Buyer participation in outsourced new product development projects: The role of relationship multiplexity

In business markets, firms increasingly participate in new product development (NPD) activities that they outsource to suppliers. Such buyer participation can be complicated by relationship multiplexity—that is, the buyer may simultaneously be a competitor, supplier, and/or partner of the supplier firm. Drawing on role theory, we theorize how relationship multiplexity moderates the effect of buyer participation on project success. We analyze 140 NPD projects that were executed by a contract R&D organization in the aerospace industry and find that buyer participation lowers the buyer's and the supplier's perceptions of project success for buyer‐as‐supplier multiplexity. Buyer participation increases the perceptions of project success in the case of buyer‐as‐partner multiplexity, but only when these partnerships are emergent, as opposed to engineered. Buyer participation decreases the buyer's perception of project success for relationships featuring buyer‐as‐competitor multiplexity but does not affect the supplier's perception in the case of buyer‐as‐competitor multiplexity. Our results provide tactical insights as to when buyer participation helps or hurts.


| INTRODUCTION
In various business markets, such as aerospace, automotive, chemicals, and software, firms increasingly outsource new product development (NPD) activities to external suppliers (Handley & Benton, 2013). These NPD projects often involve "one-off" customized products, and buyer firms participate in varying degrees during the NPD process (Sanderson & Cox, 2008). For example, Airbus outsourced the development of engines for its A350-1000 model to Rolls-Royce, while participating in Rolls-Royce's fundamental research program to develop new materials for the engine (EPSRC, 2014). Similarly, Volvo outsourced the development of its electric motor to Siemens, but rather than being a passive buyer, Volvo participated in the electric motor's inverter design (Reed & Simon, 2011).
Buyer participation is defined as the level of involvement of industrial customers in manufacturers' NPD processes (Fang, 2008). A complicating feature of buyer participation is that a buyer may take on more than one role in relation to the supplier, making the buyersupplier relationship "multiplex." Relationship multiplexity occurs when two firms are interconnected by multiple kinds of ties that represent different roles (Shipilov & Li, 2012). This is pointedly illustrated by Hamel and Prahalad (1994:40): "on any given day […] AT&T might find Motorola to be a supplier, a buyer, a competitor, and a partner." The goal of this research is to investigate to what extent "relationship multiplexity" alters the effectiveness of buyer participation in outsourced NPD. In the spirit of Hamel and Prahalad (1994), we distinguish between three forms of relationship multiplexity. First, buyer-as-competitor multiplexity occurs when a buyer that participates in the supplier's NPD process also competes with the supplier for market share in other product markets. Second, buyer-as-supplier multiplexity arises when the participating buyer and the supplier also face each other in a reversed role for activities other than the outsourced NPD project. Third, buyer-as-partner multiplexity refers to the situation when a buyer that participates in the supplier's NPD process simultaneously shares partner ties with the supplier outside of the outsourced NPD project. We distinguish between two subtypes of partner ties: engineered partner ties and emergent partner ties (Doz, Olk, & Ring, 2000).
We contribute to the literature in three ways. First, the extant literature on buyer participation suggests that buyer participation can both positively and negatively affect project success. On the bright side, it may increase project success by increasing the supplier's flexibility (Heide & John, 1990), by providing the supplier with access to buyer resources and additional development paths (Campbell & Cooper, 1999), and via better alignment with buyer needs (Fang, 2008). On the dark side, it may decrease project success by complicating coordination (Von Hippel, 1990) and creating inequity concerns (Nyaga, Whipple, & Lynch, 2010). 1 We contribute to the buyer-participation literature by adjudicating between these positions and examining when buyer participation helps or hurts.
Second, we examine how relationship multiplexity moderates the effect of buyer participation on project success. A large number of operations management and supply chain scholars study interdependencies and interactions between ties in the context of supply chain triads and networks (e.g., Choi & Wu, 2009;Dubois & Fredriksson, 2008;Pathak, Wu, & Johnston, 2014;Wilhelm, 2011). We take a different perspective and focus on interdependencies and interactions between ties within the dyadic buyer-supplier relationship. As Shipilov, Gulati, Kilduff, Li, and Tsai (2014:457) observe, multiplexity is "prevalent in the world around us, but very few studies have examined the linkages between economic actors' identities and the heterogeneity of their relationships." Thus, we address calls for more research on multiplexity (e.g., Shipilov, 2012;Tuli, Bharadwaj, & Kohli, 2010).
Third, we distinguish between the buyer's and the supplier's perceptions of project success. We thereby address a blind spot of interorganizational research, which has traditionally ignored that different stakeholders may perceive differences in outcome variables (Lumineau & Oliveira, 2018;Roh, Whipple, & Boyer, 2013). Recently, supply chain and operations management researchers started to explore asymmetries in perceptions (e.g., McEvily, Zaheer, & Kamal, 2017;Villena & Craighead, 2017). Contributing to this emerging research, we demonstrate that effects of buyer participation and relationship multiplexity on project success are indeed not always symmetric across the dyad.
Recognizing that "combining qualitative fieldwork with archival or survey data collection is vitally important for any scholar doing research within the strategic multiplexity perspective," (Shipilov, 2012:220), we first held three roundtable discussions with senior executives and conducted 18 in-depth interviews with project managers to ground our approach in managerial practice. To test our hypotheses, we composed a unique, proprietary data set on 140 outsourced NPD projects in the aerospace industry. Our data set combines proprietary archival data-project administration records, project evaluation reports (including buyer and supplier evaluations at the time of project closure), strategic cooperation plans, and procurement records-with publicly available data. In addition, we gathered survey data specifically for this study.
Our findings offer tactical recommendations to (buyer and supplier) managers who decide on buyer participation in outsourced NPD when buyer-supplier relationships are multiplex. We find that buyer participation in outsourced NPD may help or hurt, depending on the multiplexity of the buyer-supplier relationship and whether the buyer or supplier perspective is taken. Even though the supplier's perception of project success is unaffected by involving a competing buyer, the effect of buyer participation on the competing buyer's perception of project success is negative. Also, we find that involving a buyer that also has a reverse role as a supplier negatively affects both the buyer's and the supplier's perceptions of project success. On balance, it may be best to avoid buyer participation under buyer-ascompetitor or buyer-as-supplier multiplexity. Lastly, we add a caveat to the industry sentiment that involving one's partners in NPD is beneficial: we find that a positive impact only holds for emergent but not for engineered partnerships. In sum, we identify situations where buyer participation in outsourced NPD is harmful rather than helpful, in contrast to business press that articulates only the positives of buyer participation (e.g., Buchanan, 2008;Panasonic, 2014).

| Pilot study
To ground our approach in managerial practice, a qualitative phase preceded the development of the research hypotheses. The objective of this phase was to obtain more insight into suppliers' perceptions of buyer participation in multiplex relationships, as suppliers are the key decision-makers regarding the degree of buyer participation in their NPD processes (Smets, Langerak, & Rijsdijk, 2013). We held three roundtable discussions with senior supplier executives in general management, marketing, and R&D, and we conducted 18 in-depth interviews with project managers who led NPD projects for their respective supplier firms. To capture a broad set of supplier perspectives, we selected interviewees from three firms in different industries (aerospace, information technology system development, and photolithography), with annual revenues ranging from $85 million to $8.5 billion. On average, the roundtables and in-depth interviews lasted 90 and 50 min, respectively.
The roundtables underscored that buyer participation in outsourced NPD is a pervasive phenomenon and that buyers are pushing suppliers for participation in their outsourced NPD projects. Further, we learned that the suppliers' executives were concerned about letting buyers participate if these buyers also fulfilled other roles. A recurring roundtable discussion theme was that suppliers struggle with the challenges caused by multiplexity when involving buyers in NPD.
The interviewees' opinions diverged concerning consequences of participation by a competing buyer. Some interviewees argued against involving a buyer with which they compete in another product market. An information technology system development firm's general manager mentioned that "it seems wise to keep competitors at a distance" and "we tend not to involve competing buyers in their outsourced NPD projects; why enlighten them more than is absolutely necessary?" This belief contrasts with that of a photolithography firm's R&D manager, who stated: "we don't mind involving competitors, as they will get to learn the technologies and knowledge we develop sooner or later anyway. However, competitors need to bring something to the table as well; it's not a one-way street." Concerning participation by buyers that also play a supplier role for activities other than the outsourced NPD project, an aerospace development firm's marketing director explained that "it signals our intent to strengthen and deepen our relationship with this key stakeholder," while also pointing to a risk: "A bad experience with the buyer in its supplier role may negatively affect the way we deal with it in its participating-buyer role […], a situation we should try to avoid because termination of the buyer relationship would imply a lose-lose situation." Several interviewees indicated they were inclined to involve buyers with which they closely collaborate as partners outside the focal NPD project. The aerospace development firm's marketing director argued that they let such buyers participate in NPD as "they are frequently on the premises anyway," but added he was unsure whether this was the right strategy.
In sum, the roundtable discussions and in-depth interviews underscore that (a) relationship multiplexity is a pervasive phenomenon in the context of buyer participation in outsourced NPD, and (b) suppliers are either unsure or do not converge on whether they should let buyers participate if buyers also fulfill other roles. The aerospace development firm's marketing director succinctly summarized this knowledge gap: "I am not sure whether we should allow buyer X to participate in the NPD process, as we also compete against each other.
[…] Should we let buyer Y participate in the development process, as it is also our partner in a consortium? I don't know.
[…] I believe the other roles played by our buyers in relation to our firm may matter and should be taken into account when we decide on buyer participation." Appendix A1 provides an overview of the main takeaways from the roundtables and interviews. Next, we review the literature on role theory, which forms the basis of our research hypotheses.

