WHY DO FIRMS BOTH MAKE AND BUY? AN ... 2014...the make or buy decision. However, none of these...

Strategic Management JournalStrat. Mgmt. J., 28: 285–311 (2007)

Published online in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/smj.580

Received 20 December 2004; Final revision received 18 July 2006

WHY DO FIRMS BOTH MAKE AND BUY? ANINVESTIGATION OF CONCURRENT SOURCINGANNE PARMIGIANI*Lundquist College of Business, University of Oregon, Eugene, Oregon, U.S.A.

Transaction cost economics, neoclassical economics, and the firm capabilities literatures proposetheories of the firm that typically depict firm boundaries determined by a dichotomous choice:the make or buy decision. However, none of these theories presents a satisfying explanation asto why firms would concurrently source, i.e., simultaneously make and buy the same good. Thisstudy combines these organizational economics theories and compares when firms make, buy,and concurrently source through surveying small manufacturing firms. Support was shown foraspects of all three theories, with evidence indicating that concurrent sourcing is a distinctlydifferent choice, rather existing along a make/buy continuum. Copyright 2007 John Wiley &Sons, Ltd.

INTRODUCTION

As strategy scholars, we are most interested inthe behavior and performance of firms. Despitedecades of research on theories of the firm, westill struggle with the fundamental question of howfirms determine their boundaries. Firms ostensi-bly create their boundaries through procurementdecisions by choosing which goods to produceinternally; therefore most theories of the firm viewthe sourcing decision as a dichotomous choice: tomake or to buy (Williamson, 1975; Perry, 1989;Grant, 1996). However, firms can and do simul-taneously make and buy the same good, a phe-nomenon this paper terms concurrent sourcing.1 To

Keywords: concurrent sourcing; make-or-buy decisions;strategic sourcing; transaction cost theory; firm capabili-ties; vertical integration∗ Correspondence to: Anne Parmigiani, Lundquist College ofBusiness, University of Oregon, Eugene, OR 97403, U.S.A.E-mail: [email protected] I use the term ‘concurrent sourcing’ to specifically refer to onlybackward, partial vertical integration of a homogeneous good (or

institute concurrent sourcing, the firm must incurboth the costs of securing capital, allocating plantand equipment capacity, staffing, and coordinationthat accompany internal production as well as thecosts of finding, selecting, negotiating with, andmaintaining external suppliers. Once both inter-nal and external sources are in place, managingthese simultaneously can be challenging owing tothe natural comparisons, suspicions, and shirkingthat can occur between the two sources (Hen-nart, 1993). Given that concurrent sourcing is morecostly to set up and manage, why would firmsselect this sourcing mode over solely making orsolely buying? This paper attempts to answer that

service) by a single firm. The term ‘partial integration’ can referto either forward or backward integration or some combinationof these (e.g., making, buying, and selling a particular good;Porter, 1980). Likewise, Harrigan’s term of ‘taper integration’does not specifically refer to backward integration and is appliedat a broad and diverse unit of analysis (the strategic business unit;Harrigan, 1984). ‘Concurrent sourcing’ emphasizes that firms aremaking and buying the same good, in contrast to considering abroader unit of analysis and/or one with more heterogeneity.

Copyright 2007 John Wiley & Sons, Ltd.

286 A. Parmigiani

question both theoretically, by combining transac-tion cost theory, neoclassical economics, and thefirm capabilities literature, and empirically, by ana-lyzing survey data from small manufacturing firms.As such, this paper contributes to the strategy lit-erature by explicitly incorporating the effects of allthree theories in a careful empirical study, demon-strating the firm’s desire to simultaneously monitorsuppliers, produce efficiently, and improve pro-cesses.

A second and related research question that thesetheories help address is whether concurrent sourc-ing is better represented as a linear combination ofmaking and buying along a make/buy continuumor as a distinct choice that uses two different sourc-ing modes simultaneously.2 That is, whether thefirm’s primary decision is its degree of integrationor its choice to either make, buy, or concurrentlysource. The three theories mentioned above do notagree on this decision. Transaction cost economicsviews the governance choice as placed upon a con-tinuum, with market and hierarchy as the anchors(Williamson, 1985). Neoclassical economics andthe capabilities views, however, support a distinctchoice view such that a small degree of making (orbuying) can provide significant benefits. If alonga continuum, then the main decision by the firmis the percentage to produce internally; the forcesmotivating the firm to produce more would alsomotivate it to purchase less. However, if concur-rent sourcing is a distinct choice, the key firmdecision is choosing this option over solely makingor solely buying. The forces that motivate the firmtoward making may not be the same as those moti-vating it away from buying. It could be that forcespushing the firm away from making and away frombuying motivate it to concurrently source, as thelesser of the three evils. Alternately, forces maypush the firm toward making and toward buying,thus motivating it to concurrently source to gainthe different benefits of both.

These ideas relate to discussions in the litera-ture on plural forms and the distinctions between

2 A related issue involves how to define concurrent sourcing.An absolute approach would consider producing 1 percent ofthe firm’s requirements while outsourcing the rest as beingconcurrent sourcing. However, if one is interested in the typicalmode of sourcing a good, a less stringent definition is preferable.Following prior research, this study uses a threshold of 10percent for both conceptual and empirical reasons. While someinformation may be lost by using this threshold, that whichremains is more accurate and robust. For more details, see thedependent variable section.

governance modes. Firms that concurrently sourcesimultaneously use the governance modes of mar-ket and hierarchy. Bradach and Eccles have termedthis a plural governance mode, in which ‘distinctorganizational control mechanisms operate simul-taneously for the same function by the same firm.’They offer the following argument as to why firmswould use such a mode: ‘Contracting is problem-atic without in-house experience, and the maladiesassociated with hierarchy are widely recognized.The remedy for these difficulties may be the simul-taneous use of the two mechanisms, creating inessence competition between them’ (Bradach andEccles, 1989: 113; emphasis in original). It is theconcurrent use of these two mechanisms, not theextent of one or the other, which provides the ben-efits to the firm. This connects to findings fromPoppo and Zenger (1998), who proposed two dif-ferent cost functions for making and for buyingand found that ‘these two governance forms pos-sess distinctly different capacities to cope withor exploit various exchange attributes.’ Answer-ing the question of whether concurrent sourcingis a continuous or a discrete choice contributesto the field by enabling a deeper understandingof the differences between various hybrid gover-nance modes and by clarifying different theoreticalviewpoints.

Economists have provided some explanationswhy firms would concurrently source. Early workby Adelman suggested that firms concurrentlysource in times of demand uncertainty, pushingthe fluctuations in volume onto suppliers in orderto ensure full internal capacity and stable produc-tion (Adelman, 1949). Later economic scholarsconcurred with this view and posited that firmswill also concurrently source to gain an increasedunderstanding of the production process and thusbetter monitor suppliers (Cannon, 1968; Harris andWiens, 1980; Porter, 1980). Harrigan conductedempirical work investigating relationships amongstrategic business units and found evidence thatsuccessful units concurrently sourced when uncer-tainty was high and the potential for synergies wassignificant (Harrigan, 1986).

More recently, marketing scholars have inves-tigated dual distribution, in which firms simulta-neously use in-house and independent sales chan-nels, analogous to concurrent sourcing. They havefound firms use dual distribution in the presenceof performance uncertainty (Dutta et al., 1995),

Copyright 2007 John Wiley & Sons, Ltd. Strat. Mgmt. J., 28: 285–311 (2007)DOI: 10.1002/smj

Why do Firms Both Make and Buy? 287

market heterogeneity (Sa Vinhas, 2002), and infor-mation asymmetries (Heide, 2003). Franchisingscholars have also positioned the choice to bothown and franchise units as a decision to use dualdistribution in order to balance incentive and con-trol issues, as well as facilitate learning (Bradach,1997; Lafontaine and Shaw, 2005). Similar find-ings have been shown in the trucking industry(Nickerson and Silverman, 2003; He and Nicker-son, 2006).

These explanations do not fully take into accountattributes of the good being sourced or characteris-tics of the buying and supplying firms, all of whichaffect the sourcing choice. Theories that addressthese factors and thus could be included wheninvestigating this question are transaction cost eco-nomics, neoclassical economics, and the firm capa-bilities view. Although an influential theory forinvestigating strategic issues and a quintessentialtheory of vertical integration, transaction cost eco-nomics has been intriguingly silent on the questionof concurrent sourcing. This is curious, especiallygiven the scores of papers exploring hybrid formsand investigating the make/buy decision that gen-erally have supported transaction cost logic (forreviews see Crocker and Masten, 1996; Rindfleischand Heide, 1997; Boerner and Macher, 2002; andDavid and Han, 2004). Although transaction costtheory views the sourcing modes on a continuum,many of these empirical studies depict a dichoto-mous choice to make or to buy, sometimes forcingthe issue and combining concurrent sourcing withone of these two choices. For example, in the clas-sic work by Monteverde and Teece (1982), theydefine ‘make’ as when the firm produces 80 per-cent or more of its requirements and ‘buy’ as whenthe firm produces less than this amount; thus, manyof their goods were actually concurrently sourced(Bradach and Eccles, 1989). Likewise, firm capa-bility theories have neglected concurrent sourc-ing. These theories suggest that firms will con-duct competence-related activities internally andoutsource other activities (Prahalad and Hamel,1990), thus addressing firm boundaries, but they donot explicitly address when firms may both makeand buy.

