ArticlePDF Available

Comparing Alliance Network Structure across Industries: Observations and Explanations

June 2012
Strategic Entrepreneurship Journal 1(3‐4):191 - 209

June 2012
1(3‐4):191 - 209

Authors:

New York University

Much research in strategic entrepreneurship has focused on the consequences of network structure for firm performance. Despite this emphasis, little is known about variation in network structure across industries, or about the antecedents of this variation. In a comparative study of alliance networks in 32 industries, we demonstrate substantial variety in network structure, and develop a typology of network structures. We then endeavor to explain this variation by focusing on dimensions of the products and technologies that characterize these industries—such as technological uncertainty and dynamism, product modularity, and architectural control—and associating them with underlying characteristics of network structure. We conclude with a discussion of implications of our findings for research in strategic entrepreneurship. Copyright © 2008 Strategic Management Society.

. Descriptive statistics for the industries and their alliance networks

…

A typology of alliance network structures

…

Industries ranking high for technological dynamism: engines and turbines, household audio and video equipment, motor vehicles and equipment

…

Industries ranking low for technological dynamism: leather footwear, women's and children's outerwear

…

The pharmaceutical and communications equipment industries

…

Figures - uploaded by Melissa A. Schilling

Content may be subject to copyright.

Content uploaded by Melissa A. Schilling

Content may be subject to copyright.

Strategic Entrepreneurship Journal

Strat. Entrepreneurship J., 1: 191–209 (2007)

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/sej.33

COMPARING ALLIANCE NETWORK STRUCTURE

ACROSS INDUSTRIES: OBSERVATIONS

AND EXPLANATIONS

LORI ROSENKOPF1* and MELISSA A. SCHILLING2

1The Wharton School of the University of Pennsylvania, Philadelphia,

Pennsylvania, U.S.A.

2Stern School of Business, New York University, New York, New York, U.S.A.

Much research in strategic entrepreneurship has focused on the consequences of network

structure for ﬁ rm performance. Despite this emphasis, little is known about variation in

network structure across industries, or about the antecedents of this variation. In a com-

parative study of alliance networks in 32 industries, we demonstrate substantial variety in

network structure, and develop a typology of network structures. We then endeavor to explain

this variation by focusing on dimensions of the products and technologies that characterize

these industries—such as technological uncertainty and dynamism, product modularity, and

architectural control—and associating them with underlying characteristics of network struc-

ture. We conclude with a discussion of implications of our ﬁ ndings for research in strategic

INTRODUCTION

Recent research has demonstrated that interﬁ rm

network structure can signiﬁ cantly inﬂ uence ﬁ rm-

level performance outcomes such as growth (e.g.,

Powell et al., 1996), innovation (e.g., Ahuja, 2000a;

Schilling and Phelps, 2007), and access to venture

capital (Sorenson and Stuart, 2001). Some of the

most prominent mechanisms by which an interﬁ rm

network inﬂ uences ﬁ rm outcomes include shaping

the ﬂ ow of information and other resources between

connected ﬁ rms, providing signals of ﬁ rm quality,

and enabling reciprocity norms through shared

third-party ties (Ahuja, 2000a; Gulati and Gargiulo,

1999; Owen-Smith and Powell, 2004; Schilling and

Phelps, 2007; Stuart, 2000; Uzzi, 1997). However,

despite mounting evidence of the importance of

alliance network structures, neither the variation in

these structures nor the antecedents of this variation

are well understood. There is very little research

documenting systematic differences in alliance

network structure, presumably because of the dif-

ﬁ culty in doing so.1 Only a few databases exist that

track alliances for multiple industries in a consistent

fashion, and harvesting the data from these alliances

in a way that permits accurate network analysis is a

nontrivial task.

Similarly, most extant studies addressing the ques-

tion of where interorganizational alliance networks

come from either limit their scope to a single or few

Keywords: alliances; networks; technology; innovation; entre-

preneurship; small-world networks

*Correspondence to: Lori Rosenkopf, The Wharton School of

the University of Pennsylvania, 2000 Steinberg-Dietrich Hall,

Philadelphia, PA 19104, U.S.A.

E-mail: rosenkopf@wharton.upenn.edu

1 Some notable exceptions include Schilling and Phelps (2007),

who analyze how the characteristics of 11 industry alliance

networks affect patenting activity, and Verspagen and Duysters

(2004), who compare the network structure of the chemicals

and food and electrical equipment industries.

192 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

industries (e.g., Baum, Shipilov, and Rowley, 2003;

Stuart, Ozdemir, and Ding, 2007), or to explaining the

formation of dyadic alliances rather than the overall

network (e.g., Gulati, 1999; Stuart, 1998). Here we

have seen that alliance formations may be predicted

by prior alliance activity (Powell et al., 1996; Walker,

Kogut, and Shan, 1997; Gulati, 1995)—by network

ties in other contexts such as technical committees

(Rosenkopf, Metiu, and George, 2001), director

interlocks (Gulati and Westphal, 1999), patent-

related technology landscapes (Mowery, Oxley, and

Silverman, 1998; Stuart, 1998)—and by technologi-

cal capabilities (Gulati and Gargiulo, 1999; Ahuja,

2000b). Again, this activity has been fruitful, but

it does not address how these alliances aggregate

into an overall structure, how these structures vary,

or what broader industry characteristics may shape

these alliance decisions.

To address these issues, we ﬁ rst assess whether

there are signiﬁ cant and systematic differences in

alliance network structure across different industries

by developing and analyzing 32 industry alliance

networks. We take care to construct and analyze

the networks by consistent means, permitting us to

compare these networks on such dimensions as size,

connectivity, centralization, small-world properties,

and others. We are then able to compare alliance

network structure to other industry features, such as

number of publicly held ﬁ rms, change in total factor

productivity, research and development intensity,

separability of innovation activities, and concentra-

tion of architectural control. We combine this induc-

tive approach with theoretical reasoning to develop a

typology of alliance network structure, and propose

a set of industry factors that shape network structure.

We close by considering the implications of these

ﬁ ndings for future research.

COMPARING NETWORK

STRUCTURES

For the inductive portion of our study, we began

by constructing a sample of industry alliance net-

works. From the full list of three-digit (1987 SIC)

manufacturing industries, we excluded consum-

ables (food and beverage and tobacco) and those

industries that are ‘not elsewhere classiﬁ ed’ desig-

nations (i.e., catchall categories for miscellaneous

products that are not otherwise classiﬁ able under

the SIC system), leaving 103 candidate industries

for our study. For these industries, strategy and

entrepreneurship scholars assessed the levels of

technology-focused characteristics, such as product

modularity, value chain separability, proprietary

standards, and architectural control. Using these

assessments, we selected a total of 32 industries

that demonstrated varying levels of these charac-

teristics.

For our sample, alliance data were gathered using

Thomson’s SDC Platinum database. The SDC data

have been used in a number of empirical studies on

strategic alliances (e.g., Anand and Khanna, 2000;

Schilling and Steensma, 2001; Sampson, 2004).2 For

each industry, we gathered all alliances announced

between 2001 and 2003 that included at least one

ﬁ rm whose primary SIC code (the four-digit SIC

code in which the ﬁ rm generates the largest portion

of its revenues) matched the industry. Both public

and private ﬁ rms were included.

Alliance relationships typically last for more

than a year, but alliance termination dates are rarely

reported. This required us to make an assumption

about how long the alliances lasted. We took a con-

servative approach and assumed that alliance rela-

tionships would last for three years, consistent with

recent empirical work on the average duration of

alliances (Phelps, 2003). Other research has taken

a similar approach, using windows ranging from

one to ﬁ ve years (e.g., Bae and Gargiulo, 2003;

Gulati and Gargiulo, 1999; Stuart, 2000). Thus, we

create the alliance networks based on a three-year

window spanning 2001 to 2003. Each network was

2 Like all other large alliance databases (e.g., MERIT-CATI,

RECAP, Bioscan), SDC is incomplete in that it does not

capture all announced alliances. However, it has been demon-

strated that despite this incompleteness, the pattern of alliance

activity across the major databases is remarkably symmetric.

Furthermore, alliance network structure is highly resilient to

this incompleteness, because the alliance databases (with the

exception of Bioscan) are sampling on the links (the alliances)

rather than the nodes (the organizations). This means that the

likelihood of an organization making it into the sample is

directly related to the number of alliances it publicly announces,

reducing the likelihood of an important hub being overlooked.

Furthermore, an organization’s size and prominence is directly

related to both the number of alliances it is likely to have and

the amount of press attention it is likely to receive, further

reducing the likelihood of a major hub being overlooked, at

least in the datasets that consider all forms of organizations

(Schilling, 2007). This means that while network and main

component size are undoubtedly underestimated, dimensions

such as relative degree, centralization, clustering, etc., are fairly

reliable so long as the sampling methodology is consistent

across the networks being compared.

Comparing Alliance Network Structure Across Industries 193

DOI: 10.1002/sej

constructed as a binary adjacency matrix.3 Since we

were concerned with whether a path existed from

one ﬁ rm to another and not with the effect of multi-

plex relationships, multiple alliance announcements

between the same pair of ﬁ rms in any time window

were treated as only one link. Alliance relationships

are considered to be bidirectional, resulting in an

undirected unipartite graph (Newman, Strogatz, and

Watts, 2001). Ucinet 6.23, a leading social network

analysis software package, was used to obtain

network structure measures on each of these net-

works (Borgatti, Everett, and Freeman, 2002), and

Netdraw was used to create graphical pictures of the

networks (Borgatti, 2002).

Table 1 lists relevant statistics for all 32 of our

networks. The ﬁ rst two measures—change in total

factor productivity and research and development

intensity—capture the technological dynamism of

the industry. The former—change in total factor pro-

ductivity4—examines whether the rate of output from

a given quantity of inputs has changed in an industry

over time. Historically, total factor productivity has

increased signiﬁ cantly across all industries, and part

of this increase is typically imputed to be due to

technological progress that enables factors of

production to be more effective or efﬁ cient (Crafts,

1996; Griliches, 1990; Terleckyj, 1980). This

measure is available at the four-digit SIC level

from the Bertelsman-Gray database available at the

National Bureau of Economic Research, and we used

a weighted average method to aggregate the data

up to the three-digit level for the purposes of our

study. The latter measure, research and development

intensity5, has been repeatedly associated with the

importance of innovation and technological change

in an industry. To measure it, we used industry-level

research and development expenditures divided by

industry-level sales, both obtained from Compustat.

The next measure, average separability of inno-

vation activities, captures the degree to which the

industry is considered to be characterized by innova-

tion activities that can be separated across multiple

ﬁ rms (as, for example, when the industry is character-

ized by interﬁ rm product modularity). To create this

measure, we developed two rating instruments. First,

we created a list of the 103 manufacturing industries

described previously. The two researchers indepen-

dently rated every industry as high in separability (1),

low in separability (−1), or neither (0). The resulting

set of ratings exhibited a coefﬁ cient alpha of 0.71,

suggesting very high interrater reliability (Nunnally,

1978). Thus, we aggregated these scores across the

two researchers to create a single index ranging from

−2 to 2. Second, we gave copies of the list to a set of

13 scholars in strategy and entrepreneurship, asking

them to identify the 10 they felt exhibited the highest

levels of separability of innovation activities.6 These

3 A binary adjacency matrix is a square matrix with nodes

(e.g., ﬁ rms) as rows and columns. The entries in the adjacency

matrix, xij, indicate which pairs of nodes are adjacent (i.e.,

have a relationship). In a binary matrix, a value of 1 indicates

the presence of a relationship between nodes i and j, while a 0

indicates no relationship.

