Information Systems ResearchVol. 26, No. 4, December 2015, pp. 656–674ISSN 1047-7047 (print) � ISSN 1526-5536 (online) http://dx.doi.org/10.1287/isre.2015.0591
Amrit TiwanaUniversity of Georgia, Athens, Georgia 30602, [email protected]
Stephen K. KimIowa State University, Ames, Iowa 50011, [email protected]
The information technology (IT) governance literature predominantly explains firms’ IT governance choices, butnot their strategic consequences. We develop the idea that a firm’s IT governance choices induce adeptness at
strategically exploiting IT only when they are discriminatingly aligned with its departments’ knowledge outsidetheir specialty. Discriminating means that governing the two undertheorized classes of IT assets—apps andinfrastructure—requires “peripheral” knowledge in different departments. Analyses of data from 105 firms supportour middle-range theory.
Keywords : IT governance; IT infrastructure; IT applications; discriminating alignment; IT agility; Garen method;IT strategy; endogeneity; IT asset classes
History : Samer Faraj, Senior Editor; William Kettinger, Associate Editor. This paper was received April 23, 2012,and was with the authors 14 months for 3 revisions. Published online in Articles in Advance October 16, 2015.
1. IntroductionCitibank was the first to introduce an iPhone app thatlet customers cash a check by uploading its image toCitibank’s servers. The innovation created unprece-dented convenience for its customers and reduced boththe fees that Citibank paid to out-of-network ATMs andits need for new ATMs. Yet, like many information tech-nology (IT) innovations, rivals soon copied it. Citibankwas unsurprised, for it had learned to expect that sincebetting $100 million in 1977 to introduce ATMs intoits industry. By seizing an opportunity to differentiateitself and reduce costs, Citibank created a temporaryadvantage over its rivals. Its competitive weapon wasnot its IT, but its agility to use IT to consistently createa series of temporary advantages, introducing a newone before rivals could even finish copying the lastone. Such strategic IT agility is not unique to banking.Firms such as Zara, Delta Air Lines, and UPS are theCitibanks of their industries: consistently using IT tostrategically outdistance rivals, who are constantlyplaying second fiddle. Why are some firms more adeptat using IT in their pursuit of strategic opportunities?
As IT grows into firms’ largest capital expense, theyincreasingly demand—and expect—IT to be strate-gically responsive yet economical (Brynjolfsson andSchrage 2009, Feld and Stoddard 2004). IT is “expectedto perform miracles,” as Heller (2012, p. 33) laments.These conflicting demands appear hard to reconcileuntil we weigh the idea in the practitioner literature thatdifferent types of IT assets must be governed differently(Aral and Weill 2007, Xue et al. 2008). The distinctionbetween two broad “classes” of IT assets—IT apps and
IT infrastructure—widespread in the IT practitionerliterature exists primarily as a powerful descriptivetaxonomy whose theoretical properties have remainedunderdeveloped (e.g., Agarwal and Sambamurthy 2002,Ross 2002, Weill and Ross 2005).
The belief is that the secret sauce for exploiting ITfor strategic agility is how it is governed, i.e., whichdepartment makes what IT decisions (Heller 2012,p. 3; Weill and Ross 2004, p. 14). That implies thatsome IT decisions not be made by the IT unit (Ross2002). Pundits therefore urge IT and line functionsto learn to speak each other’s language (Weill andRoss 2004, p. 68), a suggestion that is at odds with thespecialization needed for functional division of laborwithin a firm. Paradoxically, departmental specializa-tion breeds ignorance, yet such ignorance is preciselywhat permits departmental specialization. A marketingdepartment making IT decisions sounds as unrealisticas IT making branding or advertising decisions. Afterall, functions specialize in different activities (Alonsoet al. 2008). A middle ground might be selectivelyreducing cross-departmental ignorance, where onefunction speaks the other’s language but not vice versa.
However, the IT governance literature, preoccupiedwith explaining IT governance centralization and decen-tralization choices (e.g., Brown and Magill 1994, 1998;Sambamurthy and Zmud 1999), has yet to explorewhen which department must speak the other’s lan-guage or its interplay with a firm’s IT governancechoices. (We refer to one department’s knowledge inthe other’s domain but outside its own as peripheralknowledge; Tiwana and Keil 2007.) Thus, although
we have considerable insight into the broad drivers offirms’ IT governance choices, these insights are notcumulative because prior studies inconsistently use ITgovernance to refer to just one class of IT decisions(e.g., Brown and Magill 1998, Brown 1997, Olson andChervany 1980) or both lumped together (Lewis andByrd 2003, Ross 2002, Tiwana and Konsynski 2010,Weill et al. 2002, Xue et al. 2011).
In summary, we know little about how the nuancedinterplay between a firm’s IT governance choices andthe IT unit’s and line functions’ peripheral knowl-edge shapes IT strategic agility. Which department hasperipheral knowledge might have value dependingon which department makes what IT decisions. Putdifferently, the alignment between the IT unit’s andline functions’ peripheral knowledge and classes ofgoverned IT assets must be more nuanced, or “dis-criminating.” Failing this, firms risk either makingstrategically numb IT investments or diluting depart-mental specialization. This study addresses this gapguided by the following research question: How doesthe interplay between firms’ IT governance choices anddepartmental peripheral knowledge influence IT strategicagility?
We theoretically develop the idea that IT governanceamplifies firms’ IT strategic agility only when it is“discriminatingly” aligned with the IT unit’s and linefunctions’ peripheral knowledge. To develop the dis-crimination idea, we disaggregate IT governance intotwo classes (apps and infrastructure) and peripheralknowledge into two types (IT unit’s business knowl-edge and line functions’ technical knowledge). Usingthe work of Jensen and Meckling (1992) (JM) as aspringboard for our middle-range theory, we use thesame theoretical apparatus for both classes of governedIT. We theorize that greater technical knowledge inline functions but greater business knowledge in the ITunit respectively increase returns to agility from ITapp and IT infrastructure governance. Our recognitionof the endogenous nature of IT governance permitscumulativeness with the vast literature on IT gover-nance choice without making any normative assertions.
Econometric tests using matched-pair data from 105firms provide considerable support for the proposedideas.
Our distinctive contribution is a middle-range theoryof how firms’ IT strategic agility is predicated in adiscriminating alignment of their IT governance choicesand their departments’ peripheral knowledge. Thedepartment whose peripheral knowledge enhances suchagility depends on how the firm governs its IT appsand IT infrastructure. Subsequent sections theoreticallydevelop these ideas (§2); describe the methodology (§3),analyses, and results (§4); and discuss our contributionsand implications (§5).
2. Theoretical DevelopmentOur research model in Figure 1 predicts IT strategicagility (construct definitions in Table 1). We defineIT strategic agility as the degree to which the IT unitfurthers a firm’s pursuit of strategic business opportuni-ties, building on prior IT governance studies’ emphasison firms’ strategic use of IT (Segars and Grover 1998),IT agility (Lu and Ramamurthy 2011, Sambamurthyet al. 2003, Tiwana and Konsynski 2010), and using ITto exploit business opportunities (Brown and Magill1998, Hann and Weber 1996). This conceptualizationrepresents an outcome manifested in firm-level IT-related actions rather than a capability. The overarchingidea in the model is that IT strategic agility is explainedby a nuanced interplay between which departmentmakes what IT decisions and which department hasknowledge outside its own domain.