Investigating the nature of the concurrent sourc-ing decision assists in our understanding of hybridmodes of organizing. Many theoretical lenses,including transaction cost economics and the capa-bilities view, have been used to explore hybridgovernance modes, which combine aspects of

both market and hierarchy. But confusion existsgiven the myriad of hybrid forms which includesalliances, joint ventures, supply chain networks,relational contracting, and franchising, among oth-ers (Hodgson, 2002). Concurrent sourcing repre-sents a simple and clean hybrid sourcing mode,involving a single firm and a single good, so itsstudy can help resolve some of this confusion.3

Supporting the need for a study that would dis-tinguish between the continuum and dual views,Dutta and colleagues suggested that a ‘congenialcontext for such a study (of sourcing modes) wouldbe an industrial purchasing decision where buy-ers engage in buy-only, make-only, and make-plus-buy choices . . . (since) data including allthree forms would thus afford a clearer separa-tion of hybrid form (continuum) effects from dualform effects’ (Dutta et al., 1995: 203). Steensmaand Corley concur with this position, advocatingfor ‘discrete analyses of governance mode deci-sions . . . conducted between the use of marketand hybrid and between hybrid and hierarchy’(Steensma and Corley, 2001: 288).

This study answers these calls and, througha holistic approach, integrates several theoreticalperspectives and builds on work that has combinedtheories but has assumed a dichotomous sourcingchoice (e.g., Argyres, 1996; Poppo and Zenger,1998; Combs and Ketchen, 1999; Steensma andCorley, 2001; Leiblein and Miller, 2003). Throughthis approach, the effects of both production andtransaction costs can be better understood, asfirms simultaneously strive to produce efficiently,improve their processes, and monitor suppliers.This approach also allows a comparison of theo-ries that imply a continuum view (transaction costeconomics) with those that imply a discrete choiceview (neoclassical economics and capabilities). Assuggested above, this study employs an industrialpurchasing context by analyzing survey data fromsmall manufacturing firms regarding their sourcingdecisions for production tooling and services. Thisunique and fine-grained dataset of over 800 obser-vations includes occurrences of all three sourc-ing modes as well as the percentage of internalproduction where applicable. The continuous vs.discrete choices can therefore be empirically exam-ined using different modeling techniques. Aspects

3 While alliances and other more complex hybrid modes areimportant, this study focuses on concurrent sourcing since it is amode available to nearly all firms in most industries, includingsmall firms in mature environments.


288 A. Parmigiani

from all three theoretical perspectives were shownto affect the sourcing choice, with greater supportshown for the discrete over the continuum view.

The next section of this paper provides thetheoretical background and hypotheses for sourc-ing choices, describing how transaction cost eco-nomics, neoclassical economics, and the capabili-ties view explain why or under what circumstancesfirms would concurrently source. An empiricalsection follows that describes the research design,context, and methodology. The results segmentdiscusses the findings of several models and howthese relate to the hypotheses. A final sectionpresents a summary, practical implications, andsuggestions for future work.

THEORIES AND HYPOTHESES

All of the literature used in this paper is rootedin economics but each stream proposes distinctarguments as to why a firm would concurrentlysource. The theoretical background and hypothesesin this section will suggest specific relationshipsbetween this sourcing strategy and attributes of thegood, the environment, suppliers, and the sourc-ing firm. Each theory and its associated predictionsalso implies either a continuous or discrete view,addressing the related research question of whetherconcurrent sourcing is a mid-point on the make-or-buy continuum or whether it is a distinct anddiscrete choice. For a given variable, if the firmcan gain the benefit of concurrent sourcing evenwith a slight percentage of its requirements out-sourced (or insourced), then the discrete view issupported over the continuum view. This view isalso less stringent than a strict continuum, whichpositions concurrent sourcing in between makingand buying, thereby intermediate levels of explana-tory factors lead to this result. The argumentsbelow, along with empirical modeling techniques,strive to unravel these issues.

Transaction cost economics and concurrentsourcing

Transaction cost economics (TCE) has been estab-lished as a dominant lens to view firm boundarydecisions. In this theory, the firm considers theex ante and ex post costs of exchange as the pri-mary determinant of whether to conduct an activ-ity internally or externally, as these are distinct

governance structures (Coase, 1937; Williamson,1975). Due to opportunism and bounded ratio-nality, asset specificity and uncertainty are keytransaction cost drivers, as they increase the costsof market exchange, motivating the firm to pro-duce internally (Williamson, 1975, 1985). TCEhas been applied broadly in the strategy, man-agement, and organization theory literatures, aswell as in non-business disciplines. Boerner andMacher offer an extensive review of over 600papers based upon TCE principles demonstratingthe wide range of questions investigated, frominternational entry modes for multinational corpo-rations, to rock band membership and prenuptialagreements (Boerner and Macher, 2002). They,along with other scholars, find the tenets of TCE tobe generally supported empirically, particularly theinfluence of asset specificity (see also Rindfleischand Heide, 1997). David and Han’s review of 63empirical papers was less sanguine in supportingTCE, but still find considerable explanatory powerfor this theory in answering its canonical questionand most frequently examined dependent variable:the make-or-buy decision (David and Han, 2004).Various forms of hybrid arrangements have beenalso studied using TCE, including alliances andjoint ventures (e.g., Oxley, 1997, 1999), franchis-ing (e.g., Minkler and Park, 1994), relational con-tracting (e.g., Heide and John, 1990; Dyer, 1997),and network forms (e.g., Eccles, 1981).

Given the wealth of contexts and questions thatTCE has addressed, it is surprising that scholarshave not investigated the rather obvious questionof why firms would concurrently source. One rea-son for this neglect could be the assumption of amake-or-buy continuum, such that the key variableof interest is the percentage produced internally.Williamson describes this continuum as ‘discretemarket exchange at the one extreme to central-ized hierarchical organization at the other, withmyriad of mixed or intermediate modes filling therange in between’ (Williamson, 1985: 16). Giventhe key dependent variables in TCE, this suggestsa fairly linear relationship, such that the greaterthe degree of asset specificity (or uncertainty), thegreater the percentage of the firm’s requirementsit would produce internally. This view also sets upmarket and hierarchy as mutually exclusive, withforces pushing it toward one form pulling it awayfrom the other, and thus not considering why afirm may choose to use both of these governanceforms simultaneously (Dutta et al., 1995).



TCE does consider the possibility of a firmsourcing both internally and externally whereby afirm produces more custom goods internally andoutsources the more generic ones. As describedby Williamson, ‘where firms are observed to bothmake and buy an identical good or service, theinternal technology will be characterized by higherasset specificity than will be external technology,ceteris paribus’ (Williamson, 1985: 96). But thesegoods aren’t truly identical if produced by differenttechnologies. Suppose the good is homogeneousand the firm is purposefully splitting the volumerequirements between their internal facility andoutside suppliers. The TCE/continuum view doesnot address this possibility and, moreover, cannotdistinguish this case of using market and hierarchysimultaneously from the case of using these twogovernance modes for two related goods.

Another reason why TCE may have neglectedconcurrent sourcing is its emphasis on transactioncharacteristics. As such, TCE implies that concur-rent sourcing would result in cases of intermedi-ate asset specificity. In this situation, Williamsonsuggests that we will observe ‘mixed governance,in which some firms will be observed to buy,others to make, and all expressing “dissatisfac-tion” with their current procurement solution’(Williamson, 1985: 94). As in the continuum caseabove, this theory doesn’t address the possibil-ity of the same firm both making and buying thesame good, which may lead to less dissatisfac-tion. TCE stresses transaction costs as the driverfor vertical integration and suggests the firm would‘never integrate for production cost reasons alone’(Williamson, 1985: 94). This may be true for fullintegration, but the case of partial integration, i.e.,concurrent sourcing, is not as clear.

Perhaps aspects of production costs, such aseconomies of scale, are influential in this deci-sion. It is important to consider these factors andeither control for them or incorporate them intothe theoretical framework. In this paper, hypothe-ses presented in later sections predict the effectsof firm and supplier scope economies, as well asfirm and supplier expertise, all of which will affectproduction costs. While not hypothesized but latercontrolled for, goods with a high minimum effi-cient scale and high volume requirements would beunlikely candidates for concurrent sourcing since itwould be inefficient to split production over mul-tiple suppliers.

As one of the key determinants of transactioncosts, asset specificity should affect the choice toconcurrently source. Asset specificity refers to thedegree of idiosyncrasy of an investment required toproduce the good for the sourcing firm; the moreidiosyncratic the investments required, the morelikely the firm will prefer to produce the gooditself since the costs of protecting against poten-tially opportunistic suppliers is greater than thecost of producing internally (Williamson, 1975).TCE theory suggests a continuum view such thatfor higher degrees of asset specificity the percent-age of requirements insourced would increase up tothe point of full integration. This positive relation-ship has been supported empirically in scores ofstudies (see Crocker and Masten, 1996; Rindfleischand Heide, 1997; Boerner and Macher, 2002; andDavid and Han, 2004; for reviews). What has notbeen investigated is the prediction that concur-rently sourced goods would be moderate in theirdegree of specific investment. This logic leads tothe first hypothesis:4

Hypothesis 1tce: The greater the asset specificityof the good, the higher the percentage of itsrequirements the firm will produce internally.Therefore, moderately asset-specific goods willbe concurrently sourced.