4 Total factor productivity (TFP) growth is a version of the

Solow residual (Solow, 1957). A series of studies of eco-

nomic growth conducted at the National Bureau of Economic

Research attracted attention to the role of technological change

when it was shown that the historic rate of economic growth

in GDP could not be accounted for entirely by growth in labor

and capital inputs. Though many researchers have attempted

to explain this residual away in terms of measurement error,

inaccurate price deﬂ ation, or labor improvement, in each case

the additional variables were unable to eliminate this residual

growth component (Terleckyj, 1980). Gradually a consensus

emerged that despite idiosyncratic measurement effects, the

residual largely captures technological change (Crafts, 1996;

Terleckyj, 1980). Though this residual is primarily used at

the national level, a number of researchers have used TFP

growth at the industry level to model the relationship between

such productivity growth and other variables (e.g., Griliches

and Lichtenberg, 1983; Jorgenson, 1984; Siegel and Griliches,

1991). The TFP growth index is calculated as the growth rate

of output (real shipments) minus the revenue-share-weighted

average of the growth rates of capital, production worker hours,

non-production workers, non-energy materials and energy. The

TFP growth measure used here is the percent average com-

pound growth rate from 2000–2005 for each industry (based

on two-digit SIC) as obtained from the Bureau of Economic

Analysis. This measure has been used in a number of other

studies to capture the rate of industry-level technological

change (e.g., Nelson, 1988; Sterlacchini, 1989).

5 Industry-level research and development intensity (R&D

intensity) captures the degree to which ﬁ rms in an industry

focus on innovation activities and is an oft-cited predictor of

technological innovation rates (e.g., Godoe, 2000; Mariani,

2004). For the industry-level measure used here, we gathered

R&D expenditure and sales data on every publicly held ﬁ rm

in each of the industries. For each industry, we divided the

average R&D expenditures by the average sales. It was impor-

tant to calculate the averages for R&D expenditures and sales

prior to dividing the former by the latter to prevent unusually

large outliers from biasing the measure. This is because if R&D

intensity is calculated at the ﬁ rm level and then averaged, the

resulting number can be skewed upward dramatically by ﬁ rms

that have spent on R&D but not yet earned revenues.

6 In the survey, respondents were asked to nominate the 10

industries with the highest level of either a) ‘. . . modularity

within the products that characterize the industry. Modularity

is deﬁ ned as the degree to which a product’s components may

be separated and recombined’, or b) ‘. . . separability of innova-

tion activities along the value chain (product design, process

design, manufacturing, marketing, etc.) for the products that

characterize the industry. Separability is deﬁ ned as the degree

to which these activities can be distributed among multiple

actors/facilities.

194 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

Table 1. Descriptive statistics for the industries and their alliance networks

Industry Chg in TFP,

1997–20021

R&D

intensity,

2001

Avg

separ-

ability

score

Avg

arch.

control

score

Alliance networks

Nodes Industry

ﬁ rms in

alliances

Avg

alliances

per

industry

ﬁ rm

Alliance

part.

rate2

Network

central-

ization

Harmonic

mean path

length

Weighted

overall

clust.

coef.

Small-

world

quot.

Nodes

in main

comp.

Broadwoven fabric mills, cotton (221) −0.004 0.00 −0.76 −1.06 17 7 1.06 0.30 6.67% 14.29 0.00 0.00 3

Carpets and rugs (227) 0.002 0.00 −0.56 −0.76 3 1 2.00 0.06 0.00% 1.00 1.00 2.37 3

Women’s and children’s outerwear (233) −0.015 0.23 0.29 −1.06 15 7 1.33 0.27 21.98% 8.40 0.38 4.72 5

Fur goods (237) −0.029 0.00 −0.16 −0.76 2 1 1.00 0.17 0.00% 1.00 0.00 0.00 2

Logging (241) 0.000 0.00 −1.12 −0.70 6 3 1.67 0.27 40.00% 1.61 0.00 0.00 6

Wood buildings and mobile homes (245) −0.024 0.00 0.15 0.00 2 1 1.00 0.11 0.00% 1.00 0.00 0.00 2

Household furniture (251) 0.009 0.32 0.51 −0.50 15 9 1.33 0.19 21.98% 8.33 1.50 18.79 5

Public building and related furniture (253) 0.003 1.62 −0.40 0.00 7 3 1.14 0.33 20.00% 4.76 0.00 0.00 3

Paper mills (262) 0.017 0.68 −1.12 −0.40 23 16 1.04 0.24 4.76% 20.41 0.00 0.00 3

Industrial inorganic chemicals (281) −0.017 1.60 −0.76 −0.15 167 59 1.89 0.74 9.82% 50.00 0.49 6.96 18

Drugs (283) −0.014 13.77 −0.40 1.26 913 568 1.62 1.08 1.80% 71.43 0.20 22.11 248

Tires and inner tubes (301) 0.003 3.06 −0.61 −0.15 29 17 1.80 0.41 11.80% 9.62 0.27 2.57 8

Footwear, except rubber (314) −0.030 0.00 0.15 −0.76 12 3 1.00 0.08 0.00% 10.99 0.00 0.00 2

Flat glass (321) 0.012 0.00 −1.12 −0.70 15 9 1.73 1.00 13.46% 5.88 0.27 1.95 5

Cement, hydraulic (324) −0.028 0.00 −1.12 −0.40 30 15 2.00 0.20 11.08% 13.16 0.91 5.07 6

Iron and steel foundries (332) 0.004 0.19 −1.12 −0.05 76 30 1.61 1.30 7.39% 17.86 0.21 4.98 26

Nonferrous rolling and drawing (335) 0.000 1.63 −1.12 −0.70 55 20 1.14 0.19 22.69% 11.11 0.36 5.37 23

Cutlery, handtools, and hardware (342) −0.003 1.62 −0.20 −0.40 24 11 1.00 0.50 15.81% 7.14 0.29 3.57 11

Engines and turbines (351) 0.041 3.47 0.51 1.56 104 23 1.00 0.96 9.22% 11.11 0.64 5.60 36

Computer and ofﬁ ce equipment (357) 0.075 6.93 2.02 0.36 383 113 1.00 0.34 11.19% 18.52 0.37 19.94 173

Refrigeration and service machinery (358) −0.002 1.60 0.20 0.00 57 29 1.00 0.50 10.36% 33.33 0.62 8.91 8

Electric lighting and wiring equipment (364) 0.001 1.20 0.51 −0.70 84 22 1.00 0.33 14.90% 8.33 0.40 5.92 41

Household audio and video equipment (365) 0.007 5.58 1.62 0.30 278 58 1.00 0.82 15.25% 7.63 0.05 6.62 180

Communications equipment (366) −0.057 13.52 1.22 0.76 342 120 1.33 0.48 5.53% 41.67 0.56 14.53 82

Electronic components and accessories (367) −0.019 11.01 0.66 0.20 503 228 1.06 0.38 6.62% 33.33 0.37 27.76 162

Motor vehicles and equipment (371) 0.007 3.93 1.12 1.16 464 239 1.54 0.88 7.74% 21.74 0.35 22.74 203

Aircraft and parts (372) 0.001 4.77 1.17 1.56 143 62 1.58 1.77 5.73% 43.48 0.62 7.50 17

Ship and boat building and repairing (373) 0.028 1.54 −0.25 −0.05 40 18 1.47 0.53 16.73% 14.29 0.63 6.09 10

Guided missiles, space vehicles, parts (376) −0.011 3.15 0.15 2.06 52 13 1.00 2.60 33.26% 5.00 0.54 5.44 32

Medical instruments and supplies (384) −0.008 9.99 −0.20 0.45 222 80 1.44 0.41 2.60% 125.00 0.34 7.84 9

Photographic equipment and supplies (386) 0.004 6.24 0.32 0.20 45 2 1.27 0.05 6.67% 10.53 0.51 5.12 18

Watches, clocks, watchcases, and parts (387) −0.012 0.40 0.47 0.30 26 15 1.43 1.00 20.33% 6.62 0.57 3.78 9

1 Bertelsman-Gray TFP data for 1997 to 2002; weighted averages (weighted by value of shipments) used to aggregate four digit up to three digit.

2 Alliance participation rate calculated as number of industry ﬁ rms participating in alliances divided by number of publicly held ﬁ rms listed in that primary SIC per the Compustat Global

database.

Comparing Alliance Network Structure Across Industries 195

DOI: 10.1002/sej

nominations were then aggregated into a count for

each industry (industries that received no nomina-

tions received a zero). These scores were compared

to the index created by the researchers, yielding a

coefﬁ cient alpha of 0.76—again suggesting very high

interrater reliability. Therefore, we normalized the

count and aggregated it with the researcher index to

create a single composite average separability score.

Not surprisingly, computers and household audio

and video equipment, which are often used as arche-

typal examples of interﬁ rm product modularity, both

scored very highly on this separability measure.

We performed a similar procedure for the next

measure, average architectural control. This measure

attempts to capture when a few ﬁ rms have signiﬁ -

cant power over the technology design or component

compatibility within an industry. To measure this,

we again used a researcher rating index (the coef-

ﬁ cient alpha between ratings by the two researchers

was 0.75), and nominations7 from the set of scholars

(the coefﬁ cient alpha between the scholar count and

the researcher index was 0.64). We again normal-

ized the latter and aggregated it with the former

to create a single measure of average architectural

control. The three industries that scored highest in

terms of architectural control were guided missiles,

aircraft and parts, and engines and turbines—all

industries that exhibit very high minimum efﬁ cient

scale and may necessitate large players for several

stages of product development, manufacturing, or

system integration.

The remaining measures are network statistics.

The measures indicate that there are wide differ-

ences in the size and structure of alliance networks

across the industries. The ﬁ rst measure—nodes—is

a count of all of the players implicated in the indus-

try alliance network, while the second—industry

ﬁ rms—is a count of the ﬁ rms in the network for

whom that industry is their primary SIC. As shown,

our networks range in size from a single industry

ﬁ rm engaged in a single alliance with a nonindustry

ﬁ rm (e.g., wood buildings and homes) to networks

with hundreds of industry ﬁ rms engaged in alliances

both within and beyond the industry. The next two

measures, number of alliances per industry ﬁ rm and

alliance participation rate, both tap into the emphasis

on alliance activity in the industry. The ﬁ rst is the

average number of alliances individual industry ﬁ rms

engage in, measured as a direct count of the number

of alliances engaged in by each industry ﬁ rm, aver-

aged over the industry. The second is a measure of

the relative rate of participation in alliances by ﬁ rms

in the industry, measured as the number of industry

ﬁ rms engaged in alliances divided by the number

of ﬁ rms in that industry worldwide, as reported

by Compustat Global.8 The average number of alli-

ances ranges from a low of 1.00 to a high of 3.75

(engines and turbines), and participation ranges

from a low of 0.05 to a high of 2.60 (for guided

missiles, where more ﬁ rms in the industry partici-

pate in alliances than there are publicly held ﬁ rms

in the industry).

