Digital trading networks and competitive advantage in a buyer-seller network

Abstract

Purpose

Firms embedded in networks of relations are theorized through Gnyawali and Madhavan’s (2001) (G&M) structural embeddedness model to gain competitive advantage from topological characteristics. Empirical studies to support their theory have never been executed in full. Our study provided a full empirical test of their model in a digital trading network to achieve a higher degree of certainty that those network structural characteristics can have a major impact on the degree to which certain firms lead to competitiveness in a digital trading network environment.

Design/methodology/approach

To examine how firms respond in competitive situations, we chose the hyper-active digital trading network, eBay as our empirical context. We used eBay auction data to analyze how the network characteristics of eBay resellers impact their competitive behaviors.

Findings

Our study found strong support for the G&M model of competitiveness. We offer explanations for where support was not as strong as the Gynawali and Madavan theory proposes.

Research limitations/implications

Our research is limited by our chosen context and findings in support of part of G&M model. Future studies in other digital contexts are needed to enhance the modeling of network topologies and further study the impacts of network density and structural autonomy on competitive action.

Practical implications

Our study suggests that managers proceed cautiously in forming partnerships, weighing circumstances where the firm can find itself with increased information power and avoiding, to the greatest extent possible, situations where the playing field is roughly equal.

Social implications

Theory-making in this domain has begun as well as initial empirical testing. Much more needs to be accomplished, though, before embeddedness modeling can be thought of as being well established.

Originality/value

The G& M Model of competitiveness is an SNA explanation of why some competitive units succeed and others do not. Our study is the first, full blown empirical analysis of the theory.

Citation

Straub, D., Warkentin, M., Rai, A. and Ding, Y. (2024), "Digital trading networks and competitive advantage in a buyer-seller network", Journal of Electronic Business & Digital Economics, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/JEBDE-11-2023-0029

Introduction

It has been posited that networks have a significant impact on resources that firms can access and, consequently, on a firm’s ability to launch competitive actions and reactions (Gnyawali & Madhavan, 2001; Gnyawali, He, & Madhavan, 2006; Sanou, Le Roy, & Gnyawali, 2016). Firms increasingly find themselves sharing resources with other firms to complement their resources (Dyer & Singh, 1998), presumably in order to improve their individual competitive positions. Moreover, firms frequently both compete and cooperate in these networks, with some studies finding as many as a half of new alliances being formed with competitors (Harbison & Pekar, 1998; Wolff, Wältermann, & Rank, 2020). This concept, that is to say, both competing and cooperating at the same time, has been aptly labeled “co-opetition” by Brandenburger and Nalebuff (1996).

Whereas one could blithely assume that firms do not foolishly engage in such resource exchange and that these linkages must ultimately benefit the firm, we need far more research to verify this truth claim, in particular, the conditions under which it might be true and those under which it is not true. Therefore, one key and central question for research remains: Does the configuration of inter-organizational networks impact a firm’s competitiveness and if so, how and why?

The relevant literature base for discussing such issues is known variously as network perspectives on competitive dynamics, relational view of the firm, or embeddedness modeling (Gnyawali & Madhavan, 2001; Klimas, Ahmadian, Soltani, Shahbazi, & Hamidizadeh, 2023; Sanou et al., 2016; Zaheer & Geoffrey, 2005). The underlying assumption of such work is that firms are embedded in networks of firms. There are several levels of critical interaction. At the firm level, firms can either be a focal point in the network or be sparsely connected to other nodes, and everywhere in between. Moreover, firms also cooperate (or compete) with other individual firms. They form a set of relationships, and characteristics of these relationships, in comparison to the pattern of relationships formed by others, can affect competitive performance (Moran, 2005; Tang, Wang, Fan, & Liu, 2023; Wang, Ren, & Jiang, 2023). Finally, at the overall network level, networks demonstrate connectivity, wherein a highly connected network is said to be dense and its opposite sparse. These differences in network density lead to variability across networks in the access to resources and, consequently, in the competitive performance of firms (Andrevski & Ferrier, 2019; Gnyawali et al., 2006; Sanou et al., 2016; Sparrowe, Liden, Wayne, & Kraimer, 2001; Wiewel & Hunter, 1985).

Firms embedded in networks of relations are theorized to profit from topological characteristics (i.e., network structures). Advantages in competitive performance come from the relative position of the firm in the network, that is, the asymmetric resources that flow to a firm because of its distinctive position in the network. They also come from the relationship configuration, that is, having network partners that differ from those of competitors. Finally, such advantages accrue from the density of the network itself, that is, networks that are sparser carry less information than those that are denser and therefore make it easier for an individual firm to outperform competitors. Thus, the embeddedness of firms in networks affects resource flows and competitive performance in a variety of ways, depending on these structural characteristics.

The above logic of how a firm’s embeddedness in a network impacts its competitive performance is effectively formalized in Gnyawali and Madhavan’s (2001) structural embeddedness model (hereafter referred to as the G&M model). The G&M model is based on the premise that levels of the network cooperate on: (1) assets, (2) information, and (3) status, with information sharing being a primary explanatory factor for competitive success [see also Dyer and Singh (1998)]. While it pays to be a lynchpin in networks based on the position that a firm occupies, an over emphasis on firm positioning by all parties can lead to deleterious performance because the playing field becomes level. Certainly the ground-breaking work of Gnyawali and Madhavan (2001) theorizes how the position of the firm, the configuration of the relationships in a network, and the network density operate to impact competitive performance of a firm. However, empirical studies (Gnyawali et al., 2006; Sanou et al., 2016) to provide evidence for their theory have been limited forthcoming to date.

