Explaining unequal returns to social capital among entrepreneurs Mabel L. Abraham

advertisement
Explaining unequal returns to social capital among entrepreneurs
Mabel L. Abraham
∗
mlba@mit.edu
November 2014
Please do not cite or circulate without the author's permission
Abstract
In this paper, I challenge the dominant claim that women generate less value from using social
ties because they tend to be embedded in less resource-rich networks. Although lacking access to
valuable others constrains actors' ability to exploit network resources, it is unclear whether men and
women embedded within the same networks would generate equal benets. This study leverages a
unique opportunity to compare men and women with access to the same social capital to address this
question. I study business referral networks to examine how referrals, or connections to potential new
clients, are exchanged among entrepreneurs. I nd a gender dierence in the referrals men and women
receive from using social ties exists under conditions where a resource-holder anticipates preferences of
a third-party contact. Simply stated, men and women are equally likely to receive business directly
from their network contacts, however women receive far fewer connections to the clients, family, and
friends of their network contacts, or third-party referrals. I further nd that women in male-dominated
occupations receive the greatest penalty. This study suggests a new network mechanism explaining gender
inequalityanticipatory third-party biaswhere expectations that a client, friend, or family member has
a preference for men over women leads actors to disproportionately exchange resources with male network
contacts.
∗
Massachusetts Institute of Technology, Sloan School of Management. 100 Main Street, Cambridge, MA 02142. This research has
beneted from the comments of Roberto Fernandez, Susan Silbey, Emilio Castilla, Ray Reagans, Ezra Zuckerman, Kate Kellogg,
Sameer Srivastava, Gino Cattani, Tristan Botelho, Julia DiBenigno, Aruna Ranganathan, Phech Colatat, and numerous seminar
and conference participants. Rosa Taormina provided excellent research assistance. I also thank the Kauman Foundation and
the American Association of University Women for their generous nancial support. All remaining errors are my own.
1
Introduction
Despite fty years of political mobilization and dedicated public policy, American working women maintain
a consistently lower economic position relative to similarly qualied men (e.g., Bird and Sapp, 2004; Elvira
and Graham, 2002; Petersen and Morgan, 1995).
Women are particularly disadvantaged in terms of
entrepreneurial outcomes. While there have been recent increases in the rate at which women establish new
businesses (U.S. Small Business Administration, 2001), women own less than 30 percent of all U.S. rms
(U.S. Department of Commerce, Economics and Statistics Administration, 2010, p. 1) and women-owned
businesses continue to be smaller and less lucrative than rms owned by men (U.S. Department of Commerce,
Economics and Statistics Administration, 2010; see also Bird and Sapp, 2004; Hundley, 2001; Loscocco and
Robinson, 1991; Merrett and Gruidl, 2000; Robb and Coleman, 2009). Since individuals commonly choose
entrepreneurship as a means for career attainment and economic growth (cf. Sorensen and Sharkey, 2014),
uncovering the mechanisms contributing to gender dierences in entrepreneurial outcomes is of particular
importance.
These gender dierences, coupled with the central role of social networks for entrepreneurs (e.g. Hallen,
2008), have resulted in a burgeoning literature focused on the role of social ties for the pervasiveness of
gender inequality in entrepreneurship. As Stewart (1990: 149) stated, For many entrepreneurs, their most
signicant resource is a ramifying personal network. But like the gender dierences in the social networks
of men and women more broadly, it has been argued that female entrepreneurs have less valuable networks
(e.g., Aldrich, 1989; Aldrich, Elam, Reese, 1997; Aldrich, Reese, and Dubini, 1989; Cromie and Birley, 1992;
Renzulli, 1998; Ruef, Aldrich, and Carter, 2003). The sorting of women into lower status occupations and
organizational positions embeds them in networks with fewer high status and inuential contacts than those
of men (Brass, 1985; Campbell, 1988; Ibarra, 1997; Marsden, 1987; McGuire, 2000, 2002; Moore, 1988,
1990; Smith, 2000) and, as a consequence, women generate lower returns from social capital than do men
(Braddock and McPartland, 1987). Because high status actors occupy central positions that facilitate access
to network resources (Lincoln and Miller, 1979), people with higher status contacts receive greater advantages
when using their social ties. Thus, women's dierential network structures and composition is posited as
a key factor in the pervasive under-representation and lower success rates of women among entrepreneurs
2
(Aldrich, 1989; Cromie and Birley, 1992; Katz and Williams, 1997; Renzulli, Aldrich, and Moody 2000).
Although lacking access to others' resources constrains women's opportunities, it is unclear whether men
and women with the same access to resources through social ties would receive equal benets. The mere
presence of social ties does not necessarily guarantee that resources will be exchanged. For networks to yield
benets, actors possessing resources must decide to share those resources with network contacts, in eect,
behaviorally activating their social ties (Lin 2001; cf. Smith, 2005). In other words, resource-holders serve as
evaluators deciding with whom to share resources, or activate a social tie. Despite the central role of resource
exchange within networks for generating benets from social ties, we do not have a clear understanding of
how gender aects resource sharing, and whether this resource allocation process has implications for gender
inequality. Scholars have argued that in many situations diuse status characteristics such as gender inuence
evaluations of actors in a way that disadvantages women (e.g., Castilla and Benard, 2010; Castilla, 2008;
Correll and Benard, 2006; Correll, Benard and Paik, 2007; Ridgeway, 2011; Turco, 2010), suggesting that
gender may also play a role in how resources are exchanged among network contacts. These gender-based
expectations have been found to be particularly salient in male-dominated areas, such as entrepreneurship
(cf.
Gupta et al., 2009).
Female entrepreneurs are commonly perceived to be less competent (Buttner
and Rosen, 1988; Thébaud, 2010) and to lack credibility (Carter and Cannon, 1992; Moore and Buttner,
1997). Thus, to develop a more complete understanding of how networks contribute to gender inequality
among entrepreneurs, it is critical to examine how the process of resource allocation within networks impacts
gender inequality. This research moves beyond structural network accounts of gender inequality to determine
whether men and women with the same access to social ties receive equal benets, or returns to social capital.
In this paper, I show that women receive fewer benets through social ties than do similar men in the
same social structure. I study business referral network groups to examine how referrals, or connections to
1
potential new clients, are exchanged among entrepreneur-members . A comparison of the referrals accruing to
men and women with the same access to social ties reveals that female entrepreneurs receive fewer business
referrals.
Consequently, this study identies that access to valuable social ties alone will not necessarily
eliminate gender dierences in outcomes produced through social ties. Resource allocation plays a role in
1I
use entrepreneur-members, members, network group members, and network contacts interchangeably to refer to members of
these business referral groups. To dierentiate between the person deciding whether to give a referral to a fellow member and
the potential recipient of these referrals I use resource-holder and resource-seeker respectively.
3
perpetuating gender inequality as women receive fewer benets through social ties relative to similar men
with access to the same social ties.
Further, I posit a new network-based mechanism to explain this gender inequality, which I term anticipatory
third-party bias. I show that while members do not favor men when directly connecting to other members in
their network group, they do exhibit a male preference when connecting these fellow network group members
to their outside contactsclients, friends, and family members. Where entrepreneur-members expect that a
client, friend, or family member has a preference for men over women, they disproportionately give referrals
to male rather than female network group members.
In other words, men and women are equally likely
to receive business directly from their network contacts, but women receive far fewer connections to the
clients, family, and friends of their network contacts, or third-party referrals.
Further, I nd that this
gender dierence in receiving third-party referrals is most pronounced among women in male-dominated
occupations.
Thus, despite having equal access to network resources through social ties, women receive
lower returns to their social capital because resource-holders respond to what they believe are the gender
preferences of others, or third-parties.
To demonstrate how resource allocation, rather than dierential network composition, contributes to
gender inequality, this study compares the resources acquired through social ties by similar men and women
embedded in the same network groups. It is only possible to identify whether and how resource exchange
within networks aects gender inequality by holding network structure constant. In this respect, the tendency
for networks to be gender-segregated poses an empirical challenge. To overcome this challenge, I collected six
years of referral data from a popular business networking club organized to help entrepreneurs generate new
business through social ties. This setting oers a unique opportunity to observe how resources are exchanged
among male and female entrepreneurs in the same networks. Entrepreneurs in these clubs represent a wide
range of occupations and meet weekly to exchange business referrals that connect recipients with potential
clients. I unobtrusively collected all referral data, observed three groups for two years, and interviewed 18
entrepreneur-members. In my analysis, I focus on 416 cases where a member exits a group and is replaced
by a new member in the same occupation.
