NETWORK STRUCTURE OF ADVANTAGE · 2020. 1. 6. · 2. Network Structure of Advantage, 2013-07-10 DRAFT, Page 4 phrases to reference whole paragraphs of text with which colleagues are

NETWORK STRUCTURE OF ADVANTAGE

July, 2013 © Ronald S. Burt* University of Chicago Booth School of Business, Chicago, IL 60637

Tel: 312-953-4089, [email protected]

*This is a draft chapter for a book tentatively titled “Structural Holes in Virtual Worlds.” I am grateful to the Booth School of Business for financial support during my work on the chapter, and support for the work from Oxford University’s Centre for Corporate Reputation. The text benefitted from discussion at the 2012 meeting of the Strategy Research Initiative at Columbia University, a 2013 workshop at Bocconi University, and particularly from discussion with Martin Kilduff, with whom I co-authored a review of social network analysis for the Annual Review of Psychology, some of which appears here (Burt, Kilduff, and Tasselli, 2013). This chapter also contains portions of a paper scheduled to appear in Research in the Sociology of Organizations (Burt and Merluzzi, 2013).

2. Network Structure of Advantage, 2013-07-10 DRAFT, Page 2

Two Network Structure of Advantage

This is a review of how certain network structures create advantage. I focus on replicated research results (occasionally presenting new results that link replicated results) concerning the two core principles about network advantage: brokerage and closure. Brokerage is about innovation and growth, expanding into new rewarding activity and perspectives. The argument is that information and practice become sticky within dense social clusters such that network brokers — the people who connect across the structural holes between clusters — have information breadth, timing, and arbitrage advantages that make brokers more likely and able to detect and develop rewarding opportunities. The brokerage principle is that achievement is fostered by access to structural holes. Closure is about governance and the status quo; maintaining stability and safety while people get better at what they already do. The argument is that dense communication channels in a closed network make it more likely that behavior and opinion inconsistent with standards in the network will be detected and discussed. With detection and discussion more likely, reputations emerge and bad behavior is less likely within the network, which lowers the risk of trust, thereby increasing the probability of trust. The closure principle is that trust and reputation are fostered within closed networks. The two principles are closely linked in that the achievements associated with brokerage require the trust and reputation associated with closure. The audience of people asked to accept a broker’s information have to accept the would-be broker as a reputable source. Thus the duality of network advantage: new ideas and growth come from people who bridge across groups, while system governance via trust and reputation comes from closed networks within groups. Closure mechanisms of bandwidth versus echo are discussed, along with network-advantage implications for leadership, and contingencies more generally.

The unit of analysis is a person, ego, surrounded by a network of contacts within a

broader market or organization. This image of ego in her network is often discussed as

an “ego-network” (Wellman, 1993), but was initially discussed as a “social atom,” the

minimum image that locates the individual in the surrounding social. Here is Jacob


Moreno (1937:213), the father of American network analysis, writing early in his work on

sociometry (cf. Figure 2.1 below): “. . . we arrive at the concept of the psychological

geography of a community. Viewing the detailed structure of a community we see the

concrete position of every individual in it, also, a nucleus of relations around every

individual which is ‘thicker’ around some individuals, ‘thinner’ around others. This

nucleus of relations is the smallest social structure in a community, a social atom.” My

focus in this chapter is how the network around ego creates for her advantage. Network

forms associated with advantage constitute social capital (Coleman, 1988; Burt, 1992;

Portes, 1998; Putnam, 2000; Lin, 2002; Burt, 2005), but I here put aside the social

capital abstraction to speak simply in terms of advantage. The gist of the story in this

chapter is that network structure can be studied as a proxy for the distribution of variably

sticky information in a population, the network around ego indicates her advantaged or

disadvantaged access and control in the distribution, then ego acting on her advantage

is rewarded with recognition, compensation, and promotion for moving otherwise

unknown or misunderstood information to places where it has value. I begin with

information foundations then turn to argument and evidence on advantage.

INFORMATION FOUNDATIONS Network models of advantage use structure as an indicator of how information is

distributed in a system of people. The models build on two facts established in social

psychology during the 1940s and 1950s (e.g., Festinger, Schachter, and Back, 1950;

Katz and Lazarsfeld, 1955; Coleman, Katz, and Menzel, 1957): (1) People cluster into

groups as a result of interaction opportunities defined by the places where people meet;

the neighborhoods in which they live, the organizations with which they affiliate, the

projects in which they are involved. (2) Communication is more frequent and influential

within than between groups such that people in the same group develop similar views of

the history that led to today, similar views of proper opinion and behavior, similar views

of how to move into the future. People tire of repeating arguments and stories

explaining why they believe and behave the way they do. They make up short-hand


phrases to reference whole paragraphs of text with which colleagues are familiar.

Jargon flourishes. Not only jargon, but a whole system of phrasing, opinions, symbols

and behaviors defining what it means to be a member of the group. Beneath the

familiar arguments and experiences labeled are new, emerging arguments and

experiences awaiting a label, the emerging items more understood than said within the

group. What was once explicit knowledge interpretable by anyone becomes tacit

knowledge meaningful only to insiders. With continued time together, new

combinations and nuances emerge to make the tacit knowledge more complex, making

knowledge more difficult to move to other groups. Information in the group becomes

“sticky” (von Hippel, 1994). Much of what we know is not easily understood beyond the

colleagues around us. Explicit knowledge converted into local, tacit knowledge makes

information sticky such that holes tear open in the flow of information. These holes in

the social structure of communication, or more simply “structural holes” (Burt, 1992), are

missing relations that inhibit information flow between people.

Figure 2.1 illustrates the resulting network image as a “sociogram” (Moreno, 1934)

of individuals variably connected as a function of prior contact, exchange, and attendant

emotions. Lines indicate where information flows more routinely, or more clearly,

between people represented by the dots. Solid (dashed) lines indicate strong (weak)

flow. Figure 2.1 is adapted from (Burt, 2005:14) where discussion of the figure can be

found in more breadth and detail. The defining feature in Figure 2.1 is clusters

demarked by line density greater within clusters than between. Within a cluster, people

share certain explicit and implicit understandings, the knowledge sticky to their cluster.

Empty space between clusters indicates a structural hole. The structural hole between

two groups need not mean that people in the groups are unaware of one another. It

means only that people focus on their own activities over the activities of people in the

other group. A structural hole is a buffer, like an insulator in an electric circuit. People

on either side of the hole circulate in different flows of information. Structural holes are

the empty spaces in social structure. We know where a hole is by where it is not. This

is not to say that people in the same cluster always think the same, or that people

separated by a structural hole always think differently. But when significant differences


in understanding occur, they are more likely between people in separate clusters than

between people in the same cluster. Structural holes define non-redundant sources of

information, sources that are more additive than overlapping.

——— Figure 2.1 About Here ———

An attractive feature of the network-information link is that network models of

advantage are easy to move across levels of analysis. The people in Figure 2.1 cluster

into groups, but the clusters themselves cluster into three macro clusters — one to the

northwest, one to the northeast, and one to the southeast. The three macro clusters

could be organizations, each containing groups of people coordinated around a central

cluster of senior people (indicated by dense areas toward the center of Figure 2.1), and

the cluster-link between the northeast and southeast clusters would contain people in a

joint venture. Or, the dots in Figure 2.1 could be organizations. The three macro

clusters then would be markets, or “institutional fields,” in which individual organizations

cluster in market niches around a central cluster of typical organizations, and the

cluster-link between the northeast and southeast clusters would contain organizations in

a hybrid market (DiMaggio and Powell, 1983; Powell et al., 2005; Padgett and Powell,

2012). The dots in Figure 2.1 could just as well be communities. The three broad

clusters then would be geographic regions in which individual cities are variably linked

as satellites around three hub cities (e.g., Eagle, Macy, and Claxton, 2010). I focus

here on individual people as the dots in Figure 2.1, but the network mechanisms to be

described generalize across levels of analysis.

NETWORKS ACROSS CLUSTERS: BROKERAGE, CREATIVITY, AND ACHIEVEMENT

People can play either of two roles in Figure 2.1: specialize within a cluster (closure), or

build bridges between clusters (brokerage). Closure is about strengthening connections

within a cluster to remove information differences and so better focus on what we

already know. I will return to closure in a moment.


Brokerage is about the benefits of exposure to variation in opinion and behavior,

benefits provided by connecting across structural holes to engage diverse information

that will take us beyond what we know. Disconnected people are more likely than

connected people to operate with different ideas and practices. The more disconnected

the contacts in a network, the more likely the network spans structural holes. People

whose contacts are all in the same group know only their own group’s opinion and

practice. People whose relationships connect across structural holes (call the people

network brokers, connectors, hubs, or entrepreneurs) are exposed to the diversity of

surrounding opinion and behavior so they are more likely to detect productive new

combinations of previously segregated information, and more likely to see alternative

sets of people whose interests would be served if the new combination were brought to

fruition.1

Robert and James in Figure 2.1 illustrate the difference provided by connections

across clusters. The two men have the same number of contacts, six strong ties and

one weak tie, but different structures surround them. James is connected to people

within group B, and through them to friends of friends all within group B. Like James,

Robert is tied through friends of friends to everyone within group B. In addition,

Robert's link with contact 7 is a network bridge connection for information from group A,

and his link with 6 is a bridge for information from group C.

Relative to James, Robert is three ways advantaged by his network: information

breadth, timing, and arbitrage. With respect to breadth, Robert's bridge relations give

him access to less redundant information. With respect to timing, Robert is positioned

at a crossroads in the flow of information between groups, so he will be early to learn

1Several network concepts emerged in the 1970s on the advantages of bridges: Granovetter

(1973) on weak ties (when they are bridges across clusters), Freeman (1977, 1979) on network centrality as a function of being the connection between otherwise disconnected people, Cook and Emerson (1978; Cook et al., 1983) on the advantage of having alternative exchange partners, Burt (1980) on the advantage of disconnected contacts, later discussed as access to structural holes (Burt. 1992), and Lin, Ensel, and Vaughn (1981) on the advantage of distant, prestigious contacts, later elaborated in terms of having contacts in statuses diverse and prominent (Lin, 2002). Application of these models to predict performance differences in representative cross-sections of managers began in earnest in the 1980s and 1990s, encouraged by earlier images of boundary-spanning personnel (Allen and Cohen, 1969; Aldrich and Herker, 1977; Tuchman, 1977, with Brass, 1984, a key transition showing the empirical importance of the more general network concept).


about activities in the three groups, and often be the person introducing to one group

information on another. Robert is what early diffusion research identified as an opinion

leader, a person responsible for the spread of new ideas and behaviors (Katz and

Lazarsfeld, 1955, on opinion leaders; Burt, 1999, on opinion-leaders as network

brokers). Third, Robert is more likely to know when it would be rewarding to bring

together separate groups, which gives him disproportionate say in whose interests are

served when the contacts come together. More, the structural holes between his

contacts mean that he can broker communication while displaying different beliefs and

identities to each contact. A certain amount of self interest can be expected, but there

is much more: Opinions and behaviors within a group are often expressed in a local

language, a dialect fraught with taken-for-granted assumptions shared within a group.

The local language makes it possible for people in the group to exchange often-

repeated data more quickly. The more specialized the language within groups,

however, the greater the difficulty in moving ideas between groups. Robert's

connections across social clusters give him an advantage in translating opinion and

behavior familiar in one group into the dialect of a target group. People who connect

across structural holes (call those people network brokers, connectors, hubs, or

entrepreneurs) are exposed to the diversity of opinion and behavior in the surrounding

organization and market. Such people are presented with opportunities to coordinate

people otherwise disconnected, and derive ideas or resources from exposure to

contacts who differ in opinion or the way they behave. Thus, a structural hole is a

potentially valuable context for action, brokerage is the action of coordinating across the

hole with bridge connections between people on opposite sides of the hole, and network

entrepreneurs, or more simply, brokers, are the people who build the bridges. Network

brokers operate somewhere between the force of corporate authority and the dexterity

of markets, building bridges between disconnected parts of markets and organizations

where it is valuable to do so. Relations with contacts in otherwise disconnected groups

provide an advantage in detecting and developing rewarding opportunities.


Distinguishing Network Brokers Figure 2.2 illustrates network metrics often used to distinguish the brokers in a network.

The computations are simple, typically described in introductory works, and social

network analysis software is readily available.2 Ego’s contacts are indicated by grey

circles in Figure 2.2. Lines indicate connections between contacts (here a simple 0,1

binary measure, but the measures all easily handle continuous measures of connection

strength). To keep the sociograms simple, ego’s relations with each contact are not

presented.


A network is closed to the extent it is small (providing few contacts that could be

separated by a structural hole) and the contacts in it are interconnected (indicating that

the contacts are already coordinating with each other). In Figure 2.2, network size (also

discussed as “degree”) increases down the figure, from networks of three contacts at

the top, to networks of five, to networks of ten at the bottom. Connectivity between

contacts increases from left to right, from networks at the left in which none of ego’s

contacts are connected (labeled “broker networks”), to the networks on the right in

which all of ego’s contacts are connected (labeled “clique networks”). Network density

is the average strength of connection between ego’s contacts, which in Figure 2.2 is the

number of connections divided by the number possible (multiplied by 100 to be a

percentage). Density is zero for networks in the left column, where no contact is

connected with others. Density is 100 for networks in the right column, where every

contact is connected with every other.

2There are general and specialist introductions to social network analysis (Borgatti and Foster

2003, Cross & Parker 2004, Borgatti et al. 2009, Kilduff & Brass 2010, Kadushin 2012, Rainie & Wellman 2012, Prell 2012, Knoke 2012), Freeman’s (2004) history of SNA development through the 20th century, introductions to network computations (Scott 2000, Hanneman & Riddle 2005, Hansen, Shneiderman and Smith, 2011), advanced introductions to computations (Wasserman and Faust, 1994; de Nooy, Mrvar and Batagelj, 2005; Carrington, Scott and Wasserman, 2005), textbooks providing an integrative view for people at the rich interface between computer science and the social sciences (Goyal, 2007; Jackson, 2008; Easley and Kleinberg, 2010; Newman, 2010), and encyclopedic handbooks covering topics introductory through sophisticated reviews(Scott and Carrington, 2011). Readers interested in data strategies can go to the work cited above, or more specifically to Marsden (2011) on sociometry, Russell (2011) on social media. With respect to software, UCINET (Borgatti, Everett, and Freeman, 2002) and Pajek (de Nooy, Mrvar, and Batagelj, 2005) are widely used, but many useful software options can be found at the INSNA website (http://www.insna.org). In this book, I often use the free software NetDraw to compute indices and draw sociograms (Borgatti, 2002).


A second way contacts can be connected, closing the network around ego, is by

mutual connection with a central person other than ego. This is illustrated by the

“partner networks” in the middle column of Figure 2.2. Partner networks are a

substantively significant kind of closure useful in detecting diversity problems in a

population (Figure 2.9c below). The middle column networks in Figure 2.2 are

characterized by no connections between contacts except for all being connected with

contact A. The networks are centralized around A, making contact A ego’s “partner” in

the network. This kind of network is detected with an inequality measure, such as the

Coleman-Theil disorder measure in the third row of each panel in Figure 2.2 (Burt,

1992:70-71). Hierarchy varies with the extent to which connections among ego’s

contacts are all with one contact. There is zero hierarchy when contacts are all

disconnected from one another (first column in Figure 2.2) or all connected with each

other (third column). Hierarchy scores are only non-zero in the middle column. As

ego’s network gets larger, the partner’s central role in the network becomes more

obvious and hierarchy scores increase (from 7 for the three-person network, to 25 for

the five-person network, and 50 for the ten-person network).

The graph in Figure 2.2 provides a sense of the population distributions from

which manager networks are sampled. The graph plots hierarchy scores by density

scores for two thousand manager networks in six management populations. The

populations, analyzed in detail elsewhere (Burt, 2010), include stock analysts,

investment bankers, and managers across functions in Asia, Europe, and North

America. The large, open networks of brokers are in the lower left of the graph, low in

density and low in hierarchy. Closure can involve simultaneous hierarchy and density,

but the extremes of either exclude the other. To the lower right are clique networks, in

which there is no hierarchy because all contacts are strongly connected with each other.