Uncertainty also affects the firm’s sourcing choice.Uncertainty includes both the potential for environ-mental change and the unpredictability of a part-ner’s behavior (Williamson, 1985). Greater uncer-tainty can lead to adaptation problems and to dif-ficulties in evaluating performance, both of whichmay motivate the firm to internalize the activity,since hierarchical authority enables better coor-dination and monitoring, as well as protectionagainst supplier opportunism. Transaction cost the-ory suggests that greater uncertainty in conjunctionwith a non-trivial level of asset specificity willlead to increased vertical integration, but empir-ically the findings have been mixed (Rindfleischand Heide, 1997). One reason for this may be

4 Each hypothesis number includes a subscript representing itstheoretical connection (‘tce’ for TCE, ‘neo’ for neoclassical eco-nomics, and ‘cap’ for capabilities). When the same variable isemployed by different theories, the hypothesis number is thesame, but the subscript differs. For example, the volume uncer-tainty hypothesis connected to TCE logic is Hypothesis 2tce,while the volume uncertainty hypothesis connected to neoclas-sical economics is Hypothesis 2neo.


290 A. Parmigiani

that uncertainty consists of a number of distinctconstructs, such as demand unpredictability, envi-ronmental volatility, and measurement difficulty(Sutcliffe and Zaheer, 1998; Leiblein and Miller,2003; David and Han, 2004). A better understand-ing of how each of these types of uncertaintyaffects the sourcing decision can be reached byexploring separate hypotheses for each.

Volume uncertainty is one type of environmentaluncertainty that should affect the sourcing deci-sion. This uncertainty refers to unpredictability ofdemand and an accompanying inability to accu-rately forecast and schedule production. Volumeuncertainty will make it more difficult to con-tract with suppliers since volume requirements willdirectly affect their costs, which will generallybe lower when volume is smooth and predictable(Pindyck and Rubenfeld, 1995). Volume uncer-tainty will also lead to more misunderstandingsand inventory coordination problems, increasingthe need to communicate and impede adaptation(Williamson, 1985); thus, firms will produce moreof their requirements internally in this situation.In addition, suppliers will be less likely to investin process improvements since fluctuating volumeswill complicate the pay-offs from such invest-ments. Following this logic, concurrently sourcedgoods will be characterized by moderate levelsof volume uncertainty. Restated, this is the nexthypothesis:

Hypothesis 2tce: The greater the good’s vol-ume uncertainty, the higher the percentage ofits requirements the firm will produce internally.Therefore, goods with moderate levels of volumeuncertainty will be concurrently sourced.

Another type of environmental uncertainty thatcan affect a firm’s sourcing decision is when thetechnological future is uncertain. The TCE lit-erature is split on whether greater technologicaluncertainty will promote internalization or out-sourcing (Mahoney, 1992). On one hand, greatertechnological change will result in more significantadaptation and coordination challenges, leading toa higher degree of internalization (Williamson,1985). However, significant technological uncer-tainty will also raise the risk of obsolescence andtherefore firms would prefer to not invest, lettingsuppliers take this risk instead (Balakrishnan and

Wernerfelt, 1986). Moreover, technological uncer-tainty can drive the firm away from internaliza-tion and away from outsourcing, complicating aTCE prediction which is based upon a continuum(Poppo and Zenger, 1998). Empirical findings havealso been mixed (Rindfleisch and Heide, 1997).Therefore, no TCE hypothesis will be made forthis variable.

Performance uncertainty, or measurement diffi-culty, also affects a firm’s sourcing decision. Thisuncertainty refers to the difficulty in predictinghow a good will perform in the firm’s subse-quent production processes. The firm will havean information disadvantage relative to its poten-tially opportunistic suppliers if it knows little aboutupstream processes (Alchian and Demsetz, 1972;Barzel, 1982). Performance uncertainty will begreater for more complex goods, especially thosethat involve multiple components and technolo-gies, since complexity increases the difficulty inevaluating quality through inspection prior to use(Coles and Hesterly, 1998; Bensaou and Ander-son, 1999; Novak and Eppinger, 2001). If qualityis difficult to evaluate, then monitoring a supplier’scompliance will be problematic, as will determin-ing when to enforce sanctions. TCE logic sug-gests that greater internalization will solve thesemonitoring problems and remove the potential forsupplier opportunism, since the firm’s hierarchi-cal structure will improve information flows andcoordinate incentives. Thus, concurrently sourcedgoods would have a moderate level of perfor-mance uncertainty. These arguments support thefinal TCE-based hypothesis:

Hypothesis 3tce: The greater the performanceuncertainty of the good, the higher the per-centage of its requirements the firm will pro-duce internally. Goods with a moderate levelof performance uncertainty will be concurrentlysourced.

Neoclassical economics and concurrentsourcing

The standard neoclassical economic explanationfor concurrent sourcing involves hedging againstdemand uncertainty. In this case, a firm can keepits internal plant at full production by using sup-pliers to handle fluctuating additional volumes,thereby running more efficiently due to havingthis flexibility in capacity (Adelman, 1949; Carl-ton, 1979; Porter, 1980). This position assumes a



robust spot market, consisting of a large numberof qualified external suppliers vying for the firm’sbusiness, although these suppliers will have higherbase costs (Adelman, 1949). The actual prices theycharge the firm may be even higher, due to the riskthey are bearing by having unused capacity duringslack times and by not knowing when the ‘lowprobability’ demand will occur (Carlton, 1979).Indeed, suppliers may charge premiums for lowervolumes and short lead times since they know theyare merely ‘overflow outlets’ for the firm (Harri-gan, 1984; Hill, 1994). Firms may be willing to paythese premiums rather than invest in additional,and potentially underutilized, capacity.

The percentage of a firm’s requirements thatwould be satisfied by its internal production maynot necessarily depend upon volume uncertaintyper se, but perhaps upon other aspects of thefirm’s production processes and equipment, suchas the minimum efficient scale or demand for otherproducts that share these resources. In the over-flow case above, when firms are making the goodand are near capacity at the usual demand levels,then volume uncertainty would result in a smalldegree of taper or a relatively small amount of out-sourcing relative to internal production. But, moregenerally, the target internal production percentagecould be great or small, since this depends uponother aspects of the production process. The keyis that neither completely making nor completelybuying will sufficiently resolve the uncertainty. Incontrast to the TCE argument that proposes thefirm is motivated toward greater making in casesof volume uncertainty to protect against supplieropportunism, this line of reasoning suggests thatthe firm is motivated both toward making, to fullyutilize capacity, and toward buying, for flexibilityin meeting demand. This implies a discrete choicerather than a continuum view. Restated, these argu-ments suggest the next hypothesis:

Hypothesis 2neo: The greater the good’s volumeuncertainty, the more likely the firm will concur-rently source.

Neoclassical economists also suggest that firmswould use concurrent sourcing to reduce infor-mation asymmetry, such as that caused by per-formance uncertainty. By using both internal andexternal suppliers, the sourcing firm will learnmore about production technology and have greateraccess to cost-saving measures (Adelman, 1949;

Harris and Wiens, 1980; Porter, 1980; Dutta et al.,1995). The firm will gain an enhanced understand-ing of the production process in terms of qualityand be in a better position to spur both internal andexternal suppliers to improve their offerings (Can-non, 1968; Harrigan, 1984; Heide, 2003). Firmscan also use concurrent sourcing as a sanctioningdevice since it enables the firm to credibly threatento switch suppliers by either totally vertically inte-grating or completely outsourcing its requirements,thus disciplining both internal and external suppli-ers (Hennart, 1993). The firm does not need toproduce a majority of its requirements in order togain this understanding and be well positioned tomonitor suppliers; a pilot plant producing smallquantities can suffice (Oster, 1994). Therefore, asin the case of volume uncertainty and again con-trary to TCE, there is not a continuous relationshipbetween the degree of performance uncertainty andthe percentage of internalization. In this case, thefirm is motivated both to make, to better under-stand the good, and to buy, to be able to bench-mark. Thus, another hypothesis results:

Hypothesis 3neo: The greater the performanceuncertainty of the good, the more likely the firmwill concurrently source.

Neoclassical economics emphasizes the produc-tion function and its associated costs as the moti-vator for a firm’s sourcing decision (Perry, 1989).One relevant characteristic of this function is thepotential for scope economies, which involve areduction in overall production costs from produc-ing two goods simultaneously, leading to a fullerutilization of sharable upstream inputs (Panzar andWillig, 1981). These sharable inputs include spe-cialized equipment and human capital, neither ofwhich is sufficiently fungible to easily sell excesscapacity in established markets, due to a lack ofpotential customers (Teece, 1982). The greater thescope economies for the firm, the less the marginalcost of production of a particular good, since pro-ducing this good in concert with its usual port-folio of products reduces the firm’s total costs.Therefore, a firm would be motivated to producesuch goods internally. This also suggests a con-tinuous and linear relationship between a firm’sscope economies and its percentage of internaliza-tion. Likewise, if suppliers can enjoy significantscope economies, this will be reflected in lowerprices that will motivate the firm to outsource.


292 A. Parmigiani

In other words, there will be a continuous andnegative relationship between the suppliers’ scopeeconomies and the firm’s internal production.

However, the firm considers both its and its sup-plier’s scope economies in sourcing decisions. Thefirm and supplier may experience different levelsof scope economies, since these depend upon theother goods each one produces. They may use dis-similar production processes and thus each has adifferent type of excess sharable input that theywish to more fully utilize (Barney, 1991; Helfatand Eisenhardt, 2004). For example, a suppliermay have excess capacity on a particular machinewhile the firm may have slack internal engineeringresources, both of which could be better utilized bythe firm and the supplier producing the good. Inthis way, scope economies can provide symmet-ric incentives for the firm and supplier. Therefore,if both the firm and the supplier could reducetheir production costs through scope economies,the firm would be motivated to make and to buyin order to enjoy both lower internal costs and rel-atively low supplier prices. This leads to the nexthypothesis:

Hypothesis 4neo: The greater scope economiesfor both the firm and its suppliers to produce thegood, the more likely the firm will concurrentlysource.