Leveraging these replacements, particularly those involving a
change in gender, allows for a comparison of men and women in the same occupations, with access to the
4
same social ties, thus providing better identication of any gender dierence in returns to social capital.
I proceed as follows. First, I develop hypotheses about whether and why women receive fewer resources
than do men embedded in the same social networks.
I explain how gender preferences, or performance
expectations, may prevent women from generating the same returns to social capital, despite having the
same access to social ties, as men. Second, I introduce the research setting and analytical method used to
test whether women receive lower returns to social capital and to identify the role of expectations about
third-party preferences in perpetuating gender inequality. Third, I present results which demonstrate that
actors' anticipation about the gender preference of their contacts prevents sharing referrals with women
under certain conditions.
This nding suggests a new mechanism by which networks perpetuate gender
inequality: anticipatory third-party bias in social capital activation. Therefore, access to valuable social ties
alone will not necessarily mitigate gender dierences in outcomes from using social ties. I conclude with a
discussion of the theoretical and organizational implications of the research ndings for our understanding
of how social networks contribute to gender inequality among entrepreneurs and more broadly.
THEORY
Generating network advantages through activation
Social networks are often conduits of valuable resources, such as information and inuence, which can be
leveraged to achieve desirable outcomes (Burt, 1995; Granovetter, 1995; Lin, 2001; Podolny and Baron,
1997). As such, social capital refers to the potential and actual set of resources made available through an
actor's direct and indirect social ties to others (Bourdieu, 1986; Coleman, 1988; Lin, 2001). The advantages
associated with social ties, however, are not automatic.
An actor's network structure and composition
provides the potential set of resources available through social ties, but within this structure actors exercise
agency, and preferences (Emirbayer and Goodwin, 1994; Gulati and Srivastava, 2014; Zeng and Xie, 2008).
The mere presence of social ties alone does not necessarily mean that resources will be exchanged. In his
theory of social capital, Lin (2001:29) makes a clear distinction between resources that are available through
social ties and resources that are accessed and/or mobilized in purposive action . Therefore, for networks
5
to yield benets, social capital must be both available and activated such that a resource-holder shares
resources with a network contact (Lin, 2001; cf. Smith, 2005; Marin, 2012). In this way, network structure
and composition constrain choice (Zeng and Xie, 2008) by providing the potential social capital, or pool of
resources, to which actors have access and the opportunity to exploit.
While we still lack a comprehensive understanding of how people use social ties to generate benets (e.g.,
Ibarra, Kildu and Tsai, 2005), recent research has shown that individuals often activate only a subset of the
social relations to which they have access. For example, limitations in the cognitive recall of relationships
(Smith, Menon, and Thompson, 2012) and reputation concerns (Smith, 2005; Marin, 2012) may prohibit
actors from activating social ties.
In considering factors that may prevent resource-holders
2 from giving
resources to particular network contacts, labor market studies examining when people provide their social
ties with job information are particularly illustrative.
For example, in a study of the black urban poor,
decisions about whether to share job-related information with potential job-seekers are largely inuenced by
expectations of how doing so will impact the information holder's reputation (Smith, 2005). Similarly, among
professional workers, information holders protected their reputation with their social ties by refraining from
sharing information about job opportunities unless their contacts had explicitly asked or the job was a clear
match (Marin, 2012). The importance of activation for understanding how social ties generate benets has
been clearly established. Some scholars have begun to examine how activation contributes to inequality in
benets (e.g. Smith, 2005), however, extant research has not focused on gender inequality.
The role of resource-holders' gender beleifs in evaluations
In social networks, resource-holders act as evaluators assessing the quality of network contacts before
deciding whether to share sought-after resources (cf.
Smith, 2005).
Ascriptive characteristics, such as
gender, inuence the evaluation of actors in ways that generally disadvantage women (e.g., Botelho and
Abraham, 2014; Castilla 2008; Castilla and Benard, 2010; Correll and Benard, 2006; Correll, Benard and
Paik, 2007; Ridgeway, 2011; Turco, 2010). Commonly, evaluators are unable to observe the actual quality
2 Choice
can be exercised by either the resource holder in deciding whether or not to share resources (Smith 2005; Marin 2012)
or the resource seeker in deciding whether or not to express needs (Hurlbert, Haines, and Beggs 2000; Srivastava, 2014). In my
setting, all actors are similarly seeking and expressing needs for resources, which I discuss further in the results section. Thus, I
focus on the role of the resource holder in deciding whether to share resources with others and how this process impacts gender
dierences in the benets men and women receive through social ties..
6
of the individuals being assessed, forcing them to rely on more accessible indicators of expected quality
(Podolny, 1993, 2005), often historically marked by gender, race, religion and social class.
Both status-based and statistical theories of discrimination posit that, because gender is conventionally
understood to convey information about an actor's quality, or expected performance, evaluators incorporate
gender into their assessments to remedy this problem of uncertainty, that is, lacking knowledge of actual
quality. Status-based preferences stem from beliefs that gender is a status characteristic, where men are more
highly valued, and generally perceived as more competent, than are women (Berger et al., 1977; Correll and
Ridgeway, 2003). These performance expectations are often incorporated into evaluations of male and female
actors (Correll and Benard, 2006) such that men's performances are evaluated more positively than those
of similar women (e.g., Ridgeway and Correll, 2004; Ridgeway and Smith-Lovin, 1999; Wagner and Berger,
1993) leading to a male social advantage (Webster, Murray and Hysom, 1998). Similarly, economic theories of
statistical discrimination posit that if quality or performance is dicult to observe directly and has dierent
distributions for men and women, it may be rational for an evaluator to prefer men (for reviews, see Correll
and Benard 2006; England 1994, pp. 6063). Therefore, if an evaluator has had more positive experiences
with men than with women, it would be logical to use this information when comparing a male and female
candidate to conclude that the man is more likely to be of higher quality.
However, reliance on gender as signal of quality may persist even when actual, more relevant performance
information is freely available (Botelho and Abraham, 2014; Berger et al., 1977; Correll and Ridgeway, 2003).
Double standards theory, a particular status-based account of discrimination, suggests that gender serves as a
distinguishing characteristic in assessments of competence even when men and women with equal performance
are compared (e.g., Foschi, 1989). According to this perspective, discrimination arises because stereotypes
about the relative expected performance of men and women inuence evaluations beyond freely available
evidence of equal past performance, particularly when the evaluator is sorting through a large number of
candidates (e.g. Botelho and Abraham, 2014).
While the role of gender for network activation remains unclear, these theories of discrimination suggest
that resource-holders may favor men when assessing whether or not to share resources with network contacts.
In job search, for example, using networks tends to channel women into low-wage, female-dominated jobs
7
because women commonly receive job information from female contacts (Drentea, 1998, Hanson and Pratt,
1991; Mencken and Wineld, 2000).
Based on this observed association, it is often assumed that women
lack access to social contacts possessing more valuable job information. It is plausible, however, that despite
having access to high status network contacts with information about better jobs, women are not receiving
information from these contacts. In other words, if individuals with valuable job information are choosing
to share this information with their male, but not female, contacts, women may be relegated to relying on
a subset of their network for job information.
Thus, resource-holders may take gender into account when assessing whether or not to share resources
with other entrepreneurs in their network such that:
Hypothesis 1:
Female entrepreneurs will receive fewer resources from network contacts than
will similar male entrepreneurs in the same network.
The role of third-parties in evaluations
The arguments presented so far highlight how resource-holders' gender beliefs or stereotypes may drive them
to disproportionately favor male network contacts over comparable women. This challenges the assumption
that men and women with access to the same social ties will receive the same benets.
Alternatively,
expectations about the gender preferences of relevant others, irrespective of an evaluator's personal beliefs
or preferences, may also lead an evaluator to favor men. Networks are social structures where actors may
be inuenced not only by their personal beliefs, but also by how they believe their actions will be perceived
by a third party.
