Determinants of Control System Design in Divisionalized Firms
Author(s): Margaret A. Abernethy, Jan Bouwens and Laurence van Lent
Source: The Accounting Review, Vol. 79, No. 3 (Jul., 2004), pp. 545-570 Published by: American Accounting Association Stable URL: http://www.jstor.org/stable/3203269 . Accessed: 03/09/2014 09:38 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. . American Accounting Association is collaborating with JSTOR to digitize, preserve and extend access to The Accounting Review. http://www.jstor.org This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions THE ACCOUNTING REVIEW Vol. 79, No. 3 2004 pp. 545-570 Determinants of Control S y stem Desig n in Divisionalized F irms Marg aret A. Abernethy University of Melbourne Jan Bouwens Ny enrode University Laurence van Lent Tilburg University ABS TRACT: We investig ate two determinants of two choices in the control sy stem of divisionalized f irms, namely decentralization and use of perf ormance measures. The two determinants are those identif ied in the literature as important to control sy stem desig n: (1) inf ormation asy mmetries between corporate and divisional manag ers and (2) division interdependencies. We treat decentralization and perf ormance measure- ment choices as endog enous variables and examine the interrelation among these choices using a simultaneous equation model. Using data f rom78 divisions, our results indicate that decentralization is positively related to the level of inf ormation asy mme- tries and neg atively to intraf irm interdependencies, while the use of perf ormance mea- sures is af f ected by the level of interdependencies among divisions within the f irm, but not by inf ormation asy mmetries. We f ind some evidence that decentralization choice and use of perf ormance measures are complementary . Key words: org anization desig n; control sy stem; interdependencies; inf ormation asy m- metry ; perf ormance measures. JEL Classif ication: D23; D82; L22; M12; M4. I. INTRODUCTION Aconsiderable body of research exists on the use of perf ormance measurement sy s- tems (Ittner and Larcker 2001; Luf t and S hields 2003). However, there are f ew attempts to understand the relation between the use of these perf ormance measures and other control choices. We provide evidence on this relation by examining two choices We thank Willem Buijink, Peter Easton, Chris Ittner, Ranjani Krishnan, Christian Leuz, Anne Lillis, Venky Nag ar, David Otley , Mark Penno, F rank S elto, Peter Wy socki, and workshop participants at the Universities of Melbourne, Ny enrode, New S outh Wales (S y dney ), and Tilburg f or helpf ul comments. The authors also benef itedf rom sug - g estions during the 2001 EAA annual meeting in Athens, andthe 2002 AAA Manag ement Accounting S ection meeting . The editor and ref erees of The Accounting Review have provided most constructive comments on the paper. This study was supportedby g rants f romthe NIVRA andthe Australian Research Council. Editor's note: This paper was acceptedby Terry S hevlin, S enior Editor. S ubmitted Aug ust 2002 Accepted December 2003 545 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent made by corporate manag ement in the control sy stem of divisionalized f irms: (1) decen- tralization and (2) use of divisional summary perf ormance measures. Decentralization oc- curs when corporate manag ement assig ns decision rig hts to lower-level manag ers. We op- erationalize decentralization as the extent to which decision-making authority over key decisions is vested with divisional manag ers (compared with their superiors). Divisional summary perf ormance measures (DS Ms) have two def ining characteristics. F irst, they sum- marize the action choices of divisional manag ers into one perf ormance metric (e.g ., prof it, ROI) and, second, they measure perf ormance at the divisional level. Divisional perf ormance can also be measured using specif ic, non-summary metrics and f irm-level summary mea- sures. S pecif ic, non-summary metrics provide inf ormation about particular activities carried out within the division (e.g ., quality statistics or delivery targ ets) and theref ore do not ref lect the overall perf ormance of a divisional manag er, while f irm-level summary measures cap- ture overall f irmperf ormance rather than divisional-level perf ormance (e.g ., f irmROI or stock price). We operationalize the use of DS Ms as the weig ht these measures receive relative to f irm-level summary measures and specif ic, non-summary measures when a su- perior assesses a division manag er's perf ormance. While a f irmmakes multiple control choices, we examine DS Ms and decentralization choices f or several reasons. F irst, they are considered two of the most important desig n choices made by corporate manag ement in divisionalized f irms (Ittner and Larcker 2001; Luf t and S hields 2003; Jensen and Meckling 1992).' S econd, the two choices are tig htly coupled. Jensen (2001) arg ues that corporate manag ement will only decentralize when it can implement a perf ormance measurement sy stem that captures the decision rig hts allo- cated to divisional manag ers. DS Ms are desig ned to summarize the decisions over which a divisional manag er has responsibility . They provide corporate manag ement with the pos- sibility to assess the contribution of the division to f irmvalue and to determine if divisional manag ers are using decision rig hts optimally (S olomons 1965; Zimmerman 1997). F inally , we f ocus on DS Ms, as these measures are present in the majority of divisionalized f irms (Vancil 1979; Homg ren et al. 1999). Drawing on prior literature, we predict that decentralization and the use of DS Ms are complementary choices and that these choices are jointly determined by : (1) inf ormation asy mmetries between corporate manag ement and divisional manag ers and (2) interdepen- dencies among divisions. Our study contributes to the literature on the economics of or- g anization desig n by extending and integ rating prior empirical research (Baiman et al. 1995; Bushman et al. 1995; Keating 1997; Nag ar 2002). Earlier studies examined decentralization in relation with compensation sy stems, but did not consider perf ormance measurement issues (e.g ., Nag ar 2002; Baiman et al. 1995). Those studies that f ocus on perf ormance measurement sy stems have either ig nored their relation with decentralization (Keating 1997) or have assumed decentralization to be exog enously determined (Abernethy and Lillis 2002; Gordon and Naray anan 1984; Chenhall and Morris 1986). Yet theoretical work arg ues that the decentralization and perf ormance measurement are complementary control choices made by manag ement; that is, they are endog enous (e.g ., Milg rom and Roberts 1992). To Jensen and Meckling (1992) include incentive sy stems as the third key control choice. Milg rom andRoberts (1992, 403), however, do not make a distinction between incentive sy stems and perf ormance measurement sy stems as they consider perf ormance measurement sy stems as "the heart of any incentive sy stem." Our approach is consistent with Milg rom andRoberts (1992) approach. We f ocus on one aspect of incentive sy stems, namely , perf ormance measurement sy stems, while Nag ar (2002) and others f ocus on the variable pay component of incentive sy stems. As such, our study represents a partial equilibriumanaly sis of decentralization and perf orm- ance measurement choices. The Accounting Review, July 2004 546 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stem Desig n in Divisionalized F irms the best of our knowledg e this is the f irst study to examine the interrelation between per- f ormance measurement sy stem and decentralization choices. A second contribution is the inclusion of both inf ormation asy mmetry and interdependencies into the model. While both of these determinants have been identif ied as important in prior research (Baiman et al. 1995; Bushman et al. 1995; Keating 1997), ours is the f irst study that has assessed their joint ef f ect on control sy stem choices. The study also contributes methodolog ically . In recent y ears there has been serious criticismnot only about the use of inappropriate proxies to capture theoretical constructs, but also about the lack of consistency between the levels of theory , analy sis, and measure- ment (Ittner and Larcker 2001; Luf t and S hields 2003; Lanen 1995). Consistent with our hy potheses development, we measure the constructs at the divisional level. We use either prior established survey instruments or, where necessary , develop measures to overcome validity concerns with available measures. We address concerns relating to reliability and validity of each of our constructs by avoiding the use of "perceptive" measures (i.e., we ask respondents to provide specif ic data on org anizational practices) and by collecting additional data to test the construct validity of each of the constructs. We examine the relation between decentralization and the use of DS Ms using a si- multaneous equation model. This approach allows us to treat decentralization and the use of DS Ms as control choices (i.e., as endog enous variables). We attempt to reduce concerns about the endog eneity of our main independent variables (i.e., interdependencies and in- f ormation asy mmetry ) by our sample selection procedures and additional tests. We test our model using survey data f roma representative sample of divisionalized f irms listed on the AmsterdamS tock Exchang e. We choose divisionalized f irms because this enables us to treat interdependencies as exog enous. The decision to create a divisionalized structure is g enerally considered to precede the desig n and implementation of a f irm's control sy stem. As arg ued by Milg rom and Roberts (1992, 546-549) f irms initially def ine divisions to "minimize the connections among them" and to "avoid the need f or co-ordination across divisional boundaries." Once the divisional structure is implemented, however, interdepen- dencies will remain. Our interest is how these interdependencies are manag ed throug h org anizational desig n choices. We also explore the ef f ect of relaxing the exog eneity as- sumption with reg ard to inf ormation asy mmetry and show that our orig inal results are robust. The simultaneous