Similarly, in regards to the
extended risk-based MCS framework, it is expected that all control sys-
tems and risk-based control systems also be found to be complementary
and interdependent rather than substitutes. However, in an explorative
sense, there will be no ex ante assumptions what these relationships look
like. This can be formally expressed as:
Again, this relationship is not only tested for the base MCS framework,
i.e. Simons’ LOC framework, but also for the risk-based MCS framework.
In a form similar to H5a, the following proposition is examined:
H6a: Firms with a better fit between their MCS and their risk profile ex-
perience a superior perceived firm performance.
H7a: Firms with a better fit between their MCS and their risk profile ex-
perience a greater perceived usefulness of its MCS.
H8a: Firms with a better fit between their MCS and their risk profile ex-
perience a higher level of EBIT.
H6b: Firms with a better fit between their risk-based MCS and their risk
profile experience a superior perceived firm performance.
H7b: Firms with a better fit between their risk-based MCS and their risk
profile experience a greater perceived usefulness of its MCS.
H8b: Firms with a better fit between their risk-based MCS and their risk
profile experience a higher level of EBIT.
Firms with a better fit between their risk-based MCS and their risk profile experi-
ence a …
H6b: superior perceived firm performance.
H7b: greater perceived usefulness of its risk-based MCS.
H8b: higher level of EBIT.
105
Empirical studies in management accounting rely on data, including field studies, case
studies, field experiments, laboratory experiments, and surveys, but excluding mathe-
matical modeling and computer simulations (Birnberg et al., 1990, p. 33).
106
The initial sample was drawn from the Bureau van Dijk database “Orbis” on April 11th
2016. According to the database, the last update for this data was performed on April 8 th
2016. In a first step, all available companies in Austria and Germany were selected,
which resulted in a number of 4,278,573 potential firms. As a next step, the temporary
sample was restricted to companies with a primary NACE Rev.2 Code of C28, which
comprises companies in the field of mechanical engineering and led to a temporary
sample of 32,927 firms. As a final restriction, the search proceeded for medium-sized
companies with a number of employees between 50 and 250, which resulted in the final
initial sample of 2311 companies.
107
From the initial sample of 2311 companies, a number of 122 firms or 5.3% had to be
removed for the following reasons: (1) 57 firms or 2.5% due to duplicate listings, (2) 49
tion has been chosen to enable the use of a sample of rather homogene-
ous firms to control for certain aspects.
Firstly, cultural and institutional aspects have been shown to have a great
influence on the design of MCS (e.g., Abernethy & Chua, 1996; Birnberg
& Snodgrass, 1988; Chow et al., 1996; Hofstede, 1993; Morden, 1995;
Van der Stede, 2003). To control for this aspect, companies located in
Austria or Germany were chosen for the target population as they repre-
sent a satisfactory number of companies expected to operate within simi-
lar cultural contexts.
Secondly, to make the control system comparable, I decided to study a
single industry, as the research on risk-based control systems is still a
rather novel concept in literature (Otley, 2012), and it is expected that
both the design and use of MCS (Messner, 2016; Van der Stede, 2016)
and the risk profile (Kaplan & Mikes, 2012) can vary significantly between
different industries. Furthermore, the mechanical engineering industry
was chosen because of the importance and structure of the industry. In
Austria and Germany, the mechanical engineering industry represents
one of the largest industries in terms of companies and employees (e.g.,
VDMA, 2017).
Finally, it was decided to analyze medium-sized companies as this study
is following a single respondent approach surveying one person in a
company. At a certain organizational size, it is considered problematic to
assume that one manager has all necessary information regarding organ-
izational aspects and practices, while on the other hand, the sample firms
had to be large enough to ensure that a formal MCS was in place and
that all examined variables apply (Bouwens & Abernethy, 2000; Davila,
2005). In addition, it was recognized that the “role of MCS in smaller or
medium-sized entities has received little attention in the contingency-
based MCS literature” (Chenhall, 2007, p. 183). As a result, companies of
108
between 50 and 250 employees were selected in order to increase the
validity of this survey.
firms or 2.1% due to bankruptcy or winding up of business activities, and (3) 16 firms or
0.7% due to missing data.
108
The European Commission (2003) defines medium-sized enterprises as enterprises
which employ fewer than 250 persons with an annual turnover not exceeding EUR 50
4.2. Data collection 131
million and/or an annual balance sheet total not exceeding EUR 43 million, but with a
minimum of 50 employees and exceeding annual turnover and/or annual balance sheet
total of EUR 10 million.
109
For extensive reviews of the survey method in control systems research, see Van der
Stede et al. (2006) and Young (1996).
110
From these studies 98% are cross-sectional (Van der Stede et al., 2005).
132 4. Methods
111
For discussion of the terms ‘construct’ and ‘construct validity’ see Section 4.3.2.
4.2. Data collection 133
Link 1
Conceptual A B
Link 2 Link 3
Operational C D
Link 4
Link 5
Controls
4.2.2 Survey
Data was collected through a structured online questionnaire sent to the
CEOs of the 2189 companies in the final sample. The CEO was chosen
as informant as the position is considered to ensure a holistic perspective
as well as profound knowledge regarding the risk profile, MCS design
and use, and business strategy of the firm. The CEO thereby serves as a
proxy for the whole organization. The personalized email addresses of
the CEOs were hand-collected by scanning the homepages and search-
ing online sources in order to increase response rate. The questionnaire
included 52 items and consisted almost exclusively of closed questions
112
with use of a seven-point-Likert scale. The survey could be carried out
in German or English and took approximately 15 minutes to complete.
In general, items from previous research were used whenever possible.
Where necessary, new constructs have been carefully developed follow-
ing the guidelines from Bisbe et al. (2007) regarding conceptual specifica-
tion. Furthermore, the translation process from the English language, in
which the questionnaire was originally developed, to the German lan-
guage was monitored by two independent language experts to ensure
113
conformity. After completion, the questionnaire was pre-tested in two
steps by three academics and three practitioners, which resulted in minor
changes in terms of wording, layout, and structure of the questionnaire.
Additionally, interviews were conducted with two managing directors to
increase understandability of the survey. Finally, guidelines on survey
114
research presented by Dillman et al. (2014) were widely followed. In
112
For a complete presentation of survey constructs and items please refer to the Appendix.
113
Dillman et al. (2014) suggest several methods to test web-based questionnaires, i.e.
conducting expert reviews, cognitive interviews, experimental evaluations, and pilot stud-
ies. These procedures of quality control aim to ensure technical functionality as well as
obtain feedback in regard to wording, question order, visual design, and navigation. As a
consequence, the optimization of clarity and understandability and the avoidance of am-
biguity particularly increases face validity of the survey instrument (Dillman et al., 2014).
114
In addition to methods of testing questionnaires (see footnote above), guidelines for
designing surveys were also followed carefully, e.g., creating interesting welcome and
closing screens, deciding what number of questions should be presented simultaneously
and how they should be arranged, using a consistent layout and visually emphasizing
essential information, and allowing the respondents to back up in the survey or stop the
survey in order to finish it at a later time (Dillman et al., 2014).
4.2. Data collection 135
115
doing so, special attention was paid to question order, aiming at mini-
mizing measurement errors due to question order effects and motivating
respondents to complete the questionnaire (Dillman et al., 2014).
The survey was started in September 2016 and lasted until November
2016. As a first contact, the target persons for this survey were sent a
th
postal letter on September 19 explaining the project and its goals, the
necessity of addressing the CEOs directly and informing of the upcoming
online questionnaire via email in the following days. One week later, on
th
September 26 , an email was sent with a personalized link for accessing
116
the online questionnaire and referring to the postal letter. As an incen-
tive to respond, an executive summary of the findings was offered. In the
following weeks, four reminder emails were sent to the non-respondents
th
before closing the questionnaire on November 18 . This process was
inspired by Dillman et al. (2014) and resulted in a total of 362 responses
117
received and a response rate of 16.5% , which is comparable to the 10-
25% range reported in similar studies (e.g., Baines & Langfield-Smith,
2003; Henri, 2006a; Widener, 2007).
According to Visser et al. (2000), there are four distinct sources which
add up to the total survey error, i.e. coverage error, sampling error, non-
response error, and measurement error. While coverage error may arise
in the case of not covering the whole population, sampling error occurs
through random differences between sample and population (Visser et
al., 2000). Because of the approach of this investigation, both effects
should be of no concern for this study.
To ensure validity of results, non-response bias was examined by inves-
tigating sample characteristics and comparing respondents to non-
115
To avoid question order effects Dillman et al. (2014) recommend, among others, to
group related questions covering similar topics, to carefully choose the first question
which should be easy to answer and arouse interest, and to place sensitive questions at
the end of the survey (Dillman et al., 2014). Also, to reduce effects from respondents
guessing the relationships being studied, the dependent variables have been placed ear-
lier in the study than the independent variables in order to minimize common method bi-
as (Podsakoff et al., 2003).
116
The online questionnaire was conducted using the web-based tool 2ask
(www.2ask.com).
117
The response rate is calculated as the number of usable returned questionnaires divided
by the number of companies in the final sample, i.e. 362/2189 = 16.54%.
136 4. Methods
𝑋1 − 𝑋2
𝑡=
𝑠2 𝑠2
√ 𝑝+ 𝑝
𝑛1 𝑛2
Equation 1: t-statistic
The value of t is then compared against the maximum value one would
expect if the null hypothesis which hypothesizes that the samples have
119
been drawn from the same population was true. The results of the
independent-samples t-test show no statistically significant differences
120
between non-respondents (n = 1827) and respondents (n = 362) for
121
the financial measures annual turnover and total assets as proxies for
the size of the organizations (see Table 2). Furthermore, to examine
whether the location of respondents and non-respondents statistically
differ from each other, a chi-square test is conducted. Pearson’s chi-
square test is used to examine if there is a relationship between two cat-
egorical variables through comparing observed to expected frequencies
118
𝑋1 − 𝑋2 is the difference between the overall means of the two sample, 𝑠𝑝2 the pooled
variance estimate which is the weighted average variance in which each variance is mul-
tiplied by its degrees of freedom and then divided by the sum of the two degrees of free-
dom, and 𝑛𝑖 the corresponding sample size. In fact, the equation tests the null hypothe-
sis μ1 = μ2 (Field, 2013, p. 365-368).
119
Critical values of the t-distribution can be found in statistical literature, e.g., Field (2013,
p. 893).
120
The group of non-respondents comprise the 2189 firms in the final sample minus the 362
respondents.
121
These measures were obtained from The Bureau van Dijk database “Orbis” (see Section
4.2.3).
4.2. Data collection 137
122
The odds ratio that a firm located in Austria responded to the survey is calculated by the
number of respondents located in Austria divided by the number of non-respondents lo-
cated in Austria: oddsAUTresponse = 52 / 124 = 0.42. Similarly, the odds ratio that a firm lo-
cated in Germany responded to the survey is calculated by the number of respondents
located in Germany divided by the number of non-respondents located in Germany:
oddsGERresponse = 310 / 1825 = 0.17. Consequently, the odds ratio that a firm responded to
the survey if located in Austria than if located in Germany is calculated by dividing the
odds of response of an Austrian firm by the odds of response of a German firm: odds ra-
tio = 0.42 / 0.17 = 2.47.
123
The author suspects that the fact that the survey was presented as a project of an Aus-
trian university played a role in this outcome.
124
While there are different definitions for early-respondents and late-respondents in litera-
ture, this study follows the procedure conducted, for example, by Chenhall & Langfield-
Smith (1998) and Bedford et al. (2016) and attributes the first 20% of the responders as
early-respondents and the last 20% of the responders as late-respondents, respectively.
Therefore, both groups of respondents comprise of 73 firms, which is a rounded figure of
362*0.2 = 72.4.
138 4. Methods
2
(X (1) = 0.03, p = .868) showed no significant difference between early
and late respondents. Overall, the results, presented in Table 2 below,
support the absence of significant non-response and early-late-response
bias.
4.2.3 Database
In order to gain additional financial data for the companies which re-
sponded to the survey, the Bureau van Dijk database was used as a
source to collect further information. As mentioned above, the measures
annual turnover and total assets have been recorded for the firms in the
final sample to compare for differences between respondents and non-
respondents. In addition, in order to gain data for actual firm performance
and to be able to examine hypotheses H8a and H8b, the financial meas-
125
However, it is pointed out that the Harman’s single-factor test is in fact appealing but
goes along with several limitations. Most importantly, the “likelihood of finding more than
one factor increases as the number of variables increases. Thus, the single-factor test
becomes increasingly less conservative as the total number of variables increases”
(Podsakoff & Organ, 1986).
140 4. Methods
ure EBIT (earnings before interest and taxes) was collected for the re-
sponding firms.
The following table provides an overview of the figures recorded as well
126
as respective availability:
As the measure EBIT showed extreme scores resulting in a high value for
standard deviation, a percentage-based rule was applied (Field, 2013)
and the 10% of highest and lowest scores deleted in order to achieve a
trimmed measure for EBIT, i.e. EBIT_trimmed. As a result, outliers from
the measure EBIT were eliminated to reduce the impact of potential bias.
Construct ƞ1 Multidimensional
Construct ξ1
ɛ ɛ ɛ
1 2 3 S1 S2 S3
Reflective model Latent model
130
See Bisbe et al. (2007), Jarvis et al. (2003), and Law et al. (1998) for detailed guidelines
regarding which models apply in specific situations.
4.3. Variable measurement 145
ζ1 ζ1
Multidimensional
Construct ƞ1 Construct ƞ1
φ31
Formative model Emergent model
131
tent validity is therefore not necessarily undermined. In regard to form-
ative models (or higher-order emergent models in multidimensional set-
tings respectively), an incomplete set of indicators (or dimensions) would
lead to serious misspecifications problems as the full census of constitu-
tive indicators (or dimensions) is required. Consequently, content validity
would be undermined since the domain of the construct would not be
completely covered (Bisbe et al., 2007).
In addition, as the construct does not explain the variances-covariances
of the operational variables in formative models, “traditional reliability
evaluation tools based on internal consistency […] and others derived
from Classical Test Theory […] are meaningless, illogical and inappropri-
ate” (Bisbe et al., 2007, p. 803). Therefore, important statistical methods
such as factor analysis and Cronbach’s alpha “work on the implicit basis
on reflective models” (Bisbe et al., 2007, p. 806). Consequently, an ade-
quate specification of the epistemic relationship is a precondition in order
to avoid inappropriate conclusions about construct relationships (Bisbe et
al., 2007; Edwards & Bagozzi, 2000; Jarvis et al., 2003).
131
For further discussion of content validity, please refer to Section 4.3.2.
4.3. Variable measurement 147
132
Factor analysis is a technique to identify clusters of variables through reducing a set of
variables into smaller sets of dimensions (Field, 2013).
148 4. Methods
133
different points in time in order to test reliability (Field, 2013). An im-
portant foundation of reliability is traditional measurement error theory
which considers internal consistency and relative absence of systematic
errors in measurement (Bisbe et al., 2007). In this study, the reliability of
constructs was assessed through (1) use of well-established constructs
with various items whenever possible and (2) calculation of Cronbach’s α,
i.e. internal consistency of the scales (Field, 2013).
As the most common measure for reliability (Field, 2013), Cronbach’s α
(Cronbach, 1951) is calculated by multiplying the number of items
squared with the average covariance between items and dividing it by the
sum of all the item variances and item covariances (Field, 2013, p. 708):
𝑁 2 𝑐𝑜𝑣
𝛼= 2
∑ 𝑠𝑖𝑡𝑒𝑚 + ∑ 𝑐𝑜𝑣𝑖𝑡𝑒𝑚
Equation 2: Cronbach’s α
∑𝑛1 𝑖𝑡𝑒𝑚𝑖𝑛
𝐶𝑂𝑁𝑆𝑇𝑅𝑈𝐶𝑇𝑖 =
𝑛
Equation 3: Final score of construct measures
th
In this equation, CONSTRUCTi is the measure for the i construct varia-
th th
ble, itemin is the response value to the n item of the i construct variable,
th
and n the number of items of the i construct variable.
133
This kind of test, which is often unfeasible due to practical reasons is also called test-
retest reliability (Field, 2013, p. 13).
134
For an overview of the final items for construct measurement see Table 10, and for an
overview of all survey items see Appendix.
4.3. Variable measurement 149
135
According to Bisbe et al. (2007) a “construct is defined as multidimensional when it
refers to several distinct but related dimensions that can be connected parsimoniously
and meaningfully into one single holistic concept” (Bisbe et al., 2007, p. 803).
136
In addition, other aspects of emergent multidimensional constructs also exist (see Sec-
tion 4.3.1). Notably, these dimensions are not interchangeable and do not necessarily
covary, dropping a dimension would alter the conceptual domain of the construct, i.e.
Kaplan & Mikes’ (2012) risk profile, and the direction of causality flows from dimensions
to the construct (Bisbe et al., 2007; Jarvis et al., 2003; Law et al., 1998).
137
Widener measures operations risks which result “from a breakdown in core internal
business processes such as manufacturing or processing and impedes the firm’s ability
to implement its strategy” (Widener, 2007, p. 769), the definition of which is similar to
Kaplan & Mikes’ (2012) conceptualization of preventable risks.
