Management control system design, ownership, and performance in professional service organ

Description
The objective of this study is to investigate the implications for organisational performance of the interplay
between ownership and management control system design in professional service organisations.
Based on transaction cost economic (TCE) theory, we expect that low ownership by professionals working
in a professional services organisation will be more efficiently managed with a boundary MCS archetype
and high ownership by an exploratory MCS archetype. Of direct relevance, we predict that a failure to
conform to these optimal archetypes will manifest in relatively poorer performance.

Management control system design, ownership, and performance
in professional service organisations
Robyn King
a,?,1
, Peter Clarkson
a,b
a
UQ Business School, The University of Queensland, Brisbane, QLD 4072, Australia
b
Beedie School of Business, Simon Fraser University, Burnaby, B.C. V5A 1S6, Canada
a r t i c l e i n f o
Article history:
Received 22 October 2013
Revised 17 June 2015
Accepted 24 June 2015
Available online 13 July 2015
Keywords:
Control systems
Ownership
TCE
Primary healthcare organisations
a b s t r a c t
The objective of this study is to investigate the implications for organisational performance of the inter-
play between ownership and management control system design in professional service organisations.
Based on transaction cost economic (TCE) theory, we expect that low ownership by professionals working
in a professional services organisation will be more ef?ciently managed with a boundary MCS archetype
and high ownership by an exploratory MCS archetype. Of direct relevance, we predict that a failure to
conform to these optimal archetypes will manifest in relatively poorer performance. The study was
conducted based on a survey of 120 practice managers of primary healthcare organisations in
Australia. These results provide empirical support for the stated prediction.
Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction
We investigate the implications for organisational performance
of the interplay between ownership and management control sys-
tem (MCS) design in professional service organisations. The con-
textual setting for our investigation is the primary healthcare
sector in Australia. Primary healthcare organisations (PHOs) are
small ‘for pro?t’ organisations where general practitioners (GPs)
provide a ?rst point of contact with the healthcare system (DHA,
2013). PHOs present a considerable control challenge because
GPs are highly trained professionals who work independently to
produce an intangible output and have preferences that con?ict
with bureaucracy. Early organisational theorists predict that
ownership is an effective solution to this challenge (Fama & Jensen,
1983; Greenwood & Empson, 2003).
2
However, in Australia we
observe differences in the level of GP ownership across PHOs (IBIS,
2011). The performance implications of this variation have not been
investigated to date. A related question is whether differences in the
MCS design can mitigate these differences.
We structure our analysis around Transaction Cost Economics
(TCE), a holistic MCS design theory that allows for the possibility
of misalignment and resultant performance effects (Hakansson &
Lind, 2007). We argue, consistent with Speklé (2001), that within
the PHO context which exhibits the characteristics of high
uncertainty and high asset speci?city, the ef?cient MCS design
for organisations with low GP ownership is the boundary arche-
type and for those with high GP ownership, it is the exploratory
archetype. The boundary archetype features administrative con-
trols emphasising behaviours to be avoided whereas the explora-
tory archetype features less formal controls that are engaged in
creating and preserving information sharing. Importantly, we
predict that conforming to these archetypes will result in relatively
higher performance (Speklé, 2001).
We employ data from an online survey of practice managers
that provided 120 useable responses (a 26.6% response rate). We
identify the empirical ideal MCS for PHOs that differ in ownership
via a two-stage cluster analysis using percentage of ownership and
MCS effectiveness (Gerdin, 2005). We measure ?t as the Euclidean
distance of the organisation’s MCS pro?le from its empirical ideal
MCS based on the top performing organisations within the cluster.
Since TCE predicts the most ef?cient MCS given ownership, we
measure performance as ?nancial performance relative to peers.
The results support our prediction of a positive relationship
between ?t and organisational performance. Sensitivity analyses
using an objective measure of performance based on gross fee rev-
enue and using ?t measured relative to the cluster average MCS
pro?le reveal results to be robust to the choice of performance
measure and choice of benchmark to de?ne the ideal MCS design.
As a by-product, we also ?nd the organisations that self-assess ashttp://dx.doi.org/10.1016/j.aos.2015.06.002
0361-3682/Ó 2015 Elsevier Ltd. All rights reserved.
?
Corresponding author.
E-mail addresses: [email protected] (R. King), [email protected].
edu.au (P. Clarkson).
1
This study is based on Robyn King’s PhD thesis completed in the UQ Business
School at the University of Queensland.
2
Limiting conditions are scale, complexity, capital intensity, commodi?cation,
litigation and social trends (Greenwood & Empson, 2003).
Accounting, Organizations and Society 45 (2015) 24–39
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er . com/ l ocat e/ aos
having effective MCS conform with Speklé’s (2001) theoretical
ideal ownership-archetype pro?les whereas those that reported
an ineffective MCS do not.
Our study extends the MCS literature by investigating owner-
ship as a relevant contextual variable when evaluating the impact
of MCS design on performance. Our evidence indicates that there
are different optimal combinations of GP ownership and MCS
design with similar ?nancial performance outcomes, consistent
with the concept of equi?nality (Gresov & Drazin, 1997). Our ?nd-
ings also contribute to the ongoing debate about the suitability of
TCE as a holistic theory of MCS design (Speklé, 2001). From a prac-
tical perspective, our evidence has the potential to assist managers,
owners and advisors optimise MCS design for the organisation’s
given level of ownership. While we conduct our study within the
context of the Australian primary healthcare sector given its eco-
nomic and social signi?cance and the signi?cance of the control
problem within this sector, our results are applicable more
broadly. Similarities among the primary healthcare sectors in
Australia, the U.K., and the U.S. make our ?ndings of interest inter-
nationally. Further, since the conceptual foundation of our study is
not restricted to one particular context, our ?ndings are also appli-
cable to other professional service sectors.
The remainder of this study is organised as follows. We next
discuss the literature on ownership and MCS design in professional
services organisations, and describe the Primary Healthcare sector
in Australia. The third section develops our hypothesis, and the
fourth section describes our research design and sample data. The
?fth section reports our results and the sixth section concludes.
2. Control in the professional services sector
2.1. The role of ownership in professional service organisations
Early organisational theorists propose ownership as the ideal
control solution for professional service organisations. As owners,
professionals will have the residual rights to control and the
incentive to make decisions that will create, maintain and improve
the organisation (Hansmann, 1996). Ownership also reduces
the likelihood of the professional leaving and is a form of cultural
control that encourages mutual monitoring (Merchant & Van der
Stede, 2007). Consistent with this view, Greenwood, Deephouse,
and Li (2007) compare the performance of large management
consultancies and ?nd that private corporations and partnerships
outperform public corporations.
If these predictions and ?ndings hold, we would expect to see
all professional service organisations owned by the professionals
working in them.
3
However, there are two arguments as to why,
in practice, ownership rights may represent an incomplete solution
to their control challenge. First, a necessary condition for owner-
ship to develop as a complete solution is a stable regulatory and
institutional setting (Mintzberg, 1979). In a dynamic environment,
ef?cient ownership may not be quickly achieved due to the long
term nature of the ownership arrangements, limits to the cognitive
abilities of the contracting parties, and the costs of changing
arrangements (Richter & Schroder, 2008). Further, if the industry
is, in some sense, relatively immature, ownership measured at a
point in time can be considered as exogenous (Larcker &
Rusticus, 2007).
Second, even given a stable setting, there are a number of
limiting factors at the organisational level. These factors include
differences in the amount of capital the individual owners can pro-
vide, their requirements for division of returns, and their priorities
including pro?t generation, employment security and working
time (Greenwood & Empson, 2003). Due to these differences, there
will be varying degrees of alignment between personal and organ-
isational goals, leaving a residual control problem (Ittner, Larcker,
& Pizzini, 2007). With diffused ownership, there also is the possi-
bility of shirking (Gaynor & Gertler, 1995), as well as the need to
co-ordinate decision-making among multiple owners and to con-
trol individual activities to achieve ef?cient outcomes. As a result,
Richter and Schroder (2008) propose that internal governance,
speci?cally MCS design, can augment ownership to arrive at a
more complete control solution.