equation model, tog ether with the sample selection procedures, beg ins to address some of the endog eneity problems in manag ement accounting research. However, we recog nize that some problems are likely to remain and our results should be interpreted according ly . Our f inding s indicate that decentralization is positively related to the level of inf or- mation asy mmetries and neg atively to intraf irm interdependencies. On the other hand, the use of DS Ms is only af f ected by divisional interdependencies. We f ind some evidence that the relation between decentralization choices and use of DS Ms is complementary . Corporate manag ement relies more on DS Ms when they have deleg ated authority to divisions. On the other hand, we f ind no evidence that a g reater use of DS Ms af f ects decentralization. The f inding s presented here allow us to increase our rather limited understanding of the complex interrelation between control choices as well as the context within which those choices are made (Luf t and S hields 2003). The remainder of this paper is structured as f ollows. S ection II provides the theoretical basis f or the empirical model tested. We describe the data and research method in S ection III, and present the results of our tests in S ection IV. In S ection V we provide a summary and concluding remarks. The Accounting Review, July 2004 547 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent II. HYPOTHES IS DEVELOPMENT Accounting researchers have devoted considerable ef f ort to establishing the determi- nants of the use of perf ormance measures (Ittner and Larcker 2001). We investig ate the f actors that inf luence the importance placed on DS Ms, as opposed to f irm-level summary measures (e.g ., f irm-wide prof it) or specif ic, non-summary measures (e.g ., delivery targ ets).2 We are interested in the relation between decentralization choices and the use of DS Ms by corporate manag ement f or evaluating divisional manag ers. We draw on Milg rom and Roberts (1992, 549) and others (Melumad et al. 1992; Ag hion and Tirole 1997; Jensen and Meckling 1992) to arg ue that g ranting decision rig hts and using DS Ms are complementary actions. We thus expect that the choice to decentralize and corporate manag ement's use of DS Ms will be jointly determined and that the relation between the two will be positive. The f ollowing two equations summarize our conceptual model. Divisional summary measures (DS Ms) = f (decentralization, inf ormation asy mmetry , interdependencies, control variables). Decentralization = f (DS Ms, inf ormation asy mmetry , interdependencies, control variables). This model is used to assess the interrelations between decentralization and DS Ms and to assess the impact of divisional interdependencies and inf ormation asy mmetry on these choices. We complete the model by including a set of control variables in each equation based on prior literature.3 Theoretical justif ication f or the model f ollows. Determinants of the Use of Divisional S ummary Measures (DS Ms) Decentralization Corporate manag ement decentralizes decision making to encourag e divisional manag ers to initiate and implement decisions that will increase f irmvalue. To avoid or mitig ate potential control loss problems, they will implement perf ormance measures that both sum- marize perf ormance into a metric and ref lect the decision rig hts allocated (Jensen 2001; Wruck and Jensen 1994). Divisional summary measures (DS Ms) serve this purpose. They summarize perf ormance based on decision rig hts deleg ated to division manag ers and thus provide a means of monitoring their action choices. Ceteris paribus, DS Ms will be pref erred to other perf ormance metrics. F irm-level summary measures are too noisy because they are not only af f ected by actions taken at the division level, but also by actions taken elsewhere in the f irm (Keating 1997; Bushman et al. 1995). Using specif ic, non-summary measures to monitor divisional manag ers ef f ectively transf ers authority f romdivision manag ers to corporate manag ement. Monitoring divisional perf ormance on measures that ref lect specif ic activities of a division reduces the authority of division manag ers to make trade-of f s among these activities necessary to optimize overall divisional perf ormance (Jensen 2001). In this sense, specif ic, non-summary measures are inconsistent with decentralized decision making . We thus expect that as decentralization increases, corporate manag ement will increasing ly rely on DS Ms to measure divisional perf ormance. 2 Prior literature has assessed the relative importance of f irm summary measures versus divisional summary measures (see Keating 1997) anda number of studies have examinedthe trade-of f s between summary measures (e.g ., prof it, ROI) vis-a-vis other more detailednonf inancial measures (e.g ., quality , turnaroundtimes, delivery targ ets, etc.). While these alternative measures are not the f ocus of our study , we do undertake additional analy sis to assess the impact on f irm summary measures anddivisional specif ic measures when the use of DS Ms chang es. 3 The choice of these control variables also enabledthe model to be identif iedas there was no reason to believe theoretically that the set of variables includedin each equation would be the same. The Accounting Review, July 2004 548 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms Inf ormation Asy mmetry Inf ormation asy mmetry occurs when lower-level manag ers have specif ic knowledg e about the f unctioning of the division that is either not available to corporate manag ement or is too costly f or corporate manag ement to obtain (Christie et al. 2003).4 The use of DS Ms by corporate manag ement assumes either that these measures f ully capture mana- g erial perf ormance or top manag ement has suf f icient knowledg e about the activities per- f ormed by divisions to allow f or the incompleteness of these measures and to detect possible manipulations of the measures. There is some support that demonstrates the reluctance of corporate manag ement to rely on DS Ms when inf ormation asy mmetry exists between cor- porate and divisional manag ement (Reichelstein 1992; Dunk 1993; Rockness and S hields 1984). In this situation these esrse measures are inadequate measures of manag erial ef f orts and their use will lead to dy sf unctional behavior (Chow et al. 1988). We thus expect to observe a neg ative relation between inf ormation asy mmetry and use of DS Ms.5 Interdependencies Interdependencies occur when demand f unctions of divisions are dependent or when divisions have joint supply and cost f unctions (see Milg rom and Roberts 1992, 108-109, 113-116). These interdependencies create a perf ormance measurement problem as they increase the "noisiness" of DS Ms. It is only in a setting where a f irmhas a "perf ect" transf er-pricing sy stem that the impact of intermediate product transf ers between divisions will not inf luence the costs and revenues included in divisional prof it or ROI calculations (Bushman et al. 1995; Keating 1997; Lambert 2001). There are two ty pes of interdepen- dencies af f ecting the use of DS Ms: (1) when the f ocal division is inf luenced by the activities of other divisions and (2) when the f ocal division inf luences the perf ormance of other divisions. As both ty pes of interdependencies increase, the use of DS Ms will decrease, but f or somewhat dif f erent reasons (Lambert 2001; Keating 1997). When the f ocal division's perf ormance is inf luenced by the activities and decisions of other divisions, the measure is "noisy " and thus less inf ormative to corporate manag ement. The measure not only captures activities within the f ocal division, but it also captures actions of the other divisions. S im- ilarly , if a f ocal division's actions and decisions inf luence the activities of another division, then the use of their own DS Ms will encourag e the manag er to take actions that will optimize his/her own perf ormance, rather than encourag e cooperation and optimization of the potential sy nerg ies among the divisions. We, theref ore, expect a neg ative relation be- tween both ty pes of interdependencies and the use of DS Ms. 4 It is also possible that corporate manag ement has specif ic knowledg e that is not available to the division manag er. However, this will not relate to the operating activities of the division. The inf ormation asy mmetry of concern here is inf ormation relating to divisional activities. 5 Hig h levels of inf ormation asy mmetry between corporate anddivisional manag ement will require the use of alternative f orms of control. Control throug h perf ormance measurement is unlikely to be completely ef f ective reg ardless of the characteristics of the perf ormance measurement sy stems. Corporate manag ement will needto resort to less obtrusive f orms of controls such as selection procedures and/or training prog rams as a means of creating a corporate culture that encourag es employ ees to behave in a manner consistent with org anizational g oals (Brickley et al. 1997; Kreps 1990). The Accounting Review, July 2004 549 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent Control Variables: Perf ormance Measurement Characteristics We control f or the ef f ects of the properties of DS Ms as sug g ested by Luf t and S hields (2003). We include two perf ormance measure characteristics, namely sensitivity and pre- cision of the measure.6 Banker and Datar (1989) arg ue that the use of a perf ormance measure is positively related to the sensitivity of the measure to the action choices of a manag er and also positively to the precision of the measure. Datar et al. (2001) question whether sensitivity of a measure is an important determinant and demonstrate that the use of a measure is inf luenced by its importance in achieving org anizational objectives (see also, Lambert 2001). Theref ore, we do not make a prediction f or the sig n of sensitivity , but we expect that the use of DS Ms will be positively related to the precision of the measure. Determinants of Decentralization Use of Divisional S ummary Measures (DS Ms) There is relatively strong theoretical support in the literature that the use of DS Ms and decentralization are complementary choices (Melumad et al. 1992; Jensen 