At the most basic level, the exchange of resources between network contacts, or social
capital activation, involves at least two social actors.
In many cases, however, the sharing of resources
within networks is a triadic process (cf. Rubineau and Fernandez, 2014). In addition to a resource-holder
and a resource-seeker, there are often third-parties such as other actors or organizations (e.g. employers),
involved in, or observing, resource exchanges between network contacts. For example, the exchange of job
information involves an information holder who knows of a vacancy, a job seeker receiving the information,
and the employing organization looking to ll that vacancy.
In this example, the employer represents a
third-party involved in the direct sharing of job information between two network contacts.
8
Generally, third-party observers, or audiences, play an important role in shaping actor behavior. The
presence of an audience serves to impose expectations of what is considered appropriate behavior in that
particular context (Zuckerman et al., 2003). This is especially pronounced when the consequences, success
or failure, of an actor's decision is contingent on the reaction of others. Under these conditions, a decision
maker is challenged to satisfy not her personal quality criteria, but rather the expectations of her audience
(Clark, Clark, and Polborn, 2006; Emerson, 1983; Jensen, 2006; Ridgeway and Correll, 2006). For example,
employment agencies, acting as hiring intermediaries, are strongly motivated to anticipate their client'sthe
hiring rm'sideal candidate in order to select the most appropriate applicants from the overall pool
(Fernandez-Mateo and King, 2011).
Specically, if decision makers know the conclusion that their audience wants them to reach, they tend to
bias their choices in an eort to satisfy the audience (Lerner and Tetlock, 1999). When there is uncertainty
about a third-party's preferences, third-order inferences about status, or assessments of who or what most
other people believe to be of higher status, suggest that it is likely for this actor to prefer men (Correll,
Ridgeway, Zuckerman, Jank, Jordan-Block, 2014).
Thus, insofar as a decision-maker is dependent on
approval from a third-party and anticipates that the third-party is aware of the status structure (i.e. men
higher status), it may be rational for a decision-maker to favor high-status options, even when the audience's
true preferences are unknown and the decision-maker personally does not endorse these beliefs.
There is evidence suggesting that anticipating the preferences of an audience, or third party, may serve
to perpetuate gender inequality.
Economists have long posited that inequality in labor market outcomes
is at least partly a function of customer discrimination (e.g.
Neumark, Bank, and Van Nort, 1996).
In
terms of gender inequality in hiring, expectations of customer or client preferences for male employees leads
hiring agents to disproportionately select male candidates (Fernandez-Mateo and King, 2011).
Similarly,
the likelihood that a female attorney is promoted to partner is signicantly greater when the law rm has
women-led clients (Beckman and Phillips, 2005).
Therefore, in social networks, expectations about the gender preferences of relevant third-parties is
proposed to play a role in resource-holders decisions of whether to share resources, such that:
Hypothesis 2:
Female entrepreneurs will receive fewer resources from network contacts than
will similar male entrepreneurs in the same network when the exchange of resources involves
9
a third-party.
Importantly, the proposed role of third-parties and the accountability they evoke can exist both under
conditions where the resource-holder has personal status beliefs or biases and under conditions where the
resource-holder has more egalitarian preferences (c.f.
Correll et al.
2014).
For anticipatory third-party
biases to play a role, the success of the resource-holder's social capital activation decision must simply be
contingent on approval from a third-party.
Gendered occupations and bias
The degree to which a third-party has a preference for men is likely to be context dependent. More generally,
the potential for bias is magnied when an actor occupies a role that is incongruent with his or her gender
(Foschi, 1989). The notion of the ideal worker (Acker, 1990; Williams, 1999) oers an explanation for the
heightened use of gender in evaluations of actors in gender incongruent roles. The ideal worker image denes
the characteristics of the individuals who are expected to be strong performers (Gorman, 2005). As such,
individuals are commonly evaluated against this norm (Heilman, 1983). To the extent that the ideal worker
image is gendered, it is increasingly likely that evaluations favor members of one gender group over another
as gender is particularly salient in these contexts (cf. Turco, 2010).
Consistent with this view, female disadvantage has been posited to be most pronounced when women are
in roles or perform tasks typed as male (Foschi, 1989) or where they are the numerical minority (Kanter,
1977).
The numerical representation of men and women in a particular occupation is a primary driver
of perceptions that the role is typed as male or female. For example, masculine personality and physical
attributes are perceived as essential for success in occupations historically dominated by men (Cejka and
Eagly, 1999). Some research has shown that evaluators tend to give lower assessments to women occupying
an occupational role that is gender incongruent, or male-typed (Eagly, 2004). As compared to their male
counterparts, women tend to be evaluated as less competent in roles traditionally held by men (Davison
and Burke, 2000; Eagly and Karau, 2002), with their successes commonly attributed to luck rather than
skill (Swim and Sanna, 1996). Therefore, as related to hypothesis 2, the degree to which a female network
contact occupies a role that is gender incongruent is likely to have implications for resource-holder's decisions
10
of whether to share resources when the information sharing involves a third-party such that:
Hypothesis 3a:
When the exchange of resources involves a third-party, there will be a greater
female penalty in receiving resources from network contacts among women in male-dominated
occupations, such that female entrepreneurs will receive even fewer resources relative to similar
male entrepreneurs in the same network when in male-dominated occupations.
It is less clear whether men occupying gender incongruent roles face similar penalties. Male minorities have
been posited to reap benets from their position as tokens ascending to high status positions within these
occupations more quickly than do women (Kmec, 2008; Williams, 1992, 1995). Despite these advantages, men
are posited to encounter discrimination from people outside of their profession (Williams 1992). Therefore, it
may be reasonable for a resource-holder to anticipate that third-party contacts have a preference for women
in female-typed occupations. For example, a resource-holder may expect that his third-party contact has
reservations about a male child-care provider, an occupation traditionally held by women, and as a result,
refrain from making referrals in this case. Further, recent audit studies have suggested that in occupations
where women are the clear majority, such as secretarial positions, applications from women are favored over
applications from men (Booth and Leigh, 2010; Riach and Rich, 2006). Therefore, the degree to which a male
network contact occupies a role that is gender incongruent is likely to have implications for resource-holder's
decisions of whether to share resources when the information sharing involves a third-party such that:
Hypothesis 3b:
When the exchange of resources involves a third-party, there will be a male
penalty in receiving resources from network contacts among men in female-dominated occupations,
such that male entrepreneurs will receive fewer resources relative to similar female entrepreneurs
in the same network when in female-dominated occupations.
One possible explanation for these mixed ndings regarding the outcomes for men in female-dominated roles
stems from the overarching status ordering of gender. In general, gender is a status characteristic, where men
are more higher status and, as a result, generally more highly valued and perceived as more competent, than
are women (Berger et al., 1977; Correll and Ridgeway, 2003). Since gender is a diuse status characteristic
where men are typically seen as the higher status gender, even when men are in female-typed roles, there
may be a higher tolerance for this inconsistency.
Thus, when a resource-holder is assessing the gender
preferences of a third-party, the threshold for assuming a female preference is likely higher. For example,
11
a resource-holder may be more apt to take a chance and refer his client to a male child-care provider than
to a female contractor. As a result, the degree to which a network contact occupies a role that is gender
incongruent is likely to have greater implications for resource-holder's decisions of whether to share resources
for women than for men such that:
Hypothesis 4:
There will be a greater penalty in receiving resources from network contacts for
women in male-dominated occupations than for men in female-dominated occupations when
the exchange of resources involves a third-party.
METHODS
Research Setting and Data
Data for this study were collected from RefClubs, a popular organization providing a forum for entrepreneurs
to exchange information and business referrals.
Entrepreneurs pay annual dues to become members of
RefClubs and gain access to a networking group of other entrepreneurs in the same geographic region. The
primary purpose of the organization is to bring together individuals seeking to grow their businesses and
provide them with a forum for sharing information about potential new clients.
These are instrumental
networking groups where individuals join with the explicit goal of establishing and leveraging social ties with
other members of the group to generate new business.