| Role theory
The theoretical root of our study resides in role theory, which is the study of characteristic behavioral patterns ("roles") of actors within contexts (Biddle, 1986;Katz & Kahn, 1966). Roles are evoked by the situation in which actors find themselves. Based on these roles, actors form expectations about the behaviors that are appropriate in a specific situation (Grayson, 2007). Table 1 summarizes role research relevant to buyer-supplier relationships. Three issues are noteworthy.
First, extant research has almost exclusively applied role theory at the individual level (e.g., the salesperson level). Two notable exceptions are Heide and Wathne (2006) and Dong, Ju, and Fang (2016), who elevate role theory to the organizational level. Heide and Wathne (2006:91) define an organizational role as "an organizational identity" or a "collective mind," which "provides the foundation for shared perceptions and coordinated decision making." In a similar vein, we adopt an organizational level of analysis to functional role theory (Biddle, 1986). In functional role theory, the identity of an actor hinges on its position within a social system, that is, the function of the actor. In our context, the use of terms such as "buyer," "competitor," "supplier," and "partner" are applications of functional role theory (Heide & Wathne, 2006). Second, research using role theory has analyzed the multiple roles an actor plays in relationships with different parties (e.g., a salesperson experiencing role conflict due to addressing a buyer's needs while also having his/her employer's best interest in mind). A largely unaddressed issue pertains to how the roles played in different ties between the same parties-that is, in a multiplex relationship-may affect one another. Building on functional role theory, we distinguish between three types of relationship multiplexity, each combining two roles: buyer-as-competitor multiplexity (when a participating buyer also shares a competitor tie with the supplier in other product markets, where they compete for market share), buyer-as-supplier multiplexity (when a participating buyer also shares a role-reversal tie with the supplier and supplies parts, materials, or other resources to the supplier for activities other than the outsourced NPD project), and buyer-as-partner multiplexity (when a participating buyer also shares a partner tie with the supplier and closely collaborates with the supplier outside of the outsourced NPD project).
Third, with one exception, all studies in Table 1 examined self-directed consequences, such as an individual's job satisfaction (e.g., Michaels, Day, & Joachimsthaler, 1987;Singh, 1998) or a firm's organizational or supply chain performance (e.g., Dong et al., 2016;Goolsby, 1992). Only Solomon et al. (1985), in a conceptual piece, considered the consequences on the other party in the relationship. We contribute by examining the impact of specific role combinations on perceptions of project success at both sides of the dyad. Drawing on the pilot study insights, and building on role theory, we argue that the nature of the multiplex relationship (buyer-as-competitor, buyer-as-supplier, or buyer-as-partner) moderates the impact of buyer participation in outsourced NPD on project success.

| The impact of buyer participation on project success
Buyer participation is the level of "customer involvement in the manufacturers' NPD process" (Fang, 2008:91). Buyer participation in outsourced NPD projects varies in terms of the number of activities performed-which can range from fundamental research to product design and prototype testing-and the intensity with which those activities are performed by a buyer-which can range from mere assistance in tasks executed by the supplier to more central responsibilities in the NPD process (Appleyard, 2002). 2 The role expectations of a participating buyer are to share resources with the supplier to perform these activities (Campbell & Cooper, 1999). In buyer-supplier relationships, which inherently contain elements of both cooperation and competition (Wilhelm, 2011), buyer participation thus tilts the balance toward cooperation.
It is not clear how increased buyer participation affects project success. In its participating role, a buyer is expected to share resources with the supplier. Depending on the resources shared, a participating buyer can serve as an information resource or as a codeveloper (Fang, 2008). A buyer can share market information (Chatterji & Fabrizio, 2014), including detailed insight into buyer preferences, end-user demand, and/or competitor activities. These insights help the supplier to better understand a buyer's needs at the NPD process outset (Mishra & Shah, 2009) and the evolution of these needs across the NPD stages as the buyer is confronted with new ideas, concepts, and prototypes (Fang, Palmatier, & Evans, 2008). The common language developed due to information sharing may further reduce errors in later phases of a project (Dyer, 1996). A participating buyer may also share technical knowledge, materials, equipment, and machinery, opening alternative development paths otherwise not accessible (Campbell & Cooper, 1999). Finally, the collaborative nature of buyer participation gives rise to normative expectations of flexibility and solidarity (Heide & John, 1992) that help curb opportunistic behavior (Das & Teng, 1998), further contributing to project success.
On the other hand, buyer participation increases coordination requirements in the buyer-supplier relationship (Von Hippel, 1990). The parties must determine which tasks to execute independently and which ones jointly. Coordination takes time and effort, creating inefficiencies in carrying the project to completion. Also, buyer participation may lead to feelings of inequity when one party believes it invests more than it should, lowering trust and openness in the buyer-supplier relationship, thereby decreasing project success (Nyaga et al., 2010).
In light of these countervailing forces, we do not offer a hypothesis for the (main) effect of buyer participation on project success. Instead, we take a contingency approach and examine when buyer participation may help or hurt. Using role theory, we argue that the effect of buyer participation on project success depends on the other ties in the multiplex buyer-supplier relationship. We visualize our contingency framework in Figure 1. 3.2 | The moderating effects of relationship multiplexity 3.2.1 | Buyer-as-competitor multiplexity Buyer-as-competitor multiplexity occurs when a participating buyer also competes with the supplier for market share in a product market outside the focal NPD project and, therefore, is a form of coopetition (Lado, Boyd, & Hanlon, 1997;Li, Liu, & Liu, 2011). Firms that compete in the same product market solve comparable sets of problems (Chen, 1996), leading to similar knowledge-processing systems (Lane & Lubatkin, 1998), which may increase their ability to internalize the knowledge that the counterparty shares (Cohen & Levinthal, 1990). Thus, buyer-as-competitor multiplexity may facilitate the application of a buyer's shared knowledge in the NPD project and improve project success.
However, this potential benefit may not be realized as it may be overshadowed by role conflict. Under buyer-ascompetitor multiplexity, a buyer plays two roles: a participating-buyer role and a competitor role. In its participating-buyer role, a buyer is expected to share knowledge with the supplier (Dyer & Singh, 1998). In its competitor role, a participating buyer may be concerned with knowledge appropriation (Lavie, 2006): it may fear that some of the knowledge it shares could be opportunistically appropriated by the supplier to improve its position in the competitive tie (Hamel, Doz, & Prahalad, 1989;Lado, Boyd, & Hanlon, 1997). The role expectations (to share knowledge) linked to a participating-buyer role may conflict with the role expectations (to protect knowledge) linked to a competitor. This conflict may hamper open communication and information provision, complicating the supplier's task to align the end product with the buyer's needs (Kelly, Schaan, & Joncas, 2002).
Further, interactions between a participating buyer and the supplier may overly focus on knowledge ownership. A buyer may only be willing to share knowledge when it can protect that knowledge from opportunistic appropriation. As the supplier dedicates more resources to cope with buyer appropriation concerns, the supplier can devote fewer resources to complete the task successfully (Nygaard & Dahlstrom, 2002;Yan & Dooley, 2013). Appropriation concerns may increase formalization, which may jeopardize the flexibility typically required in project work (Tatikonda & Rosenthal, 2000).
Finally, under buyer-as-competitor multiplexity, a buyer may not only be concerned with knowledge appropriation but also with passive supplier opportunism, such as quality shirking (Wathne & Heide, 2000). When the supplier fails to put in its best efforts because a buyer is also its competitor, this may lead to lower project success. In sum, even though buyer-as-competitor multiplexity increases the supplier's ability to internalize buyer knowledge, motivational concerns may hinder the realization of these benefits, decreasing project success. We hypothesize: H1 Buyer-as-competitor multiplexity negatively moderates the association between buyer participation in outsourced NPD and project success.

| Buyer-as-supplier multiplexity
Buyer-as-supplier multiplexity refers to a participating buyer sharing (one or more) role-reversal ties with the supplier. Role-reversal ties arise when the buyer and the supplier face each other in reciprocal roles (Bagchi, Koukova, Gurnani, Nagarajan, & Oza, 2016;Gimeno & Woo, 1996): buyer A and its supplier B share a rolereversal tie if A supplies parts, materials, or other resources to B for use in activities other than the focal F I G U R E 1 A contingency framework of the buyer participation-project success relationship Note: The signs indicate the direction of the proposed moderating effects NPD project. In business markets, role-reversal ties often occur when firms operate in multiple market segments. For example, Nokia is known as a device manufacturer, but it is also a leading provider of network equipment. A firm may simultaneously supply components to Nokia for its device production and purchase network equipment from Nokia. Similarly, Intel is a buyer of photolithography equipment, while it also provides silicon wafer mask design services to its suppliers of photolithography equipment.
In its participating-buyer role, a buyer is expected to commit resources to the project. In contrast, in its reverse role as a supplier, it is expected to use its bargaining power to guard its resources and maintain product specifications. As a result, the buyer may be unsure how strictly it can impose its demands on the supplier in the focal NPD project. Should the buyer be exigent and exercise its bargaining power, following the prototypical "the customer is king" notion according to which the supplier is expected to adhere to buyer demands (Cannon & Perreault Jr, 1999)? Or should the participating buyer be compliant and not exert its bargaining power, given the presence of a role-reversal tie and the risk of negative reciprocal actions (Bagchi et al., 2016)? Role reversal ties may thus create role ambiguity, in that parties have difficulty defining their role expectations. The lack of clear expectations may hinder effective communication (Kelly et al., 2002), undermine the information-sharing benefits of buyer participation, and lower project success.
In addition, role ambiguity reduces firms' "effort-toperformance" expectancy (Jackson & Schuler, 1985). When uncertainty regarding role expectations reduces a buyer's confidence in its participation efficacy, the buyer may become discouraged to fully deploy its innovative potential in the project (Solomon et al., 1985). Also, meetings between the buyer and the supplier risk being dominated by sorting out role expectations, detracting from the project at hand (Jackson & Schuler, 1985;Nygaard & Dahlstrom, 2002), engendering unproductive and misdirected behaviors (Dong et al., 2016;Van Sell, Brief, & Schuler, 1981), and lowering project success. We hypothesize: H2 Buyer-as-supplier multiplexity negatively moderates the association between buyer participation in outsourced NPD and project success.