We tested the G&M model in the context of digital trading networks, where firms compete and collaborate in a virtual environment. These networks have rapidly expanded since the advent of the internet and a significant and growing proportion of commercial activity is now conducted in these environments (see Figure 1).

Examples of digital trading networks include online auction environments, such as eBay, B2B markets for small suppliers, such as Ali Baba, and electronic markets for securities trading. Business processes in these networks are virtualized, and information about competitive actions and reactions are captured in real-time and made visible to participants based on the rules that govern the exchange. As a result, digital trading networks are an important context to evaluate the impact of network characteristics on a firm’s competitive actions.

In the context of digital trading networks, sellers, buyers, and resellers exchange information though digital channels, signaling competitive actions and reactions. Economists have made convincing arguments that information advantages are important considerations in maximizing the revenues from networks like auctions (Engelbrecht-Wiggins, Milgrom, & Weber, 1983; Klemperer, 1999; Milgrom, 1981; Onur, Bruwer, & Lockshin, 2020; Saeedi, 2019). Another point of evidence that resellers in digital trading networks are using the information they glean from other bidders in their network would be that the pattern of bidding and winning auctions of Simonsohn and Ariely’s (2008) subject pool did not lead to a “happy hour effect” that occurs when there is an irrational effect like “herding.” Therefore, it is more likely that a rational effect like superior information is what is driving performance.

The specific research questions these considerations raise are: What is the impact of the different structural characteristics of digital trading networks, as captured in the G&M model, on a firm’s competitive actions? How does network density moderate the relationship between a firm’s position in a digital network and competitive action and between the configuration of relationships and competitive action? This study addresses both questions.

Although Gnyawali et al. (2006) and Sanou et al. (2016) have provided partial empirical test of the G&M model, research is still needed to investigate other parts of the model such as the effects of structural equivalence. As the first empirical test of the full G&M model in the context of digital trading networks, we gathered extensive data from eBay trading of laptop computers. In this market, many active traders are buyers and sellers, and also resell computers to other sellers or to final consumers. This market, including the complete bid histories, was documented on eBay’s website. All participants in the market could view bids by identified players, see winners and losers, and consider the prices at which computers were sold.

In the data analysis, the G&M model was examined from three standpoints. First, occupying a central and autonomous position in the network is thought to produce competitive advantages. Second, pair-level structural equivalence is predicted to lead to significantly higher level of competitive action. Third, network level densities distinguishing the separate product markets in laptops was hypothesized to lead to fewer competitive actions in dense networks, and to affect the impact of firm centrality, structural autonomy, and structural equivalence on competitive action via moderation.

Results from hierarchical regression analyses show that G&M’s model (2001) was verified in all cases except for the direct effects of structural autonomy and network density on competitive action. These results provide evidence of the enabling effect of centrality and the dampening effect of structural equivalence on competitive action. They also support the moderating role of network density on the effect of firm centrality, structural autonomy and structural equivalence on competitive action.

This paper proceeds in the usual manner. First, we examine the literature on network structure and firm capabilities. Next we posit hypotheses for testing the essential elements in the model. This is followed by a description of the methods we used to measure constructs and gather data. After this is the data analysis and a discussion of the interesting findings.

Literature review

It is widely held that firms succeed in the marketplace because they have superior resources (Barney, 1991). Early research in this vein focused on both physical and intangible resources and the discussion evolved from Pfeffer and Salancik’s resource dependency theory (2003); originally published in 1978) to the resource-based view of the firm (RBV). RBV asserts that an organization can create above-average returns and superior value by developing and deploying unique and costly-to-imitate resource bundles to exploit environmental opportunities or to neutralize threats (Barney, 1991). Resource-based theory purports that firms are made up of bundles of resources, such as input factors, assets and capabilities (Penrose, 1959; Wernerfelt, 1984). Under certain conditions, organizations can use their unique resource endowments to gain a competitive advantage (Barney, 1991; Dierick & Cool, 1989; Lippman & Rumelt, 1982; Peteraf, 1993). Interestingly, the most important ingredients of the resource endowment are not tangible (i.e., physical assets) nor are they intangible (i.e., reputation). Rather, rent generation comes from capabilities, which combine resources in ways that lead to core competencies, tend to accumulate over time, and are idiosyncratic in nature (Dierick & Cool, 1989). If a firm possesses a capability that is better than its rivals, that capability becomes its core competence.

Only resources with certain characteristics are able to create and sustain above average economic rents (Amit & Schoemaker, 1993). These resources, or strategic assets, must be simultaneously valuable, rare, imperfectly imitable and nonsubstitutable (Barney, 1991). The theory purports that differences in firm performance are the result of unique capabilities engaged as each firm interacts with its environment. Specifically, RBV asserts that ownership or control of strategic assets determine which firms earn superior profits and which firms do not.

What is important to recognize is when later work such as Dyer and Singh (1998) began considering the extension of RBV into strategic alliances with other firms, some of whom were competitors, they had to broaden the concept of inimitable resources to include relational resources. These resources are “embedded” in the relationships firms create with each other in the sense that the pattern or configuration of connections has a major impact on competitive actions and competitive performance. Dyer and Singh (1998) stress that information sharing is a key element of relationship management.

Information sharing is also central to the arguments of Gnyawali and Madhavan (2001) in their model of structural embeddedness. In fact, information derived from the network gives firms knowledge about their competitors even when there is no flow of physical assets, which is a second form of resource flow upon which they attribute meaning. We focus in this paper on digitally displayed information, in particular, the ways players in digital trading networks offer up information to each other as part of the cooperative rules of the marketplace. This sharing of information forms the network structures of the study.