138
In the original scale, Widener (2007) measures operations risks using four items, i.e.
indicators about safety, quality, reliability, and efficiency of operations. As the indicator
regarding efficiency of operations was shown to have the weakest loading in Widener’s
(2007) scale, it was dismissed in this study for the reflective construct PREVRISK result-
ing in a Cronbach α of.86 compared to.66 in Widener’s (2007) analysis.
139
This approach is in line with the definition of risk as “effect of uncertainty on objectives”
(ISO, 2009a, p. 1).
150 4. Methods
not at all (1) to to a great extent (7). Responses to these items are aver-
aged to form the final score for the construct (see Equation 3; for an
overview of final measurement items see Table 10). While the exact
meaning and conceptual domain is defined in Section 2.1.3.2,
PREVRISK is conceptually specified as a reflective construct as shown in
Figure 25, as the indicators are manifestations of an underlying construct
(Widener, 2007). As shown in Table 4 below, exploratory factor analysis
reveals the unidimensionality of PREVRISK and demonstrates that all
items chosen ex ante to measure PREVRISK load on one factor with
factor loadings greater than .88 and cross-loadings of less than .22. The
Cronbach’s α for PREVRISK is .86, which is clearly above the threshold
of .7 (Field, 2013) and indicates high reliability for this scale.
4.3. Variable measurement 151
Indicators
Risk Strategy
execution risks Expected returns
profile
Behavior of competitors
External
Behavior of customers
risks
Major political or
environmental changes
Factors (66.8%)
Preventable Strategy execution
risks risks
Risk profile
Quality of operations .891 .043
Safety of operations .884 -.124
Reliability of operations .885 .087
Risk appetite -.087 .806
Expected returns .214 .487
High-risk investments -.071 .829
Cronbach’s alpha .863 .544
Explained variance of the constructs is reported in parentheses in the top line.
For the ease of presentation, factor loadings > 0.300 are highlighted in bold.
Table 4 reports the results of exploratory factor analysis for all relevant
items of the risk profile, i.e. PREVRISK and STRATRISK, but excluding
the items for EXTRISK as the unidimensionality criterion does not apply
for formative constructs (Bisbe et al., 2007). As extraction method a prin-
141
cipal component analysis was conducted on the remaining 6 items with
140
Grabner & Speckbacher (2016) use a slightly adapted scale to measure environmental
uncertainty that is based on Moers (2006). Moreover, Bedford et al. (2016) use a similar
measure termed environmental dynamism adapted from Gordon & Narayanan (1984)
and based on a formative scale.
141
While in literature some authors strictly differentiate between factor analysis and princi-
pal component analysis, this study considers the latter as legitimate extraction method to
conduct a factor analysis (Field, 2013).
154 4. Methods
142
oblique rotation (direct oblimin). All factors with eigenvalue over Kai-
143
ser’s (1960) criterion of 1 have been extracted, resulting in two factors
which explain 66.84% of the variance. As the reliability of factor analysis
depends on sample size (Field, 2013), the Kaiser-Meyer-Olkin measure
144
of sampling adequacy (KMO) has been examined (Kaiser, 1970). For
this factor analysis, the KMO statistic is .71 and all KMO values for indi-
vidual items were greater than .58 which is above the minimum criterion
of .5 (Field, 2013) and indicates the sample adequacy to be appropriate.
142
Factor rotation uses various methods to discriminate between factors and effectively
rotate initial axes in order to score variables to be maximally loaded to only one factor
(Field, 2013). Generally there are two types of rotation, i.e. orthogonal rotation and
oblique rotation. While the former method presumes independent factors, with oblique
rotation methods factors are allowed to correlate (Field, 2013). As the factors are ex-
pected to be related and not independent, oblique rotation is chosen, with direct oblimin
to be one method to do so.
143
According to a recommendation by Kaiser (1960) all factors with eigenvalue greater than
1 should be retained as they represent the amount of variation explained by a factor and
a value of 1 is regarded as substantial amount of variation (Field, 2013).
144
The KMO statistic which varies between 0 and 1 reports the ratio of squared correlation
to squared partial correlation between variables with a threshold of .5 for acceptable
sampling adequacy (Field, 2013; Kaiser, 1970). High values close to 1 for the KMO sta-
tistic indicate compact patterns of correlations and expect factors to be distinct and relia-
ble (Field, 2013).
145
For definition of and further information about multidimensional constructs please refer to
Section 4.3.1.
146
In addition, other aspects of emergent multidimensional constructs also exist (see Sec-
tion 4.3.1). Notably, the dimensions are not interchangeable and do not necessarily
4.3. Variable measurement 155
Dimensions Indicators
covary, dropping a dimension would alter the conceptual domain of the construct, i.e.
Simons’ LOC framework, and direction of causality flows from dimensions to the con-
struct (Bisbe et al., 2007; Jarvis et al., 2003; Law et al., 1998).
156 4. Methods
Indicators
Dimensions
Progress towards goals
Beliefs
systems
Monitor results
Reflective construct
Compare outcomes
Boundary
Emergent model systems
Review key measures
Simons‘ Reflective construct
MCS
Interactive use at the
Diagnostic management level
Multidimensional control systems
construct
Reflective construct Face-to-face challenge and debate
147
Bisbe et al. (2007) argue that an interactive control system “is a higher-order emergent
multidimensional construct formed by a series of defining constitutive dimensions” (Bisbe
et al., 2007, p. 808), i.e. intensive use by top management, intensive use by operating
managers, face-to-face challenge and debate, focus on strategic uncertainties, as well
as non-invasive, facilitating and inspirational involvement (Bisbe et al., 2007, p. 809).
The indicators for the interactive control systems in this study were largely chosen by fol-
lowing the argumentation of Bisbe et al. (2007), and incorporate aspects of other pub-
lished scales (Henri, 2006a; Widener, 2007). Despite the view of Bisbe et al. (2007), the
conceptual specification of the epistemic relationship for interactive control systems as
reflective construct is widely used (e.g., Bisbe & Malagueno, 2009; Henri, 2006a; Wid-
ener, 2007) and resulted in a highly reliable Cronbach’s α of .92 for this study.
4.3. Variable measurement 159
Factors (79.5%)
Diagnostic Interactive
Beliefs Boundary
control control
systems systems
systems systems
Design attributes of MCS
Mission statement
.732 .123 .088 .043
communicates values
Top managers
.765 -.019 -.024 -.085
communicate values
Workforce is aware of values .902 -.013 .083 .122
Mission statement inspires .860 -.004 -.102 -.110
Defines appropriate behavior .066 .876 .093 .121
Informs about off-limits behavior -.059 .950 .059 .020
Communicates risks to be avoided .020 .742 -.114 -.233
Workforce aware of code of conduct .060 .896 -.007 .020
Table 5 above reports the results of exploratory factor analysis for all
relevant items of MCS, i.e. BELIEF, BOUND, DIAGNOST, and
INTERACT. As extraction method a principal component analysis was
conducted on the 16 items with oblique rotation (direct oblimin). All fac-
tors with eigenvalue over Kaiser’s (1960) criterion of 1 were extracted,
resulting in four factors which explain 79.46% of the variance. As the
reliability of factor analysis depends on sample size (Field, 2013), the
Kaiser-Meyer-Olkin measure of sampling adequacy (KMO) has been
160 4. Methods
examined (Kaiser, 1970). For this factor analysis, the KMO statistic is .91
and all KMO values for individual items were greater than .86 which is
well above the minimum criterion of .5 (Field, 2013), thus indicating the
sample adequacy to be appropriate.
Furthermore, in an extended risk-based MCS framework setting, two
additional dimensions are specified, considering the extent to which the
148 149
design attributes and attention patterns are risk-based. Therefore,
the risk-based MCS framework comprises six dimensions: beliefs sys-
tems, boundary systems, diagnostic control systems, interactive control
systems, risk-based formal controls, and risk-based use of controls (see
Section 3.1). Similarly to the LOC framework, the risk-based MCS
framework can conceptually be specified as an emergent multidimen-
sional construct, as all six dimensions are defining and constitutive facets
150
(see Figure 28 below).
148
Design attributes refer to the emphasis on beliefs and boundary systems (see also
Section 2.2.3.1 and 2.2.3.2).
149
Attention patterns refer to the use of controls, i.e. diagnostic or interactive use (see also
Section 2.2.3.3 and 2.2.3.4).
150
Similar to Footnote 146, also other aspects of emergent multidimensional constructs
exist (see Section 4.3.1 above).
4.3. Variable measurement 161
Dimensions
Beliefs
systems
Reflective construct
Indicators
Boundary
systems Communication of risk appetite
Reflective construct
Emergent Workforce is aware of risk appetite
model Diagnostic
control
systems
Risk-based Communication of risk limits
Reflective construct
MCS
Interactive
Multi- control Workforce is aware of risk limits
dimensional systems
construct Reflective construct
Correct deviations
Risk-based considering risk information
formal
controls Review key measures
Reflective construct considering risk information
Factors (73.9%)
Risk-based Risk-based
formal controls use of controls
Risk-based dimensions of MCS
Communication of risk appetite .862 -.013
Workforce is aware of risk appetite .890 -.132
Communication of risk limits .782 .143
Workforce is aware of risk limits .776 .145
Correct deviations considering risk information .018 .893
Review key measures considering risk information -.001 .853
Face-to-face challenge and debate
-.066 .921
considering risk inform
Considering risk information when getting involved .108 .764
4.3.5 Strategy
According to Porter (1985), competitive strategy is “the search for a fa-
vourable competitive position in an industry” (Porter, 1985, p. 1) leading
to unique competitive advantages which enable profitable and sustaina-
ble positions against the forces that determine industry competition (Dent,
1990; Porter, 1985). While there are several classifications of strategy in
literature (e.g., Dent, 1990), this study follows Porter’s (1980, 1985) con-
164 4. Methods
151
For a definition of and for further information regarding multidimensional constructs,
please refer to Section 4.3.1.
152
In addition, other aspects of emergent multidimensional constructs are also present (see
Section 4.3.1 above). Notably, the dimensions are not interchangeable and do not nec-
essarily covary, dropping a dimension would alter the conceptual domain of the con-
struct, i.e. Porter’s (1980, 1985) conceptualization of strategy, and the direction of cau-
sality flows from dimensions to the construct (Bisbe et al., 2007; Jarvis et al., 2003; Law
et al., 1998).
4.3. Variable measurement 165
Indicators
Factors (60.3%)
Cost leadership Differentiation
strategy strategy
Strategy
Low cost of service .765 -.090
Cost efficient procedures .870 .010
High asset utilization .733 .118
Introduction of new services or products .188 .733
Uniqueness of services or products .021 .838
Range of services or products offered -.105 .617
153
Although this value is below the threshold of .7 (Field, 2013) it is above values reported
in similar studies (e.g., Judge et al., 1999).
154
Factor analysis was redone after dismissal of items low cost of services and range of
services or products offered with no significant changes in results and sufficient loadings
and lack of cross-loadings.
4.3. Variable measurement 167
Table 7 above reports the results of exploratory factor analysis for all
relevant items of strategy, i.e. COSTSTRAT and DIFFSTRAT. As extrac-
tion method a principal component analysis was conducted on the 6
items with oblique rotation (direct oblimin). All factors with eigenvalue
over Kaiser’s (1960) criterion of 1 have been extracted which resulted in
two factors which explain 60.31% of the variance. As the reliability of
factor analysis depends on sample size (Field, 2013), the Kaiser-Meyer-
Olkin measure of sampling adequacy (KMO) has been examined (Kaiser,
1970). For this factor analysis, the KMO statistic is .62 and all KMO val-
ues for individual items were greater than .58 which is above the mini-
mum criterion of .5 (Field, 2013) and indicates the sample adequacy to
be appropriate.
4.3.6 Performance
As a proxy for organizational performance, two survey constructs are
155
used in this study. First, the perceived firm performance (PERCPERF)
156
is measured using a scale adapted from Henri (2006a) and based on
Govindarajan (1988) and Govindarajan & Fisher (1990), consisting of
indicators regarding sales volume, return on investment, and results of
the organization. Based on Henri’s (2006a) conceptualization, respond-
ents were asked to rate the performance on the indicators of the organi-
zation compared to the strongest competitors in the last three years on a
seven-point Likert scale ranging from not at all satisfactory (1) to out-
standing (7). Responses to the items were averaged to form the final
score for the construct (see Equation 3; for an overview of final meas-
urement items see Table 10). While the construct measures the subjec-
tive firm performance on financial measures, PERCPERF is conceptually
specified as a reflective construct as shown in Figure 30, as the indica-
tors are manifestations of an underlying construct (Henri, 2006a). As
155
In literature, some authors argue that neither objective nor subjective performance
measures are superior for valid and reliable performance assessment (e.g., Dess & Rob-
inson, 1984; Henri, 2006a; Venkatraman & Ramanujam, 1987). In practice, in the con-
text of questionnaires, objective performance measures are often difficult to access.
156
For this study there have been minor changes to the wording of the published construct
by Henri (2006a).
168 4. Methods
Indicators
Sales volume
Perceived firm
Return on investment
performance
Reflective construct
Result
157
For measuring the reflective construct in the scale published, Sandino (2004, 2007) uses
5 items, which were reduced to 4 in this study due to efficiency reasons.
4.3. Variable measurement 169
Indicators
Reflective construct
Ability to operate timely and efficiently
Table 8 below reports the results of exploratory factor analysis for all
relevant items of performance, i.e. PERCPERF and USEFULMCS. As
extraction method a principal component analysis was conducted on the
7 items with oblique rotation (direct oblimin). All factors with eigenvalue
over Kaiser’s (1960) criterion of 1 have been extracted which resulted in
three factors which explain 81.35% of the variance. As the reliability of
factor analysis depends on sample size (Field, 2013), the Kaiser-Meyer-
Olkin measure of sampling adequacy (KMO) has been examined (Kaiser,
1970). For this factor analysis, the KMO statistic is .70 and all KMO val-
ues for individual items were greater than .61 except for one item, i.e.
ability to operate timely and efficiently, which was dismissed from the final
158
measurement model. Therefore, in exceeding the minimum criterion of
158
After dismissal of the item ability to operate timely and efficiently, the exploratory factor
analysis was redone and disclosed no significant changes in the results.
170 4. Methods
.5 (Field, 2013) for the remaining items, the sample adequacy is indicated
to be appropriate.
Factors (81.4%)
Perceived
Perceived firm Flexibility
usefulness
performance of MCS
of MCS
Performance
Sales volume .744 .042 -.027
Return on investment .940 -.019 -.001
Result .956 -.024 .017
Facilitation of company growth .093 .822 .080
Information to new threats and opportuni-
-.087 .932 -.060
ties
Protection from loss or excessive risk .014 .879 -.008
Ability to operate timely and efficiently -.016 .002 .997
159
Daily & Dollinger (1992) differentiate in their study in ‘professionally-managed firms’ and
‘family-owned and -managed firms’ and report significant differences across structural,
process, and performance dimensions. Furthermore, the definition of the European
Commission (2009) of family businesses includes, among other aspects, the require-
ment that at least one family member is formally involved in the governance of the firm
and that, at least for listed companies, 25 per cent of the decision-making rights are in
possession of the family. Thus, the operationalization of ownership structure in this study
reflects these aspects.
172
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
PREVRISK (1) .863
STRATRISK (2) .231 ---
EXTRISK (3) .405 .335 ---
BELIEF (4) .352 .228 .252 .852
BOUND (5) .349 .248 .278 .564 .907
Table 10: Descriptive statistics for survey items and constructs (continued)
174 4. Methods
th th
In this model, Yi is the i observation of the outcome variable, Xni is the i
th th
observation of the n predictor variable, bn is the coefficient of the n
predictor variable, b0 is the constant, and εi is the error, i.e. the difference
th
between the predicted and the observed value of the outcome, for the i
observation (Field, 2013). Thus, the model can be described entirely by
the coefficients associated with each predictor and by a constant. These
160
parameters are estimated by minimizing the residual sum of squares,
which is commonly achieved through a mathematical technique called
ordinary least squares (OLS) resulting in a regression model that best fits
the data (Field, 2013).
2
In order to assess the goodness of fit of the model, the measure R is
computed and analyzed, which “represents the percentage of the varia-
160
The residual sum of squares describes the degree of inaccuracy when the best model is
fitted to the data and is calculated by using the differences between the regression mod-
el and observed data (Field, 2013).
176 4. Methods
tion in the outcome that can be explained by the model” (Field, 2013, p.
2
302). The measure R is calculated by dividing the sum of squares for the
model (SSM) by the total sum of squares (SST) (Field, 2013, p. 302):
𝑆𝑆𝑀
𝑅2 =
𝑆𝑆𝑇
The sum of squares for the model describes the improvement in predic-
tion resulting from using the regression model rather than the mean as a
base model which leads to a reduction in inaccuracy of the model by
fitting the regression model to the data. It is calculated by using the dif-
ferences between the mean value of the output and the regression line.
The total sum of squares is the sum of squared differences between ob-
served values and values predicted by the mean which is the most basic
2
model to the data. A higher value for R , which can theoretically reach a
maximum of 1 implying perfect predictability, therefore indicates a greater
explanatory power of the model (Field, 2013).