| Buyer-as-partner multiplexity
Buyer-as-partner multiplexity occurs when a buyer that participates in the supplier's NPD process simultaneously shares (one or more) partner ties with the supplier outside of the NPD project. A partner tie is a collaborative arrangement among two or more organizations, intended to jointly acquire and utilize information and resources to develop new products (Mishra & Shah, 2009). In partnerships, firms collaborate as equals in the pursuit of mutually beneficial outcomes and develop normative expectations that help curb opportunistic behavior (Das & Teng, 1998).
In contrast to buyer-as-competitor multiplexity (which creates role conflict) and buyer-as-supplier multiplexity (which creates role ambiguity), buyer-as-partner multiplexity may lead to role synergy. Partners tend to develop knowledge-sharing routines (Dyer & Singh, 1998), that is, institutionalized processes purposefully designed to facilitate knowledge exchange. Especially when the knowledge to be shared is rich (e.g., because it relates to sophisticated technology), routines greatly benefit knowledge transfer (Reagans & McEvily, 2003). Knowledge-sharing routines created in the partner tie are also likely to facilitate knowledge exchange in the focal NPD project, expanding the supplier's array of development options during the NPD process (Fang, 2008). The larger number of alternative development paths the supplier can consider may lead to an improved product that better meets buyer needs. Therefore, project success is likely to be higher when a buyer and its supplier are also partners outside of the focal NPD project.
This effect may, however, not hold equally for all partnerships. Based on their formation process, we distinguish between two types of partnerships: engineered and emergent partnerships (Doz, Olk, & Ring, 2000;Koza & Lewin, 1999). Both are prominent in innovation (Millson, Raj, & Wilemon, 1996;Wallsten, 2000). Engineered partnerships are orchestrated by an overarching initiator (Doz et al., 2000), which often is a governmental or other nonprofit organization that wants to leverage innovation (Link & Scott, 2010). These topdown induced innovation projects are announced publicly, and organizations are invited to join. Partnering firms are teamed up without their full control of and consent to the match-making process. As a result, the partners have to cooperate, regardless of their intrinsic desire to do so. Sematech, an R&D partnership between 14 semiconductor manufacturers that has been initiated by the U.S. Department of Defense, is an example of an engineered partnership (Browning, Beyer, & Shetler, 1995). Other examples include the R&D consortia that develop new technologies under the European Framework Programmes supported by the European Commission. Engineered partnerships are typically formed in response to environmental changes that managers of different firms either interpret differently or overlook (Doz et al., 2000), requiring a triggering entity to make firms coalesce into a partnership.
In an emergent partnership, the partners initiate the partnership. They know each other's identity in advance and deliberately agree to develop new technology jointly. Examples of emergent partnerships include Nike and Philips Electronics jointly developing audio sports products and USCAR, the collaboration between Ford, Chrysler, and GM for automotive research. Emergent partnerships are typically initiated in response to environmental changes when managers of different firms recognize they have similar interests (Doz et al., 2000), and therefore tend to have precisely formulated objectives (Koza & Lewin, 1999). Thus, emergent partners may more easily coordinate and achieve consensus over the domain of collaboration (Doz et al., 2000).
Since engineered partners' different interpretations of their environment complicate communication (Doz et al., 2000), knowledge-sharing routines are less likely to emerge or are likely to emerge more slowly, in engineered partner ties than in emergent partner ties. Therefore, we expect stronger synergistic effects in the case of buyer-as-emergent-partner multiplexity than in the case of buyer-as-engineered-partner multiplexity. We hypothesize: H3 Buyer-as-engineered-partner multiplexity positively moderates the association between buyer participation in outsourced NPD and project success.
H4 Buyer-as-emergent-partner multiplexity positively moderates the association between buyer participation in outsourced NPD and project success.

H5
The association between buyer participation in outsourced NPD and project success is more positive for buyer-as-emergent-partner multiplexity than for buyer-as-engineered-partner multiplexity.

| Comparing buyer and supplier perceptions
Implicit in the theoretical arguments above is the idea that the buyer's and the supplier's perceptions of project success are affected similarly. However, as each party occupies a unique position in the buyer-supplier dyad, they may not always hold similar perceptions (Lumineau & Oliveira, 2018). We speculate that when buyer-as-competitor multiplexity increases, buyer participation is perceived more negatively by the buyer than the supplier. A project manager interviewed in our pilot study indicated that he deals with buyer-as-competitor multiplexity by "sharing what to do, but not how to do it." This may cause a buyer to believe (rightly or wrongly) that the supplier shirks and fails to share its latest knowledge and technologies, which may lower the buyer's perception of project success, over and above the negative effect on actual project success brought about by role conflict.
While there is no theoretical reason to expect systematic under-or overstated perceptions of project success under buyer-as-supplier multiplexity, we do speculate that perceptions of project success may be somewhat overstated under buyer-as-partner multiplexity. An interviewee in our pilot study indicated that "project managers tend to let buyers participate that are already partners outside the focal NPD project, without questioning their motives." This statement suggests that when buyer-as-partner multiplexity increases, the buyer's and the supplier's feelings of identification with each other may strengthen, akin to individuals who gradually develop feelings of identification with employers or educational institutions (Ashforth & Mael, 1989). Identification leads to overstated perceptions of project success (cf. Hewstone, Rubin, & Willis, 2002), above the beneficial effects on actual project success brought about by role synergy. 3 Yet, even though buyer-as-partner multiplexity might lead to overstated perceptions of project success, there is no reason to assume that one party's perception is overstated more than the other's.

| Empirical context
We obtained access to a unique sample of NPD projects carried out by a contract R&D organization (the supplier) for a variety of buyers. The contract R&D organization is a global firm operating in various high-technology markets associated with the aerospace industry, including aerospace craft, materials, onboard equipment, and avionics software. The firm is sufficiently large, ensuring an adequate sample size, and its buyers are sufficiently diverse to provide variation in our key constructs. Sourcing data from a single (supplier) firm allows for analyzing rich project data that otherwise would be difficult to obtain (cf. Easton & Rosenzweig, 2015;Linderman, Schroeder, & Choo, 2006). 4 Also, it "provide(s) for greater comparability of key dependent and independent variables" across projects (Chandy, 2003:353).
The aerospace industry is ideal for testing our hypotheses. First, the development of components, products, and technologies is often outsourced to specialized suppliers. For example, in developing the Boeing 787 Dreamliner, as much as 70% of the research and manufacturing of the plane was outsourced (Kotha & Nolan, 2005). Second, as several firms in this industry are active in multiple market domains, the prevalence of buyer-as-competitor multiplexity and buyer-as-supplier multiplexity is high. Partner ties are common too: several new products are developed together with buyers that closely collaborate with the supplier outside of the focal NPD project, in emergent or engineered partnerships.

| Sample
Our sampling frame consists of the project administration records for all 311 projects executed between 2001 and 2009 by project managers who were with the project from beginning to end (see, e.g., Tatikonda & Rosenthal, 2000 for a similar practice). 5 The project administration records contain key descriptors of the projects, including identification of the buyer, duration (start date and end date), and budget. The restriction of only including projects by project managers who had served through the life of the project assures that the project managers "could provide data on elements at different points in time in the project effort" (Tatikonda & Rosenthal, 2000:410).
We obtained permission from the executive board of the contract R&D firm to study a random sample of 150 NPD projects. We ensured that the sample was representative of the firm by comparing the 150 randomly selected NPD projects with the firm's portfolio of 311 eligible projects. Along dimensions we could measure for all projects (project budget, project duration, and the buyer's and the supplier's perceptions of project success), we found no significant differences between the sampled and the nonsampled projects (p > .10). Hence, it is unlikely that our sample is biased toward more successful projects. After accounting for nonresponse in the primary data collection (see Section 4.3.3), our final sample consists of 140 NPD projects. 6 All NPD projects in the sample are related to aerospace vehicles, systems and applications, and air transport operations. The average budget of the 140 sample projects is approximately $250,000, with a total of $35 million. The median start date of the projects is January This time length is not unusual in technology development: technology exploration followed by the development of technology applications and extensive testing takes time (Rosenberg, 1990). The 140 projects are executed for 40 buyer firms, by 72 project managers. Table 2 provides a summary of the NPD projects, buyers, and project managers in our sample.

| Data sources
The firm facilitated our research by providing us with a select set of internal documents, and by granting access to several managerial levels ranging from the project managers to the corporate office. We thus combined data from a wide variety of secondary and primary data sources.

| Secondary data sources
The secondary data sources include (a) project evaluation reports for our sample of 140 NPD projects, which are filed by the firm's quality assurance officers, and include the buyer's as well as the supplier's evaluations of project success, (b) the firm's strategic cooperation plans for the years 2000-2009 (2,500+ pages in total), which describe all partnerships it engaged in, and (c) the firm's procurement records between 2000 and 2009, which contain all of its purchase orders and thus provide insight into the firm's role-reversal ties. To construct the control variables (see Section 4.4.3 for details), we supplement these data with (d) company data from Compustat, Amadeus, and Mergent Intellect, (e) patent data divulged via the Derwent Innovations Index, (f) cultural data, taken from Hofstede's (2013) Value Service Module, and (g) economic data disclosed via the EU, the Federal Reserve, and other country-specific institutions.