2001; Ag hion and Tirole 1997; Milg rom and Roberts 1995; Jensen and Meckling 1992). S olomons (1965) was one of the f irst to arg ue that decentralization is conditional on the use of perf ormance measures that capture the contribution of a division to f irmvalue (see also, Vancil 1979, 88). DS Ms allow corporate manag ement to g ive divisional manag ers "a substantial deg ree of f reedomin their administration of the resources entrusted to them" (S olomons 1965, 9). More recently , Brickley et al. (1997) describe several cases where innovations in division perf ormance measurement f ail because concurrent decentralization chang es were not made. When the use of DS Ms increases, ceteris paribis, we assume that corporate manag ement is better able to judg e the actions of divisional manag ers. In turn, they will be more willing to deleg ate decision rig hts. In contrast, a reduced use of DS Ms will result in a reduction in the decision rig hts deleg ated to divisional manag ers. Corporate manag ement will make this choice when DS Ms are limited in capturing the perf ormance of divisions-centralizing decision making becomes a viable control alternative (Milg rom and Roberts 1992, 32). We thus expect that the use of DS Ms will be positively related to the level of decentralization. Inf ormation Asy mmetry Ceteris paribus, corporate manag ement will deleg ate decision rig hts to divisional man- ag ers who possess specif ic knowledg e (Jensen and Meckling 1992; Brickley et al. 1997). Conversely , if knowledg e is not impacted at lower levels, corporate manag ement has the inf ormation necessary to select optimal action choices f or the division. There is no net benef it associated with deleg ating decision rig hts (Baiman et al. 1995; Jensen and Meckling 1992). We, theref ore, expect a positive relation between the level of decentralization and the level of inf ormation asy mmetry . Divisional Initerdependencies Interdependencies among divisions can result in neg ative externalities f or the f irmwhen decision making is decentralized. When divisional manag ers are deleg ated decision rig hts ' We describe above how interdependencies andinf ormation asy mmetry af f ect the inf ormativeness of DS Ms and hence their use. It seems reasonable to assume that other f actors af f ect the inf ormativeness of DS Ms as well. Note that Banker and Datar (1989) arg ue that inf ormativeness is a f unction of precision and sensitivity . Our control variables, precision and sensitivity , attempt to capture any remaining dif f erences in the inf ormative- ness (use) of DS Ms, af ter accounting f or the ef f ect on inf ormativeness of inf ormation asy mmetry and interdependencies. The Accounting Review, July 2004 550 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms they have incentives to optimize their own division's perf ormance and ig nore the conse- quences of their actions on other divisions (F isher 1994; Bushman et al. 1995; Christie et al. 2003). Corporate manag ement will attempt to minimize these externalities by central- izing decision making . Ceteris paribus, we expect that the net benef its associated with decentralization will decrease when interdependencies are hig h; that is, we expect the re- lation between interdependencies and decentralization to be neg ative. Control Variables: Division-S pecif ic F actors Prior research arg ues that the operating environment of a division inf luences corporate manag ement's decision to deleg ate decision rig hts (Brickley et al. 1997; Milg rom and Roberts 1995; Nag ar 2002). We, theref ore, control f or specif ic contextual f actors relating to the division itself , namely the g rowth opportunities and size. We do not make a prediction about the sig n of the association f or these two variables due to conf licting arg uments and empirical evidence.7 F ollowing prior research, we also control f or the inf luence of manag erial characteristics on control sy stem desig n (Nag ar 2002). Empirical evidence provides some support that corporate manag ement's choice to deleg ate decision rig hts is inf luenced by the ability of the division manag er.8 We also expect, ceteris paribus, that corporate manag ement will deleg ate decision rig hts only when a certain level of trust is developed between corporate and divisional manag ers.9 Our model controls f or the ef f ect of moral hazard on decentralization choices. While there is some evidence that mitig ates this concern (Baiman et al. 1995; Christie et al. 2003), it is possible that the combination of inf ormation asy mmetry and size results in sig nif icant moral hazard costs. Inf ormation asy mmetry provides the ag ent with the inf ormation to behave opportunistically and size increases the rents ag ents are able to extract f romthis behavior (Milg rom and Roberts 1992, 573). Thus, when both these conditions exist (i.e., larg e levels of inf ormation asy mmetry and larg e size) the cost of decentralization may outweig h the benef its. If this is the case, then we will not observe the main ef f ects of inf ormation asy mmetry and size on decentralization. S ummary of Expectations Table 1 provides a summary of our expectations. III. METHOD, MODEL, AND ECONOMETRIC IS S UES S ample We use a randomly selected sample of f irms listed on the AmsterdamS tock Exchang e. We f irst contacted the f inancial controller to obtain permission to undertake the study and to identif y names of relevant divisions. Once the f irm ag reed to participate, we visited 7Nag ar (2002) arg ues that g rowth opportunities will increase the level of decentralization as this f acilitates the involvement of divisional manag ers in scanning and environment and exploiting the opportunities available. However, others arg ue that corporate manag ement will retain decision-making power to ensure that opportunities f or g rowth are optimized (Ag hion andTirole 1997). S imilar ambig uity exists concerning the ef f ect of size on the decentralization. It is of ten arg ued that size is relatedto the level of decentralization (Bolton and Dewatripont 1995; Vancil 1979; Lawrence andLorsch 1967) and y et empirical evidence is mixed (Miller 1987). 8 F or example, Baker et al. (1994) andothers (Medof f andAbraham 1980) demonstrate that more talented, hig hly educatedindividuals are promoted to hig her-level positions andthus are g iven more responsibilities within the f irm. 9 This relation has not been empirically examined directly . However, research in the manag ement andeconomics literatures provide theoretical support that this would be the case (see Nooteboomet al. 1997; Lorenz 1999). The Accounting Review, July 2004 551 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent TABLE 1 Predictive Ef f ects Investig ated in this S tudy Divisional S ummary Decentralization Measures (DS Ms) Choice Variables Decentralization + Divisional S ummary Measures + Test Variables Inf ormation Asy mmetry + Interdependencies Control Variables Division S pecif ic Characteristics: Growth NP S ize NP Moral Hazard Divisional Manag ers Characteristics: Ability + Trust + Perf ormance Measure Characteristics: S ensitivity NP Precision divisional manag ers within the f irm to administer the questionnaire. This method of data collection is particularly usef ul as it enables us to ensure that we have the e correct respon- dent, that the f irmhas at least two operating divisions (i.e., necessary to test our expectations with respect to interdependencies), that divisions are suf f iciently larg e, and f inally to estab- lish that divisions are autonomous and report to either the Chief Executive Of f icer or Chief Operating Of f icer of the f irm. Our analy sis is based on a sample of 78 divisional manag ers. Twenty -f ive percent of the f irms contacted areed a to participate. Tests f or response bias indicate that our sample is drawn f roma representative set of f irms.'0 Once f irms ag reed to participate, we achieved a 92 percent response rate f rom the divisional manag ers contacted. Measures The measurement of the constructs is discussed in turn. Where appropriate we f ollow Ittner and Larcker's (2001) recommendation and use "harder" data (i.e., more objective data) to support the validity of our measures. Where multiple items are used to measure a construct, we use the f actor score." All Likert-scale variables were standardized bef ore '0 We assessed if the sample was representative of all other f irms listed on the AmsterdamS tock Exchang e. We conductedt-tests (Wilcoxon-tests) f or dif f erences of means (medians) on variables relating to net income, sales, total assets, andmarket value of common equity . There were no sig nif icant dif f erences in the mean (median) values f or any of these variables except f or market value andthis was only sig nif icant at the 0.09 level. l The f actor analy sis used oblique rotation andretainedf actors with an eig envalue g reater than unity (Nunnally andBernstein 1994). The results of the f actor analy sis are not provided in the paper, but are available f romthe authors. Researchers can either compute a composite variable that captures the underly ing construct by averag ing the standardizeditemscores (Nunnally and Bernstein 1994) or use the f actor scores. We use f actor scores; however, the results are robust ag ainst using either method (the two methods produce variables that are correlated at 0.99 in all cases). The Accounting Review, July 2004 552 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms entering theminto the f actor analy sis. The survey instruments used are reproduced in the Appendix. Choice Variables Level of decentralization (DECEN). We measure the level of decentralization using an adapted version of the Gordon and Naray anan (1984) instrument. We attempt to capture what Ag hion and Tirole (1997) ref er to as "real authority " by asking division manag ers to indicate their inf luence relative to their superior's inf luence on a rang e of f ive key decisions (i.e., strateg y , human resource manag ement, operations, marketing , and investments) along a Likert-ty pe scale of 1 to 7 where 1 = the division has all the inf luence and 7 = the superior has all the inf luence. We reverse-code this measure such that hig h values indicate hig h levels of decentralization. We use the results of f