Built to Last but Falling Apart:
Cohesion, Friction and Withdrawal from Interfirm Alliances•
Henrich R. Greve
1 Ayer Rajah Avenue
Tel: +65 67995388; Fax: +65 67995399
Joel A. C. Baum
Rotman School of Management
University of Toronto
105 St. George Street
Toronto, Ontario, M5S 3E6, Canada
Faculty of Business and Commerce
Minato-ku, Tokyo 108-8345 JAPAN
Rotman School of Management
University of Toronto
105 St. George Street
Toronto, Ontario, M5S 3E6, Canada
• Research support from the Norwegian Research Council is gratefully acknowledged (Grant number 86469), as are
the helpful comments of Martin Gargiulo, Dovev Lavie, and participants at the Academy of Management Meetings.
An earlier version of this manuscript won a Distinguished Paper Award from the BPS Division in 2008.
Built to Last but Falling Apart:
Cohesion, Friction and Withdrawal from Interfirm Alliances
While models of alliance network dynamics have focused primarily on alliance formation, this
paper advances research by investigating member withdrawal from alliances. We develop a
model of cohesion and friction at the relationship, network, and market levels and propose cross-
level cohesion and time-varying effects on member withdrawal, giving predictions that are
distinct from those of alliance formation studies. Our analysis of alliances in the global liner
shipping industry showed greater withdrawal rates as a result of market competition and time-
dependent effects of prior direct and third-party ties, suggesting that member withdrawal has
both social and task-related causes. (97 words)
Networks formed by interfirm alliances have received considerable attention from strategy and
organization scholars. Research on alliance outcomes has shown that preferred positions in
interorganizational networks confer on firms such advantages as higher innovation rates (Powell,
White, Koput, & Owen-Smith, 2005), lower financing costs (Uzzi & Gillespie, 2002), and higher
performance (Burt, 1992; Baum, Calabrese, & Silverman, 2000). The relationship between
network structure and advantage has prompted a large body of research on the patterns of
alliance partner selection that propel network emergence (Gulati, 1995; Gulati & Garguilo, 1999;
Powell, Koput, & Smith-Doerr, 1996) and change (Baum, Rowley, Shipilov & Chuang, 2005;
Baum, Shipilov & Rowley, 2003). Much is now known about partner selection criteria and
collaboration buildup (Larson, 1992; Li & Rowley, 2002; Oxley, 1997; Simonin, 1997).
Although alliance breakups and member withdrawal are also common, running as high as
50 percent in some industries, much less is known about these events (Broschak, 2004: 608; Burt,
2000; Podolny & Page, 1998). Nevertheless, an understanding of them is basic to the
development of realistic and informative models of alliance behavior and network dynamics
(Kogut, 1989). Alliance breakups alter the network positions of the firms involved, their
immediate partners, as well as more socially distant firms as an initial breakup triggers others
(Nohria, 1992; Powell et al., 2005). Findings on network tie initiation do not translate directly
into implications for network change unless alliance breakups occur either infrequently or at
random. In this study, our focus is therefore on alliance member withdrawal. In contrast to
alliance initiation models, which are like fairytales that end “happily ever after,” our model of
alliance member withdrawal is more akin to a nonfiction account of “for better or worse.”
Alliance formation research frequently emphasizes resource munificence and fit, with
partner selection favoring firms that possess ample resources and well-matched interests
(Eisenhardt & Schoonhoven, 1996; Levine & White, 1961; Li & Atuahene-Gima, 2002; Pfeffer
& Salancik, 1978). An embeddedness perspective (Granovetter, 1985) in which alliances are
viewed as economic transactions that occur in a social context is also common. Previous studies
suggest that partner selection favors formation of alliances with firms that are relationally
embedded through prior direct ties and structurally embedded though network connections to
common third partners (Gulati & Garigulo, 1999; Podolny, 1994). This is not surprising given
the role of social interaction among firms and their members for development of the trust and
cohesion necessary for mutually beneficial relationships (Larson, 1992; Li & Rowley, 1999;
Mohr & Spekman, 1994; Ring & Van de Ven, 1992; Seabright, Levinthal, & Fichman, 1992).
It is possible that member withdrawal is shaped, oppositely, by these same factors, but
this has not yet been tested. Whereas the munificence of complementary resources nurture
alliance formation, resource scarcity and misalignment of alliance partners’ resources and
interests create friction that may trigger member withdrawal (Baker, Faulkner & Fisher, 1998;
Koka, Madhavan & Prescott, 2006; Rowley, et al., 2005). Similarly, member withdrawal may
occur when embeddedness is low. Findings on alliance formation do not translate so neatly into
implications for member withdrawal, however. In particular, past work has not examined
whether alliances vary in the level of friction among partners, and thus appears to have settled on
a model where alliance breakup and withdrawal result from low embeddedness rather than high
friction (Das & Teng, 1998; 2000). Friction among partners also affects alliances, however, and
there is already work showing that firms are likely to withdraw from relations with resource
misalignment at the relationship level and power inequality at the network level (Rowley, et al.,
2005). Therefore, the purpose of this study is to examine whether and how factors that initiate
alliances also reduce withdrawals from alliances by theorizing member withdrawal as a joint
function of embeddedness and friction.
In addition to examining whether and how factors affecting alliance formation also affect
member withdrawal, we extend existing theory in three ways. First, we extend the embeddedness
and resource perspectives, typically conceived at the relationship and network levels, to the
market level. To the embeddedness perspective, we add mutual forbearance theory, which
indicates that firms with contacts across multiple markets gain familiarity with each other and
have incentives to avoid provocative actions (Karnani & Wernerfelt, 1985). To the resource
perspective, we add market overlap as a source of friction because cooperation is more difficult
in the face of resource scarcity and competition (Baker, Faulkner & Fisher, 1998; Koka,
Madhavan & Prescott, 2006).
Second, in the common model of alliance formation, relational embeddedness in prior
direct ties and structural embeddedness through ties to common third parties are thought to
increase trust or control opportunistic behavior (e.g., Gulati, 1995; Gulati & Gargiulo, 1999;
Kale, Singh & Perlmutter, 2000; Kogut, 1989; Podolny, 1994; Ring & Van de Ven, 1992). It is
assumed that these effects are additive across levels when the outcome of interest is alliance
formation because each additional source of control adds to the confidence in the potential match.
However, it is not clear whether the effects are additive across levels once firms have initiated
alliances. If alliance stability does not call for embeddedness beyond a certain threshold, it is
likely that when one of the factors generates sufficient embeddedness to hold the partnering
firms together, the effects generated by embeddedness at the other levels become redundant.
Hence we examine substitution between embeddedness at relational and network, as well as
Third, revelation of new information during collaboration has earlier been used to suggest
that cohesion will grow over time within alliances (Dwyer, Schurr & Oh, 1987; Larson, 1992;
Levinthal & Fichman, 1988; Seabright et al., 1992). We make a related, but slightly different,
argument for collaboration duration. Alliances gain initial cohesion through relational and
structural embeddedness. After initiating an alliance, firms learn about their partners’ behavior
through the collaboration, and this within-alliance learning is so important for continuation
decisions that variables predicting the initial cohesion lose their explanatory power. Hence, the
alliance cohesion attributable to relational- and network-level factors will weaken over time.
Such changes over time are crucial. Network dynamics would be very different if alliances
initiated in the context of prior ties became more stable over time or, alternatively, if they became
less stable over time.
In sum, our model predicts that the risk of member withdrawal decreases as relation-,
network-, and market-level embeddedness foster cohesion among members, and increases as
resource scarcity and incompatibility at these three levels generate friction among members. We
analyze member withdrawals from alliances in the global liner shipping industry. Shipping
alliances are made to deliver a scheduled transportation service – a route connecting a set of
destinations – using ships pooled from, and orders accepted by, alliance member firms. They are
open-ended collaborations whose termination is costly and unplanned (Makino, Chan, Isobe, &
Beamish, 2007), and are thus more meaningful for investigating drivers of member withdrawal
than temporary collaborations such as R&D alliances and underwriting syndicates.
Requirements of regular and predictable liner shipping services mean that routes entail
commitments of substantial resources that must be reassigned if an alliance breaks up or
members withdraw. Despite these costs, there is considerable change in shipping alliances as
member firms withdraw individually or in groups, and routes are ended.
COHESION, FRICTION AND MEMBER WITHDRAWAL
Alliance formation is a joint action by multiple firms, but withdrawal can be done unilaterally by
a member firm. Because an alliance is simultaneously a relation between two or more firms, an
element in an interfirm network of alliance ties, and a commercial activity with effects on market
competition, this decision involves considerations of relation-, network-, and market-level
factors. The firms are not only goal directed actors with managers who consider the balance of
inducements and contributions when deciding on continuation of an alliance (e.g., March &
Simon, 1958), but also subject to influences of social structures at each of these levels, resulting
in greater persistence of the relations than of market exchanges (Granovetter, 1985; Baker, 1990;
Uzzi, 1996). Models of member withdrawal and alliance termination thus need to capture both
cohesion as a result of social relations and friction as a result of instrumental concerns of task
execution and goal conflicts as firms balance the two.
The tension between forces of friction and cohesion can be conceptualized as a force-field
model (Lewin, 1947) in which withdrawals occur when frictions become powerful enough to
overcome cohesive forces built into the relationship. Cohesive forces thus establish a threshold
for how much friction alliance members are willing to tolerate before breaking up, so a given
level of friction may lead to a breakup for an alliance with low cohesion, but not for one with
high cohesion. Empirical work has already demonstrated the utility of modeling friction and
cohesion together. Seabright et al. (1992: 126) and Levinthal and Fichman (1988) considered the
effects of both binding and separating forces in predicting terminations of auditor-client
relationships. Bresser (1988) and Oliver (1990) argued that trade associations, as a form of
interorganizational collaboration, operate on a similar principle: their existence requires not only
shared political interests as inducements for collective action but also the absence of (or low
level of) conflict and friction caused by competition in markets. There does not seem to be prior
work that examines the tension between cohesion and friction at multiple levels, however.