Research model and hypotheses

Figure 2 presents a model of network structural variables and their effects on competitive behavior. Exactly what do these constructs mean and how do they affect competitive choices of firms?

Firm centrality

Centrality is defined as the extent to which a node has ties to other nodes, given all the feasible links that a firm could form. In the social actor network literature, nodes are points of contact between network members (either human actors or computers), which in the business literature generally corresponds to firms. The larger the number of these ties and the more that a firm stands out in this respect from competitors, the more it is thought that the firm is able to make strategic plays within the network (Wasserman & Faust, 1994; Polidoro, Ahuja., & Mitchell, 2011, Cui, Yang, & Vertinsky, 2018).

In our context, the key point is that central players lie at the intersection of information flows (Galaskiewicz, 1979) and that they occupy a strategic position by being involved in many ties. They enjoy positive resource asymmetries by virtue of their structural position and, therefore, have access to more information resources from the network than their competitors (Gulati & Garino, 2000). The network-based information advantage comes in two forms. Central players get information faster from the network than their competitors (Rogers, 1995). They also receive new information that non-central players may not get (Valente, 1995). Thus, the positive effect of centrality on competitive action derives from the volume and speed of information flows to a focal firm due to its ties with others in the network. The more information it is privy to about the other players, the more effectively it can assess the competitive motives and behaviors of others. The expanded information base also provides a key resource for it to broaden its feasible set of competitive actions (Chaudhry et al., 2022).

Gnyawali and Madhavan (2001) point out that there are downsides to being central in a network that could conceivably result in a loss of competitive edge. Sources of information can also be sinks, and information can “leak” to competitors more easily if one is well connected (Harrigan, 1985). Competitors would have to be extremely well coordinated to extract such information and, moreover, have ties that the focal firm lacks in order to fully profit from such leaks (Gnyawali & Madhavan, 2001). On the whole, therefore, centrality should be a virtue in digital trading networks and allow firms that possess it to engage in competitive action.

In their study of examining impact of co-opetition on firm competitive behavior, Gnyawali, He, and Madhavan (2006) found that centrality of a firm is positively related to the “volume of competitive actions”. In another empirical study, Sanou et al. (2016) found that firm centrality contributed to the “firm’s competitive aggressiveness” through “increased volume and variety of competitive actions” to improve a firm’s market performance.

The hypothesis that will be tested for this linkage is:

H1.

The more central is a firm in its network, the more it will engage in competitive actions.

Firm structural autonomy

Structural autonomy is best defined through Burt’s work on structural holes (1982). A structural hole is the lack of a connection. It is desirable for a strategic player to itself have connections to other members of the network (i.e., to be highly central and have no-to-few structural holes), but at the same time for its competitors to lack such ties (i.e., to have many structural holes). Burt (2004) calls actors who have a high degree of structural autonomy “brokers” and these brokers can generate more ideas than their competitors. It is clear, moreover, that they engage in more interactions, have a more diverse base of contacts, and receive information sooner than competitors. In fact, competitors who suffer from structural holes depend on them not only for resources but also for indirect contact with other firms. Knowledge is clearly power in this situation and those who enjoy access to what others are doing can exploit this for superior rents. Thus, structural autonomy creates advantages of network efficiency and effectiveness that arise from non-redundant ties. Unlike firm centrality, which is a simple, absolute concept, structural autonomy is more complex and relative in that a firm is said to be more or less structurally autonomous than its rivals. Gnyawali et al. also found that a firm’s structural autonomy is positively related to the diversity of its competitive actions in their 2006 empirical study. The hypothesis aligned with this line of reasoning and existing empirical findings is straight-forward:

H2.

The more structurally autonomous is a firm in its network, the more it will engage in competitive actions.

Structural equivalence

As articulated by Valente (1995), structural equivalence is a description of the most basic of network properties, the similarity between any two network nodes with respect to their connections to other nodes. Calculated on a pairwise basis, two firms are said to be structurally equivalent if they have similar connection profiles. The reasoning about information flows and structural equivalence is that if two firms have basically the same configurations of linkages, they have similar flows of resources and know similar things about their complementors as well as competitors in the network. As a result, they develop similar attitudes, resources, and behaviors due to similar socialization and model themselves after each other due to a similar base of resources and social influence. Therefore, because structurally equivalent pairs are resource-symmetric, Gnyawali and Madhavan (2001) argue that they will actively monitor each other’s moves and coordinate resources, with the result being that they cannot engage in competitive actions against each other. To this date, there is no empirical study which has examined the effects of structural equivalence (Gnyawali et al., 2006). Our hypothesis that derives from the comparative equality of pattern of relationships is:

H3.

The more a firm is structurally equivalent to other nodes in a network, the less it will engage in competitive actions.

Network density

Density is defined as the extent to which the network as a whole is interconnected. If every node is connected to every other node, the density is at the highest level, 1.0. If few nodes are connected to each other and approaching zero, the network is said to be “sparse.” Clearly, there is a much lower volume of information passing through such a network. Because of network density is a factorial, most real-world networks are very sparse. As soon as several dozen nodes are defined in a network, the likelihood that they would all be connected to each other falls exponentially.

