sally widener aos 2007

Accounting, Organizations and Society 32 (2007) 757–788

www.elsevier.com/locate/aos

0361-3682/$ - see front matter ! 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.aos.2007.01.001

An empirical analysis of the levers of control framework

Sally K. Widener ¤

Jones Graduate School of Management, Rice University, Room 331 – MS 531, 6100 Main Street,Houston, TX 77005-1892, United States

Abstract

The purpose of this paper is to use the levers of control framework to explore the antecedents of control systems –various facets of strategy that drive the use of controls; to explore the relations among control systems; and to explorethe costs and beneWts of control systems – costs in terms of consumption of a constrained resource (i.e., managementattention) and beneWts (i.e., learning). Using data from a survey of 122 Chief Financial OYcers, this study tests a struc-tural equation model that relates strategic risk and uncertainty to control systems (i.e., beliefs, boundary, diagnostic, andinteractive control systems), which in turn are hypothesized to aVect learning and attention, and ultimately Wrm perfor-mance. The evidence suggests that there are multiple inter-dependent and complementary relations among the controlsystems. I Wnd that strategic risk and uncertainty drive both the importance and use of performance measures in diag-nostic or interactive roles. Moreover, it appears that in certain strategic conditions information processing needs aresuch that Wrms use performance measures both interactively and diagnostically. Finally, I conclude that although thereis a cost of control, there is a positive eVect on Wrm performance.! 2007 Elsevier Ltd. All rights reserved.

Introduction

The purpose of the management control system(MCS) is to provide information useful in deci-sion-making, planning, and evaluation (Merchant& Otley, 2006). While the management accountingliterature is replete with studies that investigatecontrol systems, many focus on only one controlsuch as the use of performance measures (Ittner &Larcker, 1998). However, it is well-recognized that

the MCS is comprised of multiple control systemsthat work together (Otley, 1980). Simons (2000), inhis levers of control (LOC) framework, posits thatfour control systems – beliefs (e.g., core values),boundary (e.g., behavioral constraints), diagnostic(e.g., monitoring), and interactive (e.g., forward-looking, management involvement) – work togetherto beneWt a Wrm. The LOC framework asserts thatstrategic uncertainty and risk drive the choice anduse of control systems, which in turn, impact theorganization through organizational learning andthe eYcient use of management attention (Simons,2000). The theoretical framework is illustrated in

* Tel.: +1 713 348 3596; fax: +1 713 348 6296.E-mail address: [email protected]

758 S.K. Widener / Accounting, Organizations and Society 32 (2007) 757–788

Fig. 1. The purpose of this paper is to use the LOCframework to investigate the antecedents of con-trol systems (i.e., strategic uncertainty and risk);the associations among the control systems; andthe costs and beneWts of control systems (manage-ment attention, learning, and Wrm performance).

Using data from a survey of 122 Chief FinancialOYcers (CFOs), I perform a three-stage analysis.First, I estimate a trimmed structural equationmodel (SEM) based on Fig. 1, which provides evi-dence on the holistic LOC framework. Second, Iuse coeYcients from the trimmed SEM to provideevidence on three sets of hypotheses: (1) the rela-tions among the control systems, (2) the relationsbetween both strategic risk and uncertainty andeach of the control systems, and (3) the relationsbetween each of the control systems and outcomes(i.e., attention, learning, and performance). Finally,since the existence of a well-Wtting SEM does notensure that it is the only appropriate model (Kline,1998), I generate six alternative models for com-parison against the base model.

This study makes several contributions to theliterature. First, it is generally well-accepted thatcontrol systems are inter-dependent (Milgrom &Roberts, 1995); however, it is unclear whetherthey are complements or substitutes. This studyWnds that when Wrms emphasize the beliefs system,they also emphasize each of the three other con-trol systems. In addition, the use of performancemeasures in the interactive system is associated

with the use of performance measures in the diag-nostic system and emphasis on the boundarysystem. The evidence suggests that the inter-dependencies are complementary. Thus, this studyprovides empirical evidence on the relationsamong the control systems in the LOC frameworkand contributes to a small but growing body ofwork that investigates relations among controlsystems (e.g., Anderson & Dekker, 2005; Kennedy& Widener, 2006).

Second, this study investigates both costs andbeneWts of control systems. An assumption in theliterature is that Wrms implement controls onlywhen the beneWts received outweigh the costs.However, little evidence exists to support thisassumption (e.g., for a review of performance mea-surement (PM) literature see Ittner & Larcker,1998). A limitation is that this research oftenfocuses on an aggregate measure of Wrm perfor-mance without delineating speciWc costs and bene-Wts. A recent study extends the literature andprovides evidence that a strategic PM system isassociated with a speciWc beneWt (i.e., decreasedrole stress) (Burney & Widener, 2007); however,the cost of controls is still largely ignored. I Wndthat control systems are associated with both abeneWt (organizational learning) and a cost (con-sumption of management attention); but, overall,have a positive eVect on Wrm performance.

Third, not withstanding a line of research thathas investigated the alignment between strategy

Fig. 1. Theoretical model.

Strategic Elements Control Systems Costs and Benefits Performance

Attention

Beliefs System

Learning

Performance

Boundary System

Diagnostic Controls

Interactive Controls

Strategic Uncertainties

Strategic Risks

S.K. Widener / Accounting, Organizations and Society 32 (2007) 757–788 759

and a Wrm’s MCS (Ittner & Larcker, 1997; Lang-Weld-Smith, 1997), LangWeld-Smith (1997) con-cludes that knowledge is limited since studiesonly investigate single facets of a multifacetedconstruct. Moreover, Chenhall (2003) argues thataccounting studies may suVer from outdatedstrategy constructs. This has spurred studies toincorporate additional strategic facets such asstrategic resources (see e.g., Henri, 2006) andcompetitive advantage (Widener, Shackell, &Demers, 2006). This study investigates two ele-ments of strategy that Simons’ (2000) argues playa central role in the LOC theory – strategicuncertainties and strategic risk, and Wnds thatboth are associated with the use of control sys-tems.

Finally, business are competing with complex,rapidly changing, and knowledge-intensive busi-ness models driving the need to better understandthe role of PM systems and how they can bettermeet managerial needs. A control system can func-tion in diVerent roles (i.e., either interactively ordiagnostically). This study sheds insights onthe role of the PM system and Wnds generallythat internal (external) strategic factors are associ-ated with diagnostic (interactive) controls. How-ever, under certain circumstances, performancemeasures are used in both roles. These results addto a growing body of literature that investigatehow the role of control systems diVers (e.g., Abern-ethy & Brownell, 1999; Bisbe & Otley, 2004; Henri,2006).

This study is organized as follows. Section“Theory development and hypotheses: control sys-tems” provides an overview of the control systemsthat comprise the levers of control framework anddevelops hypotheses for the inter-dependenciesamong the control systems. Section “Theory devel-opment and hypotheses: drivers and outcomes ofcontrol systems” develops the hypotheses for thedrivers (strategic risk and uncertainty) and out-comes (learning, attention, and performance) ofthe control systems. Section “Methods” discussesthe research method and measurement of the vari-ables. The analyses and results are presented inSection “Results”. Finally, conclusions, limita-tions, and extensions are discussed in Section“Conclusions”.

Theory development and hypotheses: control systems

Overview of control systems

The LOC framework consists of four controlsystems: beliefs, boundary, diagnostic, and interac-tive. The beliefs system is “the explicit set of orga-nizational deWnitions that senior managerscommunicate formally and reinforce systemati-cally to provide basic values, purpose, and direc-tion for the organization” (Simons, 1995, p. 34).That is, a beliefs system communicates core valuesin order to inspire and motivate employees tosearch, explore, create, and expend eVort engagingin appropriate actions. However, in dynamic envi-ronments there must be some restraint placed onemployees to stop them from engaging in high-riskbehaviors. This restraint is the boundary system,which acts in opposition to the beliefs system. Aboundary system “delineates the acceptabledomain of strategic activity for organizational par-ticipants” (Simons, 1995, p. 39). The boundary sys-tem communicates the actions that employeesshould avoid. Its purpose is to allow employeesfreedom to innovate and achieve within certainpre-deWned areas. The boundary and beliefs sys-tems are similar in that they both are intended tomotivate employees to search for new opportuni-ties; however, the boundary system does so in anegative way through the constraint of behaviorwhile the beliefs system does so in a positive waythrough inspiration (Simons, 1995). Often, Wrmscommunicate beliefs through a mission or visionstatement and boundaries through a code of con-duct. Appendix “Examples of beliefs and bound-ary control systems” provides examples for twoWrms in this study.

A Wrm’s critical success factors are embedded inits diagnostic system and communicated to itsemployees. The diagnostic system is intended tomotivate employees to perform and align theirbehavior with organizational objectives. It reportsinformation on the critical success factors whichallows managers to focus their attention on theunderlying organizational drivers that must bemonitored in order for the Wrm to realize itsintended strategy. The diagnostic control system


also enables managers to benchmark against tar-gets. Similar to the boundary system, the diagnos-tic system acts as a constraint on employeebehavior (Simons, 2000).

While the diagnostic system allows managersto manage results on an exception basis, aninteractive system is forward-looking and charac-terized by active and frequent dialogue among topmanagers. The interactive system is intended tohelp the Wrm search for new ways to strategicallyposition itself in a dynamic marketplace. Topmanagers choose which control system (e.g., PM,brand management, budget process) they wantto use in an interactive manner. This study inves-tigates the PM system since it is used extensivelyin practice and because the results of this studycan contribute to a long stream of research on per-formance measures (see e.g., Ittner & Larcker,1998).

It is well-accepted in the literature that controlsystems are inter-dependent (e.g., Merchant &Otley, 2006; Otley, 1999). Otley (1999) argues thatresearchers must approach research studies with aholistic perspective of control systems since diVer-ent Wrms may use diVerent conWgurations of con-trol systems. Simons (2000, p. 301) posits that inthe LOC framework all four control systems,working together, are necessary to provide aneVective control environment. He argues that anintegrated control environment eVectively facili-tates the quest for competitive sustainability andstrategy implementation and states,

Control of business strategy is achieved byintegrating the four levers of beliefs systems,boundary systems, diagnostic control sys-tems, and interactive control systems. Thepower of these levers in implementing strat-egy does not lie in how each is used alone,but rather in how the forces create a dynamictension . . .

This notion of dynamic tension is consistentwith Milgrom and Roberts (1995) who analyticallydemonstrate that control features can be comple-mentary, that is, increasing the emphasis on onecontrol component increases the beneWts receivedfrom increasing the emphasis on other controlcomponents.

Simons (1995) suggests that the four levers cre-ate tension in that two of the levers – the beliefsand interactive control system – create positiveenergy, while the remaining two levers createnegative energy. Henri (2006) empirically teststhis proposition. He posits that managers useperformance measures in both a diagnostic andinteractive role and that doing so results in a desir-able state of dynamic tension that will enhanceorganizational capabilities. Henri (2006) Wndssome evidence that together the two levers of con-trol result in dynamic tension that is positivelyassociated with performance. Similar to Henri(2006), in the section that follows I posit that thediagnostic and interactive uses of the performancemeasurement system are complementary. How-ever, Henri (2006) does not investigate whether orhow the levers inXuence one another.1 That is, hedoes not investigate the relations among the leversof control.2 In the following section, I hypothesizethe strongest relations supported in the underlyingliterature and leave other relations for explorationin Section “Comparison to alternative models –Step 3”.3

1 Tuomela (2005) uses a case study to investigate how theinteractive or diagnostic use of the performance measurementsystem aVects the beliefs and boundary system.

2 I argue that the interactive use of performance measuresinXuences the diagnostic use of performance measures, whileHenri (2006) models an independent-variables interaction (Luft& Shields, 2003). That is, Henri (2006) argues that how much thediagnostic (or interactive) use of performance measures aVectscapabilities is conditional on the other use of performance mea-sures.

3 Simons (2000) asserts that the control components in theLOC framework are integrated to form an eVective control sys-tem. This can lead to various interpretations of inter-dependen-cies including the most complete model in which all fourcontrols inXuence each of the other controls. Simons (2000,p. 305) also presents other perspectives including that the beliefsand boundary, boundary and diagnostic, diagnostic and inter-active, and interactive and belief systems have bi-directionalrelations, that is, they inXuence each other. However, Simons(2000, p. 32) also contains a diagram in which actions Xow fromthe mission, thus indicating that the mission inXuences all else.Due to the lack of extant evidence regarding the nature of theassociations of the relations among the control components, Ichoose to hypothesize only those links that can be strongly ar-gued. However, I test several other likely associations amongcontrol components when testing speciWcations of alternativemodels in Section “Comparison to alternative models – Step 3”.


Hypotheses among control systems

Organizational theory suggests that the beliefssystem is critical to high-performing organizations(Calfee, 1993; Analoui & Karami, 2002). Pearce(1982, p. 24) states

It [a mission statement] thus provides man-agement with a unity of direction that tran-scends individual, parochial, and transitoryneeds. It promotes a sense of shared expecta-tions among all levels and generations ofemployees. It consolidates values over timeand across individuals and interest groups.

In this section I Wrst posit that the more a Wrmemphasizes the beliefs system to communicateintended strategy and inspire employees to searchfor opportunities, the more the Wrm will emphasizethe other three components of the LOC frame-work – the boundary and diagnostic systems toreign in and monitor employee behavior and theinteractive system to watch for threats and oppor-tunities thus allowing for emergent strategy.