| Primary data sources
We used two sources of primary data. First, the board allowed us to ask six senior managers to identify all competitors of the contract R&D firm among the buyers in the sample. Second, we had permission to administer a brief (one-page) survey to the project managers in early 2010, in which we asked them to report on the extent of buyer participation by consulting their confidential project plans and status reports. To assess respondents' abilities to retrieve data from these plans/ reports, the survey was pretested among five project managers who were not involved in the sampled projects. All project managers received an email from executive management endorsing the study and the online survey, stating the survey's purpose was "to gain an understanding of the effects of involving third parties in innovation." After 2 weeks, we sent a reminder. We carefully screened returned surveys for completeness. In the case of missing data, we asked the project managers to consult their project plans and status reports again and provide the missing information.

| Nonresponse
Our response rate was high: for 140 out of 150 projects, the project managers returned the survey. We assessed nonresponse bias by comparing the 10 projects for which the survey was not returned with the 140 remaining projects in terms of project budget, project duration, relationship multiplexity, and the buyer's and the supplier's perceptions of project success. We did not find statistical differences (all p's > .10). Hence, we do not find evidence of response bias.

| Measures
Appendix A2 reports measures and data sources for all variables in this study.

| Dependent variables
Our dependent variables capture the evaluation of an NPD project right after project completion, consistent with the NPD (e.g., Shenhar, Dvir, Levy, & Maltz, 2001), outsourcing (e.g., Palvia, King, Xia, & Palvia, 2010), and R&D literature (e.g., Pearson, Nixon, & Kerssens-van Drongelen, 2000). At project completion, both the supplier's project manager and the buyer (i.e., matched dyads) formally appraised the project's success by independently completing a survey administered by the Profile of project managers (n = 72 project managers) supplier's quality assurance officers. In our setting, subjective measures of project success are preferable to operationally defined measures as technologies and competitive environments that vary across projects make operational measures difficult to compare (Bozarth & Edwards, 1997). We received the project evaluation reports from the supplier's quality assurance officers.

Perceptions of project success
The buyer's perception of project success (BUYSUC) is measured using two 5-point items (see Tatikonda & Montoya-Weiss, 2001, for a similar measure). The items are filled out by the project manager's primary contact at the buyer firm in an online survey administered by one of the supplier firm's quality assurance officers immediately following project closure. The supplier's perception of project success (SUPSUC) is measured through a single 5-point item that is completed by the focal supplier's project manager in a survey immediately following project closure, again at the request of one of the supplier's quality assurance officers. This measure is similar to the single item used by Petersen, Handfield, and Ragatz (2003). Practitioners often favor single-item measures to minimize respondent refusal due to over-surveying and respondent resentment due to being asked seemingly repetitive questions (Bergkvist & Rossiter, 2007;Drolet & Morrison, 2001). These considerations were also underlying the supplier's decision to use short measures. Although conventional measurement wisdom advocates multi-item scales, there is mounting evidence that singleitem measures have adequate reliability levels (Wanous & Hudy, 2001;Wanous, Reichers, & Hudy, 1997), and validity levels that are equal to multi-item measures when the construct has a simple, clear object (e.g., a project, a job, or a brand) and attribute (e.g., success, satisfaction, or liking) (Bergkvist, 2015;Rossiter, 2010). 7

Buyer participation
The extent of buyer participation (PART) in the supplier's NPD process is carefully documented in the project managers' project plans and status reports, which are subject to buyer approval. The project plan is a formal document written by the project manager, defining the project's objectives and activities, estimating the time and resources required to complete each activity, and specifying the parties to be involved in each activity. During the project, the project manager provides regular updates in project status reports. Together, the project plan and the project status reports describe whether, when, and how the buyer participates in NPD. As the supplier firm did not permit us to analyze these confidential documents (due to the presence of intellectual property and personnel information that would not only require the consent of the supplier but also of the buyer), we asked project managers to report on buyer participation in a brief (onepage) survey while they consulted their project plans and status reports, to retrieve precise information on the activities a buyer participated in.
Following Fang et al. (2008), our measure of buyer participation is a formative index that captures both breadth (number of activities in the supplier's NPD process in which a buyer participates) and depth of buyer participation (how deeply a buyer participates in each of these activities). Based on the NPD literature and preparatory interviews with five supplier project managers, we identified 12 activities that can be part of an NPD project executed by the contract R&D firm. For each of the 12 activities, we asked the project managers to check whether the activity was included for each sample project and, if so, whether the buyer participated in that activity (0 = did not participate, 1 = did participate). We used the ratio of the number of activities in which the buyer participated relative to the number of activities present in the project to measure the breadth of customer participation. If the buyer participated, we subsequently asked about buyer-participation depth using a 7-point Likert scale, ranging from "very superficially involved" to "very deeply involved." We determined overall participation depth across the activities a buyer was involved in by averaging the completed items. To combine breadth and depth into a single formative measure of buyer participation, we standardized them to obtain comparable scales and then took the average.

Relationship multiplexity variables
We operationalize buyer-as-competitor multiplexity (MPX_COMP) at the firm level as the share of the 17 departments of the supplier firm that compete with the buyer firm. We asked six senior supplier managers, for every buyer they were knowledgeable about, to indicate which of the 17 departments compete with the buyer. We received 3.1 evaluations per buyer, on average. We measure buyer-as-supplier multiplexity (MPX_SUP) as the number of projects in which a buyer serves as a supplier to the contract R&D firm, for activities other than the outsourced NPD project, during the focal project period. We coded buyer-as-supplier multiplexity based on the firm's procurement records between 2000 and 2009, which contain all of its purchase orders and thus provide insight into the firm's role-reversal ties. For buyer-as-partner multiplexity, we coded the firm's strategic cooperation plans (2,500+ pages in total), which describe all partnerships-engineered and emergent-the firm engaged in during this period. We measure engineered partner ties (MPX_ENG) and emergent partner ties (MPX_EM) as the number of engineered and emergent technology development projects in which the contract R&D firm and a buyer collaborated during the focal project period, excluding the focal project. We log-transform MPX_SUP, MPX_ENG, and MPX_EM.

| Control variables
We include control variables relating to the (a) stage of participation, (b) project, (c) project manager, (d) buyer firm, (e) buyer firm's industry, (f) buyer-supplier relationship, and (g) triad levels to isolate potentially spurious effects. First, the stage of codevelopment may affect the success of an NPD project (Chang & Taylor, 2016;Fang, Lee, & Yang, 2015). We control for whether the buyer participates in preparatory, core development, and/or completion project activities. To this end, we categorize the 12 activities that can be part of an NPD project executed by the contract R&D firm into these three stages. We define three dummy variables (PART_ST1, PART_ST2, and PART_ST3) that equal one if the buyer participates in at least one activity at that stage.
At the project level, we account for the project's launch year (PR_YEAR) and the project's size (PR_SIZE), measured by its budget. We also control for the project's complexity (PR_COMP), measured by the number of activities included in the project, and the technological maturity at the start of the project (PR_TECH). We measure the latter using the project's Technology Readiness Level (TRL), a measure developed for and used by the aerospace industry. The TRL captures the development stage of a technology or product and has been recognized by operations management scholars as a useful tool for scouting and assessing technologies (e.g., Krishnan, 2013). A higher TRL means that the buyer and the supplier are looking for an applicative solution in the scope of the project, while a lower TRL indicates a more fundamental project. Projects with a lower TRL at the start thus tend to be more novel than those with a higher TRL.
Next, numerous operations management studies emphasize the importance of controlling for traits of human agents when studying interorganizational dynamics (e.g., Tangpong, Hung, & Ro, 2010;Zhang, Viswanathan, & Henke Jr, 2011). Hence, at the project manager level, we follow Doney and Cannon (1997) and control for the project manager's experience (MAN_EXP), which we measure as the number of projects s/he completed in the 10 years preceding the start of the focal project, and his/her clout or influence within the supplier firm (MAN_CL), which we proxy by the average budget of the projects for which s/he was responsible in the preceding 10 years.
At the buyer firm level, we account for the buyer firm's size (BUY_SIZE), measured as the number of employees at the project start, and its technological innovativeness, measured as the number of forward citations to its patents (BUY_INN). Similar to Liu, Yeung, Lo, and Cheng (2014), we identified the patents that a buyer firm obtained over the 5 years preceding the focal project using the Derwent Innovations Index. We then counted the number of forward citations by other firms in the years after each patent was registered successfully. Except for PR_YEAR, PR_COMP, and PR_TECH, all project, project manager, and buyer firm level variables are logtransformed.
At the buyer firm's industry level, we control for demand uncertainty (IND_UNC) and demand growth (IND_GROW). We collect annual data on the gross value added (a measure of demand) in a buyer's two-digit NAICS industry in its home country for the 10 years before project completion. To measure demand uncertainty, we regress the industry-level gross value added for the 10 years preceding the project closing on the year variable. Subsequently, we divide the SE of the regression slope by the mean gross added value (Dess & Beard, 1984;Patel, 2011). Demand growth is operationalized as the average annual growth of gross value added in a buyer's industry during the project (Mishra, Modi, & Animesh, 2013). We also control for unobserved industryspecific effects through four industry dummy variables (IND_D1-IND_D4), based on two-digit NAICS codes.
At the buyer-supplier relationship level, we control for relationship-multiplexity scope (REL_SCOPE), which reflects the number of roles a buyer plays in relation to the supplier in addition to the buyer role. This variable ranges from zero (if a buyer only takes on the buyer role) to four (if a buyer also assumes a competitor, supplier, engineered partner, and emergent partner role, in addition to the buyer role). Playing multiple roles could lead to lower project success due to the increased need to sort out role responsibilities-the "overload" that arises when one tries to juggle too many roles. We also control for the geographic distance (REL_GEO) and the cultural distance (REL_CUL) between a buyer and the supplier, to take into account that distant exchange partners may encounter more communication and coordination difficulties (Handley & Benton, 2013). We measure geographic distance (log-transformed) based on FreeMapTools.com, and cultural distance in line with Kogut and Singh (1988). We further control for the firms' relationship history (REL_HIS) by including the proportion of months with at least one ongoing buyer-supplier project in the 5 years preceding the start of the focal project. Relationship history may make outsourced NPD projects easier to execute, improving project success (Fang, 2008).
Next, we control for the number of concurrent projects the supplier executes for a buyer. The length of technology development (Rosenberg, 1990) increases the occurrence of concurrent outsourced NPD projects for the same buyer. Executing multiple projects for the same buyer stimulates reciprocity (Kogut, 1989) and cooperativeness between buyer and supplier (Heide & John, 1990), which may increase the effectiveness of buyer participation. Consistent with the multiplexity variables, we control for (a) the main effect of the number of concurrent projects executed for the buyer and (b) its interaction with buyer participation. We measure concurrent projects (REL_CONC) as the logarithm of the number of other projects (besides the focal project) the supplier executes for a buyer during the focal project period.
Finally, triads-subsets of three firms and the ties between them-may affect cooperative-competitive relationships (Choi & Wu, 2009;Dubois & Fredriksson, 2008;Wilhelm, 2011). We measure triadic competition as the number of a buyer's competitors that the supplier serves during the focal project period. Serving a buyer's competitors may alter the effectiveness of buyer participation in the focal outsourced NPD project, although it is not clear upfront in which direction. On the one hand, the supplier gains knowledge on a buyer's industry by working for its competitors (cf. Chen, 1996;Lane & Lubatkin, 1998). The supplier's increased ability to internalize knowledge shared by a buyer (Cohen & Levinthal, 1990) may increase the effectiveness of buyer participation. On the other hand, when the supplier serves a buyer's competitors, a buyer may fear knowledge leakage, reducing the effectiveness of buyer participation (Hernandez, Sanders, & Tuschke, 2015). Therefore, similar to the focal multiplexity variables in our analyses, we control for both the main effect of triadic competition and its interaction with buyer participation. We operationalize triadic competition by counting the number of competitors to a buyer for which the supplier executed a project during the focal project period (REL_TRIAD), based on their four-digit NAICS codes and the supplier's project administration records. 8 4.4.4 | Common method and single respondent bias, memory bias, and social desirability bias Our research design, which uses a wide variety of data sources, circumvents common method and single respondent bias concerns. Although the questions to capture SUPSUC and PART are both answered by the supplier's project managers, the variables are (a) taken from different data sources (SUPSUC: supplier's project evaluation records vs. PART: authors' survey) and (b) measured on different occasions (SUPSUC: directly after project closure vs. PART: early 2010), reducing the possibility for the supplier's project managers to make systematic attributions. Also, for all hypotheses, the dependent variables and moderators come from different data sources, further reducing the probability of common method bias (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003).
Memory bias is minimized in various ways. First, for our dependent variables, we use data that were collected by the supplier firm immediately following project closure. Second, for the measurement of buyer participation, we asked the project managers to consult their project plans and project status reports. Third, we use archival data for all but one relationship multiplexity variable (the exception is buyer-as-competitor multiplexity) and for all our control variables.
Finally, we do not expect social desirability to bias our measure of the buyer's perception of project success for three reasons. First, the supplier firm's contact who reached out to the buyer to administer the survey was not the project manager, but the supplier firm's quality assurance officer. As the quality assurance officers had not been interacting with the buyer firms' project managers, there was little social pressure to conform. Second, we do not find response range compression in our sample, which would have been a sign of social desirability bias (Podsakoff et al., 2003). Third, supported interaction hypotheses are less subject to response bias because it is unlikely that respondents have an "interaction-based" theory in their minds when responding (Aiken & West, 1991). We find many significant interaction effects, as reported in Section 5.2, which offers further evidence against this bias (see Chen, Neubaum, Reilly, & Lynn, 2015, for a similar argument).