Following Richter and Schroder (2008) and Empson and
Chapman (2006), we propose a role for the MCS as part of the con-
trol solution. If due to constant changes in the environment, own-
ership is not yet in equilibrium, there should be variation not only
in the observable ownership but also in MCS design. In circum-
stances where ownership is in some sense sub-optimal, the man-
ager can more readily adjust the MCS design to achieve ef?cient
performance. Of direct relevance, if the MCS is designed in such a
way that it is optimal for the level of ownership, taken together
ownership and the MCS should reduce overall control costs and
enhance organisational performance. There is some evidence that
professional partnerships and public corporations can be equally
effective if systems and structures are suitably constructed, with
the caveat that members must be strongly committed to the pro-
fessional interpretive scheme (Empson & Chapman, 2006). There
is also the possibility that a mismatch between the MCS design
and ownership might occur in the short run with negative perfor-
mance implications (Empson & Chapman, 2006).
2.2. Primary healthcare
The Australian health and aged care sector represents one-tenth
of the economy and is predicted to grow to one-eighth in the next
twenty years (NHHRC, 2009). There is universal health coverage
with one main funding body, Medicare. The GP is the ?rst point
of contact for a majority of patients, providing 88% of their required
care and is the recognised ‘‘gatekeeper’’ as a referral is required to
access specialist, secondary and tertiary care (IBIS, 2011). GPs work
primarily in small privately held PHOs that employ nurses, admin-
istrators and increasingly practice managers (DHA, 2005). Over the
last two decades, PHOs have grown from a majority having one or
two GPs in 1994, to a majority having ?ve or more in 2010–2011
(AIHW, 2012). Since 1998, there has been a shift towards corporate
ownership by publicly listed companies that currently have 12% of
the market, and approximately 72% of GPs now work in PHOs they
do not own (Kron, 2012). Payment is mostly on a fee-for-service
basis, although since 2000 there has been an increase in blended
payments known as Practice Incentive Payments (PIPs). PIPs are
a group reward for PHOs that require collective action of their
GPs and represent 9% of income (ANAO, 2010). To receive PIPs,
PHOs must be accredited to Royal Australian College of General
Practice (RACGP) standards every three years and meet the
requirements of the thirteen PIP categories (DHS, 2011).
4
3
There is some evidence of clustering of ownership structures. In a professional
services setting, Richter and Schroder (2008) ?nd size, service standardisation, capital
requirements and risk to be determinants of ownership, and conclude that it is a
combination of these factors that determines the optimal allocation of ownership
rights. There are two provisos. First, the dif?culty in raising capital and the limited
capacity of employees to absorb risk pose limits to internal ownership. Second,
internal ownership constrains the size of ?rms.
4
The RACGP standards for general practice cover ?ve areas: practice services;
rights and needs of patients; safety, quality and improvement; practice management;
and physical factors (RACGP, 2011). The amount of the PIP is based on the number of
full time equivalent GPs, whole patient equivalents and the meeting of a number of
performance measurement targets such as delivery of after-hours care, the use of
information technology, teaching, rurality, preventative services for at risk patients,
and quality prescribing habits.
R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39 25
The Australian primary healthcare system shares similarities
with both the U.K. and the U.S. The U.K. also has a policy of universal
health coverage with one main funding body, the NHS, where GPs
act as gatekeepers and work in GP-owned PHOs that are increasing
in size (Addicott & Ham, 2013). Unlike Australia, it is the PHO and
not the individual GP that is entitled to payments which are a
mix of capitation, fee-for-service and pay for performance. In the
U.S. there is a multi-payer system where the GP is not the gate-
keeper (Kirchhoff, 2013).
5
There is mostly fee-for-service but also
some capitation payments. Historically, U.S. GPs have operated in
small GP-owned PHOs but over the last two decades solo practice
and GP ownership has also declined (Kane & Emmons, 2013). In all
three countries, GP payments account for approximately 20% of total
health care spending but GPs direct as much as 90% of the total
(Kirchhoff, 2013). There is also a similar emerging phenomenon of
‘‘Lifestyle preferences with younger doctors more willing than their
predecessors to work for an outside institution to secure a set sched-
ule and salary. . . Physicians may be having a harder time ?nding doc-
tors to buy or join a small practice, as management becomes more
complex and average compensation declines’’ (p. 2, Kirchhoff, 2013).
2.3. The control dilemma for managers in primary healthcare
The primary role for the PHO manager is to ensure that the GP
provides the necessary volume and quality of service to achieve
and maintain PHO pro?tability. In Australia, the Medicare fee has
not increased in line with in?ation and so increasingly PHOs are
reliant on blended payments (PIPs) (Richardson, Walsh, & Pegram,
2005). In this funding environment, the focus of ‘for pro?t’ PHOs is
to provide services that meet the RACGP and PIP standards
ef?ciently.
PHO managers are faced with variance in GP behaviours, a dif?-
cult monitoring environment and resistance to bureaucracy. The
innate nature of the GP output is heterogeneous as there are differ-
ences in work habits and pace that result in differences in the qual-
ity and value of services produced and resources consumed (Town,
Wholey, Kralewski, & Dowd, 2004). Direct monitoring is impossible
as the consultationtakes place ina soundproof consulting roomand
GPs usually treat patients independently. Managers cannot readily
assess GP output given its intangible nature (Merchant & Van der
Stede, 2007). In?uencing GP behaviour is problematic because their
preferences con?ict with bureaucracy, including a need for auton-
omy and a need to preserve the social status of their profession
(Barley, 2005). A seminal U.S. case study ?nds employed GPs rely
on their professional expertise to achieve suf?cient dominance
and authority to pursue their own goals, reinforcing the extent of
the control challenge facing practice managers (Freidson, 1975).
2.4. The healthcare ownership, MCS and performance literature
Existing literature on MCS and performance within the health-
care sector includes studies of hospitals and primary healthcare
organisations. MC studies of hospitals categorise ownership as
‘private for pro?t’, ‘private not for pro?t’ or ‘government owned’
and then investigate performance (Eggleston, Shen, Lau, Schmid,
& Chan, 2008; Shen, Eggleston, Lau, & Schmid, 2007). These studies
have varying results, with some concluding that ‘for pro?t’ hospi-
tals adopt more management techniques but with lower quality
outcomes. Our construct of ownership, GP ownership, has only
recently emerged in the U.S. hospital literature as a potential
incentive alignment mechanism.
6
There is also little evidence of an association between
ownership, MCS design and performance in the primary healthcare
sector (APHCRI, 2010). Reviews reveal that various components of
governance are well documented but there is a lack of systematic
mapping of these components on contingent factors and outcomes
(Stewart, 2002; Tollen, 2008). Individual characteristics considered
include culture (Smalarz, 2006), standardised clinical practice,
performance measurement, and transparency (Audet, Doty,
Shamasdin, & Schoenbaum, 2005), leadership (Casalino, Devers,
Lake, Reed, & Stoddart, 2003), goal setting (Curoe, Kralewski, &
Kaissi, 2003), planning and accountability (Rittenhouse &
Robinson, 2006), incentive systems (Mehrotra, Epstein, &
Rosenthal, 2006), and selection of workforce and patient centered-
ness (Rittenhouse & Robinson, 2006). However, this research does
not combine these attributes, nor does it indicate the directionality
between these attributes and ef?ciency (Tollen, 2008). We extend
the literature by empirically examining the relationship between
ownership, MCS design and organisational performance appealing
to TCE theory to make our predictions.
3. Theoretical foundation and hypothesis development
TCE has been proposed as a theory capable of predicting
ef?cient MCS design while allowing for the possibility of
‘misalignment’ and its effect on performance (Hakansson & Lind,
2007). Speklé (2001) argues that TCE is useful because it is focused
at the micro-analytical level, adopts the behavioural assumptions
of bounded rationality and opportunism, and is based on
minimising transaction costs, and hence can explain why a
particular MCS design is ef?cient for achieving an organisation’s
goals without having to specify the goals (i.e., instrumental
effectiveness).
7
The existing TCE intra-organisational MCS research is in its
infancy and there is minimal empirical evidence on which to rely
when making predictions about MCS design (Macher & Richman,
2008).
8
Speklé (2001) theorises optimal MCS archetypes based on
the characteristics of transactions identi?ed in TCE, asset speci?city,
uncertainty, frequency and post hoc information impactedness.