I have studied the exchange of referrals among entrepreneur-members within 37 individual RefClubs
network groups using archival records including each and all individual exchanges. I collected, coded, and
analyzed records of referrals from these network groups for the years 2007- 2013. To deepen my understanding
of this empirical context, between October 2011 and October 2013 I spent 40 hours observing individual
weekly group meetings across 10 unique groups and conducted 18 semi-structured interviews with individual
entrepreneur-members.
The quantitative data I analyze include complete information on the referrals generated and received
by each of the 2,310 members in these 37 unique network groups. These data oer two key advantages for
12
testing the hypotheses I have outlined. First, through analysis of network records, I unobtrusively observed
the actual exchange of resources among entrepreneurs allowing me to move beyond common approaches
relying on self-reports. Self-reports have been found to be tainted by the subjective perceptions of survey
respondents (Bernard et al., 1981; Quintane and Kleinbaum, 2011) calling into question the reliability of
network surveys (Bernard, Killworth, and Sailer, 1981), although recently, studies have made progress in
avoiding respondent biases by using email exchange records (e.g. Kleinbaum, 2012; Kleinbaum et al. 2013;
Srivastava, 2014).
Second, in these referral networks, men and women are co-located within the same network groups and
thus have the same structural opportunity to access resources. This allows me to hold network structure
constant and identify how resource exchange within networks aects gender disparities. Without knowledge
of the opportunity structure for accessing resources, or the set of social ties that could have been activated, it
is not possible to distinguish between structure- and activation-based explanations. Self-reports, for example,
do not capture the entire network, or the opportunity structure, from which resources are generated. Without
holding constant network structure, observations of women's reduced benets could be the result of women's
dierential access to valuable contacts or of women's inability to mobilize social ties (see McDonald (2011)
for a similar critique questioning whether feminized networks are less valuable).
-INSERT FIGURE 1 ABOUT HERE-
Figure 1 depicts the structure and composition of these network groups. Within each group there are
an average of 40 entrepreneur-members.
The gender composition of these network groups mirrors that
of women's presence in entrepreneurship more generally (e.g., U.S. Department of Commerce, Economics
and Statistics Administration, 2010, p.
1), with approximately 35 percent of members being female.
Entrepreneur-members work in a wide range of occupations, as illustrated in Figure 2A. Within each network
group, there is only one member from each occupational specialty, thus reducing the role of competition
among members in referral exchange.
The specication of occupational specialty, as opposed to broader
occupation or profession categories depicted in Figure 2A, is deliberate as members may overlap in the
latter. For example, a group may have two members from the legal occupation, both of whom are attorneys,
where one specializes in estate planning and the other in family law. While these members are in the same
13
broad occupation, they are not in competition as they are not vying for the same customer base.
The
non-competing nature of these groups reduces the likelihood that observed gender dierences in referral
patterns result from men and women being dierentially impacted by competition.
Weekly group meetings have a routinized format for entrepreneur-members to learn about each other's
businesses and to exchange business referrals.
Meetings run from 90 to 120 minutes and are scheduled
outside of normal business hours, typically held either in the early mornings or evenings. This time is largely
structured with an opportunity for informal interaction among members before the start of the meeting.
During meetings each member is allotted a time slot to speak publicly to the group about specic business
and referral needs. This short presentation provides each member the opportunity to educate other members
about their business in order to facilitate referral exchange. Since all members partake in this request for
referrals, coupled with the instrumental nature of these groups, this setting minimizes concerns that observed
gender dierences are due to women's tendency not to ask for help (e.g., Babcock, 2003).
It is also during these weekly meetings that business referrals, or information about potential new
clients, is exchanged among members.
Referrals generated in these groups provide a non-trivial source
of revenues for members. On average, each member receives approximately 23 referrals per year contributing
to approximately $10,000 in new business per member per year. Each member has the opportunity to activate
a tie with, or pass information about a potential new client to, another member during the meetings. Making
a referral to another member involves completing a form where a copy of this information goes directly to
the member receiving the referral and a second copy is entered into the groups' records by an administrator.
In addition, all business referrals exchanged between members and successful business transactions that
resulted from earlier referrals are announced verbally to the group.
Through this process of publicizing
resource exchange, members may become aware of the relative value that each other member brings to the
group in terms of generating referrals and business revenues.
There are two distinct types of business referrals exchanged among entrepreneur-members, which I refer
to as direct use and third-party referrals.
In direct use referrals, a member expresses interest in directly
hiring another member in the group to provide a service. For example, the accountant in Group A (Figure
1) indicates to the real estate agent, also in Group A, that she would like to hire him to help her nd new
14
oce space. Since direct use referrals reect the personal consumption of members, these referrals represent
the referring member's, or resource-holder's, personal gender preferences or beliefs. Alternatively, third-party
referrals involve a member providing another member with information about a third-party contactclient,
family member, or friendwho is not a member of the group but may be interested in hiring the member.
In third-party referrals, the accountant activates the tie to the real estate agent by connecting the real
estate agent to one of her clients, whom the real estate agent would not have access to otherwise. Therefore,
third-party referrals are subject to expectations of the gender preferences of a referring member's third-party
contact.
-INSERT FIGURE 2A AND 2B ABOUT HERE-
Empirical Strategy
To identify whether women receive fewer referrals than men with access to the same social ties, the natural
approach would be to count the number of referrals male and female entrepreneurs receive and estimate a
negative binomial regression with gender as the main predictor variable of interest. By design, the empirical
setting I use in this study provides an opportunity to compare men and women co-located within the same
networks. In this setting all members within an individual networking group have the same opportunity to
obtain business referrals. As a result, I am able to hold network structure constant by including network
group xed eects allowing for a comparison of the number of referrals men and women within the same
network group receive.
In addition, to account for potential gender dierences in contributions to other
members and participation in the group, this model would include relevant controls including number of
referrals given to others, attendance, and tenure for each entrepreneur-member.
However, this empirical approach would fail to consider the possibility of occupational sorting.
It is
well-established that men and women tend to be sorted into dierent occupations (Fernandez and Sosa,
2005), particularly among entrepreneurs where men and women historically own businesses in dierent
industries (Rosenfeld, 2002).
As Figure 2B reveals, the representation of female entrepreneur-members
varies across occupations with women being underrepresented in some and overrepresented in others. Since
15
approximately 35 percent of members are female, female members are underrepresented in occupations at the
far bottom-left of Figure 2B and overrepresented in occupations at the top-right. Thus it is plausible that,
despite having access to the same social capital, observations that women receive fewer referrals than do men
in the same group are a function of occupational sorting. For example, entrepreneur-members may refer to
female members in their group less frequently not because they are women but rather because there is less
demand for the particular businesses that women tend to represent, such as education and training, than
for businesses dominated by men, such as construction and trades. As aforementioned, within any single
group in my setting there can only be one member from each detailed occupational category.
Therefore,
simply comparing the number of referrals men and women receive does not directly allow for a comparison
of men and women in the same occupations who also have access to the same social ties. One way to begin
to take occupational sorting into account is to compare men and women in the same occupations across
network groups using occupation xed eects. While this approach accounts for occupational heterogeneity,
and reveals a similar female penalty in referrals received (see Table 4, Model 3), it does not compare men
and women with the same access to social ties.
To more carefully compare similar men and women, I use a unique identication strategy that simultaneously
accounts for occupation and network group.
I analyze cases where a member exits from a group and is
replaced by a member in the same detailed occupation ( replacers ). This oers a natural laboratory for
examining my research question. Leveraging these replacers allows for a comparison of men and women in
the same occupations, with access to approximately the same social ties, providing controlled conditions for
better identication of the gender dierences in returns to social capital where they exist.
While at any
given time each group has only one person in each detailed occupation (e.g. one real estate attorney), once
a member decides to exit the group (i.e. leaver) it is possible for another person in that same occupation
to join (i.e. replacer). As Figure 1 reveals, without focusing on these replacers, achieving either of these
objectives would come at the expense of the other. Therefore, comparing the returns to social capital for
members involved in a replacement event provides an opportunity for cleaner identication of the gender
dierences in returns to social capital.