| Estimation
We estimate two equations with, respectively, the buyer's and the supplier's perceptions of success in project i executed for buyer j as the dependent variables: Z ij is a vector of control variables. The effects of these variables on the buyer's and the supplier's perceptions of project success are captured through the vectors θ and Φ, respectively. We use mean-centering before forming the interactions in Equations (1) and (2) to ease interpretation.
Three factors complicate our model estimation: endogeneity of buyer participation, clustered observations within buyers and project managers, and simultaneity of the buyer's and the supplier's perceptions of project success.

| Addressing endogeneity
Buyer participation (PART) may be chosen strategically to optimize project success. Ignoring these endogeneity concerns could lead to biased and inconsistent model parameters. We address these concerns by using the control function approach (Petrin & Train, 2010). As the exclusion instrument, we use the average level of buyer participation in the buyer's industry (defined per the four-digit NAICS code), excluding the buyer participation of the buyer firm itself (see, e.g., Han, Mittal, &Zhang, 2017, andPanagopoulos, Mullins, &Avramidis, 2018, for a similar practice). We expect the average industry-level buyer participation to be a strong instrument, as it is likely that a buyer's level of participation in an NPD project is similar to the participation level of its peers in the same industry, as they face the same market conditions (Germann, Ebbes, & Grewal, 2015). Empirically, our instrument correlates strongly with the buyer-participation variable (F = 21.34, p < .001, which surpasses the threshold of F > 10 suggested by Staiger & Stock, 1997). In terms of validity, the average participation level of other buyers is not likely to be related to the focal project's success and, therefore, the error term.
We perform a first-stage regression with PART as the dependent variable and the two instruments as the independent variables and obtain the predicted residual. Next, we add this residual as a correction term to both Equations (1) and (2). The correction term in the supplier equation is significant (p < .05), indicating the need to correct for endogeneity.

| Addressing clustered observations
Our sample contains more than one project for 21 buyers. The inclusion of multiple projects for the same buyer in the sample is representative for the contract R&D firm's order book and the aerospace industry in particular, and other business markets in general-including automotive (Liker & Choi, 2004), pharmaceuticals (Wuyts, Dutta, & Stremersch, 2004), and telecommunications (Wu & Choi, 2005). In addition, our sample contains more than one project for 34 project managers. To account for the dependence of the observations, we use a two-way robust clustered-error term procedure, which adjusts the SEs to make them heteroskedasticity-consistent (Cameron, Gelbach, & Miller, 2011). The clustering dimensions are the buyer firm and the project manager.

| Addressing simultaneity
If unobserved factors explain variation in both the buyer's and the supplier's perceptions of project success, the error terms are correlated. Seemingly unrelated regression (SUR) accounts for potentially correlated errors across the two equations. 9 We estimate the SUR models in STATA 14.2 through Roodman's conditional mixed process approach, which uses a maximum likelihood estimator (Roodman, 2011), accounting explicitly for clustered observations as well as the likely correlation between the buyer's and the supplier's perceptions of project success.

| Descriptive insights
Buyers participate in 132 of the 140 projects, in line with our pilot study's observation that buyers push suppliers to let them participate in their outsourced NPD projects. On average, buyers participate in 73% of the activities in these 132 projects. Their average depth of participation is above the scale midpoint (4.5 on a 7-point scale). Buyer participation differs dramatically between projects: the percentage of activities in which buyers participate ranges from 9 to 100%, and their depth of participation ranges from 1 to 7 on a 7-point scale. Multiplex relationships are commonplace: 54 of the 140 projects are executed for buyers that also compete with the contract R&D firm in another product market; buyer-as-supplier ties occur in 79 out of 140 projects; 87 (101) projects are executed for buyers that also hold an engineered (emergent) partner role in one or more projects outside of the focal NPD project. Table 3 provides descriptives.