Network centralization captures the degree to

which some ﬁ rms have many more alliances than

others in the industry, scaled by the maximum value

this measure can take on. It is measured as:

Max c c

max

−

()

−

()

∑∑

where cmax is the maximum degree centrality of any

node in the network, and ci is the degree centrality

of node i. This measure can take on a value of 0.00

(where all nodes have the same degree centrality)

to 100 percent for a star graph where one node is

connected to all the others nodes, but those nodes

are connected only to the center of the star. In our

networks, this measure ranges from 0.00 to values

as high as 33.26 percent (for guided missiles) and

40 percent (for the logging industry). Both of these

high centralization industries are depicted in Figure

1. The graphical visualization of the guided missile

network is particularly exquisite, showing four large

cliques connected by two main hubs (Lockheed

Martin and Thales SA).

7 In the survey, respondents were asked to nominate the 10

industries with either a) ‘. . . the highest concentration of archi-

tectural control among the ﬁ rms in the industry. Architectural

control refers to the ability of a ﬁ rm(s) to deﬁ ne speciﬁ cations

for both the individual subsystems of a product as well as the

integration of these subsystems to form the end product,’ or

b) ‘. . . the highest level of proprietary technological standards

governing the products that characterize the industry. Propri-

etary technological standards refer to interface speciﬁ cations

that are company owned.’

8 Since there is no deﬁ nitive count of the number of private

and public ﬁ rms that compete in an industry worldwide, we

divided the number of industry ﬁ rms engaged in alliances by

the number of publicly held ﬁ rms in the industry worldwide

as reported by Compustat Global. While the latter decidedly

understates the size of the industry by counting only publicly

held ﬁ rms, it provides some scaling for industry size that at

least enables us to get a sense of relative participation rates.

196 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

The next three measures pertain to small-world

properties. The ﬁ rst, harmonic mean path length,

measures path length in a disconnected network

by scaling the inﬁ nite distances (those between

nodes for which there exists no path) to the size

of the network (Newman, 2000).9 The second, the

weighted overall clustering coefﬁ cient, represents

the percentage of a ﬁ rm’s alliance partners that are

also partnered with each other, weighted by the

number of each ﬁ rm’s partners, averaged across

all ﬁ rms in the network. The third, the small-world

quotient, is a ratio of the weighted overall clustering

a) Guided missiles

) Logging

Figure 1. Highly centralized networks: guided

missiles, logging

9 The length of the path connecting two nodes, A and B, is the

minimum number of intermediate nodes that must be traversed

in reaching A from B. If no path exists between a pair of

nodes, the path length is said to be inﬁ nite. To accommodate

the fact that many of the nodes in our networks have inﬁ nite

path length, we calculate the harmonic path length by ﬁ rst

inverting each distance between every pair of nodes, averaging

all of these inverted distances, and then inverting this average.

Because the inverse of inﬁ nity is well deﬁ ned (zero), the har-

monic mean path length provides a measure of the distance

between nodes that scales for the number of pairs of nodes that

are not connected by any path.

coefﬁ cient of the network over the clustering coef-

ﬁ cient that would be expected in a random graph of

similar size and degree, divided by the ratio of the

harmonic path length of the network over the path

length that would be expected in a random graph of

similar size and degree:

Weighted overall clustering coef cient

coe

actual /

Clustering ffcient

Harmonic path length

Path length

random

actual











aandom











As shown, there is tremendous range on these mea-

sures across the industries, going from small-world

quotients of 1.00 (e.g., footwear, logging) to 27.76

(electronic components). Finally, the last column

gives the number of nodes in the largest component

of the network. This ranges from two (for networks

consisting only of disconnected dyads) to 248 (for

pharmaceuticals).

Graphical visualizations of networks

Visual depictions of alliance network structure are

relatively rare, as most researchers rely on reporting

some summary statistics of their networks, or simply

analyze the more microlevel alliance formations as

described earlier. Some notable exceptions include

Powell et al.’s (2005) ‘network movie,’ which

shows the yearly evolution of the set of collabora-

tions among multiple actors in the life sciences arena

over a 12-year period, and Rosenkopf and Padula’s

(2007) four snapshots of alliance network structure

in their study of networks in the mobile communi-

cations industry. The graphics in both studies are

complex and fascinating, yet the very feature that

enables the development of the graphics—that is,

the intensive study of one industry—also leaves the

reader wondering about the generalizability of the

results. Do other industries demonstrate these same

structures? Is the structure in Rosenkopf and Padula

similar to or different than the one Powell et al. dem-

onstrate? If so, how? Are the observable differences

due to design choices on the part of each research

team as they chose to portray their networks, or do

they reﬂ ect more substantive differences? The only

way to assess this effectively is through systematic

comparative study of multiple networks.

To this end, nine network structures are displayed

in Figure 2. The graphs are created by spring embed-

ding the nodes based on their path lengths from one

ﬁ

Comparing Alliance Network Structure Across Industries 197

DOI: 10.1002/sej

another. This process brings nodes closer together

that are directly connected or share a number of

mutual connections, while pushing apart nodes that

are not connected or are connected via a long path.

Due to space constraints, this process often results

in a ring (or crescent) of the dyadic pairs of nodes

that are not connected to any other nodes, as is most

obvious in the bottom row of networks (computers,

communications, motor vehicles). The large spider-

shaped webs in these networks represent single large

components (sets of nodes that are connected to each

other by a path).

Contrasts between the three rows are immedi-

ately apparent, and we label three network types

accordingly. The ﬁ rst row of networks (broadwoven

cotton, paper mills, leather footwear) is character-

ized by small size (between 12 and 23 members)

and very low connections among the nodes (average

network degree between 1.00 and 1.06).10 Hence,

we call networks of this type disconnected. In

sharp contrast, the bottom row (computers, com-

munications, motor vehicles) contains networks of

221 Broadwovencotton 262 Paper mills 314 Leather footwear

357 Computer and office

ment

366 Communications

ment

371 Motor vehicles and

ment

372 Aircraft and parts 281 Industrial inorganic

chemicals

384 Medical instruments

and supplies

Figure 2. Nine industry alliance networks—graphical visualizations

10 Average network degree is not shown in Table 1 due to space

constraints. The number of alliances per industry ﬁ rm differs

from average network degree in that 1) it is constrained only to

industry ﬁ rms, and 2) it counts an alliance announcement only

once irrespective of the number of parties to the alliance.

198 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

large size (between 342 and 464 members) with a

much higher level of connectivity among the nodes

(average network degree between 2.26 and 2.50).

This connectivity gives rise to an identiﬁ able main

component, and due to its shape, we call networks

of this type spiderwebs. The middle row (aircraft,

chemicals, medical instruments) contains networks

of moderate size (between 143 and 222 members)

with a moderate level of connection among the

nodes (average network degree between 1.35 and

1.89). Here, no component dominates the graph, but

many separate clusters of nodes are identiﬁ able. We

call networks of this type hybrids.

The small-world statistic appears to vary with

the size and average degree of these network types.

As size and average degree increase, so too does

the small-world statistic (zero for all three dis-

connected networks, between 6.96 and 7.84 for

hybrid networks, and between 14.53 and 22.74 for

spiderweb networks). The sources of this variation

in the small-world statistic, however, differ across

the three types. For disconnected ﬁ rms, there is no

clustering. The zero-valued small-world statistic is

generated by the zero-valued clustering coefﬁ cient.

In contrast, the hybrid and disconnected networks

exhibit higher, and largely indistinguishable, clus-

tering coefﬁ cients (between 0.34 and 0.62 for hybrid

networks and between 0.35 and 0.56 for spiderweb

networks). Here, the source of the variation in the

small-world statistic among these two network types

is the harmonic path length, which obtains a range

of moderate values (2.87 to 3.03) for spiderweb net-

works, but a range of high values (5.98 to 7.11) for

hybrid networks.

One other topic of note with regard to these net-

works types and their differences pertains speciﬁ -

cally to the main component. While the size of the

main component appears to vary with the size of

the network (between two and three for discon-

nected networks, nine and 18 for hybrid networks,

and 82 and 203 for the spiderweb networks), once

we control for the size of the network, the percent

of network nodes in the main component follows

a different pattern. Speciﬁ cally, the lowest relative

size of the main component is found in the hybrid

networks (ranging from four to 12 percent of nodes),

while the disconnected networks obtain a moder-

ate relative size (ranging from 13 to 18 percent of

nodes). Not surprisingly, the spiderweb networks

exhibit the highest relative size for the main com-

ponent, ranging from 24 to 45 percent.

It is also worth noting that the network central-

ization does not separate as neatly among the three

network categories as most of the other measures.

Speciﬁ cally, among these nine networks, central-

ization ranges from 0.0 to 6.7 percent for the dis-

connected networks; 5.5 to 11.19 percent for the

spiderwebs, and 2.6 to 9.8 percent for the hybrids

Table 2 summarizes the distinctions between

the three types of networks along the dimensions

we have examined. Though the typology relies on

arbitrary distinctions between levels of connectivity

to create categories, it provides intuition for us to

visualize seemingly different networks. In the next

section, we examine some industry characteristics

that may determine these differences.

HOW DO TECHNOLOGY

CHARACTERISTICS DETERMINE

ALLIANCE NETWORK STRUCTURE?

Alliances enable ﬁ rms to pool, exchange, and jointly

create information and other resources (Eisenhardt

and Schoonhoven, 1996; Gulati 1998), and, thus, are

an important factor in technological innovation. Col-

laborating can enable a ﬁ rm to obtain necessary skills

or resources more quickly than developing them in-

house. It is not unusual for a company to lack some

of the complementary assets required to transform a

body of technological knowledge into a commercial

product. Given time, the company can develop such

complementary assets internally. However, doing

so extends cycle time. Instead, a company may be

Table 2. A typology of alliance network structures

Type Network

size

Alliance

intensity

Small-world

statistic

Size of

main

component

Relative

size of main

component

Disconnected Low Low Low Low Medium

Hybrid Medium Medium Medium Medium Low

Spiderweb High High High High High

Comparing Alliance Network Structure Across Industries 199

DOI: 10.1002/sej

able to gain rapid access to important complemen-

tary assets by entering into strategic alliances or

licensing arrangements (Hamel, Doz, and Prahalad,

1989; Pisano, 1990; Shan, 1990; Venkatesan, 1992;

Schilling and Steensma, 2001). Second, collaborat-

ing with partners can be an important source of

learning for the ﬁ rm. Close contact with other ﬁ rms

can facilitate both the transfer of knowledge between

ﬁ rms and the creation of new knowledge that indi-

vidual ﬁ rms could not have created alone (Baum,

Calabrese, and Silverman, 2000; Liebeskind et al.,

1996; Mowery et al., 1998; Rosenkopf and Almeida,

2003). By pooling their technological resources

and capabilities, ﬁ rms may be able to expand their

knowledge bases, and do so more quickly than they

could in absence of collaboration. Third, one of the

primary reasons that ﬁ rms collaborate on develop-

ment project is to share the costs and risks of the

project. This can be particularly important when a

project is very expensive or its outcome is highly

uncertain (Hagedoorn, Link, and Vonortas, 2000).