Another measure of “how much the model has improved the prediction of
the outcome compared to the level of inaccuracy of the model” (Field,
2013, p. 302) is the F-ratio, which is calculated by dividing the mean
161
squares for the model (MSM) by the residual mean squares (MSR):
𝑀𝑆𝑀
𝐹=
𝑀𝑆𝑅
Equation 6: F-ratio
The magnitude of this F-ratio is evaluated by using critical values for the
162
corresponding degrees of freedom of the F-distribution. In addition, the
161
The mean squares for the model are calculated by dividing the model sum of squares by
the degrees of freedom, which is the number of variables in the model. The residual
mean squares are calculated by dividing the residual sum of squares by the degrees of
freedom, which is the number of observations minus the number of parameters being es-
timated (Field, 2013).
162
Critical values of the F-distribution can be found in statistical literature (e.g., Field, 2013,
p. 894).
4.4. Data analysis 177
2
F-statistic can be used to evaluate the significance of R by testing the
2
null hypothesis that R is zero which would imply no improvement in the
sum of squared errors after fitting the model (Field, 2013, p. 303):
(𝑁 − 𝑘 − 1)𝑅2
𝐹=
𝑘(1 − 𝑅2 )
Equation 8: Fchange-statistic
163
An alternative measure to compare models is the Akaike information criterion (AIC)
which penalizes the model for having more variables and thus approaches the issue that
the measure R2 always goes up if more predictors are added (Field, 2013).
178 4. Methods
there was no effect and hence the null hypothesis was found to be true
(Field, 2013). In assessing significance, the significance level α is crucial
which “represents the probability the researcher is willing to accept that
the estimated coefficient is classified as different from zero when it actu-
ally is not” (Hair et al., 2014, p. 156). According to Hair et al. (2014) the
typical significance level α is .05, that is the probability, i.e. the p-value,
that the observed value of t would occur if the value of b was, in fact,
zero, and the p-value has to be below this critical value in order to be
significant (Field, 2013; Hair et. al, 2014). Furthermore, in interpreting the
effects of different predictors on the outcome variable the β coefficients –
which are the standardized regression coefficients and thus report the
number of standard deviations the outcome changes as a result of one
standard deviation change in the predictor – are crucial. Therefore, by
allowing for the comparison of different coefficients, the β coefficient pro-
vides insight into the importance of a predictor in the model (Field, 2013).
In order to be able to generalize the findings outside of the sample and
assess the quality of the model itself, several assumptions must be met,
i.e. additivity and linearity, independent errors, homoscedasticity, and
normally distributed errors (Field, 2013; Hair et al., 2014). First of all, the
assumption of additivity and linearity which demands the outcome varia-
ble to be “linearly related to any predictors and, with several predictors,
their combined effect [to be] best described by adding their effect togeth-
er” (Field, 2013, p. 311) is imperative for justifying applying the linear
model. Secondly, the assumption of independent errors implies that for
“any two observations the residual term should be uncorrelated” (Field,
2013, p. 311). This eventuality, which is also referred to as a lack of auto-
correlation, is necessary for confidence intervals and significance tests to
be valid. Furthermore, the assumption of homoscedasticity means that
the variance of the residual term should be constant at each level of the
predictor variables. Finally, the assumption of normally distributed errors
is met if the residuals in the model are random, normally distributed vari-
ables with a mean of zero (Field, 2013; Hair et al., 2014).
Furthermore, there are several other conditions for using multiple regres-
sions models (Field, 2013). First of all, the predictor variables must be
measured at the interval level or categorical with two categories, they
4.4. Data analysis 179
1
𝑉𝐼𝐹 = ∗
1 − 𝑅2
Equation 9: Variance inflation factor (VIF)
2*
In this equation, R is the amount of the independent variable explained
2*
by all remaining predictors and (1-R ) is a measure of tolerance. There-
fore, to examine this relationship, each independent variable is made a
dependent variable, which is regressed against all remaining predictors.
The square root of the VIF measure indicates the degree to which the
164
standard error increases due to multicollinearity.
The outcome variables, on the other hand, should be independent and
must be measured at the interval level, continuously and with no con-
straints on the variability.
164
For example, if 75% of the variance of an independent variable is explained by other
independent variables, the measure of tolerance is .25, the value of VIF is 4 and the
square root of VIF is 2, indicating that the standard error has been doubled due to multi-
collinearity (Hair et al., 2014).
180 4. Methods
165
Hierarchical procedures “involve a series of n - 1 clustering decisions (where n equals
the number of observations) that combine observations into a hierarchical or a treelike
structure” (Hair et al., 2014, p. 439). While in agglomerative methods each observation
starts as its own cluster and most similar clusters are joined until a single cluster re-
mains, in divisive methods the starting point is a single cluster which is divided until each
observation is a single-member cluster (Hair et al., 2014).
166
Nonhierarchical clustering procedures “assign objects into clusters once the number of
clusters is specified“ (Hair et al., 2014, p. 443) and thus do not involve treelike construc-
tion processes as in hierarchical clustering procedures (Hair et al., 2014).
4.4. Data analysis 181
1
𝑃(𝑌) =
1 + 𝑒 −(𝑏0+𝑏1𝑋1𝑖 +𝑏2 𝑋2𝑖 +⋯+𝑏𝑛 𝑋𝑛𝑖)
Equation 11: Logarithmic regression model
th
In this equation, P(Yi) is the probability that Y occurs for the i case, and
th
Yi the actual outcome for the i case. Therefore, the smaller the values of
log-likelihood, the less unexplained observations occur. Closely related to
the log-likelihood measure is the deviance which is referred to -2LL and
has a chi-square distribution (Field, 2013, p. 763):
184 4. Methods
While the baseline model only features a constant which displays the
frequency with which the outcome occurred and thus provides the best
prediction only if the outcome is known, the improvement of logistic re-
gression models incorporating predictor variables can be examined hier-
archically by adding these variables and assessing the corresponding
likelihood-ratio which has a chi-square distribution with degrees of free-
dom equaling the difference of the number of parameters of both models.
2
Finally, and in a manner similar to the measure R in linear regressions,
there are some measures in logistic regression analysis which provide
information on the substantive significance of the model and thus of how
2
well the model fits the data. For example, Nagelkerke’s R N reflects “the
amount of variation accounted for by the logistic model” (Hair et al., 2014,
p. 324) and is based on the deviance of the new and base model, as well
as the sample size (Field, 2013, p. 766):
(−2𝐿𝐿(𝑛𝑒𝑤) − (−2𝐿𝐿(𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒)))
1 − 𝑒𝑥𝑝( )
2
𝑅𝑁 = 𝑛
−2𝐿𝐿(𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒)
1 − 𝑒𝑥𝑝(− )
𝑛
Equation 15: Goodness of fit measure Nagelkerke’s RN2
2
The measure RN varies between 0 and 1 with the latter indicating perfect
outcome predictability and thus perfect model fit (Field, 2013; Hair et al.,
2014).
4.4. Data analysis 185
167
In contrast to the term probability, odds are defined as “the probability of an event occur-
ring divided by the probability of that event not occurring” (Field, 2013, p. 767).
186 4. Methods
Risk profile RQ1: Design and use of MCS Simons’ MCS framework Strategy
Preventable risks Link 1 Beliefs systems Link 3 Size
Strategy execution risks Boundary systems Age
External risks FIT Diagnostic control systems Ownership
Interactive control systems
Kaplan & Mikes (2012) Link 2 Porter (1980)
Simons (1995b)
Conceptual
controls
Performance Risk-based use
se of
Link 4 Firm performance controls RQ2:
Link 7
Usefulness of MCS Packages
of MCS
MCS_categoryrym
4.5. Research framework of this study
rbMCS_categoryn
Otley (1980)
O Bisbe et al. (2007) Link 5
Link 6
Mulitple regressions Cluster analyses Items
Link 8
L Link 10
Discriminant analyses L Secondary data
Logistic regressions
Items Items
Auzair &
FIT = 1 Link 9
Widener (2007), Moers (2006) Widener (2007), Henri (2006a) Langfield-Smith (2005)
nal
if actual MCS
Operational
Items
In this chapter the results of the empirical analysis of this study are pre-
sented and discussed. As a consequence, evidence is provided for these
hypotheses which are supported by data. For analysis of data, the statis-
tical software IBM SPSS Statistics Version 24 is used. Subsequently, in
the first section risk profile contingent design and use of MCS will be ex-
amined in order to reveal associations between types of risks and specific
control systems. As a next step, the analysis focuses on risk profile con-
tingent packages of MCS to gain insight into MCS configurations which
are put in place in practice, and how and whether they are influenced by
specific types of risks. Finally, a risk profile contingent performance anal-
ysis is conducted, to shed light on the hypothesized proposition that firms
that have MCS design and use consistent with their risk profile show per-
formance superior to that of firms that have MCS design and use incon-
sistent with their risk profile.
Explanatory Explained
variables variables
Preventable
risks
Modeln
Multiple
Strategy regression
Control Control
execution
systemn variables
risks
݉݁ݐݏݕ݈ܵݎݐ݊ܥ ൌ ܾ ܾଵ ܵܧܼܫ ܾଶ ܧܩܣ ܾଷ ܱܹܰ ܾସ ܶܣܴܱܶܵܶܵܥ
External
risks ܾହ ܶܣܴܶܵܨܨܫܦ ܾ ܴܲܭܵܫܴܸܧ
ܾ ܴܵܶܭܵܫܴܶܣ ଼ܾ ܭܵܫܴܶܺܧ ߝ
Figure 33: Theoretical model for analyzing risk profile contingent design and use of MCS
(Source: own illustration)
In the theoretical model in Figure 33 above, the measures for the levers
of control, i.e. BELIEF, BOUND, DIAGNOST, and INTERACT, each
serve as dependent variables in Model 1 - 4 of the analysis. Furthermore,
in order to examine the risk profile contingent design and use of the risk-
based MCS framework, the measures for the risk-based dimensions, i.e.
rbFORMALMCS and rbUSEMCS, act as dependent variables in Model 5
- 6 of the analysis. The main independent variables in these models are
the risk profile variables, i.e. PREVRISK, STRATRISK, and EXTRISK. To
improve validity of the models, a set of control variables is included. First
5.1. Risk profile contingent design and use of MCS 191
𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑆𝑦𝑠𝑡𝑒𝑚𝑛𝑖 = 𝑏𝑛0 + 𝑏𝑛1 𝑆𝐼𝑍𝐸𝑖 + 𝑏𝑛2 𝐴𝐺𝐸𝑖 + 𝑏𝑛3 𝑂𝑊𝑁𝑖 + 𝑏𝑛4 𝐶𝑂𝑆𝑇𝑆𝑇𝑅𝐴𝑇𝑖
+ 𝑏𝑛5 𝐷𝐼𝐹𝐹𝑆𝑇𝑅𝐴𝑇𝑖 + 𝑏𝑛6 𝑃𝑅𝐸𝑉𝑅𝐼𝑆𝐾𝑖 + 𝑏𝑛7 𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾𝑖
+ 𝑏𝑛8 𝐸𝑋𝑇𝑅𝐼𝑆𝐾𝑖 + 𝜀𝑛𝑖
th
In this equation, ControlSystemni is the i observation of the measure for
th
the n control system, resulting in Model 1 - 6 of the analysis. As a meth-
od for variables to be entered into the models, the hierarchical entry
168
method is chosen, in which the hierarchically predefined sets of varia-
bles are forced step by step into the model. This method requires a deci-
sion on the sequence of entering the predictors into the model, and thus
relies on theoretical reasoning (Field, 2013). According to Field (2013),
known predictors are entered into the model first, in order to assess the
influence of newly tested variables. Therefore, in the first step of each of
the multiple regression analyses, the control variables SIZE, AGE, and
168
Other methods for entering variables into the model of a multiple regression analysis are,
for example, the forced entry method and stepwise methods. The forced entry method
forces all predictors into the model simultaneously, making no decision about the order
of the variables. In stepwise methods, for example forward or backward methods, purely
mathematical criteria decide in which order the predictors are entered into the model
(Field, 2013).
192 5. Results
OWN are introduced into the analyses, as these common structural vari-
ables have proven influential since the beginning of contingency theory
on organizational control systems and MCS (Chenhall, 2003; Davila,
2005). In a second step, the strategy variables COSTSTRAT and
DIFFSTRAT are entered into the models, as research has provided
strong evidence that strategy is related to the design and use of MCS
(Bedford et al., 2016; Langfield-Smith, 2007). Finally, the third and last
step of the analyses adds the risk profile variables PREVRISK,
STRATRISK, and EXTRISK in order to demonstrate the influence and
additional explanatory power of these predictors for MCS design and use.
This procedure takes into account that it is crucial for predictors to be
“based on a sound theoretical rationale or well-conducted past research”
(Field, 2013, p. 321). The significance of the model after each step is
2
assessed by using the F-statistic, which evaluates the significance of R
and thus the overall fit of the model as a whole. In addition, to evaluate if
a step has significantly improved the model, the measure Fchange is com-
2
puted, which assesses the significance of the change in R . Finally, the
significance and the impact of each independent variable is evaluated by
assessing the unstandardized regression coefficients bi and the standard-
ized regression coefficients βi.
In order to verify whether the assumptions of the multiple regression
analyses have been met (see Section 4.4.1), a number of tests were
conducted for each model. First of all, to test for the assumptions of line-
arity and homoscedasticity, the suggestions by Field (2013) are followed
and the plot of standardized residuals against standardized predicted
169
values as well as the partial plots of the residuals of the outcome vari-
170
able and each of the predictors were visually examined. To test the
169
In creating a scatterplot of the values of the residuals against the values of the outcome
it can be visually examined whether there is a systematic relationship between errors
and predictions of the model. Usually, these values are converted to z-scores, to obtain
the standardized values of the errors and the outcome (Field, 2013). If there is no non-
linearity or heteroscedasticity than “this graph should look like a random array of dots, if
the graph funnels out then that is a sign for heteroscedasticity and any curve suggests
non-linearity” (Field, 2013, p. 348).
170
Similar to assessing the plot of standardized residuals against standardized predicted
values (see Footnote 169), in visually examining the partial plots non-linear relationships,
heteroscedasticity, and influential outliers can be detected as well. In these plots, the
5.1. Risk profile contingent design and use of MCS 193
171
assumption of normally distributed errors, the histograms and P-P
172
plots were also visually examined. While in general no serious signs for
heteroscedasticity, non-linearity, or non-normality have been found, there
are a few graphs which show ambiguous results. Therefore, results of the
multiple regression analyses are validated by performing bootstrapping
173
analyses, which is a robust method to generate confidence intervals
and significance tests of the model parameters (Field, 2013). Field ex-
plains that the “main benefit of the bootstrap confidence intervals and
significance values is that they do not rely on assumptions of normality or
homoscedasticity” (Field, 2013, p. 352). As a result, all findings in each
model have been confirmed by bootstrapping analyses. Finally, the as-
sumption of independent errors is examined by conducting the Durbin-
Watson test, which tests if adjacent residuals are correlated and thus
tests for autocorrelation (Durbin & Watson, 1951; Field, 2013). The test
statistic can vary between 0 and 4, with a value of 2 indicating that the
residuals are uncorrelated, with values less than 1 and greater than 3
being problematic (Field, 2013). The Durbin-Watson test was computed
for each of the following models, i.e. Model 1 - 6 and Model 1A - 6A, with
a lowest value of 1.769 (Model 1) and a highest value of 1.946 (Model 6),
which means that there is no cause for concern in regard to independent
errors or autocorrelation.
clouds of dots should be evenly spaced around a line indicating homoscedasticity and
linear relationships (Field, 2013).
171
To test for normality of residuals the histograms, i.e. the frequency distributions, should
look like normal distributions, i.e. symmetrical and approximately bell-shaped.
172
The P-P plots (probability-probability plots) show the cumulative probability of a variable
against the cumulative probability of a particular distribution, which is the normal distribu-
tion in this case. When the actual z-score is plotted against the expected z-score, the
graph should ideally fall on the diagonal of the plot, which indicates normal distribution.
In case the line is consistently above or below the diagonal, then the kurtosis of the dis-
tribution deviates from normal distributions. Furthermore, a graph which is S-shaped in-
dicates an issue of skewness (Field, 2013).
173
Bootstrapping analysis estimates the properties of the distribution from the sample data.
In this study, each bootstrapping analysis draws 1000 parameter estimates, which are
calculated from each bootstrap sample. The limits within which 95% of the drawn pa-
rameter estimates fall are used to estimate limits of the 95% confidence interval of the
parameters, which is called a percentile bootstrap confidence interval. Furthermore, the
standard deviations of the parameter estimates can be calculated from the bootstrap
samples and used as standard error of the parameter estimates (Field, 2013).
194 5. Results
5.1.2 Analysis
The analysis for examining risk profile contingent design and use of MCS
starts with examining the relations for the LOC framework and proceeds
by applying the risk-based MCS framework. Finally, additional results for
both frameworks of MCS are provided.