Combining the force field idea with the three levels of factors, we arrive at the theoretical model
in Table 1. Below we explain each prediction of the model to understand individual firms’
decisions to withdraw from alliances.
=====TABLE 1 ABOUT HERE=====
Cohesion: Prior Ties. A key finding in alliance research is that firms tend to initiate additional
alliances with their current alliance partners (Gulati, 1995; Gulati & Gargiulo, 1999). The
leading explanations of this finding are trust and incentives. Trust is viewed as vital to alliance
initiation because firms that trust one another will be better able to solve problems that occur
during the alliance (Parkhe, 1993). A structural source of trust is prior ties among alliance
members. Prior ties give alliance members experience with each other, possibly including
experience with how they react to problems and conflicts (Gulati, 1995), so they facilitate
development of trust through learning and accommodation (Larson, 1992; Kale, et al. 2000).
Prior ties also create a structural incentive to avoid non-cooperative behaviors, because partners
might retaliate with noncooperative behaviors in other alliances. Kogut (1989) and Park and
Russo (1996), for example, found that the presence of concurrent ties among partners reduced
the likelihood of international joint venture failures by facilitating reciprocity.
Thus, research suggests that relational embeddedness creates cohesion. Although repetition
of alliances is consistent with trust and incentives, the same behavior would also occur if firms
merely found it easier to discover opportunities to form alliances through local search among
their past alliance partners (Li & Rowley, 2002). A full test of the trust hypothesis thus requires
study of member withdrawal as well, as high trust prior to joining an alliance will reduce the
likelihood of leaving the alliance. Local search would also produce such stability if it remained
local after the tie initiation, but not if new opportunities became known to the firm. If a manager
who initiated an alliance as a result of local search later learns that a better partner had been
overlooked, withdrawal from the alliance may well follow. Thus, the effect of prior ties on
alliance cohesion should be to reduce member withdrawal if prior alliance ties promote trust or
create incentives against leaving for each alliance member (Olk & Young, 1997).
Hypothesis 1: Member firms are less likely to withdraw from alliances with more prior ties
Friction: Resource Incompatibility. Firms use alliances instrumentally to achieve strategic
goals. The primary goal of production-oriented alliances is to pool resources from multiple firms
and increase economies of scale beyond the capacity of single firms (Aiken & Hage, 1968;
Williamson, 1975). Such alliances help firms overcome resource scarcity and expand the range
of products and services offered to customers with their currently available limited resources.
Failure to accomplish these goals will create friction in the alliance. Accordingly, a basic source
of relationship-level friction in production-oriented alliances is resource incompatibility, which
occurs when the resources that alliance members possess are insufficiently interchangeable with
those held by other alliance members to produce uniform output quality and quantity. Machine
tools with different tolerances and R&D departments with different design tools are examples of
potentially incompatible resources. For example, the loss of the Mars Climate Orbiter was
attributed to a failure to convert between English and metric measurement units during design
collaboration. Similarly, the estimated $5 billion loss accrued developing the new A380 Airbus
was attributed to the incompatibility of fuselage and wiring suppliers’ computer systems.
Resource incompatibility has two consequences. First, it causes inconsistent quality of
services and products, decreasing reliability from the viewpoint of customers who prefer
products and services of stable quality. Second, it implies that some partners may offer more or
better resources than others. The inequality of resources allocated to alliances becomes
problematic when the outcomes of alliances are not distributed according to the contributions.
When production-oriented alliances entail many environmental contingencies, accurate
assessment of contributions and fair distribution of outcomes is difficult and costly because firms
need to renegotiate terms of contracts to decide outcome distributions (Oxley, 1997; Williamson,
1975), and the costs increase when the resources are heterogeneous. Hence, resource
incompatibility not only creates dissatisfied customers but also friction among alliance members.
Although the concept of resource incompatibility is general, its sources will differ across
industries. In our research setting, the liner shipping industry, ships are crucial resources and we
can assess resource incompatibility through the variation in ship size, speed, and age. In liner
shipping, routes are scheduled in advance and need to be regular, reliable, and frequent (Ryoo &
Thanopoulou, 1999). Pooling of ships for joint service results in a shipping route where (say) a
weekly scheduled departure uses ships from various firms. If the ships are of unequal size,
customers will plan regular shipments with respect to the smallest available ship, and firms
operating larger ships will be forced to try to fill them with incidental traffic. This is likely to
result in under-utilization of the larger ships, creating a potential source of friction.
Similarly, ships in a route move in lock step, so the slowest ship determines the maximum
speed. Fast ships are more expensive to operate than slow ones even when going at low speed
because of the weight and size of the propulsion machinery, so speed differences may also cause
friction. Ships of different ages have different reliability and maintenance schedules,
complicating provision of backup ships for scheduled or unscheduled maintenance. Moreover,
ships of different age are equipped with different energy saving technologies, making it difficult
for alliance members to equitably share operating costs. It follows that alliance members will
experience greater friction when the resources that they pool are incompatible, resulting in a
higher risk of withdrawal because alliance partners will be interested in finding alternative
partners with resources that are more compatible than those of their current partners. Conversely,
when the resources that alliance members pool are compatible, the alliance will operate smoothly,
and there is a lower likelihood of finding alternative partners with more compatible resources.
Hypothesis 2: Member firms are more likely to withdraw from alliances in which they have
more resources that are more incompatible with those of the other members.
Cohesion: Triadic Closure. Two unconnected firms are more likely to form a tie if both have
ties with a common third firm. Firms’ decision makers establish relationships after considering
available information about potential partners’ capabilities, reliability and motives (Oxley, 1997;
Pfeffer & Salancik, 1978; Williamson, 1975). While exogenous factors such as proximity may
play a role in partner selection, decision makers demonstrate a preference for extending existing
relationships and forming new relationships with their partners’ partners based on referrals
(Baker, 1990; Gulati, 1995; Gulati & Gargiulo, 1999; Podolny, 1994; Uzzi, 1996). The result is a
prevalence of closed triads characterized by collaborative ties in all three member pairs.
While studies typically focus on the effects of common third-party ties on the initiation of
relations, Krackhardt (1998, 1999) redirected the focus to effects on the stability of the resulting
triads. When the interests of two partners in a triad are in conflict, the third can serve as mediator.
Conversely, in a triad, when a firm acts badly toward one of its partners it risks not only harming
its relationship with that partner, but also the other partner. Impeachment for bad behavior thus
doubles in triads, giving disincentive to act opportunistically, and encouraging firms to be
concerned with their reputations (Walker, Kogut, & Shan, 1997; Rowley, Behrens, & Krackhardt,
2000). Triads can thus enforce shared norms of behavior and serve as social constraints that lead
members to act in accordance with each other’s expectations and to expect the same from other
members. For these reasons, Krackhardt (1998, 1999) predicted stickiness of relationships
structurally embedded within triadic ties. In support of this conclusion, Burt (2000, 2002)
showed that, absent closure, triadic relationships connecting bankers otherwise disconnected
within their network tend to more quickly decay and dissolve.
Triads are developed in two ways: closure of three dyadic alliances or a single alliance with
three or more firms. We expect reputation to generate cohesion in either case, but the mediator
principle to operate only in the latter case. In other words, even when a triad is composed of
three dyadic ties, rather than an alliance of three firms, negative reputation can quickly
disseminate and constrain opportunistic behavior. It follows from these arguments that:
Hypothesis 3: Member firms are less likely to withdraw from alliances in which they have
more third party ties with the other members.
Friction: Position Power Inequality. When network ties are sources of value, occupying a
better network position than other alliance members provides a power base from which a firm
can press for a high share of the value creation. Position power in networks derives from control
of information and resources – it is the ability to cut partners off that makes the threat of
withdrawal effective (Burt, 1992). Another source of network position power is the ability to
withdraw without being sanctioned. In a conflict, a weaker firm may use its ties to spread
information about tough negotiating tactics (Coleman, 1988), harming the reputation of a more
powerful firm. But, such disciplining tactics are ineffective if the tough firm’s network position
is sparsely structured so that each of its contacts has difficulty reaching the others (Burt, 1992).
Whether partners stay in a relationship hinges on the value they create working together
and the distribution of the created value between them. The distribution of value is a result of
negotiation among the members, and each member’s contribution to the partnership is a resource
that can be used to demand a high share of the created value (Bleeke & Ernst, 1991; Pfeffer &
Salancik, 1978). By making the problem of allocating value among members more complicated,
power inequality creates friction that makes tie dissolution more likely. Imbalance of power
among allying partners is thus a key factor in predicting network-level friction (Provan &
Skinner, 1989; Steensma & Lyles, 2000).
Managers of more powerful firms can exercise their power to obtain preferable outcomes in
reconsidering terms of alliances (Bae & Gargiulo, 2004), whereas managers of less powerful
firms try to avoid being exploited by disconnecting ties with powerful partners. Power imbalance
is particularly likely to create friction when contracts are reviewed as a result of unexpected
contingencies, and some firms can claim to have contributed more to their solution than others.
Powerful alliance members can extract greater benefits by threatening to withdraw from the
alliance and operate alone or with a new partner, or to withhold valuable resources in order to
reserve them for their private purposes. These potential threats give low-power partners an
incentive to discontinue relations with powerful firms. There may be other benefits from
associating with a powerful firm, such as status transfer, but even status transfer is costly because
a high-status firm will extract resources from a low-status firm in exchange for its association
with it (Benjamin and Podolny, 1999; Castellucci and Ertug, forthcoming). Thus, the cost of
associating with a powerful firm is not likely to be offset by other advantages.
In liner shipping alliances, the exercise of power is most visible when members renegotiate
ports of call and route schedules as a result of changes in customers’ demands and preferences.