At a high level of abstraction there are two rea-sons why the beliefs system is likely related to eachof the other control levers. First, it is important thatall four levers are “balanced” in order to manage“the inherent tensions between (1) unlimited oppor-tunity and limited attention, (2) intended and emer-gent strategy, and (3) self-interest and the desire tocontribute” (Simons, 1995, p. 28). Second, there isconsiderable evidence that the beliefs system maynot be eVective unless it is strongly supported withalternative mechanisms. For example, Pearce andDavid (1987, p. 109) note that while a Wrm’s mis-sion statement provides a foundation for the orga-nization, it is simply a “starting point”. Moreover,Bart, Bontis, and Taggar (2001) Wnd that whilethere is a positive relation between Wrm perfor-mance and the emphasis placed on the missionstatement, that internal structure, policies, and pro-cedures are related to employee behaviors whichare more strongly related to Wrm performance. Thissuggests that the emphasis placed on the boundary,diagnostic, and interactive systems may complementthe emphasis placed on the beliefs system.

There are also speciWc reasons at the controllevel that provide a foundation for expecting

inter-dependencies. The beliefs and boundarysystems both inform organizational membersabout opportunities to explore, create, and inno-vate. While the beliefs system motivates employ-ees, the boundary system sets the limits ofexploration. Thus the boundary system Wts insidethe beliefs system, that is, the boundary systemconstrains the exploration motivated by thebeliefs system (Simons, 1995). Using case studies,Simons (1994) illustrates the interplay betweenthe beliefs and boundary systems as one in whichmanagers implement formal boundary systems inorder to counterbalance the inspirational andmotivational message communicated in thebeliefs system. The beliefs system is also embod-ied in the diagnostic control system since the lat-ter captures the critical success factors associatedwith the core values espoused in the beliefs sys-tem. The more motivated employees are to worktowards achieving the Wrm’s ideals and core val-ues, the more emphasis management will place onmeasuring the appropriate critical success factorsin order to ensure that employees’ actions arealigned with the Wrm’s strategy. Finally, once theWrm’s strategy has been clariWed and communi-cated through the mission and vision containedin its beliefs system, top managers will realizewhere potential threats and opportunities mayreside and can implement an interactive system.Thus I expect that the emphasis Wrms place onthe beliefs system is inter-dependent with theemphasis they place on each of the boundary,diagnostic, and interactive systems. I formallyhypothesize:

H1a: The emphasis Wrms place on the beliefssystem is positively associated with theemphasis they place on a boundary system.

H1b: The emphasis Wrms place on the beliefssystem is positively associated with theemphasis they place on the use of performancemeasures in a diagnostic control system.

H1c: The emphasis Wrms place on the beliefssystem is positively associated with theemphasis they place on the use of perfor-mance measures in an interactive control sys-tem.


The beliefs and interactive systems are the posi-tive energy levers that inspire employees to seek,explore, and create. The relation between these twois captured above in H1c. Conversely, the bound-ary and diagnostic levers work in tandem with oneanother as negative forces to reign in behavior, toconstrain the space that employees have to explore,and to ensure compliance with organizationalobjectives. These levers counterbalance the positiveforces of the beliefs and interactive control system(Simons, 1995). A boundary system is considered anecessary component of most organization’s struc-ture; it is required legally both by the stockexchanges and through the enactment of the Sar-banes-Oxley Act; considered in environmentalactions; and called for by shareholder and stake-holder groups (Paine, Deshpande, Margolis, &Bettcher, 2005). Thus I posit that Wrms will imple-ment a boundary system; as they rely more on theboundary system they will also rely more on thediagnostic system in order to maintain the neces-sary balance needed for the control structure. I for-mally hypothesize that:

H1d: The emphasis the Wrm places on theboundary system is positively associated withthe emphasis the Wrm places on the diagnos-tic system.

Interactive controls provide top managers witha mechanism to learn of new strategic opportuni-ties. As strategy emerges through the interactivecontrol structure, objectives and critical successfactors must be redeWned and conveyed through-out the organization. Thus, interactive controlsbecome eVective through a support structure(Chenhall & Morris, 1995). Within the LOC frame-work, I posit that the interactive use of perfor-mance measures inXuences the diagnostic use ofperformance measures since the latter provides thenecessary structure that enables the interactivecontrol system to be eVective. As the organizationadjusts to the strategy that emerges through theinteractive system, the diagnostic PM system mustalso adjust in order to reXect the Wrm’s new strate-gic position and critical success factors.

Simons (1994) Wnds that the four managers inhis case study that attempted to turn around theirWrm’s strategy use the diagnostic system to com-

municate the new critical success factors. That is,the diagnostic system is used to communicate thestrategy that emerges through the interactive sys-tem. In related empirical work, Chenhall and Mor-ris (1995) Wnd that organic decision processes,similar in nature to interactive controls, are moreeVective when coupled with a formal managementcontrol system (see also Henri, 2006). Simons(2000, p. 305) summarizes the relation of the diag-nostic and interactive systems when he says, “theinformation and learning generated by interactivesystems can be embedded in the strategies andgoals that are monitored by diagnostic control sys-tems”. Thus, I argue that Wrms will use both aninteractive and diagnostic system, and the moretop managers rely on the interactive control sys-tem, the more they will rely on the diagnostic con-trol system to provide the structure necessary toenable the interactive system to be eVective. Thissupports the following hypothesis:

H1e: The emphasis Wrms place on the use ofperformance measures in an interactive con-trol system is positively associated with theemphasis they place on the use of perfor-mance measures in a diagnostic control sys-tem.

Theory development and hypotheses: drivers and outcomes of control systems

This section is comprised of two parts. The dis-cussion is guided by Fig. 1. The Wrst sub-sectiondiscusses the relation between strategic factors andeach of the beliefs and boundary systems; and thenturns to a discussion of the relation with each ofthe diagnostic and interactive systems. The secondsub-section discusses the costs and beneWts associ-ated with the use of each of the control systems.

Strategic factors

A feature of the LOC framework is that manag-ers must decide how much emphasis they will placeon each of the four types of control systems (Mer-chant & Otley, 2006). The contingency frameworkholds that environmental variables are important


to consider in the design of the management con-trol system (Chenhall, 2003). In the LOC frame-work one type of environmental variable isstrategic uncertainty, which is deWned as “theemerging threats and opportunities that couldinvalidate the assumptions upon which the currentbusiness strategy is based” (Simons, 2000, p. 215).That is, they are changes in competitive dynamicsor internal competencies that top managers moni-tor. Uncertainty implies that there is a gap betweenthe information known and desired (Galbraith,1973). Thus the more uncertainty, the more moni-toring is necessary to reduce the information gap(Simons, 2000).

A second type of environmental variable in theLOC is strategic risk, which is deWned as “an unex-pected event or set of conditions that signiWcantlyreduces the ability of managers to implement theirintended business strategy” (Simons, 2000, p. 255).Risk is a source of potential harm to the organiza-tion. Strategic risk stems from both operations (i.e.,a malfunction in core internal operating proce-dures) and external factors (i.e., a mass disruptionin customer base or intense rivalry among compet-itors). Similar to strategic uncertainty, strategicrisk requires increased information processing toassess the likelihood of risk and the magnitude ofany resultant harm.

Firms use both the beliefs and boundary sys-tems to manage risk since they help ensure thealignment of employee behavior, which minimizesthe possibility that the organization can beharmed. Empirical research suggests that thesecontrol systems are also beneWcial in managingstrategic uncertainty. Merchant (1990) Wnds thatproWt center managers are more likely to manipu-late earnings when conditions are uncertain imply-ing that the likelihood of engaging in unethical actsis higher in environments characterized by uncer-tainty. Strong boundary and beliefs systems areintended to counteract undesirable behavior andminimize the negative behavioral eVects associatedwith strategic uncertainty. Moreover, Simons’(1994) case study shows that when organizationsface strategic change (and thus heightened uncer-tainty) the beliefs system is important for commu-nicating the vision and core values while theboundary system delineates the appropriate area

for pursuing opportunities. Related researchgrounded in contingency theory Wnds that Wrmsuse sophisticated control systems when facingenvironmental uncertainty (e.g., Chenhall, 2003),market competition (Bromwich, 1990; Khand-walla, 1972; Mia & Clarke, 1999), and environmen-tal hostility (Otley, 1978). Chenhall (2003, p. 138)summarizes this stream of literature by saying“hostile and turbulent conditions appear, in themain, to be best served by a reliance on formalcontrols . . .” Thus, I propose that:

H2a: The extent to which Wrms face strategicrisks and uncertainties is positively associ-ated with the emphasis they place on a beliefssystem.

H2b: The extent to which Wrms face strategicrisks and uncertainties is positively associ-ated with the emphasis they place on aboundary system.

In the analyses that follow, I investigate threetypes of strategic uncertainty: operating uncer-tainty (e.g., scale eVects, internal product innova-tion), competitive uncertainty (e.g., new industryentrants), and technological uncertainty (e.g., newtechnology); and three types of strategic risks thatSimons’ (2000) identiWes: operating risk (e.g.,safety of operations), asset impairment risk (e.g.,accounts receivable turnover), and competitive risk(e.g., marketplace factors).4

Firms use diagnostic control systems to manageboth strategic uncertainty and risk (Galbraith,1973; Simons, 2000). In order to reduce the infor-mation processing burden for top managers, deci-sion rights can be delegated throughout theorganization (Galbraith, 1973). Performance mea-sures embedded in a diagnostic control systemthen provide direction to these empoweredemployees, which help ensure that their behaviorsare aligned with organizational goals. Further-more, diagnostic controls facilitate informationprocessing through the provision of exceptionreporting. In the face of uncertainty and risk, theneed to process information can become signiWcant.

4 Note that Porter’s (1980) “Wve forces” are encompassed inSimons’ (2000) treatment of competitive market dynamics.


Galbraith (1973, p. 15) suggests that once Wrmshave implemented goal setting their next step is toeither “reduce the need for information process-ing” or “increase the capacity to process the infor-mation”. One way to increase informationprocessing capacity is to engage in a vertical infor-mation system (Galbraith, 1973), similar to aninteractive control system. Information is spreadvertically throughout the organization to operat-ing managers and lower-level employees, whichspurs action, attention, and dialogue. Simons(1995, p. 102) characterizes the interactive controlsystem as follows:

By choosing to use a control system interac-tively, top managers signal their preferencesfor search, ratify important decisions, andmaintain and activate surveillance through-out the organization. All subordinate manag-ers will engage in the interactive dialogue tothe extent demanded by their position. Thus,the system may remain interactive downthree or four levels in the organization . . .

In order to use a performance measure diagnos-tically it must be possible to set a goal, measureoutputs, and compute variances (Galbraith, 1973;Simons, 2000). This implies that the properties ofthe measures must be stable, with low noise andvariation. Managers can assess operations, stan-dardize processes, and implement procedures toensure safety and quality (Simons, 2000). Measuresof internal strategic factors such as safety, quality,internal innovation, and cost of inputs lend them-selves to being captured in a more routine type ofPM system which allows managers to use thesemeasures diagnostically, look for exceptions, andgather feedback information. In addition, there aremeasures available to provide managers withinformation regarding the risk that assets mightbecome impaired (e.g., monitoring of inventoryturnover to watch for slow-moving items). On theother hand, measures of competitor tactics, cus-tomer switching costs, and new technology oftenhave greater variation and contain more noise.Environmental upheaval, such as uncertainty, willcause measures to be less precise and thus noisier(Banker & Datar, 1989), thus not lending them-selves to use as a diagnostic control. While these

measures may facilitate discussion among topmanagement and lend themselves to an interactiveuse, they may not be stable or routine enough touse diagnostically.5

Empirical research shows that interactive sys-tems are eVective in Wrms facing various types ofrisk and uncertainty, including competitive, mar-ket, and technological risk and environmentaluncertainty (see e.g., Bisbe & Otley, 2004; Simons,1991). Bisbe and Otley (2004) conclude that Wrmsthat face high degrees of innovation risk anduncertainty have higher Wrm performance when acontrol system is used interactively. In Wrms withmanagers that clearly understand their strategy,Simons (1991) Wnds that uncertainties related toproduct technology, new product introductions,and market competition are associated with theuse of interactive controls. In addition, Simons(2000, p. 261) states “interactive control systemsare essential to monitor competitive risks in a cul-ture that could potentially create barriers toimpede the free Xow of information about emerg-ing threats and opportunities”. Abernethy andBrownell (1999) Wnd that the interactive use ofbudgets enhances performance in hospitals facingstrategic change.

Based on the above discussion, I posit that Wrmswill use diagnostic control to manage risk anduncertainty when more precise measurement islikely to be available. I also posit that Wrms willrely more on interactive controls when facinghigher levels of strategic risk and uncertainty. I for-mally hypothesize the following:

H2c: The extent to which Wrms face strategicrisks and uncertainties, speciWcally operatingrisk, operating uncertainty or the risk of assetimpairment, is positively associated with the

5 Table 2 shows that the standard deviation for operationaluncertainties is 1.04 while the standard deviation for competi-tive and technological uncertainty is 1.20 and 1.41, respectively.The standard deviation for operations risk is 0.78 while thestandard deviation for competitive risk is 1.38. Assuming thatthe properties of performance measures that capture the strate-gic factors exhibit the same type of variance, the data suggestthat measures of external factors are likely to be noisier thanmeasures of internal strategic factors.


emphasis they place on the use of performancemeasures in a diagnostic control system.

H2d: The extent to which Wrms face strategicrisks and uncertainties is positively associ-ated with the emphasis they place on the useof performance measures in an interactivecontrol system.

Costs and beneWts of management control: organizational learning and attention

Organizational learningOrganizational learning originates in historical

experiences which are then encoded in routines(Levitt & March, 1988). Based on historical experi-ences, the organization adopts and formalizes“routines that guide behavior” (Levitt & March,1988, p. 320). The beliefs, boundary, and diagnosticsystems are formalized routines (Simons, 2000)intended to guide behavior, and as such, facilitateorganizational learning. Simons (1994) Wnds thatafter undertaking a strategic turn around, manag-ers issue new mission and vision statements tocommunicate ideas and information to employees.The Wrms develop the new routines (i.e., belief sys-tem) based on their historical experience duringthe turn around. Marginson (2002) investigatescontrol components in a UK telecommunicationsWrm and Wnds that the beliefs system opens thedoors for new ideas, actions, and initiatives. More-over, he Wnds that the organization’s boundarysystem motivates a search for new ideas within theprescribed acceptable domain.