To the upper left are partner networks, in which density is below 50% because there are

no connections between contacts other than their mutual strong connection with ego’s

partner.

Network constraint is a summary index of closure around ego. Intuitively the

percent of ego’s network time and energy consumed by one group, constraint


decreases with extent to which ego has many contacts (size), increases with the extent

to which ego’s network is closed by strong connections among ego’s contacts (density),

and increases with the extent to which ego’s network is closed by a partner strongly

connected with all of ego’s contacts (hierarchy).3 A maximum constraint score of 100

indicates no access to structural holes (ego had no friends, or all of ego’s friends were

friends with one another). Across the networks in Figure 2.2, network constraint

increases from left to right with closure by hierarchy or density (e.g., 20 points for the

five-person disconnected network versus 65 points for the five-person clique network),

and decreases from top to bottom with increasing network size (e.g., 93 points for the

three-person clique network versus 10 points for the ten-person clique network).

Figure 2.2 includes two additional metrics often used to distinguish network

brokers. “Nonredundant contacts” is a count of ego’s contacts discounting contacts

redundant with ego’s other contacts — in essence a count of the clusters to which ego

is attached.4 For the networks of disconnected contacts in the first column of Figure

2.2, nonredundant contacts equal network size. Every contact is disconnected from the

others and so nonredundant with the others. For the clique networks in the third column

3Constraint measures the extent to which ego’s network is concentrated in a single group (Burt,

1992:54-65; 2010:294-297). Begin with a measure of the extent to which ego’s relations connect back to contact k: cek = (pek + Σj pejpjk)2, e ≠ j ≠ k, where pek is the proportion of ego’s network time and energy spent directly with contact k, pek = [zek + zke] / (Σj [zej + zje]), where summation is across ego’s contacts j, and variable zej measures the strength of connection from ego to contact j (0 ≤ zej ≤ 1), so contact-specific constraint cek varies from zero to one with the extent to which ego cannot avoid contact k, either directly (pek) or indirectly (Σq peqpqk). Network constraint is the sum of the cek for each of ego’s contacts. The sum is an index that varies from zero to one — for all but very small networks — with the extent to which ego’s network time and energy is concentrated in a single group indicating that ego has no access to structural holes. The index is ill-behaved for social isolates and maximum-density networks of three or fewer contacts. The index can exceed one in such small networks. Since such networks provide no access to structural holes, I round their constraint scores to one. Also, constraint is undefined for social isolates because proportional ties have no meaning (zero divided by zero). Some software outputs constraint scores of zero for isolates. That would mean that isolates have unlimited access to structural holes when in fact they have no access, as is apparent from the low performance scores observed for managers who are social isolates. Recode network constraint to the maximum for social isolates. As displayed in Figure 2.2, scores are multiplied by 100 to discuss constraint in terms of points of constraint.

4The index is computed as follows (Burt 1992:51-54): Begin with contact j in ego’s i’s network discounted for the strength of j’s relations with other contacts in ego’s network: 1 - Σq piqmjq, where piq is the proportional strength of ego’s network k relation with contact q (ziq/[Σk zik]), zij varies between zero and one with the strength of connection between i and j, and mjq is contact j’s marginal strength of connection with contact q [zjq/[max zjk for j relations with contacts k]). Sum across ego i’s contacts j to define ego’s number of nonredundant contacts (Σj [1 - Σq piqmjq]).


of Figure 2.2, ego has only one nonredundant contact regardless of increasing network

size, because every contact is redundant with the others. The final metric in Figure 2.2

is Freeman’s betweenness index. The index is a count of the structural holes to which

ego has monopoly access. 5 Two disconnected contacts provide one opportunity to

broker a connection. Four people disconnected from one another provide six

opportunities to broker connections. For the networks of disconnected contacts in the

first column of Figure 2.2, betweenness equals the number of possible connections

between contacts because all are disconnected (e.g., betweenness is 10.0 for the

broker network of five contacts because none of the 10 possible connections between

ego’s five contacts exist). For the clique networks in the third column of Figure 2.2,

betweenness is zero because there are no structural holes between ego’s contacts. In

the middle column of Figure 2.2, ego shares access to structural holes with her partner.

For example, ego has access to the disconnect between contacts B and C in the three-

person network, but so does contact A, so ego’s betweenness score is .5, half of one

structural hole. Ego has access to six holes between contacts in the five-person partner

network, but access is shared with the partner, so ego’s betweenness score is 3.0, half

the number of holes to which ego has access.

5Network betweenness was initially proposed as a centrality index that improved prediction in the

Bavelas-Leavitt experiments (Freeman, 1977). The index measures the extent to which none of ego’s contacts are connected in a network except through ego. The strength of connection between contacts j and k through ego is the product zejzek, where zej measures the strength of connection from ego and contact j (0 ≤ zej ≤ 1). The total connection between contacts j and k is the sum of direct connection between them, zjk, plus all indirect connections through others in ego’s network, Σi zijzik, j ≠ i ≠ k. The ratio of the j-k connection through ego, zejzek, divided by the total connection between j and k varies from one (if ego is the only connection between j and k) down toward zero (if ego provides only a small proportion of the total connection). Sum the ratio across all pairs of contacts j and k. The resulting index varies up from zero counting the pairs of ego’s contacts for whom ego is the only connection. Two cautions: (a) If Freeman’s betweenness index is used as a measure of access to structural holes, a control has to be added for network size. Freeman (1977) proposed dividing by the number of possible contacts that ego could broker, which is a function of network size. (b) Betweenness scores in Figure 2.2 are computed from ego’s direct access to structural holes. When scores are computed across people beyond ego’s network, as they often are, the index measures ego’s direct and indirect access to structural holes and the index is better interpreted as a measure of network centrality corresponding to degree or the network eigenvector (Bonacich, 1972).


Evidence of Broker Advantage Figure 2.3 presents illustrative evidence of network broker advantage. The horizontal

axis in each graph is network constraint, varying from open broker networks at the left to

closed networks at the right.

Illustrating the fact that network brokers have an advantage in detecting and

developing opportunities, Figure 2.3A is taken from an analysis of the social origins of

good ideas (Burt, 2004). The population is supply-chain managers in a large electronics

company. The discussion network around each manager defines network constraint on

the horizontal axis. Managers were asked to describe their best idea for improving the

value of the company’s supply chain organization. Two senior executives evaluated the

merit of each idea. Average evaluations vary up the vertical axis in Figure 2.3A. The

graph shows a strong negative, nonlinear association in which brokers are likely to have

their ideas evaluated as good and worth pursuing, in contrast to managers in closed

networks who are likely to have their ideas dismissed. These results are attractive for

displaying a continuous quantitative association between a person’s access to structural

holes and the acknowledged value of their ideas. More depth to the association is

available from ethnographic network studies of creativity (Obstfeld, 2005; Lingo and

O’Mahony, 2010; Leonardi and Bailey, 2011) and more authoritative evidence is

available from network analyses of archives: For example, Uzzi and Spiro (2005) show

that broadway shows are more often "hits" and received "rave" reviews during years in

which production teams are more composed of people who were bridge connections to

other production teams. Fleming, Mingo, and Chen (2007) show that more creative

patents come from inventors who co-author with collaborators disconnected from one

another (while later citations to the creative patents are more likely from people in

closed networks). Fleming and Waguespack (2007) show that people accepted as

leaders in the innovative Internet Engineering Task Force community tend to be network

brokers between technology areas. Phillips (2011) shows that original jazz was more

often produced in hub cities connecting otherwise disconnected cities in the bandleader

recording network, but was developed by receptive audiences in cities “disconnected”

from the production clusters around hubs.



The data in Figure 2.3B illustrates the fact that network brokers are compensated

for their work decoding and encoding information to move it between clusters. The

graph shows a strong negative, nonlinear association with network constraint similar to

the network association with idea quality. Discussed in detail elsewhere (Burt,

2010:26), Figure 2.3B contains stock analysts, investment bankers, and managers from

diverse functions in Asia, Europe, and North America. The vertical axis is adjusted for

controls within each management population so zero is performance typical for a

manager’s peers, with respect to which an individual manager can be performing higher

(positive z-scores) or lower (negative z-scores). For the investment bankers,

performance is measured by bonus compensation. For the stock analysts, performance

is measured by industry recognition with election to the Institutional Investor’s All

America Research Team. For the managers, performance is measured by

compensation, annual evaluations, or early promotion to higher job rank. The graph

shows a network brokers paid more than their peers, receiving more positive

evaluations and recognition than their peers, and getting promoted more quickly than

peers.

The performance association in Figure 2.3B is replicated by numerous studies

reporting performance metrics higher for network brokers (reviews in Burt, 2005, 2010).

Aral and Van Alstyne (2011) is a particularly important replication emphasizing the

information foundation for network advantage. Using data on the information content of

email traffic between people in a small headhunter organization, Aral and Van Alstyne

show that network brokers distinguished in the usual way by sociometric data do indeed

engage in diverse information exchanges (see Aral and David, 2012, for replication).

Headhunters in closed networks who exchange diverse information with their contacts

also have high performance metrics. In short, information diversity is the key factor

predicting performance, not the network. Holes in ego’s network are merely an indicator

of ego’s access to diverse information.

More, bridge relations are not equally valuable, nor always valuable. Value

depends on the surrounding interests and experience. Novel combinations of existing

opinion or practice are worthless in the wrong time and place. Network brokerage is not


a guarantee. It is a probability: Connecting across structural holes increases the risk of

productive accident — the risk of encountering a new opinion or practice not yet familiar

to colleagues, the risk of envisioning a new synthesis of existing opinion or practice, the

risk of finding a course of action through conflicting interests, the risk of discovering a

new source for needed resources.

The Mechanism More Precisely The advantage of ego’s “access to diverse information” seems to come from processing

information more than from getting information. The broker information advantages of

breadth, timing, and arbitrage are easily understood in terms of getting information.

Brokers know early about things from diverse sources and so have the early mover’s

advantage of spinning stories about what the new information means for a target

audience. But if getting information was the reason for broker advantage, then

advantage should spill over to people connected to brokers. A person whose contacts

were brokers would have better access to diverse information than would a person

whose contacts are each in a closed network. That turns out not to be true. Advantage

is concentrated in the network immediately around ego. The networks around ego’s

contacts are irrelevant to ego’s advantage. What matters for ego’s advantage are the

holes between ego’s direct contacts. In other words, advantage is not a direct result of

access to structural holes. Advantage is a by-product. Broker advantage does not

result from getting diverse information so much as it results from personal abilities

exercised and developed while engaged in diverse information. In communicating

between discordant understandings across a structural hole, network brokers become

skilled in recombinant methodology of analogy and metaphor for helping people

understand a point of view inconsistent with their own. Detailed argument and evidence

for the above paragraph are given elsewhere (Burt, 2010). The implication for personal

development is that an individual needs to directly engage diverse understandings to be

successful as a network broker (Burt and Ronchi, 2007). The summary brokerage

principle remains that people with access to structural holes have an advantage in

detecting and developing rewarding opportunities.


NETWORK LEADERSHIP Among the positive outcomes of brokerage is recognition as a leader. Network brokers

tend to be the recognized leaders in a population. The essence of leadership is

coordinating others. The leadership can involve many or few people. It can be with

respect to deep or shallow issues. It can derive from authority, energy, charisma, or

something else. It can serve diverse ends, from improving economic performance to

infusing work with meaning. Whatever its volume, depth, source, or purpose,

leadership is about coordination. A structural hole is an opportunity to coordinate

across the hole, so every hole to which a person has access is an opportunity to

exercise leadership. The link between brokerage and leadership is useful in two ways.

It is useful for distinguishing leaders in fact from leaders who are leaders only in title.

More important, linked brokerage and leadership are each a portal into better

understanding the other. Network brokerage can be better understood by studying

coordination efforts by leaders. Leadership can be better understood by shifting from

people as the unit of analysis (she is a great leader) to acts of leadership (that was a

great act of leadership). An act of leadership is the thread of behavior begun in

coordinating across a structural hole and concluding with the brokerage outcome. This

is easy to state in concept, but empirical acts of leadership co-occur and bleed into one

another’s beginning, course, and outcome. Nevertheless, acts of leadership are keyed

directly to coordination across specific structural holes. Persons as a unit of analysis, in

contrast, are heterogeneous bundles of actions, some perhaps acts of leadership, most

not. The preliminary question is whether people recognized for their leadership can be

distinguished by their access to structural holes. The answer, “yes,” is illustrated in

Figure 2.4.

Leaders in the Formal Organization The graph to the left in Figure 2.4 shows people at higher job ranks having more access

to structural holes. The data describe twelve hundred relatively senior people in four

organizations: 346 investment bankers across the world, 117 commercial bankers in the

United Kingdom, 331 Asia-Pacific sales and regional managers in a software company,


and 454 American supply chain managers in an electronics company. Network data on

the investment bankers are from annual 360 evaluations, from which network constraint

is computed, and network brokers are the bankers to the left in the graph (Burt,

2010:Chp. 4). Constraint is low to the extent that a banker has frequent and substantive

business contact with many colleagues who rarely contact one another. Network data

for the other three populations were obtained with a network survey (Burt, 2010:Chp. 3,

Appendix A). Constraint is low to the extent that a manager has strong work discussion

relations with many colleagues who have weak relations with one another.


The histograms in Figure 2.4A describe the distribution of network constraint in

three broad job ranks. The “most senior job ranks” contain people who report to the

CEO or report to someone who reports to the CEO. People in the next-lower two or

three job ranks comprise the middle histogram in Figure 2.4A. People in lower job

ranks comprise the bottom histogram.

The figure shows the distribution of network constraint shifting from more to less

closed networks up the top job ranks. The shift is not absolute. There is substantial

variation within the three job-rank categories and substantial overlap between the

separate categories. However, the center of gravity in the distributions moves from

more to less closed networks. In lowest category of job ranks, the average level of

network constraint is 56.4 and the right-end of the distribution is thick with people in

closed networks. Closed networks are less common in the middle job-rank category,

and average network constraint is a lower 41.9 (-14.21 t-test, P < .001). Continuing up

to the people in the most senior job ranks, only 6% are beyond 60 points of network

constraint, average constraint decreases to 29.5 (-6.48 t-test, P < .001), and the center

of gravity in the distribution shifts down to a modal constraint of 15 points (highest bar in

top histogram). In short, network brokers are more common in higher job ranks.

Leaders in the Informal Organization Leaders in the formal organization are easily recognized by their job titles. However,

Figure 2.4A shows considerable network variation between people in the same rank:

people with similar titles can differ widely in the extent to which they coordinate others.


Speaking more colloquially, people at the same job rank often differ in the extent to

which they hold the emotional territory around their work. Some are widely known and

respected. Some are just tolerated. Such differences refer to social standing in the

informal organization.

And Figure 2.4B shows that people with more access to structural holes tend to be

higher in the informal organization. Network constraint is measured on the horizontal

axis, network brokers to the left and closed networks to the right. Social standing in the

informal organization is measured on the vertical axis by network status. In the early

years of social network analysis, social standing was measured by choice status

(Moreno, 1934:102; Jennings, 1937, 1943). A person has high choice status to the

extent that the person is widely cited as a preferred contact — indicating ego’s

popularity and likely social influence. Choice status evolved into more sophisticated

concepts of network centrality in which choices were weighted by the social standing of

the source, a condition ultimately captured by the left-hand eigenvector of a network:

The more ego is cited by people who are themselves widely cited, the higher ego’s

status in the network.6 Eigenvector models were used extensively in the 1970s and

1980s to measure centrality and power within elite networks (e.g., Mizruchi, Mariolis,

Schwartz, and Mintz, 1986). Podolny (1993) renovated use of the measure with a new

interpretation: the eigenvector measures status, which is valuable as a signal of quality.