Capabilities and concurrent sourcing

The capabilities view (e.g., Teece, Pisano, andShuen, 1997; Eisenhardt and Martin, 2000), incor-porating the resource- and knowledge-based viewsof the firm, complements the above two literaturesby considering how attributes of the firm and itssuppliers affect the sourcing decision. A firm willproduce goods that are close to its area of exper-tise, core to its business, and related to items italready produces, as it uses past experience andresources as ‘stepping stones’ into related areas(Wernerfelt, 1984; Barney, 1986; Prahalad andHamel, 1990). This expertise is broader and deeperthan just the sum of the firm’s prior experience inproduction as it incorporates its understanding ofthe base technology and the firm’s related skills.Every firm is different, so some goods will be abetter fit with its resource or knowledge base thanothers. If the good is a poor fit, it will be moreefficient to outsource (Rubin, 1973; Kogut andZander, 1992; Conner and Prahalad, 1996; Grant,

1996). Suppliers’ costs will depend upon their rel-ative expertise, resources, and capabilities and thuseach individual supplier will offer a somewhat dif-ferent blend of price, delivery, quality, and otherattributes for the sourcing firm to consider (Pen-rose, 1959; Barney, 1991). This suggests a contin-uum such that the greater a firm’s expertise abouta good, the greater its degree of internalizationand the greater the supplier’s expertise, the greaterthe degree of outsourcing. If either the firm or thesupplier has relatively greater expertise, then inter-nalization or outsourcing, respectively, will result(Jacobides, 2004). However, if both the firm andthe supplier have significant expertise, the firm willbe motivated both to make, to take advantage ofits own expertise, and to buy, to learn from sup-pliers. This case is analogous to the neoclassicalcase of combined scope economies with concur-rent sourcing being a logical choice, since the firmis motivated to both make and buy. This logic sup-ports the next hypothesis:

Hypothesis 5cap: The greater the expertise ofboth the firm and its suppliers, the more likelythe firm will concurrently source.

When a firm concurrently sources, learning willbe enhanced, since it gains both the deep tacitknowledge of internal production and the broader,more diverse understanding from external supplyrelationships. Making the good improves the firm’sunderstanding, since it gains a deeper tacit knowl-edge of the good and its processes (Kogut and Zan-der, 1992; Darr, Argote, and Epple, 1995; Grant,1996). The knowledge and skills the firm accu-mulates through this ‘learning by doing’ cannotbe replicated through outside supply relationships(Pisano, 1994). Buying the good provides vicariouslearning from the outside suppliers who are con-nected through customer relationships to the firm’scompetitors and to firms in other industries. Sincesuppliers may not necessarily provide their goodsat lower prices, a key benefit of concurrent sourc-ing is the ‘net gain in the know-how and the tradeconnections’ (Adelman, 1949: 116). This sourcingmode allows the firm access to both external sup-pliers’ research and technology developments andto its own internal knowledge of the good and itsrelated production processes (Porter, 1980; Harri-gan, 1984; Cassiman and Veugelers, 2006).



While it is clear how concurrent sourcing bene-fits the buying firm, it is not as obvious why suppli-ers would choose to be part of this relationship. Insome cases, the buyer may be large, powerful, andnearly a monopsonist, forcing the supplier to par-ticipate. This is the base case for much of the neo-classical economics literature, since it implicitlyassumes robust markets, such that the larger firmcan always find suppliers upon which to offloadtheir fluctuations in demand (Porter, 1980). Butmany commercial markets are relatively thin, withbuyers and suppliers having more equal power.Suppliers may not necessarily view the buyer’sinternal production as threatening, but rather asa signal that it is trying to gain a tacit under-standing of the good. This is knowledge that thesupplier already possesses. Suppliers could viewthe buyer’s internal production unit as an attemptto create competition, since it could not obtainthe desired quality and cost levels through pit-ting existing suppliers against each other (Harri-gan, 1985). This can give suppliers a sense ofreassurance and an indication of their significantbargaining power, especially if they are fairly wellestablished in the market and the buyer is justbeginning its own production of the good.

Moreover, suppliers benefit by having knowl-edgeable customers who are better positioned toevaluate supplier offerings (Lincoln, Ahmadjian,and Mason, 1998). Just as suppliers are a use-ful source of knowledge for customers, so too arecustomers for their suppliers (von Hippel, 1988).This motivates suppliers to participate in this bilat-eral relationship, since they can learn vicariouslyfrom the firm’s internal production unit. Theirshared understanding provides a foundation uponwhich the firms can learn from each other (Tunisiand Zanfei, 1998). It also provides a depth ofunderstanding and facilitates knowledge transferbetween the firm and its suppliers due to theincreased similarity of the firms’ knowledge bases(Lane and Lubatkin, 1998). Since the firm and sup-pliers both can gain from complementary knowl-edge, their incentives are aligned, suggesting thatlearning could drive the firm to choose concurrentsourcing and the supplier to participate. Firms canalso swap managerial and technical innovationsbetween internal manufacturing units and suppli-ers, improving both (Bradach and Eccles, 1989).

Learning is imperative when the progress oftechnology is difficult to predict. Firms will needto both gather a broad range of knowledge about

the technology and have the capacity to under-stand, interpret, and act upon whatever changesoccur (Cohen and Levinthal, 1990). By concur-rently sourcing internally and externally, firms willhave a wider range of knowledge sources andadaptive responses, enabling them to exploit theircumulative knowledge and explore a broader setof technologies from suppliers (March, 1991). Thefirm’s knowledge base will become more diverse,potentially overcoming inertia and relying solelyon one technological approach (Sorensen and Stu-art, 2000). If a firm cannot accurately predict thetype of change forthcoming, having both types ofsourcing available will improve the firm’s likeli-hood to succeed by being ‘ambidextrous’ and ableto deal with both suppliers and internal develop-ment groups in the face of technological change(Afuah, 2001). Since suppliers also face thesechanges, they too will want to be connected to thesourcing firm, making the incentives symmetric.

Another way to consider the value of learningfrom internal and external sources is through a realoptions framework (Bowman and, Hurry 1993;Trigeorgis, 1995; Mahoney, 2005). By having bothinternal and external suppliers, the firm gainsthe option to switch between them, resulting ingreater process flexibility (Kulatilaka and Marks,1988). In times of greater technological change anduncertainty, these options become more valuableand therefore concurrent sourcing should be morelikely (Sa Vinhas, 2002). However, the firm cangain these options and understand the environmentsufficiently by outsourcing (or insourcing) a smallamount of its requirements, implying a discreterather than a continuous view. Thus, the benefitsof learning from having two streams of knowledgeprovide the basis for the final hypothesis:

Hypothesis 6cap: The greater the good’s techno-logical uncertainty, the more likely the firm willconcurrently source.

Taken together, these hypotheses advance orga-nizational economic theory by providing a morerealistic and holistic approach to how firms makesourcing decisions. These predictions incorporatetransaction cost and production cost effects byincluding aspects of asset specificity, various typesof uncertainty, qualities of the good’s techno-logical production process, and the expertise ofboth the firm and its suppliers. Some of these


294 A. Parmigiani

Table 1. Summary of predictions and results

Hypothesis and variable Prediction (if high) Logic Models Resultsa

H1tce Asset specificity Produce greater %internally

Make to protect againstsupplier opportunism

1, 2, 3, 4 Not supported

Models 5, 6 supportMake > Buy andCS > Buy

H2tce Volume uncertainty Produce greater %internally

Make to better able tocoordinate and adapt

1, 2, 3, 4 Not supported

H3tce Performanceuncertainty

Produce greater %internally

Make to align incentivesby using authority

1, 2, 3, 4 Supported

Models 5, 6 supportMake > Buy andMake > CS

H2neo Volume uncertainty Concurrentlysource

Make to fully utilizecapacity

5, 6 Not supported

Buy to gain flexibilityH3neo Performance

uncertaintyConcurrently

sourceMake to better specify and

evaluate5, 6 Not supported

Buy to benchmarkH4neo Firm and supplier

scope economiesConcurrently

sourceMake to enjoy lower costs 5, 6 Supported

Buy to gain lower prices Firm: Make > CS >Buy

Supplier: Make < CS< Buy

Combined: Make >Buy and CS > Buy

H5cap Firm & supplierexpertise

ConcurrentlySource

Make to leveragecompetencies

5, 6 Supported

Buy to learn fromsuppliers

Firm: Make > CS >Buy

Combined: Make <Buy and CS >Make

H6cap Technologicaluncertainty

ConcurrentlySource

Make to understand andinterpret

5, 6 Supported

Buy to gain diverse views CS > Make

a CS, concurrently source.

hypotheses suggest a continuum, while others pro-pose a discrete choice, with concurrent sourcingnot necessarily in between making and buying.Table 1 summarizes the predictions and previewsthe results. The following empirical study alongwith ordinal and discrete modeling techniquesassist in providing evidence both about when firmswould concurrently source and about the nature ofthis decision.