The diagnostic control system provides manag-ers with information on outcomes that are notmeeting expectations, and as such, is an example ofsingle loop learning (Argyris, 1977). Simons (1994)contends that diagnostic controls communicateagendas and translate strategy through the identiW-cation of critical success factors. This is consistentwith Kaplan and Norton (1996) who assert thatPM systems inform employees of the Wrm’s strat-egy. Kloot (1997) uses two case studies to investi-gate the link between management control systemsand organizational learning. She concludes thatmanagement control systems, in particular PMsystems, facilitate organizational learning. Although

Henri (2006) Wnds a negative relation betweendiagnostic controls and organizational learning, heuses a narrow deWnition of the diagnostic use ofperformance measures. Furthermore, his univari-ate statistics reveal a positive relation betweendiagnostic controls and learning.

Organizations must be oriented to learning inorder to learn (Hult & Thomas, 1998). An organi-zation is oriented to learning when it has “a cultureamenable to learning” and thus is able to improveits understanding of the environment over time(Galer & Van Der Heijden, 1992, p. 11). As dis-cussed above, I expect that the beliefs, boundary,and diagnostic systems will facilitate an organiza-tion’s orientation to learning and formally hypoth-esize:

H3a: The emphasis Wrms place on a beliefssystem is positively associated with an orga-nization’s orientation to learning.

H3b: The emphasis Wrms place on a bound-ary system is positively associated with anorganization’s orientation to learning.

H3c: The emphasis Wrms place on the use ofperformance measures in a diagnostic systemis positively associated with an organiza-tion’s orientation to learning.

An interactive control system is a double looplearning system, which is a more diYcult type oflearning tool than is a single loop system (Argyris,1977). The purpose of interactive controls is toenhance managers’ abilities to anticipate and eVec-tively manage future uncertainties (Simons, 2000)yet organizational learning is predicated on learn-ing from past events (Levitt & March, 1988). Levittand March (1988) posit that the lack of experienceand complexity of a given situation can inhibitlearning. Since interactive systems are intended toengage managers in scanning and seeking behav-iors that may result in emergent strategy (i.e., newbehaviors and experiences), interactive systemslikely help managers handle situations that arehigh in complexity and with which the managersmay have little experience.

However, Simons (1990, 1991, 2000) speciWcallysingles out the interactive system as a facilitator oforganizational learning. It is a system that Wrms


implement to ease the information processingdemands and facilitate the learning process byusing vertical channels throughout the organiza-tion (Galbraith, 1973). The control system shapesnew strategies, suggests new ideas and possibilities,and promotes curiosity and seeking behavior(Dent, 1990; Hopwood, 1987; Simons, 1994). Italso provides a signal downward in the organiza-tion regarding the important arena for proposingand implementing new ideas (Simons, 1990, 1991).Abernethy and Brownell (1999) provide empiricalsupport of the relation between interactive con-trols and organizational learning. In a study of 63hospitals they Wnd that an interactive control sys-tem facilitates organizational learning and thatorganizational learning is greater when the budget-ing system is used interactively rather than diag-nostically (see also Henri, 2006). Therefore, Ihypothesize the following:

H3d: The emphasis Wrms place on the use ofperformance measures in an interactive con-trol system is positively associated with anorganization’s orientation to learning.

Management control and top management attentionTime and processing capability are two con-

straints faced by top managers (Schick, Gordon, &Haka, 1990; Simons, 1990, 2000). Schick et al.(1990, p. 215) state that information overload“occurs when the demands on an entity for infor-mation processing time exceed its supply of time”.Monitoring multiple control systems can requiretremendous managerial attention, thus top man-agement has to choose where to focus their atten-tion. One of the top managers in Marginson’s(2002, p. 1026) case study states, “There is a con-stant need to prioritize issues that need addressing.There is never suYcient time or resources to see allissues simultaneously”.

Top management attention is needed to processthe information required to properly manage stra-tegic concerns. Simons (2000) asserts that relianceon the beliefs, boundary, and diagnostic systemscan facilitate the eYcient use of managementattention. Strategic concerns that are more routinein nature have a smaller information deWcit (i.e.,the diVerence between the amount of information

needed to manage the concern and the amount ofavailable information) and can be managed usingthe beliefs, boundary, and diagnostic systems thatalign employee behavior, deWne limits of accept-able behaviors, and use exception reporting,respectively. On the other hand, the interactivecontrol system is designed to help the organizationmanage those issues that have larger informationdeWcits and therefore, are more costly to the Wrm interms of the managerial attention they consume(Galbraith, 1973). Simons (1991) in his Weld studyof 30 health care products Wrms, Wnds that manag-ers experience information overload quickly whenfocusing attention on a broad spectrum of controlattributes. He documents that managers limitinteractive systems to include only one systemsince it consumes such large amounts of top man-agement attention. Overall, the interactive controlsystem is more “costly” that the other systems, dueto the information deWcit the system is intended toaccommodate. This discussion supports the follow-ing hypotheses:

H4a: The emphasis Wrms place on a beliefssystem is positively associated with theeYcient use of management attention.

H4b: The emphasis Wrms place on a bound-ary system is positively associated with theeYcient use of management attention.

H4c: The emphasis Wrms place on the use ofperformance measures in a diagnostic con-trol system is positively associated with theeYcient use of management attention.

H4d: The emphasis Wrms place on the use ofperformance measures in an interactive controlsystem will consume management attention.

Learning, attention, and Wrm performance

Organizations must be oriented to learning inorder to learn. The extant literature maintains thatorganizational learning is associated withimproved performance (Levitt & March, 1988;Slater & Narver, 1995). It is well-accepted thatorganizational learning is critical to maintainingcompetitive advantage in today’s global, competi-tive business world and some believe that learning


is the only way to compete in the long-term (Hult& Thomas, 1998; Slater & Narver, 1995). Tippinsand Sohi (2003) provide empirical evidence thatorganizational learning is positively associatedwith Wrm performance. They investigate informa-tion technology competencies and Wnd thatimproved Wrm performance results in the presenceof organizational learning. Chenhall (2005) pro-vides additional evidence directly relevant to themanagement accounting arena. He Wnds supportthat organizational learning enhances delivery out-comes. Furthermore, he concludes that organiza-tional learning mediates the relation betweencustomer dimensions of a PM system and servicedelivery. This leads to

H5a: Orientation to learning is positivelyassociated with Wrm performance.

Top managers are only able to process a limitedamount of information; they also have a con-strained attention span (Schick et al., 1990;Simons, 2000). An environment characterized byan ineYcient use of management attention willresult in ineVective managerial actions. Conversely,eYcient use of managerial attention allows topmanagers to focus more strongly on the criticalsuccess factors and core competencies of the orga-nization that are associated with enhanced perfor-mance. This leads to the following hypothesis:

H5b: EYcient use of management attentionis positively associated with Wrm perfor-mance.

Methods

Sampling frame

Data sources include Compustat and surveydata. For inclusion in the sampling frame, Wrmshad to have less than $2 billion in sales. This studyinvestigates the use of control systems by top man-agement and surveying one person regarding orga-nizational practices in Wrm with greater than $2billion in sales is considered problematic.6 To

enable analysis of non-response bias, selected Wrmsare required to have certain archival data availablein Compustat. I also discarded all non-US Wrms.To enhance the likelihood of being able to general-ize results, I use a random sample of 1000 Wrmsfrom the sampling frame of 4400 Wrms.

Survey and respondents

I use a three-page survey instrument, which Ipre-tested on three business professors and WveWnancial/accounting oYcers for clarity, under-standability, ambiguity, and face validity (Dillman,1978). Upon revision, I sent the survey, a personal-ized cover letter, and a stamped return envelope tothe chief Wnancial oYcer (CFO) of each company.I chose CFOs as informants since they are knowl-edgeable about the Wrm’s MCS. In addition,because of the restriction on Wrm size, they occupya position in which they have knowledge aboutstrategic issues and other items asked in the survey.As an incentive to respond, I promised to providethe participants an executive summary of theresults. In an attempt to increase the response rate,I sent two follow-up mailings. I was unable to con-Wrm a relevant mailing address either via the tele-phone or via the internet for 24 Wrms, leaving asample of 976 Wrms.

The mailing process resulted in 122 responses(12.5% response rate). I investigated non-responsebias by comparing non-respondents to respon-dents across several Wnancial accounts. The resultsare presented in Table 1, Panel A. I randomly sam-pled the population of Compustat Wrms duringearly Spring of 2004. The 2002 Compustat datawas the most recent data available at the time ofthe survey. There are no statistically signiWcantdiVerences between early and late respondents fortotal assets, sales, number of employees, or prop-erty, plant and equipment. I also compared earlyrespondents to late respondents, based on returndate, for all survey constructs. The results areshown in Table 1, Panel B. With the exception ofoperating risk I found no signiWcant diVerences.Early respondents had signiWcantly higher operat-ing risk than did late respondents (mean 6.11 vs.5.74, p < 0.05). Overall, the results support theabsence of signiWcant non-response bias. To assess6 For precedent see Fullerton and McWatters (2002).


the extent of common method bias, I ran a Hart-man’s one-factor test on the 54 survey questionsused to form the constructs. The factor solutionyielded 14 factors with eigenvalues > 1.0. The Wrstfactor explained 30% of the total variance. Overall,the results support the absence of signiWcant sin-gle-source bias (PodsakoV & Organ, 1986).

In unreported descriptive statistics, the patternof SIC classiWcations for respondents mirrors thesample frame. The two largest groups of respon-dent Wrms are manufacturers (54 Wrms) and serviceWrms (33 Wrms). There are also 10 mining Wrms, 10Wnancial services Wrms, eight transportation Wrms,Wve retailers, and two wholesalers. On average, thesample respondents have sales of $281.60 million,total assets of $455.13, and 1334 employees.

Variable measures

To establish validity of the survey variables, Iassess both content and construct validity (Nun-nally, 1978). I assess content validity through (1) areview of questions for face validity, (2) the processof variable construction, and (3) computation ofan empirical measure of internal consistency. I

assess construct validity through (1) specifying anappropriate domain of observables underlying theconstruct and using validated measures where pos-sible, (2) factor analyses, which reveal relationsamong the observables and support the uni-dimen-sionality of constructs, (3) multiple question load-ings in excess of 0.30 in support of convergentvalidity, and (4) lack of signiWcant cross-loadingsin support of discriminant validity. I took severalsteps in performing these assessments. First, I thor-oughly reviewed the existing literature to establishappropriate domains. Second, I used previouslyvalidated questions where appropriate. Third, Idiscussed the topic with several Wnancial andaccounting managers with the express purpose ofgaining more knowledge about the domain beingmeasured. Fourth, several academicians, alongwith the Wnancial and accounting managers,reviewed the questions for face validity. Fifth, Iconducted empirical tests (i.e., reviewed range ofresponses, calculated Cronbach’s Alpha, and fac-tor analyses) suggested by Nunnally (1978) to helpestablish both content and construct validity. Anabbreviated version of the survey is found inAppendix “Abbreviated survey questions” (note

Table 1Non-response bias

a The sample was divided into early, middle, and late respondents based on return date of survey.¤¤ Means are signiWcantly diVerent at p-value < 0.05.

Variable Respondents (n D 122) Non-respondents (n D 864)

Panel A: Respondents vs. non-respondentsTotal assets (in millions) 455.13 371.95Net sales (in millions) 281.60 248.90Number of employees (in thousands) 1.33 1.75Property, plant and equipment (in millions) 122.32 98.61

Construct Early respondents (n D 37) Late respondents (n D 38)Panel B: Early respondents vs. late respondentsa

Operational strategic uncertainties 4.93 4.93Competitive strategic uncertainties 4.69 4.54Technological strategic uncertainties 5.51 5.63Operations risk 6.11¤¤ 5.74¤¤

Competitive risk 4.12 3.86Boundary controls 5.70 5.63Belief controls 4.74 5.01Diagnostic controls 5.29 5.23Interactive controls 5.07 4.93Attention 4.91 4.90Learning 5.10 5.26Performance 4.52 4.20


that several questions are reverse-coded). The fac-tor analyses used to construct the variables arereported in Table 2. Responses to these surveyquestions are averaged to form the Wnal score forthe variable. Descriptive statistics for the multi-item variables are reported in Table 3.

Strategic riskSince this study is cross-sectional in nature, I cap-

ture a variety of types of strategic risk using Simons(2000) identiWcation of three types of strategic riskWrms’ face. Operations risk results from a break-down in core internal business processes such asmanufacturing or processing and impedes the Wrm’sability to implement its strategy (Simons, 2000). Imeasure operations risk (OPRSK) through four sur-vey questions that ask respondents how critical theWrm’s safety, quality, reliability, and eYciency oftheir operations are to organizational success. Com-petitive risk (CPRSK) is the risk inherent in themarketplace. Customers, suppliers, the existence ofsubstitute products, and new entrants to the market-place can pose problems that make it diYcult for aWrm to successfully create value and remain compet-itive (Simons, 2000). I measure CPRSK using sixquestions that capture supplier, customer, and com-petitor pressure. Asset impairment risk (ASRSK) isthe risk that balance sheet assets such as inventory,receivables, and investments will become impaired.Based on the discussion in Simons (1990), I use datafrom Compustat to calculate days sales outstanding(accounts receivable (data12)/sales (data2)¤365),inventory turnover (COGS data41/inventory(data3)), intangible intensity (intangibles (data33)/total assets (data6)), and capital intensity (net ppe(data8)/total assets (data6)).