Finally, ﬁ rms may also collaborate on a develop-

ment project when such collaboration would facili-

tate the creation of a shared standard. Collaboration

at the development stage can be an important way

of ensuring cooperation in the commercialization

stage of a technology, and such cooperation may be

crucial for technologies in which compatibility and

complementary goods are important.

In this section, we examine three technology

dimensions that we argue have a signiﬁ cant inﬂ u-

ence on the formation of alliances and, hence, the

structure of the overall network. First, we consider

how technological dynamism and uncertainty in

an industry encourage ﬁ rms to form alliances and,

thus (other things being equal) , may lead to more

alliance activity overall. Second, we examine how

the degree of separability of innovation activities

(due to, for example, product modularity) enables

coordination between ﬁ rms via alliances rather than

hierarchical integration. Third, we anticipate that the

degree to which architectural control in an industry

is governed by a small number of ﬁ rms will signiﬁ -

cantly inﬂ uence the structure of the overall alliance

network. Each of these three dimensions is discussed

in turn.

Technological dynamism and uncertainty

Rosenkopf and Tushman (1994) proposed that

rates of interorganizational linkage formation are

greatest during eras of technological ferment. When

technology is changing rapidly, ﬁ rms may make

greater use of alliances in their innovation activi-

ties. As mentioned previously, alliances can be an

important mechanism for ﬁ rms to obtain knowl-

edge or other complementary assets more quickly

than developing them in-house (Liebeskind et al.,

1996; Rosenkopf and Almeida, 2003; Schilling and

Steensma, 2001; Shan, 1990; Venkatesan, 1992).

Furthermore, to the degree that ﬁ rms are uncertain

about the direction of technological change, alliances

may become more attractive because they provide

considerable ﬂ exibility compared to in-house inte-

gration of activities. The ﬁ rm can choose between

partners that differ in their competencies, increasing

the ﬁ rm’s range of production options. Addition-

ally, because alliances are often nonexclusive and

temporary, the ﬁ rm can change its mix of partners

over time, both increasing the ﬁ rm’s ﬂ exibility and

exposing its partners to some of the discipline of

market incentives. Through an alliance, a ﬁ rm can

establish a limited stake in a venture while main-

taining the ﬂ exibility to either increase the com-

mitment at a later date or shift these resources to

another opportunity (Kogut, 1991; McGrath, 1997).

In essence, ﬁ rms can use these modes as transitional

governance forms and as a means to gain an early

window on emerging opportunities that they may

want to commit to more fully in the future (Mitchell

and Singh, 1992). Finally, alliances enable ﬁ rms to

share the risk of a venture, which can be particularly

important when a technology requires large-scale

investment or faces a highly uncertain future.

All the arguments suggest that when the indus-

try environment is characterized by dynamism and

uncertainty, ﬁ rms may be motivated to make greater

use of alliances, leading to 1) a higher proportion of

ﬁ rms in the industry being actively engaged in alli-

ances, and 2) a higher rate of alliance activity per

ﬁ rm. To explore the effects of technological dyna-

mism and uncertainty, we examine both the rate of

change of total factor productivity and the level of

research and development intensity for each of our

32 industries. Five industries obtain high11 levels of

both rate of change of total factor productivity as

well as level of R&D intensity. The convergence

of these two different indicators suggests that these

industries experienced signiﬁ cant technological

11 For each measure, we divided our set of industries into ter-

tiles: high, medium, and low. Thus, 10 industries were included

in each of the high and low categories.

200 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

dynamism and uncertainty during the study period.

These industries include: computer and ofﬁ ce equip-

ment, engines and turbines, household audio and

video equipment, motor vehicles and equipment,

and photographic equipment and supplies. Notably,

the pharmaceuticals and communication equipment

industries both had exceptionally high R&D intensi-

ties (13.77 percent and 13.52 percent, respectively),

but both had experienced declines in total factor

productivity over the 1997 to 2002 period.

Consistent with the arguments we made earlier,

all but one (photographic equipment and supplies) of

the ﬁ ve industries that rank highly on both our mea-

sures of technological dynamism also obtain high

levels of alliances per industry ﬁ rm, and all but two

(computer and ofﬁ ce equipment and photographic

equipment and supplies) obtain a high percent of

industry ﬁ rms that participate in alliances. Visual

inspection of the networks for the engines and tur-

bines, household audio and video equipment, and

motor vehicles and equipment industries in Figure 3

shows that each exhibits a hybrid or spiderweb struc-

ture. On the other end of the spectrum, there were

ﬁ ve industries in the low tertile for both measures of

technological dynamism (hydraulic cement, leather

footwear, fur, women’s and children’s outerwear,

and wood buildings), and all except one of these

(hydraulic cement) also scored in the lowest tertile

for both alliance participation rates and number of

alliances per industry ﬁ rm. Visual representations of

the leather footwear and the women’s and children’s

outerwear industry networks are shown in Figure

4. These networks are clearly of the disconnected

type.

As mentioned earlier, the pharmaceutical and

communications equipment industries both had very

high R&D intensity, but had negative changes in total

factor productivity. As shown in Figure 5, both have

very extensive alliance networks with spiderweb

properties (indeed, the pharmaceutical industry has

a signiﬁ cantly larger alliance network than any other

industry we examined). The pharmaceutical industry

makes the top tertile for alliance participation rate,

but not number of alliances per industry ﬁ rm, and

the communications equipment network makes the

top tertile for number of alliances per industry ﬁ rm,

but not for alliance participation rate.

In sum, both theory and evidence suggest the

following:

Proposition 1: Technological dynamism and

uncertainty will be positively related to the

proportion of the ﬁ rms in the industry that engage

in alliances.

Proposition 2: Technological dynamism and

uncertainty will be positively related to the

average number of alliances formed by each ﬁ rm

in the network.

a) Engines and turbines

) Household audio and video equipment

c) Motor vehicles and equipment

Figure 3. Industries ranking high for technological

dynamism: engines and turbines, household audio and

video equipment, motor vehicles and equipment

Comparing Alliance Network Structure Across Industries 201

DOI: 10.1002/sej

Separability of innovation activities

Whereas technological dynamism and uncertainty

provide motivation for ﬁ rms to pool their efforts

and break down the complexity of a technological

system into more manageable pieces, it is the sepa-

rability of innovation activities that determines the

ease or effectiveness of doing so (Baldwin and Clark,

2000; Schilling, 2000; Schilling, 2004). Tushman

and Rosenkopf (1992) distinguished between

assembled product systems and the less complex

nonassembled ones, arguing that assembled systems

engender more community activity. In some product

systems, components may require such extensive

interaction—and that interaction may be so directly

inﬂ uenced by the design or nature of that compo-

nent—that any change in the component requires

extensive compensating changes in the other compo-

nents of the system, or functionality is lost (Sanchez

and Mahoney, 1996).12 For such systems, it can be

very difﬁ cult to separate innovation activities in a

way that permits multiple ﬁ rms to act in parallel.

In other product systems, however, the components

(or processes involved in the development of those

components) are relatively independent—permitting

either sequential stages or parallel activities to be

performed by separate ﬁ rms. For example, in a study

of the development of the B-2 ‘Stealth’ bomber,

Argyres (1999) describes how four ﬁ rms (Northrop,

Boeing, Vaught, and General Electric) were able to

use advanced information technology and a set of

negotiated standards to increase the separability of

the aircraft’s design. As a result, each company was

able to assume design responsibility for a differ-

ent section of the aircraft and achieve coordination

through a shared ‘technical grammar,’ rather than

hierarchical control.

One of the key factors that can increase the sepa-

rability of innovation activities is product modular-

ity. Modularity is a continuum describing the degree

to which a system’s components may be separated

and recombined (Schilling, 2000). It refers both to

the tightness of coupling between components and

the degree to which the rules of the system archi-

tecture enable (or prohibit) the mixing and matching

of components (Baldwin and Clark, 2000). Since all

systems are characterized by some degree of coupling

(whether loose or tight) between their components,

and very few systems have components that are

completely inseparable and may not be recombined,

almost all systems are, to some degree, modular.

Some systems are, however, much more modular

than others. For example, though personal comput-

ers were originally introduced as all-in-one packages

(such as Intel’s MCS-4, the Kenback-1, the Apple II,

or the Commodore PET), they rapidly evolved into

modular systems that enable the extensive mixing

a) Leather footwear

) Women’s and children’s outerwear

Figure 4. Industries ranking low for technological

dynamism: leather footwear, women’s and

children’s outerwear

12 An excellent example of this may be seen in software

systems. In many software systems, there are thousands of

interdependent programs because of redundancies in the code,

or because of shared data. Any design change results in a

cascade of required changes in other programs, known as a

‘ripple effect’ (Fichman and Kemmerer, 1993). Because of

this, many stakeholders in this industry are advocating the

adoption of object-oriented programming, despite the major

investment in new skills and new systems this will require

for many ﬁ rms. With object-oriented programming, software

modules are designed to be encapsulated so that they do not

require the sharing of data, and the range of their interde-

pendencies with other modules is limited to those intended

by the interface. Encapsulation allows information within the

module to be hidden—modules can interact without requiring

full knowledge of the contents of each module.

202 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

and matching of components from different vendors.

Standardized interfaces in both hardware and soft-

ware enable many different producers to develop

compatible components with relatively little coor-

dination. In its extreme, modularity could lower the

rate of alliance formation by eliminating the need

for ﬁ rms to interact at all (e.g., when compatibil-

ity can be achieved by conformance to a perfectly

standardized nonproprietary interface), but it is far

more typical for interfaces to be imperfect or to

incorporate elements of proprietary control, requir-

ing ﬁ rms to engage in some form of negotiation and

collaboration (such as licensing, standards consor-

tiums, etc.).13

Four of the ﬁ ve industries discussed in the previ-

ous section as ranking high on technological dyna-

mism also exhibited high separability or modularity

of the technology, as ranked by both sets of raters:

computer and ofﬁ ce equipment, household audio

video equipment, motor vehicles and equipment,

and engines and turbines. As observed previously,

three of these industries (household audio video

equipment, motor vehicles, and engines and tur-

bines) demonstrate high levels for percent of indus-

try ﬁ rms that participate in alliances and number of

alliances per industry ﬁ rm. On the other extreme,

of the 10 industries ranked lowest in terms of sepa-

rability and modularity by both sets of raters, ﬁ ve

were also ranked in the low tertile for either alliance

participation rate or number of alliances per indus-

try ﬁ rm, though only one—paper mills—ranked in

the bottom tertile for both. As previously shown in

Figure 2, it exhibits a disconnected network with

only a single triad.

Based on our theoretical discussion and anecdotal

evidence, we expect that separability of innovation

activities will be positively associated with both the

proportion of ﬁ rms in the industry that are engaged

in alliance activities and the rate of alliance activity

per ﬁ rm.14

Proposition 3: Separability of innovation activi-

ties will be positively related to the propor-

tion of the ﬁ rms in the industry that engage in

alliances.

Proposition 4: Separability of innovation activi-

ties will be positively related to the average

number of alliances formed by each ﬁ rm in the

network.

Concentrated architectural control

Even in industries in which innovation activities are

highly separable, the control over the architecture of

the ﬁ nal system may be highly concentrated within

the hands of a single (or few) ﬁ rms. When an industry

is characterized by such concentrated architectural

control, this will have a signiﬁ cant inﬂ uence over its

13 Notably, the two industries in our study most commonly

cited as examples of industries with strong network externali-

ties—communications equipment and computers—exhibit low

alliance participation rates, but high rates of alliances per ﬁ rm.