RESEARCH DESIGN

The context for this study involves the sourcingdecisions of metal stamping and powder metal

firms for production tooling and services. Thesetwo sectors of the metal forming industry con-sist of numerous, independent, mature, small firms.The attributes of these firms help to rule outalternative explanations for concurrent sourcing,such as inertia, corporate parent influence, or slackresources (Penrose, 1959). Both the firms andtheir suppliers are relatively small, which assistsin controlling for explanations of sourcing strat-egy based upon a power differential, such as alarge, powerful buyer that can act without con-sidering a supplier’s response, since it knowsthe supplier will comply (Porter, 1980). Thesefirms rarely use buyer/supplier alliances or othermore sophisticated mechanisms, simplifying the



sourcing choice. These firms are predominantlynon-union, which controls for labor contracts thatcan restrict a firm’s sourcing options (Argyres andLiebeskind, 1999). Both of these sectors sharefairly homogeneous production processes and thetechnology involved is relatively mature, enablingbetter identification and characterization of thegoods and a more conservative test of the hypothe-ses.

After on-site interviews with managers from 11metal forming firms, a mail survey was createdto collect data on the sourcing decisions of fiveproduction-related goods: die design, die building,die maintenance, end-part machining, and end-part surface coating. These goods were commonto all firms, covered all three sourcing modes,were strategically significant, and were sourcedrelatively often. The 24-page survey booklet con-tained six sections, one for data on each good andone for overall firm information, with most itemsusing a seven-point ‘true/untrue’ scale (Fowler,1995). Four stages of pre-testing included eval-uations by academic colleagues, managerial inter-views which incorporated cognitive interviewingtechniques (Campanelli, 1997), reviews by indus-try association executives, and a pilot test withmanagers which replicated final survey conditions(Babbie, 1990). In all, 10 survey revisions weremade from its initial development through the finalmailing in fall 2002. Initial lists of firms wereobtained from the respective industry associations(Precision Metalforming Association and MetalPowder Industry Federation). Screening calls weremade to verify address information and to identifythe best respondent, typically the plant or generalmanager. After these calls, a viable mailing listof 453 firms, 366 in stamping and 87 in powdermetal, was established. The survey packet includedcustomized cover letters co-signed by the industryassociation executive and a window decal incen-tive. Follow-up messages were made by fax, elec-tronic mail, and phone, reflecting a mixed modedesign, with between two and six contacts per firm(Dillman, 2000).

Variable operationalization

Dependent variable: Sourcing mode choice

This variable reflects the firm’s procurement choicefor the good over the last year. For questionsthat seek to understand behavior, including a time

period is essential (Fowler, 1995). Prior work hasoften asked respondents to nominate goods; thiscan result in a selection bias because respondentstend to mention the most recent, important, or oth-erwise memorable decision (e.g., Bottum, 1992;Heriot and Kulkarni, 2001). By asking all firmsabout the same five goods, this bias is elimi-nated. The dependent variable was measured usingone item that asked if the firm sourced this iteminternally, externally, or concurrently (both inter-nally and externally; see the Appendix for theitem used). For this last option, the percentageof internal production was obtained; rather thanrelying entirely on a respondent’s ability to remem-ber this percentage, six choices were provided.This improves accuracy by cueing the respondent’smemory, pushing him or her to consider the dif-ferent options and choosing the best one, thusreducing response distortion (Fowler, 1995).

For the analysis, the answer to this dependentvariable item was interpreted in two ways. To testthe continuum view, each of the eight options wasplaced in order from 0 to 100 percent producedinternally and an ordered logit model was used.For the discrete view, three categories were cre-ated for goods that were made, bought, and con-currently sourced and a multinomial logit modelwas used. Consistent with prior work (e.g., Har-rigan, 1986; Sa Vinhas, 2002), a 10 percent cut-off was used such that items that were producedinternally 90 percent or more often were consid-ered ‘made,’ those that were outsourced 90 percentor more often were considered ‘bought,’ with therest considered concurrently sourced. While per-haps intuitively pleasing to assign the cases of 1to 99 percent internally sourced into the concur-rent sourcing category, there are several problemswith this approach. First, respondents may havebeen hesitant to classify a good as being com-pletely insourced or outsourced since they mayhave recalled those few unusual cases in whichthe other mode was used. Since we are interestedin the typical mode of sourcing, it makes senseto ignore these odd cases and assign the overallchoice to the usual mode (make or buy). More-over, respondents may have had trouble recallingwhether absolutely all requirements were producedinternally or outsourced, and so chose the concur-rent sourcing option with a very small (or large)percentage to compensate for their memory andappease the researcher (Fowler, 1995). A thirdproblem is econometric because the data were not


296 A. Parmigiani

well balanced by type of good and sourcing mode.For example, only three of the 169 cases of partcoating were produced 100 percent internally andonly seven of the 164 cases of die maintenancewere 100 percent outsourced. This is termed asparse cell count problem and, if left uncorrected,can bias parameter estimates in multinomial logitmodels (Agresti, 1996). By using the 90 percentcut-off and reclassifying a relatively small numberof the observations (82 out of 805), we can havemore confidence in the multinomial logit mod-els that compare making, buying, and concurrentsourcing.

Independent variables

All of the independent variable items were mea-sured on a seven-point Likert-type scale. Items foreach variable are listed in the Appendix.

Asset specificity. This variable was operational-ized as market thinness, since goods requiringhighly specific assets often result in a small num-ber of willing suppliers (Williamson, 1985). Whenthere are few capable suppliers, switching is diffi-cult and costly. Three items relating to this conceptwere adapted from prior work and included in thesurvey (Walker and Weber, 1984; Heide and Weiss,1995; Poppo and Zenger, 1998).

Volume uncertainty. Volume uncertainty wasmeasured by asking respondents about forecastinaccuracies and unpredictability in volume pat-terns. Two items adapted from prior work wereincluded on the survey to measure this attribute(John and Weitz, 1988).

Technological uncertainty. This variablemeasures the likelihood that technological changewill occur. This industry is fairly mature, so mea-suring the likelihood of change was more relevantthan measuring its magnitude. Goods based uponmature technologies and stable processes will beless prone to this form of uncertainty as they areunlikely to change significantly in the future. Thelikelihood of technological change can refer to theinnovation potential for either the good or the pro-cesses used in its production. Three items fromprior work were adapted to measure this variable(Heide and Weiss, 1995; Bensaou and Anderson,1999).

Performance uncertainty. While difficulties inestimating downstream performance have some-times been operationalized in terms of complex-ity, this is not as relevant in this context. Diesmay consist of scores of components and seem-ingly be quite complex, but if the die can be eas-ily described to suppliers and accurately evaluatedfor quality, then downstream performance can beassured. A better operationalization for this contextmeasures information asymmetry, such as whensimple inspection techniques are not adequate toevaluate quality, when production problems occurthat cannot be traced to a specific cause, and whenit is difficult to compare goods from different sup-pliers. Five items were developed to measure thisattribute, some of which were adapted from priorwork (Anderson and Schmittlein, 1984; Anderson,Glenn and Sedatole, 2000; Bottum, 1992; Duttaet al., 1995).

Firm and supplier scope economies. Scopeeconomies for the firm will be independent fromthose of its suppliers since each has its own estab-lished product mix and resource base. Two distinctvariables were created to estimate firm and supplierscope economies. Measures reflected the extent towhich overall costs were reduced by producing thegood along with its other products (Dutta et al.,1995). Firm scope economies were measured bytwo items, while supplier scope economies weremeasured by one item.

Firm and supplier expertise. Variables were cre-ated for both firm and supplier expertise, reflect-ing the extent to which either has considerableskills and capabilities for producing the good andan understanding of the underlying technology.Due to different experience bases, firm and sup-plier expertise will be independent. Four itemswere used for firm expertise and five for supplierexpertise; some of these items were borrowed oradapted from prior work (Walker and Weber, 1984;Noordewier, John, and Nevin, 1990) and otherswere original.

Controls

Firm control variables included the number ofemployees and firm age since these have oftenbeen related to greater internalization (Perry,1989). A binary variable for unionization was alsoincluded, as unionized firms may be more likely to



produce internally due to contractual commitmentsor may be less likely to do so due to higher costs. Abinary variable for firm type, powder metal or not,was included, as were dummy variables for four ofthe goods. Controls were also included for volumerequirements and scale economies as these canaffect production costs, typically toward greateroutsourcing (Pindyck and Rubenfeld, 1995). Ameasure for similarity was included to address thepotential for more custom goods being producedsimultaneously with more generic ones, which isat odds with this paper’s definition of concurrentsourcing but is consistent with transaction costlogic (Williamson, 1985).

Survey response

Nearly half of the firms replied (218), delivering193 usable surveys. This equates to a 43 per-cent usable response rate, significantly higher thanthe typical rate for firm-level studies of about20 percent (Paxson, Dillman, and Tarnai, 1995).The demographics for the respondent firms wererather unique as compared to those found in atypical organizational survey. Respondent firmswere small (95% employed fewer than 500 peo-ple), non-union (86%), and fairly old (averageage: 44 years). No indication of non-response biaswas found when comparing respondents to non-respondents by firm type and size (Armstrongand Overton, 1977). Since the sample is basedupon industry association listings that represent theoverall firm populations, sample selection bias isunlikely (Tomaskovic-Devey, Leiter, and Thomp-son, 1994).