When quality is difficult to determine, network status can be used as a visible signal of

quality: A person or organization widely sought out by experts, who themselves are

6Building on Moreno’s (1934) choice status, Katz (1953) proposed weighting choices, and

Bonacich (1972) provides a succinct summary of the eigenvector model. Given a network of relations zji, where zji is the strength of connection from person j to person i, the status of person i is high to the extent that i has strong connections from persons j, who themselves have high status: si = Σ j zji sj. When the zji are normalized to sum to one in each row, the maximum eigenvalue is one and the si are elements in the first eigenvector of the network. Status scores are computed numerically. With status on both sides of the equation, there is no absolute value of status; it has to be defined with respect to a numeraire. Popular options are to divide by the highest score (so each person is a fraction of the maximum), or the sum of scores (so each person’s score is a proportion of the total). I use average status. Coleman (1972, 1990:Chp. 25) offers a sophisticated version of the eigenvector model in which relations are expressed as the intersection of control and interest with the eigenvector measuring actor power to define equilibrium structure (see Taylor and Coleman, 1979), but the Coleman model is sophisticated beyond most available data. Bonacich (1987) proposes a more general version of his 1972 model, but the 1972 version is sufficient for most substantive research and readily available in network analysis software. Eigenvector scores in this chapter were obtained with NetDraw (Borgatti, 2002).


widely sought out, must be of high quality. When in doubt, look for the expert to whom

experts turn. Whatever the interpretation, the eigenvector for a network measures the

extent to which an individual is the object of strong relations with many colleagues who

themselves have strong relations with many others. In Figure 2.4B, I have computed

each person’s network status within work discussion relations, using the average

person for a comparison point. Status scores are multiples of the average, fractions for

people with status below average, 1.0 for people of average status, 2.0 for people of

status twice the average, and so on.7 The graph in Figure 2.4B shows the highest

levels of status concentrated to the left in the graph, over network brokers. Moving from

left to right, across people increasingly constrained by their network, status rapidly

decreases to below-average at the right in the graph, over the closed networks.

Leadership Behavior The graphs in Figure 2.4 show that network brokers tend to be the recognized leaders

in an organization, but being recognized as a leader is not the same as being a leader.

From personal contact with many of the people in the graph, I know that some of them

exercise significant leadership in the sense that they coordinate across their

organization. At the same time, there are people in the graph whose leadership is

limited to a closed network of specialists within their organization. There is nothing right

or wrong, good or bad, about the differences in leadership. However, to support the

proposed link between leadership and access to structural holes, it would be good to

see evidence of leadership behavior associated with network brokers.

Such evidence is available from the classic Bavelas-Leavitt experiments on

leadership in small groups. The experiment evidence is less compelling than the Figure

2.4 data on actual people at the top of their organizations in that the experiment

evidence describes college students doing a contrived task in a contrived environment.

On the other hand, the experiment evidence is more compelling for its detail and clarity

(relative to the variety of unknown factors combined in the survey data used to define

discussion networks around the senior people). For this chapter, the experiment is

7See previous footnote on measuring network status.


attractive for corroborating the link illustrated in Figure 2.4 between leadership and

access to structural holes.


Figure 2.5 displays the four network structures for which the Bavelas-Leavitt

experiments are often cited (for historical context see Leavitt, 1996; Freeman, 2004:66-

71). The networks were selected to test the intuition that coordination in a group is

more effective when someone is centrally positioned to facilitate coordination. The

structural intuition came from Bavelas (1948, 1950), the coordination experiment from

Sidney L. Smith, and published results from Leavitt’s doctoral dissertation (1949, 1951).

For each of the four networks, five college students were assigned at random to a

position. Within a five-person group, each person was given a card on which five of the

following six symbols were printed: circle, triangle, star, square, plus, diamond. Each

symbol appeared on four of the five cards, but only one symbol was on all five cards.

The group coordination problem was to determine, as quickly as possible, which symbol

was on all five cards. People were able to communicate over the connections displayed

in Figure 2.5. They did not know about any connections other than the ones in which

they were personally involved. After a group solved their initial coordination problem,

they were presented with another, and another, until they had solved fifteen problems.

Each problem involved the same six symbols, but the one held in common varied from

problem to problem. A total of five groups, each composed of five students, was run

through each of the four networks in Figure 2.5 — providing data on 15 trials for 25

students in each of the 20 network positions. At the end of the experiment, each person

was asked to complete a questionnaire describing his experience.

The empirical results support Bavelas’ intuition. Networks are arranged in Figure

2.5 in order of centralization: leadership is most centralized in the WHEEL (position C

has access to six structural holes, the other four have access to none) and most

distributed in the CIRCLE (everyone has access to one structural hole). Summary

results from Leavitt (1951) are given in the table below the sociograms. Groups in the

WHEEL network solve the problem more quickly (32.0 seconds versus 50.4 for the

CIRCLE), but finish least pleased with the experience (44.4 average survey response


for people in the WHEEL on 100-point “How did you like your job in the group?” versus

65.6 average for people in the CIRCLE).

More specifically, Figure 2.6 shows outcomes by variation in a person’s access to

structural holes — highest for position C in the WHEEL, lowest for positions with only

one communication channel (positions A and E in the CHAIN, positions A, B, and E in

the Y-NETWORK, and positions A, B, D, E in the WHEEL). These are extreme

networks in that connections are all or nothing (no partial connections) and access to

each structural hole is all or nothing (no shared access). The result is that measures of

access to structural holes are highly correlated: the number of a person’s connections

equals his number of non-redundant contacts, is .95 correlated with the number of holes

to which he has access (ego-network betweenness), and -.92 correlated with his level of

network constraint (-.97 with log constraint). Connection and constraint scores for each

network position are given in Figure 2.5.

The graph in Figure 2.6A shows that the tendency for central positions to deliver

messages is concentrated in a specific kind of message — what Leavitt coded as

“answer” messages. An answer message is one in which a person proposes the

answer to the coordination problem, e.g., “Triangle is the symbol we have in common.”

The solid line in Figure 2.6A shows that people with more access to structural holes (to

the left in the graph) are more likely to deliver “answer” messages. For example, people

in position C in the WHEEL network have the highest access to structural holes (6

holes, constraint score of 25) and deliver the highest number of answer messages (64.2

messages on average). Position C people in the Y-NETWORK have the next highest

access (3 holes, constraint score of 33) and deliver the next highest number of answer

messages (44.4 on average). People in positions with no access to holes almost never

deliver an answer (constraint score of 100, 0.29 answer messages on average).8

8Average numbers of answer versus information messages for Figure 2.6A are taken from Table

30 in Leavitt’s dissertation (1949). The table contains 20 averages for answer messages and 20 averages for information messages (five groups through each of the four networks). Leavitt distinguishes other kinds of messages, but most messages are answer or information. While the lines in Figure 2.6A are a good description of differences between network positions, there is also a significant adjustment for increased activity in the CIRCLE network. Predicting the 20 means for answer messages yields a -13.08 t-test for log network constraint (more answer messages delivered by network brokers) and a -2.58 t-test for positions in the CIRCLE network (fewer answer messages sent). Predicting the 20 means for



In contrast, network brokers are not particularly active in sending what Leavitt

coded as “information” messages. An information message is something like: “My card

contains symbols circle, triangle, star, square, diamond.” The dashed line in Figure

2.6A shows that people in the positions with the most and least access to structural

holes delivered a similar number of information messages (respectively 20.6 and 19.4

information messages). The people who most often send information messages are the

people with access to a single structural hole (constraint score of 50; positions B and D

in the CHAIN, position D in the Y-NETWORK, all positions in the CIRCLE). These

people send about twice the number of information messages sent by others (51.6

messages on average), often forwarding to one party information just received from the

other.

The graph in Figure 2.6B shows that the tendency for people in central positions to

enjoy the experiment required time to develop. People randomly assigned to a

leadership position did not enjoy it initially. The summary enjoyment scores in Figure

2.5 were obtained by asking people at the end of the experiment to mark on a

continuous line between “disliked it” to “liked it” how much they liked their job in the

group (Leavitt, 1949:89; 1951:46, responses coded 0 to 100). People were also asked

to recall how they felt after each of the 15 trials. Responses were recorded 1 to 5 and

mean scores define the vertical axis in Figure 2.6B. At the end of the first trial, the

dashed line in Figure 2.6B shows that people in the positions with most access to

structural holes, people who had performed a disproportionate amount of the

coordination work, are least happy. By the end of the 15th trial, the same people are the

happiest. At the other extreme, people with no access to structural holes, people who

provided little of the group coordination, begin as the most pleased with their job

(dashed line) but finish the least happy (solid line).9

information messages yields a -2.49 t-test for log network constraint, and a 4.68 t-test for the larger number of information messages sent in the CIRCLE network.

9Average enjoyment scores for Figure 2.6B are taken from Table 29 in Leavitt’s dissertation (1949). The 20 mean scores show no significant increase for people in the CIRCLE network beyond what is predicted from their individual access to structural holes. Predicting the 20 mean enjoyment scores at the end of the first trial (dashed line in Figure 2.6B) yields a 4.39 t-test for log network constraint and a


Finally, returning to the tendency illustrated in Figure 2.4 for network brokers to be

the recognized leaders in an organization, Figure 2.6C shows who tends to be

recognized as a leader. People were asked at the end of the experiment: “Did your

group have a leader? If so, who?” With five people in a group and five groups assigned

to each network, each network position could receive a maximum of 25 citations as

group leader. The graph in Figure 2.6C shows a clear association between the citations

a person receives as group leader and the number of structural holes to which the

person’s position has access. Whatever it is the people had in mind to distinguish

leaders, the bulk of it is captured by leader access to structural holes. Note that the

people who most often delivered “information” messages are not often cited as group

leaders (average of 3.7 cites to people with access to one structural hole, constraint

score of 50). The people most often cited as leaders are the people who most often

delivered “answer” messages (average of 17.0 and 23.0 cites respectively to people in

position C of the Y-NETWORK and WHEEL network).10

In summary, the Bavelas-Leavitt experiments show that the people with access to

structural holes, network brokers, tend to (a) provide answers to coordination problems,

(b) eventually emerge the most pleased with their work, and (c) tend to be recognized

as leaders.

Figure 2.4 through 2.6 are significant for research on flexible forms of organization.

From the nineteenth century image of hierarchical bureaucracy emerged matrix

management distinguishing “solid” line supervision from “dotted” line supervision, and

beyond to more collaborative, flexible, shifting leadership (Pearce and Conger, 2003;

negligible 1.31 t-test for positions in the CIRCLE network. Predicting the 20 mean enjoyment scores at the end of the experiment (solid line in Figure 2.6B) yields a -3.49 t-test for log network constraint and a negligible 1.63 for positions in the CIRCLE network.

10Table 8 in Leavitt’s dissertation (1949) is a tabulation of leadership citations received by each position in each network. The lines in Figure 2.6C are good description of the leadership citations, but as in the previous footnote there is again a significant adjustment for the lack of recognized leadership in the CIRCLE network. Predicting the 20 sums of leader citations (solid line in Figure 2.6C) yields a -7.03 t-test for log network constraint (network brokers tend to be recognized as leaders) and a -3.12 t-test for the lack of citations to any positions in the CIRCLE network. About half of the people in CIRCLE networks did not name anyone as a leader (12 of 25). The key to be recognized as a leader is providing answers. Predicting the 20 sums of leader citations from number of answer messages sent, number of information messages sent, and a dummy variable distinguishing positions in the CIRCLE network yields a significant association with number of answer messages sent (12.10 t-test), but negligible associations with the other two predictors (-.69 and -1.39 t-tests respectively).


Contractor et al., 2012). The organizational change is especially apparent in “open”

online communities wherein productive collaboration depends on shared leadership

emerging among the participants (Fleming and Waguespack, 2007; Luther, Fiesler, and

Bruckman, 2008; Jira, Waguespack, and Fleming, 2012; Luther and Bruckman, 2013).

Of course, even in a virtual world, someone has to be on point. There can be multiple

leadership roles, but someone has to play them to get things done.

There are at least two reasons to shift to a network definition of leadership in terms

of access to structural holes. One is rigor. Broad, metaphorical distinctions between

distributed versus centralized leadership can be replaced with network measures of

access to structural holes that in Figures 2.4 and 2.6 predict the people recognized as a

leader. A second reason is clarity. Leadership can be described as a set of behaviors

and beliefs, but that is a complex route into the phenomenon because there are so

many ways individuals play their leadership roles. It is far simpler to focus on the

leadership potential of a person’s situation — access to structural holes — and the

outcome of leadership — coordination. Whatever leadership is, the Bavelas-Leavitt

experiments show that it is recognized when people coordinate answers across

structural holes. Those recognized leaders were not designated by some higher

authority; their recognition emerged over the course of the experiment.

The network focus is no more than a return to the early use of sociometric data

and choice status to identify and describe leadership (Jennings, 1943). Network

leadership is the spirit of Reagans and Zuckerman’s (2001) analysis showing higher

team performance when team members have strong network connections to colleagues

diverse in tenure, or Sparrowe, Liden, Wayne, and Kraimer’s (2001) analysis showing

higher individual performance for people more central in their team’s advice network, or

Mehra, Dixon, Brass, and Robertson’s (2006) analysis showing that people with higher

network status and more centrality in their team turn in higher performances and are

more widely recognized for leadership. The fact is that leadership exists in shades of

gray between completely distributed and completely centralized. The more an

individual has access to structural holes, then the more the individual is in a position to

coordinate, and so at risk of being recognized as a leader. A situation rich in leadership


is composed of people ready and able to coordinate across any of the diverse structural

holes to which they individually have access.

CONTINGENT ADVANTAGE The above discussion is about network advantage in terms of production. Access to

structural holes improves ego’s odds of detecting and developing opportunities.

However, benefitting from those opportunities involves an audience, a set of people who

have to accept the broker as a source of information, a purveyor of good ideas.11

Certain questions are to be expected from the audience: Is the broker known for

competence in the proposed idea? Will he look after my interests if complications arise

after accepting his proposal? How will it look to my colleagues if I accept a proposal

from the would-be broker? Is the would-be broker an appropriate source of actionable

information for a person like me? These are questions about trust, reputation, and

social propriety.

Job Rank as a Signal In the short run, people have to rely on visible signals to decide whether a would-be

broker is appropriate and trustworthy (Gambetta and Hamill, 2005; Pentland, 2008).

Two commonly used signals are job rank and network status. Job rank indicates social

standing in a formal organization: Who is in charge? Network status indicates social

standing in an informal organization: Who is the expert sought out by other experts?

Would-be brokers attractive on either signal are more likely to be accepted as brokers.

Job rank is a known contingency factor for brokerage. Table 2.1 offers illustration.

Discussion networks around supply-chain managers in a large electronics firm were

11Brokerage can be one- or two-way with respect to the audience. Academics are often one-way

brokers. A scholar brings an idea from somewhere to solve puzzles for a target audience. A familiar example for people reading this chapter would be the behavioral economists who gained prominence by using psychology content to solve puzzles for a target audience of economists. Two-way brokers negotiate exchange between multiple audiences. An example would be Kellogg's (2012) description of case managers in community hospitals providing valuable coordination between hospital lawyers and doctors. One- and two-way brokerage can involve distinct skills and behaviors, but they have in common that the broker’s value depends on successfully playing to an audience.


obtained by survey, from which annual salary is predicted. Results in Table 2.1 are

taken from a larger model in the published analysis (Burt, 2004:371). Five job ranks are

distinguished: executives, senior managers, and three lower levels of managers. With

level-three salary as a reference, average salary is $35,707 lower for level-one

managers, and $61,930 higher for people in the executive rank. Access to structural

holes has no advantage for managers in the first two ranks: Among level-one

managers, there is a negligible $1 average drop in salary for a one-point increase in

network constraint. Among level-two managers, there is a larger, but still negligible, $47

average drop in salary for a one-point increase in network constraint. Advantage begins

with level-three managers and increases to a maximum for executives: a one-point

increase in network constraint on executives is associated with a $697 decrease in

annual salary. An executive who operated as a social isolate could expect to earn a

salary less than the average level-three manager ($69,700 expected drop in salary

wipes out $61,930 average difference between executive and level-three salaries).