During my study window, I observed 416 of these replacement events where a member exits a group
16
and is replaced by another member in the same occupational specialty. To ensure that the replacers and
the leavers had access to approximately the same social capital and were employed in the same detailed
occupation, I used strict boundaries for dening replacement events. First, I dene replacements using the
most narrow denition of occupational specialty available. For example, a replacement event is dened as a
real estate attorney leaving and being replaced by another real estate attorney, as opposed to an attorney
being replaced by an attorney. Since men and women are sorted into dierent detailed occupations, using
this more ne-grained occupational denition accounts for gender dierences even within broad occupation
categories. Second, I include only replacement events where the replacers enter a group within 12 months
of the leaver exiting the same group, with a mean length of 133 days between the exit and replacement,
to maximize the overlap in social ties available to each of the involved members. In any 12 month period,
3 resulting in the leaver and the replacer having approximately
greater than 70 percent of members remain,
identical access to resources through social ties in this context.
-INSERT TABLE 1 ABOUT HERE-
Table 1 illustrates the types of replacement events that I observed.
Of the 416 replacement events
meeting the criteria outlined above, approximately 60 percent involved a same-gender replacement. While
these cases alone do not provide a means for comparing the relative number of referrals women receive relative
to comparable men, these provide a baseline for comparing the replacement cases involving a gender switch
enabling me to rule out a key alternative explanation. Without these same-gender replacements, it would
not be possible to disentangle whether an observation that women replacing men received fewer referrals was
due to the replacer's gender or to a newcomer eect. Comparing dierences between the referrals received
by replacers and leavers across the dierent categories provides a means for isolating gender eects from the
3 One
way to improve the overlap in the social ties available to leavers and replacers is to shorten the period between the
leaver exiting and the replacer entering the group. For example, within a six month window approximately 87 percent of
the group remains intact. But it takes groups several months (approximately four months on average) to replace an exiting
member. Therefore, shortening this window results in a drastic reduction in the number of cases of a leaver being replaced by
a new member in the same detailed occupation. The results are substantively robust to the following alternative specications:
shortening the window to a six month period while broadening the denition of same occupation and shortening the window
to a nine month period while using the same ne-grained occupational denitions without the detailed occupation xed eects.
Because the goal of this empirical strategy is to more carefully compare men and women within the same occupations, the
analyses focuses on replacements within the same detailed occupation and occurring within 12 months of a leaver's exit .
17
eects of being a new member to the group.
4
-INSERT TABLE 2 ABOUT HERE-
Importantly, the subsample of entrepreneur-members involved in a replacement event are representative
of the study population in a number of key ways.
Replacements occurred in each of the 37 chapters in
my study population and represent most occupation specialties present.
Unsurprisingly, the occupation
specialties not represented among replacement events are those that are least common in this setting, each
representing less than ve percent of all members.
The gender composition of the members involved in
the replacements is identical to the study population, and to women's presence among entrepreneurs more
generally, with 35 percent being female. Furthermore, the distribution of men and women to occupations
among this subsample mirrors the broader pattern, with most occupations being within two percentage
points. As shown in Table 2, the pattern of referrals received overall and by referral type (i.e. direct versus
third-party referrals) is largely identical among the replacements and the study population. Replacers, on
average, give slightly fewer referrals to, and generate less money for, other entrepreneur-members in their
5
group per year . All models control for the referral giving behavior of the replacer and the leaver to be sure
that these dierences are not aecting the results.
Variables and Empirical Model
The analysis in this study examines whether there are gender dierences in the number of referrals each
member received per year among male and female entrepreneurs in the same network groups. This analysis
uses three dependent variablesdirect use referrals received, or the total number of times a member receives
requests for services directly from other members in the network group for their own consumption (i.e.
personal or business use) per year, third-party referrals received, or the total number of times a member
4 This
does not rule out the possibility that there may be dierences in the tenure and years in a member's life cycle with the
group between the leavers and the replacers. For these dierences to impact my results, the dierences between leavers and
replacers would need to vary across replacement types in a way that led to underestimating the referrals that replacers received
in the male-to-female group but not in the other groups. These dierences do not appear to be sizable nor to dier across
groups. To minimize this concern, I also estimate models limited to replacement events where the absolute dierence in tenure
for the replacer and leaver was less than six months. The results are robust to this alternate specication as presented in the
appendix, Table A1.
5 The observed dierence in referrals given and dollars generated by the replacer group relative to the broader study population is
the result of slight tenure dierences between these groups. Replacers have nearly one year shorter tenure on average. Limiting
the study population to those with the same tenure as the replacer group eradicates this dierence.
18
receives connections to the third-party contacts (e.g. clients) of other members in the network group per
year, and total referrals received, or the total number of direct and third-party referrals a member receives
per year. Dierentiating between direct referrals received and third-party referrals received allows for testing
the mechanism that resource-holder's expectations of the gender preferences of others drive observed gender
dierences, as stated in hypothesis 2. Whereas third-party referrals received involve a third-party, the decision
to make a direct referral is based only on the preferences of the resource-holder. Thus, comparing the referrals
women receive relative to men across these two types of referrals provides a way to dierentiate whether
anticipated third-party gender preferences or personal preferences are driving observed gender dierences.
All models include controls for the following individual characteristicsthe number of total referrals
given by a member per year; the total number of additional meetings a member partook in with other
members in addition to weekly meetings per year; tenure with the group; the total number of weekly
meetings a member missed, or absences, per year; and the total dollar amount a member generated for other
members in thousands, or dollars generated, per year. To examine gender dierences within network group
and occupation, I also include xed-eects for network group and occupation, using the Bureau of Labor
6
Statistics Standard Occupational Classication (SOC) codes, in some models as noted.
The rst set of models leverage the full study population to establish whether there is evidence of an
overall female disadvantage. I estimate negative binomial regression models to predict total referrals received,
which is a skewed count variable.
The remaining analyses focus on the replacement events to more carefully isolate how resource exchange
among social ties dierentially benets men and women. In these models, I compare men and women in the
same detailed occupations, with access to approximately the same social ties, to determine whether there
are gender dierences in the number of referrals men and women receive.
Limiting the analysis to these
replacement events helps to rule out the alternative that observed gender dierences are due to occupational
sorting.
This analysis of the entrepreneur-members involved in replacement events focuses on the dierence in
the average number of referrals received by a replacer and the leaver that was replaced in terms of each
6 In
my main analyses I use the SOC major groups codes. My main results are robust to using the more ne-grained SOC minor
group codes. Because the intercoder reliability for coding occupations into the major codes was higher, 97 percent versus 88
percent, I report results based on models using the SOC major codes.
19
of the three measures of referrals. For each measurethe average number of total referrals received by a
member, the average total number of direct use referrals received by a member, and the average total number
of third-party referrals received by a memberI calculate the dierence in the number of referrals received
between the replacer and the leaver that they replaced as follows,
each of the three types of referrals,
r
indexes the replacer and
l
Y r−l = Y r − Y l ,
where
indexes the leaver.
Yr−l
represents
As such, a negative
value in these dependent variables indicates that the replacer received fewer referrals on average than did
the leaver whom he replaced.
The key independent variables are the four types of replacement eventsmale-to-male, female-to-female,
female-to-male, and male-to-female. I created four dummy variables to capture each of these categories with
male-to-male as the reference group. The eect of each of the three remaining categories represents how
the dierence between the referrals received by a replacer and leaver in the given category compares to the
dierence between male replacers and the male leaver they replace in the reference category. All models in
the analysis of replacements include the aforementioned controls for individual characteristics at both the
level of the leaver and the replacer.
While replacers are compared directly to the leaver that they replace, which by denition is within
network group and detailed occupation, comparisons across categories of replacement are not necessarily
within group or occupation.
Therefore, in a nal model for each measure of referrals received I include
xed-eects for network group and occupation, using bureau of labor statistics SOC codes, to account for
the eect of potential variation across chapters and occupations in these comparisons.
I estimate two sets of ordinary least squares (OLS) regressions predicting the dierence between what
the replacer received relative to the leaver he replaced: rst overalltotal referrals received and then by
referral typedirect referrals received and third-party referrals received.
Results: Women get lower returns to social capital
Given that the focus of this study is to uncover the network mechanisms that contribute to gender dierences
in the resources male and female entrepreneurs receive, Table 3 compares male and female members in the
study population in terms of each of the relevant variables.