A key informant single-respondent approachwas used for the survey. While in some cases it ispreferable to have multiple survey respondents, thesmall size of these firms and the technical and spe-cialized nature of the survey made it preferable torequest information from one very knowledgeablerespondent. The key informant approach is appro-priate when one can identify respondents who,by virtue of their positions in an organization’shierarchy, are able to provide opinions and per-ceptions that are valid reflections of those of otherkey decision-makers in the firm (Li and Atuahene-Gima, 2002; Phillips, 1981). Due to the fact thatthese firms are relatively small, that screeningcalls were made to determine the most appropriaterespondent, and that 95 percent of the respon-dents were professionals, with 53 percent being

executives, it is likely that these key informantswere the most appropriate respondents.

Consistency artifacts and a common methodvariance bias can result from collecting depen-dent and independent variables from the samerespondent. While this is a significant drawbackof this type of survey design, efforts were madeto avoid consistency artifacts by placing moresubjective items (e.g., supplier expertise) beforeobjective ones (e.g., firm size) (Salancik and Pfef-fer, 1977). Common method variance was inves-tigated by conducting the Harman one-factor test,which involves entering all the independent anddependent variable items into a factor analysis(Podsakoff and Organ 1986). A principal com-ponent factor analysis of all measurement itemsyielded seven factors with eigenvalue exceedingone. These factors accounted for 57 percent ofthe variance. The factor with the greatest eigen-value accounted for 15 percent of the variance.Because no single factor emerged as a dominantfactor accounting for most of the variance, com-mon method variance is unlikely to be a seriousproblem in the data.

METHODOLOGY

In order to statistically relate the survey items tothe sourcing mode decisions, two types of anal-ysis were conducted. First, a measurement modelwas created to assess validity and determine itemweights so a composite variable score could becomputed. Then, these independent variable scoreswere related to the sourcing decision in two dif-ferent ways. An ordered logit model was used toinvestigate the continuous, transaction cost-basedhypotheses. This model is preferred over OLSregression, since the data are ordinal rather thaninterval in nature; this method better reflects thedependent variable data as they were collected aseight discrete choices (Agresti, 2002). The orderedlogit model is also more appropriate for the datathan a Tobit model, since the data do not necessar-ily reflect an underlying but censored construct, arenot strictly continuous, and require some incorpo-ration of firm effects (Kennedy, 1998). The analy-ses used to test the other hypotheses, which werediscrete choice in nature, are multinomial logitmodels that relate the independent variables to thethree distinct sourcing options: make, buy, andconcurrently source. By comparing the exploratory


298 A. Parmigiani

power of the ordered and multinomial logit mod-els, we can infer whether concurrent sourcing canbetter be depicted as a point along a make/buycontinuum or as a separate and distinct alternative.

Since multiple items were used for the variables,exploratory factor analysis (EFA) and then con-firmatory factor analysis (CFA) were employed toinvestigate the relationships between the items andthe variables (Anderson and Gerbing, 1988). TheEFA results indicated that the each of the variableswere unidimensional and distinct. CFA submod-els for each individual variable were created andthen aggregated into a final, full model. All of theCFA models were estimated using full informationmaximum likelihood and were evaluated using sixdifferent indices (Hu and Bentler, 1998). The finalmodel sufficiently fit the data (χ 2 = 1521.642, 421degrees of freedom, p < 0.001, χ 2/d.o.f. = 3.614,TLI = 0.972, CFI = 0.976, RMSEA = 0.057). Allparameter estimates were significant (p < 0.02),supporting convergent validity, and none of thecovariances was significantly close to 1, supportingdivergent validity (Bollen, 1989; Bagozzi, 1994).Reliability estimates for the hypothesized vari-ables, other than supplier expertise, were all over0.60, suggesting adequate consistency among theitems (Nunnally, 1967). The loadings from thisfinal CFA model were used as item weightingfactors to construct aggregate scores for each vari-able (Pedazhur and Schmelkin, 1991); these scoreswere used in the subsequent analysis.

Since the dataset of 805 sourcing choices orig-inates from 193 firms, each of which provideddata on one to five goods, it is possible thatthe observations will not be independent, a clus-tering phenomenon common in survey research(Hosmer and Lemeshow, 2000). Due to the rel-atively small number of inputs vs. the larger num-ber of firms, a fixed-effects model could not beused because of the problem of perfect prediction(Greene, 1997). However, incorporating robuststandard errors adjusted for repeat observations byfirm does address this problem, and this techniquehas been used by other scholars for similarly struc-tured data (e.g., Mizruchi and Stearns, 2001).

RESULTS AND DISCUSSION

Descriptive statistics provide an initial indicationof the nature of sourcing decisions in this setting.

Table 2. Sourcing modes: overall, by firm, and by inputtype

Make Buy ConcurrentSource

Total

Overall 292 287 226 80536.3% 35.7% 28.0%

By firm typeStamping 227 226 169 622

36.5% 36.3% 27.2%Powder metal 65 61 57 183

35.5% 33.3% 31.1%

By input typeDie design 71 38 62 171

41.5% 22.2% 36.3%Die build 43 69 61 173

24.9% 39.9% 35.3%Die maintenance 110 12 40 162

67.9% 7.4% 24.7%Part machining 62 19 50 131

47.3% 14.5% 38.2%Part coating 6 149 13 168

3.6% 88.7% 7.7%

Table 2 presents the distribution of each sourc-ing mode in the entire dataset, by firm type, andby good sourced. All three modes are presentin the data, with little difference between firmtypes, but considerable difference among goods.Analysis indicated a fairly even distribution ofobservations in the middle categories, suggestingthat the data will be amenable to both orderedand unordered analysis. Table 3 provides corre-lations and descriptive statistics. Firm expertiseand scope economies were significantly positivelycorrelated with the percentage produced internally(0.704 and 0.514, respectively) whereas supplierexpertise and supplier scope economies were sig-nificantly negatively correlated with this dependentvariable (-0.494 and −0.387). This supports boththe neoclassical and capabilities views that firmswill be more likely to produce goods which theycan produce efficiently and effectively. Support-ing capabilities logic, firm and supplier expertisewere significantly negatively correlated (−0.536),suggesting that firms find suppliers that have skillsunlike their own.

One potential issue with the sourcing mode ofconcurrent sourcing is its stability. Critics mayassume that this mode of sourcing is transitory,used for a time while moving between mak-ing and buying (Nickerson and Zenger, 2002).To test for this, an item measuring longevity



Tabl

e3.

Des

crip

tive

stat

istic

san

dco

rrel

atio

nsa

Mea

nS.

D.

12

34

56

78

910

1112

1314

15

1C

oncu

rren

t(1

/0)

0.28

10.

450

12

Perc

ent

inte

rnal

0.51

90.

438

0.05

71

3A

sset

spec

ifici

ty4.

070

1.26

60.

028

0.04

21

4V

olum

eun

cert

aint

y3.

814

1.44

80.

026

0.01

20.

013

15

Tech

unce

rtai

nty

4.46

41.

232

0.04

9−0

.055

0.04

00.

088∗

16

Perf

orm

unce

rtai

nty

2.94

90.

928

−0.0

510.

024

0.06

60.

297‡

−0.0

171

7Sc

ale

econ

omie

s3.

475

1.66

4−0

.011

−0.0

22−0

.014

0.01

1−0

.039

0.02

01

8V

olum

ere

quir

ed2.

643

1.47

4−0

.090

∗−0

.034

−0.0

00−0

.192

‡−0

.044

−0.0

53−0

.029

19

Firm

scop

eec

on.

4.21

41.

571

0.08

2∗0.

514‡

0.03

20.

042

−0.0

390.

013

0.00

90.

050

110

Supp

lier

scop

eec

on.

4.70

31.

361

−0.0

52−0

.387

‡−0

.083

∗−0

.001

−0.0

080.

066

0.05

40.

001

−0.2

00‡

111

Firm

expe

rtis

e4.

722

1.74

20.

268‡

0.70

4‡−0

.026

0.00

1−0

.016

−0.0

12−0

.032

−0.0

370.

544‡

−0.3

25‡

112

Supp

lier

expe

rtis

e3.

723

1.06

0−0

.115

†−0

.494

‡0.

010

−0.0

320.

056

0.06

30.

039

−0.0

10−0

.401

‡0.

396‡

−0.5

36‡

113

Firm

size

2.50

11.

100

0.02

60.

045

0.06

1−0

.093

∗0.

052

−0.1

27‡

−0.0

75∗

0.28

0‡0.

042

−0.0

410.

055

−0.1

05†

114

Uni

on(1

/0)

0.14

00.

347

0.04

2−0

.024

−0.1

34‡

0.06

40.

037

0.05

8−0

.005

0.01

60.

022

0.05

70.

014

0.04

80.

024

115

Inpu

tsi

mila

rity

3.50

91.

804

0.02

5−0

.009

−0.1

77‡

−0.1

80‡

−0.1

80‡

0.07

0∗0.

026

0.06

5−0

.041

−0.0

230.

009

−0.0

34−0

.063

−0.0

511

an

=80

5∗p

<0.

05;

†p

<0.

01;

‡p

<0.

001.