Powerful firms attempt to adjust the terms of collaboration to favor their own customers, so less
powerful firms pursuing alliance stability may endure a situation in which the preferences of
their customers are more difficult to satisfy. Power imbalance thus is a source of network-level
conflict, increasing the risk of withdrawal. We therefore predict:
Hypothesis 4: Member firms are more likely to withdraw from alliances in which they have
a weaker power position than the other members.
Cohesion: Multimarket Contact. Multimarket contact refers to a “situation where firms
compete against each other simultaneously in several markets” (Karnani & Wernerfelt, 1985: 87).
Multimarket theory predicts that the more contacts a firm has in other markets with its rivals in a
given market, the less competitive pressure it receives from the rivals in that market (Bernheim
& Whinston, 1990; Jayachandran, Gimeno, & Varadarajan, 1999). Firms that meet each other in
multiple markets have frequent competitive interactions that permit them to learn to take actions
that weaken competition between them (Baum & Korn, 1996, 1999; Boeker, Goodstein, Stephan
& Murmann, 1997; Greve, 2000; Haveman & Nonnemaker, 2000). For example, by establishing
footholds in geographic or product markets valuable to a competitor, a firm can signal its desire
to compete softly, but, if provoked, to retaliate broadly across many markets or selectively in
markets where it can harm a rival’s interests at a low cost to itself (Bernheim & Whinston, 1990).
Multimarket contact induces mutual forbearance, a situation where interdependent firms
mutually recognize the potential harm that can come from a failure to cooperate.
A market with a high degree of multimarket contact is thus less competitive and more
profitable, which increases its attractiveness. Markets with a high degree of multimarket contact
also tend to be more stable because participants in such markets avoid actions that may provoke
retaliation. This stability is particularly beneficial for firms participating in the market through
alliances because the multifirm governance structure of an alliance encumbers responses to
others’ competitive moves. The members of an alliance need to agree on a response, adding one
step to the decision making, and because each member may have different market relationships
with the firm that made the original competitive move they may prefer different responses.
In our empirical context, the market is divided into several trade routes (e.g., transatlantic
or transpacific) and regional markets (e.g., East Asia), and there is variance in the level of
multimarket contacts across markets. We expect a high level of multimarket contact in markets
served by an alliance to reduce the likelihood of its members withdrawing from the alliance.
Hypothesis 5: Member firms are less likely to withdraw from alliances operating in markets
with higher multimarket contact.
Friction: Market Overlap. Whereas mutual forbearance is a source of market stability, market
overlap is a source of instability (Baum & Korn, 1996). Market overlap increases with the
number of rivals that a firm faces in each of its operating markets. A high degree of market
overlap not only lowers firms’ performance, but also increases the probability that they will make
market position and other changes in response to the poor performance, or the competition itself
(Baum & Singh, 1996; Dobrev & Kim, 2006; Greve, 1998). For firms participating in multiple
markets, each market is a source of competitive pressure, and a failure to compete successfully in
any market may lead to adjustments in the focal market (e.g., withdrawal to avoid competition)
or in other markets (e.g., withdrawal to focus resources in the failing market). Liner shipping
firms experience greater competition when serving markets populated by many rivals, creating
incentives to alleviate the pressure with strategic changes in its market participation through
market entry and exit, change of ports of calls and schedules, increase or decrease of service
frequency, and withdrawal from an alliance. An alliance member’s degree of market overlap
should thus be a predictor of its likelihood of withdrawal:
Hypothesis 6: Member firms operating in markets with higher market overlap are more
likely to withdraw from alliances.
During alliance formation, the cooperativeness of prospective partners is evaluated based on the
trust developed through prior interaction or the monitoring and incentives provided through
preexisting relations, shared third parties, and multimarket contact. In choosing among potential
alliance partners, more direct sources of trust or incentives are preferred, so the parsimonious
assumption that these factors have additive effects is reasonable (Podolny, 1994). For the
decision to continue or withdraw from an alliance, however, information obtained during the
collaboration must be weighted against these sources of initial cohesion. On the one hand, a
manager who considers withdrawal is likely to be more reluctant when there are multiple sources
of cohesion, such as a history of collaboration in other alliances and ties to common third parties.
Thus we would expect each additional source of cohesion to decrease the rates of member
withdrawal regardless of the level of the others.
On the other hand, it is also plausible that once firms initiate resource exchange and pooling
in alliances, stability does not require cohesion beyond a threshold. This is particularly true for
production-oriented or resource pooling alliances such as ones that we investigate in this study
because there is less potential for potential partners’ ex-ante intentional misrepresentation of
resource possession and ex-post ambiguities in monitoring partners’ resource inputs in alliances.
Even when firms find that their partners are suboptimal for achieving alliance goals, they are
unwilling to take on the cost and risk of searching for new partners, given that they too may end
up being imperfect. Routines that firms develop for resource pooling and exchange add further
structural stability. Member withdrawal is less likely if cohesion generated by any factor at the
three levels satisfies minimum requirements for holding firms together in alliances. If this is the
case, then the role that network and market cohesion plays in stabilizing alliances becomes
marginalized with relational cohesion.
Podolny (1994) argued that when assessing and judging whether other firms are appropriate
for alliance partners, firms’ decision makers resort to more tangible information sources such as
prior ties than to intangible ones such as high status when both are present. In similar vein, we
predict that the presence of direct, visible, and tangible foundations of cohesion (i.e., prior
alliance ties) weakens cohesion generated by indirect, less visible, and intangible foundations
(i.e., triads and multimarket contact) to maintain ongoing relationships between allying firms. In
turn, a lack of prior ties increases the strength of effects that network and market cohesion brings
about. It follows from these arguments that:
Hypothesis 7: The effects of network and market cohesion on member firms’ withdrawal
decrease with relational cohesion.
Dynamics of Cohesion
Because our focus is on member withdrawal rather than formation, we are concerned with
cohesion not as an antecedent to alliances, but as a process unfolding within them. We argue that
the cohesion of alliances grows over time, making firms less likely to withdraw. This increase
occurs as a result of trust buildup and commitment processes that operate within each alliance as
the firms accumulate experience with the responses of other firms. Firms are reluctant to
immediately commit to open exchange with new partners, preferring instead to gradually
escalate trust-giving as a result of observing trustworthy behaviors by the partner (Levinthal &
Fichman, 1988; Larson, 1992). Time allows partners to learn about each other as boundary
spanners invest in personal relationships that serve as conduits for information and uncertainty
reduction, and promote alliance stability (Geletkanycz & Hambrick, 1997; Gulati & Westphal,
1999; Luo, 2001; Seabright et al., 1992). Familiarity and mutual understanding in turn facilitate
development of cooperative norms of exchange based on the expectation of future interaction
(Axelrod, 1984; Larson, 1992). Time also aids development of routines and other relationship-
specific investments that facilitate the joint planning, information exchange, and conflict
resolution required for successful coordinated action in multiparty alliances (Inkpen & Dinur,
1998; Levinthal & Fichman, 1988; Simonin, 1997; Zollo, Reuer & Singh, 2002).
The increased cohesion obtained through the process of collaboration is distinct from the
cohesion among firms that have prior ties or common third-party ties. Each alliance faces unique
challenges that could cause a breakdown of collaboration. Prior ties or third-party ties help firms
overcome general distrust that could prevent well-functioning collaborations in any context, but
do not necessarily translate into an expectation that the partners will be able to collaborate well
in the focal alliance. The process of collaboration thus increases cohesion even among alliance
partners who have relational or network cohesion at the start of the alliance. However, the
process effects on cohesion are greater in alliances in which the firms initially start at a lower
level of cohesion because each instance of trustworthy behavior is more informative when the
level of trust is low. Consequently, while relation and network sources of cohesion initially have
a strong effect on member withdrawal, these effects attenuate as process sources of cohesion
allow alliances with low initial cohesion to catch up. The relation and network factors are
gradually supplanted by alliance-specific trust. This leads to:
Hypothesis 8: The effects of relational and network cohesion on member firms’
withdrawal decrease with alliance duration.
Liner shipping means operation of regularly scheduled routes among commercial ports, as
opposed to the specially ordered journeys in bulk shipping. Most routes use ships that are made
to exclusively carry containers, but some have specialized ships such as car carriers or
refrigerated cargo carriers. Container ships are very expensive, in large part because they become
more efficient the larger they are. The need for efficiency and increased world trade has made
container ships that can transport 8000 twenty-foot equivalent unit (TEUs) containers and cost
100 million USD common on the main routes, and even larger ships are in use.
In spite of the cost, a single ship is not useful. Customers value frequent departures,
preferably weekly ones, which leads to a requirement of at least five ships for a route across the
Pacific and eight for a route between Asia and Europe. For instance, Wan Hai, a Taiwanese firm,
operated a weekly trans-Pacific route in 2003, using five vessels to serve the following rotation:
Xiamen, Yantian, Hong Kong, Kaosiung, Los Angeles, Oakland, and Xiamen. Capacity
requirements increase with the number of ports served by a route, but to be commercially viable,
routes need multiple port calls in each end. The risk and financial burden of such a resource
commitment may be too much for a single company, and it may not have a sufficient customer
base to support a route of the desired regularity. Alliances are formed to provide sufficient
operating resources and to feed the route with traffic. For example, in 2003 NYK, OOCL, and
P&D Nedloyd cooperated on a weekly South China Sea route, providing two vessels each to
serve the following rotation: Singapore, Laem Chabang, Hong Kong, Kaohsiung, Los Angeles,
Oakland, Kaohsiung, and Singapore. As in the airline industry, constellations such as the ‘Grand
Alliance’ formed by Hapag-Lloyd, MISC, NYK, and OOCL have emerged. Unlike the airline
industry, however, these constellations have been unstable and have not restricted members’
ability to partner with nonmembers (Ryoo & Thanopoulou, 1999). Hence a single route is the
meaningful unit for analyzing shipping alliances.