Multiple regressions
LOC framework
Model 1 Model 2
Explanatory Explained variable: Explained variable:
variables BELIEF BOUND
Step 1:
b β p-value b β p-value
Control variables
Constant 4.295 3.714
SIZE 0.120 .113 .038* 0.260 .173 .001**
AGE 0.012 .010 .851 0.009 .005 .919
OWN 0.008 .004 .943 -0.574 -.195 .000***
Step 2:
+ Strategy profile
Constant 1.991 0,954
SIZE 0.064 .060 .247 0.190 .126 .014*
AGE 0.046 .039 .456 0.035 .021 .682
OWN 0.045 .021 .668 -0.508 -.173 .000***
COSTSTRAT 0.241 .220 .000*** 0.426 .274 .000***
DIFFSTRAT 0.180 .189 .000*** 0.085 .063 .222
Step 3:
+ Risk profile
Constant 0.897 -0.490
SIZE 0.029 .028 .587 0.146 .097 .053+
AGE 0.063 .053 .289 0,059 .035 .481
OWN 0.125 .060 .232 -0.407 -.138 .006**
COSTSTRAT 0.072 .066 .283 0.203 .131 .031*
DIFFSTRAT 0.108 .114 .037* -0.011 -.008 .879
PREVRISK 0.277 .223 .000*** 0.362 .206 .001**
STRATRISK 0.104 .125 .020* 0.112 .095 .071+
EXTRISK 0.076 .059 .311 0.129 .071 .219
The results of the multiple regression analyses show that Model 1 ex-
plains 17.5% of the variance in the outcome variable BELIEF and is sig-
2 2
nificant after step 2 (R = .120, F = 9.747, p < .001) and step 3 (R = .175,
F = 9.388, p < .001) of the analysis. Furthermore, Model 1 significantly
improved after step 2, i.e. after adding the strategy profile variables
2
COSTSTRAT and DIFFSTRAT (R change = .107, Fchange = 21.662, p <
.001), and significantly improved further after step 3, i.e. after adding the
2
risk profile variables PREVRISK, STRATRISK, and EXTRISK (R change =
.055, Fchange = 7.853, p < .001). After adding all predictors in step 3, the
variables PREVRISK (b = 0.277, t = 3.631, p < .001), STRATRISK (b =
0.104, t = 2.342, p = .020), and DIFFSTRAT (b = 0.108, t = 2.094, p =
.037) are significant predictors of BELIEF. In addition, the variable
PREVRISK (β = .223) shows the greatest impact in predicting the out-
come variable BELIEF, followed by STRATRISK (β = .125) and
DIFFSTRAT (β = .114).
The multiple regression analysis in Model 2 explains 20.6% of the vari-
2
ance in the outcome variable BOUND and is significant after step 1 (R =
2
.074, F = 9.473, p < .001), step 2 (R = .161, F = 13.626, p < .001), and
2
step 3 (R = .206, F = 11.421, p < .001) of the analysis. Specifically, after
step 2, i.e. after adding the strategy profile variables COSTSTRAT and
2
DIFFSTRAT, Model 2 significantly improved (R change = .087, Fchange =
18.469, p < .001), and significantly improved further after step 3, i.e. after
adding the risk profile variables PREVRISK, STRATRISK, and EXTRISK
2
(R change = .045, Fchange = 6.661, p < .001). After adding all predictors in
step 3, the variables PREVRISK (b = 0.362, t = 3.409, p = .001), OWN (b
= -0.407, t = -2.793, p = .006), and COSTSTRAT (b = 0.203, t = 2.165, p
= .031) are significant predictors of BOUND and the variable PREVRISK
(β = .206) shows the highest degree of importance in the model, followed
by OWN (β = -.138) and COSTSTRAT (β = .131).
The following table shows the results of the multiple regression analyses
on the attention patterns of the feedback and measurement systems of
MCS, i.e. the dependent variables DIAGNOST and INTERACT.
198 5. Results
Multiple regressions
LOC framework
Model 3 Model 4
Explanatory Explained variable: Explained variable:
variables DIAGNOST INTERACT
Step 1:
b β p-value b β p-value
Control variables
Constant 4.656 3.985
SIZE 0.226 .167 .002** 0.190 .137 .011*
AGE -0.033 -.022 .680 -0.014 -.009 .864
OWN -0.460 -.174 .001** -0.419 -.155 .003**
Step 2:
+ Strategy profile
Constant 1.550 0.727
SIZE 0.148 .110 .029* 0.108 .078 .121
AGE 0.001 .001 .989 0.022 .014 .778
OWN -0.393 -.149 .002** -0.349 -.129 .008**
COSTSTRAT 0.434 .311 .000*** 0.451 .316 .000***
DIFFSTRAT 0.140 .115 .023* 0.150 .120 .018*
Step 3:
+ Risk profile
Constant 0.521 -0.011
SIZE 0.115 .085 .089+ 0.091 .066 .190
AGE 0.015 .010 .840 0.032 .020 .680
OWN -0.280 -.106 .031* -0.263 -.097 .051+
COSTSTRAT 0.289 .207 .001** 0.358 .250 .000***
DIFFSTRAT 0.064 .053 .316 0.086 .069 .194
PREVRISK 0.151 .096 .111 0.016 .010 .870
STRATRISK 0.115 .109 .037* 0.010 .010 .854
EXTRISK 0.182 .111 .051+ 0.305 .182 .002**
Multiple regressions
Risk-based MCS framework
Model 5 Model 6
Explanatory Explained variable: Explained variable:
variables rbFORMALMCS rbUSEMCS
Step 1:
b β p-value b β p-value
Control variables
Constant 4.010 4.481
SIZE 0.168 .143 .008** 0.214 .178 .001**
AGE 0.009 .007 .903 -0.090 -.067 .206
OWN -0.336 -.146 .005** -0.422 -.179 .001**
Step 2:
+ Strategy profile
Constant 2.173 1.617 .002**
SIZE 0.121 .103 .050+ 0.141 .117 .018*
AGE 0.024 .018 .723 -0.064 -.048 .329
OWN -0.289 -.126 .013* -0.352 -.149 .002**
COSTSTRAT 0.298 .245 .000*** 0.452 .363 .000***
DIFFSTRAT 0.043 .041 .445 0.079 .073 .144
Step 3:
+ Risk profile
Constant 1.240 0.736
SIZE 0.094 .079 .129 0.117 .097 .049*
AGE 0.038 .029 .570 -0.051 -.038 .428
OWN -0.208 -.090 .080+ -0.265 -.113 .020*
COSTSTRAT 0.161 .132 .036* 0.329 .264 .000***
DIFFSTRAT -0.024 -.022 .689 0.012 .011 .832
PREVRISK 0.177 .129 .042* 0.118 .084 .156
STRATRISK 0.072 .077 .155 0.054 .057 .268
EXTRISK 0.150 .105 .080+ 0.211 .145 .010*
1𝑖𝑓𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾𝑖 ≥ 𝑀𝑒𝑑𝑖𝑎𝑛(𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾)
𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾_𝑑𝑢𝑚𝑚𝑦𝑖 = {
0𝑖𝑓𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾𝑖 < 𝑀𝑒𝑑𝑖𝑎𝑛(𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾)
All models follow the same procedures as presented in the previous sec-
tions. However, as step 1 and step 2 are identical to the results reported
in the corresponding models above, they are omitted in the following ta-
bles. For sake of clarity the subsequently presented models are labeled
with the same digit as the corresponding models above but with the suffix
174
A.
The table below shows the additional results of the multiple regression
analyses on the design attributes of formal controls of MCS, i.e. BELIEF
and BOUND.
174
For example, while the original results for beliefs systems are reported in Model 1, the
additional results for beliefs systems are reported in Model 1A.
204 5. Results
Multiple regressions
LOC framework
Model 1A Model 2A
Explanatory Explained variable: Explained variable:
variables BELIEF BOUND
Step 3:
b β p-value b β p-value
+ Risk profile
Constant 1.153 -0.254
SIZE 0.039 .037 .468 0.158 .105 .037*
AGE 0.059 .050 .318 0.056 .033 .502
OWN 0.135 .065 .195 -0.415 -.141 .005**
COSTSTRAT 0.072 .065 .286 0.207 .133 .028*
DIFFSTRAT 0.106 .111 .041* -0.010 -.008 .886
PREVRISK 0.281 .227 .000*** 0.367 .209 .001**
STRATRISK_dummy 0.325 .150 .004** 0.246 .080 .116
EXTRISK 0.076 .059 .312 0.138 .076 .185
The table below shows the additional results of the multiple regression
analyses on the attention patterns of the feedback and measurement
systems of MCS, i.e. DIAGNOST and INTERACT.
Multiple regressions
LOC framework
Model 3A Model 4A
Explanatory Explained variable: Explained variable:
variables DIAGNOST INTERACT
Step 3:
b β p-value b β p-value
+ Risk profile
Constant 0.794 0.105
SIZE 0.125 .093 .061+ 0.090 .065 .190
AGE 0.011 .007 .881 0.028 .018 .716
OWN -0.274 -.104 .035* -0.219 -.081 .101
COSTSTRAT 0.289 .207 .001** 0.346 .242 .000***
DIFFSTRAT 0.062 .051 .332 0.077 .062 .242
PREVRISK 0.155 .099 .099+ 0.015 .009 .877
STRATRISK_dummy 0.332 .120 .017* 0.275 .097 .055+
EXTRISK 0.184 .112 .048* 0.281 .167 .004**
Multiple regressions
Risk-based MCS framework
Model 5A Model 6A
Explanatory Explained variable: Explained variable:
variables rbFORMALMCS rbUSEMCS
Step 3:
b β p-value b β p-value
+ Risk profile
Constant 1.453 0.900
SIZE 0.099 .084 .103 0.121 .100 .040*
AGE 0.034 .026 .609 -0.055 -.041 .398
OWN -0.184 -.080 .119 -0.245 -.104 .031*
COSTSTRAT 0.156 .128 .041* 0.324 .260 .000***
DIFFSTRAT -0.029 -.027 .620 0.007 .007 .895
PREVRISK 0.179 .130 .038* 0.119 .085 .149
STRATRISK_dummy 0.321 .133 .012* 0.253 .103 .038*
EXTRISK 0.139 .098 .099+ 0.202 .138 .013*
Finally, in line with the results of the previous models, the additional re-
sults of the multiple regression analyses of Model 5A and Model 6A on
rbFORMALMCS and rbUSEMCS in Table 16 show that the significance
of the improvement of the models after each step and the significance of
the models as a whole remain unchanged too. However, in Model 5A the
variable STRATRISK_dummy (b = 0.321, t = 2.540, p = .012) is a signifi-
cant predictor and shows the greatest impact (β = .133) on predicting the
outcome variable rbFORMALMCS in contrast to the results of the corre-
5.1. Risk profile contingent design and use of MCS 207
Model 1 / 1A
Beliefs
systems
(BELIEF) Cost leadership
strategy
Model 2 / 2A (COSTSTRAT)
Preventable
.206** Boundary
risks
(PREVRISK) systems
(BOUND) Differentiation
strategy
Model 3 / 3A (DIFFSTRAT)
Diagnostic
Strategy control systems
(DIAGNOST) Organizational
execution risks size
STRATRISK) (SIZE)
Model 4 / 4A
Interactive
control systems
(INTERACT) Company
age
External
Model 5 / 5A (AGE)
risks
(EXTRISK) R
Risk-based
formal controls
(rbFORMALMCS) Ownership
structure
Model 6 / 6A (OWN)
Risk-based
use of controls
(rbUSEMCS)
Figure 34: Graphical depiction of significant results on risk profile MCS design and use
(Source: own illustration)
The first set of hypotheses, i.e. H1a - H1d, predicts associations between
preventable risks and certain types of control systems. Firms facing pre-
ventable risks to a greater extent focus on beliefs systems (p < .001),
boundary systems (p < .01), and risk-based formal controls (p < .05)
more intensively than firms facing preventable risks to a lower extent,
which provides support for H1a, H1b, and H1d. On the other hand, H1c is
not supported, since diagnostic control systems are not associated with
5.1. Risk profile contingent design and use of MCS 209
preventable risks. These findings are consistent with Kaplan & Mikes’
(2012) argument that for controlling preventable risks, an integrated cul-
ture-and-compliance model is most effective. Furthermore, these results
are partly consistent with Widener’s (2007) study, which found that opera-
tions risks are associated with beliefs systems and diagnostic control
systems. Finally, also the additional results on MCS design and use
(Model 1A - 6A) provide unisonous support. To summarize, data support
hypotheses H1a, H1b, and H1d, but no evidence could be found for sup-
porting H1c.
The second set of hypotheses, i.e. H2a - H2e, targets associations be-
tween strategy execution risks and certain types of control systems.
Firms facing strategy execution risks to a greater extent focus on beliefs
systems (p < .05) and diagnostic control systems (p < .05) more inten-
sively than firms facing strategy execution risks to a lower extent, which
provides support for H2a and H2b. Surprisingly, H2c, H2d, and H2e are
not supported by Model 1 - 6, since strategy execution risks were found
not to be associated with interactive control systems, risk-based formal
controls, and risk-based use of controls. These findings led to the produc-
tion of additional results on MCS design and use modifying operationali-
zation of strategy execution risks, since reliability of the measure may be
weak. Interestingly, the additional results on MCS design and use show
at least some support for H2c, since there is a strong tendency for signifi-
cance between the relationship of strategy execution risks and interactive
control systems (p = .055). In addition, Model 1A - 6A provide support for
H2d and H2e since the exposure to strategy execution risks is positively
associated with the emphasis on risk-based formal controls (p < .05) and
risk-based use of controls (p < .05). Finally, there is consistent support for
H2a and H2b by the additional results on MCS design and use as well. In
regard to H2a, the finding is consistent with Simons’ argument that beliefs
systems are intensively related to opportunity-seeking behavior (Simons,
1995b) and with Marginson’s findings that beliefs systems affect a com-
panies’ strategic climate and thus affect managers’ strategic endeavors in
influencing ideas and initiatives (Marginson, 2002). Furthermore, as there
is some evidence that strategy execution risks are associated with inter-
active control systems, the finding in regard to H2c is in line with Kaplan
210 5. Results
& Mikes’ (2012) argument that for controlling strategy execution risks
interactive discussions are essential. Similarly, Simons’ view that interac-
tive control systems focus on strategic uncertainties and changing cir-
cumstances (Simons, 1995b) is supported as well. In addition, there is
conformity with empirical findings of Simons, Bisbe & Otley, and Aber-
nethy & Brownell, who found relations between interactive control sys-
tems and product introduction and technology risks (Simons, 1991), inno-
vation risks (Bisbe & Otley, 2004), and strategic change (Abernethy &
Brownell, 1999). In regard to H2b, the finding is consistent with Widener’s
conclusion that diagnostic and interactive control systems are comple-
mentary (Widener, 2007) and that there is a dynamic tension by joint
usage of both (Henri, 2006a). In addition, the result is in line with evi-
dence in literature that structure is necessary for interactive control sys-
tems to be effective (Chenhall & Morris, 1995; Widener, 2007), which is
also suggested by Simons (2000). In summary, for all hypotheses of the
second set there is support (H2a, H2b, H2d, and H2e), or at least some
support (H2c), through data.
The third set of hypotheses, i.e. H3a - H3c, predicts associations be-
tween external risks and certain types of control systems. Firms facing
external risks to a greater extent focus on interactive control systems (p <
.01) and risk-based use of controls (p < .05) more intensively than firms
facing external risks to a lower extent, which provides support for H3b
and H3c. In addition, there is some support for H3a since there is a
strong tendency that external risks are positively associated with diagnos-
tic control systems (p = .051). Finally, the additional results on MCS de-
sign and use (Model 1A - 6A) provide support for all hypotheses H3a -
H3c, hence also for the proposition that external risks are positively asso-
ciated with diagnostic control systems (p < .05). Conclusively, these find-
ings are consistent with Kaplan & Mikes’ (2012) argument that for control-
ling external risks interactive control models are necessary. Furthermore,
results are consistent with empirical findings that competitive, market,
and technological risks are associated with interactive control systems
(Bisbe & Otley, 2004; Widener, 2007). In addition, Chenhall’s (2007)
proposition in summarizing empirical findings in contingency-based MCS
research that increased uncertainty is associated with more open, exter-
5.1. Risk profile contingent design and use of MCS 211
i.e. after adding the strategy variables to the multiple regressions in Mod-
el 1 - 6 and Model 1A - 6A, it is found that while a focus on cost leader-
ship is associated with all four levers of control (p < .001) and both risk-
based dimension of the extended risk-based MCS framework (p < .001),
a focus on differentiation is associated with an emphasis on beliefs sys-
tems (p < .001), diagnostic control systems (p < .023), and interactive
control systems (p < .018). While the findings for differentiation strategies
are in line with findings for prospector-like strategies in literature, i.e. con-
trol systems fostering creativity and innovation (Abernethy & Brownell,
1998; Chenhall, 2007), there is an unexpected broad impact of cost lead-
ership strategies on MCS design and use which are connected to more
tight control systems in literature (e.g., Chenhall & Morris, 1993; Go-
vindarajan, 1988). After adding the risk profile variables as predictors, i.e.
after step 3 of the multiple regression analyses, cost leadership strategies
are associated with boundary systems (p < .05), diagnostic control sys-
tems (p < .01), interactive control systems (p < .001), risk-based formal
controls (p < .05), and risk-based use of controls (p < .001), and differen-
tiation strategies are associated with beliefs systems (p < .05). Thus, the
findings confirm the critical role of strategy as a predictor in contingency-
based MCS literature (Langfield-Smith, 2007), although some of the as-
sociations become non-significant after adding the predictor variables
preventable risks, strategy execution risks, and external risks to the anal-
ysis. Interestingly, while cost leadership strategies are important determi-
nants for tight use of control systems and both dimensions of risk-based
MCS, the results on formal controls clearly reflect theoretical (Miles &
Snow, 1978; Porter, 1980) and empirical (Chenhall, 2007) arguments and
evidence as cost leadership strategies are associated with boundary
systems but not with beliefs systems, and vice versa for differentiation
strategies.