Our model does not assume any normatively idealIT governance structure. Instead, we recognize that afirm’s IT governance choices are endogenously drivenby (i) already-known strategic, internal, and externaldrivers of IT governance choices; (ii) the IT unit’s priortrack record; and (iii) internal and external knowledgeintegration mechanisms already in place. Our modeltherefore uses a firm’s IT governance choices as pre-dicted by these factors, departing from its historically
Table 1 Summary of the Key Constructs and Their Role in the Model
Construct Definition Role Representative guiding references
Focal theoretical constructsIT strategic agility The degree to which a firm’s IT unit furthers a firm’s pursuit of
strategic business opportunitiesDependent
variableSambamurthy et al. (2003), Segars and Grover
(1998)IT app governancedecentralization
Degree to which decision rights for IT applications lean towardthe line functions vis-à-vis the IT unit
Endogenouspredictor
Baschab and Piot (2003, p. 81), Brown and Magill(1998), Hann and Weber (1996), Lewis and Byrd(2003), Tiwana and Konsynski (2010)
IT infrastructuregovernancecentralization
Degree to which decision rights for IT infrastructure leantoward the IT unit vis-à-vis the line functions
Endogenouspredictor
Brown and Magill (1998), Hann and Weber (1996),Star and Ruhleder (1996), Tiwana and Konsynski(2010), Weill et al. (2002)
Line functions’technical knowledge
The line functions’ knowledge about IT Predictor Bassellier et al. (2003), Mitchell (2006), Tiwanaand Konsynski (2010)
IT unit’s businessknowledge
The IT unit’s business knowledge about the organization’s linefunctions’ routines, rules, heuristics, opportunities, andthreats
Predictor Bassellier and Benbasat (2004), Mitchell (2006),Tiwana and Konsynski (2010)
Instrumental variablesIndustry dynamism Rapidity of introduction of new market offerings by the firm’s
competitorsInstrument Brown (1997)
IT unit size The employee count of the firm’s IT unit Instrument Brown (1997)Uncertainty Degree of uncertainty in the work to be done by the IT unit Instrument Hann and Weber (1996)Strategic importance ofIT
Degree to which the firm’s business strategy depends on IT Instrument Brown and Magill (1998), Hann and Weber (1996)
Cross-unit IT synergy Degree to which substantial benefits can potentially be realizedby centrally coordinating IT activities across the firm
Instrument Brown and Magill (1998), Sambamurthy and Zmud(1999)
IT formalization Degree to which the work of the IT unit is codified intoformalized procedures
Instrument Ranganathan and Sethi (2002)
External knowledgeintegration
Degree to which knowledge from outside the firm is activelyutilized in the firm’s IT activities
Instrument Mitchell (2006), Reich and Benbasat (2000)
Internal knowledgeintegration
Degree to which knowledge dispersed in the line and ITfunctions is jointly utilized in the firm’s IT activities
Instrument Mitchell (2006)
IT unit performance Degree to which the IT unit met line functions’ needs Instrument Nelson and Cooprider (1996)
exogenous treatment (e.g., Tiwana and Konsynski 2010;Xue et al. 2008, 2011).1
2.1. Theoretical Foundation: Jensen and Meckling’s(1992) Theory
Our model’s foundation is Jensen and Meckling’s (1992)theory, the crux of which is that decision rights mustbe colocated with the knowledge needed to make thosedecisions. A decision right specifies who in a firm hasthe authority to make what decisions (Nault 1998).When the two are not colocated, either (a) decisionrights must be moved to the department where the rel-evant knowledge resides (JM’s “delegation” solutionfollowing Alonso et al. 2008) or (b) the relevant knowl-edge must be moved to the locus of decision rights(JM’s “transmission” solution). Figure 2(a) illustratesthese two solutions, which Jensen and Meckling (1992)view as alternative choices.
We conceptualize IT governance as comprising two“classes” of IT decisions, IT apps and IT infrastructure,building on the notion of IT governance structure
1 We follow knowledge-based theories in assuming that IT and linefunctions have deep knowledge of their own domains (Baldwin2008, p. 166; Becker and Murphy 1992).
as the specification of IT decision rights (Brown andMagill 1998, Sambamurthy and Zmud 1999, Tiwana andKonsynski 2010). A firm can locate an IT decision rightanywhere along the continuum between centralizationwith the IT unit or decentralization to the line functions(Weill and Ross 2004, p. 59).2
The problem with either JM solution—delegationor transmission—is the assumption that there existsone department with the entirety of decision-relevantknowledge (Demsetz 1992). This assumption is oftenviolated in IT decisions; effective IT decisions simultane-ously require both line functions’ business knowledgeand the IT unit’s technical knowledge (Mitchell 2006,Sambamurthy and Zmud 1999). Therefore, both central-ization and decentralization will always fail to colocatesome necessary knowledge with IT decision rights. JM’sdelegation solution then requires tandem recourse toJM’s transmission solution, i.e., moving the separatedknowledge to the decision rights-holding department.Figure 2(b) illustrates this. Unlike Figure 2(a), where a
2 The focus in our conceptualization of governance as decision rightsis on IT decisions rather than IT activities, following Aghion andTirole (1997). Our conceptualization mirrors Alonso et al. (2008),where decentralized (centralized) governance entails decision rightsallocated primarily to line functions (the IT unit).
Figure 2 (a) JM’s Two Alternative Colocation Solutions (b) Must be Used in Tandem When Decision-Relevant Knowledge Is Dispersed Across Departments
Decisionrights
Knowledge
DELEGATION
solution
TRANSMISSION
solution
or Decisionrights
Knowledge(ka)primary
Knowledge(kb)complementary
and
DELEGATION
solution
TRANSMISSION
solution
(a) (b)
firm chooses either JM’s delegation solution or trans-mission solution, it must use both the delegationsolution and the transmission solution in Figure 2(b)when knowledge is dispersed across departments.
The challenge in realizing JM’s transmission solu-tion is that IT and line functions speak different lan-guages by virtue of departmental specialization (Ross2002; Star and Ruhleder 1996; Weill and Ross 2004,p. 68). A Wall Street Journal article called this “a glasswall” (Basu 2008). The delays in one understanding theother can imperil IT strategic agility. Swiftly buildingconsensus between them requires one department tobe able to converse in the language of the decision-rights-holding department (Bucciarelli 1994, p. 148).Then, increasing one department’s knowledge in theother department’s domain but outside its own (hence,“peripheral knowledge”; Tiwana and Keil 2007) easescommunication by allowing it to speak more of theother’s language (Faraj and Xiao 2006).3 Such peripheralknowledge, however, violates the spirit of departmentalspecialization (Garicano 2000), making it important toknow when its substantial upfront costs are commensu-rate with its benefits in facilitating JM’s transmissionsolution.
2.2. Discriminating AlignmentDiscriminating alignment refers to the nuanced align-ment between which department makes what IT deci-sions and which department has peripheral knowledge.It is discriminating because a specific department’speripheral knowledge is valuable only for a specificclass of IT decisions but not the other. Table 2 summa-rizes our discriminating alignment logic for IT appgovernance (§2.2.1) and IT infrastructure governance(§2.2.2).
We first theorize which department JM’s delegationsolution nudges decision rights toward, recognizing
3 Others describe such knowledge outside one’s specializationas transspecialist understanding, general knowledge (Becker andMurphy 1992), cross-domain knowledge, out-of-area knowledge, orcommon knowledge; all implicitly assume symmetric interdepart-mental overlaps.
that IT apps and infrastructure require different types ofknowledge (Messerschmitt and Szyperski 2003, p. 200).Capitalizing on emergent opportunities demands speed(McGrath 2013, p. 12); increasing a firm’s IT strategicagility therefore requires minimizing the delays fromknowledge transfers between IT and line functions(Jensen and Meckling 1992, p. 254).
Decision rights should be allocated to the bestinformed department, regardless of the need for coor-dination (Alonso et al. 2008, p. 162). JM’s delegationsolution, i.e., shifting them toward the locus of theknowledge on which that class of IT decisions primarilydraws (ka in Figure 2(b)) accelerates IT decisions byminimizing such delays (Bester and Krähmer 2008).4
To apply the now-separated complementary knowl-edge (kb) to those decisions, the firm must then—intandem—use JM’s transmission solution. This iswhere discriminating alignment comes in. Increasingperipheral knowledge in the department that has lesserauthority over a class of IT decisions facilitates JM’stransmission solution.
Our conceptual separation of both (a) IT app from ITinfrastructure governance and (b) two types of periph-eral knowledge is theoretically significant. First, priorstudies use the term IT governance to refer to threeconceptually different things: (i) IT apps (Brown andMagill 1998, Brown 1997, Olson and Chervany 1980),(ii) IT infrastructure (Xue et al. 2011), and (iii) bothlumped together (e.g., Hann and Weber 1996, Lewisand Byrd 2003, Tiwana and Konsynski 2010, Weillet al. 2002). By contrast, we use the apps versus infras-tructure demarcation in the practitioner literature as astepping stone to develop their underdeveloped theoret-ical properties (e.g., Agarwal and Sambamurthy 2002,Ross 2002, Weill and Ross 2005). Similarly, separatingthe two peripheral knowledge types departs from anisomorphic notion of “shared knowledge” that assumesaway asymmetry in departmental knowledge overlaps(e.g., Nelson and Cooprider 1996, Reich and Benbasat2000).
4 Delegation in our theory represents centralization of IT infrastructureand decentralization of IT app governance.
Table 2 Overview of the Theoretical Logic for Discriminating Alignment
Class of IT decisions
Apps Infrastructure
Knowledge primarily needed (locus) Business knowledge (line functions) Technical knowledge (IT unit)JM’s delegation solution leads to decision rights 0 0 0 Decentralization CentralizationJM’s transmission solution requires moving 0 0 0 Technical knowledge → Line functions Business knowledge → IT unitPeripheral knowledge conducive to JM’s transmission solution Business knowledge in IT unit Technical knowledge in line functions
Notes. JM’s delegation solution = decision rights → knowledge locus; JM’s transmission solution = knowledge → decision rights locus.
2.2.1. Discriminating Alignment Between ITApplications Governance and Peripheral Knowledge.IT app governance refers to how decision rights for ITapps are divvied between the line functions and the ITunit.5 IT apps are often uniquely tailored to various linefunctions, and thus draw primarily on knowledge ofline functions’ specialized activities, business processes,and problems (Agarwal and Sambamurthy 2002, Ross2002, Weill and Ross 2005). Since such knowledge—byvirtue of departmental specialization—is concentratedin the line functions (Messerschmitt and Szyperski 2003,p. 201), agility-imperiling delays can be minimizedusing JM’s delegation solution, i.e., by shifting appdecision rights toward line functions. By contrast, cen-tralizing IT app decision rights would impede strategicagility by requiring time-consuming business knowl-edge transfer from various line functions to the IT unit(Athey and Roberts 2001).