To understand modern shipping alliances, we conducted semi-structured interviews with
managers of two Japanese and two Norwegian shipping firms and a manager of a Japanese
container terminal operator. We learned from the interviews that there are two primary forms of
interfirm collaboration in this industry: slot purchases or slot swaps and vessel sharing
agreements. The former means that the operator purchases or trades a certain amount of freight
capacity on a route served by another operator. The route is still operated by the ships of the main
operator, who maintains full authority over their use, so they do not involve asset pooling. The
purchasing operators do not have any authority to change service frequencies of routes, ports of
call, and shipping schedules, but can negotiate price and transportation volumes with the
operating operators. This type of interfirm collaboration is an arm’s length tie (North, 1990)
rather than an alliance.
Vessel sharing agreements, on the other hand, are joint route operations in which the
operators pool vessels in a route in order to increase capacity and service frequency, and have
shared authority over the vessel use. This is an embedded relationship (Uzzi, 1996: 677),
involving fine-grained information transfer and joint problem-solving. Such alliances give the
route an economical scale of operations with less investment from each alliance member, and lets
alliance members pool orders from their different customer bases. Moreover, alliance members
share their access to local port facilities and services, negotiate jointly with third-party service
providers, and jointly adapt to fluctuations in customer demand. Alliances in this context are
thus more than simple contractual arrangements. These benefits are gained at the expense of
more complex decision rules as a result of the shared authority. Because only vessel sharing
agreements involve joint decision making and resource pooling, we study member withdrawal
from vessel sharing agreements (i.e., alliance ties) only, not slot purchases or swapping (i.e.,
arm’s length ties). Alliances are made when arm’s length ties are thought to be less beneficial,
and an arm’s length tie is a possible option for a firm that is currently a member of an alliance.
The cost of member withdrawal is high because it may cause reconsideration of shipping
schedule and calling ports, adjustment of transportation capacities by reassigning vessels from
other routes or leasing additional vessels, or search for new alliance partners. These actions have
substantial effects on firms’ operations, cost structures, and reliability of service. While we focus
on member withdrawal from alliance ties only, we do consider the effect of arm’s length ties as
The alliance network changes annually through formation of new alliances and breakup or
withdrawal from old ones, but examining a graph of a single year is still useful for assessing the
overall structure. Figure 1 shows the 2003 network, with circles indicating firms and lines
indicating alliance relations. Both symbols are scaled by their degree, so firms with many routes
are larger, and lines connecting firms with more alliances are thicker. It has some clustering,
which could be the result of trust buildup and meshing of organizational routines making
repeated collaborations attractive. The network also shows a remarkable variation in the
alliances of the individual firms, suggesting that the needs of specific routes are also influential.
Structural holes are also seen. For example, although the Japanese firms MOL and NYK work
together on some routes and have shared collaboration partners (such as P&O), each also has
exclusive relations such as the NYK-Hapag collaboration and the MOL-Hyundai collaboration.
=====FIGURE 1 ABOUT HERE=====
The sample consisted of 666 alliances involving 171 shipping line operators originating from 46
nations between 1991 and 2004. The data extend from 1988 through 2005, but the 1988-90 data
are only used to create independent variables and the 2005 data are only used to detect
withdrawals from the 2004 alliances, so the analysis contains 14 years of observations. All
alliances existing in the focal years are in the data. The source of alliance data is the
International Transportation Handbook published annually by Ocean Commerce Ltd., a
Japanese publisher specializing in the liner shipping industry. It includes all line operators having
cross-national routes connected to at least one port in Japan and those partnering with them. The
data include operators that do not have routes connected to Japan, but ally with operators having
routes connected to Japan. Hence the data is a snowball sample design with all operators serving
a Japanese port as the seed and all their contacts as the snowball. The data are highly reliable, as
only information original from line operators is compiled, and the data source has a
comprehensive list of routes operated by line operators. The world’s leading line operators have
routes connected to Japan, which is one of the largest economies in the world and a major
exporter of industrial goods. Approximately 8 % of the world-wide container traffic originated in
or was bound for Japan in 1996 (Containerisation International Yearbook, 1998), but routes
connected to Japan usually visit the other main Asian manufacturing locales as well. We used
Containerisation International Yearbook to verify the route information and Lloyd’s Registry
Fairplay to obtain data about ships.
Our observations begin in 1988 following a major shift in industry alliance practices that
occurred during the mid-1980s. Prior to the observation period, collaborations were arranged
through the ‘conference system,’ which operated as a form of cartel, collectively fixing tariffs
and controlling volumes of trades. The U.S. government, which is normally hostile to cartels,
had tolerated the system to foster the development of strong trade and commerce systems. In
1984, however, the U.S. Merchant Shipping Act was passed, which excluded shipping
conferences from anti-trust regulations but placed strict limitations on price fixing. Together with
the containerization revolution and globalization, the Act weakened the ability of the conference
system to enforce price agreements. In parallel with this U.S. change, the European Commision
also increased its scrutiny of freight rates and price agreements.
We constructed our network of preexisting alliances by coding all routes operated jointly
by multiple operators. We regard joint operation of a route as a network tie between the operators.
Thus, the original list of operators of each route becomes an affiliation network (two-mode
network) where one or multiple operators are affiliated with a route. This is transformed into a
regular one-mode network by letting operators have a network tie with strength equal to the
number of routes that they jointly operate. The one-mode network is used to calculate the
network measures with Ucinet 6 (Borgatti, Everett, & Freeman, 2002).
Modeling Framework and Dependent Variable
Our dependent variable is member withdrawal from the alliance, coded as 1 when a given
member of the alliance operating the route at time t is no longer part of the alliance operating the
route in t+1 and as 0 otherwise. Our unit of analysis is thus the firm-alliance-year. We adopt this
operationalization with its requirement that the route still operates next year in order to
distinguish member withdrawal from route failure (i.e., alliance collapses). Theoretically the
determinants of route failure should be different from those of member withdrawal, with high
competition or low demand being prominent cases, so this distinction is important in order to
isolate the mechanisms that lead to withdrawal. However, resource-poor and high-competition
routes may also be more subject to withdrawals, suggesting that a link between route failure risk
and member withdrawal should be captured as well. Thus, we apply a selectivity model
(Heckman, 1979) with route failure as the first step and member withdrawal as the second. The
selectivity model calculates the inverse Mill’s ratio with Lee’s (1983) correction. In order to
obtain the best possible estimates of the selectivity model, it was estimated on the failure of all
routes, including routes with one operator. The estimates show that new routes and routes with
high competition are more likely to fail (estimates are available from the authors). To verify that
alliance routes had failure patterns just like those of all routes pooled, a model of alliance route
failure was estimated on only the alliance routes with the inverse Mill’s ratio entered. The
inverse Mill’s ratio was highly significant, and only one of the 16 other variables in the model
reached five percent significance, suggesting that the selectivity control based on the pooled
model fully accounted for the failure of alliance routes.
Withdrawals from alliances can occur in continuous time, but the annual measures of
alliance participation lead to a discrete time measurement of whether a member leaves a given
alliance in a given year. To model this process, we apply a conditional log-log specification,
which is the equivalent of an exponential hazard rate model (Allison, 1982). To control for all
annually varying influences that are shared across the observations, we enter indicator variables
for each year. Having thus controlled for time, we are left to model only the effect of covariates
specific to the route, the alliance, and the member firm (including its relation to the others in the
alliance). To incorporate the influence of alliance tenure, we enter the logarithm of the months
since the alliance was established, setting new alliances to have tenure of 6 months. Withdrawal
in the first year is rare, so we enter an indicator variable set to one in the first year of an alliance.
We incorporate time-varying effects through interactions with the logarithm of the tenure.
Relationship level. To test hypothesis 1, which predicts the effect of relational cohesion,
we calculate two measures. The first is the number of alliance ties that a firm had with members
of the focal alliance at time t-1. The second is the number of arm’s length ties that the firm had
with members of the focal alliance at time t-1. The expectation is that an alliance with partners
that also meet in other alliances will be more stable. We consider both alliances and arm’s length
ties because any kind of interfirm relationships can be a source of cohesion (Beckman &
Haunschild, 2002; Marsden & Campbell, 1984). Although both may be a source of cohesion, it
is of theoretical interest that the alliance ties are relatively stronger with more interaction and
joint problem solving, whereas arm’s length ties imply a transaction but not joint operation of the
route. Hence the trust account of relational cohesion applies mainly to alliance ties, while the
monitoring and incentives account applies to both types of ties. Indeed, it is easier to terminate
arm’s length ties, so termination to punish an uncooperative partner is a better option for those.
The myopic search account of tie initiation with prior contacts could also potentially apply to
both types, but it seems more likely that local search for new alliance partners takes place
primarily, if not exclusively, among prior alliance partners.
To test hypothesis 2 regarding the effect of resource incompatibility, we compute the
absolute difference of the firm’s resources from that of the mean of the focal alliance’s partners
(Wang & Zajac, 2007). We consider compatibility on three ship characteristics: age, maximum
speed (in knots), and capacity (in TEU). The variables compare the fleet average of the firm and
those of the focal alliance partners, so it does not take into account focal-firm variation in these
characteristics. Our interviews with managers in the shipping industry confirmed that these three
ship characteristics are typically used to assess partner resource fit and compatibility, although
one manager pointed out that he does not investigate ship age directly but infers it from ship type.
Because shipping firms’ fleet composition change annually, even if initially high, resource
compatibility among firms participating in an alliance can decline over time.
Network level. Hypothesis 3, which predicts the effect of triadic closure is tested using the
logarithm of the number of closed triads (i.e., common third-party ties) involving the focal firm
and the alliance partners (Krackhardt, 1998, 1999). Hypothesis 4 on position power inequality is
tested by measuring the difference between a firm’s network constraint and the average
constraint of its partners in the focal alliance. A firm’s ego network is constraining if it has few
partners or if its partners are connected to each other and thus can coordinate against it. A
constrained firm lacks structural holes that might provide opportunities to control the flow of
information among its partners, and thus its partners are better able to detect and jointly
sanctioning its behaviors. A more constrained firm is thus less powerful because it is more
dependent on and controlled by its partners (e.g., Gargiulo & Benassi, 2000). More formally,
network constraint, ci, is measured as (Burt, 1992: 55):
where pij, piq, and pqj are the proportional strength of i’s relationship with j, that with i’s other
contacts q, and that of q’s relationship with j, respectively. The number of alliances between two
firms is used as the strength measure.