The exploratory factor analysis reveals that theten questions used to measure strategic risk loadon three factors. I label the Wrst one OPRSK sincethe four questions chosen ex ante to measure oper-ational risk load on factor 1. Three of the six ques-tions related to competitive risk – ease of entry tothe industry, customer switching costs, and natureof the competition – load on the second factorwhich I label CPRSK. Two of the remaining threequestions load on a third factor, which has anunreliable Cronbach’s Alpha of 0.39 and is notused in the Wnal analysis. The sixth question cross-

loads on two factors and is removed from the Wnalanalysis (DeVellis, 1991). Overall, the constructscapture 53% of the explained variance. The Cron-bach’s Alpha for operational and competitive riskis 0.66 and 0.62, respectively. Thus, the analysisuses two survey proxies for strategic risk, alongwith archival measures of days sales outstanding,inventory turnover, intangible intensity, and capi-tal intensity.

Strategic uncertaintiesStrategic uncertainties are uncertainties that top

managers monitor to ensure organizational goalachievement (Simons, 1990). Similar to strategicrisk, I try and measure a variety of types of strate-gic uncertainties (STRUNC) by asking respondentsto assess the extent to which top managers monitorvarious uncertainties in order to ensure that thegoals of the Wrm are achieved. I include 12 diVerenttypes of uncertainties in order to capture uncertain-ties relevant across Wrms in the cross-section thatmay follow diVerent strategies (Simons, 1990). Thefactor analysis reveals three constructs. DiVusion ofknowledge, scale and scope eVects, input costs, andinternal product innovations load on factor 1,which I label operational uncertainties (OPUNC).Product introductions in adjacent industries, com-petitor market tactics, and new industry entrantsload on factor 2, which I label competitive uncer-tainties (COUNC). I label the third factor techno-logical uncertainties (TEUNC) since changes inproduct technology and new technology load. Twoquestions are not used in the Wnal analysis sincethey cross-load on two factors. Overall, the con-structs capture 63% of the explained variance. TheCronbach’s Alphas for the three factors are 0.78,0.75, and 0.92, respectively. Thus, the analysis usesthree proxies to capture strategic risk.

Control systemsThis study is based on the LOC framework; there-

fore, I measure the extent Wrms’ emphasize boundary,beliefs, diagnostic, and interactive controls (Simons,2000). Boundary systems (BOUND) are the controlsthat reign in the workforce and establish the out-of-bounds for employee actions. I measure BOUNDusing four questions that ask respondents to indicatethe organization’s use of a code of business conduct


Table 2Construct validity

Factors (0.63)

Operationaluncertainties

Competitiveuncertainties

Technologicaluncertainties

Strategic uncertaintiesChange in product technology ¡0.12 0.010 0.951New technology ¡0.032 ¡0.001 0.959Change in buyer tastesa 0.304 0.494 0.146Product introductions adjacent industries 0.031 0.625 0.257DiVusion of knowledge 0.635 ¡0.002 0.097Marketing innovationsa 0.544 0.315 0.020Scale eVects (product depth) 0.805 ¡0.125 0.130Scope eVects (product breadth) 0.945 ¡0.159 0.001Input costs 0.577 ¡0.185 ¡0.219Internal product innovations 0.679 ¡0.094 ¡0.097Competitor market tactics 0.070 0.812 ¡0.070New industry entrants ¡0.146 0.920 ¡0.082Std. Cronbach’s Alpha (variance extracted) 0.78 (54%) 0.75 (66%) 0.92 (93%)

Factors (0.53)

Operations Competitive Product

RiskQuality 0.761 ¡0.43 0.101Reliability 0.749 ¡0.068 ¡0.053Safety 0.667 0.033 0.059EYciency 0.604 0.160 ¡0.029Substitute productsa ¡0.086 0.382 0.715DiYcult to change suppliers 0.055 ¡0.104 0.777Customers set price 0.118 ¡0.249 0.523Ease of entry to industry (rev) ¡0.018 0.824 0.090Customer switching costs (rev) 0.196 0.707 ¡0.191Extent competition is fragmented ¡0.064 0.633 0.138Std. Cronbach’s Alpha (variance extracted) 0.66 (50%) 0.62 (57%) 0.39 (62%)

Factor (0.73)

Boundary systemDeWnes appropriate behavior 0.797Informs about oV-limits behavior 0.904Communicate risks to be avoided 0.820Workforce aware of code of conduct 0.881Standardized Cronbach’s Alpha 0.87

Factor (0.78)Belief system

Mission statement communicates values 0.844Top managers communicate values 0.911Workforce is aware of values 0.921Mission statement inspires 0.860Standardized Cronbach’s Alpha 0.91

Factors (0.66)

Diagnostic Interactive

Performance measurementProgress towards goals 0.655 0.249Monitor results 0.638 0.293


and systems that communicate areas/actions thatshould be avoided. Beliefs systems (BELIEF) are thecontrols that inspire the workforce to take desiredactions. I measure BELIEF using four questions thatassess the use of an organizational mission statement

and communication of core values. Factor analysesreveal that BOUND and BELIEF are uni-dimen-sional with explained variance of 73% and 78%,respectively. The Cronbach’s Alphas are 0.87 and0.91, respectively.

Table 2 (continued)

This table reports the results of factor analyses by broad construct. I use principal components with promax rotation to estimate thefactor analyses and extract all factors with eigenvalues > 1. The variance extracted for each factor analysis is reported in parentheses inthe top line of each table. I report the standardized Cronbach’s Alpha for each factor. For constructs with multiple factors, I report thevariance extracted for each factor in parentheses following the standardized Cronbach’s Alpha. For ease of presentation,loadings > 0.300 that are used in the Wnal measurement of constructs are highlighted in bold.

a Question loaded > 0.30 on two factors and was dropped from the Wnal measurement of constructs.

Factors (0.66)

Diagnostic Interactive

Compare outcomes 0.755 0.137Review key measures 0.775 0.091Enable discussion 0.911 ¡0.091Enable continual debate 0.918 ¡0.099Provide common view 0.936 ¡0.070Tie organization together 0.900 ¡0.044Focus on common issues 0.951 ¡0.082Focus on critical success factors 0.934 ¡0.085Develop common vocabulary 0.803 0.015Top mgrs pay little attention (rev) ¡0.058 0.821Top mgrs rely heavily on others (rev)a ¡0.456 0.590Operating mgrs involved infrequent (rev) 0.138 0.542Top managers pays day-to-day attention 0.049 0.819Top managers interpret information 0.197 0.600Operating managers frequently involved 0.288 0.621Std Cronbach’s Alpha (variance extracted) 0.96 (73%) 0.84 (62%)

Factor (0.73)

AttentionFocus attention 0.893EVectively leverage time 0.893Reduce need for constant monitoring 0.745W/O control attention spread thinly 0.873Standardized Cronbach’s Alpha 0.87

Factor (0.80)

LearningLearning is the key to improvement 0.920Values include learning 0.913No learning endangers our future 0.903Learning is an investment 0.843Standardized Cronbach’s Alpha 0.92

Factor (0.69)

PerformanceOrganizational performance 0.865Organizational proWtability 0.854Relative market share 0.825Productivity 0.784Standardized Cronbach’s Alpha 0.85


Table 3Descriptive statistics for survey items

Min Mean Median Max Std. dev.

Operational uncertainties 1.0 4.84 4.8 7.0 1.04DiVusion of knowledge 1.0 4.74 5.0 7.0 1.58Scale eVects (product depth) 1.0 4.63 5.0 7.0 1.36Scope eVects (product breadth) 1.0 4.78 5.0 7.0 1.34Input costs 1.0 5.03 5.0 7.0 1.41Internal product innovations 1.0 5.03 5.0 7.0 1.55

Competitive uncertainties 1.0 4.63 4.7 7.0 1.20Product introductions adjacent industries 1.0 4.31 4.0 7.0 1.55Competitor market tactics 1.0 4.99 5.0 7.0 1.42New industry entrants 1.0 4.61 5.0 7.0 1.58

Technological uncertainties 1.0 5.49 6.0 7.0 1.41Change in product technology 1.0 5.53 6.0 7.0 1.52New technology 1.0 5.45 6.0 7.0 1.43

Operations risk 3.5 5.93 6.0 7.0 0.78Quality 4.0 6.39 7.0 7.0 0.81Reliability 4.0 6.43 7.0 7.0 0.71Safety 1.0 4.83 5.0 7.0 1.91EYciency 3.0 6.07 6.0 7.0 0.96

Competitive risk 1.0 4.06 4.0 7.0 1.38Ease of entry to industry (rev) 1.0 3.18 3.0 7.0 1.75Customer switching costs (rev) 1.0 4.63 4.5 7.0 1.92Extent competition is fragmented 1.0 4.36 5.0 7.0 1.89

Beliefs system 1.0 4.74 4.75 7.0 1.34Mission statement communicates values 1.0 4.90 5.0 7.0 1.58Top managers communicate values 1.0 4.98 5.0 7.0 1.46Workforce is aware of values 1.0 4.98 5.0 7.0 1.49Mission statement inspires 1.0 4.07 4.0 7.0 1.64

Boundary system 2.0 5.58 5.8 7.0 1.07DeWnes appropriate behavior 1.0 5.53 6.0 7.0 1.32Informs about oV-limits behavior 2.0 5.88 6.0 7.0 1.17Communicate risks to be avoided 2.0 5.26 6.0 7.0 1.32Workforce aware of code of conduct 1.0 5.64 6.0 7.0 1.29

Diagnostic 1.0 5.21 5.4 7.0 1.14Progress towards goals 1.0 5.22 6.0 7.0 1.51Monitor results 1.0 5.61 6.0 7.0 1.34Compare outcomes 1.0 5.59 6.0 7.0 1.35Review key measures 1.0 5.43 6.0 7.0 1.38Enable discussion 1.0 5.18 5.0 7.0 1.35Enable continual debate 1.0 4.99 5.0 7.0 1.42Provide common view 1.0 4.94 5.0 7.0 1.26Tie organization together 1.0 4.88 5.0 7.0 1.34Focus on common issues 1.0 4.93 5.0 7.0 1.31Focus on critical success factors 1.0 5.26 6.0 7.0 1.36Develop common vocabulary 1.0 5.07 5.0 7.0 1.33

Interactive 1.0 5.00 5.0 7.0 1.22Top mgrs pay little attention (rev) 1.0 5.11 5.0 7.0 1.59Operating mgrs involved infrequent (rev) 1.0 4.83 5.0 7.0 1.60Top managers pays day-to-day attention 1.0 4.85 5.0 7.0 1.65


Diagnostic systems (DIAG) provide routineinformation to managers about key measures andprogress towards goals. In contrast, interactive sys-tems (INTER) require tremendous top manage-ment involvement. Although Wrms may use manycontrol systems either interactively or diagnosti-cally, this study investigates the use of the PM sys-tem for two reasons. First, an eVective PM systemis viewed as being increasingly important to thesuccess of Wrms in today’s competitive environ-ment thus, in the cross-section, the PM systemshould be important across Wrms. Second, theresults of this study can contribute to a long streamof research on performance measures (see e.g., Itt-ner & Larcker, 1998).

I measure both DIAG and INTER using ques-tions and concepts from Henri (2006), Simons(2000), and Kaplan and Norton (1996). Since bothsets of questions enquire about the PM system, Iuse an exploratory factor analysis across all ques-tions to ensure discriminant validity. Eleven ques-tions about goals, monitoring, commonalities, andkey success factors load on factor 1, which I labeldiagnostic controls (DIAG).7 Six questions that

speciWcally mention the involvement of top and/oroperating managers load on factor 2, which I labelinteractive controls (INTERACT). The distin-guishing feature between DIAG and INTERACT isthe involvement of top and operating managers.Bisbe, Batista-Foguet, and Chenhall (2007) identifyWve dimensions that deWne the theoretical proper-ties of an interactive control. That is, Simons’notion of an interactive control appears to be com-prised of a set of Wve linked dimensions. In thisstudy, interactive control corresponds to two of theproperties having to do with the involvement oftop and operating managers.8 The three (i.e., face-to-face challenges, focus on strategic uncertainties,and inspirational involvement) remaining proper-ties are omitted in this study (see also Section“Validity tests for statistical robustness”). Overall,the constructs capture 66% of the explained vari-ance. The Cronbach’s Alphas for DIAG andINTERACT are 0.96 and 0.84, respectively.

Costs, beneWts, and performanceBased on the discussion in Simons (2000) I mea-

sure management attention (ATTEN) using four

7 These questions are from Henri (2006); however, he usesthese questions to measure two constructs: interactive and diag-nostic uses. In this study, all nine questions load on one factor.Also see the sensitivity test in Section “Validity tests for statisti-cal robustness”.