However, IT app decisions must also be cognizant offirmwide technical constraints and integration withthe firm’s existing IT assets (Basu 2008; Weill andRoss 2004, p. 147), of which the IT unit likely is moreknowledgeable. Otherwise, app decisions risk beingoblivious to technical constraints (Messerschmitt andSzyperski 2003, p. 63), leading to agility-impedingiteration and rework. The IT unit’s inability to swiftlyand comprehensibly communicate such knowledge toline functions can then impede IT strategic agility. Appgovernance decentralization alone will not enhance ITstrategic agility unless this knowledge separation isovercome. This requires the IT unit to contribute itstechnical knowledge as an input into IT app decisions,i.e., a recourse to JM’s transmission solution. (Weilland Ross 2004, p. 54, call these “decision right inputs.”)However, the linguistic chasm between IT and linefunctions makes it challenging for the IT unit to com-municate its technical knowledge in a form that the line
5 IT applications (“apps”) are business analysis and transactionsystems that the firm’s individual line functions use for their coreactivities and functional business processes (Baschab and Piot 2003).They include customer-facing apps, production support apps (e.g.,supply chain, logistics, inventory, warehousing), and business supportapps (e.g., accounting and payroll; Baschab and Piot 2003, p. 81;Weill and Ross 2005). Many prior studies subsume IT apps under anall-encompassing definition of IT infrastructure (e.g., Lewis and Byrd2003, Weill et al. 2002).
functions can readily use in making app decisions (Feldand Stoddard 2004). The solution then is increasingthe IT unit’s business knowledge, which allows it toprovide inputs in a language that line functions canmore readily understand and apply in making appdecisions. Simply increasing line functions’ technicalknowledge alone would not substitute for the IT unit’speripheral knowledge because line functions are lesslikely to possess the IT unit’s holistic knowledge ofthe firmwide IT portfolio with which an app musteventually interoperate. Therefore, IT app governancedecentralization will enhance IT strategic agility onlyin combination with greater business knowledge in theIT unit. Increasing one increases the marginal returnsof the other. This represents discriminating alignmentbetween IT app governance and peripheral knowl-edge, which leads to our first discriminating alignmenthypothesis.
Hypothesis 1 (H1). An increase in IT unit businessknowledge increases the marginal returns to IT strategicagility from decentralizing IT app governance.
2.2.2. Discriminating Alignment Between ITInfrastructure Governance and Peripheral Knowledge.IT infrastructure governance refers to how decision rightsfor IT infrastructure decisions are divvied betweenthe line functions and IT unit.6 Firmwide interoper-ability, security, and scale economies are critical inprovisioning IT infrastructure because it provides afirmwide foundation shared by various line functions’IT apps (Agarwal and Sambamurthy 2002; Weill andRoss 2004, p. 35; Xue et al. 2011). IT infrastructuredecisions therefore require deep technical expertiseas well as a holistic understanding of the firm’s ITassets. Since such knowledge—by virtue of depart-mental specialization—is concentrated in the IT unit,
6 A firm’s IT infrastructure represents a foundation needed to run ITapps, or a substrate that knits together various IT apps (Feld andStoddard 2004, Lewis and Byrd 2003, Mitchell 2006, Sambamurthyet al. 2003, Weill and Ross 2005). IT infrastructure encompassesdigital communication and networks (e.g., email, Internet access),data management, IT operations and maintenance, IT procurement(e.g., hardware), and IT support (Baschab and Piot 2003, p. x; Weillet al. 2002). IT infrastructure therefore serves the entire firm and isnot unique to individual line functions.
agility-impeding delays can be minimized using JM’sdelegation solution, i.e., by shifting IT infrastructuredecision rights toward the IT unit. This representscentralization of IT infrastructure governance. By con-trast, decentralizing IT infrastructure decision rightsimpedes IT strategic agility due to knowledge transferdelays. Furthermore, piecemeal infrastructure decisionsby myriad line functions can exacerbate IT infrastruc-ture complexity, making apps progressively more timeconsuming to implement.
However, IT infrastructure decisions must also bemade in cognizance of myriad line functions’ variedIT needs (Andersson et al. 2012, Star and Ruhleder1996), which the line functions likely know betterthemselves. Line functions’ inability to swiftly andcomprehensibly communicate their needs and prioritiesto the IT unit can then impede IT strategic agility.Centralizing infrastructure governance alone will notenhance IT strategic agility unless this knowledgeseparation is overcome. This requires line functionsto contribute their business knowledge as an inputto IT infrastructure decisions, i.e., a recourse to JM’stransmission solution. However, the linguistic chasmbetween IT and line functions makes it challengingfor the line functions to communicate their businessknowledge in a form that the IT unit can readilycomprehend in making IT infrastructure decisions(Feld and Stoddard 2004, Star and Ruhleder 1996).A solution, then, is increasing line functions’ technicalknowledge, which allows them to provide relevantinputs (e.g., function-specific needs and priorities) in alanguage that the IT unit can more readily understandand apply in making IT infrastructure decisions. Simplyincreasing the IT unit’s business knowledge would notsubstitute for this because the IT unit is less likely topossess the depth of business knowledge in multiplefunctional areas. Therefore, IT infrastructure governancecentralization will enhance IT strategic agility only incombination with line functions’ technical knowledge.Increasing one increases the marginal returns of theother. This represents discriminating alignment betweenIT infrastructure governance and peripheral knowledge.This leads to our second discriminating alignmenthypothesis.
Hypothesis 2 (H2). An increase in line functions’ tech-nical knowledge increases the marginal returns to IT strategicagility from centralizing IT infrastructure governance.
3. MethodologyOur unit of analysis is the firm’s IT function, consistentwith prior studies of IT governance (Brown and Magill1998, Lu and Ramamurthy 2011, Sambamurthy et al.2003). We collected matched-pair data from seniorIT managers and line function managers in 105 U.S.firms to test the proposed ideas. Data on IT strategic
agility were collected from line function managersand the remaining variables from IT managers, whowere the key informants for our constructs. This par-titioned approach was necessary because our studyused a matched-pair design and response rates wouldhave suffered from attempting a lengthy matched-pairinstrument involving non-IT managers. The samplingframe was a random sample of 800 firms in the Dun &Bradstreet database, of which we sent the survey tomanagement information system (MIS) managers andmultiple line function managers in 620 firms where wecould either directly precontact via telephone or reachvia voicemail. Three follow-ups yielded matched-pairassessments from both an MIS manager and at leastone line function manager in 105 firms, for a 16.94%response rate (105/620).
The firms represented a variety of industries includ-ing retail, construction, services, and manufacturing.The primary respondents were highly experienced(average IT experience in the firm was 7.3 years). Onaverage, the firms invested 5.05% (SD, 4.45) of theirannual revenues on IT, and employed 34 (SD, 80.6)individuals in the IT unit and 519 (SD, 907) employeesin total. Thirty-four percent of the firms were public.Furthermore, T -tests comparing the early (first 32) andlate (last 32) respondents on the principal constructsrevealed no evidence for nonresponse bias (IT infras-tructure centralization, t = 1059; IT app decentralization,t = −0075; IT business knowledge, t = 0034; line techni-cal knowledge, t = 1043; IT unit performance, t = 0080;IT strategic agility, t = 1021; all nonsignificant (n.s.)).
3.1. Construct Operationalization andScale Development
We measured all principal constructs using reflectivemulti-item Likert scales using the firm’s IT function asthe unit of analysis. New scales were developed fortwo of the model’s key theoretical constructs, IT appand IT infrastructure governance, and the rest wereadapted (see Appendix A). Descriptions of firm-leveldecisions pertaining to IT apps and IT infrastructure inthe extant literature were used as a starting point forscale development (Baschab and Piot 2003, Ross 2002,Sambamurthy and Zmud 1999, Star and Ruhleder 1996,Weill and Ross 2004). We refined the preliminary itempool via interviews with a convenience sample of 11 ITmanagers and six academic experts. This ensured thatthe scale items clearly captured the theoretical domainof each construct and were meaningful in the broadpool of sampled industries.