9
Post hoc information impactedness is a derivative of uncertainty
and opportunism, and is related to ‘‘the extent to which the organi-
zation is able to observe and to assess perceptively the true quality
of actually delivered contributions’’ (p. 431, Speklé, 2001). High
information impactedness exists when transaction information is
known to one party but is costly or even impossible for others to
obtain. This situation might arise if information is withheld oppor-
tunistically or if the second party lacks the specialised knowledge
to understand the information and there is a high cost to assess
the true level of actually delivered inputs.
5
The U.S. equivalent title for GP is primary care physician. For consistency and
clarity, the term GP is used here (Kirchhoff, 2013).
6
In contrast, in Japan physician hospital ownership studies have associated it with
con?icts of interest (Rodwin & Okamoto, 2000).
7
Principal agent theory is a possible alternative theoretical foundation that takes
an ex ante perspective assuming rationality and so predicts organisations are
optimising with no direct capacity to test empirically for performance effects (Luft &
Shields, 2007). We prefer TCE which takes an ex post perspective assuming bounded
rationality allowing for the possibility of ‘misalignment’ between the chosen ex post
solution, the MCS and the transaction characteristics. This ability to provide a causal
map has been identi?ed as a desirable attribute of any theory of MCS design (Luft &
Shields, 2007).
8
TCE combined with contingency theory was the basis for a study of the
relationship between strategic human capital and MCS design evidencing a positive
relationship between personnel, non-traditional results control and use of strategic
human capital (Widener, 2004).
9
These archetypes have been subject to testing by Kruis (2008) who employs a
cross-sectional survey to investigate the relationship between the transaction
characteristics, ?ve of Speklé’s (2001) MCS archetypes and effectiveness. Kruis ?nds
some support for Speklé’s arm’s length control archetype but small sample sizes
precluded testing others. There is also a case study using TCE as its foundation that
investigates changes in MCS related to the restructuring of the Shell chemical
businesses (Van den Bogaard & Speklé, 2003).
26 R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39
Against this backdrop, we argue that within our setting, the
PHO transaction of interest, namely the provision of care to
patients by the GP, is characterised by both high asset speci?city
and high uncertainty. Further, we theorise that information
impactedness is directly related to the level of GP ownership.
Speci?cally, asset speci?city is high as the GP’s knowledge of the
patient and the PHO’s systems develops over time and is not easily
transferred (Sturmberg & Martin, 2008). Further, if the GP leaves
the practice, this valuable knowledge asset is lost and is not readily
replaced (WHO, 2006). There is also high uncertainty as the organ-
isation cannot ex ante prescribe GP actions within a consultation.
Consultations are becoming more dif?cult due to the increasing
incidence of chronic and complex conditions, and the uncertainty
surrounding the ef?cacy of treatment options (Holmberg, 2006).
Uncertainty is also increased because at all times, the patient can
have ‘‘input to or cause disruption in the production process’’ (p.
139, Chase, 1978).
In terms of an association between ownership and information
impactedness, we argue that when all professionals are owners,
there are reduced incentives for withholding information, suf?-
cient specialised knowledge to understand the information, and
greater incentives to monitor each other’s performance. Hence,
information impactedness will be low since the knowledge of per-
formance can be spread throughout the organisation at relatively
low cost. Alternatively, if ownership is fully allocated to
non-professionals, information asymmetry remains high and infor-
mation impactedness will likely remain high.
10
Lastly, we present ownership and MCS as acting jointly as part
of the control solution. In this regard, Speklé (2001) notes the level
of information impactedness is not necessarily an exogenous vari-
able treated as a given transaction characteristic but may be ‘‘and
often is – the product of control structure choice’’ (p. 431).
However, given the frequent funding policy changes made by suc-
cessive Australian governments over an extended period, we view
the Australian primary healthcare sector as immature in terms of
its ownership equilibrium and combined with the manager’s
inability to directly in?uence the level of GP ownership, treat own-
ership as exogenous in our context.
11
This turbulence in the regula-
tory setting and the resultant variation in GP ownership in
Australian PHOs provides a contextual setting to investigate our pre-
dictions based on TCE without the need for additional controls for
industry effects.
Of direct relevance here, given a setting of high asset speci?city
and high uncertainty, Speklé (2001) predicts that with high infor-
mation impactedness (low ownership), the optimal MCS design is
a boundary archetype whereas with low information impactedness
(high ownership), the optimal MCS design is an exploratory
archetype.
12
With high uncertainty, it is not possible to prescribe
accurately ex ante the actions required of the contributor. If there is
also high asset speci?city, there are high costs associated with ex post
monitoring due to the specialised character of the information on
contributions. Given this combination, high information impacted-
ness makes explicit contracting for concrete actions infeasible. The
aim of control then shifts to the prevention of undesired actions
and outcomes. The boundary archetype features administrative con-
trols achieved through interdictions, emphasising behaviour to be
avoided including proscriptive codes of conduct and often carrying
stringent penalties for non-compliance. In contrast, with low infor-
mation impactedness, the optimal MCS design is one that facilitates
the ?ow of information within the organisation. This MCS is consis-
tent with an exploratory archetype which features controls that are
engaged in creating and preserving information sharing, and in
re-adjusting and re-aligning perceptions on progress throughout
the life of the contract. Suggested controls include information shar-
ing entrenched in organisational structure and process design, perfor-
mance evaluation based on emergent standards, rewards through
promotion (including periodic salary revision) based on long term
performance and little emphasis on formal instruments of control.
13
In sum, accepting the link between ownership and information
impactedness, we reframe Speklé’s (2001) predictions as they
apply to our setting as follows: given a setting of high uncertainty
and high asset speci?city, with low ownership the optimal MCS design
is a boundary archetype whereas with high ownership it is an explora-
tory archetype.
Finally, predictions based on TCE represent the ef?cient
matches between ownership and MCS design because they min-
imise total costs and maximise organisational performance. To
illustrate, ?rst consider the situation where the professionals
retain all ownership rights. Here, the costs of reduced diversi?ca-
tion and a limited investment base are best countered by the adop-
tion of the lower cost, less formal, exploratory MCS (Richter &
Schroder, 2008). Alternatively, when ownership is opened to out-
siders, the higher cost of the more formal, boundary MCS is offset
by both the ability of the outside owners to diversify and the
greater access to capital. In reality, however, it is possible that
not all organisations will have their predicted ef?cient MCS design
at all times. Such a ‘mismatch’ or lack of ‘?t’ might occur because of
bounded rationality wherein some managers are better able,
within a complex setting, to identify their optimal MCS design
(Merchant & Van der Stede, 2007). Equally, it might occur because
of a ‘‘lag’’ between the recognition of the need for a change in the
MCS and the change occurring due to a lack of available resources,
because of a ‘lumpy’ change such as introducing a system with the
capacity for future growth, or because there is an unexpected
change in the operating environment. Such lags or lumpiness will
likely result in periods when the adopted MCS is less ef?cient (Luft,
1997). Thus, irrespective of cause, the result of a lack of ‘?t’ will be
a control loss, de?ned as the difference between performance that
is theoretically possible and that expected given the MCS in place
(Merchant & Van der Stede, 2007). Our hypothesis, stated in the
alternate, then directly follows as:
H1. Performance is positively related to the extent of ‘?t’ between
level of professional service worker ownership and MCS design.
4. Research design and sample data
4.1. Operationalising the MCS construct
Our interest is in the MCS as a whole, rather than in sub-groups
of controls. Grabner and Moers (2013) propose that ‘‘MC practices
form a system if the MC practices are interdependent and the
10
In reality, there will likely be a continuum of ownership ranging from 0% to 100%.
We restrict our arguments to the extremes because the form of the relation between
ownership and information impactedness is as yet untested. The level of information
impactedness is likely also in?uenced by the size of the organisation as the costs of
sharing information on inputs to transactions will increase with the number of
professionals working in them (Williamson, 1975). Size is therefore included as a
control in our analyses.
11
For example, starting in the late 1990’s, ?nancial incentives were provided for
PHOs to amalgamate, resulting in a trend towards larger PHOs (IBIS, 2011), in 1996,
the payment system changed from purely fee for service to include PIPs, and since
2000, the RACGP standards have changed substantively three times and there have
been 33 key changes to criteria for PIPs (ANAO, 2010).
12
This line of reasoning applies more broadly to professional service transactions as
they are likely characterised by high asset speci?city as knowledge of the client and
organisation’s systems is not easily transferred and not readily replaced and high
uncertainty because the organisation cannot ex ante prescribe what the professional
needs to do in each transaction.