175
producing clusters, two-step cluster analyses are conducted and the
176
Schwarz's Bayesian Criterion, i.e. BIC criterion, for selecting the num-
ber of clusters is applied, in which the number of clusters is automatically
177
detected by the statistics software.
To validate the results of the two-step cluster analyses, discriminant
analyses are conducted and the significance of the discriminant models
as well the number of correctly classified cases is evaluated. In addition,
predictive validity is assessed by examining differences on variables not
included in the cluster analyses but possessing strong theoretical support
to vary across the clusters (Hair et al., 2014). Furthermore, the validity of
analysis is tested by assessing the representativeness of the sample and
the impact of multicollinearity (Hair et al., 2014). First of all, as the sample
shows a satisfactory size to be representative for Austrian and German
mechanical engineering companies (see Section 4.1), generalizations to
other industries or countries remain problematic, as the data may not be
representative for these populations. Secondly, to avoid bias from multi-
collinearity, the predictor variables should not correlate too highly, i.e.
above .80 or .90 (Field, 2013; Hair et al., 2014), and the variance inflation
175
In two-step cluster analyses categorical as well as metric variables can be processed
and the number of clusters for the optimal cluster solution is detected automatically.
Compared to the hierarchical cluster analysis this method is preferred for rather large
sample sizes (Bühl, 2014).
176
The Schwarz's Bayesian Criterion, i.e. BIC criterion, is a goodness-of-fit statistic that is
comparable to the AIC statistic (see Footnote 163) but more conservative in its proce-
dure. That is, the BIC criterion is corrected for model complexity and penalizes additional
model parameters more harshly than the AIC. The statistic should be applied when
sample sizes are rather large and the number of parameters rather small, and enables
comparisons of different models with smaller values of BIC indicating better fit of the da-
ta (Field, 2013).
177
The software automatically detects the number of clusters using the BIC criterion (see
Footnote 176). While some authors advise against such a procedure as the automatic
determination may not produce the best solution possible (e.g., Bühl, 2014, p. 665), oth-
ers point out that no standard objective selection procedure exists and that “the selection
of the final cluster solution requires substantial researcher judgement and is considered
by many as too subjective. […]. It is thus imperative to employ whatever objective sup-
port is available and be guided by reasoned judgment, especially in the design and in-
terpretation stages” (Hair et al., 2014, p. 425). Furthermore, and most importantly, clus-
ter solutions should have theoretical validity assessed through external validation and all
clusters should be significantly different in regard to their clustering variables (Hair et al.,
2014, p. 446-448).
216 5. Results
5.2.2 Analysis
The analysis for examining risk profile contingent packages of MCS starts
with examining the relations for the LOC framework, and proceeds by
applying the risk-based MCS framework. Finally, additional results for
both frameworks of MCS are provided.
Cluster analysis
LOC framework
N % of Combined
Cluster 1 201 55.5
2 161 44.5
Combined 362 100.0
BELIEF BOUND
Mean Std. Deviation Mean Std. Deviation
Cluster 1 5.50 0.74 5.76 0.82
2 4.30 0.97 3.61 1.21
Combined 4.96 1.04 4.81 1.47
DIAGNOST INTERACT
Mean Std. Deviation Mean Std. Deviation
Cluster 1 6.07 0.72 6.07 0.72
2 4.72 1.51 4.72 1.51
Combined 5.47 1.32 5.47 1.32
Discriminant analysis
LOC framework
Panel A: Canonical discriminant function coefficients and tests of equality of group means
unstandardized
F df1 df2 Sig.
coefficients
BELIEF 0.28 177.65 1 360 .000
BOUND 0.75 403.89 1 360 .000
DIAGNOST 0.17 125.35 1 360 .000
INTERACT 0.43 183.58 1 360 .000
Constant -7.92
Table 18: Discriminant analysis for the cluster solution of the LOC framework
Finally, in Panel B the classification results are obtained and original and
predicted group membership is indicated, which leads to the finding that a
highly satisfactory number of 96.1% of the firms are correctly classified.
Additionally, as the discriminant analysis as a multivariate technique is
178
based on a number of assumptions, results are validated by perform-
ing bootstrapping analysis, which is a robust method to generate confi-
dence intervals and significance tests of the model parameters (Field,
2013). All findings of the model have been confirmed by bootstrapping
analysis.
178
Key assumptions for the discriminant function are “multivariate normality of the inde-
pendent variables and unknown (but equal) dispersion and covariance structures (matri-
ces) for the groups as defined by the dependent variable” (Hair et al., 2014, p. 250).
While the former is tested through tests for normality of individual variables, the latter is
tested by using the Box’s M test which assesses the significance of differences in the
matrices between the groups (Hair et al., 2014, p. 250). As some of these tests show
ambiguous results, bootstrapping analysis is conducted in order to verify the results.
222 5. Results
Logistic regression
LOC framework
Model 7
Predictors Explained variable: MCS_cluster
Step 1:
b EXP(b) p-value
Control variables
Constant 1.554
SIZE -0.327 0.721 .011*
AGE 0.039 1.040 .760
OWN -0.496 0.609 .021*
Step 2:
+ Strategy profile
Constant 7,108
SIZE -0.226 0,798 .099
AGE 0.013 1,013 .921
OWN -0.461 0,631 .046*
COSTSTRAT -0.864 0,421 .000***
DIFFSTRAT -0.158 0,854 .164
Step 3:
+ Risk profile
Constant 9.081
SIZE -0.189 0.828 .191
AGE -0.005 0.995 .972
OWN -0.216 0.806 .378
COSTSTRAT -0.655 0.519 .000***
DIFFSTRAT -0.034 0.966 .780
PREVRISK -0.169 0.844 .361
STRATRISK -0.266 0.767 .011*
EXTRISK -0.374 0.688 .036*
To assess the predictive validity of the clusters and to test if types of risks
influence MCS cluster membership, a logistic regression analysis is con-
ducted, which is presented in Table 19 above.
The logistic regression analysis shows a model which accounts for 25.6%
of the amount of variation in the categorical outcome variable
MCS_cluster, which is coded as 0 for membership of C1LOC and as 1 for
membership of C2LOC. The logistic regression analysis is significant after
2 2 2 2
step 1 (RN = .049, χ (3) = 13.481, p = .004), step 2 (RN = .210, χ (5) =
2 2
61.915, p < .001), and step 3 (RN = .256, χ (8) = 76.942, p < .001). Fur-
thermore, the logistic regression model significantly improved after step
1, i.e. after adding the control variables SIZE, AGE, and OWN, and is
able to correctly classify 58.0% of the firms compared to 55.0% prediction
accuracy without any predictors. After step 2, i.e. after adding the strate-
gy profile variables COSTSTRAT and DIFFSTRAT, the model significant-
2
ly improved further (χ change(2) = 48.434, p < .001) and is able to correctly
classify 66.6% of the firms. Finally, through adding the risk profile varia-
bles PREVRISK, STRATRISK, and EXTRISK in step 3, the logistic re-
2
gression model significantly improved further (χ change(3) = 15.027, p =
.002) and is able to correctly classify 67.1% of the firms. After adding all
2
predictors in step 3, the variables COSTSTRAT (b = -0.665, z (1) =
2
13.649, p < .001), STRATRISK (b = -0.266, z (1) = 6.447, p = .011), and
2
EXTRISK (b = -0.374, z (1) = 4.389, p = .036) make a significant contri-
bution to the prediction of the outcome, i.e. MCS cluster membership,
and a change in the predictor COSTSTRAT (Exp(b) = 0.519) decreases
the odds of the outcome occurring the most, followed by EXTRISK
(Exp(b) = 0.688) and STRATRISK (Exp(b) = 0.767).
Finally, after validating the results of the cluster analysis, the following
can be derived from analyzing the corresponding figures:
Cluster 1 (C1LOC) is slightly larger and 55.5% of the participants in the
survey belong to this cluster. The CEOs of the firms in C1LOC seem to
put more emphasis on each of the four types of control systems, i.e.
beliefs systems, boundary systems, diagnostic control systems, and
interactive control systems, than the CEOs of the firms in C2LOC. Fur-
thermore, firms in this cluster face strategy execution risks and exter-
nal risks to a greater extent than firms in the other cluster. Additional-
224 5. Results
Cluster analysis
Risk-based MCS framework
N % of Combined
Cluster 1 192 53.0
2 170 47.0
Combined 362 100.0
BELIEF BOUND
Mean Std. Deviation Mean Std. Deviation
Cluster 1 5.58 0.66 5.76 0.88
2 4.26 0.94 3.73 1.25
Combined 4.96 1.04 4.81 1.47
DIAGNOST INTERACT
Mean Std. Deviation Mean Std. Deviation
Cluster 1 6.05 0.75 6.05 0.75
2 4.82 1.51 4.82 1.51
Combined 5.47 1.32 5.47 1.32
rbFORMALMCS rbUSEMCS
Mean Std. Deviation Mean Std. Deviation
Cluster 1 5.43 0.73 5.68 0.70
2 3.96 1.04 4.31 1.18
Combined 4.74 1.15 5.04 1.18
Discriminant analysis
Risk-based MCS framework
Panel A: Canonical discriminant function coefficients and tests of equality of group means
unstandardized
F df1 df2 Sig.
coefficients
BELIEF 0.41 242.643 1 360 .000
BOUND 0.53 322.348 1 360 .000
DIAGNOST 0.01 99.898 1 360 .000
INTERACT 0.34 175.863 1 360 .000
rbFORMALMCS 0.23 245.629 1 360 .000
rbUSEMCS 0.21 186.542 1 360 .000
Constant -8.42
Table 21: Discriminant analysis for the cluster solution of the risk-based MCS framework
Logistic regression
Risk-based MCS framework
Model 8
Predictors Explained variable: rbMCS_cluster
Step 1:
b EXP(b) p-value
Control variables
Constant 1.525
SIZE -0.355 0.701 .006**
AGE 0.127 1.136 .323
OWN -0.610 0.543 .005**
Step 2:
+ Strategy profile
Constant 6.888
SIZE -0.263 0.769 .054
AGE 0.112 1.119 .406
OWN -0.591 0.554 .010*
COSTSTRAT -0.837 0.433 .000***
DIFFSTRAT -0.148 0.863 .192
Step 3:
+ Risk profile
Constant 9.332
SIZE -.219 .803 .131
AGE .093 1.098 .495
OWN -.308 0.735 .209
COSTSTRAT -.591 .554 .001**
DIFFSTRAT -.004 .996 .974
PREVRISK -.216 .806 .252
STRATRISK -.344 .709 .001**
EXTRISK -.410 .663 .023*
Table 22: Logistic regression for prediction of risk-based MCS cluster membership
5.2. Risk profile contingent packages of MCS 229
The logistic regression analysis shows a model that accounts for 27.7%
of the amount of variation in the categorical outcome variable MCS clus-
ter membership compared to 25.6% when applying the LOC framework.
In this model, rbMCS_cluster is coded as 0 for membership of C1rbMCS
and as 1 for membership of C2rbMCS. The logistic regression analysis is
2 2 2
significant after step 1 (RN = .064, χ (3) = 17.772, p < .001), step 2 (RN
2 2 2
= .211, χ (5) = 62.326, p < .001), and step 3 (RN = .277, χ (8) = 84.108,
p < .001). Furthermore, the logistic regression model significantly im-
proved after step 1, i.e. after adding the control variables SIZE, AGE, and
OWN, and is able to correctly classify 58.8% of the firms compared to
53.0% prediction accuracy without any predictors. After step 2, i.e. after
adding the strategy profile variables COSTSTRAT and DIFFSTRAT, the
2
model significantly improved further (χ change(2) = 44.554, p < .001) and is
able to correctly classify 66.0% of the firms. Finally, through adding the
risk profile variables PREVRISK, STRATRISK, and EXTRISK in step 3,
2
the logistic regression model significantly improved further (χ change(3) =
21.782, p < .001) and is able to correctly classify 67.7% of the firms com-
pared to 67.1% in the base model. After adding all predictors in step 3,
2
the variables COSTSTRAT (b = -0.591, z (1) = 10.952, p = .001),
2
STRATRISK (b = -0.344, z (1) = 10.407, p = .001), and EXTRISK (b = -
2
0.410, z (1) = 5.205, p = .023) make a significant contribution to the pre-
diction of the outcome, i.e. risk-based MCS cluster membership, and a
change in the predictor COSTSTRAT (Exp(b) = 0.554) decreases the
odds of the outcome occurring the most, followed by EXTRISK (Exp(b) =
0.663) and STRATRISK (Exp(b) = 0.709).
Finally, after validating the results of the cluster analysis, the characteris-
tics of C1rbMCS and C2rbMCS are very similar to the characteristics of the
clusters identified when applying the LOC framework (see Section
5.2.2.1). However, the importance of strategy execution risks and exter-
nal risks increases and the importance of cost leadership strategies de-
creases as a predictor for risk-based MCS cluster membership compared
to the results when applying the LOC framework. In general, 53.0% of the
firms belong to C1rbMCS, applied tighter MCS (Merchant & Van der Stede,
2012), and face risks more intensively than firms belonging to C2rbMCS.
230 5. Results
To sum up, it can be concluded that the analysis for the risk-based MCS
framework revealed similar clusters compared to the analysis for the LOC
framework. While both cluster solutions show significant discrimination
between the clustering variables and satisfactory predictive validity, the
cluster solution of the risk-based MCS framework discriminates better
and the predictability of cluster membership is superior.
Logistic regression
LOC framework
Model 7A
Predictors Explained variable: MCS_cluster
b EXP(b) p-value
Step 3:
+ Risk profile
Constant 8.455
SIZE -0.212 0.809 .143
AGE 0.004 1.004 .979
OWN -0.213 0.808 .385
COSTSTRAT -0.648 0.523 .000***
DIFFSTRAT -0.035 0.966 .775
PREVRISK -0.192 0.825 .301
STRATRISK_dummy -0.720 0.487 .004**
EXTRISK -0.376 0.686 .034*
Table 23: Logistic regression – additional results on predictability of MCS cluster member-
ship
Logistic regression
Risk-based MCS framework
Model 8A
Predictors Explained variable: rbMCS_cluster
Step 3:
b EXP(b) p-value
+ Risk profile
Constant 8.572
SIZE -.252 0.778 .086
AGE .109 1.115 .434
OWN -.294 0.745 .232
COSTSTRAT -.582 0.559 .001**
DIFFSTRAT -.003 0.997 .978
PREVRISK -.254 0.775 .182
STRATRISK_dummy -1.010 0.364 .000***
EXTRISK -.405 0.667 .024*
Table 24: Logistic regression – additional results on predictability of risk-based MCS cluster
membership
In sum, it is demonstrated that while all other findings have been con-
firmed, the additional results show that the significance of the explanatory
power of strategy execution risks to predict MCS cluster membership and
risk-based MCS cluster membership, respectively, improved compared to
the preceding findings. In addition, in contrast to the previous findings,
STRATRISK_dummy is the most important predictor in these models.
consistent with Widener’s (2007) evidence that in fact all four levers of
control are interdependent and complementary.
Furthermore, hypothesis H4b proposes that also in regard to the risk-
based MCS framework, there are configurations of risk-based MCS in
place. The cluster analysis in Table 20 provides evidence on H4b and
found similar relationships as in regard to the LOC framework. In prac-
tice, there are two different and distinct configurations of risk-based MCS
in place, C1rbMCS and C2rbMCS, and 53.0% of the firms belong to C1rbMCS
and 47.0% of the firms to C2rbMCS. While firms in C1rbMCS put more em-
phasis on all four levers of control – beliefs systems, boundary systems,
diagnostic control systems, interactive control systems – and both risk-
based dimensions of the extended risk-based MCS framework – risk-
based formal controls and risk-based use of controls – the opposite is the
case for firms in C2rbMCS. Similarly to the findings in regard to H4a, evi-
dence was found for sound predictive validity in regard to these clusters,
as the predictors cost leadership strategy (p < .01), strategy execution
risks (p < .01), and external risks (p < .05) significantly predict cluster
membership, i.e. the greater the emphasis on these variables, the more
likely that a firm belongs to C1rbMCS. Thus, support for H5b is also provid-
ed, since types of risks significantly influence cluster membership.
Finally, the additional results supported all findings, and reported strategy
execution risks (p < .001) to be more influential than in the original model.