Following prior research (Brown 1999, Brown andMagill 1994, Sambamurthy and Zmud 1999), IT gover-nance centralization and decentralization were used asend points of the continuum and referred respectivelyto the degree to which the authority over an IT decisionleaned toward the IT unit or the line functions. We
assume that, irrespective of which department makesIT decisions, the IT unit implements them. IT app gover-nance decentralization used four items that assessed thedegree to which the primary responsibility for decisionsabout app planning, initiating new projects, managingkey projects, and app development activities resided toa greater degree with the line functions vis-à-vis the ITunit. IT infrastructure governance centralization used fouritems that assessed the degree to which the primaryresponsibility for decisions about IT communicationsand networking, IT operations and maintenance, pro-curement of hardware/software, and end-user supportresided to a greater degree with the IT unit vis-à-visthe line functions. IT unit’s business knowledge used sixitems based on prior measures (Bassellier and Benbasat2004, Tiwana and Konsynski 2010) that assessed theextent to which members of the IT unit understoodthe firm’s day-to-day business routines, business rulesand heuristics, business opportunities and threats,business strategy, and had a holistic understandingof its business. Line functions’ technical knowledge usedseven items that assessed the extent to which theline function managers understood systems design,database structures, programming languages, IT projectmethodologies, software testing and debugging, dataprocessing procedures, and application developmenttools, informed by prior scales (see Bassellier et al. 2003,Tiwana and Konsynski 2010). We used MIS managersas key informants for three reasons. Collecting thesedata from line managers (a) was infeasible because itwould have required matched responses from manyline functions in each firm, (b) they would have beenpoor informants of line functions other than their own,and (c) they would have been even more vulnerableto self-reporting bias. We focused on IT managersperceptions about line managers because (a) they aremore likely to interact with line managers (the keydecision makers with authority in line functions) acrossthe firm and (b) managers in individual line func-tions are unlikely to be good informants about all(potentially thousands) employees in other functionaldepartments. The scale is therefore a reasonable proxyfor line functions’ technical knowledge. The adaptedscales were refined to be generalizable to the broadswath of industries in our sampling frame, based onfeedback from our expert panel. IT strategic agility wasadapted from Segars and Grover’s (1998) strategicinformation systems (IS) “planning alignment success”measure that tapped into the line function managers’assessment of the degree to which their firm’s IT unithad been successful in identifying IT opportunitiesto support the firm’s strategic direction, educatingtop management on the importance of IT, assessingthe strategic importance of emerging technologies,and adapting technology to changing business needs.(Four of their items held up in the factor analysis,
narrowing the meaning of our measured construct.)Their conceptualization of this construct was the firm’ssuccess in deploying IT that was congruent with thefirm’s evolving strategic needs, which is conceptuallysimilar but broader than IT strategic agility. The itemsand sources for the controls and instrumental variablesappear in Appendix A.
3.2. Psychometric PropertiesTable 3 summarizes correlations, scale reliability, means,and standard deviations. The acceptable Cronbach�’s (≥0079) and eigenvalues (>201) provide the firstassurance that the scales had high convergent validity.The Varimax-rotated exploratory factor analysis (EFA)matrix in Appendix B further shows that the loadingsof each indicator on the corresponding theoreticalconstruct exceeded the recommended threshold of0.6 and had low cross loadings (<003). Overall, thissuggests psychometric adequacy.
4. Analysis and Results4.1. Endogeneity in Firms’ IT Governance ChoicesIt is erroneous to view IT governance as a propertyfirms have rather than choose. Almost all prior stud-ies, however, treat IT governance as exogenous (e.g.,Tiwana and Konsynski 2010, Xue et al. 2008). To iso-late the proposed theoretical relationships, we mustfirst econometrically account for factors that drivefirms’ IT governance choices but do not directly affectIT strategic agility. Not accounting for endogeneityassumes that firms’ IT governance choices are randomrather than rational choices. Recognizing IT governanceendogeneity also means that we eschew any assertionsabout normatively ideal IT governance structures (e.g.,the pervasive assertion that centralization is optimalfor IT infrastructure and decentralization for IT appgovernance).
Our analyses must econometrically account for threesources of endogeneity in IT governance: (a) selec-tion effects due to omitted variable bias, (b) existingknowledge integration mechanisms already in placeto facilitate JM’s transmission solution independentof peripheral knowledge, and (c) reverse causality(i.e., IT performance causing changes in IT governancestructure).
First, three sets of drivers of IT governance choiceobserved in prior IT governance studies must beaccounted for before any attempts to explain its conse-quences. These include (a) the strategic roles of IT for afirm, (b) properties of the firm’s internal organization,and (c) properties of the firm’s external environment.Strategic roles of IT include the strategic importanceof IT to the firm (Agarwal and Sambamurthy 2002,Sambamurthy and Zmud 1999), premising that greaterstrategic importance of IT will encourage firms to
centralize IT decisions, and potential for cross-unitIT synergies that encourage centralizing IT decisions(Brown and Magill 1998, Sambamurthy and Zmud1999). Internal properties include IT formalization, withthe rationale that greater formalization will lead firmsto increase centralization of IT decisions, and IT unitsize, with the rationale that larger IT units are morelikely to be centralized to ensure coordination (Brown1997, Sambamurthy and Zmud 1999). External environ-mental properties include industry dynamism, becausefirms in more dynamic industries are less likely tocentralize IT decisions (Brown 1997), and uncertainty,premising that greater uncertainty about IT needswill lead firms to decentralize IT decisions (Hann andWeber 1996). Information intensity and departmentalinterdependence do not affect IT governance choice(Brown 1997) and were therefore not included. Firmsthus choose how much to centralize IT governancebased on the strategic role of IT in their business model,their internal structure, and the firm’s external envi-ronment. Our model therefore culminates on the vastprior literature on IT governance choices.
Second, it is naive to assume that firms will not haveanticipated the knowledge separation problem poten-tially addressed by JM’s transmission solution andalready accounted for it in their IT governance choices.This requires accounting for knowledge integrationmechanisms already in place in a firm that mightserve as alternatives to peripheral knowledge to realizeJM’s transmission solution. We therefore use twoinstruments: the degree to which a firm actively scansand integrates knowledge (a) from outside its bound-aries in its IT activities (external knowledge integration)and (b) across departments within the firm (internalknowledge integration; Alonso et al. 2008, Mitchell 2006,Reich and Benbasat 2000).
Third, we must acknowledge the defining role thatpast IT performance might have played in shapingcurrent IT governance. Firms are likely to attempt tocorrect performance shortfalls or reinforce strong ITunit performance by further tweaking IT governancestructures. IT unit performance can therefore lead toreallocation of authority over IT decisions; this historicalreverse-causal explanation must directly be considered.We therefore use it as an instrument.
4.1.1. Econometric Modeling Approach for Endog-enizing IT Governance. The solution to account forendogeneity is Garen’s (1984) two-stage economet-ric technique. Unlike two-stage least squares (2SLS;which also produced consistent results), this approachpermits us to model unobserved heterogeneity overa range of IT governance choices and also allowsdrawing more nuanced theoretical inferences. Ouranalysis first accounted for IT governance endo-geneity (Stage 1) and then tested the hypotheses
(Stage 2). In Stage 1, we estimated two reduced-form “IT governance choice models” (Equations (1)and (2)) to construct endogeneity-correcting �s forboth IT app and infrastructure governance, which areincluded in the subsequent “strategic agility model”(Equations (3)–(5)).
4.2. Analysis
4.2.1. Stage 1: IT Governance Choice Model andEndogeneity Correction �s. We first evaluated whetherendogeneity was a concern in our model using theHausman (1978) endogeneity test. The results suggestedthat IT app decentralization was endogenous (Hausmant = 1094, p < 0005), but IT infrastructure centralizationwas not endogenous (Hausman t = 0043, n.s.). Forconsistency, we account for endogeneity in both models.(Appendix C demonstrates how failure to endogenizeleads to misleading conclusions.)
In this stage, we first estimated the endogenousvariables—IT app governance decentralization and ITinfrastructure governance centralization—and theirresiduals (�app and �inf) using the two governancechoice equations, (1) and (2). As instruments, we used(a) the six predictors of IT governance choice spanningstrategic, internal, and environmental drivers fromprior studies; (b) internal and external knowledgeintegration mechanisms; and (c) IT unit performance.Their definitions and guiding sources appear in Table 1.Stage 1 (Equations (1) and (2)) results appear in Table 4.(Stata 12 was used.)
IT app governance decentralization
=�0 +�1 ×industry_dynamism+�2 ×IT_unit_size
+�3 ×uncertainty+�4 ×strategic_importance_of_IT
+�5 ×cross-unit_IT_synergy+�6 ×IT_formalization
+�7 ×external_knowledge_integration
+�8 ×internal_knowledge_integration
+�9 ×IT_unit_performance+�3 (1)
IT infrastructure governance centralization
=�0 +�1 ×industry_dynamism+�2 ×IT_unit_size
+�3 ×uncertainty+�4 ×strategic_importance_of_IT
+�5 ×cross-unit_IT_synergy+�6 ×IT_formalization
+�7 ×external_knowledge_integration
+�8 ×internal_knowledge_integration
+�9 ×IT_unit_performance+�0 (2)
The results in Table 4 show that the models for both ITapp decentralization and IT infrastructure centralizationare significant, and that firms systematically choosetheir IT governance based on the strategic importance of
Performance IT unit performance −0002 (−0.20) −0006 (−0.54)R2 (%) 16.93 16.03R2
adj (model F) 8014% (1.93∗) 7014% (1.80∗)
Note. N = 105 firms.∗p < 0005; ∗∗p < 0001 (one-tailed test; significant values are bold).
IT to the firm, potential for cross-unit IT synergies, levelof IT formalization, uncertainty, and preexistence ofexternal knowledge integration mechanisms.7 However,firms’ IT governance choices are not shaped by industrydynamism, IT unit size, internal knowledge integration,or IT unit performance.
4.2.2. Stage 2: IT Strategic Agility Model Account-ing for IT Governance Endogeneity. Our concep-tualization of discriminating alignment correspondsto alignment-as-interaction in Venkatraman’s (1989)framework, which is appropriate when alignment hashigh theoretical specificity and is anchored to a specificcriterion variable (IT strategic agility) predicted by theinteraction between a small number of variables (ITgovernance and peripheral knowledge). Interactionterms are therefore used to test the two hypotheses.