——— Table 2.1 About Here ———

Job-rank contingency has either of two interpretations: production capability, or

social acceptance. With respect to production, people in more senior positions do more

political, less routine, kind of work that requires understanding and coordinating the

interests of others (Burt, 1997; 2005: 156-162). Senior rank also carries bureaucratic

authority. The boss might not be competent or trustworthy, but he is certainly culpable

and in charge. “The boss asked me to . . .” is a perfectly adequate explanation to

colleagues for your acceptance of the boss as broker. More often, he is competent and

doing a kind of work likely to involve brokerage. Together, authority and less routine

work mean that brokerage is both more valuable and readily accomplished for people in

more senior job ranks, as illustrated in Table 2.1.

Network Status as a Signal

Recall the concept of network status as a signal of quality (introduced in the discussion

of Figure 2.4B). When quality is difficult to determine, network status can be used as a

visible signal of quality: When in doubt, look for the expert to whom experts turn.


Interpreted as a signal of quality, network status is related to familiar concepts of

audience reaction such as reputation and legitimacy. Reputation is what the audience

expects of the person — she known to be trustworthy, he is aggressive, she is an expert

in her field. Legitimacy is also about audience expectations, but primarily the boundary

between who is appropriate to take action versus who is not — she is board-certified to

do this kind of operation, he is out of his element here. As a network metric, status is no

more than an index of prominence in social structure, but its interpretation in terms of

quality is grounded in an audience observing the structure. The audience sees the

structure and draws inference about the higher quality of elements toward the top of the

structure. Podolny (2005:13-21) is careful to distinguish status as a network concept

from reputation as a behavioral concept, but the interpretation of status as an indicator

of quality is no less an expectation of behavior than is reputation or legitimacy. Ego is

known for her reputation. Network status is a visible characteristic of her position in a

network, from which inferences about her can be drawn. Network status is at once a

visible result of, and a source of, inference about reputation. Status is no more than a

measure of prominence in social structure, but that prominence is correlated with

various audience reactions discussed as quality, reputation, legitimacy, and other

concepts (allowing too that audiences simultaneously read other signals, Podolny,

1993:834).

Network status should be a contingency factor for brokerage much as job rank is a

contingency factor. High-status people are visible, reputable, known for ability and

integrity. Because of past trustworthy behavior responsible for high status today, or the

high status at risk of being lost tomorrow if behaved in an untrustworthy manner, a

would-be broker’s high status can allay concerns about the broker, and allay concerns

about a broker’s proposal. In contrast, low status makes a person unattractive, perhaps

illegitimate, as a broker.

In a sense, reputation is intrinsic to brokerage. Consider Nee and Opper

(2012:211) on Chinese entrepreneurs building reputation in the course of brokering

connections: “Through personal introductions and fine-grained information passed

through social networks, the ‘broker’ typically signals trustworthiness and reputation of


the prospective business partners. Moreover, it is in the broker’s interest to make good

recommendations, as most business partners will tend to reward their networking

contacts in one way or another. Such introductions can span the social gaps, or

‘structural holes’ between groups. The owner of a Ningbo-based automotive company,

for example, found her new business partner through a close friend working in the local

highway construction business. The friend introduced her to a firm in Beijing that was

looking for a reliable production partner in the Ningbo area.”

Rider (2009) offers quantitative evidence in his study of placement agents, the

people who broker connections between investors and venture funds. Across a

thousand ventures funds from 2001 to 2006, higher-status brokers have preferred

access to higher status funds (Rider, 2009:593-595). Rider (2009:578-579) goes on to

infer the contingency being discussed here: “a broker’s reputation for consistently

representing actors of high quality is a valuable, intangible asset that enables a broker

to realize future rents on the brokerage position. . . If a positive reputation reduces the

costs of assuaging potential exchange partners’ concerns, then the returns to brokerage

should be positively related to a broker’s reputation.”

Zuckerman et al. (2003) document the phenomenon in reverse. In contrast to

showing that status amplifies broker success, Zuckerman et al. show that low-status

people are more successful when they are not brokers. Using data from the Internet

Move Database on English-language films, the diversity of movie genres in which a

person acted in 1992-1994 is used to predict the person's later success finding

employment in 1995-1997. Later success is more likely for young (low status) actors

who concentrate their acting within a single genre.

A familiar illustration is provided by Merton’s (1968) discussion of the “Matthew

Effect” in science. Status autocorrelation can be explained in multiple ways (Podolny

and Phillips, 1996; Podolny, 2005:Chp. 4; Burt, 2005:Chp. 4, 2010:Chp. 6; Bothner,

Podolny, and Smith, 2011; Bothner, Kim and Smith, 2012), but Merton’s focus on status

and new ideas in science is particularly relevant to contingent returns to brokerage:

prominent scientists are more likely to have their new ideas recognized and acted upon,

which subsequently enhances prominence (cf. Podolny, 2005:Chp. 2). Merton


(1968:60) argues that ideas proposed by prominent scientists receive disproportionate

attention: “a single discovery introduced by a scientist of established reputation may

have as good a chance of achieving high visibility as a multiple discovery variously

introduced by several scientists no one of whom has yet achieved a substantial

reputation.” Disproportionate attention increases the likelihood of productive result (p.

62): “since it is probably important, it should be read with special care; and the more

attention one gives it, the more one is apt to get out of it.” Couple Merton’s discussion

with the fact that people whose networks bridge structural holes are disproportionately

the source of good ideas (illustrated in Figure 2.3A above), and one has another reason

to expect network status to be a contingency factor for brokerage.

Thus, network status is associated with network brokerage in concept, in fact, and

in effects. As job rank indicates high social standing in the formal organization

embedding a structural hole, network status indicates high social standing in the

informal organization in which a structural hole is embedded. As job rank is associated

with more access to structural holes and higher returns to brokering across holes,

network status is associated with more access to structural holes and higher returns.12


Figure 2.7 illustrates brokerage advantage contingent on status. The data to the

left describe people in a large software company shortly after launching a new product

in Asia-Pacific markets. The data to the right describe HR officers in a large American

commercial bank. Within each organization, people are categorized as high versus low

status (divided at median) and within each category, compensation on the vertical axis

(adjusted for job rank and other controls, Burt and Merluzzi, 2013), is predicted by

12I have reasoned from the perspective of an audience reacting to a broker. One could reason

instead from the broker’s perspective: Are the kinds of people drawn to brokerage also likely to achieve high status? For example, self-monitoring, a psychological concept of adapting one’s behavior to the social situation, is correlated with access to structural holes (Mehra, Kilduff, and Brass, 2001; see Burt et al., 2013, for review), and people tend to re-create the same kinds of networks across different roles (slightly more than a third of the variance in access to structural holes is consistent across roles, see Chapter 4). Given personality correlated with access to structural holes, and correlation between status and access to structural holes, status should be correlated with the personality characteristics of brokers. However, returns to brokerage seem to be independent of the network variance attributed to personality (Chapter 4), so personality-induced correlation between status and access to holes cannot explain the contingent returns to network advantage illustrated below in Figures 2.8 and 2.9. I therefore focus in the text on status and access to structural holes directly affecting the advantage that each can provide.


network constraint on the horizontal axis. I averaged compensation and network

constraint scores within five-point intervals of network constraint to define the dots in

Figure 2.7. Thin lines through hollow dots show the association between compensation

and network constraint for low-status people. Bold lines through solid dot show the

association for high-status people. The difference is striking. For high-status people,

compensation drops sharply with decreasing access to structural holes (-.96 and -.98

correlations with network constraint). For low-status people, access to structural holes

has no association with compensation (-.03 and -11 correlations).13

An Instance of Local-Structure Cue to Global Structure The severe contingency illustrated in Figure 2.7 highlights the importance to brokers of

local-structure cues to global structure. Kleinberg (2000) distinguished the existence of

bridge relations from their detection. Bridge relations are likely to exist and are easily

identified by people who have a bird’s eye view across a network. But how do people

limited to local knowledge find the bridge relations that link beyond their immediate

social circle? The problem can be solved if local structure contains cues to global

structure. Kleinberg’s (2000) analysis implies that bridges should be most readily

detected in networks of small, linked clusters, but does not go into the details of what

constitutes a local-structure cue.14

13It is clear that network status and access to structural holes are complementary assets closely

related in concept and fact, such that advantage is more clearly revealed when the two variables are analyzed together as complements defining network advantage. It is unfortunate that the two concepts have developed with so little reference to one another. There are exceptions — Podolny (2001), then recently Rider (2009), Shipilov, Li and Greve (2011), and Ferrin, Parker, Cross and Dirks (2012) — but for the most part research papers report on one or the other concept. For example, the 2012 Annual Review of Sociology contains a chapter on brokerage and a chapter on status. In the chapter on status, there is no mention of brokers, brokerage, or structural holes (Sauder, Lynn, and Podolny, 2012). Status is mentioned several times in the chapter on brokerage, but as a qualitative attribute, not as a network correlate (Stovel and Shaw, 2012). I hasten to note that Stovel and Shaw speculate about brokers achieving status, anticipating the strong status-broker association in Figure 2.7 (see pages 146 and 153-154). Saying that the two Annual Review chapters are independent says nothing negative about either chapter. The point is that network status and access to structural holes are rarely discussed together.

14These are not Kleinberg’s words, so let me quickly link the text to Kleinberg’s model. Kleinberg locates individuals in a lattice; everyone is connected to their left-right and up-down neighbors. The probability that a bridge connects ego to some person k selected at random is set to r-α, where r is the lattice distance between ego and k (1 to nearest neighbors, 2 to diagonal neighbors, etc.), and α is a clustering coefficient (α ≥ 0). Fractional values of the clustering exponent mean that local structure is a poor indicator of global structure; near and distant contacts are likely to be bridges. As the clustering


The graph connecting status and brokerage in Figure 2.4B displays for network

advantage a local-structure cue to global-structure: seeing a person behave locally as a

network broker is a cue that the person has status in the broader network (which could

be one reason why we resent people behaving like a broker locally when we know in

fact that they do not have status in the broader network). In a related vein, Everett and

Borgatti (2005) link local and global access to structural holes. They measure access

with Freeman’s (1977) betweenness index, which is a count of the structural holes to

which ego has monopoly access. Everett and Borgatti compute a local-structure

betweenness score for a person’s direct contacts and a global-structure betweenness

score for the person’s direct and indirect contacts across the broader network. (I follow

Podolny in this chapter by measuring status with the network eigenvector, which is often

discussed with global betweenness as a measure of network centrality.) Everett and

Borgatti report correlations of .88 to 1.00 between local and global betweenness scores

for several small networks taken from prior research. They report correlations of .86 to

.99 for random networks of 200 to 500 nodes. Most management study populations

contain a few hundred people with ego networks varying from zero to a few dozen

contacts, so I expect strong correlation in most management populations between

advantage indices computed from local versus global network structure.

Beyond the existence of a local-structure cue to global structure, the results in

Figure 2.7 illustrate the point that network advantage depends on the cue. Returns to

network advantage are negligible when local-structure access to holes is not associated

with global-structure status (or global-structure access to holes, borrowing from Everett

and Borgatti). Returns are most apparent when local-structure access to holes occurs

with global-structure status.

exponent increases, bridges are concentrated in near neighbors, so the network is a system of small clusters with near neighbors providing bridges to other clusters. Kleinberg (2000) shows that the quickest distribution of information occurs when the clustering exponent equals two, which concentrates bridges in near neighbors. Therefore I say in the text that Kleinberg’s model implies that bridges should be most readily detected in networks of small, linked clusters.


Reputation Is Sufficient Network status scores are readily available from a variety of data, so status is attractive

to use as a contingency factor for analyses such as the one illustrated in Figure 2.7.

But status is a visible signal with multiple interpretations. Status is correlated with

reputation as the behavior for which a person is known, but status is just a signal. A

person can have a positive reputation with many or few people. Status is by definition

about having reputation with many people, so it presumes some prior achievement. Is

reputation without status sufficient to be accepted as a broker?


I can answer “yes” for at least one population in which I have status and reputation

data. Figure 2.8 displays reputation, status, and performance outcomes for senior

investment bankers in a large financial organization. As defined in the organization,

banker reputation is measured by the average annual evaluation he or she received

from colleagues. Reputation varies in the years for which I have data from a high of

colleagues agreeing that a banker is outstanding (4.0 average evaluation), down to

agreement that he is terrible (low average evaluations). Status is measured by the

number of people evaluating a banker, weighted by their status (zji in footnote 4 is one if

colleague j evaluated banker i, zero otherwise). The graph to the left in Figure 2.8 is a

box-and-whisker plot of reputation scores on the vertical axis across increasing banker

status on the horizontal. The graph illustrates three features of the association between

reputation and status. First, increasing horizontal bars within the boxes show average

reputation increasing slightly with status (.20 correlation). Second, change in the width

of the boxes and whiskers show that variation in reputation decreases with increasing

status. High-status bankers tend to have positive reputations, so high status is a good

signal of positive reputation. But low-status bankers vary widely in reputation, from the

very worst reputations in the graph, to the most positive. Low status is less a negative

signal than it is an ambiguous signal. Reputation is both very high and very low among

low-status bankers. The third point illustrated in the graph is that reputation does not

increase to universal positive at the highest level of status. The most positive

reputations occur among the low-status bankers. As bankers rise in status prominence,

minimum reputation becomes less negative, but maximum reputation also becomes


less positive. The higher a person rises, the more likely their audience includes

disgruntled individuals.

Regardless, reputation alone is sufficient to facilitate brokerage. Brokers do not

have to be widely known, just well respected. The graph in Figure 2.8B is constructed

just like the graph in Figure 2.7B: annual compensation is regressed across network

constraint for bankers with positive versus poor reputations. The solid dots describe

compensation to the bankers with the top third of reputation scores. A bold line through

the solid dots shows compensation drops as a banker has decreasing access to

structural holes (-.74 correlation with constraint in Figure 2.8B, -3.08 t-test for individual

bankers holding network status constant). In comparison, the dashed line through

hollow dots shows no network advantage for bankers with poor reputations (bottom third

of reputation scores; -.27 correlation in the graph, -0.36 t-test for individual bankers with

status held constant).

Diagnostic Contingency Contingent brokerage creates an unobtrusive organization diagnostic with which difficult

people problems in an organization can be addressed in a rigorous, analytical way.

Given returns to brokerage contingent on trust and reputation, categories of people not

trusted in an organization can be identified by looking for places where rewards are not

enhanced by brokerage. This does not mean that everyone should always be

successful as a broker. It means that consistent failure by a category of people signals

a problem. Diagnostic methodology is presented elsewhere (Burt, 2010:Chp. 7). The

point is illustrated in Figure 2.9.

Figure 2.9A is an executive-development example. Network data were obtained

on discussion relations for managers in two upper-middle job ranks in three divisions of

a large electronics company. Network constraint scores locate managers on the

horizontal axes in Figure 2.9A. The vertical axes distinguish managers by annual

compensation. The top graph shows that compensation is higher for network brokers in

the first two of the company’s three divisions (-5.66 t-test). The bottom graph shows no

compensation association with brokerage in the third division (1.05 t-test). Further

analysis showed that the strongest predictor of compensation in the third division (after


job rank) was years of service. The longer a person worked in the division, the higher

his compensation. Years in service was associated with compensation in the first two

divisions, but network constraint was a stronger predictor.


The promotion issue surfaced because the vice-president managing the third

division complained to top management about his people not being promoted to higher

levels in the organization. Promotions went to the other two divisions. The two graphs

in Figure 2.9A explain why. Network brokers were rewarded in the first two divisions;

people developed the skills needed to exercise leadership higher in the organization

and were promoted. In the third division, network brokers received no compensation

above what was appropriate for their years of service. People in the third division were

being developed as good supervisors, not leaders. On seeing the evidence, top

management removed the third-division vice-president — an unpleasant outcome for

him, but less significant than the misfortune he left for the company in the many people

at the bottom of Figure 2.9A unprepared for higher office.

Figure 2.9B is an example of post-merger integration. The two graphs in Figure

2.9B are the same as in Figure 2.9A, except these are managers in the regional

operations of a large computer company six months after one company acquired

another. Managers are distinguished by legacy organization in Figure 2.9B. Managers

in the top graph originated in the company that made the acquisition. Network brokers

are well compensated (-4.92 t-test). Managers in the bottom graph originated in the

acquired company. There is no compensation association with brokerage (1.06 t-test).