Market level. Following the International Transportation Handbook, we define market
areas as corresponding to trade routes and treat variations in port visits on these routes as product
differentiation within the global market. For example, North Asia-West Europe is a trade route
served by many different routes, often with port calls in Japan, China, Taiwan or Korea, the U.K.,
and finally Germany or Netherlands. For hypothesis 5 we calculate multimarket contact among
firms participating in the market served by the focal alliance using the proportional multimarket
contact measure suggested by Baum and Korn (1996). This measure captures the potential for
strategic interactions among all firms in the market as the average proportion of markets each
focal firm has that is also occupied by each of its competitors. Firms will high overlap will affect
each other strongly and hence will be more likely to respond to each other’s actions. Hypothesis
6 is tested with a measure of each firm’s market overlap with other firms in the industry
(regardless of whether they are members of the same alliance or not). Market overlap is equal to
the sum of the firm density of the markets in which it participates (Baum & Korn, 1996), with
firms operating in higher density markets facing stronger competition.
Control variables. We include controls for route, alliance, firm, and market characteristics.
For routes, we control for the (logged) number of days between sailings (frequency), the number
of alliance partners in the focal alliance, and the number of charter partners in the focal alliance.
We also control for whether or not the focal alliance has feeder routes. For the focal firm, we
enter its fleet size (number of ships), an indicator variable for owning no container ships, and
capacity growth (TEUs) in the preceding year. For the market, we enter the number of
withdrawals from alliances operating in the area, as withdrawals may trigger realignment in other
alliances (Olk & Young, 1997). All independent variables are lagged one year.
=====TABLE 2 ABOUT HERE=====
Table 2 shows the descriptive statistics and correlation coefficients for the analysis. Prior
alliance ties and closed triads correlate at the .62 level because of the anchoring at zero (firms
with no ties have no closed triads), but other correlation coefficients are low.
Table 3 shows the results of the analysis of firm withdrawals from alliances. We note again that
the dependent variable is that a firm withdrew from the alliance, but other firms in the alliance
continued to offer the route served by the alliance. Route failures are censoring events in this
analysis, so the analysis shows the determinants of firms’ withdrawals from alliances serving
routes that other members in the alliance still view as commercially viable. Model 1 is a baseline
model, containing control variables only. In model 2, we add main effects of variables for testing
hypothesis 1 to 6. Models 3 and 4 incorporate cross-level cohesion effects stated in hypothesis 7
and time-dependent effects stated in hypothesis 8, respectively. Model 5 is a full model
containing all control, independent, and interaction variables. In model 6, we drop insignificant
interactions as a robustness check and find the results essentially unchanged. The findings also
remain unchanged when we test interaction effects with centering techniques and when we
entered the variables in the main-effect Model 2 in groups of one or two levels of analysis at a
time instead of all at once. As the table shows, most variables show consistent results across the
models, so we proceed to interpret the full model 5, but we also note differences in results from
the reduced models 1-4.
=====TABLE 3 ABOUT HERE=====
Hypothesis 1 on the relationship-level cohesion is tested by the coefficients for the number
of prior alliance and arm’s length ties. Alliance ties are not significant until model 3, where they
have a negative effect on member withdrawal as predicted in hypothesis 1. They remain
significant in later models that enter time-dependent effects to test hypothesis 8.
Arm’s length ties are negative and significant in model 2, but lose significance in the more
complete models. This suggests that alliance ties are more consequential for cohesion than arm’s
length ties as a result of the direct social interaction found in alliance, but not in arm’s length, ties.
It may also be that the greater ease with which arm’s length ties can be unilaterally terminated as
punishment limits their ability to foster cohesion. These findings thus contradict the incentive
interpretation of prior ties; however, they do not distinguish between the trust and the myopic-
search interpretations, which are both consistent with these findings.
Support for hypothesis 2 on relational friction, which is tested by the coefficients for fleet
differences, are mixed. Across all models, ship speed difference has a negative and significant
effect on withdrawal. Although significant only at the p < .10 level in models 4 and 5, this
contradicts hypothesis 2. Ship size difference has a positive and highly significant effect on
withdrawal, however, consistent with hypothesis 2. There is thus a tradeoff between allocating
ships of the same size and speed when alliance members have limited fleets, and the estimates
indicate that it is worse for an alliance to have mismatched ship size than ship speed. This
finding is particularly interesting when comparing with how these same firms matched ships
when initiating alliances. In alliance initiation, firms found partners with similar age and speed of
ships, reducing variance in these criteria, but the preference for these criteria led to some
alliances with high differences in ship size (author cite). Thus, the matching criterion that was
not used for alliance initiation turned out to affect withdrawal.
Hypothesis 3 on network cohesion is tested by the coefficient estimate of closed triads.
This estimate is positive and significant across all models, contrary to the hypothesis. Puzzled
by this finding, we conducted a post-hoc analysis of the distribution of closed triads in
observations with member withdrawal and without it. We found that the risk of withdrawal of
one or more members did not differ as a function of closed triads. Alliances with more closed
triads were, however, prone to multi-member withdrawal, which rarely happened in alliances
without closed triads. Hence, while closed triads do not appear to reduce the likelihood of
withdrawal events, they do increase their magnitude. This is an interesting and unexpected
illustration of the dark side of embeddedness (Granovetter, 1985). To explore it further, we
examined the networks of alliances in the data that had multi-member breakups. We found that in
many cases, they included some firms sharing many closed triads, but also some firms more
weakly connected to the rest. The multi-member breakups often involved joint withdrawal of
either the best-connected firms or the weakly connected firms. Hence, closed triads cause joint
behavior, but joint behavior can mean joint withdrawal rather than stability.
Consistent with the hypothesis 4 on network friction, coefficient estimates for constraint
difference are positive and significant in all models. Thus, less powerful (more constrained)
firms in an alliance are more likely to withdraw from it, presumably because of actual or
anticipated failures to negotiate good terms when the alliance needed to distribute rewards or
implement changes. As noted earlier, less powerful firms in an alliance may either leave
voluntarily or be pushed out, and it is difficult to distinguish which of these events occur based
on the exit rates. However, some indication can be gained from an examination of what firms do
after leaving an alliance. If most exits are involuntary, then we may expect to see that the firms
exiting will not enter new alliances right away, but will either start new routes alone or not do
any new route action. The reason is that alliance entry is a negotiated outcome that is difficult to
accomplish quickly. By contrast, if most exits are voluntary, then we should expect a greater
incidence of alliance entry, because a firm that exits voluntarily can negotiate entry into a new
alliance while it is still in the focal alliance, and can leave as soon as the negotiations are
successful. We found that 48 percent of firms leaving an alliance entered a new alliance in a
competing route in the next year, suggesting that many exits are voluntary actions by firms that
continue to operate in the same market. 19 percent entered a new alliance in another market, so
they found a use for their ships but exited the focal market. A minority of firms (27 percent)
appeared to have difficulty adjusting to alliance exits, and hence was more likely to have been
involuntary exits. The rest (6 percent) made multiple actions simultaneously, complicating
interpretation of the cause.
The coefficients for multimarket contact are significant and negative, consistent with
hypothesis 5 on market cohesion. It appears, however, that a more detailed specification that
also includes the other factors that influence member withdrawals is needed in order to tease out
the effect fully. The finding shows that firms are less likely to leave alliances serving routes in
which they have high multimarket contact. The coefficient of market overlap is positive and
significant, in support of hypothesis 6. Thus, a firm’s market overlap is a source of friction and
disruption that may cause it to leave its alliances.
Hypothesis 7 posits cross-level cohesion interaction effects of relational embeddedness
with network and market embeddedness. Of the four cross-level interaction terms, only the two
involving multimarket contact are significant. The positive sign of the interaction of prior
alliance ties and multimarket contact is opposite of the negatively signed main effects,
supporting the substitution predicted in hypothesis 7. For arm’s length ties, the main effect is not
significant; consequently, the negative interaction term suggests that arm’s length ties yield
cohesion only when the firms have multimarket contact. Although this finding is not anticipated
by the substitution argument, it can be understood from the viewpoint of multimarket theory. It
is a key assumption of multimarket theory that the firms have asymmetric relations so that each
firm can potentially harm the others in some markets, and is exposed to harm in others
(Bernheim & Whinston, 1990). Measures of multimarket contact capture the extent of the
contact instead of the potential dominance in each market, but the assumption is that firms will
rationally distribute their contacts so that each has a market in which it can attack the others.
Arm’s length ties also involve unequal power because the firm that purchases a slot acts as a
customer of the other. Each party in these ties can unilaterally fire the other, which will
inconvenience the seller if there are alternative routes operated by competitors, and the buyer if
there are few alternative routes in the area. Hence arm’s length ties operate as an extension of
the multimarket effect, but do not have an independent effect.
To test hypothesis 8, models 4 and 5 enter time-dependent interactions of prior alliance ties
and closed triads. The interaction with prior ties has the opposite sign of the main effect,
consistent with hypothesis 8. Thus prior alliance ties initially have lower withdrawal rates, but
this effect attenuates over time. Visual inspection of a plot of the interaction effects (not reported
to conserve space) shows that the prior alliance effect is short-lived, reducing withdrawal arrest
only in the early years of an alliance. Similarly, the interaction of time and structural
embeddedness is significant and has the opposite sign of the main effect, showing that the effect
of common third-party ties is also strongest in the early years of an alliance, and declines
thereafter. Visual inspection of a plot of this time-varying effect (not reported to conserve space)
indicates that closed triads initially increase the risk of withdrawal, but that, consistent with
hypothesis 8, the effect weakens as the alliance ages.1
1 In addition to the analyses shown here, we also conducted analyses with time-dependent variables for
relational and network friction, and market sources of cohesion and friction. These effects were not
hypothesized, and the coefficients testing them did not obtain significance. Hence the analysis shows
time-dependent effects for the predicted coefficients and not for other coefficients.