Conclusively, the same considerations in regard to connections to the
existing body of literature take place as discussed above. Finally, it can
be stated that the results indicate that the risk-based MCS framework
provides for an incremental development of Simons’ LOC framework
(Simons, 1995b) and the risk-based dimensions integrate well into the
existing base framework.
5.3. Risk profile contingent performance analysis of MCS 235
179
For the following analysis the measure EBIT_trimmed is used instead of EBIT, as the
measure EBIT shows extreme scores and in order to reduce the impact of outliers (see
Section 4.2.3. and Table 3).
236 5. Results
1𝑖𝑓𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = 𝑎𝑐𝑡𝑢𝑎𝑙𝑀𝐶𝑆𝑐𝑙𝑢𝑠𝑡𝑒𝑟𝑚𝑒𝑚𝑏𝑒𝑟𝑠ℎ𝑖𝑝
𝐹𝐼𝑇𝑖 = {
0𝑖𝑓𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 ≠ 𝑎𝑐𝑡𝑢𝑎𝑙𝑀𝐶𝑆𝑐𝑙𝑢𝑠𝑡𝑒𝑟𝑚𝑒𝑚𝑏𝑒𝑟𝑠ℎ𝑖𝑝
As a result, firms are classified into “High-Fit” (FIT = 1) and “Low-Fit” (FIT
= 0) and univariate tests and multiple regression analyses are conducted
to compare these two groups in terms of their performance measures.
Similar to the procedure in the previous sections, the following examina-
tions will start by applying the LOC framework and proceed with the anal-
ysis for the risk-based MCS framework. Accordingly, the following figure
illustrates the theoretical model:
180
Milgrom & Roberts (1992) counter the argument that everybody optimizes all the time
and state that: “Paradoxically, the very imperfections in the rationality of people and in
the adaptability of organizations denied by many simple economic theories are neces-
sary in proving that rationality-based theories are descriptively and prescriptively useful.
With perfect rationality, one would rarely expect to observe two organizations in substan-
tially the same circumstances making substantially different choices, so there would be
no possibility of testing what kinds of organizations perform better. […] A more defensi-
ble position […] is that people learn to make good decisions and that organizations adapt
by experimentation and imitation, so there is at least ‘‘fossil evidence’’ available for test-
ing theories (Milgrom & Roberts, 1992, p. 43; see also Ittner & Larcker, 2001, p. 399).
5.3. Risk profile contingent performance analysis of MCS 237
Preventable
risks
Logistic
Strategy regression
execution MCS_cluster
risks
ܲሺݎ݁ݐݏݑ̴݈ܿܵܥܯ ሻ
ͳ
External ൌ
ͳ ݁ ିሺబାభோாோூௌାమ ௌ்ோ்ோூௌାయா்ோூௌሻ
risks
FIT = 1
if the actual MCS configuration
equals the predicted one by the
logit model
regression
ܾସ ܶܣܴܱܶܵܶܵܥ
ܾହ ܶܣܴܶܵܨܨܫܦ ܾ ܶܫܨ ߝ
Control
Performance
variables
Figure 35: Theoretical model for risk profile contingent performance analysis of MCS
(Source: own illustration)
238 5. Results
1
𝑃(𝑀𝐶𝑆_𝑐𝑙𝑢𝑠𝑡𝑒𝑟𝑖 ) =
1 + 𝑒 −(𝑏0 +𝑏1𝑃𝑅𝐸𝑉𝑅𝐼𝑆𝐾𝑖 +𝑏2𝑆𝑇𝑅𝐴𝑇𝑅𝐼𝑆𝐾𝑖 +𝑏3𝐸𝑋𝑇𝑅𝐼𝑆𝐾𝑖 )
Equation 19: Predicted MCS cluster membership
2
Furthermore, the significance of the model is assessed by using the χ -
statistic and the overall fit of the model as a whole evaluated by compu-
2
ting the measure RN . Finally, the classification tables before and after
adding the risk profile variables as predictors are presented to show the
182
improvement of the model.
181
The reasons why logistic regression analysis was preferred over discriminant analysis
are threefold: Firstly, when computing the additional results and operationalizing strategy
execution risks, a dummy variable and thus a categorical variable is used, which is not
permitted in discriminant analysis. Secondly, comparisons of the logistic regression
analyses in Section 5.2.2.1, Section 5.2.2.2, and Section 5.2.2.3 in regard to their predic-
tive ability should be feasible. Finally, analyses with discriminant models showed that the
findings are very similar and the percentage of correctly classified cases almost the
same (for example, 64.9% of the cases are correctly classified in the base model when
using the logistic regression analysis as well as when using the discriminant analysis,
respectively).
182
In contrast to the logistic regression analyses conducted in Section 5.2.2.1, Section
5.2.2.2, and Section 5.2.2.3 in order to evaluate predictive validity of clusters and test if
the risk profile influences MCS cluster membership, the logistic regression analyses in
this section only use the risk profile variables as predictors. The reasons for this are
threefold: Firstly, as the hypotheses which are examined propose that firms with a better
fit between their MCS and their risk profile experience superior performance, it is aimed
to model fit between MCS and risk profile without being influenced by other factors, e.g.
strategy. Secondly, the control variables strategy, organizational size, company age, and
ownership structure also affect performance and therefore these variables are used in
the multiple regression analyses on performance as control variables to further validate
the results on the FIT variable instead of computing this variable. Finally, as will be
shown, the predictability of MCS cluster membership of the logistic regression analyses
in this section show similar results as the logistic regression analyses in Section 5.2.2.1,
Section 5.2.2.2, and Section 5.2.2.3 using all variables as predictors (for example, 64.9%
or 66.6% of the cases are correctly predicted using the risk profile variables as predictors
5.3. Risk profile contingent performance analysis of MCS 239
compared to 66.6% or 67.7% of the cases correctly predicted using all variables as pre-
dictors when applying the base MCS framework or the risk-based MCS framework, re-
spectively).
240 5. Results
As a next step, to test if firms with a better fit based on the logistic re-
gression models and the associated configuration of MCS perform better
than the others, univariate tests are conducted and the mean values of
the performance measures PERCPERF, USEFULMCS, and
EBIT_trimmed of both group of firms compared. To start with, an inde-
pendent-samples t-test is computed to evaluate if the means are signifi-
cantly different (see Equation 1). As the independent-samples t-test is
based on the normal distribution and relies on a number of assumptions,
to validate the results, bootstrapping analysis is performed and the Mann-
183 184
Whitney test, which is the non-parametric equivalent of the inde-
pendent-samples t-test, is computed.
Finally, multiple regression analyses are conducted with the measures for
organizational performance, i.e. PERCPERF, USEFULMCS, and
EBIT_trimmed, being the dependent variables in the analyses. These
examinations are performed for the LOC framework and the risk-based
MCS framework, and thus six regression models on performance are
conducted to examine if firms with a better fit perform better than others.
Consequently, for each regression model the following procedure is ap-
plied which can be formally expressed as:
183
The Mann-Whitney test is a non-parametric test and is based on a test statistic 𝑈 =
𝑛 (𝑛 +1)
𝑛1 𝑛2 + 1 1 − 𝑅1 , with n1 and n2 being the sample size of the groups and R1 being the
2
sum of ranks for group 1 (Field, 2013, p. 221).
184
Non-parametric tests rank the data and thus overcome the problem of the shape of the
distribution. While giving the lowest score rank 1, the next highest score rank 2 and so
on, effects of outliers are eliminated and the analysis carried out on ranks instead of ac-
tual data (Field, 2013).
5.3. Risk profile contingent performance analysis of MCS 241
addition and in line with the previous analyses, to improve validity of the
models a set of control variables is included, which have been argued to
not just affect the control structure of the firm (see Section 2.3.2.2, Sec-
tion 2.3.2.3, Section 2.3.2.4, and Section 2.3.2.5), but also the perfor-
mance of an organization (e.g., Sandino, 2007), i.e. COSTSTRAT,
DIFFSTRAT, SIZE, AGE, and OWN.
As a method for variables to be entered into the models, the hierarchical
entry method is chosen as in the previous analyses. This method re-
quires a decision on the sequence of entering the predictors into the
model and thus relies on theoretical reasoning. In general, known predic-
tors are entered into the model first in order to assess the influence of
newly tested variables (Field, 2013). Therefore, in the first step the con-
trol variables SIZE, AGE, and OWN and in the second step the strategy
variables COSTSTRAT and DIFFSTRAT are entered into the model.
Finally, the third and last step adds the dummy variable FIT (see Equa-
tion 18), in order to demonstrate the influence and additional explanatory
power of this predictor for the performance measures. The significance of
the model after each step is assessed by using the F-statistic which is
2
used to evaluate the significance of R and thus the overall fit of the
model as a whole. In addition, to evaluate if a step significantly improved
the model, the measure Fchange is computed, which assesses the signifi-
2
cance of the change in R . Finally, the significance and the impact of
each independent variable is evaluated by assessing the unstandardized
regression coefficients bi and the standardized regression coefficients βi.
In order to verify if the assumptions of the multiple regression analysis
have been met, a number of tests have been conducted for each model,
which were already outlined in Section 5.1.1 in the process of conducting
multiple regression analyses. First of all, the plot of standardized residu-
als against standardized predicted values and the partial plots of the re-
sidual of the outcome variable and each of the predictors were visually
examined, in order to test for the assumptions of linearity and homosce-
dasticity. In addition, the histograms and P-P plots (normal probability
plots) were visually examined, to test the assumption of normally distrib-
uted errors. While in general no serious signs for heteroscedasticity, non-
linearity, or non-normality have been found, there are a few graphs which
242 5. Results
5.3.2 Analysis
In order to conduct the risk profile contingent performance analysis of
MCS, it is started by examining the relations for the LOC framework and
preceded by applying the risk-based MCS framework. Finally, additional
results for both frameworks of MCS are discussed.
Logistic regression
LOC framework
Model 9
Explained variable: MCS_cluster
Panel A: Step 0
Predicted MCS cluster
Observed
1 2 Total
MSC cluster
Classification Count 1 201 0 201
table 2 161 0 161
% 1 100.0 0.0 100.0
2 100.0 0.0 100.0
Percentage
55.5
Correct
Panel B: Step 1
b EXP(b) p-value
Constant 7.004
PREVRISK -0.499 0.607 .002**
STRATRISK -0.312 0.732 .002**
EXTRISK -0.545 0.580 .001**
Table 25: Logistic regression for predicting MCS cluster membership via risk profile
5.3. Risk profile contingent performance analysis of MCS 245
2
The logistic regression analysis shows that the model is significant (RN =
2
.199, χ (3) = 58.216, p < .001) after adding the risk profile variables
PREVRISK, STRATRISK, and EXTRISK and accounts for 19.9% of the
amount of variation in the categorical outcome variable MCS cluster
membership (MCS_cluster), which is coded as 0 for membership of
C1LOC and as 1 for membership of C2LOC. All predictors in this model are
highly significant and after adding the risk profile variables, Model 9 is
able to correctly classify 64.9% of the firms (Panel B) compared to 55.5%
prediction accuracy without any predictors (Panel A). Furthermore, a
change in the predictor EXTRISK (Exp(b) = 0.580) decreases the odds of
the outcome occurring the most, followed by PREVRISK (Exp(b) = 0.607)
and STRATRISK (Exp(b) = 0.732).
According to Equation 18, the dummy variable FIT is calculated through
comparison of predicted and actual MCS cluster membership and named
FITLOC indicating that this variable was calculated when applying the LOC
framework.
The following table shows the results of the univariate performance tests
applying the LOC framework:
Univariate analyses
Risk-based MCS framework
a
: mean value in Mio. EUR
PERCPERF, USEFULMCS: n = 362, EBIT_trimmed: n = 130
+
p<.1, *p<.05, **p<.01, ***p<.001, two-tailed tests.
185
Levene’s test is a test statistic which tests the null hypothesis that the variance in differ-
ent groups is equal. A significant result of the test reports that variances are significantly
different and indicates that the assumption of homogeneity of variances is violated
(Field, 2013).
5.3. Risk profile contingent performance analysis of MCS 247
Multiple regressions
LOC framework
Model 10 Model 11
Explanatory Explained variable: Explained variable:
variables PERCPERF USEFULMCS
Step 1:
b β p-value b β p-value
Control variables
Constant 3.582 3.062
SIZE 0.241 .191 .000*** 0.266 .195 .000***
AGE -0.066 -.047 .382 0.009 .006 .909
OWN -0.016 -.006 .903 -0.205 -.077 .137
Step 2:
+ Strategy profile
Constant 2.958 0.816
SIZE 0.228 .180 .001** 0.211 .155 .003**
AGE -0.047 -.034 .533 0.040 .026 .616
OWN -0.021 -.008 .873 -0.166 -.062 .218
COSTSTRAT -0.027 -.020 .712 0.258 .183 .001**
DIFFSTRAT 0.137 .121 .028* 0.154 .126 .018*
Step 3:
+ FITLOC
Constant 2.816 0.758
SIZE 0.214 .170 .002** 0.205 .150 .004**
AGE -0.043 -.030 .570 0.042 .027 .599
OWN 0.010 .004 .936 -0.153 -.057 .257
COSTSTRAT -0.032 -.025 .652 0.256 .182 .001**
DIFFSTRAT 0.137 .121 .027* 0.154 .126 .018*
FITLOC 0.331 .128 .014* 0.136 .049 .334
Logistic regression
Risk-based MCS framework
Model 12
Explained variable: rbMCS_cluster
Panel A: Step 0
Predicted MCS cluster
Observed MSC
1 2 Total
cluster
Classification Count 1 192 0 192
table 2 170 0 170
% 1 100.0 0.0 100.0
2 100.0 0.0 100.0
Percentage
53.0
Correct
Panel B: Step 1
b EXP(b) p-value
Constant 7.710
PREVRISK -0.511 0.600 .002**
STRATRISK -0.392 0.676 .000***
EXTRISK -0.571 0.565 .001**
Table 28: Logistic regression for predicting risk-based MCS cluster membership via risk
profile
250 5. Results
2
The logistic regression analysis shows that the model is significant (RN =
2
.228, χ (3) = 67.793, p < .001) after adding the risk profile variables
PREVRISK, STRATRISK, and EXTRISK and accounts for 22.8% of the
amount of variation in the categorical outcome variable risk-based MCS
cluster membership (rbMCS_cluster), which is coded as 0 for member-
ship of C1rbMCS and as 1 for membership of C2rbMCS. Furthermore, all pre-
dictors in this model are highly significant and after adding the risk profile
variables Model 12 is able to correctly classify 66.6% of the firms (Panel
B) compared to 53.0% prediction accuracy without any predictors (Panel
A). Furthermore, a change in the predictor EXTRISK (Exp(b) = 0.565)
decreases the odds of the outcome occurring the most, followed by
PREVRISK (Exp(b) = 0.600) and STRATRISK (Exp(b) = 0.676).
According to Equation 18, the dummy variable FIT is calculated through
comparison of predicted and actual risk-based MCS cluster membership
and named FITrbMCS indicating that this variable was calculated when
applying the risk-based MCS framework.
The following table shows the results of the univariate performance tests
applying the risk-based MCS framework:
Univariate analyses
Risk-based MCS framework
a
: mean value in Mio. EUR
PERCPERF, USEFULMCS: n = 362, EBIT_trimmed: n = 130
*p<.05, **p<.01, ***p<.001, two-tailed tests.
Multiple regressions
Risk-based MCS framework
Model 13 Model 14
Explanatory Explained variable: Explained variable:
variables PERCPERF USEFULMCS
Step 3:
+ FITrbMCS
Constant 2.819 .000*** 0.783 .209
SIZE 0.214 .169 .002** 0.207 .152 .004**
AGE -0.037 -.027 .620 0.042 .028 .596
OWN 0.001 .000 .996 -0.160 -.060 .234
COSTSTRAT -0.029 -.022 .687 0.257 .183 .001**
DIFFSTRAT 0.133 .117 .032* 0.153 .126 .019*
FITrbMCS 0.300 .115 .028* 0.072 .026 .612
Table 26 / 29
Perceived firm
performance
Panel A:
(PERCPERF)
Univariate setting
Table 26 / 29
Perceived useful-
FIT ness of MCS
(USEFULMCS)
Table 26 / 29
EBIT_trimmed
Cost leadership
strategy
(COSTSTRAT)
Model 10 / 13
Perceived firm
m
D
Differentiation
performance
Panel B: strategy
(PERCPERF)
Multivariate setting (DIFFSTRAT)
Model 11 / 14
Perceived useful- O
Organizational
FIT ness of MCS size
(USEFULMCS) (SIZE)
S
Model - / -
Company
age
EBIT_trimmed (AGE)
Ownership
structure
(OWN)
Panel A: For each significant finding the t-value of the independent-samples t-test when
applying the LOC framework and, after the diagonal slash, when applying the risk-based
MCS framework is quoted as well as the corresponding significance level.
Panel B: For each significant association the standardized coefficient β of the respective
model when applying the LOC framework and, after the diagonal slash, when applying the
risk-based MCS framework is quoted as well as the corresponding significance level.
*p<.05, **p<.01, ***p<.001, two-tailed tests.