It is likely that modularity reduces the need for certain ﬁ rms

to participate in alliances.

14 Cowan, Jonard, and Zimmerman (2007) model the effect

of decomposability on collaboration structure, suggesting

that decomposability increases density and decreases struc-

tural holes. Of course, their modeling environment controls

for network size, which has strong effects on these measures.

In our case, with networks of varying size across industries,

we examine the more basic measures of size (proportion) and

degree (average alliances per ﬁ rm).

a) Pharmaceutical industry

b) Communications equipment industry

Figure 5. The pharmaceutical and communications

equipment industries

Comparing Alliance Network Structure Across Industries 203

DOI: 10.1002/sej

alliance network structure. Tushman and Rosenkopf

(1992) distinguished between ‘closed’ (geographi-

cally bounded) and ‘open’ (unbounded) assembled

systems. Clearly, unbounded systems (such as

telecommunications networks) require standardized

interfaces to operate and grow effectively. The extent

to which these unbounded systems are governed by

open or proprietary standards will determine the con-

centration of architectural control. In general, con-

centrated architectural control will be more common

in industries in which a dominant standard interface

incorporates proprietary elements.

There is often considerable ambiguity in the extant

literature about what is meant by ‘open’ versus pro-

prietary systems, largely because various domains of

management research use the term ‘open’ in different

ways (Gacek and Arief, 2004). Proprietary systems

are deﬁ ned here as those based on technology that

is company owned and protected through patents,

secrecy, or other mechanisms. In the information

systems literature, the term ‘open standards’ does

not necessarily mean that the underlying technol-

ogy is unprotected. For example, the Open Group

standard-setting body deﬁ nes open systems as com-

puters and communications environments based on

de facto15 and formal interface standards, but these

standards may be proprietary in the sense that they

were developed, introduced, and are maintained

by vendors (Chau and Tam, 1997). The degree to

which rivals or complementary goods producers can

access, augment, or distribute a proprietary technol-

ogy varies along a continuum in accordance with

the degree of control imposed by the technology’s

developer. Anchoring the two ends of the contin-

uum are wholly proprietary technologies, which are

strictly protected and may be accessed, augmented,

and distributed only by their developers, and wholly

open technologies which may be freely accessed,

augmented, and distributed by anyone. If a ﬁ rm

grants or sells the right to access, augment, or dis-

tribute its technology for commercial purposes, but

retains some degree of control over the technology,

this is termed a ‘partially open’ technology. Many

technologies that are termed ‘open’ in common

parlance are actually only partially open.

When a ﬁ rm retains some proprietary control

over a technology, it may be able to exert some

degree of architectural control over the system in

which the technology is embedded (Schilling, 2000).

As pointed out by Prybeck, Alvarez, and Gifford

(1991), a ﬁ rm that possesses proprietary control over

an important component in a system can restrict

market access by offering that component only as

part of a total product system. If potential entrants to

the industry must be able to provide an entire system

(rather than just individual components), integrated

systems may act as a signiﬁ cant barrier to entry

and lower competitive intensity, particularly if one

proprietary component of the integrated system is

highly desired by customers and can be protected

from compatibility with other providers’ compo-

nents. The ﬁ rm can also control the rate at which

the technology is upgraded or reﬁ ned, the path it

follows in its evolution, and its compatibility with

previous generations. If the technology is chosen

as a dominant design, the ﬁ rm with architectural

control over the technology can have great inﬂ uence

over the entire industry. Through selective compat-

ibility, it can inﬂ uence which other ﬁ rms do well and

which do not, and it can ensure that it has a number

of different avenues from which to proﬁ t from the

platform. The literature on increasing returns sug-

gests that these dynamics will be accentuated in

industries where there are strong forces encouraging

the adoption of a single dominant design, such as in

industries that exhibit network externalities (Arthur,

1994; Schilling, 1998).

Of course, platform leadership is but one form

of architectural control. Concentrated control over

product architecture is also found in industries where

products are assembled into systems by what can be

called integrators. So, for example, automobile and

aircraft manufacturers assemble multiple subsys-

tems to produce their products, and most of these

subsystems are outsourced to specialized producers.

These integrators control the overall design of the

system, and subsystem producers must conform to

the integrator’s system requirements since the inte-

grator controls access to the end users. For complex,

capital-intensive products, the number of integra-

tors is typically limited relative to the number of

complementors, and the integrators’ access to the

end users creates a great deal of bargaining power

for the integrators (Coff, 1999). Here, the integrators

do not need to control or even master the subsystem

technologies, yet they maintain bargaining power

through design and control of the overall system

architecture.

When one or a few ﬁ rms have signiﬁ cant archi-

tectural control in an industry, they will tend to be

15 A de facto standard is deﬁ ned similarly to a dominant

design—it is the emergence of a standard through market

selection.

204 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

engaged in a disproportionately high number of alli-

ances in the industry, creating hubs that account for

a signiﬁ cant portion of the network’s overall con-

nectivity. For example, in software, the dominance

of Microsoft’s Windows gives the ﬁ rm immense

inﬂ uence over the design of software applications

for personal computers. Makers of complementary

software must work to ensure that their products

are compatible with Windows, resulting in a large

number of joint R&D alliances and licensing agree-

ments between Microsoft and other software pro-

ducers. Thus, Microsoft is a very large hub in the

software industry.

Alliance networks tend to have skewed degree

distributions because larger and more prestigious

ﬁ rms tend to attract and sustain a far greater number

of alliances than smaller or less prestigious ﬁ rms

(Stuart, 1998). Both size and visibility tend to attract

alliance partners, while size gives the larger ﬁ rm

greater resources with which to manage alliance rela-

tionships. Concentrated architectural control ampli-

ﬁ es this effect, leading to networks with extremely

skewed degree distributions.16

Notably, such hubs can also shorten a network’s

average path length. Hubs that are connected to

many other nodes contract the network by provid-

ing a short path between all of the nodes to which

they are connected. The extreme is a star graph,

where a single node in the network connects every

other node, leading to an average path length that

will always be under two. Thus, we might expect

industries with high architectural control to also

exhibit strong small-world properties. This charac-

teristic is of particular interest to many manage-

ment researchers because the average path length

of a network is often a crucial determinant of the

dynamics of the network. In networks in which infor-

mation, power, disease, etc., diffuse, the length of the

path determines how long transmission will take, the

likelihood of the transmission being completed, and

the degree to which whatever is being transmitted

retains its integrity (Schilling and Phelps, 2007).

To explore these arguments in our data, we

examine three industries that were rated high in

architectural control: aircraft and parts, communica-

tions equipment, and motor vehicles and equipment.

All three also exhibit exceptionally skewed degree

distributions (see Figure 6). We used SPSS curve ﬁ t

estimation to assess the relationship between degree

and frequency and found that all three ﬁ t a power law

function with R squares above .90 with signiﬁ cant

scale-free degree exponents. In the aircraft industry,

Boeing and Rolls-Royce plc played key hub roles.

In the communications equipment industry, the key

hubs were Alcatel SA, Oki Electric Industries, and

Nokia. In the motor vehicles and equipment indus-

try, Toyota was, by far, the largest hub, followed

by Mitsubishi Corp.17 All three of these industries

also ranked in the top 10 for small-world quotients

(7.5, 14.53, and 22.74 respectively). In fact, of the

10 industries ranked highest in terms of architectural

control, six (aircraft, pharmaceuticals, motor vehi-

cles, communications equipment, medical instru-

ments, and computer and ofﬁ ce equipment) also

make the top tertile for small-world quotient.

At the other extreme, of the 10 industries ranked

lowest in terms of architectural control, most had

very small networks with nearly uniform degree dis-

tributions (of degree one), and six made the lowest

tertile for small-world quotients. For example, both

footwear and broadwoven cotton fabric mills (pic-

tured previously in Figure 2) fall into this category:

both are rated very low for architectural control,

have nearly uniform degree distributions whereby

nearly every node has a degree of one, and exhibit

small-world quotients of zero.

In sum, we propose the following:

Proposition 5: Industries with concentrated

architectural control will have alliances with more

highly skewed degree distributions than industries

without concentrated architectural control.

16 Scale-free networks are a particular kind of highly central-

ized network with a skewed degree distribution. Speciﬁ cally,

a network is only considered to be scale free if the distribution

of links across nodes follows a power law. There are, however,

highly centralized networks that are not scale free in that the

distribution of their links across nodes does not conform to

a power law; it is even possible to have a highly centralized

network with a uniform degree distribution. For example, con-

sider a network wherein there are no redundant paths, and a

single central node has four links to four other nodes, and each

of those nodes has three links to three other nodes, and each of

those nodes has three other links to three other nodes, and so

forth. Such a network might represent a traditional organiza-

tional hierarchy, with the founder or CEO playing the role of

the ﬁ rst node. In this network, the ﬁ rst node lies on the shortest

path between many of the pairs of nodes in the network, giving

this network high centralization even though every node in the

network has the exact same number of links.

17 Notably, despite the fact that these industries exhibited scale-

free degree distributions with large hubs, none scores high for

overall network centralization, ostensibly because there are

multiple hubs rather than a single hub.

Comparing Alliance Network Structure Across Industries 205

DOI: 10.1002/sej

Proposition 6: Concentrated architectural

control will tend to be associated with small-

world properties.

DISCUSSION

Before we discuss any implications of our study,

we must acknowledge two limitations. In order to

create 32 networks systematically in a reasonable

time frame, we made some design choices to keep

the data collection demands manageable. First,

each alliance network was generated from a three-

digit SIC code and included ﬁ rms located primarily

within the SIC code, as well as their partners (who

in many cases are located in a different primary

SIC code). Thus, from a three-digit SIC perspective,

many cross-industry alliances are included. Future

research should examine whether our ﬁ ndings are

robust when considering four-digit or even two-digit

industry classiﬁ cations, and future research should

consider whether the relative breadth or narrow-

ness of any particular SIC code may affect ﬁ nd-

ings. Second, our network data is cross-sectional,

in that only three years of alliance formations are

included for each industry. Any imputation about

network evolution can only be induced from the

between-industry variation in our sample. An ideal

study would include network snapshots at several

times over the lifecycle of an industry.

These limitations notwithstanding, our work has

demonstrated that there is substantial variance in

alliance network structure across industries. Visu-

ally, notable contrasts are observable between spi-

derweb, disconnected, and hybrid types of network

structures. Furthermore, we have suggested that

several underlying technological characteristics of

the industry, such as dynamism and uncertainty,

separability, and architectural control can be use-

fully associated with these structures.

There are several implications of these ﬁ ndings,

some of which are particular to strategic entrepre-

neurship, and others of which may be more broadly

applied to the ﬁ eld of network studies. To begin,

our research shows there are signiﬁ cant differences

in the rate (and, presumably, importance) of alli-

ance activity across the industry networks, and this

leads to signiﬁ cant differences in the overall con-

nectivity of the networks. If the recent extensive

theorizing about the network acting as a medium

for the exchange of information and other resources

is correct, the network connectivity of an industry

has important implications for the alliance strategies

pursued by new entrants. For example, in indus-

tries that are disconnected, it may not matter that

much whether or how new ﬁ rms engage in alliances.