The sparseness or density of connections between all pairs in the network of interest affects the appearance of unique information in the network (Granovetter, 1985). Dense networks, those in which relatively many nodes are connected with many others in a defined group, mean that the competitive landscape has been leveled because the nodes all know the same things (Granovetter, 1985). In such circumstances, less competitive action is possible because knowledge about new, desirable actions is widespread and all members can take equal advantage of this information about the marketplace. Information moves quickly through dense networks, and firms therefore know how their competitors are reacting to strategic opportunities so, in effect, no individual firm can benefit differentially by being in a privileged position of greater knowledge, (Gnyawali & Madhavan, 2001). Moreover, dense networks exhibit higher trust, shared norms, and homogeneous behaviors than spare networks, which increase the likelihood of sanctions against unacceptable behavior and of the negative consequences of reputation fallouts. Some existing empirical studies have shown that high network density can have an impact on a firm’s performance and inhibit business innovation (Manik, Indarti, & Lukito-Budi, 2023; Gilsing et al., 2008)

Since most real-world networks are sparse, there should be room for significant competitive movements in digital trading networks, which is another reason this domain is an excellent platform for testing these ideas. The hypothesis that expresses the dampening effect that a dense network has on competition is:

H4.

The denser the network, the less that firms within that network will engage in competitive actions.

Moderation of network density on firm centrality

Network density is also predicted to have a moderating effect on the relationship between the other IVs and competitive action. The reasoning underlying for each of these hypotheses is presented next.

In and of itself, central firms have access not only to greater volumes of information, but their access to information is also timely. But when a network becomes denser, the effect of centrality on competitive action should be dampened as other firms, too, now have easier and faster access to information that they can piece together information and use against the central firm. Thus, a dense network reduces the effects of centrality and central firms lose some of their advantage relative to other firms (Gnyawali & Madhavan, 2001). The hypothesis that captures this moderating effect is:

H5.

Network density moderates the relationship between firm centrality and competitive action, such that the relationship between centrality and competitive action is weaker when network density is higher than when it is lower.

Moderation of network density on structural autonomy

All things being equal, a firm with high structural autonomy is believed to be superior to its competitors in having more information about the competitive marketplace. Structural autonomy creates network efficiency and effectiveness benefits for the firm by capitalizing on the difference between the relatively large number of structural holes in competitors’ connections and the lack of holes in a firm’s own network. Because of this difference, an increase in network density actually should reinforce the advantage a structurally autonomous firm enjoys as other firms are now locked in redundant relationships. The gap between the two thus becomes even more pronounced when the network becomes denser (Coleman, 1990). Therefore, with respect to the moderating influence of network density, an increase in network density correspondingly increases the advantage of firms that are highly autonomous, according to the theorizing of Gnyawali and Madhavan (2001). The second hypothesis for moderating impacts is:

H6.

Network density moderates the relationship between firm structural autonomy and competitive action, such that the relationship between structural autonomy and competitive action is stronger when network density is higher than when it is lower.

Moderation of network density on structural equivalence

Gnyawali and Madhavan (2001) note that the structurally equivalent pairs, which are essentially interpreted to be resource-symmetric firms, avoid conflict through active monitoring and competitive coordination. It is through these mechanisms of monitoring and coordination that structural equivalence is theorized to dampen competitive actions. However, with an increase in density, the cohesiveness of the network increases, leading to more structurally equivalent pairs. It naturally becomes harder for firms to actively monitor and coordinate competitive moves across an expanded set of structurally equivalent pairs. Thus, an increase in network density weakens the negative effect of structural equivalence on competitive action.

The hypothesis for this moderation effect is:

H7.

Network density moderates the relationship between structural equivalence and competitive action, such that the negative relationship between structural equivalence and competitive action is weakened when network density is higher than when it is lower.

Table 1 summarizes the causal mechanisms of the main effects and the moderations that are hypothesized.

Research methods

To investigate the propensity of firms to respond in competitive situations, we chose to use the hyper-active digital trading network, eBay. eBay offers a number of advantages for gathering archival data about transactions and relationships between trading partners, which can be used to examine the relationships between structural variables related to networks and competitive actions. First, the offerings are separated into product classes, each of which, within a broader framework, can be considered to be a network in its own right. Since traders can be both competitors and buyers of each other’s wares, the best selection of a specific trading network to meet the conditions specified by Gnyawali and Madhavan (2001) would have a sufficient number of major players who were competitors. Another way of envisioning such a network is that sellers and buyers can be and often are resellers, who frequently buy from each other in order to secure a desirable inventory item. They then may, if they choose, break up lots, repackage, and attempt to resell at a price advantage either to competitors or to final customers. When selling, eBay resellers are unlike traditional auctioneers in that they act like small-business owners who need rapid turnaround on cash flows (Bandyopadhyay & Bandyopadhyay, 2006). They are, according to eBay, the “powersellers” who represent more than 100,000 persons in their customer base (Walker, 2005). When bidding, these resellers are equally tenacious. Bapna, Goes, Gupta, and Jin (2004) examined 3,121 serious bidders in online auctions for computer electronics to determine strategic bidding categories. Other than automated bidding, “participators” extracted the largest surplus (i.e., winning auctions). These were the bidders who, in general, monitored the entire bidding process by entering early and staying late.

Second, the transactions on eBay contain bid histories that allow interested parties to see who is bidding against them on lots, how rapidly the auction is moving to a conclusion, and who finally wins the bid and at what price. This information is shared widely among major players since they are actively engaged in multifarious auctions and they can read price intentions and bidding strategies of competitors, many of whom are, of course, also suppliers of their inventory.