Figure 36: Graphical depiction of significant results on risk profile contingent performance
(Source: own illustration)
256 5. Results
Hypotheses H6a and H6b predict that firms with a better fit between their
MCS and their risk profile experience a superior perceived firm perfor-
mance. In an univariate setting, evidence has been found that firms that
choose MCS design and use better suited to their risk profile have a
higher perceived firm performance when applying the LOC framework (p
< .01) and when applying the risk-based MCS framework (p < .01), re-
spectively. Additionally, multiple regression analyses report that the fit
between MCS and risk profile is a significant predictor for perceived firm
performance when applying the LOC framework (p < .05) and when ap-
plying the risk-based MCS framework (p < .05), respectively. Thus, sup-
port for H6a and H6b is provided by data.
Furthermore, hypotheses H7a and H7b predict that firms with a better fit
between their MCS and their risk profile experience a better usefulness of
their MCS. In both univariate and multivariate analyses, no support for
hypotheses H7a and H7b are provided, which undermines Chenhall’s
(2007) conclusion on the aspect of usefulness of MCS and its effects on
organizational performance stating that “there is no compelling evidence
to suggest that such links exist” (Chenhall, 2003, p. 132).
Finally, hypotheses H8a and H8b predict that firms with a better fit be-
tween their MCS and their risk profile have a higher level of EBIT. Uni-
variate analyses show that firms that choose MCS design and use better
suited to their risk profile have a higher level of EBIT when applying the
LOC framework (p < .05). However, no association between matching
MCS and risk profile and higher level of EBIT when applying the risk-
based MCS framework could be found. Furthermore, multiple regression
analyses show that in fact the EBIT is significantly influenced by the fit
between MCS and the risk profile when applying the LOC framework (p <
.05), but the multiple regression models as a whole are non-significant
and thus the models are not reported. In summary, it can be argued that
there is some support for H8a, since firms with a higher fit between MCS
and risk profile were found to have, on average, significantly higher EBIT
figures than others. Nevertheless, data provided no support for H8b.
6 Conclusions
In this final chapter, the main findings and contributions to the research
field are presented. Furthermore, as it is the case for most research, this
study is limited for a number of reasons, which are discussed subse-
quently. Finally, in order to highlight potential key avenues, implications
for future research are addressed.
resulting in a type of rather tight MCS and a type of rather loose MCS. In
addition, it is shown that the probability of belonging to the group of rather
tight MCS is increased for organizations that are exposed to strategy
execution risks and external risks to a greater extent.
Finally, evidence is provided that organizations which choose MCS de-
sign and use better suited to their risk profile perform better than others.
Specifically, organizations with a better fit between their MCS design and
use and their risk profile experience a superior organizational perfor-
mance, both with subjective and objective measures.
Thus, the following contributions for theory and practice can be derived:
First of all, this study demonstrates that the types of risks an organization
faces affect MCS design and use. Therefore, evidence is provided for
acknowledging the risk profile of an organization as essential contingency
factor in management accounting and control research next to other well
established contingencies such as strategic choice (Langfield-Smith,
2007). This result is in line with Simons’ (2000) arguments and Widener’s
(2007) findings and adds further insight to the rather less recognized
consequences of uncertainty for MCS (Otley, 2012, 2014). Moreover, as
the “basic premise of contingency theory is that a proper “fit” will result in
higher performance” (Fisher, 1995, p. 47), it was shown that an organiza-
tion’s implementation of their MCS has performance implications depend-
ing on the types of risks they are exposed to. Specifically, organizations
with matching MCS and risk profile reveal superior performance. This
finding is consistent with Mikes & Kaplan’s (2014) argument that the “in-
herent nature and controllability of the different types of risk the organiza-
tion faces” (Mikes & Kaplan, 2015, p. 40) is decisive to the effectiveness
of the control model and, as a consequence, to organizational effective-
ness.
Secondly, implications for implementing MCS are derived, as specific
types of risks drive certain aspects of MCS. It was found that preventable
risks are associated with an emphasis on formal controls - beliefs sys-
tems and boundary systems - while strategy execution and external risks
are associated with an interactive use of control systems. In line with
Widener’s (2007) findings, these results imply that formal control systems
are rather focused on the internal environment, while the uncertainty
6.1. Findings and contributions 259
The limitations of this study serve as a starting point for providing a num-
ber of directions and opportunities for future research. To overcome the
weaknesses of survey-based research, the usage of multiple research
methods, for example combinations with case studies (Kaplan, 1986), to
examine the relationships is promising to strengthen validity of the inves-
tigation (Birnberg et al., 1990). Furthermore, multi-period research de-
signs and panel analyses seem promising to further examine cause-and-
effect relations. Moreover, to check for robustness of the findings, addi-
tional well-established contingencies of MCS design and use such as
generic and contemporary concepts of technology (e.g., Chenhall, 2007)
should be integrated into the study. Finally, newly conceptually specified
constructs such as strategy execution risks should be further improved in
order to develop more precise measurements of key variables.
In addition, there are a vast number of possibilities to further examine the
interrelationship between MCS design and use, risk profile, and organiza-
tional performance. For example, the influence of the types of risks on
MCS design and use could be examined using a non-linear model as the
impact of certain levers of control may differ depending on various levels
of respective risks. Furthermore, another classification for types of risks
could be operationalized, to broaden the understanding of the proposed
relationships. Finally, the initial approach of this study to integrate as-
pects of risk-based dimension of MCS into a holistic framework could be
further developed.
Overall, there are numerous exciting avenues ahead to further shed
some light on this unresolved but crucial venture to control uncertainty.
Appendix
To what extent are the following factors critical to achieving the organization’s strategy?
(1 = not at all, 7 = to a great extent)
Item Label
To what extent are the following factors critical to achieving the organization’s strategy?
(1 = not at all, 7 = to a great extent)
Item Label
To what extent are the following factors critical to achieving the organization’s strategy?
(1 = not at all, 7 = to a great extent)
Item Label
Please indicate the extent to which the following items describe your organization:
(1 = not descriptive, 7 = very descriptive)
Item Label
Please indicate the extent to which the following items describe your organization:
(1 = not descriptive, 7 = very descriptive)
Item Label
Item Label
Item Label
Please indicate the extent to which the following items describe your organization:
(1 = not descriptive, 7 = very descriptive)
Item Label
Item Label
Item Label
Achieving lower cost of services than com- COSTSTRAT_lower cost than competitors
petitors
Making services or procedures more cost COSTSTRAT_cost efficient
efficient
Improving the utilization of available equip- COSTSTRAT_improving utilization
ment, services and facilities
Item Label
Please rate the performance of your organization compared to your strongest competitors
on each of the following dimension in the last three years:
(1 = not at all satisfactory, 7 = outstanding)
Item Label
Sales volume PERCPERF_sales
Return on investment PERCPERF_profitability
Result PERCPERF_result
Please rate the extent to which you agree or disagree with the following:
(1 = strongly disagree, 7 = strongly agree)
Item Label
How many employees do you have on your pay list at the moment?
________ employees
________ (YYYY)
Ownership (FAMILYCOMP_ownership)
No Yes, up to 0,1%-25%
Ownership (FAMILYCOMP_management)
No Partly Exclusively
References
Ahrens, T.; Chapman, C. (2004): Accounting for flexibility and efficiency: a field
study of management control systems in a restaurant chain. Contemporary Ac-
counting Research, Vol. 21, Vol. 2, 1-28.
Alvesson, M.; Kärreman, D. (2004): Interfaces of Control. Technocratic and So-
cio-Ideological Control in a Global Management Consultancy Firm. Accounting,
Organizations and Society, Vol. 29, No. 3-4, 423-444.
Andrews, K.R. (1989): Ethics in Practice. Harvard Business Review, Vol. 67, No.
5, 99-104.
Ansari, S.L. (1977): An integrated approach to control system design. Accounting,
Organizations and Society, Vol. 2, No. 2, 101-112.
Ansoff, H.I. (1979): Strategic management. John Wiley & Sons Inc., New York,
United States.
Anthony, R.N. (1965): Management planning and control systems: A framework
for analysis. Harvard Business School Press, Boston, United States.
Anthony, R.N. (1988): The Management Control Function. Harvard Business
School Press, Boston, United States.
Anthony, R.N.; Govindarajan, V.; Hartmann, F.G.H.; Kraus, K.; Nilsson, G.
(2014): Management Control Systems. McGraw-Hill Higher Education, London,
United Kingdom.
Archer, S.; Otley, D.T. (1991): Strategy, structure, planning and control systems
and performance evaluation - Rumenco Ltd. Management Accounting Research,
Vol. 2, No. 4, 263-303.
Arena, M.; Arnaboldi, M.; Azzone, G. (2010): The organizational dynamics of
Enterprise Risk Management. Accounting, Organizations and Society, Vol. 35,
No. 7, 659-675.
Argyris, C.; Schön, D.A. (1978): Organizational learning: a theory of action per-
spective. Addison-Wesley Publishing Company, Reading, United States.
Arjalies, D.-L.; Mundy, J. (2013): The use of management control systems to
manage CSR strategy: A levers of control perspective. Management Accounting
Research, Vol. 24, No. 4, 284-300.
Armstrong, J.B.; Overton, T.S. (1977): Estimating Nonresponse Bias in Mail Sur-
veys. Journal of Marketing Research, Vol. 14, No. 3, 396-402.
References 273
ment Processes. Journal of Accounting, Auditing & Finance, Vol. 23, No. 3, 311-
332.
Beck, U. (1992): Risk Society. Towards a New Modernity. SAGE Publications
Ltd., London, United Kingdom.
Bedford, D.S. (2015): Management control systems across different modes of
innovation: Implications for firm performance. Management Accounting Research,
Vol. 28, 12-30.
Bedford, D.S.; Malmi, T.; Sandelin, M. (2016): Management control effectiveness
and strategy: An empirical analysis of packages and systems. Accounting, Organ-
izations and Society, Vol. 51, 12-28.
Beer, S. (1962): Kybernetik und Management. Hanseatische Druckanstalt GmbH,
Hamburg, Germany.
Beer, S. (1966): Decision and control: the meaning of operational research and
management cybernetics. John Wiley & Sons Inc., New York, United States.
Beer, S. (1972): Brain of the firm. John Wiley & Sons Inc., New York, United
States.
Bergen, M.; Dutta, S.; Walker, O.C. (1992): Agency Relationships in Marketing: A
Review of the Implications and Applications of Agency and Related Theories.
Journal of Marketing, Vol. 56, No. 3, 1-24.
Bernstein, P.L. (1996a): Against the gods. The remarkable story of risk. John
Wiley & Sons Inc., New York, United States.
Bernstein, P.L. (1996b): The New Religion of Risk Management. Harvard Busi-
ness Review, Vol. 74, No. 2, 47-51.
Bernoulli, D. (1738/1954): Exposition of a New Theory on the Measurement of
Risk (Specimen Theoriae Novae de Mensura Sortis). Translated from Latin by
Louise Sommer. Econometrica, Vol. 22, No. 1, 23-36.
Berry, A.J., Coad, A.F., Harris, E.P., Otley, D.T.; Stringer, C. (2009): Emerging
themes in management control: A review of recent literature. The British Account-
ing Review, Vol. 41, No. 1, 2-20.
Bertalanffy, L.v. (1972): The History and Status of General Systems Theory.
Academy of Management Journal, Vol. 15, No. 4, 407-426.
References 275
Dauten, P.M.Jr.; Gammill, H.L.; Robinson, S.C. (1958): Our concepts of control-
ling need re-thinking. The Journal of the Academy of Management, Vol. 1, No. 3,
41-55.
Davila, T. (2000): An empirical study on the drivers of management control sys-
tems’ design in new product development. Accounting, Organizations and Socie-
ty, Vol. 25, No. 4-5, 383-409.
Davila, T. (2005): An exploratory study on the emergence of management control
systems: formalizing human resources in small growing firms. Accounting, Organ-
izations and Society, Vol. 30, No. 3, 223-248.
Davila, A.; Foster, G. (2007): Management Control Systems in Early Stage
Startup Companies. The Accounting Review, Vol. 82, No. 4, 907-937.
Dent, J.F. (1990): Strategy, organization and control: Some possibilities for ac-
counting research. Accounting, Organizations and Society, Vol. 15, No. 1-2, 3-25.
Dess, G.G.; Davis, P.S. (1984): Porter's (1980) Generic Strategies as Determi-
nants of Strategic Group Membership and Organizational Performance. Academy
of Management Journal, Vol. 27, No. 3, 467-488.
Dess, G.G.; Robinson, R.B.Jr. (1984): Measuring organizational performance in
the absence of objective measures: The case of the privately-held firm and con-
glomerate business unit. Strategic Management Journal, Vol. 5, No. 3, 265-273.
Diamantopoulos, A.; Winklhofer, H.M. (2001): Index construction with formative
indicators: An alternative to scale development. Journal of Marketing Research,
Vol. 38, No. 2, 269-277.
Dillman, D.A.; Smyth, J.D.; Christian, L.M. (2014): Internet, Phone, Mail, and
Mixed-Mode Surveys: The Tailored Design Method. John Wiley & Sons Inc., New
York, United States.
Ditillo, A. (2004): Dealing with uncertainty in knowledge-intensive firms: the role of
management control systems as knowledge integration mechanisms. Accounting,
Organizations and Society, Vol. 29, No. 3/4, 401-421.
Donaldson, L. (2001): The Contingency Theory of Organizations. SAGE Publica-
tions, Thousand Oaks, California.
Douglas, M. (2001): Dealing with Uncertainty. Ethical Perspectives, Vol. 8, No. 3,
145-155.
Drazin, R.; Van de Ven, A.H. (1985): Alternative Forms of Fit in Contingency
Theory. Administrative Science Quarterly, Vol. 30, No. 4, 514-539.
References 281
Flamholtz, E.G.; Das, T.K.; Tsui, A.S. (1985): Toward an integrative framework of
organizational control. Accounting, Organizations and Society, Vol. 10, No. 1, 35-
50.
Floricel, S.; Miller, R. (2001): Strategizing for Anticipated Risks and Turbulence in
Large-Scale Engineering Projects. International Journal of Project Management,
Vol. 19, No. 8, 445-455.
Foster, G.; Young, S.M. (1997): Frontiers of Management Accounting Research.
Journal of Management Accounting Research, Vol. 9, No. 1, 63-77.
Freeman, J.; Carroll, G.R.; Hannan, M.T. (1983): The Liability of Newness: Age
Dependence in Organizational Death Rates. American Sociological Review, Vol.
48, No. 5, 692-710.
Galbraith, J.R. (1973): Designing Complex Organizations. Addison-Wesley Pub-
lishing Co., Reading, United States.
Galbraith, J.R. (1977): Organization Design. Addison-Wesley Publishing Co.,
Reading, United States.
Gatewood, R.D.; Carroll, A.B. (1991): Assessment of Ethical Performance of
Organization Members: A Conceptual Framework. The Academy of Management
Review, Vol. 16, No. 4, 667-690.
Gerdin, J. (2005): The impact of departmental interdependencies and manage-
ment accounting system use on subunit performance: A second look. European
Accounting Review, Vol. 14, No. 2, 335-340.
Gerdin, J.; Greve, J. (2004): Forms of contingency fit in management accounting
research. Accounting, Organizations and Society, Vol. 29, No. 3/4, 303-326.
Gerdin, J.; Greve, J. (2008): The appropriateness of statistical methods for testing
contingency hypotheses in management accounting research. Accounting, Or-
ganizations and Society, Vol. 33, No. 7/8, 995-1009.
Giglioni, G.B.; Bedeian, A.G. (1974): A Conspectus of Management Control The-
ory: 1900-1972. Academy of Management Journal, Vol. 17, No. 2, 292-305.
Göstl, P.; Schwaiger, W.S.A. (2016): Risk-based Planning and Control Systems:
Integration of different risk types into strategic, financial and operational man-
agement systems. Working paper presented at Management Control Association
Conference, University of Antwerp, from 07th to 09th September 2016, 1-16.
284 References
Gond, J.-P.; Grubnic, S.; Herzig, C.; Moon, J. (2012): Configuring management
control systems: Theorizing the integration of strategy and sustainability. Man-
agement Accounting Research, Vol. 23, No. 3, 205-223.
Gong, M.; Ferreira, A. (2014): Does consistency in management control systems
design choices influence firm performance? An empirical analysis. Accounting
and Business Research, Vol. 44, No. 5, 497-522.
Gordon, L.A.; Miller, D. (1976): A contingency framework for the design of ac-
counting information systems. Accounting, Organizations and Society, Vol. 1, No.
1, 59-69.
Gordon, L.A.; Narayanan, V.K. (1984): Management accounting systems, per-
ceived environmental uncertainty and organization structure: An empirical investi-
gation. Accounting, Organizations and Society, Vol. 9, No. 1, 33-47.
Gordon, L.A.; Loeb, M.P.; Tseng, C.Y. (2009): Enterprise risk Management and
firm performance: A contingency perspective. Journal of Accounting and Public
Policy, Vol. 28, No. 4, 301-327.
Goold, M.; Quinn, J.J. (1993): Strategic control. Milestones for long-term perfor-
mance. Addison-Wesley Publishing Co., Reading, United States.