In fact, there is an empty space in the northwest of the graph at the bottom of Figure

2.9B where the high-compensation network brokers should be. The story here is that

the merged companies both had strong cultures. Leaders in the acquiring company felt

uncomfortable giving leaders in the acquired company the discretion enjoyed before the

merger. Acquired executives were given titles, but little flexibility within the merged

operations. Seeing the way things were, network brokers from the acquired company

soon left for jobs in more welcoming organizations. Inefficient operations and poor

morale plagued the merged operations.


Figure 2.9C is a labor-diversity example. Data on discussion and collaboration

relations were obtained from a probability sample of managers in a large electronics

company. Network constraint scores locate managers on the horizontal axis of Figure

2.9C. The vertical axis distinguishes managers by whether they were promoted to their

current rank ahead of peers. The top graph describes promotions among the senior

men. Network brokers were more likely to be promoted early (-5.56 t-test). The bottom

graph shows that the same is not true for women and junior men. The positive

association with network constraint in the bottom graph of Figure 2.9C shows that

promotions for women and junior men were systematically delayed for network brokers

(3.38 t-test). Sometimes punishment comes to people who try to be brokers when they

are not accepted as such by the people whose collaboration is being brokered. This is

where the partner networks in Figure 2.2 are useful. Persons whose status does not

entitle them to be brokers form a partner network through which they achieve sponsored

access to structural holes. Detailed discussion of this example, plus examples in which

men need a partner, is available elsewhere (Burt, 2010:Chp. 7).15

NETWORKS WITHIN CLUSTERS: CLOSURE, TRUST, AND REPUTATION

The importance of trust and reputation to network advantage shifts attention to the

internal dynamics of dense social clusters. That is where trust and reputation are

15The fact that returns to brokerage are contingent on broker reputation can inform contextual

studies of brokerage. For example, Vasudeva, Zaheer, and Hernandez (forthcoming) report that returns to brokerage are higher for firms that operate out of corporatist countries. Firms in the fuel cell industry are compared for their innovativeness (measured by patent volume and citations) and the extent to which their alliance networks span structural holes. The firms are then distinguished by the extent they operate out of a “corporatist” country — which means that trustworthiness and cooperation are commonly espoused as proper behavior. Sweden, Denmark, Germany, and Japan top the corporatist culture scale used by Vasudeva et al (forthcoming:8 [page is in “articles in advance” reprint]). The United Status, United Kingdom, and Canada are at the bottom of the scale. The authors show that innovation has no association with alliances that span structural holes, unless the broker firm operates out of a corporatist country. For firms operating out of corporatist countries, innovation increases significantly with access to structural holes. Regardless of the countries in which partners operate, innovation increases with access to structural holes as long as the broker firm operates out of a corporatist country (Vasudeva et al., forthcoming:14-15). In other words, as illustrated for the bankers in Figure 2.8B, broker trustworthiness is critical for returns to brokerage.


produced, providing a governance mechanism in social networks. Argument and

evidence on the governance provided by closed networks is available elsewhere (Burt,

2005:Chps. 3-4; 2010:Chp. 6). Here I focus on closure as it generates the trust and

reputation on which returns to brokerage are dependent.

Research on closed networks was energized by Granovetter’s (1985, 1992)

argument analyzing economic transactions by the extent to which they are embedded in

a social context: “Relational” embedding refers to a transaction between people who

have history and investment with each other. “Structural” embedding refers to a

transaction between people who have many mutual contacts, i.e., people in a closed

network. The core of the closure argument is that embedding facilitates trust by

creating a reputation cost for bad behavior. To the extent that two people know each

other well (relational embedding), and have many mutual friends in the closed network

around them (structural embedding), bad behavior by either person is likely to become

known to the other. The strong relations between and around the two people create a

wide bandwidth for information flow. Knowing that bad behavior will be discovered,

each person is less likely to behave badly for fear of the reputation damage that would

result, which lowers the risk of trust between them, which increases the probability of

trust. The analogy is life in a city versus life in a village. In a village, dense connections

between people make it likely that bad behavior will be discovered and discussed, to the

detriment of the misbehaved person. As Granovetter (1992:44) put it: ‘‘My mortification

at cheating a friend of long standing may be substantial even when undiscovered. It

may increase when the friend becomes aware of it. But it may become even more

unbearable when our mutual friends uncover the deceit and tell one another.’’ With the

likelihood of bad behavior decreased, the risk of trust decreases, so the probability of

trust increases.

Thus, closed networks facilitate trust by creating a reputation cost for behaving in

an untrustworthy manner, which aligns opinion and behavior, making collaborations

possible that would be otherwise difficult or unwise. Examples abound online: eBay’s

reputation system, the gender shelter dontdatehimgirl.com, hotel industry watchdog

oyster.com. Barker (1993) provides ethnographic description of closed-network


governance in an organization (cf. Blau, 1955). Bernstein offers thick description within

a legal framework of reputation governance in the cotton industry (Bernstein, 2001, esp.

pp. 1745-1762), and in the diamond trade (Bernstein, 1992, esp. pp. 138-145) as both

evolved from transactions within a cohesive, geographically concentrated social group

to transactions across a dispersed group embedded in electronic connections. Other

widely-cited closure arguments are offered by economist Greif (1989) arguing that trust

within closed networks facilitated medieval trade in the Mediterranean, and sociologist

Coleman (1988) arguing that closed network are social capital. Coleman (1988: S98)

summarizes: ‘‘Social capital is defined by its function. It is not a single entity but a

variety of different entities having two characteristics in common: They all consist of

some aspect of social structure, and they facilitate certain actions of individuals who are

within the structure. Like other forms of capital, social capital is productive, making

possible the achievement of certain ends that would not be attainable in its absence.’’

Adapting Coleman’s social capital metaphor, political scientist Putnam (1993:167)

proposed an influential explanation for regional differences in civic government: ‘‘Social

capital here refers to features of social organization, such as trust, norms, and

networks, that can improve the efficiency of society by facilitating coordinated action.’’

Positive and Durable Relationships Figure 2.10 contains illustrative evidence of closure’s effect on relations. The data

come from four years of annual reviews within a large financial organization. Each year,

employees eligible for bonus compensation go through a roster of colleagues, cite those

with whom the employee had frequent and substantive business contact during the

year, and rate each person for the quality of working with them. Divisions could use

additional rating dimensions relevant to their business, but all included a summary

evaluation: “outstanding,” “good,” “adequate” (negative evaluation akin to a grade of C

in graduate school), or “poor” (persons receiving multiple poor evaluations were

encouraged to look for a more compatible employer). Outstanding to poor are my

synonyms for the words actually used in the evaluations. Evaluations receive numeric

values of 4 to 1, respectively. The average evaluation of a person then goes to bonus

and promotion personnel as a measure of the person’s reputation with colleagues. The


graphs in Figure 2.10 summarize 46,231 colleague evaluations by analysts and bankers

in the organization (see Burt, 2010:174, 181 for more detailed discussion of the graphs;

see Rivera et al., 2010, for broader review of closure effects on relationships).


Figure 2.10A shows relational and structural embedding associated with positive

relationships. The horizontal axis is structural embedding this year (number of people

with whom the employee and the colleague are both connected by citations). The

dependent variable is whether the employee cites the colleague as good or outstanding

next year. The bold line in the graph shows that the citation probability next year

increases with the number of mutual contacts this year. The association is strong. The

data cluster closely around the regression line (test statistic of 14.88).

The closure-trust association is strongest for new relationships. The regression

line at the top of Figure 2.10A describes the closure-trust association for relations that

survived for more than two years. Relations embedded in many mutual contacts (to the

right in the graph) are no stronger than relations between people who have no mutual

contacts (to the left; 0.81 test statistic for the regression line). In other words, these

relations in their third year have grown sufficiently strong that structural embedding no

longer matters. Employee and colleague have known one another long enough that

they do not require the reassurance of mutual friends.16

16There is a consequential methodological choice implicit in the Figure 2.10A evidence. Distinguish

two sets of relations: relations cited this year as frequent and substantive versus relations that could have been cited. The first is a “realized” risk set. The second is a “potential” risk set. Figure 2.10A is based on a realized risk set: the relations cited this year that are more embedded in closed networks are more likely to receive positive evaluations next year. Similarly, the next two bits of closure evidence are based on realized risk sets: Figure 2.10B describes the probability of decay in relations cited this year, and Figure 2.11 describes decay in reputations observed this year. However, while a realized risk set makes sense when studying decay — we see it today and study how long it takes to expire — both realized and potential risk sets can make sense when studying change that includes the formation of relations. The choice between realized and potential is consequential in that closure’s effect is likely underestimated within a realized risk set. The realized risk set excludes people who could have had relations with ego but did not. Some of those people are embedded in mutual contacts with ego, but most are not, so their exclusion results in overstating the probability of strong ties in the absence of mutual contacts. This issue is especially consequential in a population of disconnected closed networks. Imagine a population in which everyone is a member of only one group and has maximum connection with everyone else in his or her group. In such a population, the prediction in Figure 2.10A would be a step function: zero probability of a strong tie in the absence of mutual contacts, and 1.0 probability of a strong tie given one or more mutual contacts. If closure’s effect is estimated within a realized risk set, then the effect is zero: all relations included in the estimation are maximum and all are embedded in mutual contacts. If closure’s


Figure 2.10B shows the same closure effect in reverse. Instead of facilitating trust,

the closure effect is displayed in terms of preventing decay over time. The dependent

variable in Figure 2.10B is whether a relationship cited this year decays so it is not cited

next year (citation is for work connection, any level of evaluation). The horizontal axis is

relationship age. In the first year, bridge relations decay at a rate of 92% while

embedded relations decay at a lower rate of 53%. Faster decay rates make sense for

bridge relations in that bridges are not protected by obligations ensured by mutual

friends — so bridges are more subject to short-term cost-benefit analysis and more

subject to suspicions about the person on the other side (Stovel, Golub, and Milgrom,

2011). The complementarity between relational and structural embedding is evident in

effect is estimated within a potential risk set, on the other hand, closure’s effect is strong because estimation now includes the many relations of zero that occur in the absence of mutual contacts. For a concrete example, the graph at the end of this note is analogous to Figure 2.10A but computed within a potential risk set rather than the realized risk set (also in Burt, 2005:123-126). As in Figure 2.10A, the vertical axis is the probability of a banker evaluating a colleague as good or outstanding next year, and the horizontal is the number of mutual contacts shared by banker and colleague this year. In contrast to the 46,231 cited relations used to compute results in Figure 2.10A, the footnote graph is based on the 387,720 that could have occurred between 346 senior bankers who were continuous employees through the four-year observation period (129,240 potential relations from one banker to a second banker in each of three years used to predict the first banker’s next-year evaluation of the second). The 346 bankers were each at risk of being cited since they were continuously employed in senior ranks in the same organization. Of 387,720 possible citations, 377,606 were unrealized and 10,114 were realized (other of the 46,231 cited relations summarized in Figure 2.10A were to colleagues in other divisions, or lower job ranks, or new arrivals, or bankers who left the company). Of the unrealized relations, 120,454 were between bankers who had one or more mutual contacts. The footnote graph shows how including the unrealized relations lowers the probability of strong, positive evaluations in the absence of mutual contacts, which results in stronger evidence of closure’s effect (40.66 logistic test statistic here is 14.88 in Figure 2.10A). Relations cited last year, although showing the strength improvement associated with relational embedding, now also show a significant improvement associated with structural embedding (8.86 test statistic here versus 0.81 in Figure 2.10A).

Probability of citing acolleague cited lastyear (z = 8.86)

Probability citing anycolleague (z = 40.66)


Figure 2.10B from the varying gap in decay rates between bridges and embedded

relations. The larger the gap, the more important structural embedding is in preventing

decay. The gap is greatest in the first year of a relationship. The gap all but disappears

after the third year, showing that structural embedding is no longer a decay factor.

The waning value of structural embedding is substantively important because it

means that the bridge relations defining network brokers do not need closure to ensure

trust between people connected by the bridge. We know that trust and understanding

are more likely across bridge relations that have developed beyond a need for structural

embedding. Examples are Uzzi (1996) on garment manufacturers less likely to go

bankrupt if they concentrate their business in a few suppliers, Reagans and McEvily

(2003) on strong bridges facilitating knowledge transfer, Centola and Macy (2007) on

complex ideas more likely to diffuse through “wide” bridges, Tortoriello and Krackhardt

(2010) on innovation associated with strong bridges, termed “Simmelian ties,” and Sosa

(2011) on creativity associated with strong rather than weak bridges.

Stable Reputation Figure 2.11 contains illustrative evidence of closure’s effect on reputation. Reputation is

measured by the average evaluation a person received from colleagues. Figure 2.11A

shows that banker reputations are stable from year to year. This is important because

reputation without stability cannot provide the trust-facilitation illustrated in Figure 2.10.

If a person behaves badly this year, it erodes his reputation among colleagues, and they

know to avoid him next year. Knowing that will happen, the potential misbehaver has

an incentive not to behave badly. But if bad behavior is quickly forgotten, colleagues

next year have no forewarning to avoid the misbehaving person, so there would be no

reputation incentive for the misbehaver to behave well.


Figure 2.11B shows how the stability of a banker’s reputation increases with

connections among the people evaluating him. Bankers are arranged on the horizontal

axis in order of mutual contacts with colleagues evaluating the banker. To the left,

illustrated by a sociogram below the horizontal axis, are bankers evaluated by

colleagues who have no contact with one another. Banker and evaluators share no


mutual contacts. To the right are bankers evaluated by densely connected colleagues.

The vertical axis measures reputation stability (Burt, 2010:161-166). For each banker,

the correlation between reputation this year and reputation next year (Figure 2.11A) is

computed for the banker and the dozen colleagues adjacent to him on the horizontal

axis, i.e., the dozen colleagues with networks most similar to the banker’s in closure. At

the top of the vertical axis are bankers whose reputations are extremely stable. At the

bottom of the vertical axis are bankers whose reputations this year have no correlation

with their reputations next year.

Although one can find differences in the targets of positive and negative gossip

(Ellwardt, Labianca, and Wittek, 2012), Figure 2.11B shows that positive and negative

reputations are strikingly similar in their dependence on network closure. The dark dots

describe bankers with below-average evaluations this year. The white dots describe

bankers with above-average evaluations this year. There is no statistically significant

difference between the height of corresponding white and dark dots, which is to say that

positive reputations are on average no more or less stable than negative reputations.

Closure is the key stability factor. Reputation is correlated .73 from year to year

for bankers evaluated by colleagues in closed networks (upper-right corner in Figure

2.11B). At the other extreme, the reputations of bankers evaluated by disconnected

colleagues show no stability. The year-to-year correlation is a negligible .09 (lower-left

corner in Figure 2.11B). The association between reputation stability and network

closure so apparent in Figure 2.11B is a robust effect in statistical analysis of the data

(Burt, 2010:167), which supports Coleman’s (1988:S107–S108) intuition that:

‘‘Reputation cannot arise in an open structure, and collective sanctions that would

ensure trustworthiness cannot be applied.’’

Bandwidth versus Echo The results in Figure 2.11B illustrate more than reputation’s dependence on closure.

They also illustrate an important feature of the social mechanism responsible. The

horizontal axis in Figure 2.11B describes connection through mutual contacts without

distinguishing positive from negative connections. The significance of the neglect is

illustrated in Figure 2.12 with sociograms showing a colleague about to evaluate a


banker. In Figure 2.12A, colleague and banker are linked by positive indirect

connections through mutual contacts. Colleague and banker share negative opinion of

Emile and positive opinion of Marc. Marc is a source of positive stories about the

banker and a more likely discussion partner than the disliked Emile. If the colleague

and Emile find themselves in a conversation, Emile’s negative stories about the banker

strengthen the colleague’s positive opinion of the banker (my enemy’s enemy is my

friend). Figure 2.12B displays negative indirect connections. The colleague thinks well

of Catherine, who has a negative relation with the banker. Catherine is a likely

discussion partner for the colleague and she will have stories to support her negative

opinion of the banker. The colleague is less likely to gossip with disliked Philippe, but if

a conversation occurs, and Philippe shares a story about his positive relationship with

the banker, it will strengthen the colleague’s negative opinion of the banker (my

enemy’s friend is my enemy). In short, (as predicted by balance theory, Heider, 1958),

positive evaluations are expected to develop in relations embedded in positive indirect

connections (Figure 2.12A) and negative evaluations are expected in relations

embedded in negative indirect connections (Figure 2.12B).