20
The motivation for this comparison is to
determine whether male and female entrepreneurs studied dier in substantive ways that could contribute to
dierences in the number of referrals they receive. Table 3 reveals that, on average, male and female members
do not dier in terms of the number of referrals they generate for other members, the amount of money
they generate for other members through referrals, or their tenure with the group. Women engage in more
additional meetings and have fewer absences from the weekly meetings relative to men. Despite evidence
that women contribute at least comparably to men, they receive fewer referrals.
This gender dierences
in referrals seems to be largely driven by third-party referrals, with women receiving over 26 percent fewer
third-party referrals than men. While this comparison does not include any controls, it indicates that female
entrepreneurs receive fewer resources through their network ties than do male entrepreneurs on average,
despite being more involved.
-INSERT TABLE 3 and TABLE 4 ABOUT HERE-
Table 4 presents results from negative binomial regression models estimating the eect of member gender
on the total number of referrals received to determine whether women receive fewer total referrals than do
men.
Model 1 contains only control variables, revealing that on average, members receive more referrals
when they contribute more to the group. Specically, the number of additional meetings a member engages
in, the number of referrals a member gives to others, and the amount of money a member generates for
others, are each strong predictors of the benets that they receive. In line with the comparison of means,
Model 2 reveals that women received approximately 25 percent fewer referrals per year than men in the
same network group, conditional on the covariates. Model 3 introduces occupation xed eects, and removes
network group xed eects, to compare men and women across network groups who are in the same broad
occupations.
Accounting for potential gender dierences in occupations, women still received nearly 15
percent fewer referrals than men.
The reduction in the magnitude of the gender dierence suggests that
occupational sorting is a contributing factor to observed dierences. Thus, despite having access to the same
social ties, women receive fewer referrals than men who are contributing similarly to the group.
21
Replacement events
Total referrals.
To provide a precise test of hypothesis 1, this next set of analyses focuses on the subsample
of members involved in a replacement event to compare men and women in the same detailed occupations
within the same networking group. In Table 5, the constant term provides the estimated dierence in the
number of referrals that the replacer received, on average, relative to what the leaver received for the omitted
category, male-to-male replacements, conditional on the covariates. In other words, the constant tells us how
many more or fewer referrals a male replacing a male receives relative to the man he replaced. The coecients
on male-to-female, female-to-male, and female-to-female dummy variables provide estimates of whether the
dierence in what the replacer received relative to the leaver he or she replaced in each respective replacement
category is greater or less than observed dierences in the male-to-male replacement category. As Model 2
in Table 5 shows, the gap between what a replacer receives relative to the leaver being replaced is far wider
in the male-to-female category than in the male-to-male category.
Whereas a male in the male-to-male
category receives approximately the same number of total referrals as the leaver he replaced, as evidenced
by the non-signicant constant term, a female replacer in a male-to-female replacement event receives ve,
or 23 percent, fewer referrals than the man whom she replaced. The introduction of occupation xed eects
in Models 3 and 4 does not lessen this observed gender dierence. The penalty for female replacers in the
male-to-female category is even greater when more ne-grained occupation categories are included (Model
4), with women generating nearly seven fewer referrals than the men they replace.
-INSERT TABLE 5 ABOUT HERE-
The coecients of the other two replacement categoriesfemale-to-female and female-to-maleare not
signicant indicating that replacers in these categories, like the male replacers in the male-to-male category,
receive a similar number of total referrals as the person whom they replaced. I also conducted a post-estimation
F test to determine whether the eect for the male-to-female category diers signicantly from the female-to-female
and female-to-male categories.
The results of this test reveal that these dierences are also statistically
signicant. In other words, with the exception of female replacers in the male-to-female category, there is
no penalty or advantage for replacers, on average, for entering a network group relative to the person they
22
replaced. Taken together, these results provide support for hypothesis 1 revealing that resource exchange
directly contributes to gender inequality in outcomes.
Type of referral.
The previous section showed that female entrepreneurs receive fewer total referrals
than their male counterparts with access to the same social ties and in the same detailed occupation. This
section presents additional analyses aimed at uncovering the mechanism driving this gender dierence and
testing hypothesis 2. I nd that expectations that a client, family member, or friend, prefers to work with
male entrepreneurs, or what I call anticipatory third-party bias, lead members in these network groups to
disproportionately give referrals to male group members under certain conditions.
To disentangle the anticipatory third-party bias mechanism proposed in this study, Table 6 presents
estimates from models predicting the dierence in number of referrals replacers receive relative to the referrals
received by the leaver they replaced by referral typedirect use referrals (Model 1) and third-party referrals
(Model 2).
These models reveal that the gender dierence observed is solely driven by women receiving
fewer third-party referrals than their male counterparts. Women receive as many direct use referrals as the
men whom they replace, as indicated by the non-signicant coecient of male-to-female in model 1, but
over ve fewer third-party referrals as shown by the coecient of male-to-female in model 2.
Since male
leavers replaced by women receive approximately 17 third-party referrals (presented below in Table 8), this
dierence translates to women receiving 30 percent fewer third-party referrals than the men they replace.
-INSERT TABLE 6 ABOUT HERE-
Whereas third-party referrals involve one member giving the information of his outside contact to another
member, direct use referrals only involve the two members in the group. Thus, unlike direct use referrals, the
success of third-party referrals requires that the outside, or third-party, contact be satised with the referrer's
decision to make a referral. Members commonly mentioned taking the preferences of their clients, family
members, and friends into account. When discussing the criteria used to decide whether to make third-party
referrals to fellow entrepreneur-members in interviews, members commonly stressed the importance of a
good match between their fellow contacts in their network group and their third-party contacts outside of
the group. For example, one entrepreneur-member shared: You can be a pretty good judge of who is going
to hit it o with who and certain things like that. It really comes down to a t.
23
Together, these results provide support for hypothesis 2 suggesting that expectations about the gender
preferences of third-party contacts not in the group serve to restrict the number of referrals women receive
relative to men.
Gender composition of occupation.
To provide additional support that anticipatory third-party
biases are leading to fewer referrals for women, this section tests hypotheses 3a, 3b, and 4, by exploring
variation in observed gender dierences in third-party referrals based on the gender composition of the
occupation an entrepreneur-member occupies.
To the extent that members are anticipating the gender preferences of a third-party when deciding
whether or not to make referrals, it is reasonable for members to expect that their contacts are particularly
likely to have a preference for men in roles that are traditionally performed by men, as posited in hypothesis
3a.
Similarly, to the extent that third-party contacts have a preference for women, this is most likely in
female-dominated occupations, as posited in hypothesis 3b. Figures 3A and 3B graph the dierence in the
predicted number of third-party referrals received by replacers relative to the leaver they replaced when
the replacer's gender is congruent versus incongruent with the gender-type of the occupation.
As in the
models presented in Table 6, the underlying OLS models used to generate Figures 3A and 3B predict the
number of third-party referrals received based on the type of replacement and introduce dummy variables for
the sex composition of the occupationMale Occupation for Figure 3A, to test hypothesis 3a, and Female
Occupation for Figure 3B, to test hypothesis 3b. These models include both the main eect of Male or Female
Occupation and interactions between replacement type and these dummy variables. These interaction terms
estimate how dierences in the referrals received by replacers versus leavers varies based on the degree to
which the occupation that the men and women being compared occupy is gendered. The estimates from
these models were used to calculate the predicted dierence in the number of referrals that a replacer received
relative to the person they replaced based on the gender of both actors and the the gender composition of
the occupation.
Figure 3A introduces Male Occupation, which takes the value of one for occupations where 70 percent
or more of occupants are male, and compares the relative referrals received by replacers versus leavers in
the male-to-female category to the dierence for the relevant baseline group, male-to-male. Consistent with
24
hypothesis 3a, Figure 3A reveals a clear pattern: the penalty in third-party referrals received by a female
replacer relative to male leaver she replaced is most pronounced in occupations that are predominantly
male. Whereas women in occupations that are more than 70 percent male receive 4.6, or 25 percent, fewer
referrals than the men they replaced, there is no gender dierence in referrals received by men and women
in occupations that are less than 70 percent male.