13
The exploratory archetype is related to Mintzberg’s (1979) ‘adhocracy’ and the
‘organic’ organisation (Burns & Stalker, 1961).
R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39 27
design choices take these interdependencies into account. In con-
trast, MC as a package represent the complete set of control prac-
tices in place, regardless of whether the MC practices are
interdependent and/or the design choices take interdependencies
into account’’ (p. 4). As we are predicting the optimal combination
of controls that are ef?cient for a singular control problem, GP
behaviour, it is the MC system design and not a MCS package that
is the construct of interest here.
Although Speklé (2001) provides general descriptions of his
archetypes of control, he does not provide in detail the individual
MC practices that make up each of the archetypes. Since PHOs
are small organisations where managers may not have a complete
set of formal MC practices in place, we may omit a correlated vari-
able if we simply ask our respondents about the exploratory versus
boundary characteristics of their MCS. To operationalise the MCS
construct, we therefore adopt the framework proposed by Malmi
and Brown (2008) for three basic reasons. First, within our context,
the control challenge being investigated is that of directing GP
behaviour, not of strategy formulation. We view the Malmi and
Brown (2008) de?nition of MCS (‘‘Those systems, rules, practices,
values and other activities management put in place in order to
direct employee behaviour’’) as consistent with the GP control
challenge. Second, the Malmi and Brown framework describes an
extensive range of controls likely to be present in small PHOs
and so provides a comprehensive descriptive framework to
ensure survey coverage of our MCS domain. Third, it includes
cultural controls as overriding controls able to be affected by
management. These controls have been found to be important in
professional services settings (Freidson, 1975). Thus, we argue that
using the Malmi and Brown (2008) framework provides some
assurance of a complete coverage of the domain of the MCS
construct.
14
From the empirical perspective, a critical challenge presented
by the Malmi and Brown (2008) framework is that it is purely
descriptive of MC types and does not suggest how to measure each
control practice. To provide operational context, we enlist the ?rst
eight questions from the Ferreira and Otley (2009) PMS frame-
work.
15
Ferreira and Otley state the ‘‘general nature of the frame-
work enables other frameworks to be used to complement its
interpretations and insights.’’ (p. 265).
4.2. The survey questionnaire
We develop the initial questionnaire from the MCS literature,
notably Ferreira and Otley (2009). We also use input from inter-
views with managers from seven PHOs that differed on ownership
and performance. We pretested the questionnaire on thirteen
experts including experienced practice managers, GPs, and
accounting and health care academics. We then modi?ed the ques-
tionnaire and conducted a pilot study on thirty practice managers
as a further check of reliability and validity (Van der Stede, Young,
& Chen, 2005). We used the analysed pilot results and participant
feedback to improve the face and content validity of questionnaire
items.
The ?nal questionnaire consists of 111 questions presented in
nine panels. A summary is presented in Appendix A. The ?rst ?ve
panels relate to the ?ve types of controls (p. 291, Malmi &
Brown, 2008). All questions use a 7-point Likert scale and with
two exceptions are worded such that higher scores re?ect more
formal rule-based controls, representative of a boundary archetype
of control; the two exceptions are Socialise and Selection.
Following the speci?c questions relating to each of the ?ve
types of controls, we measure two additional summary constructs
(Ferreira & Otley, 2009). The ?rst construct is the overall emphasis
placed on each control type. To measure emphasis, we ask respon-
dents to re?ect upon their answers to the speci?c questions about
control practices within each control type and to indicate the
emphasis placed on that control type for managing the behaviour
of the GPs. Consistent with the wording of the underlying control
practice questions, higher scores are indicative of a boundary
archetype. The purpose of including the ?ve emphasis measures
is to compare the MCS of different PHOs at the control type level
as a summary measure of the theoretical archetypes. For the sec-
ond construct, we ask respondents to take their responses to the
questions on individual control practices into consideration and
indicate the effectiveness of each particular control type for
managing the behaviour of GPs. This question is designed to differ-
entiate organisations on the perceived effectiveness of their chosen
MCS (Chenhall, 2007; Kruis & Widener, 2009) and is a subjective
evaluation by the manager about the usefulness of the MCS design.
The ?nal four panels contain questions on organisational per-
formance, contingency factors, and both PHO and manager charac-
teristics. For performance, we collect subjective measures of the
PHO’s relative pro?tability, competitiveness, market share, growth,
innovativeness and size (Govindarajan & Gupta, 1985). For com-
pleteness, we also collect data on measures of learning (Widener,
2007), patient satisfaction (Shortell & Rundall, 2007), accreditation
scores and gross fees.
We include questions for the contingency variables ‘Size’,
‘Structure’, ‘Strategy’ and ‘Perceived Environmental Uncertainty
(PEU)’ for use in robustness testing. We sourced the items from
the literature as follows: size is the number of full time equivalent
employees including GPs (FTE) (King, Clarkson, & Wallace, 2010);
structure is measured using six items that ask about the degree
of decision making delegated to the manager (Gordon &
Narayanan, 1984); strategy is measured using a single item which
distinguishes between cost leadership and price differentiation
(Govindarajan, 1988); and PEU is measured using four items, two
on competition and two on environmental turbulence in the exter-
nal environment (Gordon & Narayan, 1984).
Finally, we collect manager and PHO characteristics to assist in
the testing of possible non-response bias and as controls for the
statistical analysis. For managers, we collect their highest level of
management quali?cation, years of experience in managing a
PHO, and relationship with the owner(s). For PHOs, in addition to
size (FTE), we collect data on years of operations (as a measure
of organisational lifecycle), percentage of private billing (as a mea-
sure of pro?t margin), and the percentage of GPs working in the
organisation that were owners (as the measure of ownership).
4.3. Sample organisations
4.3.1. The sampling frame
The identi?ed sampling frame is the population of PHOs with
three or more GPs in Australia. These organisations likely present
a greater potential control problem for managers than smaller solo
or dual GP practices (Ittner et al., 2007). Additionally, the trend has
been towards increasing practice size and greater prevalence of
practice managers which has led to an increased likelihood of
MCS implementation (DHA, 2013). We therefore focus on GP
14
Malmi and Brown (2008) de?ne organisational structure as the degree of
functional specialisation. In our pre-survey interviews, we identi?ed organisational
structure control practices as the existence and use of an organisational chart and GP
position descriptions. In contrast, the contextual variable labelled as ‘Structure’ is a
separate and distinct construct de?ned as the degree of decentralisation of decision
making (Gordon & Narayanan, 1984). The decision on the degree of decentralisation is
typically made by the owners and not readily changed by the practice manager. It is
therefore not a control practice used by the manager for controlling GP behaviour and
hence is not considered a part of the MCS but rather a contextual variable.
15
We have not selected it as our framework because with its inclusion of strategy
reformulation, its de?nition of control is broader than required here, and cultural
controls are not included in the PMS.
28 R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39
groups of three or more due to the greater need for, and capacity of
managers to design MCS to direct GP’s behaviours (Merchant &
Van der Stede, 2007). While there is no publicly available informa-
tion on how many organisations ?t these criteria, the closest
approximation is that there were 4502 PHOs with two or more
GPs in 2008 (PHCRIS, 2008). The target survey participants are
practice managers as they are likely to have the greatest knowl-
edge of the organisation’s MCS design and performance.
As ownership information for Australian PHOs is not publicly
available, following King et al. (2010), we approached the
Australian Association of Practice Managers (AAPM) to assist in
identifying and contacting suitable study participants. The AAPM
is the only recognised professional body for practice managers in
Australia. Because membership is voluntary, subject to an annual
subscription, it is likely that AAPM members are interested in
current management trends, wish to become part of a professional
network and have resources available to pay the fee. From the
limited AAPM membership data, it appears that this selection
method introduces the potential for bias as AAPM members are
likely to be from larger PHOs with greater available resources.
Notwithstanding, we consider the advantages of accessing AAPM
members and having the AAPM’s support to outweigh the potential
problem of bias (Dillman, 2000). Further, as discussed in the next
subsection, this bias does not, in fact, reveal itself in our data.