But the banker relations are not balanced in direction, only in strength. Negative

and positive evaluations are both more likely in relationships embedded in positive or

negative indirect connections (Burt, 2008). It is true that positive evaluations are more

associated with positive indirect connections than with negative (Burt, 2010:175), but

positive indirect connections also increase the likelihood of negative relationships (Burt,

2005:185), positive and negative indirect connections both protect against relation

decay (Burt, 2010:175), and positive and negative indirect connections both stabilize

reputations into next year (Burt, 2010:167).

Explanation for stability resulting from relationship magnitude — rather than the

positive or negative sentiment between people — requires digging past network

structure to the information flowing through a network. Reputations emerge from the

flow of information about opinion and behavior. Information flow can be non-reactive or

reactive. Non-reactive refers to a network in which information flows without distortion,


like water flows through pipes in a plumbing system. The more closed the network, the

more alternative channels through which information can move, making it more likely

that people are informed about one another’s opinion and behavior. For example,

eBay’s reputation system is a network of buyers and sellers through which information

on one another’s behavior is distributed. Knowing bad behavior will be reported in a

public way, buyers and sellers who wish to continue in eBay have an incentive to

behave well. The information on past behavior is presented in the same way to anyone

who looks it up. The information is not filtered such that certain viewers see more

positive or negative information. This image of information moving through a network,

non-reactive to channel, can be labeled a “bandwidth” image of closed networks; more

connections create wider bandwidth for information flow.

In contrast, information is often reactive in social networks. What one says to a

person can be affected by the person to whom it is being said. We tell stories, share

information, consistent with the emotional tone of the conversation. When a friend is

suffering in the emotional aftermath of a bad relationship, we do not share positive

stories about the friend’s former partner; we share negative stories portraying the

partner as a miserable creature from whom our friend is fortunate to be free. This is

what friends do. We support one another, strengthen our friendship, by displaying

similar orientation to surrounding people, objects, and events. More generally, this is

how we build relationships. We display to one another our similar orientation to

surrounding people, objects, and events to establish a comfortable, reassuring

connection between us. In network analysis, this is a process of connections

established and maintained with strategic displays of structural equivalence. The

implication of the process is that closed networks are a partisan echo chamber more

than a neutral distribution system. We join conversations consistent with the way we

feel and we hear echoed back the emotional tone of the conversation. The risk is

ignorant certainty. We become certain in opinions we often share with colleagues but

the opinions are based on a sample distorted by etiquette (Burt, 2005:178-181).

The above discussion draws sharp distinction between an “echo” image of closed

networks subjecting information to an etiquette filter versus a “bandwidth” image of


closed networks as a plumbing system of pipes for information flow. The distinction is

more accurately a matter of degree with contexts varying in the extent to which an

etiquette filter applies. Detailed argument and evidence is available elsewhere (Burt

and Knez, 1995; Burt, 2005:Chp.4, 2010:Chp. 6).

There are three implications for this book. A practical implication is that I do not

have to worry in the forthcoming analyses about predicting from the balance of

sentiments in a relationship so much as connection strength. This is valuable because

connection strength can be measured with network data less precise — and so more

reliable — than the data required to measure the balance of sentiment in a relationship.

I cannot predict from ego’s sentiment toward a contact what information will flow

through their connection. What flows will depend on context. But the likelihood of flow

increases with connection strength, and that is what predicts trust and reputation. The

indirect connections used to predict positive relations and relation decay in Figure 2.10

do not depend on people sharing similarly positive or negative evaluations of mutual

contacts, only the strength of their connection through third parties. The indirect

connections used to predict reputation stability in Figure 2.11 do not depend on

colleague and banker sharing similar evaluations of third parties, only the strength of

their connection through third parties.

——— Table 2.2 About Here ———

Second, echo shifts reputation ownership from ego to her contacts. The pronoun

in the phrase “your reputation” merely indicates who is going to suffer if reputation is

damaged. Under a bandwidth image of closed networks, ego owns her reputation. The

network is merely a distribution system ensuring that colleagues are informed about

ego’s opinion and behavior. If ego behaves well in eBay, for example, potential

partners will see stories about ego’s good behavior posted by the people she treats

well. Under an echo image, ego does not own her reputation. Her reputation is owned

by the people who talk about her. They own Reputation is only valuable if it persists to

facilitate entry into new opportunities, so reputation as a valuable asset is owned by

whomever ensures that reputation persists over time. The results in the lower-left

corner of Figure 2.11B show that reputations do not persist in open networks.


Reputation stability increases quickly up through a handful of mutual contacts, then

increases at a slower rate with additional mutual contacts. Reputation stability is

anchored in clusters of about five or more colleagues reinforcing one another’s opinion.

If the people talking about ego were primarily concerned with exchanging information to

accurately summarize ego, then the bandwidth and echo images of closed networks

would predict the same final reputation. However, the etiquette filter is applied so that

people select stories to build and maintain relations with one another. In different

groups, different stories can bring people together. Where some active people have a

positive view, positive stories about ego circulate so she acquires a positive reputation.

Where people are brought together by negative stories about ego, negative stories

circulate so she acquires a negative reputation. The summary distinction between

bandwidth versus echo is between closed networks as a neutral distribution system

versus closed networks as a partisan production system; not partisan in terms of for or

against ego personally, partisan in the sense of contacts pursuing their interests. In the

interest of building relations with one another, people exchange stories about ego, as

well as other people, objects, and events. The stories they select depend on what

brings them together. Ego’s reputation among them is a by-product of their selection.

Table 2.2 lists key implications of echo: reputation depends on circulating stories about

ego, stories selected because they bring discussants together, so ego’s reputation is

owned by the people circulating the stories, reputation emerges from ego’s behavior on

projects likely to be talked about, and ego has as many reputations as there are closed

networks in which she are discussed.

Third, the echo image of closure explains closure generating reputation stability for

network brokers as well as for people in closed networks. Ego’s reputation is sustained

by gossip about ego, with or without direct contact to ego. Given ego soliciting trust

from alter, or alter deciding whether to trust alter, the closure that facilitates trust is not

between ego and alter, creating bandwidth for alter's information on ego; it is closure

around alter, creating echo that solidifies alter's opinion of ego. This means that the

closure around alter can maintain the reputation of a colleague within the same network,


or a broker beyond the network. Thus, closure can maintain the reputations of network

brokers just as it maintains the reputations of people in closed networks.


To illustrate this third implication, I re-calculated the results in Figure 2.11B for two

categories of bankers: those who are network brokers versus others. Network brokers

are distinguished for the illustration as anyone with below-median network constraint

this year (predicting reputation stability into next year). The vertical axis in Figure 2.13

is the same as in Figure 2.11B; it is a subsample correlation between reputation this

year and next.

The horizontal axis in Figure 2.13 distinguishes bankers by the average number of

third parties to the evaluations of people who evaluated the banker. These are the third

parties who closed the network around evaluators, making an evaluator more certain in

his or her opinion of a banker. Consider the sociogram to the right under the horizontal

axis. Two colleagues evaluate a banker. Each colleague in turn is evaluated by three

other colleagues. The count of third parties as defined in Figure 2.11B is zero; there are

no third parties connecting the banker with either evaluator. But the two evaluators

themselves are embedded in relatively closed networks. Average numbers of third

parties embedding an evaluator’s relations are rounded in Figure 2.13 to the nearest

lower integer (e.g., an average of 2.43 third-party connections would be over the “2” on

the horizontal axis), but the regression equations in the graph are estimated from

continuous scores (routine t-tests are adjusted down for repeated observations using

the “cluster” option in STATA).

The summary point to Figure 2.13 is that the reputation stability associated with

closure is true for brokers as well as for people in closed networks. In fact, broker

reputations are slightly more stable and broker contacts are slightly more embedded in

closed networks. The greater stability is apparent in the graph from the higher

regression line for brokers.17 The deeper embedding is hinted at in the graph by the

17To put the point in perspective, regress the subsample stability correlation for a banker (vertical

axis is Figure 2.13) across log third parties, a dummy variable distinguishing broker bankers, and an interaction term between the broker dummy and the log third parties. With routine t-tests adjusted down


lack of brokers with any contacts embedded in less than two third parties on average

(broker regression line begins at two third parties). Specifically, the people evaluating

broker bankers are embedded in 7.6 third parties versus 6.6 for non-broker bankers (8.0

t-test). The deeper embedding around broker contacts can seem counter-intuitive since

brokers by definition have greater access to structural holes, but Figure 2.13 is not

about the broker’s network; it is about the networks around a broker’s contacts. The

deeper embedding indicates that brokers were evaluated more often by colleagues in

closed networks, and Figure 2.13 shows that the more-closed networks around broker

contacts are associated with more stable reputations.

CONCLUSIONS This chapter has been an introduction to the network structure of advantage. I focused

on the two core principles defining network advantage: brokerage and closure.

Brokerage is about innovation and growth, expanding into new rewarding activity

and perspectives. The argument is that information and practice become sticky within

dense social clusters such that network brokers — the people who connect across the

structural holes between clusters — have information breadth, timing, and arbitrage

advantages that make brokers more likely and able to detect and develop rewarding

opportunities. The arbitrage advantage of interpreting source information for a target

audience is particularly important. Broker advantage does not result from getting

diverse information so much as it results from personal abilities exercised and

developed while managing diverse information. In communicating between discordant

understandings across a structural hole, network brokers become skilled in recombinant

methodology of analogy and metaphor for helping people understand a point of view

inconsistent with their own. Understanding the mechanism by which brokerage

provides advantage is important to developing and managing advantage, but the

summary brokerage principle remains simply that achievement is fostered by access to

for repeated observations, the respective t-tests for the three predictors are 15.11, 5.81, and -3.02. Stability is most strongly associated with closure, but it is also significantly higher for brokers.


structural holes. Illustrative evidence includes network brokers enjoying higher

compensation and more positive recognition (Figure 2.3), being recognized as leaders

in formal and informal organization (Figure 2.4), and emerging as recognized leaders in

informal task groups (Figure 2.6).

Closure is about governance and the status quo; maintaining stability and safety

while people get better at what they already do. The argument is that dense

communication channels in a closed network make it more likely that behavior and

opinion inconsistent with standards in the network will be detected and discussed. With

detection and discussion more likely, reputations emerge and bad behavior is less likely

within the network, which lowers the risk of trust, thereby increasing the probability of

trust. Information transmission defines a distinction between two mechanisms by which

closure creates trust and reputation. Information can be reactive or non-reactive to the

channels through which it moves. When information is non-reactive, closure creates

bandwidth – more connections mean better access to information. Certain electronic

evaluation websites such as eBay’s reputation system are an illustration. Everyone

sees the same evaluations. Trust and reputation emerge from people better informed.

On the other hand, when information is reactive to the connected people sharing it,

closure creates echo – more connections mean more exposure to a biased sample of

information. Stories consistent with ongoing emotional tone and shared perspectives

are shared more often than stories contradictory to ongoing conversation, so trust and

reputation emerge distorted to fit prevailing emotion and perspective. The distinction

between bandwidth and echo has implications for managing trust and reputation (Table

2.2), but by either mechanism, the summary closure principle is that trust and reputation

are fostered in closed networks. Illustrative evidence includes relations more durable

and positive in closed networks (Figure 2.10), and reputations more stable in closed

networks (Figure 2.11). In fact, the illustrative banker and analyst reputations dissolve

in the absence of closure: reputation this year in an open network has no correlation

with reputation next year (Figure 2.11).

The creation of trust and reputation is critical for brokerage. This is a fundamental

link between brokerage and closure. The achievements associated with brokerage


require the trust and reputation associated with closure. Would-be brokers need to be

accepted as brokers by the people between whom, or to whom, connection is to be

brokered. Is the would-be broker competent, trustworthy, and appropriate as a source

of news to a person like me? The would-be broker’s job rank and network status are

two visible signals relevant to these concerns. Job rank indicates social standing in the

formal organization, and returns to brokerage are higher for people holding more senior

jobs (Table 2.1). Network status indicates social standing in the informal organization,

and returns to brokerage are higher for people with status higher in discussion networks

(Figure 2.7). However, reputation alone is sufficient: Regardless of status, bankers

with positive reputation show the advantage of access to structural holes while bankers

with below-average reputations show no advantage for the same access (Figure 2.8).

Contingent brokerage creates an unobtrusive organization diagnostic. Given

returns to brokerage contingent on trust and reputation, categories of people not trusted

in an organization can be identified by looking for places where rewards are not

enhanced by brokerage. I discussed three examples: a merger in which people from

the legacy acquired firm were distrusted as brokers, a division in which the senior

person stunted the development of new leaders by rewarding for loyalty rather than

brokerage, and a firm in which women and young men who tried to be brokers were

suspect until proven otherwise (Figure 2.9).

Brokerage contingent on closure seems a contradiction. Brokerage by definition

involves links across closed networks, so how do brokers find the trust and reputation

necessary for successful brokerage?

Two points in the review resolve the ostensible contradiction. The first is a matter

of timing. The safety offered by a closed network is less necessary as two people get to

know one another. At some point, collaboration can be maintained by relational

embedding without structural embedding. For the bankers and analysts in Figure 2.10,

the transition occurred in the second year. By the third year of a relationship, the

mutual friends provide no significant advantage for trust or decay prevention. Second,

reputation is maintained in social networks, and therefore owned, by a broker’s

audience, not the broker. Reputation implications of the difference between bandwidth


and echo are detailed in Table 2.2, and illustrative evidence was discussed showing

that the reputation stability displayed in Figure 2.11 is based on echo more than

bandwidth (see discussion of Figure 2.13). A network broker is by definition not

embedded in a closed network with the people between whom she brokers connections,

but she does have connections into closed networks within which her reputation is built

and maintained. Reputation does not require closure between broker and contacts. It

requires only that some proportion of a broker’s contacts are in closed networks.


REFERENCES Aldrich, Howard and Diane Herker. 1977. “Boundary spanning roles and organization

structure.” Academy of Management Review 2:217-230. Allen, Thomas J. and Steven Cohen. 1969. “Information flow in R&D labs.

“Administrative Science Quarterly 14:12-19. Aral, Sinan and Marshall Van Alstyne. 2011. “Networks, information and brokerage: the

diversity-bandwidth tradeoff.” American Journal of Sociology 117:90-171. Aral, Sinan and Vincent J. David. 2012. “The anatomy and dynamics of vision

advantages.” Proceedings of the 33rd Annual International Conference on Information Systems, Orlando, FL.

Barker, James R. 1993. “Tightening the iron cage: concertive control in self-managing teams.” Administrative Science Quarterly 38:408-437.

Bavelas, Alex. 1948. “A mathematical model for group structures.” Applied Anthropology 7:16-30.

Bavelas, Alex. 1950. “Communication patterns in task-oriented groups.” Journal of Acoustical Society of America 22:725-730.

Bernstein, Lisa. 1992. “Opting out of the legal system: extralegal contractual relations in the diamond industry.” Journal of Legal Studies 21:115-157.

Bernstein, Lisa. 2001. “Private commercial law in the cotton industry: creating cooperation through rules, norms, and institutions.” Michigan Law Review 99:1724-1790.

Blau, Peter M. 1955. The Dynamics of Bureaucratic Life. Chicago: University of Chicago Press.

Bonacich, Phillip. 1972. “Factoring and weighting approaches to status scores and clique identification.” Journal of Mathematical Sociology 2:113-120.

Bonacich, Phillip. 1987. “Power and centrality: a family of measures.” American Journal of Sociology 92:1170-1183.