DISCUSSION AND CONCLUSION
We have sought to show that theoretical and empirical analysis of member withdrawal from
alliances can contribute to our knowledge of alliances by providing comparison of the factors
that influence initiation and withdrawal. In addition, it provides insight into new theoretical
mechanisms that are unique to withdrawals. With results in hand, we can now consider whether
the findings indeed inform the theory and practice of alliances. First, we argued that factors
known to produce alliance formation would also reduce withdrawals from alliances. Although
this prediction naturally follows from the theory, it was not clear a priori that it would be
supported because the selection at the formation stage could be so strong that ill-fated alliances
would not occur. Our findings supported lower withdrawal rates from alliances with higher
relational embeddedness (prior alliance ties), but showed that high structural embeddedness
(third-party ties) actually increased withdrawals by making joint withdrawals more likely. The
first of these findings is a straightforward confirmation of the cohesive effect of prior ties, while
the second showed an unexpected effect of structural embeddedness in triggering multi-member
withdrawal. These findings are interesting because they suggest limitations to managerial
foresight. Partner selection theory states that embeddedness is favored because it reduces
uncertainty, but this argument implies that embeddedness increases the duration of alliance
membership only if the uncertainty perceived by managers matches the actual uncertainty. Our
findings suggest that managers correctly anticipate that direct ties will improve cooperation, but
overlook the potential for shared third-party ties for spreading conflicts.
Second, we hypothesized that withdrawal rates were also affected by the market level of
analysis, with multimarket contact making withdrawal less likely, and competition making it
more likely. This argument introduces a contextual effect on alliances that is most likely
especially prominent in analysis of withdrawals. Again, managerial foresight enters the picture.
Alliances perform tasks that result in market activities and are in turn affected by market
conditions. While managers are likely to initiate alliances under favorable market conditions,
there is no guarantee that these conditions will persist over the duration of an alliance. Indeed,
our findings suggests that some alliances experienced market conditions that triggered
withdrawals, suggesting either that market effects were overlooked at the point of initiating the
alliance or that they changed after initiation. We think that the latter is more likely, but analysis
of market effects on alliance initiation would be needed to test this conjecture.
Third, we noted that embeddedness at different levels of analysis does not necessarily “add
up” in member withdrawals. Once partners have found each other and secured sufficient trust
and incentives to initiate an alliance, events that occur during the actual collaboration affect the
continuation decision strongly – positively or negatively – regardless of the initial level of
embeddedness. This is the simple result of collaboration experience becoming available during
the course of the alliance, making even multiple layers of embeddedness relatively less important
during the alliance than before it is initiated. Evidence in support of this line of reasoning was
found in the tradeoff observed between multimarket contact and prior alliances, but overall the
support for this part of our overall model was weaker than we anticipated.
Finally, another effect of events that occur during the actual collaboration is a declining
influence of relational and structural embeddedness over time. The actual experience of
collaboration can either increase the cohesion or decrease it, depending on the content, but in
both cases the result is a weaker effect of the structures that facilitated the alliance formation in
the first place. We did in fact find weaker effects over time of both the expected positive effect
of relational embeddedness and the unexpected negative effect of structural embeddedness.
These findings yield further support to a view of alliances as a matching problem under
uncertainty (Jovanovic, 1979; author cite) because they show that the social structures that
decision makers use to reduce uncertainty prior to the alliance initiation lose their effect as
information is revealed during the course of collaboration.
The investigation has thus showed that the reasons for alliance breakups are not simply the
reverse of those leading to alliance formation. There are two sources of difference. On the one
hand, time allows a buildup of trust through ongoing interactions. So, the advantage of prior ties
for alliance formation is clear, but their role in stabilizing alliances attenuates with time, much
like other forms of initial endowment and goodwill (Levinthal & Fichman, 1988; Park & Russo,
1996; Seabright et al., 1992). On the other hand, events that are difficult to foresee can
undermine the alliance. For example, market events are important for withdrawal because it is
likely that the withdrawals occur as a result of unforeseen changes in market conditions rather
than from a failure to appreciate the market conditions at the time of initiation.
Similarly, the effect of power differences that we showed is the best known from earlier
research on network tie termination because the disadvantages from having low power are
difficult to predict in advance (Baker et al., 1998; Rowley et al., 2005). The corresponding
hypothesis for alliance initiation would be that managers anticipate problems caused by having
powerful partners, and avoid entering such partnerships. This prediction is worth investigating,
but it is already clear that the anticipation of such problems is partial at best, given that we found
sufficient variation in partner power to robustly support this prediction. Resource
incompatibility also seems an apt criterion to apply at the time of alliance formation, but again
there was sufficient variation in the data to show that some alliances broke up over differences in
the fleets operated by their members. Hence, limited managerial ability to anticipate alliance
problems is clearly shown in these findings.
Our findings imply that member withdrawal is an outcome of complex, multilevel factors
that require managerial attention. As earlier work on alliances and embeddedness has shown,
attention to managing personal interactions and socialization with partners is important (Larson,
1992; Uzzi, 1996). In addition, managers seeking interorganizational collaboration must attend
to forces beyond the alliances, including some that they have little control over. A competitor’s
market entry may, for example, alter a firm’s intensity of market overlap and multimarket contact
with alliance partners, creating unintended tension within its alliances. However, even market
conditions are to some extent controllable by a firm’s own market entry and exit decisions and
their management of multimarket ties. It is important to know that these factors affect alliances,
as we have shown, and also that obtaining a powerful network position can destabilize alliances
by spurring less powerful partners to withdraw. If managers are aware of such market and
network level effects on their alliances, they can manage the resulting tradeoffs.
A possible extension of this work is to consider interactions between the cohesion and
friction factors. Three conditions may give rise to such interactions. First, if cohesion and friction
factors share the same causal mechanism, they are likely to interact. We predicted and found
such interactions among cohesion factors related to trust building, but did not observe any shared
causal mechanism between these and the friction factors. Second, when viewing alliance
withdrawal as a decision-making problem for the (potentially) withdrawing firm, there will
normally be tradeoffs between all factors. Decision-making rules often incorporate a preference
for even strengths across criteria, suggesting interactions to incorporate effects such as a
discounting of high resource compatibility when the power balance is unfavorable. Third, when
viewing alliance withdrawal as an outcome of conflict, a single negative factor may adversely
affect the entire relation. This would suggest interactions in order to incorporate a spreading
effect. For example, the impact of friction from market overlaps may increase with high
relational cohesion because firms react more sensitively to small frictions in relations with
strongly attached partners. We focused on the first of these conditions in this work, and leave the
others for future research.
This study has some limitations that must be acknowledged. First, the data cover trans-
ocean shipping routes operated by firms originating from 37 nations, so its generalizability
across cultural and institutional contexts should be high, but it is still limited by its sample of
firms coming from one industry. Moreover, unlike most studies on interorganizational relations,
which frequently investigate R&D alliances, this study examined production-oriented alliances.
This uniqueness of the data is an asset to the study, but also suggests caution in generalization.
For instance, we hypothesized and actually found substitutive, not additive, interactions between
cohesive forces at the different levels on the basis of an idea that there is a threshold required for
stability of alliances. But, this finding may not hold for other types of alliances such as R&D
alliances, in which effective collaboration may need extremely high cohesion. Thus, future
research should test our model with the data of different types of alliances.
Second, while this study focuses on multimarket contact and market overlap to test the
market effects, there may be alternative ways of characterizing markets. For instance, we lack
data of shipping firms’ network ties with actors outside the industry, such as customers, railroad
forwarders, trucking companies, and airlines. Such linkages may impact shipping alliance
networks. More data are thus needed to capture market effects comprehensively. Third, while
several studies have examine the performance consequences of alliance formation (e.g., Baum, et
al., 2000; Powell et al., 1996; Stuart, 2000), very little is known about performance consequences
of alliance termination and member withdrawal. Future research should examine whether
managers can use alliance breakups to enhance firm performance.
We have shown some of the potential that investigation of alliance breakups has for testing
and extending models that account for the dynamics of interorganizational networks. Such work
is both an extension of past work and a corrective to it, as it leads to new hypotheses and new
tests of existing ones. Because alliance breakups are linked to failures in collaboration
particularly when alliances are built to last, it helps assess how alliances succeed or fail.
Research on alliance breakups is also needed in order to link findings on alliance formation to
alliance network evolution over time. For these reasons, and because it has seen far less
investigation than alliance initiation, it is an urgent task to continue research on when and why
firms leave alliances.
Aiken, M. & Hage, J. 1968. Organizational interdependence and intra-organizational structure.
American Sociological Review, 33: 912-930.
Allison, P. D. 1982. Discrete time methods for the analysis of event histories. In S.Leinhardt
(Ed.), Sociological Methodology 1982: 61-98. San Francisco: Jossey Bass.
Axelrod, R. 1984. The evolution of cooperation. New York: Basic Books.
Bae, J. & Gargiulo, M. 2004. Partner substitutability, alliance network structure, and firm
profitability in the telecommunications industry. Academy of Management Journal, 47:
Baker, W. E. 1990. Market networks and corporate behavior. American Journal of Sociology, 96:
Baker, W. E., Faulkner, R. R., & Fisher, G. A. 1998. Hazards of the market: The continuity and
dissolution of interorganizational market relationship. American Sociological Review, 63:
Baum, J. A. C. & Korn, H. J. 1996. Competitive dynamics of interfirm rivalry. Academy of
Management Journal, 39: 255-291.