Table 3 (continued)

Min Mean Median Max Std. dev.Top managers interpret information 1.0 5.23 6.0 7.0 1.40Operating managers frequently involved 1.0 4.98 5.0 7.0 1.59

Learning 1.0 5.06 5.3 7.0 1.19Learning is the key to improvement 1.0 5.02 5.0 7.0 1.36Values include learning 1.0 4.98 5.0 7.0 1.37No learning endangers our future 1.0 5.25 5.5 7.0 1.33Learning is an investment 1.0 4.98 5.0 7.0 1.31

Attention 1.5 4.95 5.0 7.0 1.11Focus attention 1.0 5.32 6.0 7.0 1.29EVectively leverage time 1.0 4.93 5.0 7.0 1.27Reduce need for constant monitoring 1.0 4.45 5.0 7.0 1.40W/O control attention spread thinly 2.0 5.09 5.0 7.0 1.32

Performance 2.0 4.32 4.3 7.0 1.14Organizational performance 2.0 4.37 4.0 7.0 1.39Organizational proWtability 1.0 4.11 4.0 7.0 1.69Relative market share 2.0 4.34 4.0 7.0 1.29Productivity 2.0 4.46 4.0 7.0 1.18

8 In addition, this distinction is consistent with Simons (1994,p. 171) who states that interactive control systems are “formalsystems used by top managers to regularly and personally in-volve themselves in the decision activities of subordinates” (seealso Simons, 1991, p. 49, 2000, p. 216).


questions that ask about the ability to focus atten-tion, eVectively leverage time, reduce need for con-stant monitoring, and how the control systemimpacts attention.9 Organizational learning isdependent upon whether the organization has aculture that is suitable and enables learning (Galer& Van Der Heijden, 1992). I use a previously vali-dated scale (Henri, 2006; Hult & Thomas, 1998)comprised of four questions to capture learning(LEARN).10 Performance is measured using a pre-viously validated scale (Roth & Jackson, 1995).11 Iask the respondents to assess their organizationalperformance relevant to their goals on four dimen-sions: overall, proWtability, market share, anddelivery system. I use this measure since it waspreviously validated for both service and manufac-turing Wrms and thus should be relevant in thecross-section. The survey measure of performance

correlates positively with an archival measure ofreturn-on-assets deWned as pre-tax income dividedby total assets (r D .179, p < 0.05). Factor analysisreveals that the three constructs are uni-dimen-sional and explain 73%, 80%, and 69% of the vari-ance, respectively. The Cronbach’s Alphas are 0.87,0.92, and 0.85, respectively.

Summary of constructsThe constructs, with the exception of asset

impairment risk, are measured using Likert-scalesurvey questions with an available range of 1–7.The descriptive statistics presented in Table 3 forthe survey questions reveals that all questions,with the exception of operations risk, include abroad range with a minimum of either 1 or 2 and amaximum of 7. Although operations risk has a con-stricted range, at the minimum it still encompassesa four-point scale. The factor analyses reveal uni-dimensional constructs, with acceptable explainedvariance. In addition, all constructs used in theWnal analyses have acceptable Cronbach’s Alphas(Nunnally, 1978).

Although the constructs are theoretically dis-tinct and measured using either validated scales orquestions drawn from the underlying literature, it isstill likely that they are correlated with one another.Since this study models these variables as distinctconstructs, I present a multitrait matrix in Table 4.While the use of factor analyses for risk, strategicuncertainties, diagnostic, and interactive controls

9 I also speciWcally test to ensure that ATTN and INTERACTare distinct constructs by conducting an exploratory analysisthat includes all of the underlying questions of ATTN andINTERACT. The factor analysis reveals two factors with eigen-values >1.0 and the questions appropriately load on ATTN andINTERACT with no evidence of cross-loading. Thus, I con-clude that ATTN and LEARN are not only theoretically dis-tinct constructs, but are also empirically distinct.10 The Hult and Thomas (1998) measure of learning orienta-

tion had a variance extracted D 0.49 and a constructreliability D 0.79 (>0.70 in Henri, 2006).11 The Roth and Jackson (1995) measure of performance had

a coeYcient alpha D 0.82 and validated it against return on as-sets (r D 0.24, p < 0.05).

Table 4Multitrait matrixa

a The diagonal of the matrix is the Cronbach’s Alpha for each variable. The remainder of the tables reports the bi-variate correlationcoeYcients.¤¤,¤: p-value signiWcant at <0.05, 0.01, respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

ATTEN (1) 0.87BELIEF (2) 0.440¤¤ 0.91BOUND (3) 0.359¤¤ 0.520 0.87CORSK (4) 0.073 ¡0.011 0.062 0.62COUNC (5) 0.275¤¤ 0.386¤¤ 0.250¤¤ 0.009 0.75DIAG (6) 0.533¤¤ 0.580¤¤ 0.398¤¤ 0.059 0.371¤¤ 0.96INTER (7) 0.214¤ 0.421¤¤ 0.289¤¤ 0.113 0.276¤¤ 0.633¤¤ 0.84LEARN (8) 0.363¤¤ 0.485¤¤ 0.356¤¤ ¡0.013 0.284¤¤ 0.500¤¤ 0.343¤¤ 0.92OPRSK (9) 0.196¤ 0.303¤¤ 0.230¤ 0.043 0.326¤¤ 0.376¤¤ 0.259¤¤ 0.443¤¤ 0.66OPUNC (10) 0.318¤¤ 0.359¤¤ 0.235¤¤ ¡0.094 0.498¤¤ 0.376¤¤ 0.191¤ 0.378¤¤ 0.243¤¤ 0.78PERF (11) 0.359¤¤ 0.296¤¤ 0.230¤ ¡0.145 0.156 0.375¤¤ 0.215¤ 0.384¤¤ 0.271¤¤ 0.207¤ 0.85TEUNC (12) 0.101 0.251¤¤ 0.065 ¡0.335¤¤ 0.363¤¤ 0.151 0.061 0.179¤ 0.118 0.372¤¤ 0.240¤¤ 0.92


clearly demonstrate that empirically the six con-structs are distinct, the multitrait matrix providesadditional evidence. The diagonal of the matrix(the “reliability diagonal”) contains the Cronbach’sAlpha for each latent construct and shows theinternal consistency or reliability. The remainder ofthe table is a correlation matrix between the pairsof variables. In order to demonstrate that thedimensions are distinct, the correlation coeYcientswithin a column should be less than the coeYcientalphas found in the diagonal (Churchill, 1979).Examining Table 4 shows that the internal reliabil-ity is much higher than the inter-item reliability.The correlation coeYcients range up to 0.58 andmany of the pairs of variables are signiWcantly cor-related. The Cronbach’s Alphas range between 0.62and 0.96. This analysis provides strong evidence ofdiscriminant validity since the Cronbach’s Alphasexceed the inter-item correlations in all cases. Basedon the results of both the factor analyses and themultitrait matrix I conclude that there is strongempirical support for discriminant validity.12

Results

I use the AMOS 4.0 software program, withdefault maximum likelihood estimation technique13,to estimate a system of equations described inFig. 1.14 I use SEM since it forces a decision regard-ing whether the entire model is satisfactory (Kline,

1998). It brings to the table a macro-view of thecontrol framework and provides a big picture per-spective. Simons (2000) asserts that the success ofthe control framework is contingent on the properuse of all four control systems; thus a model focusis appropriate. Due to a sample size of 122 obser-vations, the constructs are treated as manifest vari-ables15 using the summated scores described earlierin Section “Methods”.

The analysis proceeds in three steps. First, I esti-mate a base model based on Fig. 1. I trim themodel for insigniWcant links that are not hypothe-sized. I then trim the model guided in part byempirics. Model trimming is used to derive a parsi-monious, well-Wtting base model (Kline, 1998),which results in a base model that provides evi-dence on the holistic model. The results for thisstep are presented in Table 5. Second, I use thetrimmed base model arrived at in the Wrst step as abasis of discussion for the speciWc hypotheses.Third, the existence of a well-Wtting model does notensure that the model is the only appropriatemodel (Kline, 1998). Therefore, I generate six alter-native models for comparison against the basemodel. This analysis is presented in Table 6. I showa graphical depiction of the signiWcant results con-tained in the Wnal mode in Fig. 2.

Generation of trimmed base model – step 1

The analysis is exclusive of ASRSK, CORSK,and TEUNC. None of these variables have signiW-cant explanatory power. All remaining path resultsare robust to either the inclusion or exclusion ofASRSK, CORSK, or TEUNC in the model, thusthey are removed for the sake of parsimony. I usethe Chi-square, the Chi-square divided by themodel degrees of freedom (CMINDF), the com-parative Wt index (CFI), the goodness of Wt index(GFI) and the root mean square error of approxi-mation (RMSEA) as indicators of model Wt.An insigniWcant Chi-square (Joreskog, 1969), aCMINDF ratio less than 5 (Wheaton, Muthen,Alwin, & Summers, 1977), a CFI and GFI close to

12 As a robustness check of discriminant validity, I also performa technique suggested by Fornell and Larcker (1981) that com-pares the variance extracted for each construct to the squaredcorrelation. The variance extracted for the latent variables rangesfrom 0.50 to 0.92, which in all cases exceeds the squared inter-pair correlations. Thus, I conclude that the discriminant validityof the constructs is robust across multiple tests.13 Kurtosis and skewness indicate that the data is within toler-

able levels of univariate normality. Kline (1998) suggests thatskewness greater than 3.0 and kurtosis greater than 10.0 maysuggest that a problem with the data. Kline (1998, p. 83) notesmultivariate non-normality can usually be identiWed throughunivariate procedures.14 The analysis is run using data from 122 surveys. I found no

evidence of multicollinearity or heteroscedasticity. SpeciWcationtests for multicollinearity included reviewing the tolerance andvariance inXation factor for each coeYcient, and for heterosced-asticity included the review of plots of the residuals and theWhite test.

15 This technique is used when working with small sample sizessince it reduces the number of parameters that are estimatedthus accommodating smaller samples.


1 (Arbuckle & Wothke, 1999; Bentler, 1990), andan RMSEA of less than 0.08 (Browne & Cudeck,1993) indicates good Wt. As shown in Table 5, thebase model is reasonably well-Wtting.

In alternative model 1 I trim the path from com-petitive uncertainty to the diagnostic system. Thispath is not signiWcant nor is it hypothesized. TheX2 diVerence test is insigniWcant which indicates

Table 5Structural model (base)

This table presents the results of a structural equation model. The trimmed model (Alternative 1) removes the insigniWcant paths fromstrategic factors to the control systems as hypothesized in H2c and H2d. The trimmed model (Alternative 2) is guided by both underly-ing theory and empirical results and removes the paths from strategic factors to the boundary system, from the boundary to the diag-nostic control system, and from the interactive control system to learning orientation.¤¤¤, ¤¤, ¤: SigniWcant at p-value < 0.01, 0.05, 0.10, respectively (one-tailed).

Dependent variable (R2) Independent variable Base model Trimmed model (Alt. 1) Trimmed model (Alt. 2)

Coef. Coef. Coef.

DIAG (0.213) COUNC 0.023 – –OPUNC 0.129¤¤ 0.135¤¤ 0.137¤¤

OPRSK 0.112¤¤ 0.115¤¤ 0.119¤¤

BELIEF 0.278¤¤¤ 0.280¤¤¤ 0.303¤¤¤

BOUND 0.057 0.058 –INTERACT 0.462¤¤¤ 0.464¤¤¤ 0.470¤¤¤

BELIEF (0.188) COUNC 0.177¤¤ 0.177¤¤ 0.177¤¤

OPUNC 0.273¤¤¤ 0.273¤¤ 0.273¤¤¤

OPRSK 0.150¤¤ 0.150¤¤ 0.150¤¤

BOUND (0.096) COUNC 0.044 0.044 –OPUNC 0.027 0.027 –OPRSK 0.090 0.090 –BELIEF 0.478¤¤¤ 0.478¤¤¤ 0.530¤¤¤

INTERACT (0.117) COUNC 0.141¤ 0.141¤ 0.116¤

OPUNC ¡0.065 ¡0.065 –OPRSK 0.170¤¤ 0.170¤¤ 0.167¤¤

BELIEF 0.352¤¤¤ 0.352¤¤¤ 0.337¤¤¤

ATTEN (0.381) DIAG 0.538¤¤¤ 0.538¤¤¤ 0.543¤¤¤

BELIEF 0.255¤¤¤ 0.255¤¤¤ 0.256¤¤¤

BOUND 0.105 0.105 0.106INTERACT ¡0.315¤¤¤ ¡0.315¤¤¤ ¡0.316¤¤¤

LEARN (0.327) DIAG 0.303¤¤¤ 0.303¤¤¤ 0.279¤¤¤

BELIEF 0.351¤¤¤ 0.351¤¤¤ 0.350¤¤¤

BOUND 0.070 0.070 0.067INTERACT ¡0.041 ¡0.041 –

PERF (0.166) ATTEN 0.210¤¤¤ 0.210¤¤¤ 0.209¤¤¤

LEARN 0.292¤¤¤ 0.292¤¤¤ 0.292¤¤¤

Model WtX2 24.747 24.854 28.406p-value 0.053 0.072 0.163df 15 16 22CMINDF 1.650 1.553 1.291RMSEA 0.073 0.068 0.049GFI 0.962 0.962 0.957CFI 0.974 0.977 0.983

X2 diVerence testX2 diVerence (df) 0.107 (1) 2.417 (4)p-value Ns Ns


Table 6Comparison of trimmed base model to alternative models

This table presents the results of comparing the trimmed base model (Alt. 2) shown in Table 5 to alternative model speciWcations. TheWrst Wve alternative model speciWcations build the model by adding one additional path. A signiWcant X2 diVerence test indicates thatthe alternative model is better Wtting than the base model. Alternative 6, however, trims the model by removing nine paths and onlyadding four paths, for a net reduction of Wve paths. A signiWcant X2 diVerence test indicates that the alternative model does not betterWt the data; rather, the model has been trimmed too much and the base model is better speciWed.¤¤¤, ¤¤, ¤: SigniWcant at p-value < 0.01, 0.05, 0.10, respectively (one-tailed), except for the alternative model speciWcation tests, which aretwo-tailed.