| Hypotheses testing
We report the results in Table 4. We used a hierarchical regression procedure to illustrate the model's increased explanatory model when adding the variables of interest. Model 1 shows results with only control variables. Model 2 adds the main effect of buyer participation in NPD, the main effects of the multiplexity variables, and the interaction effects of buyer participation with the control variables "concurrent projects" and "triadic competition." Model 3, which adds multiplexity interaction effects and is used for hypothesis testing, outperforms Models 1 and 2 in terms of model fit. The largest variance inflation factor (VIF) among the independent variables is 3.42 (mean VIF = 2.25). Given our sample size (n = 140), multicollinearity is thus not a major concern (Guide & Ketokivi, 2015). In presenting the results, we use "b buy " and "b sup " to signify the effect on the buyer's and the supplier's perceptions of project success, respectively. We find a nonsignificant effect of buyer participation on both the buyer's (b buy = −.03, n.s.) and the supplier's perceptions of project success (b sup = −.28, n.s.). More importantly, these effects strongly depend on relationship multiplexity. Consistent with H 1 , we find a negative interaction effect between buyer participation and buyer-as-competitor multiplexity on the buyer's perception of project success (b buy = −1.05, p < .01). Simple slope analyses show that, in the absence of buyer-ascompetitor multiplexity, buyer participation is positively and significantly associated with the buyer's perception of project success (b buy_ABSENCE = .34, p < .001), but this association turns negative for high levels of buyer-ascompetitor multiplexity (b buy_HIGH = −.16, p < .001), where a high level is reflected in one SD above the mean. 10 However, for the supplier's perception of project success, we do not find support for an interaction between buyer participation and buyer-as-competitor multiplexity (b sup = −.20, n.s.).
In line with H 2 , we find buyer-as-supplier multiplexity negatively moderates the associations between buyer participation and the buyer's (b buy = −.10, p < .01) and supplier's (b sup = −.07, p < .10) perceptions of project success. All else equal, the association of buyer participation with project success is positive for relationships without buyer-as-supplier multiplexity (b buy_ABSENCE = .10, p < .01; b sup_ABSENCE = .07, p < .01), but negative for relationships with high levels of buyer-as-supplier multiplexity (b buy_HIGH = −.11, p < .01; b sup _ HIGH = −.07, p < .001). A high level is again one SD above the mean.
Contrary to H 3 , we do not find a significant interaction between buyer participation and buyer-asengineered-partner multiplexity (b buy = .04, n.s.; b sup = .02, n.s.). Consistent with H 4 , we find that buyer-as-emergent-partner multiplexity positively moderates the associations of buyer participation with the buyer's (b buy = .16, p < .01) and supplier's (b sup = .18, p < .01) perceptions of project success. Parameter distance tests (β 9 -β 8 and γ 9 -γ 8 ) reveal that the associations of buyer participation with both the buyer's (p < .05) and supplier's (p < .001) perceptions of project success are more positive for multiplex relationships consisting of emergent rather than engineered partner ties outside the focal NPD project, lending statistical support for H 5 . Simple slope analyses reveal that buyer participation is strongly negatively associated with project success in the case of no buyer-as-emergent-partner multiplexity (b buy_ABSENCE = −.21, p < .01; b sup_ABSENCE = −.23, p < .001). This negative association becomes positive for high levels of buyer-as-emergent-partner multiplexity (b buy_HIGH = .15, p < .01; b sup_HIGH = .17, p < .01). Figure 2 visualizes these analyses.
Turning to the control variables, we find a positive effect of buyer participation during the core-development stage on the buyer's perception of project success (b buy = .34, p < .01), and a negative effect of buyer participation during the project-completion stage on the supplier's perception of project success (b sup = −.28, p < .01). These results indicate that a buyer benefits more from participation in activities where major technological choices are yet to be made, allowing the buyer to more readily secure solutions that serve its specific needs (Fang et al., 2015). Buyer input during the project-completion stage may be harder for the supplier to accommodate since changing technologies in a late stage comes at great costs (Carbonell, Rodríguez-Escudero, & Pujari, 2009).
Project manager experience lowers the buyer's as well as the supplier's perceptions of project success (b buy = −.14, p < .01; b sup = −.23, p < .001). Possibly, extensive project manager experience raises both parties' expectations, which may hurt their perceptions of project success. Further, the buyer's perception of project success is negatively associated with a project's technological  maturity (b buy = −.04, p < .05) and positively with industry growth (b buy = .13, p < .001) and geographic distance (b buy = .47, p < .05). The supplier's perception of project success, on its turn, is negatively related to project size (b sup = −.10, p < .05) and relationship history (b sup = −.42, p < .10) and positively to project complexity (b sup = .08, p < .05), project manager clout (b sup = .24, p < .01), and demand uncertainty (b sup = 7.99, p < .05). Finally, and somewhat surprisingly, the association between buyer participation and the buyer's perception of project success is lower when the supplier executes more projects for the buyer concurrently (b buy = −.08, p < .10). Finally, triadic competition is positively associated with the buyer's perception of project success (b buy = .02, p < .01), as is its interaction with buyer participation (b buy = .03, p < .01). We return to these findings in the discussion section.

| Robustness checks
To validate our results, we perform several robustness checks, reported in Table 5.

| Ruling out memory bias in the buyer participation measure
Following Fang et al. (2008), our measure of buyer participation captures both breadth (the number of activities in the supplier's NPD process in which a buyer participates) and depth (how deeply a buyer participates in each of these activities). One may argue that the breadth dimension is less likely to be subject to memory bias than the depth dimension. We therefore re-estimated the model in Equations (1) and (2), replacing our two-dimensional measure of buyer participation with the breadth dimension only (operationalized as the ratio of the number of activities in which the buyer participated to the number of activities present in the project). The results (see Table 5, Model 3a) are substantively the same. 11 5.3.2 | Using alternative time windows to operationalize the multiplexity variables Consistent with our conceptualization, we measured the multiplexity variables during the focal project period. 12 We re-estimated Equations (1) and (2) with the measures for MPX_SUP, MPX_ENG, and MPX_EM based on alternative time windows, starting one, two, or 3 years before the project start date and ending at project completion. Results are substantively the same (see Table 5, Model 3b, for the results based on the time window starting 3 years before the project start date).

| Using an alternative measure for the supplier's perception of project success
The measurement items for the buyer's and the supplier's perceptions of project success were constructed by the quality assurance officers, as part of the firm's survey. These measures have the advantage that the quality assurance officers administered them immediately after project completion and, therefore, they are not subject to memory bias. Their disadvantage, though, is that they are similarly but not identically worded. Also, the quality assurance officers used two items to measure the buyer's perception, but only a single item to measure the supplier's perception. In our own survey of these project managers, we used a two-item scale identical to the one used to measure the buyer's perception of project success. We established the reliability of the original single-item measure by analyzing how it relates to the two-item alternative: a correlation of .52 indicates adequate internal consistency (Hair, Anderson, Tatham, & Black, 1998:118). We re-estimated our model using this two-item measure. Results (Table 5, Model 3c) are substantively the same.

| Including additional interactions
Different relationship multiplexity types are not necessarily mutually exclusive but can occur in combination, reinforcing or obstructing one another. We re-estimated Equations (1) and (2), including all 6 two-way interactions between the four multiplexity types. No interactions are significant (all p's > .10), nor do the focal effects change substantially. 13 6 | DISCUSSION

| Theoretical contributions
Our study contributes in three ways. First, our findings extend the literature on buyer participation in NPD. Hitherto, buyer participation research has assumed a single tie between buyer and supplier (e.g., Chatterji & Fabrizio, 2014;Fang, 2008). We recognize that many buyer-supplier relationships are multiplex in nature. Building on existing research on multiplex relationships (e.g., Ross Jr. & Robertson, 2007;Shipilov & Li, 2012Tuli et al., 2010) and using a F I G U R E 2 Simple slope analyses. † p < .10; *p < .05; **p < .01; ***p < .001 Note: High multiplexity is defined as one SD above the mean T A B L E 5 Robustness checks (n = 140 projects)   uniquely assembled database combining several sources of archival data with survey data, we show that the effects of buyer participation are contingent on whether a participating buyer is also a competitor, supplier, or partner of the supplier firm. Second, we contribute to role theory, which has mainly studied impacts of conflict, ambiguity, and, to a lesser extent, the synergy between the standards, values, and goals that a party experiences within one role (Biddle, 1986;Dong et al., 2016). Our study contributes by developing theoretical arguments that the different ties in a multiplex relationship cause a party to hold multiple functional roles, which may also trigger role conflict, role ambiguity, and role synergy. Our focus on the different roles actors play within a single relationship contrasts with the emphasis in traditional role theory on the different roles actors play across different relationships.

Conditional mixed process regression results
Third, the four types of multiplex buyer-supplier relationships identified in this study-buyer-as-competitor, buyer-as-supplier, buyer-as-engineered-partner, and buyer-as-emergent-partner-differ in their relational posture. Kim and Choi (2015) identified relational posture as one of two key dimensions in their buyer-supplier relationship typology. In their typology, relational posture ranges from adversarial to cooperative. In our context of multiplex relationships, buyer-as-competitor multiplexity maps on the adversarial end of the scale, followed by buyer-as-supplier multiplexity. Buyer-as-engineeredpartner and buyer-as-emergent-partner multiplexity map on the cooperative end of the scale. The second dimension of Kim and Choi's buyer-supplier relationship typology is relational intensity, which is captured by our central buyer participation variable as higher degrees of participation intensify the buyer-supplier relationship . Kim and Choi propose that as adversarial relationships intensify, they turn "sticky" and lead to adverse outcomes, such as less information sharing, more opportunism, and decreased commitment. As cooperative relationships intensify, they become "deep" and bring positive results such as increased relational stability, more effective communication, and lower coordination costs.
Interestingly, we find the association between buyer participation and the buyer's perception of project success ranges from strongly negative under buyer-as-competitor multiplexity (−1.05), to slightly negative under buyer-assupplier multiplexity (−.10), to neutral under buyer-asengineered-partner multiplexity (.04), and to positive under buyer-as-emergent-partner multiplexity (.16), a pattern consistent with Kim and Choi's typology. We find a similar pattern of associations between buyer participation and the supplier's perception of project success (ranging from −.20 to −.07 to .02 to .18), even though the negative effect of buyer-as-competitor multiplexity (−.20) is not significant. Significant parameter distance tests (p < .05) for five of six contiguous parameter pairs (the exception being γ 8 -γ 7 ) offer further support for the typology.
Fourth, the two constructs that are competitionrelated-buyer-as-competitor multiplexity and triadic competition-moderate the effects of buyer participation on the buyer's perception of project success in opposite ways. Being a buyer that is also a competitor to the supplier reduces the effectiveness of buyer participation in the eyes of the buyer, whereas being a buyer whose competitors are served by the supplier increases the effectiveness of buyer participation, according to the buyer. What explains this difference? If fear of knowledge leakage would be the predominant driver of these competition- Wald Chi-square 4.48 * 10 10 (39 df )*** 1.24 * 10 10 (39 df )*** 3.01 * 10 10 (39 df )*** Notes: Values in bold correspond to hypotheses. In all models, endogeneity corrections have been performed by adding the (model-specific) correction term that was obtained through the control function approach. In Model 3b, we only report the results for the 3-year lead due to space constraints. The results for the 1-and 2-year leads are similar and can be obtained from the first author. Abbreviations: BUYSUC, buyer's perception of project success; SUPSUC, supplier's perception of project success. † p < .10; *p < .05; **p < .01; ***p < .001 (two-tailed).
related effects (as we expected), we should have observed negative interaction effects for both buyer-as-competitor multiplexity and triadic competition. Instead, the positive interaction for triadic competition implies that the bright side of the supplier working with a buyer's competitorsviz., the supplier's increased industry knowledge and familiarity with industry verbiage, which may facilitate the application of a buyer's shared knowledgeoutweighs a buyer's knowledge-leakage fears. As to buyer-as-competitor multiplexity, we hypothesized that-besides knowledge-leakage fears and better application of the knowledge shared-a third factor may come into play: the fear that one's partner may shirk (e.g., by not sharing its latest technologies and knowledge). The negative interaction that we find for buyer-as-competitor multiplexity may indicate that a buyer's concerns that the competing supplier might shirk outweigh any positive effects from better knowledge application, leading to a negative (net) effect on the buyer's perception of project success.
Finally, recognizing that a pluralistic view addressing both parties' perceptions is particularly relevant when researching phenomena that are inherently multiparty (such as buyer participation in outsourced NPD), we investigated both parties' perceptions of project success. Our approach to incorporate both the buyer's and the supplier's perceptions uncovered an intriguing difference across the dyad. We find that the hypothesized effects on the buyer's and the supplier's perceptions are always highly similar, except for the two competition-related constructs: buyer-as-competitor multiplexity and triadic competition. Both constructs affect the buyer's perception but not the supplier's perception of project success. The nonsignificant effect for triadic competition might indicate that suppliers are more concerned about knowledge-leakage fears than buyers: indeed, the null effect may imply that the supplier's knowledge-leakage fears cancel out the positives from better understanding the buyer's needs. The nonsignificant effect for buyer-as-competitor multiplexity could further be explained by the fact that, in contrast to a buyer, the supplier may not see buyer shirking as a threat, since a buyer is the principal beneficiary of the project.