Baum, J. A. C. & Korn, H. J. 1999. Dynamics of dyadic competitive interaction. Strategic
Management Journal, 20: 251-278.
Baum, J. A. C. & Singh, J. V. 1996. Dynamics of organizational responses to competition. Social
Forces, 74: 1261-1297.
Baum, J. A. C., Calabrese, T., & Silverman, B. S. 2000. Don't go it alone: Alliance network
composition and startups' performance in Canadian biotechnology. Strategic Management
Journal, 21: 267-294.
Baum, J. A. C., Rowley, T. J., Shipilov, A. V., & Chuang, Y.-T. 2005. Dancing with strangers:
Aspiration performance and the search for underwriting syndicate partners. Administrative
Science Quarterly, 50: 536-575.
Baum, J. A. C., Shipilov, A. V., & Rowley, T. J. 2003. Where do small worlds come from?
Industrial and Corporate Change, 12: 697-725.
Beckman, C. M. & Haunschild, P. R. 2002. Network learning: The effects of partners’
heterogeneity of experience on corporate acquisitions. Administrative Science Quarterly,
Benjamin, B. A. and Podolny, J. M. 1999. Status, Quality, and Social Order in the California
Wine Industry. Administrative Science Quarterly, 44: 563-89.
Bernheim, B. D. & Whinston, M. D. 1990. Multimarket contact and collusive behavior. RAND
Journal of Economics, 21: 1-26.
Bleeke, J. & Ernst, D. 1991. The way to win in cross-border alliances. Harvard Business Review,
Boeker, W., Goodstein, J., Stephan, J., & Murmann, J. P. 1997. Competition in a multimarket
environment: The case of market exit. Organization Science, 8: 126-142.
Borgatti, S. P., Everett, M. G., & Freeman, L. C. 2002. Ucinet for Windows: Software for social
network analysis. Harvard, MA: Analytic Technologies.
Bresser, R. K. F. 1988. Matching collective and competitive strategies. Strategic Management
Journal, 9: 375-385.
Broschak, J. P. 2004. Managers' mobility and market interface: The effect of managers' career
mobility on the dissolution of market ties. Administrative Science Quarterly, 49: 608-640.
Burt, R. S. 1992. Structural holes: The social structure of competition. Cambridge, MA: Harvard
Burt, R. S. 2000. Decay functions. Social Networks, 22: 1-28.
Burt, R. S. 2002. Bridge decay. Social Networks, 24: 333-363.
Coleman, J. S. 1988. Social capital in the creation of human capital. American Journal of
Sociology, 94: S95-S120.
Das, T.K. & Teng, B.S.. 1998. Between trust and control: developing confidence in partner
cooperation in alliances. Academy of Management Review, 23: 491-512.
Das, T. K. & Teng, B.S. 2000. Instabilities of strategic alliances: An internal tensions perspective.
Organization Science, 11: 77-101.
Dobrev, S.D. & T. Kim. 2006. Positioning among organizations in a population: Moves between
market segments and the evolution of industry structure. Administrative Science Quarterly,
Dwyer, R., Schurr, P., & Oh, S. 1987. Developing buyer-supplier relations. Journal of Marketing,
Eisenhardt, K. M., & Schoonhoven, C. B. 1996. Resource-based view of strategic alliance
formation: Strategic and social effects in entrepreneurial firms. Organization Science, 7:
Gargiulo, M. & Benassi, M. 2000. Trapped in your own net? Network cohesion, structural holes
and the adaptation of social capital. Organization Science, 11: 183-196.
Geletkanycz, M. A. & Hambrick, D. C. 1997. The external ties of top executives: Implications
for strategic choice and performance. Administrative Science Quarterly, 42: 654-681.
Granovetter, M. S. 1985. Economic action and social structure: The problem of embeddedness.
American Journal of Sociology, 91: 481-510.
Greve, H. R. 1998. Performance, aspirations, and risky organizational change. Administrative
Science Quarterly, 44: 58-86.
Greve, H. R. 2000. Market niche entry decisions: Competition, learning, and strategy in Tokyo
banking, 1894-1936. Academy of Management Journal, 43: 816-836.
Gulati, R. 1995. Social structure and alliance formation patterns: A longitudinal analysis.
Administrative Science Quarterly, 40: 619-652.
Gulati, R. & Gargiulo, M. 1999. Where do interorganizational networks come from? American
Journal of Sociology, 104: 1439-1493.
Gulati, R. & Westphal, J. D. 1999. Cooperative or controlling? The effects of CEO-board
relations and the content of interlocks on the formation of joint ventures. Administrative
Science Quarterly, 44:473-506.
Haveman, H. A. & Nonnemaker, L. 2000. Competition in multiple geographic markets: The
impact on growth and market entry. Administrative Science Quarterly, 45: 232-267.
Heckman, J. J. 1979. Sample selection bias as a specification error. Econometrica, 47: 153-161.
Inkpen, A. C. & Dinur, A. 1998. Knowledge management processes and joint ventures.
Organization Science, 9: 454-468.
Jayachandran, S., Gimeno, J., & Varadarajan, P. R. 1999. The theory of multimarket
competition: A synthesis and implications for marketing strategy. Journal of Marketing,
Kale, P., Singh, H., & Perlmutter, H. 2000. Learning and protection of proprietary assets in
strategic alliances: Building relational capital. Strategic Management Journal, 21: 217-237.
Karnani, A. & Wernerfelt, B. 1985. Multiple point competition. Strategic Management Journal,
Kogut, B. 1989. The stability of joint ventures: Reciprocity and competitive rivalry. Journal of
Industrial Economics, 38: 183-198.
Koka, B. R., Madhavan, R., & Prescott, J. E. 2006. The evolution of interfirm networks:
Environmental effects on patterns of network change. Academy of Management Review,
Krackhardt, D. 1998. Simmelian ties: Super strong and sticky. In R.M.Kramer & M. A. Neale
(Eds.), Power and Influence in Organizations: 21-38. Thousand Oaks, CA: Sage.
Krackhardt, D. 1999. The ties that torture: Simmelian tie analysis in organizations. In
S.B.Andrews & D. Knoke (Eds.), Research in the Sociology of Organizations, 16 183-210.
Stamford, CT: JAI Press.
Larson, A. 1992. Network dyads in entrepreneurial settings: A study of the governance of
exchange relationships. Administrative Science Quarterly, 37: 76-103.
Lee, L. F. 1983. Generalized econometric models with selectivity. Econometrica, 51: 507-512.
Levine, S., & White, P.E. 1961. Exchange as a conceptual framework for the study of
interorganizational relationships. Administrative Science Quarterly, 5: 583- 601.
Levinthal, D. A., & Fichman, M. 1988. Dynamics of interorganizational attachments: Auditor-
client relationships. Administrative Science Quarterly, 33: 345-369.
Lewin, K. 1947. Frontiers in group dynamics. Human Relations, 1: 1-38.
Li, S. X. & Rowley, T. J. 2002. Inertia and evaluation mechanisms in interorganizational partner
selection: Syndicate formation among U.S. investment banks. Academy of Management.
Journal, 45: 1104-1119.
Luo, Y. 2001. Antecedents and consequences of personal attachment in cross-cultural
cooperative ventures. Administrative Science Quarterly, 46: 177-201.
Makino, S., Chan, C. M., Isobe, T., & Beamish, P. W. 2007. Intended and unintended
termination of international joint ventures. Strategic Management Journal, 28:1113-1132.
Marsden, P. V. & Campbell, K .E. 1984. Measuring tie strength. Social Forces, 63: 482-501.
Mohr, J. & Spekman, R. 1994. Characteristics of partnership success: Partnership attributes,
communication behavior, and conflict resolution techniques. Strategic Management
Journal, 15: 135-152.
Nohria, N. 1992. Is a network perspective a useful way of studying organizations? In N. Nohria
& R. G. Eccles (Eds.), Networks and organizations: Structure, form, and action: 1-22.
Boston, MA: Harvard Business School Press.
North, D. C. 1990. Institutions, Institutional Change, and Economic Performances. Cambridge:
Cambridge University Press.
Oliver, C. 1990. Determinants of interorganizational relationships: Integration and future
directions. Academy of Management Review, 15: 241-265.
Olk, P. & Young, C. 1997. Why members stay in or leave an R&D consortium: Performance and
conditions of membership as determinants of continuity. Strategic Management Journal,
Oxley, J. E. 1997. Appropriability hazards and governance in strategic alliances: A transaction
cost approach. Journal of Law, Economics, and Organization, 13: 387-409.
Park, S. H. & Russo, M. V. 1996. When competition eclipses cooperation: An event history
analysis of joint venture failure. Management Science, 42: 875-890.
Park, S. H. & Ungson, G. R. 2001. Interfirm rivalry and managerial complexity: A conceptual
framework of alliance failure. Organization Science, 12: 37-55.
Parkhe, A. 1993. Strategic alliance structuring: A game theoretic and transaction cost
examination of interfirm cooperation. Academy of Management Journal, 36: 794-829.
Pfeffer, J. & Salancik, G. R. 1978. The external control of organizations. New York: Harper and
Podolny, J. M. 1994. Market uncertainty and the social character of economic exchange.
Administrative Science Quarterly, 39: 458-483.
Podolny, J. M., K. L. Page. 1998. Network forms of organization. In K.S.Cook & J. Hagan
(Eds.), Annual Review of Sociology: 57-76. Palo Alto: Annual Reviews.
Powell, W. W., Koput, K. W., & Smith-Doerr, L. 1996. Interorganizational collaboration and the
locus of innovation: Networks of learning in biotechnology. Administrative Science
Quarterly, 41: 116-145.
Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. 2005. Network dynamics and
field evolution: The growth of interorganizational collaboration in the life sciences.
American Journal of Sociology, 110: 1132-1205.
Provan, K. G. & Skinner, S. J. 1989. Interorganizational dependence and control as predictors of
opportunism in dealer-supplier relations. Academy of Management Journal, 32: 202-212.