Govindarajan, V. (1984): Appropriateness of accounting data in performance
evaluation: An empirical examination of environmental uncertainty as an interven-
ing variable. Accounting, Organizations and Society, Vol. 9, No. 2, 125-135.
Govindarajan, V. (1986): Impact of participation in the budgetary process on
managerial attitudes and performance: universalistic and contingency perspec-
tives. Decision Sciences, Vol. 17, No. 4, 496-516.
Govindarajan, V. (1988): A contingency approach to strategy implementation at
the business-unit level: Integrating Administrative mechanisms with strategy.
Academy of Management Journal, Vol. 31, No. 4, 828-853.
Govindarajan, V.; Fisher, J. (1990): Strategy, control systems and resource shar-
ing: effects on business-unit performance. Academy of Management Journal, Vol.
33, No. 2, 259-285.
Govindarajan, V.; Gupta, A.K. (1985): Linking control systems to business unit
strategy: Impact on performance. Accounting, Organizations and Society, Vol. 10,
No. 1, 51-66.
References 285
Hopwood, A.G. (1972): An empirical study of the role of accounting data in per-
formance evaluation. Journal of Accounting Research, Vol. 10, No. 3, 156-182.
Hoyt, R.E. (2015): Evidence of the Value of Enterprise Risk Management. Journal
of Applied Corporate Finance, Vol. 27, No.1, 41-56.
Hoyt, R.E.; Liebenberg, A.P. (2011): The Value of Enterprise Risk Management.
The Journal of Risk and Insurance, Vol. 78, No. 4, 795-822.
ICAEW (1999): Internal Control. Guidance for directors on the combined code.
The Institute of Chartered Accountants. Chartered Accountants’ Hall, London,
United Kingdom.
ISO (2009a): ISO Guide 73:2009. Risk management – Vocabulary. International
Standards Organisation, Geneva, Switzerland.
ISO (2009b): ISO 31000:2009. Risk management – Principles and guidelines.
International Standards Organisation, Geneva, Switzerland.
Ittner, C.D.; Larcker, D.L. (1998): Innovations in performance measurement:
Trends and research implications. Journal of Management Accounting Research,
Vol. 10, No. 1, 205-238.
Ittner, C.D.; Larcker, D.F. (2001): Assessing empirical research in managerial
accounting: a value-based management perspective. Journal of Accounting and
Economics, Vol. 32, No. 1/3, 349-410.
Ittner, C.D.; Lanen, W.N.; Larcker, D.F. (2002): The Association between Activity-
Based Costing and Manufacturing Performance. Journal of Accounting Research,
Vol. 40, No. 3, 711-726.
Itzkowitz, G. (1996): Contingency theory: rethinking the boundaries of social
thought. University Press of America, New York, United States.
Jarvis, C.B.; Mackenzie, S.B.; Podsakoff, P.M. (2003): A Critical Review of Con-
struct Indicators and Measurement Model Misspecification. Journal of Consumer
Research, Vol. 30, No. 2, 199-218.
Jensen, M.C.; Meckling, W.H. (1976): Theory of the firm managerial behavior,
agency costs and ownership structure. Journal of Financial Economics, Vol. 3,
No. 4, 305-360.
Judge, T.A.; Higgins, C.; Thoresen, C.J.; Barrick, M.R. (1999): The Big Five per-
sonality traits, general mental ability, and career success across the life span.
Personnel Psychology, Vol. 52, 621-652.
288 References
Law, K.; Wong, C.-S.; Mobley, W.H. (1998): Toward a Taxonomy of Multidimen-
sional Constructs. The Academy of Management Review, Vol. 23, No. 4, 741-
755.
Lawler, E.E.; Rhode, J.G. (1976): Information and control in organizations. Good-
year Publishing Co., Santa Monica, California.
Lawrence, P.R.; Lorsch, J.W. (1967): Differentiation and Integration in Complex
Organizations. Administrative Science Quarterly, Vol. 12, No. 1, 1-47.
Leblebici, H.; Salancik, G.R. (1982): Stability in Interorganizational Exchanges:
Rulemaking Processes of the Chicago Board of Trade. Administrative Science
Quarterly, Vol. 27, No. 2, 227-242.
Libby, T.; Waterhouse, J.H. (1996): Predicting Change in Management Account-
ing Systems. Journal of Management Accounting Research, Vol. 8, No. 4, 137-
150.
Libby, T.; Bloomfield, R.; Nelson, M.W. (2002): Experimental research in financial
accounting. Accounting, Organizations and Society, Vol. 27, No. 8, 775-810.
Liebenberg, A.P.; Hoyt, R.E. (2003): The Determinants of Enterprise Risk Man-
agement: Evidence from the appointment of chief risk officers. Risk Management
and Insurance Review, Vol. 6, No. 1, 37-52.
Locke, E.A.; Latham, G.P.; Erez, M. (1988): The Determinants of Goal Commit-
ment. The Academy of Management Review, Vol. 13, No. 1, 23-39.
Lowe, E.A.; McInnes, J.M. (1971): Control in Socio-Economic Organizations: A
Rationale for the Design of Management Control Systems (Section I). Journal of
Management Studies, Vol. 8, No. 2, 213-227.
Luft, J.; Shields, M.D. (2003): Mapping management accounting: graphics and
guidelines for theory-consistent empirical research. Accounting, Organizations
and Society, Vol. 28, No. 2/3, 169-249.
Luft, J. (2016): Management accounting in the laboratory and in social context:
Four contrasts 1975-2014. Accounting, Organizations and Society, Vol. 49, No.1,
9-20.
Luneski, C. (1964): Some Aspects of the Meaning of Control. The Accounting
Review, Vol. 39, No. 3, 591-597.
MacCrimmon, K.R.; Wehrung, D.A. (1986): Taking risks: the management of
uncertainty. Free Press, New York, United States.
References 291
(http://www.lse.ac.uk/accounting/news/ MARG/Mikes-MARG-Presentation-
160413-From-Belief-Systems-to-Risk-Culture-short.pdf).
Mikes, A.; Kaplan, R.S. (2014): Towards a Contingency Theory of Enterprise Risk
Management. Working Paper, 1-45.
Mikes, A.; Kaplan, R.S. (2015): When One Size Doesn’t Fit All: Evolving Direc-
tions in the Research and Practice of Enterprise Risk Management. Journal of
Applied Corporate Finance, Vol. 27, No. 1, 37-41.
Mikes, A., Hall, M.; Millo, Y. (2013): How experts gain influence. Harvard Busi-
ness Review, Vol. 91, No. 7/8, 70-74.
Miles, R.E.; Snow, C.C.; Meyer, A.D.; Coleman, H.J. (1978): Organizational
Strategy, Structure, and Process. The Academy of Management Review, Vol. 3,
No. 3, 546-562.
Milgrom, P.; Robert, J. (1992): Economics, organization and management. Eng-
lewood Cliffs, New York Prentice-Hall, United States.
Miller, P. (1998): The margins of accounting. The European Accounting Review,
Vol. 7, No. 4, 605-621.
Miller, D.; Friesen, P.H. (1982): Innovation in Conservative and Entrepreneurial
Firms: Two Models of Strategic Momentum. Strategic Management Journal, Vol.
3, No. 1, 1-25.
Miller, P.; Power, M. (2013): Accounting, Organizing, and Economizing: Connect-
ing Accounting Research and Organization Theory. The Academy of Manage-
ment Annals, Vol. 7, No. 1, 557-605.
Mintzberg, H. (1973): Strategy-Making in Three Modes. California Management
Review, Vol. 16, No. 2, 44-53.
Mintzberg, H. (1978): Patterns in Strategy Formation. Management Science, Vol.
24, No. 9, 934-948.
Mintzberg, H. (1979): The structuring of organizations. Prentice‐ Hall, Englewood
Cliffs, United States.
Mintzberg, H. (1987): Crafting Strategy. Harvard Business Review, Vol. 65, No. 4,
65-74.
Mintzberg, H.; Waters, J.A. (1982): Tracking Strategy in an Entrepreneurial Firm.
Academy of Management Journal, Vol. 25, No. 3, 465-499.
294 References
Otley, D.T.; Berry, A.J. (1980): Control, organisation and accounting. Accounting,
Organizations and Society, Vol. 5, No. 2, 231-244.
Otley, D.T.; Fakiolas, A. (2000): Reliance on accounting performance measures:
dead end or new beginning? Accounting, Organizations and Society, Vol. 25, No.
4/5, 497-510.
Otley, D.T.; Soin, K. (2014): Management Control and Uncertainty. In: Otley, D.T.;
Soin, K. (Ed.): Management Control and Uncertainty. Palgrave Macmillan, Lon-
don, United Kingdom, 1-13.
Otley, D.T.; Broadbent, J.; Berry, A. (1995): Research in Management Control: An
Overview of its Development. British Journal of Management, Vol. 6, Special
Issue, 31-44.
Ouchi, W. (1977): The Relationship between Organizational Structure and Organ-
izational Control. Administrative Science Quarterly, Vol. 22, No.1, 95-113.
Ouchi, W.G. (1979): A conceptual framework for the design of organizational
control mechanisms. Management Science, Vol. 25, No. 9, 833-848.
Ouchi, W. (1980): Markets, bureaucracies and clans. Administrative Science
Quarterly, Vol. 25, No. 1, 129-141.
Ouchi, W.G.; Maguire, M.A. (1975): Organizational Control: Two Functions. Ad-
ministrative Science Quarterly, Vol. 20, No. 4, 559-569.
Pagach, D.; Warr, R. (2011): The Characteristics of Firms that Hire Chief Risk
Officers. The Journal of Risk and Insurance, Vol. 78, No. 1, 185-211.
Palermo, T.; Power, M.; Ashby, S. (2017): Navigating Institutional Complexity:
The Production of Risk Culture in the Financial Sector. Journal of Management
Studies, Vol. 54, No. 2, 154-181.
Perrow, C. (1970): Organizational analysis: A sociological view. Tavistock Publi-
cations, New York, United States.
Pfister, J. (2014): Controlling Creativity and Innovation: Paradox or Necessity? In:
Otley, D.T.; Soin, K. (Ed.): Management Control and Uncertainty. Palgrave Mac-
millan, London, United Kingdom, 134-158.
Pitkänen, H.; Lukka, K. (2011): Three dimensions of formal and informal feedback
in management accounting. Management Accounting Research, Vol. 22, No. 2,
125-137.
296 References
Ross, A. (1995): Job related tension, budget emphasis and uncertainty: a re-
search note. Management Accounting Research, Vol. 6, No. 1, 1-11.
Sandelin, M. (2008): Operation of management control practices as a package: A
case study on control system variety in a growth firm context. Management Ac-
counting Research, Vol. 19, No. 4, 324-343.
Sandino, T. (2004): Introducing the first management control systems: Evidence
from the Retail Sector. Doctoral theses, Harvard University, Cambridge, United
States.
Sandino, T. (2007): Introducing the first management control systems: Evidence
from the Retail Sector. The Accounting Review, Vol. 82, No. 1, 265-293.
Sarbanes-Oxley Act (2002): Public Law 107-204, United States.
Schall, M.S. (1983): A Communication-Rules Approach to Organizational Culture.
Administrative Science Quarterly, Vol. 28, No. 4, 557-581.
Schein, E.H. (2004): Organizational Culture and Leadership. Jossey-Bass, A
Wiley Imprint, San Francisco, United States.
Schoonhoven, C.B. (1981): Problems with Contingency Theory: Testing Assump-
tions Hidden within the Language of Contingency “Theory”. Administrative Sci-
ence Quarterly, Vol. 26, No. 3, 349-377.
Schwartz, M. (2001): The nature of the relationship between corporate codes of
Ethics and Behaviour. Journal of Business Ethics, Vol. 32, No. 3, 247-262.
Segars, A.H. (1997): Assessing the unidimensionality of measurement: a para-
digm and illustration within the context of information systems research. Omega,
Vol. 25, No. 1, 107-121.
Seibert, S.E.; Silver, S.R.; Randolph, W.A. (2004): Taking Empowerment to the
Next Level: A Multiple-Level Model of Empowerment, Performance and Satisfac-
tion. Academy of Management Journal, Vol. 47, No. 3, 332-349.
Selto, F.H.; Renner, C.J.; Young, S.M. (1995): Assessing the organizational fit of
a just-in-time manufacturing system. Accounting, Organizations and Society, Vol.
20, No. 7/8, 665-684.
Shields, M. D. (1995): An empirical analysis of firms’ implementation experiences
with activity-based costing. Journal of Management Accounting Research, Vol. 7,
No. 4, 148-166.
298 References
Speklé, R.; Kruis, A.-M. (2014): Management control research: a review of current
developments. In: Otley, D.T.; Soin, K. (Ed.): Management Control and Uncertain-
ty. Palgrave Macmillan, London, United Kingdom, 30-46.
Taleb, N.N. (2007): Fooled by Randomness. The Hidden Role of Chance in the
Markets and in Life. Penguin Books Ltd., London, United Kingdom.
Taylor, F.W. (1911/1919): The Principles of Scientific Management. Reprinted
1919 by Harper & Brothers Publishers, New York and London, United States and
United Kingdom.
Tessier, S. (2014): Managing Management Controls. In: Otley, D.T.; Soin, K.
(Ed.): Management Control and Uncertainty. Palgrave Macmillan, London, United
Kingdom, 149-161.
Tessier, S.; Otley, D. (2012a): A conceptual development of Simons’ Levers of
Control framework. Management Accounting Research, Vol. 23, No.3, 171-185.
Tessier, S.; Otley, D. (2012b): From management controls to the management of
controls. Accounting, Auditing & Accountability Journal, Vol. 25, No. 5, 776-805.
Thompson, J.D. (1967): Organizations in Action: Social Science Bases of Admin-
istrative Theory. McGraw-Hill Inc., New York, United States.
Tocher, K.D. (1970): Control. Operational Research Quarterly, Vol. 21, No. 2,
159-180.
Tocher, K.D. (1976): Notes for Discussion on "Control". Operational Research
Quarterly, Vol. 27, No. 1, 231-239.
Tuomela, T.-S. (2005): The interplay of different levers of control: A case study of
introducing a new performance measurement system. Management Accounting
Research, Vol. 16, No. 3, 293-320.
Tymon, W.G.; Stout, D.E.; Shaw, K.N. (1998): Critical Analysis and Recommen-
dations regarding the Role of Perceived Environmental Uncertainty in Behavioral
Accounting Research. Behavioral Research in Accounting, Vol 10, 23-46.
Vandenbosch, B. (1999): An empirical analysis. Accounting, Organizations and
Society, Vol. 24, No. 1, 77-92.
Van der Stede, W.A. (2001): Measuring ‘Tight Budgetary Control’. Management
Accounting Research, Vol. 12, No. 1, 119-137.
300 References
Van der Stede, W.A. (2003): The effect of national culture on management con-
trol and incentive system design in multi-business firms: evidence of intracorpo-
rate isomorphism. European Accounting Review, Vol. 12, No. 2, 263-285.
Van der Stede, W.A. (2016): Management accounting in context: Industry, regula-
tion and informatics. Management Accounting Research, Vol. 31, 100-102.
Van der Stede, W.A.; Young, C.; Chen, C.X. (2005): Assessing the quality of
evidence in empirical management accounting research: The case of survey
studies. Accounting, Organizations and Society, Vol. 30, No. 7/8, 655-684.
VDMA (2017): Mechanical engineering – figures and charts 2017. VDMA Verlag
GmbH, Frankfurt am Main, Germany.
Venkatraman, N.; Ramanujam, V. (1987): Measurement of Business Perfor-
mance in Strategy Research: A Comparison of Approaches. The Academy of
Management Review, Vol. 11, No. 4, 801-814.
Venkatraman, N. (1989): The Concept of Fit in Strategy Research: Toward Verbal
and Statistical Correspondence. The Academy of Management Review, Vol. 14,
No. 3, 423-444.
Visser, P.S.; Krosnick, J.A.; Lavrakas, P.J. (2000): Survey research, in: Reis,
H.T.; Judd, C.M. (Eds.): Handbook of research methods in social and personality
psychology. Cambridge University Press, Cambridge, United Kingdom, 223-252.
Von Neumann, J.; Morgenstern, O. (1953): Theory of Games and Economic Be-
havior. Princeton University Press, New Jersey, United States.
Walsh, J.P.; Seward, J.K. (1990): On the efficiency of internal and external corpo-
rate control mechanism. Academy of Management Review, Vol. 15, No. 3, 421-
458.
Westley, F.; Mintzberg, H. (1989): Visionary Leadership and Strategic Manage-
ment. Strategic Management Journal, Vol. 10, Special Issue, 17-32.
Whitley, R. (1999): Firms, institutions and management control: the comparative
analysis of coordination and control systems. Accounting, Organizations and
Society, Vol. 24, No. 5/6, 507-524.
Widener, S.K. (2007): An empirical analysis of the levers of control framework.
Accounting, Organizations and Society, Vol. 32, No. 7/8, 757-788.
Wiener, N. (1948): Cybernetics: or Control and Communication in the Animal and
the Machine. The M.I.T. Press, Cambridge, United Kingdom.
References 301