actor analy sis, Chi-squared tests, and the Cronbach alpha statistic (0.73) to support the use of the f ive-itemmeasure as a unidi- mensional construct. Use of divisional summary measures (DS Ms). We use Keating 's (1997) notion of own-level measures to capture the use of DS Ms. These measures are those that summarize division perf ormance in a sing le measure. In contrast to Keating (1997), our measure in- cludes both f inancial summary measures (e.g ., prof it, ROI) and other quantitative ef f iciency - ty pe measures (e.g ., output data). We ask respondents to assig n weig hts to the relative importance of DS Ms measures vis-ai-vis other perf ormance metrics.'2 These other metrics include f irm summary measures (both stock-price-related measures and f irm prof itability measures) and measures that provide perf ormance inf ormation on specif ic aspects of per- f ormance within divisions (specif ic, non-summary measures). Tog ether these ty pes of perf ormance measures cover the complete continuumof possible quantitative metrics used in a f irm. To improve the consistency of the responses, we f ollow Ittner and Larcker's (2001) advice and specif y the decision context in which perf ormance measures are used. We ask respondents to indicate the percentag e weig ht (out of a total of 100) the superior assig ns to each ty pe of perf ormance metric when s/he assesses perf ormance. Converg ent validity of the measure is established by the Pearson correlation of DS Ms with an alternative measure captured elsewhere in the questionnaire. This alternative measure asks the superior to assig n the weig ht attached to divisional prof itability measures compared to other ty pes of measures when assessing perf ormance. Prof it measures constitute an important subset of all DS Ms. The correlation between these two measures is 0.33 (p < 0.01). This provides some evidence of the validity of our measure. Test Variables Interdependencies. We adopt Keating 's (1997) instrument to measure interdependen- cies. The f irst itemasks respondents to identif y the extent to which their activities impact on other divisions' activities (IMPACT-THEM), and the other itemasks the extent to which their perf ormance is af f ected by the activities carried out by other divisions (IMPACT-YOU). To be consistent with Keating (1997), we treat each item as a separate variable. This approach is supported by Lambert's (2001) analy sis that shows that these two ty pes of interdependencies mig ht have dif f erent ef f ects on perf ormance measurement. We investig ate 12 Likert-ty pe scales can be problematic as they allow respondents to answer that all perf ormance measures are used. We wanted respondents to reveal the relative use of perf ormance measures. DS Ms can thus be interpreted as the relative weig ht division summary measures are g iven by a manag er's superior during perf ormance evaluation. The Accounting Review, July 2004 553 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent the converg ent validity of IMPACT-THEM and IMPACT-YOU by computing Pearson cor- relations with three other measures: (1) S UPPLY, (2) INDEP, and (3) CHANGE. S UPPLY is the sumof the percentag e of total incoming g oods or services sourced f romother divi- sions and the percentag e of total outg oing g oods or services provided to other divisions. This item captures operating interdependencies caused by joint cost and supply f unctions or dependent demand f unctions. INDEP asks respondents to indicate the extent to which their unit could operate as a stand-alone f irm. The respondents answer on a f ully anchored Likert scale (1 = not at all, 7 = completely ). Hig h scores on this measure indicate f ewer intraf irm dependencies. CHANGE measures the time it would take external suppliers to meet the demand of a division's current internal customers. Hig h scores on a f ully anchored Likert scale (1 = immediate delivery possible, 7 = delivery only af ter more than 6 months) indicate more interdependencies within the f irm. This variable is also a reasonable proxy f or interdependencies. If a division can make a speedy chang e to an external supplier, then the division is less dependent on other divisions within the f irm. Pearson correlations be- tween IMPACT-THEM and S UPPLY, INDEP, and CHANGE are sig nif icant and in the pre- dicted direction (i.e., 0.50, p < 0.01; -0.36, p < 0.01; 0.46, p < 0.01, respectively ). We f ind similar support f or the IMPACT-YOU construct and the three alternative measures (i.e., 0.56, p < 0.01; -0.32, p < 0.01; 0.40, p < 0.01, respectively ). Tog ether these re- sults strong ly support the converg ent validity of our interdependency measures. Note also that both S UPPLY and CHANGE ask f or "harder" responses than just perceptions of interdependencies. Inf ormation asy mmetries (INF ORAS YM). Using the six-item scale developed by Dunk (1993), manag ers rate their inf ormation relative to their superior's in their area of responsibility on a seven-point Likert-ty pe scale. The results of the f actor analy sis (not reported) and scale reliability are consistent with using the six items as a unidimensional scale (Cronbach alpha = 0.86). We investig ate the converg ent validity of our measure of inf ormation asy mmetries by computing Pearson correlations with two other variables that should proxy f or specif ic knowledg e of the division manag er: (1) the manag er's experience (in y ears) in his current position (EXPINPOS ), and (2) the manag er's ag e (AGE). These variables are considered to be reasonable proxies f or inf ormation asy mmetry as more ex- perienced and older manag ers are likely to have unique knowledg e about local circum- stances and context. This knowledg e mig ht be hard to communicate to superiors, which implies that inf ormation asy mmetries arise. We also include a proxy f or the relative level of knowledg e of the division manag er about the business vis-a-vis the corporate manag er by measuring the division manag er's experience in the industry relative to his/her superior (EXPIND). All three proxy variables are based on hard data and, thus, partially address concerns with the use of perception measures. There is a sig nif icant and positive relation between INF ORAS YM and all three alternative measures (INF ORAS YMIEXPINPOS = 0.23, p < 0.05; INF ORAS YMIAGE = 0.31, p < 0.01; INF ORAS YMIEXPIND= 0.23, p = 0.06). These results provide reasonable support f or the validity of our measure. Control Variables f or Equation (1) (Use of DS Ms) Perf ormance measurement characteristics. We measure the sensitivity of perf orm- ance measures (S ENS ) by asking respondents to distribute 100 points over all perf ormance measures in such a manner that the most "sensitive" measure receives the most points. The percentag e points assig ned to DS Ms are used to measure the sensitivity of summary mea- sures. The precision of the measure is captured in a similar manner. The percentag e points assig ned to DS Ms are used as our measure of precision (PRECIS ION). The Accounting Review, July 2004 554 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stem Desig n in Divisionalized F irms Control Variables f or Equation (2) (DECEN) Division-specif ic f actors. We developed an instrument to measure g rowth opportunities (GROWTH). The instrument asks manag ers to rate g rowth expectations f or (1) their own division and f or (2) the industry in which their division operates. Respondents rated g rowth expectations on a seven-point Likert-ty pe scale where 1 = strong decline and 7 = sig nif icant increase. F actor analy sis and the Cronbach alpha statistic (0.77) support the summation of the two items. We def ine S IZE as a categ orical variable that measures the number of em- ploy ees working in a division. The categ ories rang ed between 1 (< 50 employ ees) and 8 (more than 2,000 employ ees). To reduce scale problems, S IZE is standardized to have zero mean and a standard deviation of unity . F inally , we control f or industry ef f ects in each of the reg ressions. We do so by def ining f our indicator variables to capture potential industry ef f ects. Deg rees of f reedomconsiderations prevent us f rom using an indicator f or each industry in the sample. Instead, we reg roup the industries into f our categ ories (i.e., f oodstuf f , apparel, and textile; metal, printing and plastics; electronics; services) and include three dummy variables in each of the reg ressions. Moral hazard. We include a multiplicative termin the decentralization equation that is composed of an interaction between inf ormation asy mmetry and size. This interaction term captures the capures onditions that g ive rise to moral hazard. Inclusion of the interaction term enables us to assess whether the main ef f ects of inf ormation asy mmetry and size variables on decentralization remain positive when conditions associated with moral hazard are pres- ent (Jaccard and Turrisi 2003; Wooldridg e 2000).13 Divisional manag er characteristics. We identif y f rom prior literature the importance of two divisional manag er characteristics on the decision to decentralize, namely , the ability of the divisional manag er and the trust between corporate and divisional manag ement. Ability is captured using education as a proxy . Respondents were asked to state their hig hest level of education and responses are coded as (1) only hig h school, (2) some colleg e education, (3) complete f our-y ear university -level prog ram. Trust is captured using expe- rience as a proxy . Manag ers reported the number of y ears they were employ ed in their current position (EXPINPOS ). We standardized the variable to have zero mean and standard deviation unity . Model and Econometric Issues Model S pecif ication We test our hy potheses using a simultaneous equations model that describes the deter- minants of each of the endog enous variables and their interrelation. Our sy stem of equations is described as f ollows: DS Mi = oo + a,DECEN_Pi + a2INF ORAS YM, + oa3IMPACT-THEM, + oaIMPACT-YOU, + c5S ENS , + ox6PRECIS ION, + EDS M. (3) '3 The interaction termis the cross-product of inf ormation asy mmetry andsize. It can be represented in alg ebraic f ormas f ollows: Y = a + a2X, + a3X, + a4 X,X,, where Y = decentralization, X, = inf ormation asy mmetry