Dependentvariable

Independentvariable

Trimmedbase model

Alt. 1 Alt. 2 Alt. 3 Alt. 4 Alt. 5 Alt. 6

DIAG OPUNC 0.137¤¤ 0.140¤¤ 0.137¤¤ 0.137¤¤ 0.137¤¤ 0.137¤¤ 0.137¤¤

OPRSK 0.119¤¤ 0.074 0.119¤¤ 0.119¤¤ 0.119¤¤ 0.119¤¤ 0.119¤¤

BELIEF 0.303¤¤¤ 0.218¤ 0.303¤¤¤ 0.303¤¤¤ 0.303¤¤¤ 0.304¤¤¤ 0.303¤¤¤

INTERACT 0.470¤¤¤ 0.701¤¤ 0.470¤¤¤ 0.470¤¤¤ 0.470¤¤¤ 0.472¤¤¤ 0.470¤¤¤

BELIEF COUNC 0.177¤¤ 0.177¤¤ 0.177¤¤ 0.177¤¤ 0.177¤¤ 0.132 0.177¤¤

OPUNC 0.273¤¤¤ 0.273¤¤¤ 0.273¤¤¤ 0.273¤¤¤ 0.273¤¤¤ 0.266¤¤¤ 0.273¤¤¤

OPRSK 0.150¤¤ 0.150¤¤ 0.150¤¤ 0.150¤¤ 0.150¤¤ 0.101 0.150¤¤

BOUND BELIEF 0.530¤¤¤ 0.530¤¤¤ 0.530¤¤¤ 0.530¤¤¤ 0.461¤¤¤ 0.530¤¤¤ 0.530¤¤¤

INTERACT COUNC 0.116¤ 0.187 0.116¤ 0.116¤ 0.116¤ 0.183¤ 0.116¤

OPRSK 0.167¤¤ 0.277 0.167¤¤ 0.167¤¤ 0.167¤¤ 0.205¤¤ 0.167¤¤

BELIEF 0.337¤¤¤ 0.607 0.337¤¤¤ 0.337¤¤¤ 0.337¤¤¤ 0.116 0.337¤¤¤

ATTEN DIAG 0.543¤¤¤ 0.541¤¤¤ 0.530¤¤¤ 0.543¤¤¤ 0.543¤¤¤ 0.541¤¤¤ –BELIEF 0.256¤¤¤ 0.256¤¤¤ 0.242¤¤ 0.256¤¤¤ 0.257¤¤¤ 0.256¤¤¤ –BOUND 0.106 0.106 0.106 0.106 0.106 0.106 –INTERACT ¡0.316¤¤¤ ¡0.316¤¤¤ ¡0.315¤¤¤ ¡0.316¤¤¤ ¡0.317¤¤¤ ¡0.316¤¤¤ –

LEARN DIAG 0.279¤¤¤ 0.279¤¤¤ 0.279¤¤¤ 0.262¤¤¤ 0.279¤¤¤ 0.279¤¤¤ –BELIEF 0.350¤¤¤ 0.350¤¤¤ 0.350¤¤¤ 0.337¤¤¤ 0.350¤¤¤ 0.350¤¤¤ –BOUND 0.067 0.067 0.067 0.062 0.066 0.067 –

PERF ATTEN 0.209¤¤¤ 0.209¤¤¤ 0.209¤¤¤ 0.209¤¤¤ 0.209¤¤¤ 0.209¤¤¤ –LEARN 0.292¤¤¤ 0.292¤¤¤ 0.292¤¤¤ 0.292¤¤¤ 0.292¤¤¤ 0.292¤¤¤ –

Alternative model speciWcationINTERACT DIAG – ¡0.552 – – – – –ATTN LEARN – – 0.042 – – – –LEARN ATTN – – – 0.051 – – –BOUND INTERACT – – – – 0.163¤ – –BELIEF INTERACT – – – – – 0.222 –PERF DIAG – – – – – – 0.332¤¤¤

PERF BELIEF – – – – – – 0.087PERF BOUND – – – – – – 0.046PERF INTERACT – – – – – – ¡0.058

Model WtX2 28.406 27.851 28.182 28.077 24.669 27.950 8.680p-value 0.163 0.1444 0.135 0.138 0.262 0.142 0.563df 22 21 21 21 21 21 10CMINDF 1.291 1.326 1.342 1.337 1.175 1.331 0.868RMSEA 0.049 0.052 0.053 0.053 0.038 0.952 0.000GFI 0.957 0.958 0.957 0.958 0.962 0.958 0.983CFI 0.983 0.982 0.981 0.981 0.990 0.982 1.00

X2 diVerence testX2 diVerence (df) 0.555 (1) 0.226 (1) 0.329 (1) 3.737 (1) 0.456 (1) 19.726 (12)p-value Ns Ns Ns P < 0.10 Ns P < 0.10


that the model has not been overly trimmed (Kline,1998). Inspection of the model reveals that empha-sis on the boundary system is inXuenced by theemphasis placed on the beliefs system and not bystrategic factors the Wrm faces. In other words, theresults suggest that strategic factors inXuence theemphasis placed on the beliefs system, which inturn, inXuences the emphasis placed on the bound-ary system. Thus, in alternative model 2, I trim thepaths from the strategic factors to the boundarysystem. I also trim the path from the interactivesystem to learning. The results suggest that theinteractive use of performance measures inXuencesthe diagnostic use of performance measures, whichin turn, inXuences learning. In addition, I trim thepath from operational uncertainty to the interac-tive control system. The results indicate that thebeliefs and diagnostic systems are used to manageoperational uncertainty. Finally, I trim the pathfrom the boundary system to the diagnostic sys-tem, since strategic factors and the interactive andbeliefs systems explain the use of the diagnostic

system. I leave the insigniWcant paths between theemphasis placed on the boundary system and eachof attention and learning since eliminating thesepaths would purely be empirically driven (i.e., themodel does not contain an alternative explanationsince there are no other paths originating from theboundary system). Trimming the paths to derivealternative model 2 results in an insigniWcant X2

diVerence suggesting that the model has not beenoverly trimmed (Kline, 1998). The Wnal trimmedmodel (i.e., Model Alt. 2) is well-Wtting with aninsigniWcant X2 (p D 0.163), a CMINDF D 1.291, aGFI D 0.957, a CFI D 0.983, and an RMSEA D0.049.

In summary, I conclude that many of the rela-tions posited to exist in the LOC framework arewell-represented by the data. In general, the extentto which Wrms face strategic risk and uncertaintydrive the emphasis placed on control systems,which in turn, inXuence learning, attention, andperformance. However, contrary to the frameworkand to prior literature, I Wnd that the interactive

Fig. 2. Graphical depiction of signiWcant results. This table presents the signiWcant path coeYcients of the Wnal model shown in Table 6(Alternative 4). The squared multiple correlations for the endogeous variables are shown in parentheses. The covariances among theendogenous variables are positive and signiWcant (but not shown above).

Strategic Elements Control Systems Behavioral Responses Performance

Attention (0.387)

Beliefs System (0.214)

Learning (0.359)

Performance (0.171)

Boundary System (0.303)

Diagnostic Controls (0.591)

Interactive Controls (0.226)

Operational Uncertainties (proxy for strategic uncertainty)

Operational Risks (proxy for strategic risk)

Competitive Uncertainties (proxy for strategic uncertainty)

0.273***

0.119**

0.167**

-0.317***

0.279***

0.543***

0.350***

0.257****

0.292***

0.209**

0.116*

0.177**

0.137**

0.150**

0.303***

0.470***

0.163*

0.337***

0.461***


use of performance measures does not facilitateorganizational learning, nor does the emphasisplaced on the boundary system impact any of theoutcomes investigated in this study. Moreover, IWnd evidence of inter-dependencies among thecontrol systems. I now turn to a discussion of thespeciWc hypotheses.

Discussion of hypotheses – step 2

The path coeYcients reported in the trimmedmodel (Alt. 2), Table 5 provide evidence on thehypotheses. The Wrst set of hypotheses suggeststhe existence of relations among control systems.The emphasis Wrms place on the beliefs system ispositively associated with the emphasis they placeon the boundary system (p < 0.01), the diagnosticsystem (p < 0.01) and the interactive system (p <0.01), which provides support for H1a–H1c. TheseWndings are consistent with the argument that thebeliefs system inXuences and complements each ofthe other control systems in the LOC framework.In addition, H1e is supported since the interactiveuse of performance measures helps explain thediagnostic use of performance measures (p < 0.01).Overall, these results provide support that controlsare inter-related and are complementary. I Wnd noevidence that the controls are substitutes.

The second set of hypotheses predicts that theextent to which Wrms face strategic risk and uncer-tainties is positively associated with the emphasisWrms place on the beliefs, boundary, and interac-tive systems, and that strategic factors that likelylend themselves to relatively precise measurementare associated with the diagnostic use of perfor-mance measures. H2a is supported since Wrms fac-ing higher levels of operational risk (p < 0.05),competitive uncertainty (p < 0.05) and operationaluncertainty (p < 0.01) place more emphasis on thebeliefs system. H2b is not supported since strategicfactors are not associated with the emphasis placedon a boundary system. H2c is supported sinceWrms facing higher strategic factors of operationaluncertainty and operational risk emphasize thediagnostic use of a PM system (p < 0.05). There isalso some support for H2d since the extent towhich Wrms face competitive uncertainty is mar-ginally associated with the interactive use of a PM

system (p < 0.10). Although not hypothesized, Ialso Wnd that operational risk is associated with aninteractive use of performance measures (p < 0.05).The results provide support that the role of the PMsystem varies depending on the strategic uncer-tainty and risk the Wrm faces. The PM system isused interactively when the Wrm faces competitiveuncertainty, but not either technological or opera-tional uncertainties. It is also used interactivelywhen the Wrm faces operational risk, but not eitherasset impairment or competitive risk. One interest-ing observation is that the PM system is used bothdiagnostically and interactively in conjunctionwith operating risk. This is consistent with Margin-son’s (2002) Wnding that Wrms may use some partsof their PM system interactively while using otherparts diagnostically.

The remaining hypotheses investigate out-comes associated with the emphasis Wrms place oncontrol systems. The third set of hypotheses pre-dicts that a Wrm’s orientation to learning is higherwhen Wrms place more emphasis on their controlsystems. The coeYcients on the paths from each ofDIAG and BELIEF to LEARN provide supportfor hypotheses 3A and 3C. As Wrms place moreemphasis on diagnostic uses of the PM system andthe beliefs system, they have an organization moreamenable to learning (p < 0.01, p < 0.01, respec-tively). Emphasis of the boundary system and theinteractive use of the PM system are not associatedwith organizational learning, which indicates thathypotheses 3b and 3d are not supported.16 Simons(1990) suggests that the structure found in formal

16 As expected, the interactive use of the PM system and theboundary system are each positively correlated with an organi-zation’s orientation to learning when inspecting the bi-variatecorrelations shown in Table 4 (p-value < 0.05 for both rela-tions). This result is robust to the path model when the interac-tive use of the performance measurement system or theboundary system is modeled as the only driver of organiza-tional learning (p < 0.01); however, when all control attributesare modeled as drivers of organizational learning then the inter-active use of the performance measurement system and theboundary system do not signiWcantly inXuence organizationallearning. These results are robust using OLS as well. The VIF isless than 2.5 indicating that multicollinearity is not likely an is-sue. This indicates that an interactive control system and aboundary system are not signiWcantly associated with learningwhen controlling for the beliefs and diagnostic control systems.


management control systems facilitates organiza-tional learning. However, the results imply that theinteractive control system may be more organicand inXuence organizational learning through theformal structure of the diagnostic control system.

The fourth set of hypotheses predicts a relationbetween the eYcient use of management attentionand each of the control systems. H4a, 4b and 4cpredict that emphasis on a beliefs system, a bound-ary system, and a diagnostic control system willfacilitate the eYcient use of management attention.In other words, these types of control “free up”management attention and facilitate the eYcientuse of a constrained resource. The signiWcantcoeYcients on the paths between ATTEN and eachof DIAG and BELIEF (p < 0.01, p < 0.01, respec-tively) provide support for hypotheses 4a and 4c.Hypothesis 4b is not supported. In contrast,hypothesis 4d predicts that the interactive use ofthe PM system consumes management attention.In other words, a “cost” of using an interactivecontrol system is the use of management attention.The negative coeYcient on the path betweenINTERACT and ATTEN (p < 0.01) provides sup-port for hypothesis 4d. Hypothesis 5a and 5b pre-dicts that the eYcient use of attention and higherlevels of organizational learning are associatedwith higher levels of Wrm performance. The resultsshow that higher performance is associated withhigher levels of both attention (p < 0.01) and orga-nizational learning (p < 0.01), thus these twohypotheses are supported.

Comparison to alternative models – step 3

Although the trimmed model presented inTable 5 is well-Wtting, there is no assurance that itis the only model. The purpose of this section is tocompare the trimmed base model to alternativemodels to rule out alternative model speciWcations(Kline, 1998). The Wrst Wve alternative model speci-Wcations in Table 6 build the model by adding apath. When building a model, a signiWcant X2

diVerence test indicates that the alternative modelis better-Wtting then the base model (Kline, 1998).Alternative 1 proposes that the relation betweeninteractive and diagnostic controls is non-recur-sive. That is, not only does the interactive system

inXuence the diagnostic system, but the diagnosticsystem may also inXuence the interactive system.This stems from Simons (2000) who depicts therelation between the interaction and diagnosticsystems as being bi-directional. Alternatives 2 and3 investigate the relation between learning andattention. Organizational learning consumes atten-tion, that is, attention can be viewed as a cost oflearning. Levitt and March (1988) assert that“mechanisms of attention within a memory struc-ture” are necessary to facilitate learning. To becomplete I run two alternative models (2 and 3)which adds a path from LEARN to ATTEN, andfrom ATTEN to LEARN, respectively. Both thepath coeYcient and the X2 diVerence test on eachof these Wrst three alternative models are insigniW-

cant, which indicates that the alternative model isnot better-Wtting than the original model.