By contrast, in industries we would characterize as

having a large spiderweb network, there may be a

signiﬁ cant difference between the amount of infor-

mation and resource ﬂ ow between ﬁ rms that are

100

150

200

250

300

010203040

Degree

Frequency

Motor vehicle mfg

Communication equip mfg

Aircraft mfg

Figure 6. Degree distributions for the motor vehicles and equipment, communications equipment,

and aircraft and parts industries

206 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

connected to the main component and those that

are not. Thus, in these industries, it may be very

important that new entrants attempt to forge alli-

ances that connect them to this main component, and

better still if they can forge alliances that help them

achieve a more central position in this component.

On the other hand, the size of these networks and the

presence of large hubs tends to make these networks

highly stable over time (Schilling, 2007), lessening

the likelihood that a new entrant’s alliance strategy

will enable them to rapidly achieve a highly central

position or signiﬁ cantly alter the alliance landscape.

Thus, the entrant’s efforts to connect to the main

component act primarily as an ‘admission ticket’

(Powell et al., 1996) to have access to roughly

the same information that other participants have

access to, helping them obtain competitive parity

rather than conferring a superior position. In the

networks we have labeled hybrids, where there are

multiple signiﬁ cant clusters that are not connected

to each other, a new entrant’s alliance activity can

take on even greater signiﬁ cance. For example, if

a new entrant focuses their effort on striking alli-

ances with partners from different clusters, the

new entrant might be able to become a knowledge

broker between formerly disconnected communities

of ﬁ rms, dramatically altering the alliance landscape

(and concomitant information and resource ﬂ ow) of

the industry.

Future research should examine the entry strate-

gies utilized by ﬁ rms in various network structures,

as well as analyze their success. For example, Ahuja

and Polidoro (2003) examine both the contract terms

and the subsequent alliances formed by ﬁ rms ini-

tially entering the alliance network in the chemicals

industry. Their preliminary results suggested that

new entrants accept less attractive alliance deals,

and that these new entrants are not able to parlay

their initially low network status into more attractive

network positions. Is such a ﬁ nding generalizable to

all high-tech industries, or is it limited to industries

that are nonseparable, like chemicals? Examples of

ﬁ rms such as Electronic Arts—which accepted a

highly unfavorable alliance with Nintendo only to

parlay that into a much more attractive one with

Sega later on—raise the possibility that certain

network structures enable more movement than

others, which must be addressed by future research.

Similarly, Gawer and Cusumano (2002) have sug-

gested that in industries with concentrated architec-

tural control, three strategies exist: platform leaders,

wannabes, and complementors. For new entrants in

these industries, are all three viable strategies or are

they consigned to exploitable positions as in the

chemicals industry?

At the industry level of analysis, which industries

facilitate entrepreneurial entry, and how much of

this can be associated with network structure? Do

hybrid or spiderweb structures induce more entry?

Rosenkopf and Padula (2007) ﬁ nd that new entrants

to mobile telecommunications networks frequently

enter via multiparty deals. Again, is this a general-

izable ﬁ nding, or speciﬁ c to industries with strong

standards?

Beyond implications of various network types

for entry, it is also useful to consider how these

types may be related to performance. Schilling and

Phelps (2007) have considered effects of network

structure across 11 different industries, ﬁ nding that

network structures that exhibit both high clustering

and low average path lengths are associated with

higher levels of ﬁ rm and industry innovativeness.

Of course, studies like these are susceptible to the

endogeneity critique raised by Stuart and Sorenson

elsewhere in this issue. Nonetheless, as our methods

in these areas improve, it will become important

to more fully understand the connections between

industry performance, variance in ﬁ rm-speciﬁ c

performance, and ﬁ rm entry into these networks;

most speciﬁ cally, how these interrelated character-

istics may look very different across the structural

types.

Next, our industry-speciﬁ c ﬁ ndings may indicate

support for a weaker version of what has been called

network endogeneity (Gulati and Gargiulo, 1999).

A strict form of network endogeneity suggests that

networks are inertial over time, because as network

actors seek to form new alliances, they are enabled

or constrained by their preexisting network posi-

tions. Others have critiqued this position because it

does not provide an explanation for how nonnetwork

members might join the network (Ahuja, 2000b;

Rosenkopf et al., 2001; Rosenkopf and Padula, 2007).

To reconcile these opposing views, consider that the

aggregate network structure of an industry may be

inertial over time, even though different actors may

enter or exit the particular network positions. Since

our work highlights the technological characteristics

of the industry as determinants of network structure,

it supports the notion that network structures may be

relatively ﬁ xed over extended periods of time, while

allowing actors that are technologically advanced

(or backward) to move through these alliance

structures. Therefore, an extension of this research

Comparing Alliance Network Structure Across Industries 207

DOI: 10.1002/sej

can explore longitudinal variation in alliance

participation as well as overall network structure.

Finally, our work raises some substantial concerns

about the recent excitement about small worlds in the

social network literature. Two characteristics of this

literature merit attention. Small worlds are formally

deﬁ ned as networks that simultaneously exhibit low

average path length and high clustering, and most

researchers have focused on the main component of

a network to assess these measures (though Schilling

and Phelps (2007) represent a notable exception).

Our networks make it clear that limiting analy-

ses to the main component of a network is a risky

endeavor, even for the higher connectivity networks.

Additionally, it is theorized that small-world struc-

tures simultaneously offer social capital beneﬁ ts due

to clustering and also information transfer beneﬁ ts

stemming from low path length. Many alliance net-

works exhibit these statistical characteristics, yet

the underlying network structures generating these

characteristics can be quite different. For example,

both centralized and decentralized networks can

generate substantial small-world statistics. Yet, the

Watts (1999) characterization of small worlds quite

speciﬁ cally requires that they also be decentral-

ized—it is, after all, unsurprising to ﬁ nd that cen-

tralized networks have short path lengths. In fact, the

shortest possible path length is achieved via a star

graph whereby a single node connects all the others.

Decentralization is also a key attribute to support

the small-world theorizing about information ﬂ ow,

because the combination of decentralization and a

short path length typically means that there are many

routes for information to reach the same destination.

In contrast, in a centralized network, a single (or

few) node(s) may serve as the central connecting

point between many nodes, and these central actors

may have neither the capacity nor the motivation

or economic interest in sharing information. So in

the recent rush to classify many networks as small

worlds based on the clustering/path length ratios and

to generalize these ﬁ ndings to all industries, our data

demonstrate that many alliance networks are quite

centralized and, as such, may not offer the beneﬁ ts

that are theorized.

In summary, our focus on comparative network

structure has allowed us to demonstrate a variety of

network structures and speculate as to the determi-

nants of this structure. With these issues established,

it is our hope that researchers will continue to make

strides assessing how strategic entrepreneurship can

best occur in each of these speciﬁ c network contexts.

ACKNOWLEDGEMENTS

Our work was much improved by the provocative discus-

sion at the SEJ Launch Conference and by the thought-

ful comments of Robert Burgelman. We acknowledge

funding from the Mack Center at Wharton and the

National Science Foundation, Grant No. SES-0234075.

REFERENCES

Ahuja G. 2000a. Collaboration networks, structural holes,

and innovation: a longitudinal study. Administrative

Science Quarterly 45: 425–455.

Ahuja G. 2000b. The duality of collaboration: inducements

and opportunities in the formation of interﬁ rm linkages.

Strategic Management Journal 21(3): 317–343.

Ahuja G, Polidoro F Jr. 2003. Structural homophily or

social asymmetry? The formation of alliances by poorly-

embedded ﬁ rms. Working paper, University of Michigan.

Anand BN, Khanna T. 2000. Do ﬁ rms learn to create value?

The case of alliances. Strategic Management Journal

21(3): 295–315.

Argyres NS. 1999. The impact of information technology

on coordination: evidence from the B-2 ‘Stealth’ bomber.

Organization Science 10: 162–179.

Arthur WB. 1994. Increasing Returns and Path Depen-

dency in the Economy. The University of Michigan Press:

Ann Arbor, MI.

Bae J, Gargiulo M. 2003. Local action and efﬁ cient alliance

strategies in the telecommunication industry. Working

paper, INSEAD France.

Baldwin CY, Clark KB. 2000. Design Rules (Volume

1): The Power of Modularity. MIT Press: Cambridge,

MA.

Baum JAC, Calabrese T, Silverman BS. 2000. Don’t go it

alone: alliance network composition and start-ups’ per-

formance in Canadian biotechnology. Strategic Manage-

ment Journal 21(3): 267.

Baum JAC, Shipilov AV, Rowley TJ. 2003. Where do small

worlds come from? Industrial and Corporate Change

12(4): 697–725.

Borgatti SP. 2002. NetDraw: Graph Visualization Software.

Analytic Technologies: Harvard, MA.

Borgatti SP, Everett MG, Freeman LC. 2002. Ucinet

for Windows: Software for Social Network Analysis.

Analytic Technologies: Harvard, MA.

Burt R. 1992. Structural Holes. Harvard University Press:

Cambridge, MA.

Chau PYK, Tam KY. 1997. Factors affecting the adoption

of open systems: an exploratory study. MIS Quarterly

21(1): 1–24.

Coff RW. 1999. When competitive advantage doesn’t lead

to performance: the resource-based view and stake-

holder bargaining power. Organization Science 10(2):

119–133.

208 L. Rosenkopf and M. A. Schilling

DOI: 10.1002/sej

Cowan R, Jonard N, Zimmerman JB. 2007. Bilateral col-

laboration and the emergence of innovation networks.

Management Science 53(7): 1051–1067.

Crafts NFR. 1996. Deindustrialization and economic

growth. The Economic Journal 104(434): 172–183.

Eisenhardt KM, Schoonhoven CB. 1996. Resource-based

view of strategic alliance formation: strategic and social

effects in entrepreneurial ﬁ rms. Organization Science

7(2): 136–150.

Fichman R, Kemmerer C. 1993. Adoption of software engi-

neering process innovations: the case of object orienta-

tion. Sloan Management Review 34(2): 7–22.

Gacek C, Arief B. 2004. The many meanings of open

source. IEEE Software 21(1): 34–40.

Gawer A, Cusumano MA. 2002. The elements of platform

leadership. Sloan Management Review 43(3): 51–58.

Godoe H. 2000. Innovation regimes, R&D and radical

innovations in telecommunications. Research Policy 29:

1033–1046.

Griliches Z. 1990. Patent statistics as economic indica-

tors: a survey. Journal of Economic Literature 28(4):

1661–1707.

Griliches Z, Lichtenberg F. 1983. Interindustry technol-

ogy ﬂ ows and productivity growth: a reexamination. The

Review of Economics and Statistics 66(2): 324–329.

Gulati R. 1995. Social structure and alliance formation

patterns: a longitudinal analysis. Administrative Science

Quarterly 40: 619–652.

Gulati R. 1998. Alliances and networks. Strategic Manage-

ment Journal 19(4): 293–317.

Gulati R. 1999. Network location and learning. Strategic

Management Journal 20(5): 397–420.

Gulati R, Gargiulo M. 1999. Where do interorganizational

networks come from? American Journal of Sociology

104(5): 1439–1493.

Gulati R, Westphal JD. 1999. Cooperative or controlling?

The effects of CEO-board relations and the content of

interlocks on the formation of joint ventures. Administra-

tive Science Quarterly 44(3): 473–506.