andX, = size, andX1X, = cross-product between X, andX,. While our model includes the interaction termin order to assess its af f ect on the coef f icients relating to inf ormation asy mmetry and size, a priori we also expect the coef f icient of the interaction termto be neg ative, i.e., we expect the relation to be less positive (or more neg ative) in larg er f irms compared to smaller f irms. The Accounting Review, July 2004 555 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent DECEN, = Po + 31DS M_P + 32INF ORAS YM, + I33MPACT-THEM, + 34IMPACT-YOUi + PsS IZEi + 36GROWTHi + [7(S IZE, * INF ORAS YM,) + 38EDUCATE, + 39EXPINPOS i + DECEN. (4) S imultaneity Bias The endog enous variables DECEN and DS M are jointly determined in equilibrium; theref ore ordinary least squares (OLS ) estimation of the model may be inappropriate (Greene 1997). We use the Durbin-Wu-Hausman test (MacKinnon 1992) to determine if simultaneity bias exists.'4 We f ind no evidence of a simultaneity bias (F = 0.59 and 1.08; p = 0.45 and 0.30) and consequently report OLS results.'5 Note that these test results do not imply that DECEN and DS M are not interrelated; the test only sug g ests that OLS estimation of the sy stem is unbiased. IV. RES ULTS Descriptive S tatistics Table 2, Panel A presents the summary statistics f or each variable. All measures in the equations have variation with most of them spanning the entire set of theoretical values. Panel B presents the Pearson correlations among the variables. There is a positive and sig nif icant relation between level of decentralization and use of DS Ms (r = 0.29, p < 0.01). We also f ind that inf ormation asy mmetries are positively related with decentralization (r = 0.61, p < 0.01) but have no sig nif icant relation with DS Ms. Decentralization is neg atively related to one f ormof interdependency (IMPACT-YOU; r = -0.26, p < 0.05), whereas the use of DS Ms is neg atively associated with the other f ormof interdependency (IMPACT- THEM; r = -0.27, p < 0.05). The relations between the endog enous and control variables are as expected. Decentralization is positively and sig nif icantly associated with S IZE, GROWTH, and a manag er's education (r = 0.31, p < 0.01; r = 0.39, p < 0.01; and r = 0.35, p < 0.01, respectively ). The use of DS Ms is positively and sig nif icantly related to S ENS and PRECIS ION (r = 0.36, p < 0.01 and r = 0.28, p < 0.01). Overall these results provide some preliminary univariate evidence that is consistent with our expectations. The correlations among the exog enous variables are not suf f iciently hig h to warrant concerns with multicollinearity (Grif f iths et al. 1993). Table 3 presents inf ormation on the relative use of perf ormance metrics in our sample. Divisional summary measures are by f ar the most important perf ormance measures used (i.e., on averag e f irms put 57 percent weig ht on this measure). This is consistent with other evidence on the use of these measures (Horng ren et al. 1999) and supports our rationale 14 The Hausman test is an asy mptotic test andits properties in f inite samples are unknown. We verif iedthe small sample power of this test using a Monte Carlo simulation. The results (not reported) indicate that the null hy pothesis of no simultaneity bias needs to be rejected at lower values of the F -statistic than under asy mptotic conditions. The values of the F -statistic we f ound in our sample are, however, well below the critical values sug g ested by the Monte Carlo simulation. 15 As a robustness check we also estimate the model using two stag e least squares (2S LS ). All results obtained using 2S LS reinf orce our earlier conclusions (i.e., sig ns and sig nif icance are not af f ected) and thus are not reported here. Note that using OLS to estimate the model circumvents the problems associated with weak instrumental variables in 2S LS (Nelson andS tartz 1990; Bound et al. 1995). Moreover, g iven that our sample size is 78 observations, which is rather small g iven the number of parameters being estimated, the use of OLS would seem pref erable. The Accounting Review, July 2004 556 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions a a , TABLE 2 Descriptive S tatistics and Correlations Panel A: S ummary S tatistics S ummary statistics f or decentralization (DECEN), use of divisional summary measures (DS M), inf ormation asy mmetries (INF ORAS YM), impact of own division on perf ormance of other divisions in f irm (IMPACT-THEM), impact of other divisions in f irmon perf ormance of own division (IMPACT-YOU), g rowth opportunities (GROWTH), size of the division (S IZE), interaction of size and inf ormation asy mmetry (INTERACTION EF F ECT), sensitivity of divisional summary measures (S ENS ), precision of divisional summary measures (PRECIS ION), experience of divisional a manag er in current position (EXPINPOS ), level of education (EDUCATE). S ample consists of 78 divisions. Data obtained f rom survey of divisional ' manag ers. t Variable Mean S td. Dev. Minimum Median Maximum DECEN 0 0.84 -2.37 0.20 1.30 DS M 0.57 0.25 0 0.50 1.00 INF ORAS YM 0 0.93 -2.18 0.14 1.43 IMPACT-THEM 0 1.00 -1.77 -0.16 1.45 IMPACT-YOU 0 1.00 -1.70 -0.08 1.54 ' S S IZE 0 1.00 -1.78 -0.01 1.32 b GROWTH 0 0.77 -1.51 -0.07 1.41 INTERACTION EF F ECT 0.04 0.91 -2.10 0 2.24 PRECIS ION 0.58 0.25 0 0.53 1.00 > S ENS 0.60 0.21 0 0.60 1.00 EDUCATE 0 1.00 -2.74 0.76 0.76 I EXPINPOS 0 1.00 -0.54 -0.30 5.62 (continued on next pag e) This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions TABLE 2 (continued) Panel B: Pearson Correlations between All Variables (ref er to Table 2, Panel A f or variable def initions) Correlati'ons are based on 78 observations. 1 1.00 2 3 4 5 6 7 0.29*** 1.00 INF ORAS YM IMPACT-THEM IMPA CT- YO U S IZE GROWTH PRECIS ION S ENS EDUCATE EXPINPOS -0.17 -0.26** 0.3 1* 0.39***, 0.02 -0.01 0.35**1* 0.14 0.15 -0.27~* -0.11 0.10 0.06 0.28** 0.17 0.14 1.00 -0.03 -0.08 0.04 0.13 0.04 0.11 -0.05 0.22* 1.00 0.66*** -0.01 0.06 -0. 19* -0.20' 0.01 -0.26* * 1.00 -0.14 0.10 -0.28** -0. 19* 0.11 -0.30*~** 1.00 0.02 -0.01 0.03 0.20-h 0(.09 1.00 -0.20* -0. 2)8 * 0.09 0.05 1.00 0.34*** 0.01 0.18 1.00 -0.14 0.08 .L. IM rt -0.39*** 1.00 rllz) ll-* Z71 tz Z f ll 1--a 'S ." z Z- -t z f lz ad denotes 10%, 5%, and I1% sig nif icance levels (two-tailed), respectively . zrl rt r C C z :z Z. Z (Q I'llIC t f t :z- llz- t-1) Q Z 41.1 1. DECEN 2. DS M GoA 3. 4. 5. 6. 7. 8. 9. 10. 11. 8 9 10 11 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms TABLE 3 Relative Use (in Percentag es) of F our Ty pes of Perf ormance Metrics by a S uperior in the Perf ormance Evaluation of a Divisional Manag er (n = 78) Ty pe of Perf ormance Metric Mean S td. Dev. Median Minimum Maximum F irm-Level S ummary Measures S tock-price-related 0.04 0.09 0.00 0.00 0.50 Other f irm-level summary 0.22 0.21 0.20 0.00 0.90 measures Divisional S ummary 0.57 0.25 0.50 0.00 1.00 Measures S pecif ic, Non-S ummary 0.17 0.16 0.13 0.00 0.80 Measures f or f ocusing on these measures. Relatively little weig ht (4 percent) is g iven to stock-price- related measures. Indeed, a majority of f irms do not use this measure to evaluate the per- f ormance of divisions. F irm-level summary measures and specif ic, non-summary measures receive on averag e similar weig ht (22 percent and 17 percent, respectively ). Main F inding s Table 4 presents the OLS estimation results of the two reg ression equations.16 The results of Equation (1) (Table 4, Panel A) provide strong evidence that the use of DS Ms is positively related to the deg ree of decentralization (aLI = 0.11, t = 2.67). It appears that the choice to deleg ate decision rig hts increases top manag ement's use of DS Ms. Contrary to our expectations, inf ormation asy mmetry has no sig nif icant inf luence on the use of DS Ms, but we do f ind that divisional interdependencies are related to the use of DS Ms. However, the ef f ect depends on the ty pe of interdependency . S pecif ically , we f ind a sig nif icant and neg ative relation (CX3 = -0.07, t = -2.23) when the division's activities af f ect other divi- sions (IMPACT-THEM), but a positive and sig nif icant ef f ect (c4 = 0.07, t = 1.96) when the division is af f ected by other divisions (IMPACT-YOU). This positive relation is unex- pected and contrary to our expectations. We also f ind that the sensitivity of a perf ormance measure increases its use (5 = 0.34, t = 2.65). The precision of a measure is only mar- g inally associated with the use of DS Ms (o6 = 0.18, t = 1.52). All industry dummies are statistically insig nif icant in explaining the use of summary measures. The adjusted R2 of Equation (1) indicates that the model has reasonable explanatory power (i.e., 23.0 percent). 16 The robustness of our results is tested by (1) relaxing the assumption of exog eneity of inf ormation asy mmetry , and (2) testing f or common method variance. There is some arg ument that inf ormation asy mmetry is an endog - enous variable (Zimmerman 2001; Ag hion and Tirole 1997). We address this issue by including a third equation into our model with inf ormation asy mmetry as a dependent variable and reestimate our model using 2S LS (based on the results f roma Hausman test). The 2S LS estimates f or the three-equation model indicate that the coef f i- cients are of a similar mag nitude and sig n to our two-equation model. Our inf erences remain unchang ed. We conduct Harman's (1967) sing le-f actor test to evaluate the extent to which common method variance exists in the data. If there is a substantial amount of common method variance, then either a sing le f actor will emerg e or one g eneral f actor will account f or the majority of the covariance among the variables. The test statistic is hig hly sig nif icant (X2 = 306.5, d.f . = 199, p < 0.001), rejecting the null that one sing le f actor accounts f or the covariance among the variables. We conclude that common method bias is not a serious problem in this data set. The Accounting Review, July 2004 559 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Aberethy , Bouwens, and van Lent TABLE 4 Ordinary Least S quares Reg ressions DS Mi = oto + otlDECEN_Pi + ot2INF ORAS YMi + 3IMPACT-THEMi + ax4MPACT-YOU, + c5S ENS i + o.6PRECIS ION, + ?f S M (1) DECEN, = Po + 31DS M P + 32INF ORAS YM, + P3IMPACT-THEM, + 34MPACT-YOUi + S S IZE, + 