In that eBay is a macro market and not all trading networks within eBay can be characterized as networks of competitors, it is necessary to select a domain for study that will allow us to test our research model. Computer trading networks on eBay meet the conditions for Gnyawali and Madhavan (2001) competitively dynamic networks. This domain has designated product classes, such as laptops, servers, and printers which vary by network density. The archive has data about who is selling to whom, and the volume of transactions and reselling behavior can be determined from the bid histories. Participants in the network get to know each other by eBay name and, we would argue, derive competitive benefits from being aware of what is transpiring on the trading floor. Thus, a judicious selection of major players who are also resellers and thus competitors can also be made.

eBay as reseller network

Gynawali and Madhavan’s structural embeddedness model (2001) specifies that the resource sharing network be made up of competitors who also engage in cooperative activity. The eBay computer trading marketplace meets this requirement. This digital marketplace has rules of engagement to which the competitors agree. Bidders must register with eBay, and their longevity on the system is part of the profile information that other bidders can see. Likewise, transaction histories of the highest bids are publicly available for all bidders to see and reflect on. Part of this is the normal structure of an auction where new bids from identified bidders are logged onto the system with the highest bid showing at the top of the list. There is a means by which the reputation of a seller and a buyer can be assessed, and these reputational profiles are likewise viewable.

This trading network has both competitive and cooperative dimensions. Resellers cooperate in belonging to the eBay network and obeying its rules and by selling to each other from their inventories. They compete in trying to buy low and sell high, as in any other trading situation.

We defined resellers as those who were engaged in the activities of both buying and selling. Moreover, to ensure that information factors would have a bearing in how resellers were responding competitively, we restricted the dataset to those who bought and sold multiples of the products.

Resellers thus are sellers or bidders who return again and again to either buy or sell on the network. Their connection to each other is virtual, that is through their observation of the bids, namely, the timing of the bids as well as the size and increments of the offers. Sometimes they are observing these bid patterns when they are selling goods. In this case, they are suppliers and are cooperating with their customers not only in the norms of the auction setting, but also in offering certain lots at certain intervals and in certain ways (such as reserve or minimum prices). Clearly, at other times, they process information differently, that is when they are bidding against each other as rivals. Here they are observing how a competitor operates and executes her/his bidding strategy. Therefore we posit that they create their network positions by virtue of which and how many auctions they participate in. The more they engage in selling or bidding, the more they become hubs of information flowing through the network. Their selling and bidding actions signal others in the network about their intentions.

There is persuasive evidence that experienced resellers such as those on eBay use their knowledge to gain competitive advantage. The experience that is brought to bear in online auctions is likely information about pricing and bidding strategies in future auctions, according to the work of Zeithammer (2006). Whereas there have been studies of student subjects that have found that experience does not seem to mitigate the winner’s curse (Foreman & Murnighan, 1996), the persistence of high bid wins in the presence of information that the winner paid too much, it is not clear that real world experience does not lead to consistently efficacious bidding strategies.

Measurement of constructs

As noted above, publicly-available eBay data shows which resellers interact with other resellers through bid histories. As with other comparable auction settings (Katok & Roth, 2004; van den Berg, van Ours, & Pradhan, 2001), eBay auctions typically do not become competitive until a short time before the auction window closes. In other words, early bids in computer auctions appear to be simply testing of the waters and are rarely serious contenders. There is an inflection point in the bid history, though, when the bidders begin to exchange serious bid information and to signal to each other their desire to outlast the competition and to win the auction.

Major or serious players are competitors who participate in auctions regularly and either win or bid seriously against those who win auctions, and then often rebundle and resell products in the same product classes. In this study, we restrict the network to such serious players.

Competitive action dependent variable (DV)

Gnyawali and Madhavan (2001) make it clear that a competitive action is a “specific, observable competitive move initiated by a firm to improve or defend its competitive position” (bolding added; page 434). Therefore, the observable data that we need are bids that eventually win the auction. Players in digital trading networks are assumed to engage in the auction to buy computers and thereby improve or defend their reselling positions. This is in accordance with authors who argue that auctions of this sort are designed to allow strategic bidders to win at a price that allows them to resell at a profit.

With respect to competitive actions, the bids that represent the likelihood of a competitive action can be defined as the opening and closing bids of the winner, and the subsequent responses by major players in the competitor network. This limits the activities for testing the theoretical research model to the serious players and their reseller competitors.

Whereas Gnyawali and Madhavan (2001) speak about the “likelihood” of competitive action as their DV, likelihood does not differ from a straight-forward assessment of action in the way in which we have operationalized it. The likelihood of an event (i.e., competitive action) is assessed in terms of the actual numbers of occurrences (i.e., winning) proportional to a certain total (in our case, participation). As such, we calculate competitive action as the total number of wins divided by the total number of bids (wins + losses) by a certain reseller (see Table 1).

Independent variables (IVs)

Firm centrality is the structural and strategic position of a node in the network, particularly with respect to its ties to other nodes (1994). The more ties, the better in that this increases the firm’s ability to understand the competitive environment. In our context, a reseller establishes a link to another reseller when it buys or sells from it. Firm centrality is measured as the total links that a reseller has to other resellers (see Table 1). Degree centrality is an appropriate measure of a node’s centrality in a network when the interest lies in the immediate influence or prominence of the node based on itsr direct connections (Wasserman and Faust, 1994), which aligns with our theorization of centrality.

Structural autonomy is delineated by Burt (1992) as having fewer structural holes with respect to other nodes, set against a background of nodes that themselves have many structural holes. In short, resellers that are structurally autonomous are relatively well connected compared to their competitors. Autonomous resellers act as bridges between two other resellers that are not directly connected. We measure structural autonomy as the number of bridges that a reseller creates divided by the total number of links (see Table 2).