In the base model I hypothesize that the interac-tive use of performance measures inXuence thediagnostic use of performance measures. In addi-tion, it is possible that the interactive system inXu-ences each of the boundary and beliefs systems. Astop managers gain information through the inter-active system regarding emergent strategies, it ispossible that the Wrm adjusts its acceptable bound-aries and the Wrm’s vision or mission statement.For example, Simons (1994) Wnds that as emergingstrategy is clariWed managers draft new missionstatements and redeWne the boundary system. Tuo-mela (2005) also argues that the dialogue stemmingfrom an interactive control system can lead theWrm to place more emphasis on the boundary andbeliefs system in order to communicate the newstrategy. Thus, alternative model 4 adds a pathfrom INTERACT to BOUND and alternativemodel 5 adds a path from INTERACT to BELIEF.Although the path coeYcient and X2 diVerence teston alternative 5 are insigniWcant, there is supportfor accepting model 4. The interactive system posi-tively inXuences the emphasis Wrms place on theboundary system (p < 0.10) and the X2 diVerencetest is signiWcant (p < 0.10), indicating that alterna-tive model 4 is better-Wtting than the trimmed basemodel.

Finally, alternative model 6 trims the mediatingeVects of ATTEN and LEARN. This model allowsthe control systems to directly aVect performance.


The only signiWcant path is between DIAG andPERF (p < 0.01) indicating that diagnostic use ofthe PM system is signiWcantly associated with per-formance. However, the X2 diVerence test is signiW-cant (p < 0.10) indicating that the model has beenoverly trimmed and thus, the base model is betterWtting.

In summary, I conclude that alternative model 4in Table 6 is a better-Wtting model then the basemodel. The other alternative models are not better-Wtting than the base model and are thus rejected.The signiWcant paths in alternative model 4 aredepicted graphically in Fig. 2.

Validity tests for statistical robustness

In order to ensure that the SEM model resultsare robust I perform several validity tests. First, Idemonstrate the robustness of the relations withperformance. The measure of performance is per-ceptual; obtained from the survey respondents. Icalculate an archival measure of performance,return-on-assets (ROA). I use ROA as the archivalproxy since it may closely approximate the surveymeasure of performance which asks about perfor-mance relative to the organization’s goals duringthe past year. I run the SEM model using ROAinstead of PERF and Wnd that the statistical infer-ences are unchanged although the p-value for bothof the paths: LEARN to PERF and ATTN toPERF, are marginally signiWcant at p < 0.10. Thatthe relation is somewhat weaker is not surprisingsince there may be lags between when the eVects ofthe relations contained in the LOC model have aneVect on ROA. The archival measure of perfor-mance is skewed, thus I use bootstrapping, whichdoes not assume normality, to estimate the stan-dard errors.

Second, I replace the summated variables withfactor scores and Wnd that the statistical modelinferences are unchanged. Third, questions aboutthe extent to which the PM system enable discus-sion and enable continual debate are perhaps morecharacteristic of an interactive system althoughthey load on the diagnostic factor (Bisbe et al.,2007). I remove these two questions from the diag-nostic construct and re-run the SEM analyses. ThesigniWcance and direction of results are not

changed; therefore, I conclude that the inclusion ofthese two questions is not driving the results.Fourth, although earlier statistical checks sug-gested that normality of the model variables doesnot appear to be a concern, a strict assumption ofmaximum likelihood estimation of an SEM modelis multivariate normality. I perform an additionalrobustness check as suggested by Kline (1998) anduse bootstrapping (200 samples with replacement),which does not assume multivariate normality toestimate p-values. I Wnd that the results are qualita-tively unchanged. Finally, Kline (1998) states thata sample size of <100 may be problematic.Although, the sample size in this study is 122,which is within the guidelines stated by Hair,Anderson, Tatham, and Black (1995) and Kline(1998), I also estimate two sub-models as follows:(1) antecedents of control systems and (2) out-comes of control systems. This increases the num-ber of observations per parameter estimated to beabove the threshold of Wve as recommended byHair et al. (1995). The statistical inferences areunchanged.

Conclusions

This study provides evidence on the LOCframework. It Wnds that two types of strategic ele-ments – strategic uncertainties and strategic risk –drive the importance and role of control systems. Italso documents that each of the diagnostic andbeliefs systems facilitate the eYcient use of man-agement attention, while the interactive systemconsumes management attention (i.e., a “cost” ofcontrol). Organizational learning is enhanced byemphasis on the beliefs system as well as use of thediagnostic system. Both organizational learningand attention are positively associated with perfor-mance. Finally, it Wnds that the interactive systeminXuences the diagnostic and boundary systemsand the beliefs system inXuences each of the threeother systems.

Similar to most studies, there are limitations.This study relies on survey data from 122 respon-dents. Steps were taken to ensure the reliability ofthe data (i.e., random sample, pre-test of instru-ment, construct and content validity). All diagnostic


tests show that there is no reason to expect bias(common method test, non-response bias); more-over, I also use an archival measure of performanceto demonstrate the robustness of the results. How-ever, measures may be noisy and caution should betaken when generalizing the results to other popula-tions. In addition, this study relies on cross-sec-tional data. Although the relations in the pathmodel are substantiated by underlying theory,cause-and-eVect relations cannot be demonstratedempirically. In spite of the limitations, this studyresults in Wve implications for theory and practice.

First, this study demonstrates that many of thecontrols in the LOC framework are inter-depen-dent and complementary. SpeciWcally, the resultsshow that the interactive system is inter-dependentwith both the boundary system and the diagnosticuse of performance measures, the latter of which isconsistent with Henri (2006) who argues thatdynamic tension results from the use of perfor-mance measures in dual roles. An important impli-cation for organizations is that in order to realizethe full beneWts of the performance measurementsystem they must use them both diagnostically andinteractively. The Wndings are also consistent withChenhall and Morris (1995) who argue that struc-ture is necessary for interactive type controls to beeVective. In this study, the diagnostic system pro-vides the structure that enables the interactive sys-tem to be eVective since the diagnostic system is amechanism by which the employees learn of thenew strategy and consequently, the new goals andobjectives with which to align behavior. Theboundary system also provides structure throughdelineating the areas oV-limit to employees. I alsoWnd that use of the beliefs system positively inXu-ences all other systems. This is consistent withPearce and David (1987) who state that the beliefssystem provides the foundation for the Wrm’s iden-tity and value system. The positive relationsbetween the beliefs system and each of the threeremaining control systems positively aVect organi-zational outcomes implying that the relations arecomplementary.

The total standardized eVect of the four controlsystems on performance are 0.514; however, if thecontrol systems are isolated by removing the pathsamong them, the total standardized eVect of the

four control systems on performance decreases to0.328 (the Chi-squared diVerence test is signiWcantat p < 0.01). This result suggests that managersmust consider all four control systems whendesigning their control system. It also providesempirical evidence that the control systems arecomplementary. This is consistent with Simons(2000) who argues that an eVective control system,comprised of the four control levels working inharmony and balance, facilitates organizationalperformance.

Second, strategy not only drives the importanceof controls, but also the role of controls. I Wnd thattwo types of strategic uncertainties (competitiveuncertainty and operational uncertainty) are asso-ciated with the importance of control systems.Examining the standardized coeYcients for strate-gic uncertainties shows that operational uncertain-ties have the largest eVect on the diagnostic andbeliefs systems, while competitive uncertaintiesdrives interactive controls. This implies that theinteractive control system is used to scan the exter-nal environment while the other systems arefocused more on the internal environment. I alsoWnd that the Wrm uses the PM system both diag-nostically and interactively to manage operationalrisk. These results are consistent with Galbraith(1973) who says that Wrms implement mechanismsto process information; as uncertainty increasesthe information deWcit increases leading toincreased reliance on mechanisms that facilitatethe processing of information. The results suggestthat there are relatively precise measures of opera-tional risk and uncertainty since both are useddiagnostically. However, it is likely that measuresof competitive uncertainty are relatively less pre-cise since competitive uncertainty is managed withan interactive system. Both the diagnostic andinteractive systems are used to manage operationalrisk, which implies not only that the measures ofoperational risk are relatively precise, but that theinformation deWcit associated with operationalrisk is such that Wrms need both systems to eVec-tively manage it. The results also indicate thatorganizations may implement other types of con-trol systems outside the scope of this study to eVec-tively manage asset impairment and technologicalrisk, which future research could investigate.


Third, the interactive use of the PM system isnot associated with organizational learning. Thereis a signiWcant bi-variate correlation between orga-nizational learning and use of the interactive sys-tem. In addition, if paths between the othersystems and organizational learning are restrictedto zero in the structural equation model then thepath between the interactive use of the PM systemand organizational learning is positive and signiW-

cant. The empirical results in this paper suggestthat it is the structured, formal process of the diag-nostic system that brings to life the beneWts of theinteractive system. This Wnding illustrates theimportance of studying multiple control systems.Studies that focus only on interactive controls maycontend that organizational learning is enhanced;however, when controlling for other control sys-tems (i.e., beliefs and diagnostic controls), thedirect link between the interactive system andlearning does not contain any additional explana-tory power. Rather, the interactive system aVectslearning through the diagnostic system.

Fourth, although there is a cost of control, over-all, the four control components have a positiveimpact on performance, with a total standardizedeVect of 0.514. The results demonstrate that there isa cost associated with the interactive use of perfor-mance measures since it consumes managementattention. However, the net eVect of the four con-trols on attention is positive. Moreover, if theinteractive system is removed from the controlenvironment, the total standardized eVect of theremaining control systems on performance falls to0.438. Therefore, the beneWts of the control systemsoutweigh the costs.

Finally, this study shows that emphasis on con-trol systems inXuences performance through theiraVect on learning and management attention. Sensi-tivity tests demonstrate that the direct relationbetween control systems and performance is weak;however, the eVects become apparent when LEARNand ATTEN are included in the model. This is con-sistent with Luft and Shields (2003) who state that ifthe relation between two variables is weak thenincluding appropriate intervening or mediating vari-ables can help researchers detect eVects. Thus thispaper presents a more complete model of the rela-tion between control systems and Wrm performance.

Acknowledgement

I thank Henri Dekker, Anne Lillis, Frank Selto,and two anonymous reviewers for their helpfulcomments. I thank participants at the 2006 Man-agement Accounting Conference in Tampa, FL,the 2006 Lonestar Accounting Research Confer-ence at Texas A&M University, the 2006 EuropeanAccounting Association Conference in Dublin, Ire-land, and workshop participants at the Universityof Colorado. I gratefully acknowledge the JonesGraduate School of Management for underwritingthe survey costs.

Appendix 1. Examples of beliefs and boundary control systems

The following examples are direct quotes takendirectly from Infor’s and Manugistics corporatewebsites at the time of the development of this pro-ject in 2005/2006.

Example of Infor (formerly Mapics) mission statement

Infor’s mission is to solve the essential, industry-speciWc challenges that others cannot. We accom-plish this mission through: Our solutions – Infor’svertically focused solutions have been developed,enhanced, and proven eVective through decades ofpractical application. Our people – the average Inforconsultant has an advanced degree and over 10yearsof experience in the industries we serve. Our reach –Infor has the global infrastructure and expertise tomanage the complexity of an increasingly interna-tional marketplace. Our passion – we are committedto making you successful by being the very bestsoftware provider for the industries we serve. Ourvalues – we treat our customers, employees, andpartners with honesty, respect, and integrity.

Example of Infor (formerly Mapics) vision statement

Infor’s vision is to become the leading globalsoftware provider in all of the markets we serve.Through on-going investments in products and peo-ple, an unwavering commitment to our customers,


and execution of an aggressive worldwide growthstrategy, we will continually enhance our ability toaddress the specialized, complex, and increasinglyglobal requirements of select manufacturing anddistribution industries.

Example of Manugistics Code of Conduct

This Code of Conduct clariWes the responsibili-ties that Manugistics employees and oYcers haveto each other, to our clients, and to our communi-ties. It helps us understand the responsibilities weshare, and alerts us to important legal and ethicalissues that may arise. You will not Wnd everyManugistics rule, policy or standard here. Whatyou will Wnd are the basic values of how we chooseto do business. The spirit of this Code of Conductgoverns the interpretation of any other policies,guidelines or rules adopted by Manugistics. Allemployees and oYcers of Manugistics Group, Inc.(including its subsidiaries) are responsible for con-ducting themselves in compliance with this Codeof Conduct, other Company policies, and applica-ble laws and regulations. In addition, all membersof the Board of Directors, in regard to their Manu-gistics duties, are responsible for conducting them-selves in compliance with applicable provisions ofthis Code of Conduct and other Company policies,and applicable laws and regulations.

Example of speciWc boundary taken from Manugistics’ Code of Conduct

You [a Manugistic employee] may not acceptany discount or other preferential treatment thatyou know has been oVered to you personallybecause of your position with Manusgistics . . .

Appendix 2. Abbreviated survey questions

Technological strategic uncertainties

Q2. To what extent does top management inyour organization monitor the following strategicuncertainties in order to ensure that the goals ofthe Wrm are achieved (1 D to a small extent, 7 D to alarge extent):

(a) Changes in product technology that aVectthe relative cost/eYciency to user.

(b) New technology.

Competitive strategic uncertainties


(d) Product introductions in adjacent industries.(k) Market tactics of competitors.(l) New industry entrants.

Operational strategic uncertainties


(e) DiVusion of proprietary knowledge outsidethe org.

(g) Scale eVects (product depth).(h) Scope eVects (product breadth).(i) Input costs.(j) Internal product innovation.

Operational risk

Q1. To what extent are the following factorscritical to achieving your organization’s strategy?(1 D to a small extent, 7 D to a large extent):

(a) The safety of our operations.(b) The quality of our operations.(c) The reliability of our operations.(d) The eYciency of our operations.