Borgatti, Stephen P. 2002. Netdraw Network Visualization. Analytic Technologies: Harvard, MA.

Borgatti, Stephen P. and Pacey C. Foster 2003. “The network paradigm in organizational research: a review and typology.” Journal of Management 29:991-1013.

Borgatti, Stephen P., Ajay Mehra, Daniel J. Brass, and Giuseppe Labianca. 2009. “Network analysis in the social sciences.” Science 323:892-895.


Borgatti, Stephen P., Martin G. Everett, and Linton C. Freeman. 2002. UCINET for Windows: Software for Social Network Analysis. Harvard, MA: Analytic Technologies.

Bothner, Matthew S., Joel M. Podolny, and Edward Bishop Smith. 2011. “Organizing contests for status: the Matthew effect versus the Mark effect.” Management Science 57:439-457.

Bothner, Matthew S., Young-Kyu Kim, and Edward Bishop Smith. 2012. “How does status affect performance? Status as an asset vs. status as a liability.” Organization Science 23:416-433.

Brass, Daniel J. 1984. “Being in the right place: a structural analysis of individual influence in an organization.” Administrative Science Quarterly 29:518-539.

Burt, Ronald S. 1980. “Autonomy in a social topology.” American Journal of Sociology 85:892-925.

Burt, Ronald S. 1992. Structural Holes. Harvard University Press: Boston.

Burt, Ronald S. 1997. “The contingent value of social capital.” Administrative Science Quarterly 42:339-365.

Burt, Ronald S. 1999. “The social capital of opinion leaders.” Annals 566:37-54. Burt, Ronald S. 2004. “Structural holes and good ideas.” American Journal of Sociology

110:349-399.

Burt, Ronald S. 2005. Brokerage and Closure. Oxford: Oxford University Press. Burt, Ronald S. 2008. “Gossip and reputation.” Pp. 27-42 in Management et réseaux

sociaux: resource pour l’action ou outil de gestion?, edited by Marc Lecoutre and Pascal Lievre. London: Hermes-Lavoisier.

Burt, Ronald S. 2010. Neighbor Networks. Oxford: Oxford University Press. Burt, Ronald S., Martin Kilduff, and Stefano Tasselli. 2013. “Social network analysis,

through a focus on network advantage.” Annual Review of Psychology 64:537-547.

Burt, Ronald S. and Marc Knez. 1995. “Kinds of third-party effects on trust.” Rationality and Society 7:255-292.

Burt, Ronald S., and Jennifer Merluzzi. 2013. “Embedded brokerage.” Pp. xxx-xxx in Research in the Sociology of Organizations, edited by Stephen P. Borgatti, Daniel J. Brass, Daniel S. Halgin, Giuseppe Labianca, and Ajay Mehra. Cambridge, MA: Emerald Group Publishing.

Burt, Ronald S., and Donald Ronchi. 2007. “Teaching executives to see social capital.” Social Science Research 36: 1156-1183.

Carrington, Peter J., John Scott, and Stanley Wasserman, eds. 2005. Models and Methods in Social Network Analysis. New York: Cambridge University Press.


Centola, Damon and Michael Macy. 2007. “Complex contagions and the weakness of long ties.” American Journal of Sociology 113:702-734.

Coleman, James S. 1972. “Systems of social exchange.” Journal of Mathematical Sociology 2:145-163.

Coleman, James S. 1988. “Social capital in the creation of human capital.” American Journal of Sociology 94:S95-S120.

Coleman, James S. 1990. Foundations of Social Theory. Cambridge, MA: Harvard University Press.

Coleman, James S., Elihu Katz, and Herbert Menzel. 1957. “The diffusion of an innovation among physicians.” Sociometry 20:253-270.

Contractor, Noshir S., Leslie A. DeChurch, Jay Carson, Dorothy R. Carter, and Brian Keegan. 2012. “The topology of collective leadership.” Leadership Quarterly 23: 994-1011.

Cook, Karen S. and Richard M. Emerson. 1978. “Power, equity and commitment in exchange networks.” American Sociological Review 43:712-739.

Cook, Karen S., Richard M. Emerson, Mary R. Gillmore, and Toshio Yamagishi. 1983. “The distribution of power in exchange networks: theory and experimental results.” American Journal of Sociology 89:275-305

Cross, Rob and Andrew Parker. 2004. The Hidden Power of Social Networks. Boston, MA: Harvard Business School Press.

de Nooy, Wouter, Andrej Mrvar, and Vladimir Batagelj. 2005. Exploratory Social Network Analysis with Pajek. New York: Cambridge University Press.

DiMaggio, Paul and Walter W. Powell. 1983. “The iron cage revisited: institutional isomorphism and collective rationality in organizational fields.” American Sociological Review 48:147-160.

Eagle, Nathan, Michael Macy, and Rob Claxton. 2010. “Network diversity and economic development.” Science 138:1029-1031.

Easley, David and Jon Kleinberg. 2010. Networks, Crowds, and Markets. New York: Cambridge University Press.

Ellwardt, Lea, Giuseppe (Joe) Labianca, and Rafael Wittek. 2012. “Who are the objects of positive and negative gossip at work? A social network perspective on workplace gossip.” Social Networks 34:193-205.

Everett, Martin and Stephen P. Borgatti. 2005. “Ego network betweenness.” Social Networks 27:31-38.

Ferrin, Donald L., Andrew Parker, Robert L. Cross, and Kurt T. Dirks. 2012. “Who are structural entrepreneurs? Tertius gaudens or tertius adserviens?” Paper presented at the annual meetings of the Academy of Management, Boston, MA.


Festinger, Leon, Stanley Schachter, and Kurt W. Back. 1950. Social Pressures in Informal Groups. Stanford, CA: Stanford University Press.

Fleming, Lee, Santiago Mingo, and David Chen. 2007. “Collaborative brokerage, generative creativity, and creative success.” Administrative Science Quarterly 52:443-475.

Fleming, Lee and David M. Waguespack. 2007. “Brokerage, boundary spanning, and leadership in open innovation communities.” Organization Science 18:165-180.

Freeman, Linton C. 1977. “A set of measures of centrality based on betweenness.” Sociometry 40:35-40.

Freeman, Linton C. 1979. “Centrality in social networks: conceptual clarification.” Social Networks 1:215-239.

Freeman, Lindon C. 2004. The Development of Social Network Analysis. Vancouver, Canada: Empirical Press.

Gambetta, Diego and Heather Hamill. 2005. Streetwise. New York: Russell Sage Foundation.

Goyal, Sanjeev. 2007. Connections. Princeton, NJ: Princeton University Press. Granovetter, Mark. 1973. “The strength of weak ties.” American Journal of Sociology 78:

1360-1380. Granovetter, Mark. 1983. “The strength of weak ties: a network theory revisited.”

Sociological Theory 1:201-233. Granovetter, Mark. 1985. “Economic action and social structure: the problem of

embeddedness.” American Journal of Sociology 91:481-510. Granovetter, Mark. 1992. “Problems of explanation in economic sociology.” Pp. 29-56 in

Networks and Organizations, edited by Nitin Nohria and Robert G. Eccles. Boston, MA: Harvard Business School Press.

Greif, Avner. 1989. “Reputation and coalitions in medieval trade: evidence on the Maghribi traders.” Journal of Economic History 49:857-882.

Hanneman, Robert A. and Mark Riddle. 2005. Introduction to Social Network Methods. Riverside, CA: University of California Riverside published in digital form at http://faculty.ucr.edu/~hanneman/nettext.

Hansen, Derek, Ben Shneiderman, and Marc A. Smith. 2011. Analyzing Social Media Networks with NodeXL. Burlington, MA: Morgan Kaufmann, Elsevier.

Heider, Fritz. 1958. The Psychology of Interpersonal Relations. New York: John Wiley. Jackson, Matthew O. 2008. Social and Economic Networks. Princeton, NJ: Princeton

University Press. Jennings, Helen H. 1937. “Structure of leadership — development and sphere of

influence.” Sociometry 1:99-143.


Jennings, Helen H. 1943. Leadership and Isolation. New York: Longmans, Green. Jira, Chonnikarn, David M. Waguespack, and Lee Fleming. 2012. “Collaborative

innovation and leadership effectiveness.” Working Paper, Harvard Business School.

Kadushin, Charles. 2012. Understanding Social Networks. New York: Oxford University Press.

Katz, Elihu and Paul F. Lazarsfeld. 1955. Personal Influence. New York: Free Press. Katz, Leo. 1953. “A new status index derived from sociometric analysis.” Psychometrika

18:39-43. Kellogg, Katherine C. 2012. “Keeping it Real: Inter-Occupational Guardians and

Institutional Change in Two Community Health Centers.” Working Paper, Massachusetts Institute of Technology.

Kilduff, Martin and Daniel J. Brass. 2010. “Organizational social network research: core ideas and key debates.” Academy of Management Annals 4:317-357.

Kleinberg, Jon M. 2000. “Navigation in a small world.” Nature 406:845. Knoke, David. 2012. Economic Networks. Cambridge: Polity Press. Leavitt, Harold J. 1949. Some Effects of Certain Communication Patterns upon Group

Performance. Doctoral Dissertation, Massachusetts Institute of Technology, Boston, MA.

Leavitt, Harold J. 1951. “Some effects of certain communication patterns upon group performance.” Journal of Abnormal and Social Psychology 46:38-50.

Leavitt, Harold J. 1996. “The old days, hot groups, and manager’s lib.” Administrative Science Quartery 41:288-300.

Leonardi, Paul and Diane Bailey. 2011. “Sharing the work of brokerage: network articulation for idea generation and implementation.” Presented at Chicago-Northwestern Conference on Innovation, Organizations, and Society, Chicago, IL.

Lin, Nan, Walter Ensel, and John Vaughn. 1981. “Social resources and strength of ties: structural factors in occupational status attainment.” American Sociological Review 46:393-405.

Lin, Nan. 2002. Social Capital. Cambridge: Cambridge University Press. Lingo, Elizabeth Long and Siobhán O’Mahony. 2010. “Nexus work: brokerage on

creative projects.” Administrative Science Quarterly 55:47-81. Luther, Kurt and Amy Bruckman. 2008. “Leadership in online creative collaboration.”

Proceedings of the ACM 2008 conference on Computer Supported Collaborative Work, ACM, 343-352.

Luther, Kurt, Casey Fiesler, and Amy Bruckman. 2013. “Redistributing leadership in online creative collaboration.” Proceedings of the ACM 2013 conference on Computer Supported Collaborative Work, ACM, 1007-1022.


Marsden, Peter V. 2011. “Survey methods for network data.” Pp. 370-388 in The SAGE Handbook of Social Network Analysis, edited by John P. Scott and Peter J. Carrington. Thousand Oaks, CA: Sage.

Mehra, Ajay, Andrea L. Dixon, Daniel J. Brass, and Bruce Robertson. 2006. “The social network ties of group leaders: implications for group performance and leader reputation.” Organization Science 17:64-79.

Mehra, Ajay, Martin Kilduff, and Daniel J. Brass. 2001. “The social networks of high and low self-monitors: implications for workplace performance.” Administrative Science Quarterly 46:121-146.

Merton, Robert K. 1968. “The Matthew effect in science.” Science 159:56-63.

Mizruchi, Mark S., Peter Mariolis, Michael Schwartz, and Beth Mintz. 1986. “Techniques for disaggregating centrality scores in social networks.” Sociological Methodology 16:26-48.

Moreno, Jacob L. 1934. Who Will Survive? Washington, D.C.: Nervous and Mental Disease Publishing.

Moreno, Jacob L. 1937. “Sociometry in relation to other social sciences.” Sociometry 1:206-219.

Nee, Victor and Sonja Opper. 2012. Capitalism from Below. Cambridge, MA: Harvard University Press.

Newman, M. E. J. 2010. Networks. New York: Oxford University Press. Obstfeld, David. 2005. “Social networks, the tertius iungens orientation, and

involvement in innovation.” Administrative Science Quarterly 50:100-130. Padgett, John F. and Walter W. Powell, eds. 2012. The Emergence of Markets and

Organizations. Princeton, NJ: Princeton University Press. Pearce, Craig L. and Jay A. Conger, eds. 2003. Shared Leadership. Thousand Oaks,

CA: Sage Publications.

Pentland, Sandy. 2008. Honest Signals. Cambridge, MA: MIT Press. Phillips, Damon J. 2011. “Jazz and the disconnected: city structural disconnectedness

and the emergence of a jazz canon, 1897-1933.” American Journal of Sociology 117:420-483.

Podolny, Joel M. 1993. “A status-based model of market competition.” American Journal of Sociology 98:829-872.

Podolny, Joel M. 2001. “Networks as the pipes and prisms of the market.” American Journal of Sociology 107:33-60.

Podolny, Joel M. 2005. Status Signals. Princeton, NJ: Princeton University Press. Podolny, Joel M. and Damon J. Phillips. 1996. “The dynamics of organizational status.

Industrial and Corporate Change 5:453-471.


Portes, Alejandro. 1998. “Social capital: its origins and applications in modern sociology.” Annual Review of Sociology 24:1-24.

Powell, Walter W., Douglas R. White, Kenneth W. Koput, and Jason Owen-Smith. 2005. “Network dynamics and field evolution: the growth of interorganizational collaboration in the life sciences.” American Journal of Sociology 110:1132-1205.

Prell, Christina. 2012. Social Network Analysis. Thousand Oaks, CA: Sage. Putnam, Robert D. 1993. Making Democracy Work. Princeton, NJ: Princeton University

Press. Putnam, Robert D. 2000. Bowling Alone. New York: Simon and Schuster. Rainie, Lee and Barry Wellman. 2012. Networked. Boston, MA: MIT Press. Reagans, Ray and Bill McEvily. 2003. “Network structure and knowledge transfer: the

effects of cohesion and range.” Administrative Science Quarterly 28: 240-267. Reagans, Ray and Ezra W. Zuckerman. 2001. “Networks, diversity, and productivity: the

social capital of R&D teams.” Organization Science 12:502-517. Rider, Christopher I. 2009. “Constraints on the control benefits of brokerage: a study of

placement agents in U.S. venture capital fundraising.” Administrative Science Quarterly 54:575-601.

Rivera, Mark T., Sara B. Soderstrom, and Brian Uzzi. 2010. “Dynamics of dyads in social networks: assortative, relational, and proximity mechanisms.” Annual Review of Sociology 36:91-115.

Russell, Matthew A. 2011. Mining the Social Web. Sebastopol, CA: O’Reilly Media. Sauder, Michael, Freda Lynn, and Joel M. Podolny. 2012. “Status: insights from

organizational sociology.” Annual Review of Sociology 38:267-283. Scott, John P. 2000. Social Network Analysis. Thousand Oaks, CA: Sage. Scott, John P. and Peter J. Carrington. eds. 2011. The SAGE Handbook of Social

Network Analysis. Thousand Oaks, CA: Sage.

Shipilov, Andrew V., Stan Xiao Li, and Henrich R. Greve. 2011. “The prince and the pauper: search and brokerage in the initiation of status-heterophilous ties.” Organization Science 22:1418-1434.

Sosa, Manuel E. 2011. “Where do creative interactions come from? The role of tie content and social networks.” Organization Science 22:1-21.

Sparrowe, Raymond T., Robert C. Liden, Sandy J. Wayne, and Maria I. Kraimer. 2001. “Social networks and the performance of individuals and groups.” Academy of Management Journal 44:316-325.

Stovel, Katherine, Benjamin Golub, and Eva M. Meyersson Milgrom. 2011. “Stabilizing brokerage.” Proceedings of the National Academy of Sciences 108:21326-21332.


Stovel, Katherine and Lynette Shaw. 2012. “Brokerage.” Annual Review of Sociology 38:139-158.

Taylor, D. Garth and James S. Coleman. 1979. “Equilibrating processes in social networks: a model for conceptualization and analysis.” Pp. 257-300 in Perspectives on Social Network Research, edited by Paul W. Holland and Samuel Leinhardt. New York: Academic Press.