300 A. Parmigiani

Table 4. Ordered logit models, with percentage produced internally as the dependent variable; Models 1 and 2 useeight categories; Models 3 and 4 use three categoriesa

Model 1 Model 2 Model 3 Model 4

Asset specificity 0.053 0.057 0.058 0.058(0.071) (0.071) (0.075) (0.075)

Volume uncertainty −0.046 −0.047 −0.021 −0.022(0.056) (0.056) (0.065) (0.064)

Technological uncertainty −0.059 −0.062 −0.065 −0.065(0.069) (0.070) (0.077) (0.077)

Performance uncertainty 0.198∗∗ 0.205∗∗ 0.178∗∗ 0.184∗∗

(0.097) (0.098) (0.106) (0.106)Firm scope economies 0.326∗∗∗ 0.009 0.327∗∗∗ 0.090

(0.065) (0.168) (0.069) (0.190)Supplier scope economies −0.263∗∗∗ −0.573∗∗∗ −0.370∗∗∗ −0.601∗∗∗

(0.062) (0.165) (0.076) (0.180)Firm × Supplier scope economies 0.070∗∗ 0.052∗

(0.035) (0.038)Firm expertise 0.882∗∗∗ 0.880∗∗∗ 0.848∗∗∗ 0.931∗∗∗

(0.088) (0.313) (0.087) (0.309)Supplier expertise −0.132 −0.116 −0.179∗∗ −0.046

(0.125) (0.454) (0.123) (0.451)Firm expertise × Supplier expertise −0.001 −0.023

(0.081) (0.080)Firm age 0.007∗∗ 0.008∗∗ 0.007∗ 0.007∗

(0.004) (0.004) (0.005) (0.005)Firm size (employees) 0.042 0.040 0.084 0.085

(0.074) (0.073) (0.083) (0.083)Union −0.129 −0.138 −0.031 −0.035

(0.229) (0.229) (0.248) (0.247)Powder metal 0.074 0.071 0.102 0.104

(0.212) (0.211) (0.232) (0.232)Scale economies −0.039 −0.034 −0.030 −0.027

(0.061) (0.061) (0.066) (0.066)Volume required −0.067∗ −0.061 −0.069 −0.066

(0.051) (0.053) (0.057) (0.059)Input similarity 0.005 0.009 0.041 0.044

(0.048) (0.048) (0.052) (0.053)Likelihood ratio 755.75(19) 759.62(21) 690.30(19) 691.90(21)Pseudo R2 0.266 0.267 0.393 0.394Adjusted Count R2 0.375 0.374 0.513 0.517χ 2 (−2 log likelihood) 0.000 (Ctrl) 0.021 (Mod 1) 0.000 (Ctrl) 0.201 (Mod 3)a n = 805 for both models; parameter estimates for input controls and constants are omitted. Robust standard errors are in parenthesesbelow the parameter estimates. All tests are one-tailed since the hypotheses are directional, with ∗ p < 0.10; ∗∗ p < 0.05; and∗∗∗ p < 0.01.

was included and was compared across modes.For concurrently sourced inputs, 60 percent hadalways been sourced in this way, as compared to54 percent for inputs made, a difference that isnot statistically significant. Firms were also askedwhether they plan to change modes; for concur-rently sourced inputs, 11.9 percent plan to change,as compared to 10.3 percent for inputs made. Thedifference was not significant, supporting concur-rent sourcing as a stable, equilibrium sourcingmode.

Model 1, presented in Table 4, depicts anordered logit model that included all eight choicesfor the dependent variable.5,6 This model has rea-sonably good explanatory power, as shown by

5 As a robustness check, Tobit models were run to replicate theordered logit Models 1 and 2. The substantive results relating tothe hypothesized variables were identical.6 In some cases, scholars include interaction terms between assetspecificity and uncertainty variables (e.g., Coles and Hesterly,1998, who studied hospital procurement decisions for 15 differ-ent services). This technique is appropriate when the goods being



its pseudo R2 value of 0.266.7 Note that positiveparameter estimates for the explanatory variablessuggest that greater amounts of these variables leadto a higher percentage of the good produced inter-nally. Interestingly, it shows no significance forthe asset specificity variable, contrary to Hypoth-esis 1tce. Volume uncertainty had no effect on thepercentage produced internally, indicating no sup-port for Hypothesis 2tce. Goods with greater uncer-tainty in performance were likely to be producedinternally at higher rates, supporting Hypothesis3tce. Greater firm scope economies resulted inhigher percentages of internalization, while greatersupplier scope economies resulted in lower per-centages, following the logic of Hypothesis 4neo.Greater firm expertise resulted in higher percent-ages of internalization, following the logic ofHypothesis 5cap. The parameter estimate for tech-nological uncertainty was not significant, whichsupports Hypothesis 6cap. Since we expect greatertechnological uncertainty to result in concurrentsourcing, the coefficient for this variable will notbe directly related to the percentage producedinternally, and therefore an insignificant estimateis expected. To more directly test Hypotheses 4neoand 5cap, Model 2 incorporates interaction termsbetween the firm and supplier scope economiesand between firm and supplier expertise. Thereis a positive relationship between greater com-bined firm and supplier scope economies withgreater internalization, potentially conflicting withHypothesis 4neo. No significant results were foundfor the interaction of firm and supplier expertise,which is consistent with Hypothesis 5cap.

The modest explanatory power plus the lackof significance for the asset specificity variablesuggest that this model could be improved. Perhapsthe distinctions between the eight categories aretoo fine and fewer categories will lead to a bettermodel. The next two models are ordered logitmodels with three categories representing make,

studied have widely varying levels of asset specificity (e.g., land-scaping and respiratory services) and binary variables are used.In this study, all the goods have some non-trivial degree of assetspecificity above the threshold where it affects uncertainty andthis variable was measured continuously. Interactions were notincluded in the models since the effects on uncertainty variablescould be interpreted directly.7 Model 1 was run with only the control variables, resultingin a pseudo R2 of 0.14, an adjusted count R2 of 0.29, and alikelihood ratio of 392.87 (11). Likelihood ratio significance testsindicated that the model incorporating the explanatory variableswas considerably better.

concurrent source, and buy; concurrent sourcingis always the middle option. Model 3 has betterexplanatory power than Model 1 or 2, with apseudo R2 value of 0.393.8 The substantive resultsof Model 3 were identical to Model 1, with theexception of supplier expertise being negativelyand significantly associated with the percentageproduced internally, which follows the logic ofHypothesis 5cap. Hypotheses 1tce and 2tce were notsupported, while Hypothesis 3tce was supported,and the logic underlying Hypotheses 4neo, 5cap,and 6cap was supported. Model 4, which includesthe interaction terms for scope economies andexpertise, produced the same substantive resultsas Model 2, but does have greater explanatorypower.

Based upon the fit statistics, the models withthree categories better represent the data. How-ever, many of the hypotheses do not require therigid specification of ordering concurrent sourc-ing between making and buying. The hypothesesthat relate explanatory variables toward a greaterlikelihood of concurrent sourcing are better testedusing a multinomial logit model which comparesconcurrent sourcing with making and with buy-ing. Table 5 displays Model 5, a multinomial logitmodel that was significant with very good explana-tory power, predicting 71 percent of the sourc-ing mode choices correctly.9 The Wald combinetest indicated that sourcing mode of concurrentsourcing was statistically distinct from that ofmake and of buy. Both the Hausman and theSmall–Hsiao tests confirmed the independence ofirrelevant alternatives (IIA) assumption, furthersupporting the distinction between the three sourc-ing modes. Most of the key independent variableshad explanatory power; Wald tests indicated thatasset specificity (p = 0.017), performance uncer-tainty (p = 0.094), firm scope economies (p =0.000), supplier scope economies (p = 0.000), and

8 Model 3 was run with only the control variables, resultingin a pseudo R2 of 0.21, an adjusted count R2 of 0.33, and alikelihood ratio of 361.80 (11). Likelihood ratio significance testsindicated that the model incorporating the explanatory variableswas considerably better.9 Model 5 was run with only the control variables, resultingin a pseudo R2 of 0.22, an adjusted count R2 of 0.36, and alikelihood ratio of 379.03 (22). Likelihood ratio significance testsindicated that the model incorporating the explanatory variableswas considerably better.


302 A. Parmigiani

Table 5. Multinomial logit modelsa

Model 5 Model 6

Makevs. CS

Buy vs.CS

Make vs.Buy

Make vs.CS

Buy vs.CS

Make vs.Buy

Asset specificity −0.089 −0.344∗∗∗ 0.255∗∗ −0.091 −0.349∗∗∗ 0.258∗∗(0.094) (0.120) (0.124) (0.94) (0.120) (0.122)

Volume uncertainty −0.070 −0.115 0.045 −0.069 −0.109 0.041(0.083) (0.105) (0.109) (0.083) (0.104) (0.109)

Technological uncertainty −0.113∗ −0.060 −0.053 −0.118∗ −0.069 −0.049(0.082) (0.131) (0.146) (0.083) (0.130) (0.145)

Performance uncertainty 0.265∗∗ 0.017 0.249∗ 0.267∗∗ −0.002 0.269∗(0.123) (0.159) (0.182) (0.124) (0.161) (0.183)

Firm scope economies 0.266∗∗∗ −0.253∗∗ 0.520∗∗∗ 0.259∗ 0.205 0.055(0.081) (0.111) (0.119) (0.201) (0.300) (0.351)

Supplier scope economies −0.265∗∗∗ 0.431∗∗∗ −0.695∗∗∗ −0.291∗ 0.791∗∗∗ −1.083∗∗∗(0.088) (0.119) (0.131) (0.219) (0.264) (0.322)

Firm × Supplier scope 0.003 −0.093∗ 0.096∗(0.043) (0.058) (0.068)

Firm expertise 0.278∗∗∗ −1.095∗∗∗ 1.373∗∗∗ 0.772∗∗ −1.575∗∗∗ 2.346∗∗∗(0.110) (0.128) (0.148) (0.370) (0.527) (0.586)

Supplier expertise −0.119 0.053 −0.171 0.682 −0.471 1.153∗∗(0.142) (0.171) (0.206) (0.584) (0.615) (0.695)