4.3.2. The survey process and survey respondents
By request of the AAPM, the survey was conducted as an inter-
active web-based survey by Ultra Feedback, a commercial survey
group. The advantages of online surveys are increased speed of
response, lower cost and less data entry than mail surveys
(Crawford, Couper, & Lamias, 2001). As recommended by Dillman
(2000), the survey was accompanied by an invitation letter with
links to an endorsement letter from the AAPM president and a par-
ticipant information sheet. There were a total of two reminders,
the ?rst two weeks after the initial email and another a week
later.
16
E-mail addresses for one practice manager from each of 451
PHOs identi?ed as potentially satisfying the selection criteria were
provided to Ultrafeedback by the AAPM. Of these, 193 managers
opened the survey, and 178 ?t the selection criteria (PHOs with
three or more FTE GPs). Fifty-eight responses were identi?ed as
having signi?cant missing values, leaving a ?nal sample of 120
respondents.
17
This represents a usable response rate of 26.6%
which compares favourably with other management accounting
studies (Bisbe & Malagueno, 2012; King et al., 2010).
We screened the survey data for possible non-response bias by
comparing the ?rst and last 30 responses via t-tests (Moore &
Tarnai, 2002). We ?nd smaller PHOs with managers having greater
experience more likely to respond early, thereby raising the possi-
bility of non-response bias. To address this concern, the chosen
cluster analysis solution (Section 5) was scrutinised for differences
in the size of PHOs, revealing no statistically signi?cant differences.
We also include size as a control variable in the regression analy-
ses. We performed a Harman’s one factor test which resulted in a
17-factor solution with the ?rst factor explaining 24.77% of the
total variance. As a result, common method variance was not con-
sidered a serious threat (Podsakoff & Organ, 1986).
Descriptive demographics for the 120 respondents are provided
in Table 1. Data reveal considerable cross-sectional variation in
both ownership and gross fee revenue. GP ownership (%
Ownership) ranges from 0% to 100%, with a mean value of 38%
and a standard deviation of 26%. For the 84 PHOs that provided
the data, gross fee revenue ranges from $600,000 to $5 million,
with a mean value of $2.296 million and a standard deviation of
$1.017 million. The mean number of FTE employees is 15.7 and
the mean number of GPs working in the PHO is 6.58.
Given the exclusion of PHOs with two or fewer GPs, the sample
mean number of FTE employees is greater than the population
average of 5.73 (IBIS, 2011). For further comparison, 37.5% of the
sample had three or four GPs, and the remaining 62.5% had ?ve
or more whereas after excluding solo practices, 38.9% of the
remaining population has between two and four GPs, and 61.1%
have ?ve or more (IBIS, 2011). Similarly, the sample mean value
for gross fees exceeds the population average of $970,934 for the
2009–2010 ?nancial year (IBIS, 2011).
Finally, for the constructs with multiple measurement items,
we conducted exploratory factor analysis using PCA with orthogo-
nal rotation for each of the ?ve control types, as well as effective-
ness, PEU,
18
and structure (Tabachnick & Fidell, 2007).
19
We
eliminated four items, two for insuf?cient loadings and two for cross
loading.
20
In line with expectations, the PCA’s revealed 18 compo-
nents, six for cultural controls, two for planning, three for cybernetic,
one for rewards and compensation, and six for administrative con-
trols. While the Cronbach Alphas (CA) were below the recommended
limit of 0.6 for three components (Recruit, 0.532; Selection, 0.472;
and Policy and Procedures, 0.446), we attribute the results to the
small number of measurement items and retain these components
keeping the CA in mind (Hair et al., 2010).
Based on the results from the PCA, we create summated scores
for each of the components as they can be more easily reproduced
in future research (Hair et al., 2010). Table 2 presents descriptive
statistics for the summated scores. When compared with the factor
scores, there were consistently high correlations. Further, when the
sample is split between high and low ownership, a comparison of
the summated scores, factor scores and highest loading items
reveal the same pattern of differences (Hair et al., 2010). Our anal-
yses are therefore based on the summated scores.
21
5. Empirical methodology and results
5.1. Empirical strategy
To test H1, we adopt a con?guration/contingency approach
(Gerdin & Greve, 2004). The underlying assumption of the con?g-
uration approach is that there are ‘‘only a few states of ‘?t’ between
context and structure, with organisations having to make quantum
jumps from one state of ‘?t’ to another’’ (p. 304, Gerdin & Greve,
2004). It is similar to the systems approach and takes a holistic
view such that multiple variables are retained in the analysis
(Venkatraman & Prescott, 1990). In conjunction, a contingency
view assumes that rather than only the best-performing organisa-
tions surviving to be observed, organisations have varying degrees
of ‘?t’ with their context. Using this approach, the researcher must
demonstrate empirically that higher degrees of ‘?t’ are associated
with higher performance.
16
Copies of the AAPM cover letter and the original survey questionnaire are
available from the authors upon request.
17
Remaining missing data assessed by a t-test and Little’s MCAR test statistic
(p > 0.10) as missing completely at random (MCAR) were replaced using the
expectation maximisation EM estimation algorithm in SPSS as recommended by
Hair, Black, Babin, and Anderson (2010).
18
Consistent with the literature (King et al., 2010), we extract two factors that we
label as PEU1-competition and PEU2-dynamism.
19
For robustness, we also conducted CFA. Orthogonal rotation was chosen as the
controls in the MCS are not necessarily theoretically correlated, and the resulting
uncorrelated scores are more suitable for the subsequent analyses (Hair et al., 2010).
20
Loadings less than |0.50| were considered insuf?cient and when an item had
loadings greater than 0.45 on two factors it was considered as a cross loading (Hair
et al., 2010; Tabachnick & Fidell, 2007).
21
All analyses were also conducted using factor scores with results qualitatively
identical.
R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39 29
Pro?le deviation analysis is the method recommended to eval-
uate the association between ‘?t’ and performance (Gerdin &
Greve, 2004). It assumes that ‘?t’ is the degree of adherence to
an externally speci?ed ideal pro?le and lack of ‘?t’ will have perfor-
mance implications (Drazin & Van de Ven, 1985). Following the
majority of literature, we develop the ideal pro?le empirically.
We cannot, however, use performance to cluster because it creates
endogeneity (Jermias & Gani, 2004). We therefore develop the ideal
pro?le by forming clusters of PHOs based on GP ownership and
perceived MCS effectiveness. As the number of clustering variables
increases, it becomes more dif?cult to interpret which variable has
the greatest in?uence and thus there is greater researcher subjec-
tivity in making the choice of the most valid solution. By using only
two clustering variables standardised prior to clustering, we min-
imise researcher subjectivity in the choice of solution.
After clustering, for the clusters that have scores indicating
effective MCS, we identify their ideal empirical MCS pro?le as
the average scores for each of the 18 controls from the top
Table 1
Descriptive pro?le for a sample of 120 Australian primary healthcare organisations.
Characteristic N Mean Median Std Dev Min Max
GP 120 6.576 6.000 3.567 3 27
FTE 120 15.700 14.050 7.979 5 62
% Ownership 120 0.384 0.333 0.255 0 1
Gross Fees ($) 84 2,295,983 2,200,000 1,016,884 600,000 5,000,000
Lifecycle 119 31.910 26.000 24.966 0.30 135
Private Billings 117 42.682 40.000 23.367 0 95
Manager experience 118 11.199 11.000 6.691 1 30
Variable de?nitions: GP is the number of GP’s working in the practice; FTE is the number of full time equivalent workers;% Ownership is the percentage of FTE GPs working in
the organisation who are also owners; Gross Fees is the organisation’s total gross fee revenue; Lifecycle is the years the PHO has been operating; Private Billing is the percentage
of total gross fees derived from private (non-bulk) billings; and Manger experience is the number of years of experience the practice manager has in managing PHOs.
Table 2
Survey questionnaire item response descriptive statistics.