Competitive risk

Q4. To what extent do Wrms enter your industryeasily? (reverse coded (RC)) (1 D very easy, 7 Dvery diYcult).


Q5. To what extent is it diYcult for a customerto leave your Wrm and begin a relationship with anew Wrm in your industry? (RC) (1 D very easy tochange Wrms, 7 D very diYcult to change Wrms).

Q7a. To what extent is your competition frag-mented (i.e., fragmented is one in which many Wrmshold small relative market share) (1 D to a smallextent, 7 D to a large extent).

Product risk

Q6. To what extent is it diYcult for your Wrm toleave one supplier and begin a relationship withanother supplier? (1 D very easy to change suppli-ers, 7 D very diYcult to change suppliers).

Q7b. To what extent is your Wrm concernedabout the threat of substitute products? (1 D to asmall extent, 7 D to a large extent).

Beliefs system

Q10. Please indicate the extent to which the fol-lowing items describe your organization (1 D notdescriptive, 7 D very descriptive):

(a) Our mission statement clearly communicatesthe Wrm’s core values to our workforce.

(b) Top managers communicate core values toour workforce.

(c) Our workforce is aware of the Wrm’s core val-ues.

(d) Our mission statement inspires our work-force.

Boundary system

Q11. Please rate the extent to which you agreeor disagree with the following (1 D strongly dis-agree (SD), 7 D strongly agree (SA)):

(a) Our Wrm relies on a code of business conductto deWne appropriate behavior for our work-force.

(b) Our code of business conduct informs ourworkforce about behaviors that are oV-limits.

(c) Our Wrm has a system that communicatesto our workforce risks that should beavoided.

(d) Our workforce is aware of the Wrm’s code ofbusiness conduct.

Diagnostic control system

Q12. Please rate the extent to which your topmanagement team currently relies on performancemeasures (PM), or performance measurement sys-tem to (1 D to a small extent, 7 D to a large extent):

(a) Track progress towards goals.(b) Monitor results.(c) Compare outcomes to expectations.(d) Review key measures.(e) Enable discussion in meetings of superiors,

subordinates and peers.(f) Enable continual challenge and debate of

underlying data, assumptions, and actionplans.

(g) Provide a common view of the organization.(h) Tie the organization together.(i) Enable the organization to focus on common

issues.(j) Enable the organization to focus on critical

success factors.(k) Develop a common vocabulary in the orga-

nization.

Interactive control system

Q13. Please indicate the extent to which youagree or disagree with the following statements (1SD, 7 D SA):

(a) Top management pays little day-to-dayattention on the PM system (RC).

(b) Top management relies heavily on staV spe-cialists in preparing and interpreting infor-mation from the PM system (RC).

(c) Operating managers are involved infre-quently and on an exception basis with thePM system (RC).

(d) Top management pays day-to-day attentionto the PM system.

(e) Top management interprets informationfrom the PM system.

(f) Operating managers are frequently involvedwith the PM system.


Organizational learning


(a) Learning is the key to improvement.(b) Basic values include learning as a key to

improvement.(c) Once we quit learning we endanger our future.(d) Learning is viewed as an investment, not an

expense.

Management attention


(e) The control systems in place allow top man-agement to focus attention on critical issues.

(f) The control systems in place allow top man-agement to eVectively leverage their time.

(g) The control systems in place reduce the needfor top managers to constantly monitor Wrmactivities.

(h) Without our control systems the attention oftop managers would be spread more thinly.

Performance

Q9. For each performance indicator, circle thenumber that best indicates the degree of confor-mance to your organization’s goals over the pastyear on (1 D very poor performance, 4 D met goals,7 D very good performance):

(a) Overall organizational performance.(b) Overall organizational proWtability.(c) Relative market share for primary products.(d) Overall productivity of the delivery system.

References

Abernethy, M. A., & Brownell, P. (1999). The role of budgets inorganizations facing strategic change: an exploratory study.Accounting, Organizations and Society, 24, 189–204.

Analoui, F., & Karami, A. (2002). CEOs and development of themeaningful mission statement. Corporate Governance, 13–20.

Anderson, S. W., & Dekker, H. C. (2005). Management controlfor market transactions: the relation between transactioncharacteristics, incomplete contract design, and subsequentperformance. Management Science, 51, 1734–1752.

Arbuckle, J. L., & Wothke, W. (1999). AMOS 4.0 user’s guide.Chicago, IL: SmallWaters Corporation.

Argyris, C. (1977). Organizational learning and managementinformation systems. Accounting, Organizations and Society,2, 113–123.

Banker, R. D., & Datar, S. M. (1989). Sensitivity, precision, andlinear aggregation of signals for performance evaluation.Journal of Accounting Research, 27, 21–39.

Bart, C. K., Bontis, N., & Taggar, S. (2001). A model of theimpact of mission statements on Wrm performance. Manage-ment Decision, 39, 19–36.

Bentler, P. M. (1990). Comparative Wt indexes in structuralmodels. Psychological Bulletin, 107, 238–246.

Bisbe, J., Batista-Foguet, J., & Chenhall, R., (2007). DeWningmanagement accounting constructs: a methodological noteon the risks of conceptual misspeciWcation. Accounting, Orga-nizations and Society, in press. doi:10.1016/j.aos.2006.09.010.

Bisbe, J., & Otley, D. (2004). The eVects of the interactive use ofmanagement control systems on product innovation.Accounting, Organizations and Society, 29, 709–737.

Bromwich, M. (1990). The case for strategic managementaccounting: the role of accounting information for strategyin competitive markets. Accounting, Organizations and Soci-ety, 15, 27–46.

Browne, M. W., & Cudeck, R. (1993). Alternative ways ofassessing model Wt. In K. A. Bollen & J. S. Long (Eds.), Test-ing Structural Equation Models. Newbury Park, CA: Sage.

Burney, L. L., & Widener, S. K. (2007). Strategic performancemeasurement systems, job-relevant information and mana-gerial behavioral responses – role stress and performance.Behavioral Research in Accounting, forthcoming.

Calfee, D. L. (1993). Get your mission statement working. Man-agement Review (January), 54–57.

Chenhall, R. H. (2003). Management control systems designwithin its organizational context: Wndings from contin-gency-based research and directions for the future. Account-ing, Organizations and Society, 28, 127–168.

Chenhall, R. H. (2005). Integrative strategic performance mea-surement systems, strategic alignment of manufacturing,learning and strategic outcomes: an exploratory study.Accounting, Organizations and Society, 30, 395–422.

Chenhall, R. H., & Morris, D. (1995). Organic decision andcommunication processes and management accounting sys-tems in entrepreneurial and conservative business organiza-tions. Omega, 23, 485–497.

Churchill, G. A. (1979). A paradigm for developing better mea-sures of marketing constructs. Journal of MarketingResearch, 16, 64–73.

Dent, J. (1990). Strategy, organization and control: some possi-bilities for accounting research. Accounting, Organizationsand Society, 15, 3–25.

http://dx.doi.org/10.1016/j.aos.2006.09.010

http://dx.doi.org/10.1016/j.aos.2006.09.010


DeVellis, R. F. (1991). Scale development: Theory and applica-tions. Newbury Park, CA: Sage Publications.

Dillman, D. A. (1978). Mail and telephone surveys: The totaldesign method. New York: John Wiley & Sons.

Fornell, C., & Larcker, D. F. (1981). Evaluating structural equa-tion models with unobservable variables and measurementerror. Journal of Marketing Research, 18, 39–50.

Fullerton, R. R., & McWatters, C. S. (2002). The role of perfor-mance measures and incentive systems in relation to thedegree of JIT implementation. Accounting, Organizationsand Society, 27, 711–735.

Galbraith, J. (1973). Designing complex organizations. Reading,MA: Addison-Wesley Publishing Company.

Galer, G., & Van Der Heijden, K. (1992). The learning organi-zation: How planners create organizational learning. Mar-keting Intelligence and Planning, 10, 5–12.

Hair, J. F. Jr., Anderson, R. E., Tatham, R. L., & Black, W. C.(1995). Multivariate data analysis. Englewood CliVs, NJ:Prentice Hall.

Henri, J. (2006). Management control systems and strategy: aresource-based perspective. Accounting, Organizations andSociety, 31, 529–558.

Hopwood, A. (1987). The archaeology of accounting systems.Accounting, Organizations and Society, 12, 207–234.

Hult, G., & Thomas, M. (1998). Managing the internationalstrategic sourcing process as a market-driven organizationallearning system. Decision Sciences, 29, 193–216.

Ittner, C., & Larcker, D. (1997). Quality strategy, strategic con-trol systems, and organizational performance. Accounting,Organizations and Society, 22, 293–314.

Ittner, C., & Larcker, D. (1998). Innovations in performancemeasurement: trends and research implications. Journal ofManagement Accounting Research, 10, 205–238.

Joreskog, K. G. (1969). A general approach to conWrmatory max-imum likelihood factor analysis. Psychometrika, 34, 183–202.

Kaplan, R. S., & Norton, D. P. (1996). The balanced scorecard.Boston, MA: Harvard Business School Press.

Kennedy, F., & Widener, S. K. (2006). A control framework:Insights from evidence on lean accounting. Working Paper.Clemson and Rice Universities.

Khandwalla, P. (1972). The eVect of diVerent types of competi-tion on the use of management controls. Journal of Account-ing Research, 10, 275–285.

Kline, R. B. (1998). Principles and practice of structural equationmodeling. NY: Guilford Press.

Kloot, L. (1997). Organizational learning and management con-trol systems: responding to environmental change. Manage-ment Accounting Research, 8, 47–73.

LangWeld-Smith, K. (1997). Management control systems andstrategy: a critical review. Accounting, Organizations andSociety, 22, 207–232.

Levitt, B., & March, J. G. (1988). Organizational learning.Annual Review of Sociology, 14, 319–340.

Luft, J., & Shields, M. D. (2003). Mapping managementaccounting: graphics and guidelines for theory-consistentempirical research. Accounting, Organizations and Society,28, 169–249.

Marginson, D. E. (2002). Management control systems andtheir eVects on strategy formation at middle-managementlevels: evidence from a UK organization. Strategic Manage-ment Journal, 23, 1019–1031.

Merchant, K. A. (1990). The eVects of Wnancial controls on datamanipulation and management myopia. Accounting, Orga-nizations and Society, 15, 297–313.

Merchant, K. A., & Otley, D. T. (2006). A review of the litera-ture on control and accountability. In C. Chapman, A. Hop-wood, & M. Shield (Eds.), The handbook of managementaccounting research. Elsevier Press.

Mia, L., & Clarke, B. (1999). Market competition, managementaccounting systems and business unit performance. Manage-ment Accounting Research, 10, 137–158.

Milgrom, P., & Roberts, J. (1995). Complementarities and Wt:strategy, structure and organizational change in manufac-turing. Journal of Accounting and Economics, 19, 179–208.

Nunnally, J. C. (1978). Psychometric theory (2nd ed.). NewYork, NY: McGraw-Hill Book Company.

Otley, D. T. (1978). Budget use and managerial performance.Journal of Accounting Research, 16, 122–149.

Otley, D. T. (1980). The contingency theory of managementaccounting: achievements and prognosis. Accounting, Orga-nizations and Society, 5, 413–428.

Otley, D. T. (1999). Performance management: a framework formanagement control systems research. ManagementAccounting Research, 10, 363–382.

Paine, L., Deshpande, R., Margolis, J. D., & Bettcher, K. E. (2005).Up to code: does your company’s conduct meet world-classstandards? Harvard Business Review (December), 1–12.

Pearce, J. A., II (1982). The company mission as a strategic tool.Sloan Management (Spring), 15–24.

Pearce, J. A., II, & David, F. (1987). Corporate mission state-ments: the bottom line. Academy of Management Executive,1, 109–116.

PodsakoV, P. M., & Organ, D. W. (1986). Self-reports in organi-zational research: problems and prospects. Journal of Man-agement, 12, 531–544.

Porter, M. E. (1980). Competitive strategy. New York: The FreePress.

Roth, A. V., & Jackson, W. E., III (1995). Strategic determinantsof service quality and performance evidence from the bank-ing industry. Management Science, 41, 1720–1733.

Schick, A., Gordon, L., & Haka, S. (1990). Information over-load: a temporal approach. Accounting, Organizations andSociety, 199–220.

Simons, R. (1990). The role of management control systems increating competitive advantage: new perspectives. Account-ing, Organizations and Society, 15, 127–143.

Simons, R. (1991). Strategic orientation and top managementattention to control systems. Strategic Management Journal,12, 49–62.

Simons, R. (1994). How new top managers use control systemsas levers of strategic renewal. Strategic Management Jour-nal, 15, 169–189.

Simons, R. (1995). Levers of control. Boston, MA: HarvardBusiness School Press.


Simons, R. (2000). Performance measurement and control sys-tems for implementing strategy. Upper Saddle River, NJ:Prentice Hall.

Slater, S. F., & Narver, J. C. (1995). Market orientation and thelearning organization. Journal of Marketing, 59, 63–74.

Tippins, M. J., & Sohi, R. S. (2003). IT competency and Wrmperformance: Is organizational learning a missing link?Strategic Management Journal, 24, 745–761.

Tuomela, T. (2005). The interplay of diVerent levels of control: acase study of introducing a new performance measurementsystem. Management Accounting Research, 16, 293–320.

Wheaton, B., Muthen, B., Alwin, D. F., & Summers, G. F.(1977). Assessing reliability and stability in panel models. InD. R. Heise (Ed.), Sociology methodology. San Francisco:Jossey-Bass.

Widener, S. K., Shackell, M. B., & Demers, E. A. (2006). Insightson performance measures: an investigation of relationsamong performance measures, delegation, incentives, andstrategy. Working paper. Rice University, University ofNotre Dame, and INSEAD.

sally widener aos 2007

Documents