Figure 3B introduces Female Occupation, which takes the value of one for occupations where 70 percent
or more of occupants are female, and compares the relative referrals received by replacers versus leavers in
the female-to-male category to the dierence for the relevant baseline group, female-to-female. This gure
depicts that there is no signicant dierence between these two categories of replacements suggesting that
men are not penalized for entering female-typed occupations in this setting. As a result, I do not nd support
for hypothesis 3b. Rather, when the man replaces a female leaver he is not penalized in terms of third-party
referrals, even when occupying a role traditionally held by women.
-INSERT FIGURES 3A and 3B ABOUT HERE-
Together these results provide support for hypothesis 4, that the degree to which a network contact
occupies a role that is gender incongruent has greater implications for resource-holder's decisions of whether
to share resources with women than with men. Even stronger, these results suggest that this penalty only
impacts women. Among members involved in a replacement, the only notable gender dierence is when a
woman replaces a man in a male-typed occupation, where third-party contacts are likely to have the strongest
gender preferences.
Moving from replacement events to the overall study population, Figure 4 graphs the eect of the gender
composition of the occupation a member occupies on the predicted number of third-party referrals received
by men versus women.
The underlying model for Figure 4 predicts the number of third-party referrals
received based on gender using a negative binomial regression.
As in the models presented in Table 4,
the underlying model includes the same set of controls and introduces Percent Male Occupation as an
independent variable, which is an ordinal variable ranging from 1-7 and measuring the degree to which
actors within each occupation tend to be men. This model includes both the main eect of Percent Male
Occupation and interactions between Female and Percent Male Occupation. This interaction term estimates
25
how gender dierences in third-party referrals received varies based on the degree to which the occupation
that the men and women being compared occupy. To generate Figure 4, the estimates from this model were
used to calculate the predicted number of referrals that a man and woman will receive based on the gender
composition of their occupation.
-INSERT FIGURES 4 and 5 ABOUT HERE-
Consistent with hypothesis 4, Figure 4 reveals that women receive fewer third-party referrals than their
male counterparts in male-dominated occupations. Whereas women in occupations that are more than 70
percent male receive 63.6 percent fewer referrals than their male counterparts, there is no gender dierence
in referrals received by men and women in occupations that are less than 50 percent male. To conrm that
this pattern is due to anticipated third-party biases, Figure 5 depicts the eect of the gender composition
of the occupation a member occupies on the predicted number of direct-use referrals received by men and
women.
This gure reveals that there are no signicant dierences in the number of direct use referrals
men and women receive. Thus, women only receive fewer third-party referrals than do men and this gender
dierence is limited to comparisons of men and women in male-typed occupations. This nding provides
additional support for my proposed mechanism, namely anticipatory third-party bias. Gender dierences in
the referrals men and women receive result from expectations about the gender preferences of others.
Taken together, the results presented in this paper reveal that despite having access to the same social
ties as male entrepreneur-members, women receive fewer referrals. Further, this study provides support that
this gender dierences is driven by anticipated third-party biases.
It is only when entrepreneur-members
anticipate that their clients, friends and family members are likely to prefer to work with men over women
that they disproportionately give referrals to male members within their network group.
Robustness checks and ruling out alternatives
Screening eect.
One possible alternative mechanism for the nding that women receive fewer third-party,
but not direct use, referrals relates to potential unobserved gender dierences in on-the-job performance
or competency. The argument would be that members hire female members (i.e. direct use referrals) and
discover that these women are poor performers. As a result, they do not connect their third-party contacts to
26
the female members in their group. While this is a plausible alternative explanation, neither my quantitative
nor qualitative data provide evidence that supports this alternative explanation.
To the extent that actors in these groups are not referring to women because they hire them and realize
that they are poor performers, we would expect that as a woman receives more direct use referrals she would
receive fewer third-party referrals. The logic here is that, if women are lower quality performers, the more
direct use referrals they receive the more likely members in the group will learn of their lower quality through
informal mechanisms. Based on this logic, as other members learned about a woman's lower quality, either
by hiring the female member themselves of through word-of-mouth from other members who had hired the
female member, they would be less likely to connect their outside contacts to the female member in the
group.
-INSERT TABLE 7 ABOUT HERE-
Table 7 presents the results from a model estimating the number of third-party referrals a replacer received
as a function of the number of direct use referrals he or she received. If this alternative screening mechanism
is driving observed gender dierences, we would expect that the number of direct use referrals a woman
received would, at least partly, account for observed gender dierences in third-party referrals received. In
other words, by introducing the number of direct use referrals a woman receives as a predictor in this model,
we would expect to see a reduction in the main eect of the male-to-female coecient. This is not the case,
however, as the results are identical to those presented in Table 6. On average, women receive approximately
six, or 35 percent, fewer third-party referrals than the men that they replace, even after controlling for the
number of direct use referrals they received in that year.
In line with this quantitative evidence, my interviews do not suggest that entrepreneur-members are
using direct use referrals to screen women.
Generally, competency does not seem to be a major concern
among members. When asked about experiences with, or knowledge of, poorly performing others, members
typically stated that those who were not strong in their area of specialty typically did not last long in the
group. Furthermore, they had a dicult time recalling cases where a member was a poor performer. It was
often the case that multiple members within a networking group oered the same story of a poor performer,
suggesting that this was in fact an uncommon occurrence. As a result, members did not center their decisions
27
of whether to refer on competence. As one entrepreneur-member stated: Competency is important, but most
of the people that I've seen are competent in their eld. I haven't seen anyone in our [group] that I don't feel
can do the job.
Even when asked about the criteria used for assessing others in the group, members often did not mention
assessing quality or on-the-job performance. Rather, members discussed the central importance of learning
about other members in the group as people. As one member shared: Unless you get to know the people in
the group, it's very hard to give that referral. You have to learn who they are. Another member stresses
the importance of the characteristics of the person over their performance stating: I'm only going to make
referrals if I know the person I'm referring to.
I know the quality of who they are as people because my
prejudice is that if they are a good quality of person the quality of their work will be ne.
Taken together, both my quantitative and qualitative evidence cast doubt on the alternative that direct
use referrals serve as a screening mechanism excluding women from the consideration set for third-party
referrals. Thus, it is unlikely that members are directly hiring female members, realizing that they are poor
performers, and consequently not referring their outside contacts to the women in the group.
Dierences in the quality of leavers.
A second possible alternative mechanism for the nding that
women receive fewer third-party referrals than the men they replace relates to potential heterogeneity in
the leavers by replacement category.
Specically, if women are replacing top performing men, but men
are replacing lower performing men, it would be possible that women receive fewer referrals than the men
they replace because they are replacing the stars in the group. Similarly, if women are replacing very poor
performers they may receive less not because they are women, but rather because other members in the
group do not value the particular occupation they represent. In either case, my results could be biased, such
that a female penalty would be exaggerated.
While this is unlikely, it is also falsiable. To examine whether this could be the case, Table 8 depicts
the mean number of third-party referrals received by leavers in each of the four replacement categories. As
this table shows, male leavers receive on average approximately 16 to 17 third-party referrals and female
leavers receive approximately 13. Importantly, the mean number of referrals received by a leaver does not
dier signicantly based on whether the male leaver is replaced by a same or opposite gender member. This
28
captures potential dierences due to the quality of the leaver as well as potential dierences in the opportunity
for generating referrals for a given occupation. Thus, it is unlikely that dierences in the benets received
by the leavers account for the nding that women receive less than the men that they replace.
-INSERT TABLE 8 ABOUT HERE-
Women do not ask.
Another plausible explanation for the observed dierence in the number of referrals
received by women versus men is that women are not asking or expressing a need for referrals. Research has
often argued that women are less likely to make requests than their male counterparts (Babcock, 2003). To
the extent that women do not make their needs known, they could receive fewer referrals not because other
members in the group are favoring men, but rather because men ask, it is easier to identify opportunities to
refer to them.
In this empirical context, however, this is an unlikely explanation for two key reasons. First, by simply
joining these instrumental business networking groups, members are all seeking to grow their businesses by
gaining connections to new potential clients through business referrals. Second, by the design and structure
of the weekly meetings, all members in these groups publicly state their needs during the meeting.
As I
observed, all members leveraged the opportunity to discuss their business and express the types of referrals
they were seeking at that time. Furthermore, members can sign up for a longer presentation to pitch their
work and educate the other members on how to make referrals to them during the weekly meeting.