Measure Mean Median Std Dev 0–1.0 1.1–2.0 2.1–3.0 3.1–4.0 4.1–5.0 5.1–6.0 6.1–7.0
Cultural controls
Socialise 4.786 5.667 2.137 5 13 6 7 13 22 54
Code of conduct 4.501 5.625 1.577 0 7 16 16 30 26 25
Vision and mission 4.562 5.000 1.869 8 5 6 14 21 34 32
Dress code 5.349 5.500 1.050 0 1 2 4 28 42 43
Recruit 2.593 2.000 1.820 15 33 17 16 20 10 9
Selection 4.251 4.000 1.516 0 9 12 18 33 27 21
Planning controls
Long range planning 3.823 4.000 1.974 10 14 12 18 25 20 21
Short range planning 4.260 4.625 1.853 4 13 8 19 23 29 24
Cybernetic controls
Budgets 4.486 5.00 2.042 9 9 9 11 19 26 37
Boundary 4.242 4.667 1.999 9 11 4 12 30 22 32
Non-?nancial 2.478 2.333 1.743 21 29 19 14 24 9 4
Rewards &Comp. 2.892 3.333 1.642 27 4 17 26 38 7 1
Admin. controls
Rules 3.807 3.800 1.721 8 8 19 30 19 21 15
Position 4.279 5.000 1.589 1 7 10 11 26 35 30
Organisational committees 6.039 6.667 1.421 1 2 3 1 11 15 87
Chronic disease management 4.908 5.333 1.721 1 6 6 9 24 37 37
Policies and procedures 3.787 4.333 2.125 13 17 6 14 22 27 21
Meetings 3.217 3.000 1.446 1 21 27 23 32 9 7
Performance
Overall performance 4.946 5.000 1.399 3 2 1 14 28 41 31
Relative ?nancial performance 4.450 4.000 1.764 7 2 3 15 36 20 37
More competitive 4.720 5.000 1.704 6 6 9 27 32 21 19
Greater market share 5.030 5.000 1.655 5 5 3 29 23 32 23
Growing faster 5.030 5.000 1.700 7 2 7 20 29 33 22
More innovative 5.400 6.000 1.677 6 2 3 15 22 40 32
Larger in size 5.040 6.000 2.023 11 5 4 22 9 37 32
Gross fees ($ millions; n = 84) 2.295 2.200 1.016
Learning 5.858 6.000 1.285 3 1 2 6 23 42 43
Accreditation (n = 82) 4.107 4.000 0.750 0 1 30 23 28 – –
Patient satisfaction 5.244 5.333 0.863 0 0 0 8 32 49 31
Contextual variables
Size 15.700 14.050 7.979
Lifecycle 31.910 26.000 24.966
Private billings 42.682 40.000 23.367
Strategy 5.042 5.000 1.266 3 2 4 26 41 31 13
Structure 4.951 5.000 1.343 0 5 7 8 32 35 33
PEU1 – competition 3.009 3.333 1.323 10 14 25 31 34 6 0
PEU2 – dynamism 4.609 5.000 1.448 0 6 8 9 36 35 26
Effectiveness 3.778 3.800 1.449 3 14 16 29 28 24 6
30 R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39
performing organisations in the cluster, where the top performing
organisations are those that received a score of ‘7’ on their relative
pro?tability measure. We then calculate the degree of ‘?t’ for each
organisation in these clusters based on the deviations of their
scores on the 18 controls from those of the ideal MCS pro?le as fol-
lows (Drazin & Van de Ven, 1985):
EucD
j
¼
????????????????????
X
s
Dist
2
js
q
ð1Þ
where EucD
j
is the Euclidean distance of the jth organisation from
the ideal MCS pro?le,
Dist
js
¼ ðx
js
À x
is
Þ ð2Þ
and x
js
and x
is
are the score of the jth organisation and the average
score of the top performing organisations in the cluster, respec-
tively, for the sth control (s = 1, . . ., 18). For organisations within
clusters that alternatively identify as having an ineffective MCS,
we use the average control scores of the top performing organisa-
tions from the effective cluster with the closest GP ownership.
Finally, following Ittner and Larcker (2001), we investigate the
relationship between ‘?t’ and organisational performance using
the following model:
Perf ¼ a þ d
1
EucD þR
k
Control
k
þm ð3Þ
where EucD is the Euclidean measure of distance (?t) from Eq. (1),
and Perf and Control are the measure of performance and a vector
of ?ve control variables, respectively, both discussed below. Based
on H1, we expect the sign of d
1
to be negative (d
1
< 0). Since Perf
is measured using a 7-point Likert scale, we use ordinal logistic
regression to estimate the model.
22
For Perf, we use the pro?tability item from the measurement
instrument of Govindarajan and Gupta (1985) that asks the
respondent whether, when compared to similar organisations,
their organisation is more pro?table. Use of a subjective measure
is well established in the literature (King et al., 2010; Miller &
Cardinal, 1994) and has been argued as preferable to archival data
when there is the possibility of differences in accounting presenta-
tion (Powell, 1995), a situation that is likely with PHOs. Miller and
Cardinal (1994) provide further support, arguing ‘‘It may be that
informant data, which individuals typically give under conditions
of promised anonymity for their ?rms, basically re?ect true perfor-
mance, but archival data to a substantial degree re?ect public rela-
tions, tax, and other extraneous considerations that create noise in
the data.’’ (pg. 1661)
For robustness purposes, since this subjective performance
measure may be subject to leniency bias, we also consider a mea-
sure based on Gross Fees for the subset of the respondents who
provide the ?gure (Brownell, 1982).
23
In so doing, we concede that
Gross Fees is not well suited for our purposes as it is a recognised
proxy for size and thereby critically, not a measure of ef?ciency. A
more appropriate objective measure within our context would be a
measure such as the expenses-to-income ratio. Unfortunately, when
we attempted to collect this measure in the pilot survey, we received
an exceedingly low response rate and so did not include it in the full
survey. Given our inability to access our preferred measure we revert
to Gross Fees. Further, we rely on the subjective measure as our pri-
mary measure following the argument advanced by Merchant
(1985) that subjective measures are defensible when it is not possi-
ble to get properly matched objective data.
Finally, the ?ve control variables we include in the model are
emphasis, GP ownership, size, private billings, and life cycle. We
include measures of emphasis and ownership, arguing that these
could potentially be main effects that directly in?uence perfor-
mance. Emphasis (Emphasis) is measured as the mean of the
emphasis scores across the ?ve types of controls in the MCS. As
described, GP ownership (% Ownership) is measured as the propor-
tion of FTE GPs working in the practice who are owners. We
include size (Size) given the possibility that it has a direct relation-
ship with organisational performance (Chenhall, 2007). We include
lifecycle (Lifecycle) under the expectation that organisations with
longer operating histories are more likely to have found opera-
tional ef?ciencies. Finally, we include private billings (Private
Billings) under the expectation that organisations with a higher
proportion of fees from private billings will exhibit better perfor-
mance given the higher pro?t margin per consultation.
5.2. Cluster analysis
We conduct a two-stage cluster analysis classifying the sample
organisations according to ownership and overall effectiveness of
their MCS to identify the empirical ideal pro?le of MCS when GP
ownership varies.
24
Overall effectiveness is measured as the mean
of the effectiveness scores across the ?ve types of controls in the
MCS. We screened the data and the Pearson bivariate correlation
(À0.220, p < 0.01) reveals no threat of multi-collinearity (Hair
et al., 2010). Calculation of Mahalanobis distance revealed eight
cases as potential outliers. Analyses conducted after their exclusion
revealed results to be qualitatively unaffected and hence they were
retained.
We ?rst perform hierarchical clustering using the agglomera-
tive approach and Ward’s method (Everitt, Landau, Leese, & Stahl,
2011), and assess the output using the dendrogram, the agglomer-
ation schedule, the graph of the cluster numbers versus agglomer-
ation coef?cients and the Duda–Hart method. There was support
for a four cluster solution and this was subsequently pro?led via
an ANOVA. We then conduct a non-hierarchical analysis via
K-mean clustering prescribing a four cluster solution using cluster
seeds from the hierarchical cluster analysis (Everitt et al., 2011).
Again, there is support for a four cluster solution from ANOVA,
MANOVA and one-way discriminant analyses.
To provide context, we compare our four-cluster solution with
Speklé’s (2001) theorised optimal MCS archetypes using Multiple
Comparison Procedures (MCP) with Games–Howell tests (Hair
et al., 2010; Toothaker, 1991). The results are presented in
Table 3. A more formal proscriptive system, indicative of a bound-
ary archetype of control, is the theoretical ideal for the two clusters
with the low member ownership, Clusters #3 and #4, relative to
the two with higher GP ownership, Clusters #1 and #2. To frame
our expectations, we appeal to the mean values of the overall effec-
tiveness and ownership measures reported in Panel A to classify
clusters as either ‘effective’ or ‘ineffective’. Based on the wording
of the control questions where, in the main, high scores are indica-
tive of boundary archetypes of control, our expectation is that the
mean overall MCS effectiveness score will be lower for clusters that
have high GP ownership since their managers are expected to rely
less on formal controls and hence likely to consider formal controls
as less effective. On this basis, we ?rst note that the mean value of
the effectiveness measure for Cluster #4 at 1.800 is not only low in
absolute terms, it is also signi?cantly lower than its counterparts
for the other three clusters based on the Games–Howell test.