Hagedoon J, Link AN, Vonortas NS. 2000. Research

Partnerships. Research Policy 29: 567–586.

Hamel G, Doz YL, Prahalad CK. 1989. Collaborate with

your competitors and win. Harvard Business Review

67(1): 133–139.

Jorgenson DW. 1984. The role of energy in productivity

growth. The American Economic Review 74(2): 26–30.

Kogut B. 1991. Joint ventures and the option to expand and

acquire. Management Science 37(1): 19–33.

Koka BR, Madhavan R, Prescott JE. 2006. The evolution

of interﬁ rm networks: environmental effects on patterns

of network change. Academy of Management Review 31:

721–737.

Liebeskind JP, Oliver AL, Zucker L, Brewer M. 1996.

Social networks, learning, and ﬂ exibility: sourcing

scientiﬁ c knowledge in new biotechnology ﬁ rms.

Organization Science 7: 428–444.

Mariani M. 2004. What determines technological hits?

Geography versus ﬁ rm competencies. Research Policy

33: 1565–1582.

McGrath RG. 1997. A real options logic for initiating

technology positioning investments. The Academy of

Management Review 22(4): 974–996.

Mitchell W, Singh K. 1992. Incumbents’ use of pre-entry

alliances before expansion into new technical subﬁ elds

of an industry. Journal of Economic Behavior and Orga-

nization 18(3): 347–372.

Mowery DC, Oxley JE, Silverman BS. 1998. Technologi-

cal overlap and interﬁ rm cooperation: implications for

the resource-based view of the ﬁ rm. Research Policy

27: 507–524.

Nelson RR. 1988. Modelling the connections in the cross

section between technical progress and R&D intensity.

RAND Journal of Economics 19: 478–485.

Newman MEJ. 2000. Models of the small world. Journal

of Statistical Physics 101: 819–841.

Newman MEJ, Strogatz SH, Watts DJ. 2001. Random

graphs with arbitrary degree distribution and their appli-

cations. Physical Review 64: 1–17.

Nunnally J. 1978. Psychometric Theory. McGraw-Hill:

New York.

Owen-Smith J, Powell WW. 2004. Knowledge networks

as channels and conduits: the effects of spillovers in the

Boston biotechnology community. Organization Science

15(1): 5–21.

Phelps C. 2003. Technological exploration: a longitudinal

study of the role of recombinatory search and social

capital in alliance networks. Unpublished dissertation,

New York University.

Pisano GP. 1990. The R&D boundaries of the ﬁ rm: an

empirical analysis. Administrative Science Quarterly 35:

153–176.

Powell WW, Koput KW, Smith-Doerr L. 1996. Interor-

ganizational collaboration and the locus of innovation:

networks of learning in biotechnology. Administrative

Science Quarterly 41: 116–145.

Powell WW, White DR, Koput KW, Owen-Smith J. 2005.

Network dynamics and ﬁ eld evolution: the growth of

interorganizational collaboration in the life sciences.

American Journal of Sociology 110: 1132–1205.

Prybeck F, Alvarez F, Gifford S. 1991. How to price for

successful product bundling (part II). Journal of Pricing

Management 2: 16–20.

Rosenkopf L, Almeida P. 2003. Overcoming local search

through alliances and mobility. Management Science 49:

751.

Rosenkopf L, Metiu A, George V. 2001. From the bottom

up? Technical committee activity and alliance formation.

Administrative Science Quarterly 46: 748–772.

Rosenkopf L, Padula G. 2007. The micro-structure of

network evolution: an empirical investigation of alli-

ance formation in the mobile communications industry.

Organization Science Forthcoming.

Comparing Alliance Network Structure Across Industries 209

DOI: 10.1002/sej

Rosenkopf L, Tushman ML. 1994. The coevolution of tech-

nology and organization. In Evolutionary Dynamics of

Organization, Baum J, Singh J (eds). Oxford University

Press: New York; 403–424.

Sampson R. 2004. The cost of misaligned governance

in R&D alliances. Journal of Law, Economics, and

Organization 20(2): 484–526.

Sanchez R, Mahoney J. 1996. Modularity, ﬂ exibility, and

knowledge management in product and organizational

design. Strategic Management Journal 17(Winter): 63–

76.

Schilling MA. 1998. Technological lockout: an integra-

tive model of the economic and strategic factors driving

technology success and failure. Academy of Management

Review 23(2): 267–284.

Schilling MA. 2000. Toward a general modular systems

theory and its application to interﬁ rm product modularity.

Academy of Management Review 25(2): 312–334.

Schilling MA. 2004. Intraorganizational technology. In

Companion to Organizations, Baum J (ed). Blackwell

Publishers: Oxford, U.K.; 158–181.

Schilling MA. 2007. Understanding the alliance data.

Working paper, New York University.

Schilling MA, Phelps C. 2007. Interﬁ rm collaboration

networks: the impact of large-scale network structure

on ﬁ rm innovation. Management Science 53(7): 1113–

1127.

Schilling MA, Steensma K. 2001. The use of modular orga-

nizational forms: an industry level analysis. Academy of

Management Journal 44: 1149–1169.

Shan W. 1990. An empirical analysis of organizational

strategies by entrepreneurial high-technology. Strategic

Management Journal 11(2): 129–139.

Siegel D, Griliches Z. 1991. Purchased services, out-

sourcing, computers and productivity in manufacturing.

Working paper, #3678, National Bureau of Economic

Research,

Solow RM. 1957. Technical change and the aggregate pro-

duction function. The Review of Economics and Statistics

39(3): 312–320.

Sorenson O, Stuart TE. 2001. Syndication networks and

the spatial distribution of venture capital investments.

American Journal of Sociology 106: 1546–1589.

Sterlacchini A. 1989. R&D innovations, and total factor

productivity growth in British manufacturing. Applied

Economics 21: 1549–1562.

Stuart TE. 1998. Network positions and propensities to

collaborate: an investigation of strategic alliance for-

mulation in a high-technology industry. Administrative

Science Quarterly 43: 668–698.

Stuart TE. 2000. Interorganizational alliances and the per-

formance of ﬁ rms: a study of growth and innovation

rates in a high-technology industry. Strategic Manage-

ment Journal 21(8): 791–811.

Stuart TE, Ozdemir SZ, Ding WW. 2007. Vertical alli-

ance networks: the case of university-biotechnology-

pharmaceutical alliance chains. Research Policy 36:

477–491.

Terleckyj N. 1980. Direct and indirect effects of industrial

research and developments in productivity growth of

industries. In New Developments in Productivity Mea-

surement and Analysis, Kendrick JW, Vaccara BN (eds).

University of Chicago Press: Chicago, IL.

Tushman ML, Rosenkopf L. 1992. Organizational deter-

minants of technological change: towards a sociology

of technological evolution. Research in Organizational

Behavior 14: 311–347.

Uzzi B. 1997. Social structure and competition in interﬁ rm

networks: the paradox of embeddedness. Administrative

Science Quarterly 42: 35–67.

Venkatesan R. 1992. Strategic sourcing: to make or not to

make. Harvard Business Review 70(6): 98–107.

Verspagen B, Duysters G. 2004. The small worlds of strate-

gic technology alliances. Technovation 24(7): 563–571.

Walker G, Kogut B, Shan W. 1997. Social capital,

structural holes and the formation of an industry network.

Organization Science 8(2): 109–125.

Watts DJ. 1999. Networks, dynamics and the small world

phenomenon. American Journal of Sociology 105(2):

493–527.

Editorial: Social Network Analysis in Marketing: A Step-by-Step Guide for Researchers

Article

Oct 2022
IND MARKET MANAG

In a business-to-business setting, social networks comprise direct and indirect connections between firms that provide access to new information, knowledge, and resources that otherwise may not be available to the firms. Social network analysis (SNA), which refers to studying and mapping social structures through graph theory, has been widely used in many social science fields, including management. The interest in SNA is also growing among marketing scholars. This editorial discusses the main aspects of SNA and provides a step-by-step guide to researchers on how to conduct SNA in marketing, with a particular focus on the interorganizational context. The purpose of this editorial is to encourage social network research within marketing. We introduce key theoretical constructs in SNA, discuss their operationalization, and offer detailed instructions on constructing them using UCINET 6 (a common software package to implement SNA). The practical application of SNA is made on strategic alliance data collected from the Securities Data Company (SDC) Platinum database.

The interplay of international alliance and subsidiary portfolios: Implications for firms’ innovation and financial performance

Article

Sep 2023
J WORLD BUS

The benefits of having an internationally diverse alliance portfolio are well known. However, the challenges remain overlooked, especially the potential to curb firms’ international expansion beyond such alliances. Building on global connectivity literature, we study how firms’ international footprint through their foreign subsidiaries is affected by the geographical spread of their international alliances. Using data on a sample of U.S. high-tech firms, we find that this relationship follows a U-shaped pattern and is contingent on alliance portfolio geographic distance and firms’ absorptive capacity for internationalization. Deviations from the optimal international footprint lower firms’ innovation and financial performance.

The role of network community characteristics for firms' rapid business scaling

Article

Sep 2023
TECHNOL FORECAST SOC

Which factors lead to firms' massive rapid business scaling? The prior work has predominantly focused on factors inside the firm, assuming that the possession of internally controlled idiosyncratic resources and capabilities largely explains the phenomenon. We extend this work by arguing that the alliance network community in which a scaling firm is embedded represents a distinct valuable channel through which it can access and orchestrate resources needed for its rapid growth. Network communities are structural groups of firms that are densely connected internally and sparsely linked to firms in other groups. We focus on three key characteristics of network communities: size, density, and stability. We theorize that the access to and orchestration of diverse and complementary network community resources influence firm rapid scaling through the expansion, replication, and synchronization of resources and business practices. Using a sample of firms publicly traded on the North American stock exchanges and observed over 1990–2020, we find that these firms' business scaling positively relates to the size and density of their network communities, yet the relationship between firm scaling and network community stability is negative.

I Want to Marry Rich: Why Cash Makes You a Desirable Partner

Article

Full-text available

Jul 2023

Both academics and practitioners frequently argue that ‘excessive’ firm cash reserves are likely to be squandered and thus may actually be detrimental to firm performance. Contrary to this narrative, we show that large cash reserves can have strong benefits for firms. Specifically, being cash‐rich helps firms form more strategic partnerships because cash helps assure potential partners that the focal firm will be able to adapt to unfolding contingencies. This assurance is particularly relevant in industry conditions that make disruptions either more likely or more costly. We explain how such disruptions arise due to uncertainty about partners' behaviours and thereby extend the literature in the area of transaction cost economics by shifting the focus from intentional opportunism to encompass unintentional factors that can impede a firm's ability to honour partnership agreements. This shift highlights a beneficial role of excess cash that is often overshadowed by the popular narrative on its wastefulness.