We tested the hypotheses using a three-step hierar-chical weighted least squares (WLS) model with �appas a source variable in Stage 2. Ordinary least squares(OLS) estimation is inefficient because of heteroskedas-ticity caused by the dependence of the second stageerror term on the governance choice variables (Garen1984). WLS mitigates the subsequent risk of inefficientstandard errors affecting significance tests.
In this stage, controls were added to the model (Step1 in Table 5), then the main effects, and �app and �inffrom Stage 1, and the two � product terms described
7 Although the drivers appear to push IT governance in conflictingdirections in the “conflicting contingencies” view (Brown and Magill1998, Sambamurthy and Zmud 1999), our Stage 1 results paint a morenuanced picture when the IT governance concept is decomposedinto IT app and IT infrastructure governance; there is limited overlapin the drivers of the two classes of decision rights, and the ones thatinfluence both push them in opposite directions.
below (Step 2), and finally the mean-centered interac-tion terms (Step 3) to test the hypotheses (shaded cellsin Table 5) (�app is � from Equation (1) and �inf is �from Equation (2)). We centered the interaction termsto minimize multicollinearity; the highest varianceinflation factor was 5.02 in Step 3. Note that we usethe predicted values of IT app and IT infrastructuregovernance in Stage 2 (indicated by y, following Garen1984). By using predicted rather than observed valuesof IT governance choices, our model does not presumeany normatively ideal IT governance structure (e.g.,that app decisions ought to be decentralized and infras-tructure decisions centralized (e.g., Brown and Magill1994, Sambamurthy and Zmud 1999)). The stepwisemodel in Equations (3)–(5) was used to estimate ITstrategic agility with the results in Table 5; the modelwas significant in the second and the third steps
Notes. y , Predicted values of IT governance choice variables from Garen Stage 1 in Table 4. Significant values are in bold. The shaded cells highlight the hypothesistest results. N = 105 firms.
The � values obtained from Stage 1 are included inStage 2 to account for endogeneity while performingthe hypothesis tests. The product terms of the �s withthe predicted values of the two IT governance variablesare also included to account for unobserved heterogene-ity across the range of centralization/decentralizationchoices for the two classes of IT governance; theircoefficients indicate the direction of the endogeneitybias (Garen 1984, p. 1214).
In Table 5, the interaction effect between �app
and yapp_decentralization (Step 3) is significant (� = 0034,t-value = 2.60, p < 0001) in the full model. This revealsthe advantages of decentralizing IT app governance aswell as the penalties of centralizing it. It implies thatIT strategic agility suffers in firms that decentralize ITapp governance to a lesser degree than predicted. Bycontrast, the corresponding term for IT infrastructurecentralization is nonsignificant, suggesting that thereis no unobserved heterogeneity over the range of ITinfrastructure governance choices.
Our two hypotheses are tested in Step 3.8 Theinteraction between IT unit business knowledge andyapp_decentralization was positive and significant (�= 0041,t-value = 2.48, p < 0001), supporting Hypothesis 1. Theinteraction between line functions’ technical knowl-edge and yinfrastructure_centralization was positive and signif-icant (�= 0021, t-value = 1.90, p < 0005), supportingHypothesis 2. The nonsignificant interaction of thetwo peripheral knowledge variables suggests that, byitself, symmetrically shared knowledge between depart-ments does not enhance strategic agility (� = 0005,t-value = 0035). Furthermore, the nonsignificance of thetwo nonfocal interaction terms suggest that neithergreater business knowledge in the IT unit nor greatertechnical knowledge in the line functions respectivelycompensate for discriminating peripheral knowledgefor IT infrastructure and IT app governance. The modelexplained 43.3% (31.8% adjusted) of the variance inIT strategic agility, of which the interactions includ-ing discriminating alignment account for 23% (20%adjusted) of the explained variance. This explainedvariance, although substantial on its own, also sig-nals considerable opportunity to theoretically expandthe nomological network beyond our discriminatingalignment explanation.
To assess sample size adequacy, we analyzed effectsize and power at the 5% and 1% levels. The R2
increment due to the addition of interaction terms
8 Iacobucci (2008, p. 48) emphasizes that when the focus of the testsis on the interaction effects (as in our hypotheses), the significanceof the main effects is not of substantive interest. Therefore, thehypothesis tests do not require the main effects to be significant; nordoes our theory.
has a power of 0.99 at �= 0005 and 0.96 at �= 0001.The sample size required for the standard thresholdof power of 0.80 is 50 at �= 0005 and 67 at �= 0001.Our sample size exceeds these requirements. The valueof effect size (f 2) is 0.33, which approximates a large(f 2 > 0035) Cohen effect size.
Assessment of rival explanations. Since no prior empir-ical precursors exist for strategic agility as they dofor IT governance choice, we used a conservativeapproach of not reusing any of our nine instrumentalvariables as controls for rival explanations. (Addingthe nonsignificant instruments as control variablesdid not substantively change the results.) The onlycontrol variable we used for the second stage was ITintensity on the premise that firms investing a greaterproportion of revenues in IT will be better equipped tostrategically deploy IT more rapidly (Baschab and Piot2003, p. 48; Hann and Weber 1996).
4.3. Econometric Robustness TestsWe used the following econometric robustness tests:
1. Validity of model overidentifying restrictions. We usedBassman’s (1960) test for overidentifying restrictions.The Bassman f (0.39; p = 0082) was nonsignificant, indi-cating that the model was appropriately overidentified.
2. Instrument sufficiency tests. We used Anderson andRubin’s (1949) test, whose null is that the excludedinstruments are uncorrelated with the error term andcorrectly excluded from the equation. The Anderson-Rubin �2 was 2.00 (p = 0073; n.s.) suggesting thatthe set of instruments used in our model are validand sufficient. A Sargan test (�2 = 4075; p = 0031, n.s.)also independently confirmed the adequacy of ourinstruments.
3. 2SLS and WLS robustness. To ensure robustnessof our findings across estimation methods, we repli-cated the strategic agility model using (a) 2SLS and(b) WLS with �inf as an alternative source variable.Both hypotheses were consistently supported.
4. Selective endogeneity correction robustness. Since theHausman (1978) test did not indicate endogeneity in ITinfrastructure governance as it did for app governance,we replicated the analysis without �inf in Stage 2 of theGaren procedure. The results strongly supported bothdiscriminating alignment hypotheses (H1, � = 0039,t-value = 2.43, p < 0001; H2, �= 0020, t-value = 1.78,p < 0005), and other coefficients remained consistentwith Table 5.
5. Model robustness without reverse causality and com-mon methods bias. We reestimated the model without ITperformance as an instrument in Stage 1. All hypotheseswere supported (H1, �= 0045, t-value = 2.73, p < 0001;H2, �= 0025, t-value = 2.09, p < 0005), and the signif-icance for the remaining coefficients was consistentwith the reported results. Overall, this indicates that
the results are robust. We also failed to find evidenceof common methods bias.9
4.4. LimitationsThe results should be interpreted cognizant of threelimitations. First, our cross-sectional data cannot testcausation. Even though matched-pair data were used,line functions’ technical knowledge is measured as theIT manager’s perception of the construct. Collectingthese data from many line managers in each firmwas infeasible. Second, we did not explicitly controlfor firms’ outsourcing levels. Third, caution shouldbe exercised in extrapolating the results to larger orsmaller firms, given the midsized firms in the study.For example, would ERP systems more common inlarge firms constitute IT infrastructure as we view themor IT apps? To assess whether firm size systematicallybiases our results, we included firm size (only availablefor the 36 public firms in the data set) in both stagesof our model. It was nonsignificant as an instrument(suggesting that firm size does not influence IT gover-nance choice) and as a control (suggesting that largeror smaller firms are not systematically different intheir IT strategic agility).10 The correlation betweenfirm size and strategic agility was also nonsignificant(� = −0004; t = −0021, n.s.). Therefore, we failed to findevidence that the results would not generalize beyondthe midsized firms in our study.
5. DiscussionOur study was motivated by the increasingly strategicrole of IT but inattention to how firms can governIT to be strategic. Although over $4 trillion are spentannually on IT, some firms like Zara, Walmart, andDelta Air Lines strategically exploit IT better thantheir rivals. Prior IT governance studies have focusedprimarily on explaining firms’ IT governance choices(e.g., centralization versus decentralization), ratherthan their downstream strategic consequences. Ourunderstanding of how firms’ multifaceted IT gover-nance choices shape IT strategic agility thus remainsembryonic, even though firms aspire to it (e.g., Hann
9 Although our multi-informant design mitigates common methodsbias, we conducted two additional tests. First, the marker variabletest (using the theoretically unrelated public firm dummy and surveyidentifier number as the two marker variables) showed averagecorrelations of 0.24 (n.s.) and 0.11 (n.s.) with the principal constructs.Second, a single-factor test showed no single dominant factor in theEFA in Appendix B (the first factor explained 10.5% of the total79.4% variance).10 We also added firm size as a control variable (which was notincluded in Table 5 due to missing data for private firms) andobtained consistent results (H1, �= 0047, t-value = 2.30, p < 0005;H2, �= 0032, t-value = 2.08, p < 0005). Using firm size as an instru-ment and both as an instrument and control yielded consistentresults.