22
As explained by Borooah (2002), use of the less restrictive multinomial logit
‘‘would mean that the information conveyed by the ordered nature of the data was
being discarded.’’
23
Reassuringly, there is also evidence that objective and subjective measures of
performance are correlated (Dess & Robinson, 1984).
24
The advantage of using two stages is that the hierarchical analysis partitions the
data to determine the acceptable number of clusters and identi?es cluster centres,
while the non-hierarchical analysis ?ne tunes the membership of the clusters (Hair
et al., 2010).
R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39 31
Since Cluster #4 has a relatively low mean ownership measure
(24.8%), comparable with that for Cluster #3 (17.1%), we would
expect its mean effectiveness score to in fact be higher, not lower,
than those for the high ownership clusters, #1 and #2. We there-
fore label Cluster #4 as ‘ineffective’. For the remaining three
clusters, while the mean values for Clusters #2 and #1 at 3.158
and 3.849, respectively, are statistically smaller than the mean
value for Cluster #3 at 5.083, given their higher mean ownership
measures, we argue that this is to be expected. Thus, we label these
three clusters as ‘effective’.
Table 3
Cluster and pro?le analysis results.
Cluster
#2 #1 #3 #4 ANOVA G-H
n = 18 n = 43 n = 38 n = 21 MCP
Effectiveness/Ownership Pro?le Med/High Med/Med High/Low Low/Low
Panel A: Mean values for the primary measures
% Ownership 0.825 0.453 0.171 0.248 1115.848
???
2 > 1 > 4 > 3
Effectiveness – overall (1–7) 3.158 3.849 5.083 1.800 62.208
???
3 > 1 > 2 > 4
Culture 3.830 4.420 5.390 2.240 24.929??? 3 > 1, 2 > 4
Planning 3.350 4.260 5.390 1.760 31.707??? 3 > 1, 2 > 4
Cybernetics 2.720 3.410 5.230 1.000 25.259??? 3 > 1, 2 > 4
Rewards/compensation 2.110 2.700 3.680 1.670 25.259??? 3 > 4
Administrative controls 3.780 4.470 5.710 2.330 24.840??? 3 > 1, 2, 4:1 > 4
Performance (relative pro?tability) (1–7) 4.440 4.700 4.380 4.050 0.674 1, 2, 3, 4
Emphasis – overall (1–7) 3.144 4.008 5.207 2.162 47.214
???
3 > 1, 2 > 4
Culture 3.830 4.490 5.740 3.050 17.228
???
3 > 1, 2, 4; 1 > 4
Planning 3.500 4.440 5.550 2.000 23.576
???
3 > 1, 2 > 4
Cybernetics 2.780 3.630 5.420 1.240 26.379
???
3 > 1, 2 > 4
Rewards/compensation 1.940 2.670 3.610 1.710 5.002
???
3 > 2, 4
Administrative controls 3.670 4.810 5.760 2.810 20.539
???
3 > 1 > 2, 4
Panel B: Mean values for the 18 MC practice variables by MC type
Culture (1–7)
Socialise + 3.819 4.727 5.182 3.393 8.631
???
1, 3 > 2, 4
Code of conduct – 4.167 5.132 5.386 3.508 4.796
???
1, 3 > 4
Vision and mission – 4.741 4.476 5.307 3.239 6.361
???
3 > 1 > 4
Dress Code – 1.889 3.081 2.860 1.714 4.124
???
1 > 2; 1 > 3 > 4
Recruit – 4.847 5.485 5.743 4.786 6.106
???
3, 1 > 2;3 > 4
Selection + 4.811 4.419 4.382 3.191 4.966
???
1, 2, 3 > 4
Planning (1–7)
Long range planning – 4.263 4.191 4.426 1.838 11.862
???
1, 2, 3 > 4
Short range planning – 4.333 4.380 5.155 2.333 14.082
???
1, 2, 3 > 4
Cybernetics (1–7)
Budgets – 4.482 4.329 5.654 2.698 12.324
???
3 > 1 > 4; 2 > 4
Boundary – 2.370 2.667 2.912 1.397 3.941
???
1, 3 > 4
Non-?nancial – 4.185 4.250 5.105 2.715 7.513
???
1, 2, 3 > 4
Rewards/compensation (1–7)
Rewards/compensation – 2.185 3.333 3.360 1.746 7.693
???
1 > 2, 4; 3 > 4
Administrative controls (1–7)
Rules – 4.417 4.901 5.461 3.321 10.669
???
1, 3 > 4
Position – 3.685 4.212 3.772 3.032 1.491 1, 2, 3, 4
Organisational committees – 3.944 4.042 4.437 2.067 11.609
???
1, 2, 3 > 4
Chronic disease management – 5.222 4.744 5.553 3.810 7.042
???
2, 3 > 4; 3 > 1
Policies and procedures – 3.167 3.326 3.645 2.262 4.639
???
1, 3 > 4
Meetings – 5.796 5.876 6.605 5.556 3.377
??
3 > 1, 4
Panel C: Performance measures and contextual variables
Performance
Relative pro?tability (1–7) 4.440 4.700 4.380 4.050 0.674 –
More competitive (1–7) 4.610 4.720 5.180 4.000 2.259
?
–
Greater market share (1–7) 5.280 5.000 5.370 4.290 2.144
?
–
Growing faster (1–7) 4.940 5.070 5.570 4.050 3.911
???
3 > 4
More innovative (1–7) 5.060 5.260 6.050 4.810 3.329
???
3 > 1
Larger in size (1–7) 4.390 5.090 5.710 4.290 3.143
???
–
Gross Fees ($ millions) 2.188 2.342 2.284 2.084 0.209 –
Learning (1–7) 6.224 6.064 6.540 5.012 8.676
???
3 > 1, 4
Accreditation (1–5) 4.160 3.986 4.186 4.148 0.374 –
Patient satisfaction (1–7) 5.130 5.059 5.421 5.397 1.534 –
Contextual variables
Size (FTE) 14.014 15.497 18.072 13.270 2.103 –
Lifecycle 45.056 33.700 24.792 29.524 2.942
??
2 > 3
Private billings 39.694 46.333 41.500 40.429 0.514 –
Strategy 5.390 5.210 5.290 3.950 7.395
???
1, 2, 3 > 4
Structure 5.046 4.950 5.614 3.675 12.077
???
1, 2, 3 > 4; 3 > 1
PEU 1 3.093 3.023 3.152 2.651 0.681 –
PEU 2 4.493 4.390 4.962 4.524 1.145 –
32 R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39
Turning to the comparison, given the nature of our sample
PHOs, we view it as unlikely that they will each activate all 18
controls within the ?ve control types; rather, each will likely
select the subset of controls best suited to its situation. As such,
we argue that it is the overall emphasis measure that will best
re?ect MCS choice (boundary or exploratory archetype) and we
turn our primary attention to the results for this measure in
Panel A. Here, the mean value is 5.207 for Cluster #3, 4.008 for
Cluster #1, 3.144 for Cluster #2, and 2.162 for Cluster #4. The
F-statistic for the difference in mean values is 47.214 (p < 0.001).
Importantly, consistent with Speklé’s (2001) theorised optimal
MCS archetypes, the Games–Howell test reveals the mean value
for Cluster #3 to be signi?cantly higher than for the two other
‘effective’ clusters (Clusters #1 and #2) that have higher levels
of member-ownership. Further, all three ‘effective’ clusters place
signi?cantly greater emphasis on formal proscriptive controls
than the low effectiveness cluster, Cluster #4. As additional
support, the results for the 18 individual control variables across
the ?ve control types presented in Panel B are largely consistent
with theoretical ideal pro?les for the three effective clusters
while the mean values for Cluster #4 are almost universally in
contrast.
In sum, we view the ?ndings for the ?rst three clusters as
providing reassurance regarding the ability of our cluster analysis
to identify the ideal empirical MCS pro?les on which to base our
deviation measure. Importantly, the existence of Cluster #4 also
indicates that our sample comprises PHOs that exhibit a signi?cant
degree of mis?t with their theoretically ideal MCS pro?le.