The role of artificial intelligence in supply chain analytics during the pandemic

Article

Full-text available

Jan 2023

The global supply networks have been disrupted and weak connections exposed to an extent that few people have ever seen in their lifetime due to the COVID-19 epidemic. As a result of the severity of the crisis, every country and industry is feeling the effects, and the massive shifts in demand and supply that have happened throughout the epidemic are easily distinguishable from the effects of previous crises. We looked into the adaptability of alliance management and AI-driven supply chain analytics in the context of an ever-changing external environment. We examined four hypotheses in this area using survey data from the American auto components manufacturing industry. To do the analysis, we used Smart PLS. Alliance management capabilities, mediated by AI-powered supply chain analytics capacity, have been found to increase an organization's operational and financial performance. We also discovered, with environmental dynamics as a moderating factor, that alliance management capability has a substantial impact on AI-powered supply chain analytics capabilities. Based on our findings, we have a deep appreciation for the interplay between dynamic capacities and the relational view of organization. Finally, we pointed up the limitations of our study and offered a number of directions for future investigation that might help address some of the concerns that our results raise.

Functional Knowledge versus Strategic Knowledge: What Type of Knowledge Matters Most for the Long-Term Performance of Startups

Article

Full-text available

Jun 2023

Past studies have shown that the flow of knowledge from incumbent firms is associated with the creation of startups and their subsequent performance. While much research has focused on the mechanisms of how incumbent-to-startup knowledge transfer takes place, such as entrepreneurs pursuing opportunities that their previous employers do not want to pursue, we explore with detailed qualitative analysis of six private startups in the Chinese synthetic-dye industry what type of knowledge actually flows and what type is more important for the long-term success of startups. We discover eight types of knowledge that flow from incumbents to new firms during the foundation of startups. Abstracting these eight types of knowledge into two general categories of functional knowledge and strategic knowledge , we find the reception of strategic (not functional) knowledge shapes the long-term competitiveness of surviving startups. Receiving technical knowledge – one type of functional knowledge – during the founding period is necessary for startups’ short-term survival but insufficient for long-term success. Our findings show that the performance implications of initial knowledge flows from incumbents hinge on the type of knowledge, contributing to a more explicit explanation of how incumbent-to-startup knowledge flows affect entrepreneurial performance.

R&D Alliances and Open Innovation: Review and Opportunities

Chapter

Oct 2022

This chapter takes stock of the empirical literature on research and development (R&D) alliances, one among other open-innovation (OI) practices, and identifies opportunities to bring the traditions of research on R&D alliances and OI into closer alignment. Through a structured literature review, it finds a wealth of research on the antecedents and consequences of R&D alliances and their characteristics, and almost half the sampled articles study R&D alliances alongside other OI practices. The chapter establishes that the core of the R&D alliance literature is rich and fairly balanced. Yet, it also argues that productive opportunities exist for more comprehensively examining the interdependencies between R&D alliances and other OI practices in their effects on performance.

Network Ecology: Introduction to the Special Issue

Article

Dec 2023
SOC NETWORKS

Path dependence in evolving R&D networks

Article

Full-text available

Dec 2022
J EVOL ECON

Lorenzo Zirulia

This paper models the formation of R&D networks in an oligopolistic industry. In particular, it focuses on the coevolutionary process involving firms’ technological capabilities, market structure and the network of interfirm technological agreements. The main result of the paper is that the R&D network can work as a strong selection mechanism in the industry, creating ex post asymmetries among ex ante similar firms. This is due to a self-reinforcing, path-dependent process, in which events in the early stages of the industry affect firms’ survival in the long run. In this framework, both market and technological externalities created by the formation of cooperative agreements play a role. Although the R&D network creates profound differences at the beginning, which are reflected by an unequal distribution of links, it tends to eliminate them as it becomes denser and denser. The nature of the technological environment affects the speed of the transition and some of the characteristics of the industry in the long run.

Structure in context: A morphological view of whole network performance

Article

Jan 2023
SOC NETWORKS

Network perspectives in organizational research have focused primarily on how the embeddedness of actors shapes individual, or nodal, outcomes. Against this backdrop, a growing number of researchers have begun to adopt a wider lens on organizational networks, shifting the focus to collective, or whole network, performance. Yet, efforts to understand the relationship between whole network structure and whole network performance have produced conflicting findings, which suggests that a different approach may be needed. Drawing on macrostructural sociology, we propose a "whole network morphology" framework, which argues the whole network structure-performance relationship is contingent on other fundamental—relational and cultural—whole network dimensions. Subsequently, we undertake an application of our framework, through which we demonstrate how a morphological view helps address conflicting findings on the structure-performance relationship. We study 250 whole interorganizational networks known as Accountable Care Organizations (ACOs), which collectively comprise more than 44,000 healthcare organizations and 250,000 physicians. Consistent with previous work, we do not find a clear association between structural connectedness and performance. However, we find that a more disconnected network structure is associated with negative ACO performance when the relational strength of network ties is high. We also find evidence of better ACO performance in the presence of a physician cultural orientation when the whole network is more connected.

The Coevolution of Technology and Organization

Chapter

Mar 1994

The book presents the latest research and theory about evolutionary change in organizations. It brings together the work of organizational theorists who have challenged the orthodox adaptation views that prevailed until the beginning of the 1980s. It emphasizes multiple levels of change - distinguishing change at the intraorganizational level, the organizational level, the population level, and the community level. The book is organized in a way that gives order and coherence to what has been a diverse and multidisciplinary field. (The book had its inception at a conference held at the Stern School of Business, New York University, January 1992.)

Do firms learn to create value? The case of alliances

Article

Mar 2000

We investigate whether firms learn to manage interfirm alliances as experience accumulates. We use contract-specific experience measures in a data set of over 2000 joint ventures and licensing agreements, and value creation measures derived from the abnormal stock returns surrounding alliance announcements. Learning effects are identified from the effects of unobserved heterogeneity in alliance capabilities. We find evidence of large learning effects in managing joint ventures, but no such evidence for licensing contracts. The effects of learning on value creation are strongest for research joint ventures, and weakest for marketing joint ventures. These results are consistent with the view that learning effects are more important in situations characterized by greater contractual ambiguity. Copyright © 2000 John Wiley & Sons, Ltd.

Interorganizational alliances and the performance of firms: a study of growth and innovation rates in a high‐technology industry

Article

Aug 2000
STRATEGIC MANAGE J

Toby E. Stuart

This paper investigates the relationship between intercorporate technology alliances and firm performance. It argues that alliances are access relationships, and therefore that the advantages which a focal firm derives from a portfolio of strategic coalitions depend upon the resource profiles of its alliance partners. In particular, large firms and those that possess leading‐edge technological resources are posited to be the most valuable associates. The paper also argues that alliances are both pathways for the exchange of resources and signals that convey social status and recognition. Particularly when one of the firms in an alliance is a young or small organization or, more generally, an organization of equivocal quality, alliances can act as endorsements: they build public confidence in the value of an organization's products and services and thereby facilitate the firm's efforts to attract customers and other corporate partners. The findings from models of sales growth and innovation rates in a large sample of semiconductor producers confirm that organizations with large and innovative alliance partners perform better than otherwise comparable firms that lack such partners. Consistent with the status‐transfer arguments, the findings also demonstrate that young and small firms benefit more from large and innovative strategic alliance partners than do old and large organizations. Copyright © 2000 John Wiley & Sons, Ltd.

Network location and learning: the influence of network resources and firm capabilities on alliance formation

Article

May 1999
STRATEGIC MANAGE J

Ranjay Gulati

This paper presents a dynamic, firm‐level study of the role of network resources in determining alliance formation. Such resources inhere not so much within the firm but reside in the interfirm networks in which firms are placed. Data from extensive fieldwork show that by influencing the extent to which firms have access to information about potential partners, such resources are an important catalyst for new alliances, especially because alliances entail considerable hazards. This study also assesses the importance of firms’ capabilities with alliance formation and material resources as determinants of their alliance decisions. I test this dynamic framework and its hypotheses about the role of time‐varying network resources and firm capabilities with comprehensive longitudinal multi‐industry data on the formation of strategic alliances by a panel of firms between 1970 and 1989. The results confirm field observations that accumulated network resources arising from firm participation in the network of accumulated prior alliances are influential in firms’ decisions to enter into new alliances. This study highlights the importance of network resources that firms derive from their embeddedness in networks for explaining their strategic behavior. Copyright © 1999 John Wiley & Sons, Ltd.

Do firms learn to create value? The case of alliances

Article

Mar 2000
STRATEGIC MANAGE J

We investigate whether firms learn to manage interfirm alliances as experience accumulates. We use contract‐specific experience measures in a data set of over 2000 joint ventures and licensing agreements, and value creation measures derived from the abnormal stock returns surrounding alliance announcements. Learning effects are identified from the effects of unobserved heterogeneity in alliance capabilities. We find evidence of large learning effects in managing joint ventures, but no such evidence for licensing contracts. The effects of learning on value creation are strongest for research joint ventures, and weakest for marketing joint ventures. These results are consistent with the view that learning effects are more important in situations characterized by greater contractual ambiguity. Copyright © 2000 John Wiley & Sons, Ltd.

Alliances and networks

Article

Apr 1998
STRATEGIC MANAGE J

Ranjay Gulati

This paper introduces a social network perspective to the study of strategic alliances. It extends prior research, which has primarily considered alliances as dyadic exchanges and paid less attention to the fact that key precursors, processes, and outcomes associated with alliances can be defined and shaped in important ways by the social networks within which most firms are embedded. It identifies five key issues for the study of alliances: (1) the formation of alliances, (2) the choice of governance structure, (3) the dynamic evolution of alliances, (4) the performance of alliances, and (5) the performance consequences for firms entering alliances. For each of these issues, this paper outlines some of the current research and debates at the firm and dyad level and then discusses some of the new and important insights that result from introducing a network perspective. It highlights current network research on alliances and suggests an agenda for future research.© 1998 John Wiley & Sons, Ltd.

Don't go it alone: alliance network composition and startups' performance in Canadian biotechnology

Article

Mar 2000
STRATEGIC MANAGE J

We combine theory and research on alliance networks and on new firms to investigate the impact of variation in startups’ alliance network composition on their early performance. We hypothesize that startups can enhance their early performance by 1) establishing alliances, 2) configuring them into an efficient network that provides access to diverse information and capabilities with minimum costs of redundancy, conflict, and complexity, and 3) judiciously allying with potential rivals that provide more opportunity for learning and less risk of intra‐alliance rivalry. An analysis of Canadian biotech startups’ performance provides broad support for our hypotheses, especially as they relate to innovative performance. Overall, our findings show how variation in the alliance networks startups configure at the time of their founding produces significant differences in their early performance, contributing directly to an explanation of how and why firm age and size affect firm performance. We discuss some clear, but challenging, implications for managers of startups. Copyright © 2000 John Wiley & Sons, Ltd.

The duality of collaboration: inducements and opportunities in the formation of interfirm linkages

Article

Mar 2000
STRATEGIC MANAGE J

Gautam Ahuja

I argue that the linkage‐formation propensity of firms is explained by simultaneously examining both inducement and opportunity factors. Drawing upon resource‐based and social network theory literatures I identify three forms of accumulated capital—technical, commercial , and social —that can affect a firm’s inducements and opportunities to form linkages. Firms possessing these capital stocks enjoy advantages in linkages formation. However, firms lacking these accumulated resources can still form linkages if they generate a radical technological breakthrough. Thus, I identify paths to linkage formation for leading as well as peripheral firms. I test these arguments with longitudinal data on technical collaborative linkages in the global chemicals industry. Copyright © 2000 John Wiley & Sons, Ltd.

Design Rules: The Power of Modularity

Book