Figure 3 Interactions for High and Low (±1SD) Levels of Interdepartmental Peripheral Knowledge
6.3(a) (b)
5.7
5.2
4.7
IT s
trat
egic
agi
lity
4.2
3.6
3.1
1.3 2.1 3.0 3.9
IT applications governance decentralization
4.8 5.7 6.5
IT infrastructure governance centralization
IT s
trat
egic
agi
lity
IT unitbusiness
knowledge
3.8
4.2
4.5
4.9
5.3
5.6
6.0
1.5 2.4 3.3 4.3 5.2 6.1 7.0
HighLow
Line functions’technical
knowledge
HighLow
and Weber 1996, Sambamurthy et al. 2003, Segars andGrover 1998). Our premise was that differences in howIT activities are governed within firms can help explaindifferences in IT strategic agility across them.
Our middle-range theory developed the idea thatfirms’ IT governance choices foster IT strategic agilityonly when their alignment with departments’ periph-eral knowledge is discriminating—discriminating inthat only a specific department’s peripheral knowledgeinduces agility for a specific class (apps or infrastruc-ture) of IT decisions; which department has peripheralknowledge must be aligned with which departmentmakes what IT decisions.
Noteworthy in our theory development are decom-position of predominantly monolithic conceptualiza-tions of IT governance and incorporation of knownexplanations of IT governance choice in our econo-metric analyses. Tests of the proposed ideas usingmatched-pair data from 105 firms have two substantivetheoretical implications for both the IT governanceliterature and broader organization theory.
5.1. Contributions and Theoretical Implications
5.1.1. Discriminating Alignment. Our distinctivecontribution is that IT governance enhances IT strate-gic agility only when it is discriminatingly alignedwith departments’ peripheral knowledge. We useda singular theoretical apparatus for both classes ofIT decisions. Moving decision rights to the locus ofrelevant knowledge (JM’s delegation solution) mini-mizes agility-impeding delays from interdepartmentalknowledge transfer. IT apps and infrastructure relyprimarily on different types of knowledge concen-trated in different departments, so JM’s delegationsolution nudges them in opposite directions on the
centralization–decentralization continuum.11 However,nudging IT app governance toward decentralizationand IT infrastructure governance toward centralizationalso separates them from dispersed complementaryknowledge that either needs. This requires a tandemuse of JM’s transmission solution, which peripheralknowledge facilitates. Discriminatingly aligned periph-eral knowledge in the IT unit and line functions thenrespectively increase the marginal returns to IT strategicagility from a firm’s governance choices for IT appsand infrastructure.
This finding has two implications for the IT gover-nance literature. First, firms’ demand that IT activitiessimultaneously be economical yet fine-tuned to linefunctions’ strategic needs appears hopelessly contradic-tory and unrealistic (Cramm 2010, p. 107; Feld andStoddard 2004). This dreaded tension might simply bean artifact of confounding the two classes of IT assets intheory. It vaporizes when we theoretically disaggregatethe pervasive monolithic conception of IT governanceinto two distinctly governed classes of IT decisions.Such bifurcation permits simultaneous pursuit of twoseemingly contradictory demands: Firms can governIT infrastructure to realize scale economies and focusIT apps on the myriad line functions’ diverse needs.Appreciating what is being governed must thereforeprecede the quest for the Holy Grail of “good” ITgovernance.
Figure 3 illustrates discriminating alignment. Fig-ure 3(a) shows that decentralizing IT app governanceenhances IT strategic agility only when the IT unit’s
11 Prior studies’ inconsistent use of IT governance to refer to justone class of IT decisions (e.g., Brown and Magill 1998, Brown 1997,Olson and Chervany 1980) or lumping together both classes (Hannand Weber 1996, Lewis and Byrd 2003, Ross 2002, Tiwana andKonsynski 2010, Weill et al. 2002) makes it difficult to infer thissubtlety.
peripheral knowledge is high (the solid line; +1 SD),but not when it is low (the dotted line; −1 SD). Thus,contrary to admonitions to decentralize IT app gover-nance (Cramm 2010, p. 25; Heller 2012, p. 105; Weilland Ross 2004, p. 75), decentralization alone is impotentunless the IT unit has greater peripheral knowledge.Figure 3(b) shows that centralizing IT infrastructuregovernance enhances IT strategic agility only whenthe line functions’ peripheral knowledge is high (thesolid line), but not when it is low (the dotted line).Sambamurthy and Zmud’s (1999) caution that line func-tions lacking technical knowledge will be unpreparedfor IT decentralization thus applies primarily to ITinfrastructure governance. This implies that only in anuanced combination are IT governance and peripheralknowledge powerful enablers of IT strategic agility,revealing the inseparability of the previously uncon-nected research streams on firms’ IT governance choices(e.g., Brown and Magill 1994, 1998; Sambamurthyand Zmud 1999) and IT–line shared knowledge (e.g.,Bassellier et al. 2003, Nelson and Cooprider 1996, Reichand Benbasat 2000).
The merits of granularizing IT governance intoapps and infrastructure remained invisible until wedeveloped their theoretical properties, even though theconceptual distinction is widespread (e.g., Agarwaland Sambamurthy 2002, Ross 2002, Weill and Ross2005). Such disaggregation by itself induces IT strategicagility (indicated by the significant infrastructure×appsproduct term in Step 3 in Table 5; �= 0018, p < 0005).Thus, centralizing one class of IT decisions amplifies thebenefits from decentralizing the other. Lumping themunder the rubric of IT governance penalizes agilityby unnecessarily forcing firms to covary them. This isillustrated by the 2×2 heat map in Figure 4, which usesfour median-split subgroups representing high andlow levels of delegation of IT app and infrastructuregovernance. Delegation in this heat map is as predicted byour theory and respectively represents centralization of ITinfrastructure and decentralization of IT app governance.Three inferences can be drawn from the patterns inthis heat map.
• Complete centralization or decentralization. Central-izing all IT decisions (the lower right cell; 5.31) out-performs only decentralizing all of them (upper leftcell; 5.02). IT strategic agility is most penalized whenfirms delegate app decisions (i.e., decentralize them)but not infrastructure (i.e., do not centralize it). Thisimplies that decentralizing all IT governance is theworst possible approach to governing IT assets in termsof IT strategic agility. The pattern illustrated jointlyby the top left and bottom right cells is that firms areworse off when they delegate one class of IT decisionsbut not the other.
• Centralizing one class of IT decisions and decentral-izing the other class. Any more nuanced approach to
Figure 4 A 2 × 2 Heat Map of IT Strategic Agility Consequences fromDelegating IT App and Infrastructure Governance
5.02
5.84
5.52 5.31
(Centralized)(Decentralized)
Non-delegated Delegated
IT infrastructure decision rightsIT
app
s de
cisi
on r
ight
s
Non
-del
egat
edD
eleg
ated
(cen
tral
ized
)(d
ecen
tral
ized
)
Note. Delegation represents centralization of IT infrastructure and decentraliza-tion of IT app governance.
governing IT outperforms a completely centralizedor completely decentralized approach, as illustratedby the bottom left and top right cells of Figure 4. Putdifferently, not covarying IT app governance and ITinfrastructure governance is associated with higherlevels of IT strategic agility.
• A discriminating approach to governing IT assets. Adiscriminating alignment approach (top right cell ofFigure 4; 5.84) posited by our theory outperforms anondiscriminating approach (bottom left cell; 5.52).Delegation of both classes of IT decisions thereforeresults in the highest IT strategic agility. The smalldifference across IT strategic agility across the top rightand bottom left cells can potentially culminate overtime into larger competitive differences due to thepath-dependent nature of technology evolution.
For example, consider how Coca-Cola disaggregatesIT app from IT infrastructure governance to fosterstrategic agility. The IT app embedded in Coke’s inno-vative 125-flavor Freestyle soda dispensers collectssales data on individual Coke brands at each loca-tion. This allows leasing stores to alter the productmix at individual stores multiple times a day, usinganalytics data provided by Coke. However, Coke isable to economically leverage the app for real-timemarket experiments only because a robust, central-ized IT infrastructure also seamlessly integrates appsembedded in the geographically dispersed dispensersto Coke’s analytics group in Atlanta. Coke’s IT gover-nance approach puts it squarely in the upper right cellin Figure 4.
Second, endogenizing historically exogenized ITgovernance choices also reveals insights into the
consequences of IT governance over- and undercentral-ization, besides building on the vast IT governancechoice literature without presuming any normativeideal. Note that the endogeneity of IT app decentraliza-tion is a conditional effect because �app is nonsignificantin Step 2 but its interaction term is significant in Step 3in Table 5. This conditional effect implies that firms arebetter off erring towards over-decentralizing ratherthan over-centralizing IT app governance because theformer amplifies the effect of that choice on IT strate-gic agility. No such inference can be drawn about ITinfrastructure governance.