| Managerial implications
Our findings have tactical implications for managers deciding on buyer participation. First, on average, neither the buyer nor the supplier perceives a project to be more successful when a buyer participates more. Instead, buyer participation in outsourced NPD can help or hurt, depending on the different ties that coexist between buyer and supplier.
We started our investigation asking whether a buyer that also plays a competitor role in another product market should participate in outsourced NPD. As our pilot study indicated, the answer is not obvious. Some interviewees prefer to keep competitors at a distance and others acknowledge the positive aspects of involving them. Our empirical test shows different effects on the buyer's and the supplier's perceptions of project success. While the supplier's perception of project success does not decline under buyer-as-competitor multiplexity, the buyer's perception does. Presumably, a buyer that is also a competitor feels it always needs to watch its back when participating in outsourced NPD due to fears that the supplier may shirk by not sharing its latest technologies and knowledge (a form of passive opportunism), lowering the buyer's perception of project success. Hence, even though the supplier may not perceive this as detrimental, we advise against involving a buyer that is also a supplier's competitor in other product markets.
Should a buyer that also is a supplier to its supplier participate in outsourced NPD? Although the managers we interviewed were in doubt, our empirical results were quite clear: buyer-as-supplier multiplexity lowers the effectiveness of buyer participation. While buyer participation is positively associated with both the buyer's and the supplier's perceptions of project success for relationships without buyer-as-supplier multiplexity, these effects reverse and become negative when buyer-as-supplier multiplexity increases. We argue that the buyer's and the supplier's roles in bargaining for resources and product specifications during negotiations shift from one tie to the other. The resulting role ambiguity may decrease the buyer's ability and motivation to fully deploy its innovative potential in its participating-buyer role, thereby reducing both parties' perceptions of project success. Hence, a high degree of buyer participation can become counterproductive when the buyer and the supplier share more role-reversal ties.
Should a buyer that also plays a partner role participate in outsourced NPD? The managers we interviewed in our pilot study thought the answer was "yes." However, empirical evidence reveals big differences between engineered and emergent partner ties, which implies the initiation process underlying a partnership impacts the parties' interactions during the partnership. Specifically, while buyer participation is negatively associated with the buyer's and the supplier's perceptions of project success for relationships without buyer-as-emergent-partner multiplexity, these associations become positive for high levels of buyer-as-emergent-partner multiplexity. These findings are consistent with our arguments that knowledge-sharing routines, which allow for role synergy, are a natural outcome of emergent partnerships but not of engineered partnerships. The role synergy may make knowledge-sharing between a supplier and a participating buyer more effective, leading to higher-quality buyer participation, and improved project success. As such, more emergent partner ties between buyer and supplier calls for higher degrees of buyer participation as their benefits carry over to the focal buyer-supplier tie, increasing both the buyer's and the supplier's perceptions of project success. In contrast, buyer participation without emergent-partner ties is damaging. Buyer-asengineered-partner multiplexity, on the other hand, does not alter the effects of buyer participation on project success, suggesting that engineered partnerships do not carry the same role synergy benefits as emergent partnerships. These results demonstrate the importance of distinguishing between emergent and engineered partnerships.
Finally, two control variable findings stand out. First, when the supplier concurrently executes more projects for the buyer, the impact of buyer participation on the buyer's perception of project success is lower. Possibly, a buyer believes that the supplier should already understand its needs because of the concurrent projects, and sees further information and knowledge sharing via buyer participation as less instrumental and perhaps even a waste of its resources. Thus, involving a buyer for whom the supplier already concurrently carries out other projects is not advisable as it negatively affects the buyer's perception of project success. Second, when the supplier also serves a buyer's competitors (which we label triadic competition), the association between buyer participation and the buyer's perception of project success is strengthened. Working for a buyer's competitors is likely to enhance the supplier's ability to speak the buyer's language, which may increase the participating buyer's perception of project success. This finding points to a bright side of triadic competition: it is in the supplier's best interest to involve a buyer that competes with other buyers the supplier is serving. In a post hoc roundtable in which we discussed our findings, a senior marketing manager commented on this result: "We do work for clients that compete with each other. For example, client XYZ competes with client ABC, and we execute projects for both, often simultaneously. Per internal guidelines, we do not directly transfer knowledge from client XYZ to other clients. However, our engineers and project managers do learn from executing these NPD projects-and the body of knowledge of our firm grows. As a result, client ABC seems to benefit from our increased knowledge indirectly, especially if ABC is closely involved in the NPD process." Not every project manager deciding about buyer participation may be aware of the added roles a buyer plays in the relationship with the supplier firm. During the post hoc roundtable, several project managers indicated that, in light of our findings, it would help to have full knowledge of relationship multiplexity before project planning. We recommend extending supplier account management practices by including a check of competitor, supplier, engineered partner, and emergent partner ties with buyer firms.

| Limitations and opportunities for further research
Our study's limitations potentially offer interesting avenues for future research. First, although our sample involves multiple NPD projects and numerous buyers, it only includes one supplier firm. Follow-up studies could investigate the generalizability of our findings by studying multiple suppliers across various industries. Also, the industry examined is characterized by knowledge bases that are complex, expanding, and in a constant state of flux, making it hard to recruit and retain qualified professionals. Given the critical role of knowledge sharing in our theory development, the level of knowledge intensity in an industry may form a boundary condition to our results. In knowledge-intensive industries, the locus of innovation is typically found in interorganizational collaborations (Powell, Koput, & Smith-Doerr, 1996), rather than in individual firms. A senior project manager quote pointedly illustrates the need for buyer participation given the knowledge-intensive nature of the photolithography industry: "Nobody knows everything in this industry. It's just not possible. Knowledge develops too fast. So here we are: the buyer knows everything about his problems, and we know everything about our solutions. Involving the buyer brings those two perspectives together." Second, we were limited by the project success data collected by the supplier firm's quality assurance officers. Although used in practice, we acknowledge that these measures do not offer the richness of multi-item scales.
We suggest future research to study other outcomes, such as knowledge development, project speed, cost, and customer lifetime value. We also faced confidentiality constraints. For example, we did not get permission to consult the supplier's confidential project plans and project status reports, which might have provided more insight into the share of each activity that the buyer and the supplier performed.
Third, due to the cross-sectional nature of our study, we cannot unequivocally infer causality. Longitudinal data may allow future research to better separate cause from effect. Alternatively, a study that adopts a scenariobased experimental design, even though weaker on external validity, would help infer causality. Further, we argued that buyer participation in the case of relationship multiplexity may give rise to role conflict, role ambiguity, and/or role synergy. Although our empirical evidence is consistent with these predictions, our data does not allow for mediation tests.
Fourth, we studied the context of outsourced NPD, where the developing supplier involves the buyer in NPD. Relationship multiplexity may also occur in the case of supplier involvement in NPD when a developing firm involves a supplier in its NPD project. The theoretical arguments on which we based our hypotheses do not depend on which one of the two parties participates in the other party's NPD project. Still, a key difference between supplier participation and buyer participation is that a participating buyer benefits directly from a successful project, whereas a participating supplier benefits only indirectly-through derived demand and/or the prospect of future exchange opportunities. The different incentive structures for participating buyers versus participating suppliers may affect the amount of effort the participating party puts in and, therefore, the magnitude (rather than the directionality) of the hypothesized effects. Future research might explore to what extent our findings generalize to the supplier participation case.
Finally, while we study multiplexity in a dyadic setting, the effects of triadic competition suggest that multiplexity in triads forms a promising research avenue. One triadic situation in which multiplexity may be paramount is parallel sourcing (Ho & Ganesan, 2013;Wilhelm, 2011). In a case study on Volvo's parallel sourcing of seats from Lear and JCI, Dubois and Fredriksson (2008:175-176) describe how Lear and JCI are "competitors but also (i) R&D partners, in the sense that Lear and JCI interact with each other (and Volvo) when developing […], (ii) contract manufacturers since they assemble […] on behalf of each other […], and (iii) supply chain partners, since they operate […] plants that deliver to both their respective plants […] and also share suppliers." Future research could examine multiplexity in supply chain triads.