Structural equivalence as defined by Gnyawali and Madhavan (2001) is “a pair-level measure of how similar [are] the actors’ network patterns” (p. 437). In our context, pairs are actors who qualify as major reselling players, and who bid against each other on more than one occasion in the same product class.

In our empirical context of eBay, and more generally in the context of sparse networks, the traditional formulations of Structural Equivalence are not suitable. Traditionally, structural equivalence has been measured as either the correlation or the Euclidean distance between any two nodes (Wasserman & Faust, 1994). Given that most nodes have different sets of connections and the commonalities are few, these measures emphasize differential connections, which is highly dependent on the degree of activity, than to the common connections. Thus, these operationalizations underestimate the role of access to similar information, which is our investigative focus, and overlap with degree of activity.

To meaningfully assess structural equivalence in our context, we considered the number of common links that any two nodes have. We view this as a more meaningful measure of structural equivalence in this research setting. We, therefore, average the structural equivalence of nodes across their connections to all others. Structural equivalence is therefore calculated as the number of connections that a node shares with others averaged across all possible pairs for that node (see Table 1). This measure aligns with theoretical prediction in our nomology that that firms that are the most similar will not engage in competitive actions against each other.

Network density is likewise well understood, and calculated as: total links in the network divided by the maximum links in the network (Wasserman & Faust, 1994). We investigated 5 product classes in computer auctions over a two month period. Given that resellers tended to specialize in one kind of computer, the classes were relatively independent. In order to include network density in the regression equations, we dummy recoded them as relatively high or low density, as shown in Table 3. Therefore, there were 2 different network densities for the dataset. The density of a network with n nodes and k total links is calculated as: k/(n * (n−1)/2) (see Table 2).

Data analysis

We gathered a data set of eBay auction records within the categories of desktop PCs from March to July of 2006. Our auction data sample includes 1930 reseller auctions from five different brands of laptop PC auction categories: Apple, HP, Dell, Sony, and Alienware.

We first traced the bidding histories of the sample auction records. The auctions in our sample had a total of 32,800 bids, or, on average, 17 bids per auction. Then, for each different bidder firm, we analyzed the bidding records, winning records, and selling records. Only firms who had both selling and bidding records were included in the reseller network analysis. These firms were also organized based on the networks in which they participated. The vast majority of firms in the sample participated in only one brand network during the data collection period. This left us with an overall sample size of 523 unique firm/network combinations. The data analysis is based on these firm/network combinations and the centrality, structural autonomy, structural equivalence and competitive action variables are calculated for each combination. Network density is based on the overall brand network and is constant across the participants in each network.

For the five brands represented in our sample, we obtained a spread of values for network density, which enabled us to reasonably classify the networks as high or low in network density (See Table 3).

The summary statistics for this sample are found in Table 4, which includes means, standard deviations, and Pearson correlations for all variables in this study. Table 2 shows structure autonomy, network centrality, and structure equivalence are all significantly correlated with competitive action. Structure autonomy is significantly correlated with network centrality and structure equivalence. Network density is only significantly correlated with structure equivalence. Variance Inflation Factors (VIFs) were well within the limits specified by Hair, Anderson, Tatham, and Black (1998). With a sample size over 500, the power of detecting insignificant results is also very high.

We used hierarchical regression analyses to test our hypotheses regarding the main effects and interaction effects (see results in Table 5). As suggested by (Cohen, Cohen, West, & Aiken, 2003), we first entered the control variables, then the main effects, and finally the interaction effects. Two cases were deleted as their standardized residuals were greater than 3. The final results are displayed in Table 5, with the unstandardized coefficients and standard errors shown.

In terms of the main effects, network centrality has a significant positive effect on competitive action, supporting H1. However, structural autonomy has a significant negative effect, which is in the opposite direction to what G&M (2001) theorized and does not support H2. Structural equivalence, however, does have a significant negative impact on competitive action. This supports H3. Finally, network density does not significantly impact competitive action, and thus does not support H4.

In terms of the interactions shown graphically in Figure 3, all three hypotheses were supported. First, network density weakens the positive effect of centrality on competitive action, supporting H5. Second, the interaction effect of network density and structural autonomy on competitive action is negative, supporting H6. Third, and finally, the interaction effect of network density and structural equivalence on competitive action is positive, which supports H7.

Discussion

The eBay computer reseller network allowed us to test the main tenets of Gynawali and Madhavan’s structural embeddedness model (2001). Our results are generally supportive, as shown in Table 6. As predicted, we found that firms that were highly central were in greater control of their environment and were able to win more auctions. We believe that these firms were able to observe the back-and-forth of the bidding of other, specific competitors and thereby to anticipate their pricing strategies.

The main effect of structural equivalence was, as anticipated, negative. The more similar the composition of firms in terms of their connections to other firms, the less firms were able to stand out from the crowd and win a higher percentage of auctions. The extreme case of this would be the same exact bidders taking part in each auction, which is, of course, highly unlikely and far from true in our data. Nevertheless, it shows that the more that pairs are similar to other pairs in terms of connections, the more homogeneous the network. In such a network, it is difficult for firms to hold a differential advantage with respect to the information resource. In short, everyone knows the same things from their connections.

Our data analysis did not find evidence supporting the prediction that firms that participated in relatively denser networks did, indeed, demonstrate fewer competitive actions. Of course, there was range restriction for our measure of density. We considered reseller networks for laptops in digital auctions, a real-world network where densities are inherently low.