Tortoriello, Marco and David Krackhardt. 2010. “Activating cross-boundary knowledge: the role of Simmelian ties in the generation of innovations.” Academy of Management Journal 53:167-181.

Tushman, Michael. 1977. “Special boundary roles in the innovation process.” Administrative Science Quarterly 22:587-605.

Uzzi, Brian 1996. “The sources and consequences of embeddedness for the economic performance of organizations: the network effect.” American Sociological Review 61:674-698.

Uzzi, Brian and Jarrett Spiro. 2005. “Collaboration and creativity: the small world problem.” American Journal of Sociology 111:447-504.

Vasudeva, Gurneeta, Akbar Zaheer, and Exequiel Hernandez. forthcoming. “The embeddedness of networks: institutions, structural holes, and innovativeness in the fuel cell industry.” Organization Science 24: In Press.

von Hippel, Eric. 1994. “Sticky information” and the locus of problem solving: implications for innovation. Management Science 40:429-439.

Wasserman, Stanley and Katherine Faust. 1994. Social Network Analysis. New York: Cambridge University Press.

Wellman, Barry. 1993. “An egocentric network tale: comment on Bien et al.” 1991. Social Networks 15:423-436

Zuckerman, Ezra W., Tai-Young Kim, Kalinda Ukanwa, and James von Rittman. 2003. “Robust identities or nonentities? Typecasting in the feature-film labor market.” American Journal of Sociology 108:1018-1074.

James

Robert

1

23

54

6

7

A

B

C & D 25

1

0

100

29

Group A

Group B

Group C

Group D

Density Table

0

85

5

0

0

person 3: .402 = [.25+0]2 + [.25+.084]2 + [.25+.091]2 + [.25+.084]2

Robert: .148 = [.077+0]2 + [.154+0]2 + [.154+0]2 + [.154+0]2 + [.154+0]2 + [.154+0]2 + [.154+0]2

Network Constraint(C = Σj cij = Σj [pij + Σq piqpqj]2, i,j ≠ q)

Networkindicates

distributionof sticky

information, which defines

advantage.

Figure 2.1Network Bridge and Cluster Structure

Adapted from Burt (2005:14).

Figure 2.2 Network Metrics

Adapted from Burt (2010:298). To keep the sociograms simple, relations with ego are not

presented. Graph above plots density and hierarchy for 1,989 networks observed in six

populations (analysts, bankers, and managers in Asia, Europe, and North America; aggregated

in Figure 3 to illustrate returns to brokerage). Squares are executives (MD or more in finance, VP or more otherwise). Hollow circles are lower ranks. Executives have significantly larger, less

dense, and less hierarchical networks.

E B

D C

A

CliqueNetworks

3100093

3131311.00.0

5100065

13131313131.00.0

101000361.00.0

PartnerNetworks

367784

4420201.70.5

5402559

366666

3.43.0

102050418.218.0

BrokerNetworks

300

33

1111113.03.0

500

20

44444

5.010.0

1000

1010.045.0

SmallNetworks

size (degree)density x 100

hierarchy x 100constraint x 100

from:ABC

nonredundant contactsbetweenness (holes)

LargerNetworks



from:ABCDE


Still LargerNetworks




E B

D C

A

A

C B

A

C D

A

C B

A

C B

E B

D C

A

Network Density

Net

wor

k H

iera

rchy

Partner Networks

CliqueNetworksBrokers

Figure 2.3 Brokerage for Detecting and Developing Opportunities

Graph A shows idea quality increasing with more access to structural holes. Circles are average scores on the vertical axis for a five-point interval of network constraint among supply-chain managers in a large electronics firm (Burt, 2004:382, 2005:92). Bold line is the vertical

axis predicted by the natural logarithm of network constraint. Graph B shows performance increasing with more access to structural holes. Circles are average scores on the vertical axis for a five-point interval of network constraint within each of six populations (analysts, bankers,

and managers in Asia, Europe, and North America; Burt, 2010:26, cf. Burt, 2005:56).

Network Constraintmany ——— Structural Holes ——— few

Aver

age

Z-Sc

ore

Idea

Val

ue

Z-Sc

ore

Res

idua

l Per

form

ance

(eva

luat

ion,

com

pens

atio

n, p

rom

otio

n)

B. Yielding Performance Scores Higher than Peers(r = -.58, t = -6.78, n = 85)

A. Brokers Are More Likely to Detect & Articulate Good Ideas

(r = -.80, t = -9.67, n = 54)

Figure 2.4Network Brokers Tend To Be Recognized Leaders

Constraint and status are computed from work discussion networks around twelve hundred managers in four organizations.

A. In the formalorganization

B. And in the informal organization

Most Senior Job Ranks(29.5 mean network constraint)

Next-Lower,Middle Ranks(56.4 mean constraint)

Next-Lower,Senior Ranks(41.9 mean constraint)

r2 = .61

Net

wor

k St

atus

(S)

(Si =

Σj z

ji Sj,

divi

ded

by m

ean

so a

vera

ge is

1.0

)

Perc

ent o

f Peo

ple

with

in E

ach

Leve

l of J

ob R

anks

18%

1%



Figure 2.5Bavelas-Leavitt Group Structures and Metrics

A A

A

A

B B

B B C C

C C

D D

D

D E E E

E

WHEEL (32.0 sec)

N NC Happy

A 1 100 37.5

B 1 100 20.0

C 4 25 97.0

D 1 100 25.0

E 1 100 42.5

Avg 1.6 85.0 44.4

Most Distributed Leadership

(slow, happy)

Most Centralized Leadership

(fast, unhappy)

Y-NETWORK (35.0 sec)

N NC Happy

A 1 100 46.0

B 1 100 49.0

C 3 33 95.0

D 2 50 71.0

E 1 100 31.0

Avg 1.6 76.7 58.4

CHAIN (53.2 sec)

N NC Happy

A 1 100 45.0

B 2 50 82.5

C 2 50 78.0

D 2 50 70.0

E 1 100 24.0

Avg 1.6 70.0 59.9

CIRCLE (50.4 sec)

N NC Happy

A 2 50 58.0

B 2 50 64.0

C 2 50 70.0

D 2 50 65.0

E 2 50 71.0

Avg 2.0 50.0 65.6

The four networks are from the Bavelas-Leavitt experiments on leadership in task groups. The WHEEL is a traditional bureaucracy in which C is in charge. The other three networks involve distributed leadership (all five people in the CIRCLE; B, C, and D in the CHAIN; C and D in the Y-NETWORK). More distributed leadership is associated with more messages (M), slower task completion, and greater enjoyment (E). Speed, messages, and enjoyment scores are from Leavitt (1951). Number of contacts (N) and network constraint (NC) are computed from binary ties in the sociograms (number of contacts equals number of non-redundant contacts in these structures).

Figure 2.6Behavioral and Opinion Correlates of Network Brokers

A. Network brokers tend to distribute answers, people in moderately constrained positions tend to be conduits for informational messages.

Data are from Leavitt (1949: Table 30, following page 62).

B. Network brokers are least happy initially, but eventually become the most pleased with the experience.

Data are from Leavitt (1949:Table 29, pages 60-61; "How did you like your job in the group?).

C. By the end of the experiment, network brokers are most likely to be recognized as the unofficial group leader.

Data are from Leavitt (1949: Table 8, page 38; “Did your group have a leader? If so, who?”).

Network Constraint

Mea

n En

joym

ent S

core

Network Constraint ( )

Tim

es C

ited

as G

roup

Lea

der

Network Constraint

Mea

n M

essa

ges

Sent

Answer messagesInformation messages

Enjoyment after first trialEnjoyment after last trial

Figure 2.7 Returns to Brokerage Contingent on Network Status

Compensation and network constraint scores are averaged within five-point intervals of constraint. Correlations are for averages in the graph. "High" status is above median. Adapted from Burt and Merluzzi (2013:Figure 2).

Z-Sc

ore

Com

pens

atio

n

Network Constraint onHR Officers in a Large

Commerical Bank

r = -.96 forhigh statuspeople

r = -.03 forlow statuspeople

Z-Sc

ore

Com

pens

atio

n

Network Constraint onManagers in an Asia-Pacific

Software Launch

r = -.98 forhigh statuspeople

Figure 2.8 Reputation Alone Can Provide

the Social Standing that Facilitates BrokerageGraph A plots investment banker reputation by levels of network status. Reputation is measured by average colleague evaluation. Boxes span 25% to 75% with bold horizontal at the mean. Whiskers extend down to minimum reputation,

up to maximum. Graph B shows z-score annual compensation decreasing with banker lack of access to structural holes. Compensation and constraint scores are averaged within five-point intervals of network constraint. Correlations are for averages

in the graph. Positive versus poor reputation this year are top and bottom third of banker reputation scores this year.

Network Status(eigenvector score / mean score)

Network ConstraintZ-

Scor

e C

ompe

nsat

ion

(tota

l ann

ual)

Ban

ker R

eput

atio

n(m

ean

colle

ague

eva

luat

ion)

A. High Status is a Good Signalof Positive Reputation, but LowStatus Is an Ambiguous Signal

B. Positive Reputation is Enoughfor High Returns to Brokerage

(ignoring status in the organization)

Positive reputation

Poor reputation

r = -.27

r = -.74

Figure 2.9 Diagnostic Contingency in Three Organizations

Z-Sc

ore

Rel

ativ

e C

ompe

nsat

ion

Z-Sc

ore

Rel

ativ

e C

ompe

nsat

ion

Z-Sc

ore

Rel

ativ

e C

ompe

nsat

ion

Z-Sc

ore

Rel

ativ

e C

ompe

nsat

ion

A. Leader DevelopmentAll But One Division of Firmr = -.36, t = -5.66, P < .001

The One Other Divisionr = .09, t = 1.05, P = .30

Early

Pro

mot

ion

(in y

ears

)

B. Merger & Acquisition C. Diversity

Acquiring Managementr = -.40, t = -4.92, P < .001

Acquired Managementr = .11, t = 1.06, P = .29

Women and Junior Menr = .30

t = 3.38P < .01

Senior Menr = -.40

t = -5.56P < .001

Network ConstraintNetwork Constraint Network Constraint Network Constraint

Early

Pro

mot

ion

(in y

ears

)

Figure 2.10 Closure for Stronger, More Durable Relationships

Graphs describe 46,231 colleague relations with analysts and bankers over a four-year period. Graph A distinguishes relations on the horizontal axis by number of mutual contacts this year. Vertical axis is the proportion of relations cited next year as good or outstanding. Dashed lines connect averages and regression line is solid (z-score test statistics in parentheses are adjusted for autocorrelation between an employee's

evaluations). Graph B shows the probability that a relationship cited this year will not be cited next year. Adapted from Burt (2010:174, 181).

B. And LessLikely to Decay

Prob

abili

ty R

elat

ion

Dec

ays

befo

re N

ext Y

ear

(rel

atio

n ci

ted

this

yea

r is

not c

ited

next

yea

r)Relationship Duration

(in years, through this year)

A. Relations More Positive with Relational and Structural Embedding

Prob

abili

ty th

at R

elat

ions

hip

is C

ited

Nex

t Yea

r as

Goo

d or

Out

stan

ding

All Colleagues(z = 14.88)

Continuing Colleague(first cited two years

ago, z = 0.81)

Number of Mutual Contacts Linking Employee with Colleague this Year

10 ormore

Bridge Relationships(employee and colleaguehave no mutual contacts)

StronglyEmbeddedRelationships(employee and colleague sharesix or more mutual contacts)

92% decay infirst year

53% decay in first year

Figure 2.11 Closure Essential to Reputation

Graph A plots analyst and banker reputations this year versus next. Squares are analysts (r = .55, t = 9.78), and circles are bankers (r = .61, t = 13.16). Graph B describes for the bankers subsample correlations between positive (above average)

and negative (below average) reputations this year and next year. Adapted from Burt (2010:162, 166).

Mean Number of Third PartiesConnecting Banker with

Colleagues This Year

Mea

n C

orre

latio

n fo

rB

anke

r’s R

eput

atio

nfr

om th

is Y

ear t

o N

ext

(13-

pers

on s

ubsa

mpl

e)

B. Disappears Without ClosureR

eput

atio

n N

ext Y

ear

(ave

rage

eva

luat

ion

by c

olle

ague

s)

Reputation This Year(average evaluation by colleagues)

Bold line through white dots describes aboveaverage reputations (8.1 routine t-test). Dashedline through black dots describes reputationsaverage and below (6.1 routine t-test).

banker banker

1

2

3

4

1 2 3 4

1

2

3

4

1 2 3 4

A. Stability from Year to Year

AnalystsBankers

10 ormore

Figure 2.12Positive versus Negative Indirect Connections

through Mutual ContactsAdapted from Burt (2008:35).

A. Positive B. Negative

Colleague Banker Colleague Banker

+++

MutualContactPhilippe

MutualContact

Catherine

+- - -

-

MutualContactEmile

MutualContact

Marc

Figure 2.13 Essential Closure Is Around Contacts, Maintaining the Reputations of Brokers and People in Closed Networks

Vertical axis is same as in Figure 2.11B. Horizontal axis is average number of third party connections in the networks around banker's contacts (rounded to nearest whole number). Brokers are bankers with below-median network constraint this year. Regression lines in graph go through averages. Regression equations estimated from 894 year-to-year banker transitions. Test statistics are adjusted down for correlation between repeated observations of the same bankers using the "cluster" option in Stata.

Mea

n C

orre

latio

n fo

rB

anke

r’s R

eput

atio

nfr

om th

is Y

ear t

o N

ext

(13-

pers

on s

ubsa

mpl

e)

1

2

3

4

1 2 3 4

1

2

3

4

1 2 3 4

Mean Number of Third PartiesConnecting People in the Networksaround Banker’s Contacts this Year

banker banker

10 ormore

Brokers (8): Y = .248 + .202 log(X), n = 894, t = 13.0

Other (J): Y = -.047 + .274 log(X), n = 897, t = 15.1

Predictor Salary Prediction Standard Error Test Statistic Manager 1 -‐$35707 $3,498 -‐10.21

Manager 2 -‐$19892 $3,479 -‐5.72

Manager 3 (reference) . . . . . . . . .

Senior Manager $15484 $4,143 3.74

Execu?ve $61930 $4,835 12.81

Network Constraint -‐$1 $38 -‐.04

Constraint x Mgr2 -‐$47 $58 -‐.82

Constraint x Mgr3 -‐$159 $59 -‐2.71

Constraint x Senior Manager -‐$216 $84 -‐2.58

Constraint x Execu?ve -‐$697 $132 -‐5.29

Table 2.1 Returns to Brokerage Increase with Job Rank

Note — These are regression results for job rank and network constraint predicting dollars of annual salary for supply-chain managers in a large electronics company. Network effects are dollars of salary lost in association with a one-point increase in network constraint. The regression equation contains additional controls in the published analysis (Burt, 2004:371, Model 2, R2 = .83).

Reputation Questions:

Closure Creates Bandwidth (information non-reactive to channel)

Closure Creates Echo (information reactive to channel)

1. What makes ego’s reputation persist?

Ego’s consistent behavior, on which others are informed. Wide bandwidth in a closed network enhances information distribution and consistency.

Consistent stories circulating about ego’s behavior. The echo produced by etiquette enhances story distribution and consistency in a closed network.

2. Therefore, who owns ego’s reputation?

Ego does. It is defined directly and indirectly by ego’s behavior.

They do. It is defined by people gossiping about ego. Reputation quickly decays in open networks.

3. What implications for building reputation?

Behave well and get the word out. Put a premium on projects likely to be talked about.

4. How many reputations does ego have?

One, defined by ego’s behavior. Variation can exist from imperfect information distribution or conflicting interests, but variation is resolved by finding the true, authentic person inside ego.

Multiple, depending on gossip. Ego has as many reputations as there are groups in which ego is discussed. The reputations can be similar, but they are generated and maintained separately.

Table 2.2 Reputation Implications of Bandwidth Versus Echo

Documents

NETWORK STRUCTURE OF ADVANTAGE · 2020. 1. 6. · 2. Network Structure of Advantage, 2013-07-10 DRAFT, Page 4 phrases to reference whole paragraphs of text with which colleagues are