36GROWTH, + P7(S IZEi * INF ORAS YM,) + 8EDUCATE, + 3EXPINPOS , + EDECEN (2) Panel A: Dependent Variable: DS M (n = 78) Predicted Prob. S ig n Coef f icient S td. Error t-statistic (two-sided) Intercept 0.24 0.09 2.67 0.01 DECEN + 0.11 0.04 2.67 0.01 INF ORAS YM - -0.03 0.03 -0.82 0.42 IMPACT-THEM - 0.07 0.03 -2.23 0.03 IMPACT-YOU - 0.07 0.04 1.96 0.05 S ENS NP 0.34 0.13 2.65 0.01 PRECIS ION + 0.18 0.12 1.52 0.13 F -statistic: 3.56 Prob. (F ) = 0.001 Adj. R2 23.0% Panel B: Dependent Variable: DECEN (n = 78) Predicted Prob. S ig n Coef f icient S td. Error t-statistic (two-sided) Intercept -1.55 0.30 -5.03 0.00 DS M + 0.23 0.24 0.93 0.35 INF ORAS YM + 0.46 0.06 7.28 0.00 IMPACT-THEM - 0.04 0.08 0.53 0.60 IMPACT-YOU - -0.27 0.08 -3.21 0.00 S IZE NP 0.12 0.07 1.84 0.07 GROWTH NP 0.31 0.08 3.77 0.00 S IZE*INF ORAS YM - -0.09 0.07 -1.34 0.19 EDUCATE + 0.51 0.11 4.59 0.00 EXPINPOS + 0.06 0.06 0.99 0.32 F -statistic: 14.83 Prob. (F ) = 0.000 Adj. R2 68.3% S ample consists of 78 divisions. Data obtained f rom survey of division manag ers. Industry dummies included in both reg ression equations (not reported). Variable def initions: Decentralization (DECEN), use of divisional summary measures (DS M), inf ormation asy mmetries (INF ORAS YM), impact of own division on perf ormance of other divisions in f irm (IMPACT-THEM), impact of other divisions in f irmon perf ormance of own division (IMPACT-YOU), g rowth opportunities (GROWTH), size of the division (S IZE), interaction of size andinf ormational asy mmetry (S IZE * INF ORAS YM), sensitivity of divisional summary measures (S ENS ), precision of divisional summary measures (PRECIS ION), experience of divisional manag er in current position (EXPINPOS ), level of education (EDUCATE). The Accounting Review, July 2004 560 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms The results in Table 4, Panel B are consistent with our expectation concerning decen- tralization and inf ormation asy mmetry . The relation is positive and sig nif icant. It is inter- esting to note that this relation holds even af ter considering the potential ef f ect of moral hazard. Recall that there is some concern in the literature that as the potential f or oppor- tunistic behavior increased, corporate manag ement would be reluctant to deleg ate decision rig hts, even in the presence of inf ormation asy mmetry . Our equation includes an interaction termto proxy f or moral hazard. We f ind that even when inf ormation asy mmetry and size are larg e (the situation where we expect moral hazard problems to be most severe), the relation between inf ormation asy mmetry and decentralization remains positive and sig nif i- cant.1 Our f inding s also indicate (in Table 4, Panel B) that the use of DS Ms is unrelated to the level of decentralization, inconsistent with our expectations. We f ind evidence that interdependencies are neg atively associated with decentralization as predicted (a4 = -0.27, t = -3.21). However, this only occurs with IMPACT-YOU (the extent to which a division's perf ormance is af f ected by activities carried out in other divisions). This sug g ests that f ewer decision rig hts are assig ned to lower-level manag ers if the spillover ef f ects f romother divisions af f ect a manag er's perf ormance. S ize and g rowth opportunities are also positively correlated with decentralization (15 = 0.12, t = 1.84 and 6 = 0.31, t = 3.77). The relation between decentralization and size remains positive when we control f or the inf luence of moral hazard.'8 We f ind that more hig hly educated manag ers receive more decision rig hts (P8 = 0.51, t = 4.59). The manag er's experience in his current position, and the industry dummies, are all statistically insig nif icant at conventional levels of sig nif icance (i.e., p > 0.10). F inally , the adjusted R2 of the equation is hig h (68.3 percent). Additional Analy ses Our results indicate that the use of DS Ms increases when f irms decentralize and de- creases as interdependencies increase. This observation leaves open the question if (and how) other perf ormance measures are used to compensate f or the (de)increase in DS Ms. Recall that we measure the use of DS M as the weig ht placed on DS M vis-a'-vis f irm-level summary measures and other more specif ic, non-summary measures. We are thus able to shed some lig ht on this issue. F irst, we examine the correlation between DS Ms and the use of f irm-level summary measures (Questions 2i and 2ii in the Appendix) and between DS Ms and the use of specif ic, non-summary measures (Question 2iv in the Appendix). The cor- relation between DS Ms and the use of f irm-level summary measures is sig nif icantly neg ative (r = -0.76, p<0.01). The correlation between DS Ms and the use of specif ic, non-summary measures is also neg ative and sig nif icant (r = -0.42, p<0.01). These correlations sug g est that when corporate manag ement decreases its use of DS Ms, they put more weig ht on f irm- level summary measures and/or more specif ic, non-summary measures. To f urther remove ambig uity about how dif f erent perf ormance measures are used in relation to decentralization choices and our test variables we rerun our orig inal model, but use two alternative per- f ormance measure variables: (1) F LEVEL and (2) S PECDIV. 17 The partial relation between decentralization and inf ormation asy mmetry is representedby the expression P, + 17 * S IZE. Theref ore, of interest is a F -test of the joint hy pothesis 12 = -7 = 0. The F -statistic is 30.84 (d.f . = 2, 65; p < 0.01). S ince S IZE is a standardized variable, the coef f icient estimate 12 is the partial ef f ect of INF ORAS YM on DECEN evaluatedf or averag e-sized divisions. F or these divisions, we f indthat inf ormation asy mmetry is positively associated with decentralization. Even f or the larg est divisions, the relation remains sig nif icantly positive. 18 The partial relation between decentralization andsize is representedby the expression 15 + 17 * INF ORAS YM. Theref ore of interest is the joint test that 35s + 17 = 0. The F -statistic is 2.57 (d.f . 2, 65; p-value < 10 percent), which indicates that size anddecentralization are associated. The Accounting Review, July 2004 561 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent F LEVEL is the use of f irm-level summary measures relative to the use of DS Ms (def ined as the sumof Questions 2i and 2ii divided by the sumof Question 2i, 2ii, and 2iii). This def inition (and rescaling ) of F LEVEL allows us to conclude that any increase (decrease) in the use of this variable implies a decrease (increase) in the use of DS Ms. Prior literature (Keating 1997) has sug g ested that f irm-level summary measures will be used to provide incentives f or cooperation and reduce the noise of DS Ms when interdependencies are hig h. S PECDIV is the use of specif ic, non-summary measures relative to the use of DS Ms (def ined as Question 2iv divided by the sumof Questions 2iii and 2iv). This def inition (and rescaling ) allows us to conclude that any increase (decrease) in the use of specif ic, non- summary measures implies a decrease (increase) in the use of DS Ms. Theory sug g ests that DS Ms, rather than specif ic, non-summary measures will be positively associated with de- centralization (see also, Jensen 2001). Using more specif ic measures is inconsistent with providing decision rig hts to manag ers since asking manag ers to perf orm on specif ic mea- sures is analog ous to restricting the action choices of these manag ers. The results (not presented here) of the reg ressions are consistent with these expecta- tions. Taking tog ether the results f romthese additional analy ses and the orig inal model, we conclude that a decline in decentralization is associated with a shif t f romDS Ms to specif ic, non-summary measures and that more interdependencies are associated with a shif t f rom DS Ms to f irm-level summary measures. V. DIS CUS S ION AND CONCLUDING COMMENTS This study soug ht to develop a better understanding of the determinants of control sy stemdesig n choices and the interrelation between those choices. The evidence presented here is g enerally consistent with our expectations. Our results indicate that DS Ms are used more intensively in decentralized divisions. This is consistent with earlier empirical work that has shown decentralization choices to be an important determinant of manag ement accounting practices. Within-f irm dependencies also inf luence the use of DS Ms. If the f ocal division impacts the perf ormance of the other divisions within the f irm, then corporate manag ement's use of DS Ms decreases. This result is consistent with our expectations, with Keating 's (1997) empirical f inding s, and also with Holmstrom's (1979) inf ormativeness hy pothesis. Divisional perf ormance measures become "noisier" as interdependencies in- crease and thus are less ef f ective in evaluating the division's unique contribution. Our additional analy sis indicates that corporate manag ement shif ts f romthe use of DS Ms to f irm summary measures (Keating 1997; Bushman et al. 1995). In contrast, and contrary to our expectations, we f ind that if other divisions af f ect the perf ormance of the f ocal unit, then the use of DS Ms increases. This f inding is consistent with the inf ormativeness principle if DS Ms provide inf ormation on how well a manag er copes with interdependencies. A principal could use such inf ormation to adjust other perf ormance measures so he can "better evaluate the thing the ag ent does control" (Lambert 2001, 24). While this is post hoc rationalization, f urther theoretical and empirical research is required to f ully understand the impact of alternative f orms of interdependencies on perf ormance measurement choice. We f ind no relation between inf ormation asy mmetries and the use of DS Ms. We ex- pected that corporate manag ement would be reluctant to rely on these measures as they become increasing ly "incomplete" when they do not have the same inf ormation set as divisional manag ers. Our results do not support our expectation, but rather indicate that the use of DS Ms is only determined by the amount of decision rig hts allocated to divisional manag ers. This result needs to be interpreted in the lig ht of the evidence relating to decen- tralization choices