For the broader literature on organization design(e.g., Alonso et al. 2008, Athey and Roberts 2001,Dessein et al. 2010), this finding offers new insights intohow the underused lever of decision rights allocationcan resolve an enduring tension in organizing func-tional activities. The tension is between centralization’sadvantage of firmwide synergies and decentralization’sadvantage of being functionally bespoken (Desseinet al. 2010). Separating decision rights into two theoret-ically distinguishable classes with primarily firmwideconsequences vis-à-vis departmental consequences—asillustrated by IT infrastructure and apps here—permitsthe same department to achieve firmwide synergywhere possible without sacrificing functional bespoken-ness where critical. Our results suggest centralizingfunctional decisions that affect the entire firm, butdecentralizing those that affect only some functions.
5.1.2. Governance-Contingent Value of PeripheralKnowledge. Our second contribution is moving pasta simplistic notion of shared knowledge to show thegovernance-contingent nature of which departmentneeds peripheral knowledge. Increasing shared knowl-edge across two departments appears at odds withintrafirm division of labor. It is both distracting andcostly because it contradicts departmental specializationand overlooks plausible asymmetry in departmentalknowledge overlaps. Our middle-range theory eschewsthe oversimplified symmetric notion of shared knowl-edge. Consider an analogy. For the reader to interactwith a Japanese person, it is useful to know when theJapanese person must be able to speak English andwhen you must be able to speak Japanese; the benefitsof both becoming bilingual might not be commensu-rate with costs. This is exactly what increasing sharedknowledge between IT and line functions entails.
The implication for the IT governance literature isthat the department that needs to speak the other’slanguage depends on which department makes whatIT decisions. The benefits of a particular department’speripheral knowledge are governance contingent. Bycontrast, prior studies overlook a firm’s IT governancechoices in encouraging increasing business acumen ofIT staff (Bassellier and Benbasat 2004; Heller 2012, p. 94),line managers’ IT skills (Bassellier et al. 2003, Weill and
Aral 2006), and shared knowledge between IT and linefunctions (Nelson and Cooprider 1996; Ross 2002; Weilland Ross 2004, p. 68). The value of shared knowledgeis discriminating, not unconditional. Our results alsoimply that the IT unit’s business knowledge by itself ismore important than previously recognized (note itspositive main effect in Table 5), unlike line functions’technical knowledge, whose benefits depend entirelyon their possession of IT app decision rights. Firmsfare better erring toward overinvesting in increasingthe IT unit’s business knowledge but underinvesting inline functions’ technical knowledge, as the post hocanalyses in Appendix D further illustrate.
This offers a new insight for broader organizationtheory that recognizes the necessity of specializationfor productive intrafirm division of labor even thoughit breeds ignorance across departments (Becker andMurphy 1992, Jensen and Meckling 1992). Ignoranceitself can be a valuable asset worth protecting (Tiwana2008). Our results imply that firms can selectivelyreduce cross-departmental ignorance just enough toconduce collaboration without imperiling departmentalspecialization.
Our findings raise three questions that merit futureresearch. First, firms appear to lag in correcting ITunit underperformance by reorganizing IT governance(performance → IT governance reverse causality wasabsent); how can they reduce this lag? Second, whatintervening mechanisms—treated as a black box in ourtheory—help explain the link between discriminatingalignment and IT strategic agility? Third, the inter-play of IT apps with IT infrastructure is theoreticallyuncharted. If a firm’s IT portfolio were envisionedmetaphorically as a pizza, IT infrastructure is the crustand IT apps its toppings. It is the apps that com-petitively differentiate one firm’s IT portfolio fromanother’s, just as toppings differentiate one pizza fromanother. IT infrastructure—like a pizza’s crust—neversubstantively differentiates. Yet, a good enough ITinfrastructure is a necessary foundation to run theseapps, just as a good enough crust is needed for apizza.
In conclusion, our middle-range theory of discrim-inating IT governance scratches only the surface inshifting attention from monolithic toward more multi-faceted conceptions of IT governance. As firms investan expanding proportion of their capital into IT, ITmanagers’ ability to translate these investments into astrategic advantage is increasingly inseparable fromtheir prosperity.
Appendix A. MeasuresThe respondents were instructed that line functions referredto non-IT departments of their organization, such as sales,purchasing, manufacturing, human resources, accounting,and finance. Scale end-point anchors were 1, strongly disagree,
and 7, strongly agree, unless noted otherwise. Items droppedduring scale refinement are indicated by an asterisk.
IT app governance decentralization was measured using fouritems that tapped into how decision-making responsibilitiesfor the following IS activities were distributed in the organi-zation: (1) applications planning, (2) initiating new projects,(3) managing key projects, and (4) application developmentactivities. The anchors were as follows: 1, fully vested withIS unit; 4, shared between IS unit and line functions; 7, fullyvested with various line functions.
IT infrastructure governance centralization was measuredusing four items that tapped into how decision-makingresponsibilities for the following IS activities were distributedin the organization: (1) communications and networking,(2) IT operations and maintenance, (3) procurement of hard-ware/software, and (4) end-user support. The anchors were asfollows: 1, fully vested with various line functions; 4, sharedbetween IS unit and line functions; 7, fully vested withIS unit.
IT unit’s business knowledge was measured using six itemsthat tapped into the overall extent to which members of theIT unit understood the following about the organization:(1) its day-to-day business routines, (2) business rules andheuristics, (3) business opportunities and threats, (4) thecompany’s business strategy, (5) a “big picture” of the com-pany’s business, and (6) a holistic understanding of thecompany’s business. The anchors were as follows: 1, not atall; 4, somewhat; 7, to a great extent.
Line functions’ technical knowledge was measured usingseven items that tapped into the overall extent to whichthe company’s line managers understood the following:(1) systems design, (2) database structures, (3) programminglanguages, (4) IT project methodologies, (5) software testingand debugging, (6) data processing procedures, and (7) appli-cation development tools. The anchors were as follows: 1, notat all; 4, somewhat; 7, to a great extent.
IT strategic agility was measured using four items fromSegars and Grover’s (1998) strategic IS scale that tappedinto the extent to which the IT unit had been successful inaccomplishing the following: (1) identifying IT opportunitiesto support the strategic direction of the firm, (2) educatingtop management on the importance of IT, (3) assessing thestrategic importance of emerging technologies, and (4) adapt-ing technology to changing business needs. Dropped itemswere (5) coordinating IT initiatives in various departments∗,(6) identifying and resolving potential sources of resistanceto IT plans∗, (7) maintaining open lines of communicationwith other departments∗, and (8) avoiding overlapping devel-opment of major systems∗. Anchors were as follows: 1, veryunsuccessful; 7, very successful.
Uncertainty was measured using four items derived fromHann and Weber (1996) that tapped into the degree to whichthe key activities performed by the IT unit (1) followed aclearly known approach, (2) followed an understandablesequence of steps, (3) relied on established procedures andpractices, and (4) were guided by a clearly defined body ofknowledge. The anchors were as follows: 1, strongly agree;7, strongly disagree.
Strategic importance of IT was measured using four itemsadapted from Hann and Weber (1996) that tapped into the
degree to which, in the respondent’s organization, informa-tion systems (1) were key to competitiveness, (2) critical tobusiness success, (3) were of great strategic importance, (4)served only administrative purposes (reversed)∗, and (5)were critical to marketplace adaptability.
Cross-unit IT synergy was measured using four itemsbased on Brown and Magill (1998, p. 182) that assessedthe extent to which substantial potential benefits exist incentrally coordinating the following across all departmentsin the organization: (1) programming tools, (2) applicationsplanning∗, (3) project management, (4) systems developmentactivities, and (5) systems development resources.
Industry dynamism was measured using three items thattapped into the degree to which (1) the organization’s majorcompetitors in its industry were continually devising newstrategies, (2) the organization’s major competitors in itsindustry were continually introducing new products, and(3) technological breakthroughs had made possible numerousnew products in its industry.
IT formalization was measured using three items adaptedfrom Ranganathan and Sethi (2002) that tapped into theextent to which the IT unit (1) had clearly documented jobdescriptions for all staff, (2) used task forces and committeesto handle critical issues∗, (3) used operating rules and pro-cedures for decision making, and (4) extensively relied onstandard operating procedures such as rules, policies, andforms.
IT unit performance was measured using four items thatassessed how the organization’s IT unit generally compared inrelation to comparable IT units the respondent had observedin terms of the following in the work that it produced forthe organization’s line functions: (1) quality, (2) efficiency,(3) timeliness, (4) adherence to budgets∗, and (5) overallperformance. The end-point anchors were as follows: 1, muchworse; 7, much better.
External knowledge integration was measured using threeitems that assessed the extent to which the IT unit actively uti-lized information about the following in its ongoing activities:(1) new market developments, (2) new technical develop-ments, (3) emerging technologies, (4) competitors’ use of IT∗,and (5) ideas from business partners (e.g., suppliers, vendors,distributors)∗. The scale anchors were as follows: 1, not at all;4, somewhat; 7, to a great extent (new scale).