ACKNOWLEDGMENTS
Our sincere thanks go to the executives and the project managers who participated in the roundtable discussions, the in-depth interviews, and the survey. We further thank Katrijn Gielens, Andrew Petersen, Lisa Scheer, and Raji Srinivasan for insightful comments on an earlier version. This study was supported by research grants from the Institute for the Study of Business Markets and the Netherlands Organisation for Scientific Research (NWO). The funding organizations were not involved in the execution of the research. in NPD, and its divergent findings, is provided in Web Appendix WA1. 2 As such, similar to the extant literature on buyer participation (summarized in Web Appendix WA1), we study the quantity (rather than the quality) of buyer participation. 3 Still, these perceptual errors tend to be limited when the selfreported performance measure is collected close to the event (Richard, Devinney, Yip, & Johnson, 2009), as it is in our case (see Section 4.4.1 for details). 4 Easton and Rosenzweig (2015) study the effect of team-leader and organizational experience on the success of 152 Six Sigma projects within a single consumer products manufacturer, while Linderman et al. (2006) examine the role of goals on project performance for 128 projects within a single electronic component manufacturing firm. 5 In total, the firm completed 352 NPD projects between 2001 and 2009. For 11.6% of these projects, the project managers had left the firm before the end of the primary data collection period. 78% of the project managers left because they reached retirement age, a legal requirement in the home country of the supplier. The remaining project managers left for unknown reasons, which could be related/unrelated to project success. The 41 excluded projects do not differ systematically from the 311 projects included in our sampling frame in terms of project budget and duration, and buyer and supplier perceptions of project success (all p's > .10). Hence, it is unlikely that our sampling frame is affected by selection bias in that only successful project managers and projects were included. 6 Our sample size is comparable to that of other studies on project execution. For example, Tatikonda and Rosenthal (2000) and Petersen, Handfield, and Ragatz (2005) study a sample of 120 and 134 NPD projects, respectively, while Song and Di Benedetto (2008) examine 173 radical innovation projects. For a sample size N of 140, a power level β of .80, and a significance level α of .05, we can detect effect sizes of f 2 = .28, which corresponds to medium-sized effects (Cohen, 1988). 7 In addition, we validated these perceptual measures using an objective measure that captures whether or not the projects were completed within budget (taken from the project administration records). Out of 140 projects, 111 (79.2%) projects were within budget. As expected, we find that the proportion of aboveaverage SUPSUC projects that were completed within budget (84.6%) is higher than the proportion of below-average SUPSUC projects that were completed within budget (77.2%) (Δ = 7.4, p = .05). Similarly, the proportion of above-average CUSSUC projects that were completed within budget (83.3%) is higher than the proportion of below-average CUSSUC projects that were completed within budget (76.7%) (Δ = 6.6, p = .08). Even though the within-budget measure is a rudimentary dummy variable, these findings increase the confidence in our measures. 8 To identify these competitors, we obtained permission to consult the supplier's project administration records for all projects executed within the time frame studied, not just the records for the sampled projects. 9 As we employ clustered errors in both equations, the specification of the equations is not identical. Hence, SUR is more efficient than OLS. 10 Simple slopes are calculated while holding the other variables fixed at their mean. 11 We do not estimate a similar model with the depth dimension only, because the depth dimension captures the average intensity of participation across those activities the buyer participates in. Using the depth dimension only would therefore overestimate the true amount of buyer participation. 12 For example, buyer-as-emergent-partner multiplexity is measured as the number of projects in which the contract R&D firm and the buyer collaborate as emergent partners during the focal project period. 13 Due to space constraints, we do not report these results in  (3) Photolithography This firm is the global market leader in a fast-paced industry characterized by high innovation. • Buyer participation in NPD is common in the industry.
• Buyers insist on participating in NPD: Given the fast-paced industry, buyers do not want to wait passively. • A participating buyer may be a valuable resource due to its technical knowledge. • Multiplexity is paramount: The firm works with buyers that are also competitors, suppliers, and/or partners. The firm has a sense of "we are in this together" to push the knowledge frontiers and grow the entire industry.
2 a Marketing manager, division manager, department manager, senior project manager (2) Aerospace This firm is the national market leader, active globally, in an industry characterized by a high safety and regulation focus. • Buyers tend to participate in NPD, especially when the product developed is meant to be integrated with existing systems (e.g., a wing design needs to fit the specifications of other airplane components). • Buyer participation may be a way to ensure needed input from the buyer's operational environment. • Many firms are active in multiple market domains, which is partly due to industry consolidation and globalization. One's buyers can also be one's competitors or one's suppliers. • Many firms in the industry collaborate (either by their own choice or in an orchestrated way) with each other in large innovation projects, which leads to many buyer-as-partner multiplex relationships. asked for far too stringent specifications that it did not need for its operations, the buyer's participation wasted engineering resources and hurt the project's success. • Having local buyers participate in NPD is easier due to the absence of cultural barriers.
• Buyers often outsource their NPD to a supplier with which they also compete. The supplier does not shield its knowledge from these "buyers-as-competitors," as it believes they will obtain this knowledge sooner or later anyway. However, one needs to be clever, in dealing with "buyers-as-competitors:" it is fine to share what to do, but not how to do it. • Having buyers participate in NPD when they are also suppliers may be risky, as some suppliers may also serve competitors-The knowledge the supplier and the buyer-assupplier have built up together may then benefit competitors. • Whereas the supplier may benefit from buyer participation in NPD due to buyer inputs, buyer participation may also be frustrating due to less sensible buyer ideas to which project managers need to listen. Spending time on these ideas may cause a waste of resources. • Some buyers believe that the supplier always benefits from buyer participation, which negatively influences buyers' attitudes toward participation. • Some buyers believe their participation is wasteful, as they have to spend too many resources. • Buyer participation may increase buyer satisfaction with project success, but it is not clear when. • When multiple roles exist in a relationship, these roles are not independent.
Developments in one role may affect other roles-Although it is not clear how. • Project managers tend to let buyers participate that are already partners outside the focal NPD project without questioning their motives. • Project managers are uncertain which other buyer roles, besides the partner role, are conducive to project success. • Some buyers feel entitled to participate, especially larger firms and firms that have longstanding relationships with the supplier. • When developing new products for "buyers-as-competitors," these buyers are kept at a distance due to fears of knowledge leakage. • Project managers are not always aware of the other roles a buyer may play in a multiplex relationship.  (2), department manager (2), senior project manager (2) Aerospace • Participants confirmed that buyer participation is sometimes beneficial and sometimes damaging to project success. • Participants indicated that they had always felt that buyer-as-supplier multiplexity should be taken into account when deciding on buyer participation but had not known whether this type of multiplexity was beneficial or harmful. • A senior project manager indicated that even though direct knowledge transfer from one buyer to a competing buyer is not allowed per internal guidelines, working for buyers that compete with each other is beneficial. Engineers and project (Continues) T A B L E A 2 (Continued)

Construct Operationalization Data source
Buyer-as-competitor multiplexity (MPX_COM) The share of departments (17) of the supplier that competes with the buyer firm. On average, 3.1 evaluations from senior supplier managers per buyer.

Data provided by supplier evaluations from six senior supplier managers
Buyer-as-supplier multiplexity (MPX_SUP) The number of role-reversal ties between the supplier and the buyer during the project period of project i, log-transformed.

Data provided by supplier Procurement records
Buyer-as-engineered-partner multiplexity (MPX_ENG) The number of engineered partner ties between the supplier and the buyer during the project period of project i, log-transformed.

Data provided by supplier Strategic cooperation plans
Buyer-as-emergent-partner multiplexity (MPX_EM) The number of self-initiated partner ties between the supplier and the buyer during the project period of project i, log-transformed. Technology readiness level at the start of project i, ranging from 1 to 9.

Data provided by supplier
Project administration records (Continues) T A B L E A 2 (Continued)

Construct Operationalization Data source
Project manager experience (MAN_EXP) The number of projects managed and completed by the project manager in the 10 years preceding the start of project i, log-transformed.

Data provided by supplier
Project administration records Project manager clout (MAN_CL) The average budget of the projects managed and completed by the project manager in the 10 years preceding the start of project i, in $100,000, log-transformed.

Data provided by supplier
Project administration records Buyer firm size (BUY_SIZE) The number of buyer-firm employees at the start of project i, log-transformed.
Compustat, Amadeus, Mergent Intellect, firm websites Buyer innovativeness (BUY_INN) The number of citations to buyer patents obtained within 5 years of the start of project i, logtransformed.

Derwent Innovations Index
Geographic distance (REL_GEO) The geographic distance ("as the crow flies") between the headquarters of the buyer and the supplier firm, in thousands of miles, logtransformed.

FreeMapTools.com
Cultural distance (REL_CUL) Composite index composed of the deviation along four cultural dimensions (power distance, uncertainty avoidance, masculinity/femininity, and individualism) of each buyer firm country from the country of the supplier firm: , where I dc stands for the index for the dth cultural dimension and cth country, V d is the variance of the index of the dth dimension, f indicates the supplier country, and CD c is the difference of the cth country from the supplier country.

Relationship history (REL_HIS)
The proportion of months with an ongoing project for the buyer in the 5 years preceding the start of project i.

Data provided by supplier
Project administration records Concurrent projects (REL_CONC) The number of projects other than the focal project i the supplier carries out for the buyer during the project period, log-transformed.

Data provided by supplier
Project administration records Triadic competitive ties (REL_TRIAD) The number of competitors to the buyer the supplier executed a project for during the project period of project i, based on their fourdigit NAICS codes and the supplier's project administration records. The measure also includes buyers not included in the sample.

Data provided by supplier
Project administration records, Compustat, Mergent Intellect, firm websites Demand uncertainty (IND_UNC) Similar to Patel (2011), we first create a measure of industry demand by summing the gross value added for all firms in the buyer's two-digit NAICS industry in its home country in the 10 years preceding the closing year of project i. We then regress the 10 years industry-level gross value added against year t: Y t = b 0 + b 1 * t + ε where Y t is the gross value added, t is the year, and ε is the residual. Demand uncertainty is calculated as the SE of the regression coefficient b 1 divided by the industry-average value-added over the data period.