Ring, P.S. & Van de Ven, A.H. 1992. Structuring cooperative relationships between organizations.
Strategic Management Journal 13:438-98.
Rowley, T. J., Behrens, D., & Krackhardt, D. 2000. Redundant governance structures: An
analysis of structural and relational embeddedness in the steel and semiconductor
industries. Strategic Management Journal, 21: 369-386.
Rowley, T. J., Greve, H. R., Rao, H., Baum, J. A. C., & Shipilov, A. V. 2005. Time to break up:
The social and instrumental antecedents of exit from interfirm exchange cliques. Academy
of Management Journal, 48: 499-520.
Ryoo, D. K. & Thanopoulou, H. A. 1999. Liner alliances in the globalization era: a strategic tool
for Asian container carriers. Maritime Policy and Management, 26: 349-367.
Seabright, M. A., Levinthal, D. A., & Fichman, M. 1992. The role of individual attachments in
the dissolution of interorganizational relationships. Academy of Management Journal, 35:
Simonin, B.L. 1997. The importance of collaborative know-how: An empirical test of the
learning organization. Academy of Management Journal, 40: 1150-1174.
Steensma, H. K. & Lyles, M. A. 2000. Explaining IJV survival in a transitional economy through
social exchange and knowledge-based perspectives. Strategic Management Journal, 21:
Stuart, T. E. 2000. Interorganizational alliances and the performance of firms: A study of growth
and innovation rates in a high technology industry. Strategic Management Journal, 21:
Uzzi, B. 1996. The sources and consequences of embeddedness for the economic performance of
organizations: The network effect. American Sociological Review, 61: 674-698.
Uzzi, B. & Gillespie, J. J. 2002. Knowledge spillover in corporate financing networks:
Embeddedness and the firm’s debt performance. Strategic Management Journal, 23: 595-
Walker, G., Kogut, B., & Shan, W. 1997. Social capital, structural holes and the formation of an
industry network. Organization Science, 8: 109-125.
Wang, L. & Zajac, E. J. 2007. Alliance or acquisition? A dyadic perspective on interfirm
resource combinations. Strategic Management Journal, 28: 1291-1317.
Williamson, O. E. 1975. Markets and hierarchies: Analysis and antitrust implications. New
York: Free Press.
Zollo, M., Reuer, J. J. & Singh, H. 2002. Interorganizational routines and performance in
strategic alliances. Organization Science, 13: 701-713.
Table 1: Theoretical Model and Hypotheses
Relationship Network Market
H1: Prior ties H3: Triadic closure
H8: Time-dependent effect H8: Time-dependent effect
H5: Multimarket contact
H4: Position power
inequality H6: Market overlap
Table 2: Descriptive statistics and correlation coefficients
Variable Mean Std. Dev. 1234 5 67
1 Alliance tenure 2.755 .871 1
2 Log frequency 2.099 .403 .02 1
3 Number of members 3.183 1.433 .16 .09 1
4 Number of charters 0.206 0.609 .09 .00 .02 1
5 Feeder route .195 .396 .03 -.12 .02 .05 1
6 Firm fleet size 33.676 54.515 .02 -.07 .07 -.06 -.01 1
7 No container ships .298 .457 -.09 .11 -.13 .01 -.05 -.38 1
8 Firm TEU growth .019 .073 .01 -.04 .03 .01 .02 .07 -.17
9 Withdrawals in area 5.487 6.553 -.01 -.18 -.17 -.02 -.09 .00 -.02
10 Inverse Mills ratio 1.152 .290 .59 .05 .26 .04 .03 .08 -.17
11 Number of alliance ties 14.115 13.974 .08 -.07 .51 -.09 -.13 .10 -.23
12 Number of arm’s length ties 1.766 3.816 .05 -.06 .34 .03 -.01 .22 -.14
13 Ship age difference 2.390 2.773 .04 -.01 .06 -.05 .06 -.07 .10
14 Ship speed difference 1.742 1.745 -.03 .03 -.04 -.05 -.07 -.05 .34
15 Ship TEU difference .885 .768 .03 .02 .08 -.05 -.05 -.07 .20
16 Closed triads 1.569 1.210 .24 .13 .75 .00 -.04 .17 -.31
17 Constraint difference .000 .2196 .00 -.00 -.00 .00 -.00 .12 -.18
18 Multimarket contact .488 .321 .01 -.16 .03 -.06 -.04 .17 -.25
19 Market overlap 57.681 37.203 -.08 -.23 -.19 .01 -.01 -.06 .11
8 9 10 11 12 13 14 15 16 17 18
9 .02 1
10 -.04 -.01 1
11 .08 -.05 .24 1
12 -.01 -.02 .13 .16 1
13 -.05 -.03 .05 -.04 -.02 1
14 -.08 .04 -.07 -.14 .01 .15 1
15 -.09 .02 .11 .03 .06 .19 .29 1
16 .06 -.13 .34 .63 .32 -.02 -.12 .13 1
17 .04 -.00 .00 .02 .00 -.03 -.03 -.09 .05 1
18 .07 .05 .09 .23 .10 -.10 -.07 .08 .17 .06 1
19 -.04 .30 -.18 -.15 .03 -.07 .18 .02 -.27 -.10 -.06
Table 3: Complementary Log-log Models of Member Withdrawals
Model (1) (2) (3) (4) (5) (6)
New alliance -1.094* -1.239* -1.152* -1.323** -1.200* -1.250*
(0.468) (0.485) (0.490) (0.499) (0.505) (0.502)
Alliance tenure -0.698* -0.738* -0.708* -0.691+ -0.628+ -0.665+
(0.290) (0.301) (0.303) (0.358) (0.359) (0.360)
Log frequency 0.425+ 0.312 0.287 0.335 0.312 0.330
(0.235) (0.239) (0.235) (0.236) (0.234) (0.235)
Number of members 0.303** 0.303** 0.205 0.276** 0.189 0.301**
(0.057) (0.107) (0.132) (0.107) (0.132) (0.107)
Number of charters -0.058 -0.034 -0.064 -0.055 -0.069 -0.039
(0.147) (0.157) (0.152) (0.156) (0.152) (0.152)
Feeder route 0.956** 0.975** 0.899** 0.964** 0.904** 0.976**
(0.216) (0.222) (0.225) (0.225) (0.227) (0.224)
Firm fleet size -0.011 -0.009 -0.011 -0.010 -0.011 -0.011
(0.009) (0.009) (0.009) (0.009) (0.009) (0.009)
No container ships 0.569 0.483 0.420 0.436 0.388 0.399
(0.437) (0.440) (0.425) (0.441) (0.432) (0.435)
Firm TEU growth -1.703 -3.114+ -3.349+ -2.631 -2.931 -2.943
(1.643) (1.778) (1.878) (1.796) (1.907) (1.923)
Withdrawals in area 0.039** 0.026+ 0.027+ 0.021 0.021 0.019
(0.014) (0.016) (0.015) (0.016) (0.016) (0.016)
Inverse Mills Ratio 1.756* 1.675+ 1.607+ 1.639+ 1.608+ 1.589+
(0.865) (0.866) (0.853) (0.865) (0.858) (0.863)
Number of alliance ties -0.023+ -0.186** -0.133** -0.282** -0.235**
(0.013) (0.054) (0.047) (0.070) (0.057)
Number of arm’s length ties -0.135* 0.005 -0.370+ -0.163 0.046
(0.054) (0.126) (0.189) (0.215) (0.072)
Ship age difference -0.014 -0.003 -0.017 -0.007 -0.011
(0.042) (0.042) (0.043) (0.042) (0.043)
Ship speed difference -0.167* -0.148+ -0.155+ -0.139+ -0.135+
(0.082) (0.079) (0.081) (0.080) (0.080)
Ship TEU difference 0.505** 0.473** 0.525** 0.491** 0.489**
(0.153) (0.149) (0.153) (0.150) (0.152)
Closed triads 0.510** 0.610** 1.554** 1.625** 1.443**
(0.197) (0.203) (0.446) (0.460) (0.435)
Constraint difference 2.711** 2.617** 2.769** 2.683** 2.645**
(0.545) (0.536) (0.544) (0.539) (0.538)
Multimarket contact -0.720+ -1.197* -0.705+ -1.184* -1.222*
(0.432) (0.507) (0.427) (0.506) (0.515)
Market overlap 0.013** 0.012** 0.013** 0.012** 0.013**
(0.004) (0.004) (0.004) (0.004) (0.004)
Closed triads 0.019 0.015
X alliance ties (0.012) (0.013)
Closed triads 0.010 0.017
X arm’s length ties (0.045) (0.046)
Multimarket contact 0.180** 0.176** 0.167**
X alliance ties (0.050) (0.050) (0.048)
Multimarket contact -0.378** -0.365** -0.322*
X arm’s length ties (0.139) (0.138) (0.126)
Direct alliance ties 0.038* 0.038* 0.039**
X log(alliance tenure) (0.015) (0.016) (0.015)
Direct arm’s length ties 0.087 0.057
X log(alliance tenure) (0.058) (0.057)
Closed triads, -0.369** -0.358* -0.312*
X log(alliance tenure) (0.140) (0.144) (0.133)
Constant -5.768** -6.066** -5.128** -4.857** -4.211** -4.586**
(1.270) (1.317) (1.330) (1.532) (1.534) (1.519)
Log likelihood -514.40 -482.82 -474.23 -478.14 -470.31 -472.00
LR Chi2 against null 116.11** 179.27** 196.45** 188.63** 204.29** 200.91**
Degrees of freedom 24 33 37 36 40 37
LR Chi2 against 1 63.16** 80.34** 72.52** 88.18** 84.8**
Degrees of freedom 9 13 12 16 13
Note: + p<. 10; *p< .05; **p< .01; Standard errors in parentheses; 4,004 observations on 171 firms.
Figure 1: Alliance Network in 2003