5.3. Results for tests of H1
Table 4 presents results for our test of H1. Panel A presents
descriptive statistics for EucD and related univariate results.
Following, we formally test H1 using Eq. (3), considering four vari-
ants of the model. The ?rst, Model 1, only includes EucD while
Model 2 additionally includes Emphasis and % Ownership. Model
3 extends the model to include Size and Model 4 further includes
Lifecycle and Private Billings. All analyses are conducted using ordi-
nal logistic regression.
As revealed in the ?rst row of Panel A, for the effective clusters,
there are four top performing organisations in Cluster 2, two in
Cluster 1, and ?ve in Cluster 3. These organisations are used to
de?ne their ideal MCS pro?les. For Cluster 4, the ineffective cluster,
the top performing organisations in Cluster 3 are used to de?ne the
ideal MCS since it has the closest ownership level.
The next set of rows in Panel A present descriptive statistics
for EucD. As revealed, this measure exhibits considerable cross-
sectional variation, both for the pooled data and within each
cluster. The F-statistic for the difference in mean values (not
tabulated) is 16.705 (p < 0.001). Of note, based on the post hoc
tests, the mean value for the ineffective cluster (Cluster #4) is
signi?cantly different from the mean value of the effective cluster
that also has low ownership, Cluster #3 (p < 0.001). In conjunction,
the minimumvalue of EucD is noticeably higher for Cluster #4 than
for any of the three effective clusters. Finally, the last row of Panel
A presents the pairwise correlations between Perf and EucD. As
implied by H1, the correlations are uniformly negative and signif-
icant at the 5% level or better for the pooled sample and the three
effective clusters. Alternatively, while negative, the correlation for
the ineffective cluster, Cluster 4, is not signi?cant at conventional
levels (although it is signi?cant at the 10% level for a one-tailed
test). Thus, overall, these univariate results provide preliminary
support for H1.
More formally, turning to the ordinal logistic regression results
for Eq. (3) presented in Panel B, of central interest the coef?cient
on EucD is negative as predicted and signi?cant at better than the
1% level across all four models. Thus, consistent with H1, the
results suggest that greater mis?t is associated with reduced per-
formance. Given consistent ?ndings for EucD and all control mea-
sures, for parsimony we only detail the results for the complete
model, Model 4. To begin, the chi-square for testing the propor-
tional odds assumption is insigni?cant at conventional levels
(v
2
= 37.416; p = 0.165), thereby indicating that the assumption
the model has parallel slopes is met and use of an ordered model
is appropriate (Borooah, 2002). Next, the null hypothesis that the
coef?cients are simultaneously equal to zero is rejected at less
than the one percent level (v
2
= 20.429; p = 0.002). Of greatest
interest, the coef?cient on EucD is À0.269 (p = 0.001). Lastly, for
the remaining measures, only the Size variable is statistically sig-
ni?cant. Its coef?cient is 2.698 (p = 0.004). The coef?cients on the
remaining control variables are insigni?cant at conventional
levels.
25
Finally, notwithstanding its limitations, for sensitivity purposes
we re-ran Eq. (3) after replacing the dependent variable with an
objective measure of performance based on ‘Gross Fees’ for the
84 sample organisations that report this ?gure. Since ‘Gross Fees’
is a recognised proxy for size and thereby not directly a proxy for
the underlying construct of interest, relative pro?tability, we ini-
tially regress the natural log of ‘Gross Fees’ (lnGF) on Size and then
use the residual as the dependent variable. The results, run using
OLS, are presented in Table 5. Model A includes only EucD, Model
B extends the model to include Emphasis and % Ownership, and
Model C adds Lifecycle and Private Billings. Again, the results pro-
vide consistent support for H1. Focusing on Model C, the coef?cient
on EucD at À0.007 is negative and signi?cant (p = 0.026). Thus,
results and conclusions appear robust to the use of an objective
performance measure based on ‘Gross Fees’.
26
5.4. Alternative performance measures
Within our setting the relevant notion of performance is ?nan-
cial performance relative to peer organisations. Notwithstanding,
we also included ?ve questions that related to non-?nancial
dimensions of the PHO’s performance, asking whether compared
with similar practices, the PHO is more competitive, has greater
market share, is growing faster, is more innovative and is larger.
We also asked for the accreditation score, patient satisfaction,
and the importance of learning. To gain a sense of whether the
degree of ‘?t’ impacts these dimensions of performance, we
re-ran Eq. (3) alternatively with each of the measures as the depen-
dent variable using ordinal regression.
27
The results, presented in Table 6, are largely consistent with
expectations. We ?nd negative and signi?cant coef?cients on EucD
for the models based on competitiveness (À0.161; p = 0.027),
market share (À0.132; p = 0.071), growth (À0.276; p = 0.001),
innovation (À0.189; p = 0.011), size (À0.186; p = 0.014), and
learning (À0.147; p = 0.050). Thus, organisations with better ‘?t’
indicate that they view themselves as more competitive, having a
greater market share, growing faster, being more innovative, larger,
and fostering learning. Alternatively, we ?nd the coef?cient in the
25
To consider the potential in?uence of outliers, we trim the data at the 2.5% and
97.5% level for Dist and re-run Model 4. Here, the coef?cient on EucD 4 is À0.272
(p < 0.001). If we trim at the 5% and 95% levels, the coef?cient on EucD is À0.186
(p = 0.026). To provide further assurance, we set AbsD
j
=
P
s
|Dist
js
| and re-ran Model 4,
?nding a coef?cient on AbsD of À0.075 (p < 0.001).
26
Results are robust to the inclusion of organisation and cluster ?xed effects, and to
trimming at the 2.5% and 97.5% level for Dist.
27
For competitiveness, market share, growth, innovation, and size, the dependent
variable is the response to the relevant single item, for learning and (patient
satisfaction, it is the average summated score across the underlying questions
rounded to the next highest integer value, and for accreditation, it is the score
obtained.
R. King, P. Clarkson/ Accounting, Organizations and Society 45 (2015) 24–39 33
models based on the accreditation score and patient satisfaction to
be insigni?cant at conventional levels.
28
5.5. Robustness tests
As a ?nal step, to explore the sensitivity of our results and con-
clusions to several of our design and econometric decisions, we
undertook a number of additional analyses, ?nding in each
instance, the coef?cient on EucD remains negative and signi?cant
as predicted by H1. First, while appealing to the top performing
organisations to identify the ideal empirical MCS pro?le may ini-
tially give the appearance of introducing a bias towards H1, a priori
we do not believe that this is necessarily the case. We argue that
simply by having a lower relative performance measure, it does
not necessarily mean that the organisation has placed more or less
weight on any particular control, or in aggregate across the 18 con-
trols. This is, in fact, the empirical question being addressed in the
study – do organisations with lower performance exhibit greater
distance measures? Notwithstanding, to provide a degree of assur-
ance that our results are not being driven by use of the top perform-
ing ?rms, we repeated all analyses reported in Tables 4–6 using an
alternative measure of EucD calculated using the average score for
each of the 18 controls across all organisations within a cluster.
Here, we ?nd the results to be qualitatively similar. To illustrate,
the coef?cient on the recalculated EucD for full model in the pri-
mary analysis (Model 4) is again negative and signi?cant (À0.205;
Table 4
Results for the relation between relative performance and ‘Fit’.
Pooled Cluster 2 Cluster 1 Cluster 3 Cluster 4
(n = 120) (n = 18) (n = 43) (n = 38) (n = 21)
Panel A: Descriptive Statistics
# top performers n/a 4 2 5 n/a
EucD
Mean 9.177 8.114 10.331 7.388 10.961
Median 9.114 6.809 10.434 7.315 10.741
Std dev 2.272 3.357 2.096 1.599 2.692
Minimum 3.866 3.856 4.113 4.966 6.146
Maximum 15.258 15.258 14.246 10.863 14.686
Correlation (Perf, EucD) À0.345 À0.693 À0.368 À0.355 À0.295
(p < 0.001) (p < 0.001) (p = 0.015) (p = 0.029) (p = 0.194)
Variable Model 1 Model 2 Model 3 Model 4
Panel B: Regression results, full sample (n = 120)
Intercept 1 À4.873 À4.453 À2.363 À2.373
(