Internal knowledge integration was measured using threeitems that assessed the extent to which members of the ITunit and line functions (1) have a shared agenda∗, (2) buildon each other’s ideas, (3) engage in joint problem solving,(4) recognize each other’s constraints∗, (5) draw on eachother’s skills and expertise, and (6) have developed a sharedunderstanding about the role of IT in the organization∗. Thescale anchors were as follows: 1, not at all; 4, somewhat; 7, toa great extent (new scale).
IT unit size was measured by the number of individualsemployed in the organization’s IT unit.
IT intensity was measured as the percentage of the organi-zation’s gross revenue that was allocated to the IT unit.
Appendix C. Errors from Neglecting IT GovernanceEndogeneity
Failing to account for the endogeneity of a firm’s IT gover-nance choices results in spurious conclusions, as illustratedby the OLS model in Table C.1, which does not correctfor endogeneity and uses observed rather than predictedvalues of the two IT governance variables. As the shadedcells in the endogeneity-neglecting OLS results in Table C.1show, the interaction between IT app decentralization andIT infrastructure centralization is nonsignificant (�= −0000,t-value = −0002), completely missing the advantages of disag-gregating them. They also downplay the importance of the ITunit’s business knowledge (� of 0.27 in Table C.1 as opposedto 0.31 in Table 5, both significant at p < 0001). This highlightsthe problem in attempting to study the consequences ofIT governance centralization/decentralization without firstcumulatively accounting for how the IT governance choicepredictors from prior IT governance studies endogenize it.
Appendix D. What Peripheral Knowledge ShouldFirms Err Toward Overinvesting and Underinvestingin?
Additional slope analysis (Figure D.1) suggests that firmsshould err toward overinvesting in the IT unit’s businessknowledge and underinvesting in line functions’ technicalknowledge. In Figure D.1, the dotted line with a steeperslope illustrates that the influence of the IT unit’s businessknowledge on strategic IT agility is more pronounced whenthe line functions’ technical knowledge is lower than when itis higher (the solid line). Firms are therefore better off erringtoward overinvesting in increasing the IT unit’s businessknowledge while underinvesting in increasing line functions’technical knowledge for enhancing IT strategic agility.
Figure D.1 Variation in Line Functions’ Peripheral Knowledge Does NotSubstantively Improve IT Strategic Agility When IT UnitBusiness Knowledge Is High (±3 SD)
5.9
5.6
5.3
5.0
4.7
IT s
trat
egic
agi
lity
4.5
4.2
2.8 3.5 4.2 4.9
IT unit’s business knowledge
5.6 6.3 7.0
Linefunctions’technical
knowledge
HighLow
ReferencesAgarwal R, Sambamurthy V (2002) Principles and models for
organizing the IT function. MIS Quart. Executive 1(1):1–16.Aghion P, Tirole J (1997) Formal and real authority in organizations.
J. Political Econom. 105(1):1–29.Alonso R, Dessein W, Matouschek N (2008) When does coordination
require centralization? Amer. Econom. Rev. 98(1):145–179.Anderson T, Rubin H (1949) Estimators of the parameters of a single
equation in a complete set of stochastic equations. Ann. Math.Statist. 20(1):570–582.
Andersson H, Kaplan J, Smolinski B (2012) Capturing value from ITinfrastructure innovation. McKinsey Quart. (October):1–10.
Aral S, Weill P (2007) IT assets, organizational capabilities, andfirm performance: How resource allocations and organizationaldifferences explain performance variation. Organ. Sci. 18(5):763–780.
Athey S, Roberts J (2001) Organizational design: Decision rights andincentive contracts. Amer. Econom. Rev. 91(2):200–205.
Baldwin C (2008) Where do transactions come from? Modularity,transactions, and the boundaries of firms. Indust. CorporateChange 17(1):155–195.
Baschab J, Piot J (2003) The Executive’s Guide to Information Technology(Wiley, New York).
Bassellier G, Benbasat I (2004) Business competence of informationtechnology professionals: Conceptual development and influenceon IT-business partnerships. MIS Quart. 28(4):673–694.
Bassellier G, Benbasat I, Reich B (2003) The influence of businessmanagers’ IT competence on championing IT. Inform. SystemsRes. 14(4):317–336.
Bassman R (1960) On finite sample distributions of generalizedclassical linear identifiability test statistics. J. Amer. Statist. Assoc.55(292):650–659.
Basu A (2008) How to tap IT’s hidden potential. Wall Street J.(March 10) http://www.wsj.com/articles/SB120467900166211989.
Becker G, Murphy K (1992) The division of labor, coordination, andknowledge. Quart. J. Econom. 107(4):1137–1160.
Bester H, Krähmer D (2008) Delegation and incentives. RAND J.Econom. 39(3):664–682.
Brown C (1997) Examining the emergence of hybrid IS governancesolutions: Evidence from a single case site. Inform. Systems Res.8(1):69–94.
Brown C (1999) Horizontal mechanisms under differing IS organiza-tion contexts. MIS Quart. 23(3):421–454.
Brown C, Magill S (1994) Alignment of the IS function with theenterprise: Toward a model of antecedents. MIS Quart. 18(4):371–403.
Brown C, Magill S (1998) Reconceptualizing the context-design issuefor the information systems function. Organ. Sci. 9(2):176–194.
Brynjolfsson E, Schrage M (2009) The new, faster face of innova-tion. Wall Street J. (August 17). http://www.wsj.com/articles/SB10001424052970204830304574130820184260340.
Bucciarelli L (1994) Designing Engineers (MIT Press, Cambridge, MA).Cramm S (2010) 8 Things We Hate About IT (Harvard Business Press,
Boston).Demsetz H (1992) Comments: Specific and general knowledge, and
Lewis B, Byrd T (2003) Development of a measure for the informa-tion technology infrastructure construct. Eur. J. Inform. Systems12(2):93–109.
Lu Y, Ramamurthy K (2011) Understanding the link between infor-mation technology capability and organizational agility: Anempirical examination. MIS Quart. 35(4):931–954.
McGrath R (2013) The End of Competitive Advantage (Harvard BusinessPress, Boston).
Messerschmitt D, Szyperski C (2003) Software Ecosystem (MIT Press,Cambridge, MA).
Mitchell V (2006) Knowledge integration and information technologyproject performance. MIS Quart. 30(4):919–939.
Nault B (1998) Information technology and organization design:Locating decisions and information. Management Sci. 44(10):1321–1335.
Nelson KM, Cooprider JG (1996) The contribution of shared knowl-edge to IS group performance. MIS Quart. 20(4):409–429.
Olson M, Chervany N (1980) The relationship between organizationalcharacteristics and the structure of the information servicesfunction. MIS Quart. 4(2):57–68.
Ranganathan C, Sethi V (2002) Rationality in strategic informationtechnology decisions: The impact of shared domain knowledgeand IT unit structure. Decision Sci. 33(1):59–86.
Reich B, Benbasat I (2000) Factors that influence the social dimensionof alignment between business and information technologyobjectives. MIS Quart. 24(1):81–111.
Ross JW (2002) Six IT decisions your IT people shouldn’t make.Harvard Bus. Rev. 80(11):84–95.
Sambamurthy V, Zmud R (1999) Arrangements for technologygovernance: A theory of multiple contingencies. MIS Quart.23(2):261–290.
Sambamurthy V, Bharadwaj A, Grover V (2003) Shaping agilitythrough digital options: Reconceptualizing the role of IT incontemporary firms. MIS Quart. 27(2):237–256.
Segars A, Grover V (1998) Strategic information systems planning:An investigation of the construct and its measurement. MISQuart. 22(2):139–164.
Star S, Ruhleder K (1996) Steps toward an ecology of infrastructure:Design and access for large information spaces. Inform. SystemsRes. 7(1):111–134.
Tiwana A (2008) Does interfirm modularity complement ignorance?A field study of software outsourcing alliances. Strategic Manage-ment J. 29(11):1241–1252.
Tiwana A, Keil M (2007) Does peripheral knowledge complementcontrol? An empirical test in technology outsourcing alliances.Strategic Management J. 28(6):623–634.
Tiwana A, Konsynski B (2010) Complementarities between organiza-tional IT architecture and governance structure. Inform. SystemsRes. 21(2):288–304.
Venkatraman N (1989) The concept of fit in strategy research: Towardverbal and statistical correspondence. Acad. Management Rev.14(3):423–444.
Weill P, Aral S (2006) Generating premium returns on your ITinvestments. Sloan Management Rev. 47(2):39–48.
Weill P, Ross J (2004) IT Governance: How Top Performers Manage ITDecision Rights for Superior Results (Harvard Business SchoolPress, Boston).
Weill P, Ross J (2005) A matrixed approach to designing IT governance.Sloan Management Rev. 46(2):26–34.
Weill P, Subramani M, Broadbent M (2002) Building IT infrastructurefor strategic agility. Sloan Management Rev. 44(1):57–65.
Xue L, Ray G, Gu B (2011) Environmental uncertainty and ITinfrastructure governance: A curvilinear relationship. Inform.Systems Res. 22(2):389–399.
Xue Y, Liang H, Boulton W (2008) Information technology governancein information technology investment decision processes: Theimpact of investment characteristics, external environment, andinternal context. MIS Quart. 32(1):67–96.