There was strong support for the three posited moderation roles of network density. First, the positive effect of centrality is attenuated by network density, suggesting that dense networks make it easier for other firms to access information and reduce the resource advantage derived from centrality. Second, network density and structural autonomy negatively interact in terms of their effect on competitive action, supporting the view that the advantage gained by a firm from playing a bridging role is lower when the network is dense. Third, and finally, network density and structural equivalence negatively interact in impacting competitive action. This finding provides support for the argument that a firm in a sparse network is more negatively impeded in its competitive actions if its connection profile is similar to others in the network than a firm in a dense network.

Contributions, limitations and future research

Practitioners should be aware that they will be able to compete most effectively when they form many connections in their trading networks, but when, it is extremely important to note, their competitors do not do the same. When a network becomes too dense and firms too similar in the information they possess, it becomes difficult to trade on specialized knowledge. The greatest opportunities for competitive advantage arise from information asymmetry.

The characteristics of trading networks may not be readily calculated in that much of the data about who is trading with whom is private. Good managers sense when they have information supremacy, however, and this is likely associated with networks that have structural holes and that have unequal connectivity. When a network become denser, it may become necessary to adjust expectations downward about Ricardian rents from present networks, and to seek out other situations where the players are more heterogeneous and the firm can compete through channeling more and better information in its local network.

From a scholarly point of view, the structural embeddedness model, as articulated by Gnyawali and Madhavan (2001), has shown itself to be a powerful predictor of how organizational networks function. Why does this occur? Digital information is shared freely among all parties in many contemporary alliances, and this can reduce the usual information asymmetry among competitors. For information to be employed effectively as a competitive resource, there needs to be a disparity among competitors.

The network we examined nicely demonstrated the topological properties for which we were looking. The DV, that is, competitive actions, was calculated from the bids of active traders, those who were clearly competitors in that they resold items both to each other and to end consumers. This marketplace contained within it several mini-markets of different products, with different players operating in these various product classes. Thus, the basic qualities of networks as specified by Gnyawali and Madhavan (2001) were met by eBay reseller trading in computers. When a firm has a central position, it is easier for it to access distinctive information in a timely manner and it is more difficult for other firms to compete against it. Moreover, when a firm’s pair profile (its structural equivalence) is similar to others, then it lacks the ability to interpret signals that would differentiate its actions from those of its main competitors. The network density moderated the relationships of centrality, structural autonomy and structural equivalence with competitive action as theorized by G&M.

Two of the posited relationships were not supported, which raises interesting opportunities for theory development and future empirical research. First, we did not observe network density to have a direct dampening effect on competitive action. Is this because network density really operates as a pure moderator, as opposed to a quasi-moderator, of the relationships between the structural characteristics and competitive action? Future research should examine if network density has a direct effect on competitive action and has an interaction effect with the three structural characteristics on competitive action besides, or if it is only through the interaction effects. Future empirical research conducted in other digital contexts, and with a greater range of network densities, should inform this issue.

Second, we observed a negative effect of structural autonomy on competitive action, suggesting that there is a negative consequence of a firm playing a bridging role in the eBay auction context. This is, indeed, a very interesting finding as it raises questions about the boundary conditions for some of Gnyawali and Madhavan’s (2001) model specification. In the eBay context, the entry cost for a firm to participate in an auction and to establish links with other firms is low. Additionally, the rules for information visibility are uniformly enforced and the cost to observe the actions (e.g., bidding behaviors) of other firms is low. Do such low entry costs, standards for information visibility, and ease with which competitive actions can be monitored challenge the view that the ability to launch competitive action is promoted by bridging structural holes? Indeed, could it be that in such contexts there are penalties for firms that focus on bridging roles?

There are interesting implications for how future empirical studies should evaluate the G&M model. In our study, since product classes had a somewhat differing set of competitors, they varied in the density of the network connections, which was another requirement of the model. The online availability of open bids and bid histories allowed for signaling to major players participating in those specific auctions. This, in turn, constituted the dyadic connections that was of two kinds. Major players who bid against each other form one set of dyads. Those who sell to each other or buy from each other form the other connection.

This being said, the model can be tested more fully by looking at the in-degree and out-degree of connections (Braha & Bar-Yam, 2005). The presence or absence of a connection (dummy variables) is a standard way of modeling the structure of networks, but much more can be learned about how the strength of these ties (interval or ratio data) affects strategic behavior. Researchers should seek out datasets that will yield such information and test models accordingly.

Besides enhancements to the modeling of network topologies, future studies might probe other measures of structural characteristics to gain additional insights. For example, firm centrality can be measured with betweenness and closeness centralities, which can provide further understanding of a node’s strategic positioning and its efficiency in network connectivity. Also, new datasets should be sought out. The Internet increasing volumes of publicly-available data about firm interactions, that, properly interpreted and coded, could enable vital tests of structural models in new industry contexts. Supply chains are a fruitful area of inquiry. Suppliers and customers are well suited for tests of structural models in that they engage in both cooperative and competitive actions with one another. In some cases, it is possible to see who is buying which components from whom, moreover, and thorough coding of these interactions may be able to see if networks structures are important in other settings. In short, external validity needs to be strengthened for scientific verification.

Conclusion

Network analysis offers an interesting intellectual foundation for conceptualizing how alliances between firms has both strengths and weaknesses for further exploration. This approach suggests that managers proceed cautiously in forming partnerships, weighing circumstances where the firm can find itself with increased information power and avoiding, to the greatest extent possible, situations where the playing field is roughly equal.