Firm expertise × Supplier expertise −0.139∗ 0.119 −0.259∗∗(0.099) (0.129) (0.137)

Firm age 0.009∗∗ 0.000 0.009 0.009∗∗ 0.000 0.009(0.005) (0.007) (0.008) (0.005) (0.007) (0.008)

Firm size (employees) −0.010 −0.218∗ 0.207∗ −0.005 −0.229∗∗ 0.223∗(0.094) (0.133) (0.147) (0.095) (0.133) (0.147)

Union −0.357∗ −0.567∗ 0.210 −0.379∗ −0.538∗ 0.159(0.260) (0.390) (0.448) (0.261) (0.393) (0.439)

Powder metal −0.079 −0.356 0.277 −0.067 −0.334 0.267(0.247) (0.367) (0.409) (0.251) (0.373) (0.417)

Scale economies −0.023 0.034 −0.057 −0.023 0.034 −0.057(0.066) (0.088) (0.113) (0.066) (0.089) (0.114)

Volume required 0.064 0.248∗∗∗ −0.184∗∗ 0.056 0.257∗∗∗ −0.201∗∗(0.064) (0.099) (0.105) (0.065) (0.100) (0.105)

Input similarity −0.059 −0.193∗∗ 0.134∗ −0.056 −0.193∗∗ 0.137∗(0.059) (0.086) (0.092) (0.058) (0.087) (0.093)

Likelihood ratio 733.68(38) 738.70(42)Pseudo R2 0.417 0.420Adjusted Count R2 0.550 0.556χ 2 (−2 log likelihood) 0.000 (vs. controls) 0.040 (vs. Model 5)a CS, concurrent source; n = 805 for both models. A positive coefficient indicates that the first choice is more likely than the second.Parameter estimates for input controls and constants are omitted. Robust standard errors are in parentheses below the parameterestimates. All tests are one-tailed since the hypotheses are directional, with ∗ p < 0.10; ∗∗ p < 0.05; and ∗∗∗ p < 0.01.

firm expertise (p = 0.000) coefficients were sig-nificant for the overall model.10

10 The Wald tests indicate that variables associated with bothproduction and transaction costs were significant in determiningthe sourcing mode. As an exploratory exercise, Model 5 wasrun with controls and only production cost variables (firm andsupplier scope, firm and supplier expertise) and again with con-trols and only transaction cost variables (asset specificity, volumeuncertainty, performance uncertainty, and technological uncer-tainty). Likelihood ratio tests indicated that these exploratory

Model 5 computes the likelihood of each of thethree modes vs. the others: it compares concur-rent sourcing vs. making, concurrent sourcing vs.buying, and making vs. buying. Hypothesis 1tceis supported since firms would be more likely toconcurrently source rather than buy goods withgreater asset specificity; they also would be more

models were inferior to Model 5. This supports the importanceof both production and transaction costs in the sourcing decision.



likely to make vs. buy, which supports the usualTCE prediction. Neither Hypothesis 2tce nor 2neois supported, since volume uncertainty does notappear to affect the sourcing choice. Hypothesis3neo is not supported, as greater performance uncer-tainty will motivate firms to choose making overconcurrent sourcing and making over buying; thisfinding does support Hypothesis 3tce. The logic ofHypothesis 4neo is supported, since greater firmscope economies will lead to making over con-current sourcing, concurrent sourcing over buying,and making over buying, with similar results foundfor supplier scope economies. Likewise, the logicof Hypothesis 5cap is supported, since greater firmexpertise was associated with making over con-current sourcing, concurrent sourcing over buy-ing, and making over buying; supplier expertisewas not significant but was signed appropriately.Hypothesis 6cap is moderately supported since con-current sourcing is more likely than solely making,although not necessarily favored over solely buy-ing.

Model 6, also shown in Table 5, was created totest Hypotheses 4neo and 5cap directly by includ-ing interaction terms for the scope economy andexpertise variables. Greater scope economies forboth the firm and the supplier led to concurrentsourcing over buying, supporting Hypothesis 4neo.Likewise, greater expertise for both the firm andthe supplier led to concurrent sourcing over mak-ing, supporting Hypothesis 5cap. Interaction termsare difficult to interpret, especially in nonlinearmodels. To verify that the expertise and scopeeconomy interactions were indeed significant, themodel was converted into a dichotomous model(concurrent source or not) and the algorithm ofNorton, Wang, and Ai was used. This program

computes the correct marginal effect of a change intwo interacted variables by calculating the cross-derivative. It provides both tabular and graphicaloutput, which can be interpreted to determine thetrue interaction effect since this is conditional uponthe value of the variables (Norton, Wang, and Ai,2004). These findings supported the significanceof both of these interaction variables toward con-current sourcing; they also supported Hypothesis6cap. While the multinomial logit model suggestedpartial support for this technological uncertaintyhypothesis (concurrent sourcing over making butnot concurrent sourcing over buying), the dichoto-mous model indicated that concurrent sourcing wasmore likely for goods high in technological uncer-tainty.

An overriding question that these empirical anal-yses can address is whether or not sourcing choiceslie on a continuum. If a continuum exists, then theordered logit model should fit better than the multi-nomial logit model, which does not restrict con-current sourcing always to be the middle choice.Comparing Model 3 and Model 5 through good-ness of fit and nested model tests indicated a viola-tion of the parallel regression assumption and thatthe ordering constraint was too severe; thus themultinomial logit was the better model (Long andFreese, 2001). The explanatory power of Model 5was also superior. A graphical depiction of Model5, shown in Figure 1, also assists in answering thisquestion as it presents the marginal effect of thevariables on each of the three options: make, buy,and concurrently source. If the marginal effects aresmall, then the options will be very close together,suggesting that the variable does not affect thechoice between them. The greater the effect, thefarther the option will be from zero, as indicated

-0.25 -0.20 -0.15 -0.10 -0.05 0 0.05 0.10 0.15 0.20 0.25

fexB MM

MM

M

CB

MCB

B MM B

B M

MBM C

B C

C

MB CC

CC

C

BCMB MC

B CB C

M B

B C

mthinvolun

techunperfunscale

volumefscope

suscope

supexage

empeesunion

pmallsame

Figure 1. Marginal probabilities for making (M), buying (B), and concurrent sourcing (C) per standard deviationchange in each independent variable (based on Model 5)


304 A. Parmigiani

by the y-axis. The figure indicates that concurrentsourcing is sometimes the middle choice, such aswith firm expertise, while other times it is not,such as with technological uncertainty. It appearsthat some variables may indeed act continuously,with concurrent sourcing associated with an inter-mediate level of the attribute, while others do not.

Concurrent sourcing appears to be a distinctchoice over making or buying, rather than a linearcombination of the two sourcing modes. Perhapsconcurrent sourcing is chosen first, with the sec-ondary decision being the extent of internal pro-duction. As an exploratory exercise, models werecreated to relate the percentage of internal pro-duction for only the concurrently sourced goodswith the variables used in the other models. Thesemodels are not displayed owing to relatively poorexplanatory power (pseudo R2 = 0.093), but someinteresting results did occur. Asset specificity didnot affect the percentage made. While performanceuncertainty was positively and directly relatedto internalization, the other uncertainty variableswere not significant. Firm scope economies wererelated to the percentage produced internally, whilethe scope interaction was not. Firm and supplierexpertise affected this percentage, but the inter-action did not. These findings suggest that thepercentage produced internally could be affectedby different variables from those that influence theinitial decision to source internally, externally, orconcurrently.

Several control variables also deserve mention-ing. Unionized firms were more likely to con-currently source; it may be difficult for them tocompletely outsource due to contractual commit-ments or they desire outside suppliers as a bench-mark. Higher-volume goods were less likely to beconcurrently sourced and more likely to be out-sourced, following economies of scale logic. Morehomogeneous goods were also more likely to beconcurrently sourced, perhaps because this facili-tates the comparison between internal and externalofferings. This contradicts the TCE assumption ofheterogeneity among concurrently sourced goods(i.e., more customized goods produced internallyand more generic ones outsourced).

In summary, all three theories assisted in ex-plaining the sourcing choice. TCE logic was sup-ported as firms were less likely to buy whenmarkets were thin and more likely to make ifperformance uncertainty was great. The neoclas-sical economics prediction of greater firm and

supplier scope economies leading to concurrentsourcing was confirmed. The capabilities view wassupported as greater combined firm and supplierexpertise led to concurrent sourcing, as did greatertechnological uncertainty that may motivate firmsto search for knowledge both internally and exter-nally. Moreover, support was shown for a discretechoice between making, buying, and concurrentsourcing over a continuum view. This suggests thatfirms that concurrently source may only need tobuy or make a minor percentage of their require-ments and still get the governance benefits of bothmarket and hierarchy.

CONCLUSION AND IMPLICATIONS

This study assists in clarifying the confusion sur-rounding firm boundaries and hybrid governancemodes through investigating why firms would con-currently source, simultaneously making and buy-ing. By incorporating transaction cost, neoclassi-cal economics, and capabilities theories, a holisticview of why firms would use this sourcing mode ispresented, revealing aspects of each theory moti-vating the sourcing choice. Finding that all threetheories contribute to the choice of sourcing modeillustrates the firm’s desire to simultaneously moni-

WHY DO FIRMS BOTH MAKE AND BUY? AN ... 2014...the make or buy decision. However, none of these...

Documents

Transcript of WHY DO FIRMS BOTH MAKE AND BUY? AN ... 2014...the make or buy decision. However, none of these...