and inf ormation asy mmetry (discussed below). Inf ormation asy mmetry The Accounting Review, July 2004 562 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms is a sig nif icant explanatory variable f or corporate manag ement's choice to decentralize. Thus, it would appear that once the f it between decentralization and inf ormation asy mmetry occurs, perf ormance measures are used to monitor divisional manag ers' use of decision rig hts. That is, inf ormation asy mmetry does not directly inf luence the use of these measures, but rather indirectly throug h the assig nment of decision rig hts. As expected, we f ind that the sensitivity of the DS Ms leads to increased use by corporate manag ement. Inf ormation asy mmetry and interdependencies are both sig nif icant determinants of de- centralization. Hig her levels of inf ormation asy mmetry increase the level of decentralization even in the presence of moral hazard. This is consistent with earlier research arg uing that the costs associated with decentralization are more than outweig hed by the benef its (Baiman et al. 1995; Christie et al. 2003). Consistent with our hy pothesis, we f ind a neg ative asso- ciation between interdependencies and decentralization. However, the neg ative relation is only f ound when other divisions af f ect the f ocal division. This sug g ests that division man- ag ers perceive their decision authority declines only when other divisions af f ect their op- erations. On the other hand, there is no ef f ect on their sense of empowerment (or disem- powerment) when their division inf luences others. This result is intuitively appealing , particularly g iven that we operationalized decision rig hts using Ag hion and Tirole's (1997) notion of "real" authority , i.e., we captured divisional manag ers' assessment of their inf lu- ence over a rang e of decisions. It is entirely possible that divisional manag ers believe there is little or no "real" deleg ation of authority when there are f actors over which he/she has no control. We also f ind that larg er- and hig her-g rowth divisions tend to be more decentralized. While we took care to address methodolog ical concerns, three caveats should be rec- og nized when considering the evidence in this paper. F irst, we concede the simplicity of the model. No empirical study can investig ate all org anizational desig n choices at the same time. Theref ore, this is a partial equilibriumstudy in which modeling choices are inf ormed by prior research and theory . However, there is a potential f or specif ication error in the reg ression and f or correlated omitted variables. There may be other determinants of org a- nization desig n choices and other control mechanisms that will inf luence decentralization and perf ormance measurement choices. F or example, we do not have data on whether perf ormance measures are used in annual bonus plans. We can, theref ore, not test directly the inf luence of compensation contracts on metric use or decentralization. S econd is the potential f or measurement error, which is of ten associated with survey data that relies on the perceptions of respondents. Measurement error af f ects the consistency of the parameter estimation of the structural model and its standard errors. We tried to minimize this problem by asking respondents to provide "harder" data on org anizational practices. The use of multi-item scales also mitig ates some of the measurement error concerns. Nevertheless, some of our data may be subject to measurement error. Third, there is a possibility of cross- sectional positive serial correlation between our observations. If f irms adopt f irm-wide or- g anizational desig n practices, then multiple observations within one f irm may be correlated. Industry -wide adoption of org anizational routines mig ht not be completely controlled f or in our dummy specif ication. It should be noted, however, that any existing positive serial correlation would bias the standard errors of the parameter estimates downward. Despite these caveats, this study has the potential to contribute to our understanding of control sy stemdesig n and the f actors that inf luence corporate manag ement's choices when implementing control sy stems into divisions. We extend prior literature by examining the interrelation between decentralization and perf ormance measurement choices and also by assessing the joint ef f ect of inf ormation asy mmetry and interdependencies on those choices. The Accounting Review, July 2004 563 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions 564 Abernethy , Bouwens, and van Lent We f ind strong evidence f or the importance of inf ormation asy mmetries and interdepen- dencies as determinants of these choices. F inally , we f ind some evidence that control choices are interrelated. APPENDIX Instruments 1. Decentralization (DECEN) In this section, we would like y ou to compare y our inf luence with the inf luence of y our superior on the f ollowing decisions. i. S trateg ic decisions (e.g ., development of new products; enter and develop new mar- kets; y our unit's strateg y ) ii. Investment decisions (e.g ., acquiring new assets and f inancing investment projects; inf ormation sy stems) iii. Marketing decisions (e.g ., campaig ns; pricing decisions) iv. Decisions reg arding internal processes (setting production/sales priorities; inputs used and/or processes employ ed to f ill orders; contracting input suppliers) v. Human resources decisions (e.g ., hiring /f iring ; compensation and setting career paths f or the personnel employ ed within y our unit; reorg anizing y our unit; crea- tion of new jobs) If y ou and/or any of y our subordinates make the decision without the knowledg e of y our supervisor, y ou and/or others of y our unit are considered to have all inf luence. My superior and I have My unit about the My superior has all same has all inf luence inf luence inf luence i. S trateg ic decisions 1 2 3 4 5 6 7 n/o ii. Investment decisions 1 2 3 4 5 6 7 n/o iii. Marketing decisions 1 2 3 4 5 6 7 n/o iv. Decisions reg arding internal 1 2 3 4 5 6 7 n/o processes v. Human resource decisions 1 2 3 4 5 6 7 n/o 2. Divisional S ummary Measures (DS Ms) Indicate the weig hts y our supervisor assig ns to each of these measures to assess y our unit's perf ormance. Your answers should total 100 percent. i. S tock-price-related measures % ii. F irm-level perf ormance measures (e.g ., f irm output, f irm ROI, f irm prof it % marg ins, f irm income) iii. Measures summarizing the total perf ormance of y our unit (e.g ., y our % unit's income, unit EVA or ROI, unit output) iv. Measures that provide perf ormance inf ormation on specif ic aspects % within y our business unit (e.g ., R&D, production ef f iciency or quality prog rams, unit product costs) v. Other measures not mentioned (please specif y ) % Total 100% The Accounting Review, July 2004 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms 3. Alternative Perf ormance Measure Instruments (PROF IT) Indicate the weig hts y our supervisor assig ns to each of these measures to assess y our unit's perf ormance. Your answers should total 100 percent. i. S tock-price-related measures % ii. Prof it measures (e.g ., ROI, prof it marg ins, net prof it) % iii. Cost measures (e.g ., averag e cost price, R&D costs) % iv. Turnover measures % v. Nonf inancial perf ormance measure reg arding strateg y , marketing and % investments (e.g ., customer satisf action, market share, R&D prog ress) vi. Nonf inancial perf ormance measures related to internal processes % reg arding personnel (e.g ., productivity , quality , coaching and training projects) vii. Other nonf inancial perf ormance measures (please specif y ) % Total 100% 4. Interdependencies This section relates to the relationships between y our unit and other org anizational units. No A very impact S ome sig nif icant at all impact impact (a) To what extent 1 2 3 4 5 6 7 n/o do y our unit's actions impact on work carried out in other org anizational units of y our f irm (IMPACT- THEM) (b) To what extent 1 2 3 4 5 6 7 n/o do actions of manag ers of other units of the f irm impact work carriedout in y our particular unit (IMPACT- YOU) (c) What percentag e of y our total production is deliveredto other org anizational units of y our f irm? % (S UPPLY) (d) What percentag e of y our total production uses inputs acquired f romother org anizational units of % y our f irm? (S UPPLY) F or about Not at half the F or all all business business (e) To what extent 1 2 3 4 5 6 7 n/o could y our org anizational unit operate as a stand-alone business (INDEP) The Accounting Review, July 2004 565 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethv, Bouwens, and van Lent (f ) How long would it take external suppliers to meet the demand of y our current internal customers? (CHANGE) 1 2 3 4 5 6 They can It would It would It would It would It would deliver take themtake themtake themtake themtake them without 1 month 2 months 3 months 4 months 5 months delay 5. Inf ormation Asy mmetries (INF ORAS YM) My supervisor is much more f amiliar 1 (a) Compared to y our superior, who is in possession of better inf ormation reg arding the activities undertaken in y our org anizational unit? (b) Compared to y our superior, who is more f amiliar with the input- output relationships inherent in the internal operations of y our org anizational unit? (c) Compared to y our superior, who is more certain of the perf ormance potential of y our org anizational unit? (d) Compared to y our superior, who is more f amiliar technically with the work of y our org anizational unit? 2 3 1 2 3 My supervisor is much more certain 1 My supervisor is much more f amiliar 1 2 3 We are about equally f amiliar 4 5 6 4 56 We are about equally certain 4 5 6 We are about equally f amiliar 4 56 The Accounting Review, July 2004 n/o 7 It would take them 6 or more months n/o n/o n/o I ammuch more f amiliar 7 7 I ammuch more certain 7 I ammuch more f amiliar 7 n/o 566 2 3 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stem Desig n in Divisionalized F irms (e) Compared to y our superior, who is better able to assess the potential impact on y our activities of f actors external to y our org anizational unit? (f ) Compared to y our superior, who has better understanding of what can