Theorizing in this domain has begun as well as initial empirical testing. Much more needs to be accomplished, though, before embeddedness modeling can be seen as being well established. This can be a worthy challenge for the administrative sciences over the next few decades.

Figures

Figure 1

2002∼2019 estimated E-commerce sales of U.S. Merchant wholesalers in millions of dollars

Figure 2

A structural embeddedness model

Figure 3

Interaction results

Table 1

Underlying mechanisms for theorized effects on competitive action

Network-based properties impacting competitive action	Mechanisms for the main effects	Mechanisms for the Moderation effects
Centrality (Firm-level) Main Effect: H1 Moderation: H5	Volume and speed of information flow provide central firms with differentiated information resources, which trigger competitive action	The resource advantages from volume and speed of information flows are dampened, as there is easier access and faster flows of information to parties in dense networks compared to those in sparse networks
Structural autonomy (Firm-level) Main Effect: H2 Moderation: H6	Network efficiency and effectiveness stem from non-redundant ties, which trigger competitive action	While it may be difficult to establish structural autonomy in a dense network, once achieved, its benefits are enhanced as other firms are locked in redundant relationships
Structural equivalence (Pair-level) Main Effect: H3 Moderation: H7	Conflict avoidance with resource-symmetric peers through active monitory and coordination dampen competitive actions	With an increase in network density, it becomes harder for firms to actively monitor moves of similar peers and to coordinate competitive moves
Network Density (Network- level) Main Effect: H4	Shared information and common behavioral patterns dampen competitive actions

Source(s): Table by authors

Table 2

Formulation of constructs

Variable	Formulation	Notes
Competitive action (CA)	∑Winsi∑Bidsi	i = reference node Bids = wins + loses
Centrality (CEN)	∑Sellsi+∑Buysi	As every node is a reseller, they will have links to buyers and sellers
Structural autonomy (SA)	∑Bridgesi∑Sellsi+∑Buysi	Bridge = link to any two resellers that are not directly connected. A reseller can create a bridge between any two resellers it is connected to. Therefore, if a reseller has k links, it can create up to (k * (k-1)/2) bridges
Structural equivalence (SE)	∑j=1n−1∑k=1m−2Linki,k×Linkj,kn−1Wherei≠j≠k	n = total resellers in the network m = total nodes including non-members i = reference node j = paired node. SE is averaged across all paired nodes k = other nodes in the network Linki,k = 0 if there isn’t a link between i and k, 1 if there is one. Same for Linkj,k
Network density (DEN)	∑i=1nLinks,ixnx×(nx−1)	n = number of resellers in network x x = network i = nodes Linki,x = the total number of links that reseller i has to other resellers in the network In this equation, links are counted twice (one for each node involved), which is offset by specifying the denominator to be n*(n-1) instead of n*(n-1)/2

Source(s): Table by authors

Table 3

Network densities for resellers of five laptop brands

Brand	Number of resellers	Network density (x100)	Relative Network density
Alienware	12	3.64	H
Sony	98	2.62	H
Apple	118	1.34	H
HP	110	0.63	L
Dell	185	0.51	L

Table 4

Correlations and descriptive statistics

	Mean	Std Dev	1	2	3	4	5
Competitive Action (1)	0.528	0.384	1
Centrality (2)	0.065	0.104	0.183**	1
Structural Autonomy (3)	0.034	0.118	−0.276**	−0.150**	1
Structural Equivalence(4)	0.067	0.102	−0.13**	0.411**	0.082	1
Net Density (Binary) (5)	0.436	0.496	0.05	0.198**	−0.109*	0.105*	1

Note(s): **p < 0.01; *p < 0.05

Source(s): Table by authors

Table 5

Moderated regression results

	Main effects			Full model
	B	SE	t-value	B	SE	t-value
Constant	0.56*	0.04	14.64	0.56*	0.04	14.13
Centrality (C)	0.99*	0.17	5.77	1.43*	0.39	5.44
Structural Autonomy (SA)	−0.82*	0.14	−5.95	−0.60*	0.15	−4.09
Structural Equivalence (SE)	−1.0*	0.18	−5.59	−1.85*	0.29	6.27
Network Density (ND)	0.01	0.03	0.31	0.01	0.04	0.21
C x ND				−0.84*	0.35	−2.39
SA x ND				−1.15*	0.42	−2.75
SE x ND				1.32*	0.36	3.63
R2	0.16			0.20
ΔR2	0.16			0.04
F-change	24.57			6.49

Note(s): N = 521 (2 cases deleted as outliers)

N.B. The three control variables were first examined and all of them were insignificant. The results shown above are consistent with their inclusion. Beta values are for the unstandardized coefficients

*p < 0.01

Source(s): Table by authors

Table 6

Results of hypothesis testing

H#	Hypothesis summary	Supported?
1	The more central is a firm in its network, the more it will engage in competitive actions	☑
2	The more structurally autonomous is a firm in its network, the more it will engage in competitive actions	☐
3	The more a firm is structurally equivalent to other nodes in a network, the less it will engage in competitive actions	☑
4	The denser the network, the less that firms within that network will engage in competitive actions	☐
5	Network density moderates the relationship between firm centrality and competitive action, such that the relationship between centrality and competitive action is weaker when network density is higher than when it is lower	☑
6	Network density moderates the relationship between firm structural autonomy and competitive action, such that the relationship between structural autonomy and competitive action is stronger when network density is higher than when it is lower	☑
7	Network density moderates the relationship between structural equivalence and competitive action, such that the relationship between structural equivalence and competitive action is weakened when network density is higher than when it is lower	☑

Source(s): Table by authors