be achieved in y our org anizational unit? My supervisor has a much better understanding 1 2 3 1 2 3 We have about the same understanding I have a much better understanding 4 5 6 7 n/o 4 56 7 n/o 6. Growth i. What is y our expectation with respect to the g rowth opportunities that exist within the industry in which y ou compete? ii. What is y our expectation with respect to the g rowth opportunities y our specif ic unit f aces? S trong Decline 1 2 3 1 2 3 No g rowth 4 5 6 4 56 S ig nif icant increase 7 n/o 7 n/o 7. S ensitivity (S ENS ) Please indicate the extent to which the f ollowing perf ormance measures reveal the perf ormance of y our unit in a relatively timely manner. If the perf ormance measures are all equally timely in y our unit, then each measure (i-iv) receives 25 percent. If one measure is timelier, then it will receive a weig ht of more than 25 percent. Hence, y our answers should total 100 percent. i. S tock-price-related measures % ii. Prof itability , cost or revenue measures (e.g ., ROI, prof it marg ins, income, % production costs, R&D costs) iii. Nonf inancial perf ormance inf ormation on investments, strateg y , marketing , % internal processes, and human resources iv. Other nonf inancial measures not mentioned (please specif y ) % Total 100% 8. Precision Please indicate the extent to which the f ollowing perf ormance measures reveal the perf ormance of y our unit accurately . In this context, accurateness means that y our unit's ef f ort is well ref lected in The Accounting Review, July 2004 567 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent the perf ormance measure. We would like y ou to compare the accurateness of the measures below. If the perf ormance measures are all equally accurate in y our unit, then each measure (i-v) receives 20 percent. If one measure is more accurate, then it will receive a weig ht of more than 20 percent. Hence, y our answers should total 100 percent. i. S tock-price-related measures % ii. Measures jointly summarizing the perf ormance of y our and other units (e.g ., % f irm-wide prof it) iii. Measures summarizing the total perf ormance of y our unit (i.e., y our unit's % income, total output of y our unit) iv. Measures that provide perf ormance inf ormation on specif ic aspects within y our % org anizational unit (e.g ., R&D, production ef f iciency or quality prog rams) v. Other measures not mentioned (please specif y ) % Total 100% REF ERENCES Abemethy , M. A., and A. Lillis. 2001. Interdependencies in org anization desig n. Journal of Manag e- ment Accounting Research 13: 107-129. Ag hion, P. J., and J. Tirole. 1997. F ormal and real authority in org anizations. Journal of Political Economy 105 (11): 1-29. Baiman, S ., D. Larcker, and M. V. Rajan. 1995. Org anizational desig n f or business. Journal of Ac- counting Research 33 (2): 205-229. Baker, G., M. Gibbs, and B. Holmstrom. 1994. The internal economics of the f irm: Evidence f rom personnel data. The Quarterly Journal of Economics 109 (4): 881-919. Banker, R. D., and S . M. Datar. 1989. S ensitivity , precision and linear ag g reg ation of sig nals f or perf ormance evaluation. Journal of Accounting Research 27 (1): 21-39. Bolton, P., and M. Dewatripont. 1995. The time and budg et constraints of the f irm. European Eco- nomic Review 39 (3-4): 691-699. Bound, J., D. A. Jaeg er, and R. M. Baker. 1995. Problems with instrumental variables estimation when the correlation between the instruments and the endog enous explanatory variable is weak. Journal of the American S tatistical Association 90 (430): 443-450. Brickley , J. A., C. W. S mith, and J. L. Zimmerman. 1997. Manag erial Economics and Org anizational Architecture. New York, NY: Irwin. Bushman, R. M., R. J. Indjejikian, and A. S mith. 1995. Ag g reg ate perf ormance measures in business unit manag er compensation: The role of intra-f irm interdependencies. Journal of Accounting Research 33 (S upplement): 101-128. Chenhall, R. H., and D. Morris. 1986. The impact of structure, environment, and interdependence on the perceived usef ulness of manag ement accounting sy stems. The Accounting Review 61 (1): 16-35. Chow, C. W., J. C. Cooper, and W. S . Waller. 1988. Participative budg eting : Ef f ects of a truth-inducing pay scheme and inf ormation asy mmetry on slack and perf ormance. The Accounting Review 63 (1): 111-123. Christie, A. A., M. P. Joy e, and R. L. Watts. 2003. Decentralization of the f irm: Theory and evidence. Journal of Corporate F inance 9: 3-36. Datar, S ., S . Cohen Kulp, and R. A. Lambert. 2001. Balancing perf ormance measures. Journal of Accounting Research 39 (1): 75-92. Dunk, A. S . 1993. The ef f ect of budg et emphasis and inf ormation asy mmetry on the relation between budg etary participation and slack. The Accounting Review 68 (2): 400-410. F isher, J. 1994. Technolog ical interdependence, labor production f unctions, and control sy stems. Ac- counting , Org anizations and S ociety 19 (6): 493-505. The Accounting Review, July 2004 568 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Determinants of Control S y stemDesig n in Divisionalized F irms Gordon, L. A., and V. K. Naray anan. 1984. Manag ement accounting sy stems: An empirical evaluation. Accounting , Org anizations and S ociety 9 (1): 33-47. Greene, W. H. 1997. Econometric Analy sis. 4th edition. Upper S addle River, NJ: Prentice Hall. Grif f iths, W. E., R. C. Hill, and G. C. Judg e. 1993. Learning and Practicing Econometrics. New York, NY: John Wiley . Harman, H. H. 1967. Modern F actor Analy sis. 2nd edition. Chicag o, IL: University of Chicag o Press. Holmstrom, B. 1979. Moral hazard and observability . Bell Journal of Economics 10: 74-91. Hormg ren, C. T., G. F oster, and S . M. Datar. 1999. Cost Accounting : A Manag erial Emphasis. Upper S addle River, NJ: Prentice Hall. Ittner, C. D., and D. F . Larcker. 2001. Assessing empirical research in manag erial accounting : A value based perspective. Journal of Accounting and Economics 32 (1-3): 349-410. Jaccard, J., and R. Turrisi. 2003. Interaction Ef f ects in Multiple Reg ression. 2nd edition. Thousand Oaks, CA: S ag e Publications. Jensen, M. C., and W. H. Meckling . 1992. S pecif ic and g eneral knowledg e and org anizational struc- ture. In Contract Economics, edited by L. Werin, and H. Wijkander. Cambridg e, U.K.: Blackwell Publishers. . 2001. Value maximization, stakeholder theory , and the corporate objective f unction. Journal of Applied Corporate F inance 14 (3): 8-21. Keating , A. S . 1997. Determinants of divisional perf ormance evaluation practices. Journal of Ac- counting and Economics 24: 243-274. Kreps, D. 1990. Corporate culture and economic theory . In Perspectives on Positive Political Economy , edited by J. Alt, and K. S hepsle. Cambridg e, U.K.: Cambridg e University Press. Lambert, R. A. 2001. Contracting theory and accounting . Journal of Accounting and Economics 32 (1-3): 3-88. Lanen, W. N. 1995. Discussion of ag g reg ate perf ormance measures in business unit manag er com- pensation: The role of intra f irm interdependencies. Journal of Accounting Research 33 (S up- plement): 129-134. Lawrence, P. R., and J. W. Lorsch. 1967. Org anization and Environment: Manag ing Dif f erentiation and Integ ration. Homewood, IL: Irwin. Lorenz, E. 1999. Trust, contract and economic cooperation. Cambridg e Journal of Economics 23: 301-315. Luf t, J., and M. S hields. 2003. Mapping manag ement accounting : Graphics and g uidelines f or theory - consistent empirical research. Accounting , Org anizations and S ociety 28: 169-250. MacKinnon, J. G. 1992. Model specif ication tests and artif icial reg ressions. Journal of Economic Literature 30 (1): 102-146. Medof f , J. L., and K. G. Abraham. 1980. Experience, perf ormance and earning s. Quarterly Journal of Economics (December): 703-736. Melumad, N., D. Mookherjee, and S . Reichelstein. 1992. A theory of responsibility centers. Journal of Accounting and Economics 15: 445-484. Milg rom, P., and J. Roberts. 1992. Economics, Org anization and Manag ement. Upper S addle River, NJ: Prentice Hall. , and . 1995. Complementarities and f it: S trateg y , structure and org anizational chang e in manuf acturing . Journal of Accounting and Economics 19: 179-208. Miller, G. A. 1987. Meta-analy sis and the culture f ree hy pothesis. Org anization S tudies 8 (4): 309- 326. Nag ar, V. 2002. Deleg ation and incentive compensation. The Accounting Review 77 (2): 379-395. Nelson, C. R., and R. S tartz. 1990. The distribution of the instrumental variables estimator and its t-ratio when the instrument is a poor one. Journal of Business 63 (1): S 125-S 140. Nooteboom, B., H. Berg er, and N. G. Noorderhaven. 1997. Ef f ects of trust and g overnance on rela- tional risk. Academy of Manag ement Journal 40 (2): 308-338. Nunnally , J. C., and I. H. Bernstein. 1994. Psy chometric Theory . 3rd edition. New York, NY: McGraw- Hill. The Accounting Review, July 2004 569 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions Abernethy , Bouwens, and van Lent Reichelstein, S . 1992. Constructing incentive schemes f or g overnment contracts: An application of ag ency theory . The Accounting Review 67 (4): 712-731. Rockness, H. 0., and M. D. S hields. 1984. Org anizational control sy stems in research and develop- ment. Accounting , Org anizations and S ociety 9 (2): 165-177. S olomons, D. 1965. Divisional Perf ormance, Measurement and Control. Homewood, IL: Irwin. Thompson, J. D. 1967. Org anizations in Action. New York, NY: McGraw-Hill. Vancil, R. F . 1979. Decentralization: Manag erial Ambig uity by Desig n. New York, NY: Dow Jones- Irwin. Wooldridg e, J. M. 2000. Introductory Econometrics: A Modern Approach. Cincinnati, OH: S outh- West Colleg e Publishing . Wruck, K., and M. C. Jensen. 1994. S cience, specif ic knowledg e, and total quality manag ement. Journal of Accounting and Economics 18 (3): 247-287. Zimmerman, J. L. 1997. EVA and divisional perf ormance measurement: Capturing sy nerg ies and other issues. Journal of Applied Corporate F inance 10 (2): 98-109. .2001. Conjectures reg arding empirical manag erial accounting research. Journal of Accounting and Economics 32 (1-3): 411-428. The Accounting Review, July 2004 570 This content downloaded from 145.18.103.80 on Wed, 3 Sep 2014 09:38:56 AM All use subject to JSTOR Terms and Conditions