A
comparison of the propensity for men and women to sign up for these slots reveals that both genders are
equally likely to do so. Therefore, in this setting I do not nd evidence of gender dierences in requesting
referrals from fellow network contacts.
Discussion
Uncovering the sources of gender dierences in the benets men and women receive through social ties has
been the focus of an extensive body of research. Lower returns to social capital for women are commonly
attributed to gender dierences in the composition and structure of social networks (McGuire, 2000).
Undoubtedly, lack of access to valuable social ties contributes to women's inability to generate benets
29
from social ties. Existing research is less clear, however, about whether men and women with access to the
same social ties receive equal returns. This study addresses this gap in our understanding of how resource
exchange, or the activation of social ties, contributes to gender inequality by comparing the relative benets
received by male and female entrepreneurs with the same access to social ties.
Using a unique empirical setting where male and female entrepreneurs are members of the same business
networking groups, I nd that women receive fewer referrals through social ties than do similar men in the
same network group, but this female penalty is not ubiquitous. The results presented in this study indicate
that this female disadvantage is limited to cases where a resource-holder deciding whether to share resources
with a network contact expects that a relevant third-partysuch as a client, friend, or family memberhas
a preference for men over women, which I call anticipatory third-party bias. Whereas men and women are
equally likely to receive business directly from their network contacts, women receive far fewer connections
to the clients, family, and friends of their network contacts, or third-party referrals. Further, I show that
this gender dierence in referrals only exists among women in occupations traditionally occupied by men.
This nding provides further support for the central role third-parties play in the relative benets men and
women receive through their social ties.
This paper makes two main contributions to the study of the network sources of gender inequality. First,
by showing that women receive fewer benets through social ties than men despite having the same access,
this study provides evidence that access to social ties is insucient for eradicating gender dierences in
the benets of networks. Resource allocation also plays a role in perpetuating gender inequality. Network
scholars have long argued that specifying individual-level dierences in how actors use social ties provides
an opportunity for making theoretical progress (e.g., Ibarra et al., 2005; Renzulli and Aldrich 2005). This
study reinforces the importance of focusing on how resources are exchanged among network contacts. To
gain a deeper understanding of how networks aect gender inequality in particular, future research must
carefully account for both the structure and composition of networks and the process of resource exchange,
or activation, within networks.
Second, this paper identies a new network-based mechanism explaining gender inequality, which I call
anticipatory third-party bias.
The nding that women only receive lower returns to their social capital
30
when resource-holders anticipate the gender preferences of others, or third-parties, provides support for this
mechanism. Expectations about the gender preferences of others, or anticipatory third-party bias, inuence
resource-holders' decisions of whether to share resources with network contacts in a way that limits the
benets women receive through social ties. The fact that resource-holders only exhibit a male preference
in occupations that are traditionally held by men provides further support for this mechanism.
To the
extent that a resource-holder is anticipating the preferences of a third-party, it is reasonable to expect that
the third-party prefers men in occupations that are male-typed.
In other words, the assumption that a
third-party contact prefers a male electrician is more likely to be accurate than the assumption that a
third-party contact prefers a male nurse. Future research could further uncover the conditions that make
third-party biases more or less prevalent in evaluations. Resource-holders may dier in their propensity to
incorporate gender into their decisions of whether to share resources based on their own gender or on the
gender of their outside contact, or third-party. Understanding whether anticipated third-party biases operate
dierently for certain actors or under certain conditions would provide a more nuanced understanding of
how and when resource exchange within networks disadvantages women relative to men.
This study also makes a contribution to research on how ascriptive characteristics aect evaluations more
generally by providing empirical evidence for the role of third-parties in perpetuating gender inequality. A
challenge in identifying how anticipatory third-party beliefs contribute to gender inequality is disentangling
this mechanism from other competing explanations. Specically, existing evidence consistent with the notion
that anticipating the preferences of others leads actors to disproportionately favor men over women has been
unable to account for personal gender beliefs, or preferences (e.g. Fernandez-Mateo and King, 2011). To
develop comprehensive theories about sources of gender inequality, and to redress these inequities, it is
necessary to more carefully identify not only the potential mechanisms, but also the extent to which these
mechanisms are responsible for observed gender dierences in outcomes. This study oers a rst attempt
to disentangle these competing mechanisms by comparing the relative returns men and women receive
from social ties in cases where decisions to share resources are contingent on a third-party versus cases
where decisions are based on personal preferences. Future work could further disentangle these processes
by comparing decisions to share resources under conditions with and without a third-party made by the
31
same evaluator. This will allow for further identication of how anticipating the preferences of a third-party
impacts gender inequality in the benets received through social ties by accounting for each individual's
personal gender preferences.
In terms of identifying when third-parties are more apt to perpetuate versus reduce inequality, this
study provides evidence consistent with the importance of visibility, or transparency. Acting in accordance
with the anticipated preferences of others is particularly likely when the presence of bias in decisions is
not observable, or visible, to others.
Of course, third party audience expectations can serve to mitigate
reliance on ascriptive characteristics in evaluations (Salancik and Pfeer, 1978; Tetlock, 1992; Valian, 1998).
By imposing expectations, audiences serve to create accountability, or impose sanctions for behavior that
is in violation of established norms (Adut, 2005; Jensen, 2006).
Therefore, concerns that the third-party
will sanction biased behavior may discourage reliance on gender in decisions of whether or not to share
resources.
But an audience is only able to impose its expectations, tacitly or otherwise, onto a decision
maker if this bias is visible to the evaluating audience (see Castilla (2008) for a similar discussion about the
importance of transparency). This study provides support for the presence of bias under conditions where a
third-party is unable to perceive whether a decision-maker is exercising bias. Since the third-party contacts
of entrepreneur-members are unlikely to know that a member refrained from connecting them to another
member of the group, bias is not visible to third-parties in this setting.
In conclusion, my ndings help to explain how networks contribute to the underrepresentation and lower
success of women among entrepreneurs.
Even if female entrepreneurs have access to the same networks,
and engage similarly within these networks, as their male counterparts, I nd that they may still receive
fewer benets. Thus, to the extent that entrepreneurs strategically shape their networks (Burt, 2007; Stuart
and Sorenson, 2011), simply gaining access to valuable social ties may not lead to the same benets for
men and women.
Further, even if network contacts are not exercising overt bias, or male preference, in
deciding with whom to share resources, expectations about collective gender beliefs of others may serve to
constrain actors in a way that perpetuates gender inequality. The gender preferences of third-parties are
likely to be particularly salient in assessments of entrepreneurs because entrepreneurship is a male-dominated
area (Gupta et al., 2009).
Therefore, redressing gender inequality in the benets that male and female
32
entrepreneurs receive through social ties requires more than providing women with access to better networks
or eliminating personal gender stereotypes or beliefs. Equipping entrepreneurs with knowledge about the
role of third-parties for gender inequality provides a rst step toward diminishing gender dierences in the
benets available through social ties. Since actors are not bound to anticipating the gender preferences of
third-parties, more equitable network-based outcomes are possible if actors avoid basing decision on these
expected gender preferences, or at least learn the true preferences or expectations of others where possible.
33
References
Acker, Joan. 1990. “Hierarchies, Jobs, Bodies: A Theory of Gendered Organizations.” Gender and Society 4
(2): 139–58.
Adut, Ari. 2005. “A Theory of Scandal: Victorians, Homosexuality, and the Fall of Oscar Wilde.” American
Journal of Sociology 111 (1): 213–48.
Aldrich, Howard, Amanda Elam, and Patty Reese. 1997. “Strong Ties, Weak Ties, and Strangers: Do
Women Business Owners Differ from Men in Their Use of Networking to Obtain Assistance?” In
Entrepreneurship in a Global Context, edited by Sue Birley and Ian Macmillan, 1–25. London ; New York:
Routledge.
Aldrich, Howard E., Ben Rosen, and Bill Woodward. 1987. “The Impact of Social Networks on Business
Foundings and Profit: A Longitudinal Study.” In Frontiers of Entrepreneurship Research 1987, edited by
Neil C. Churchill, 154–68. Wellesley, Mass.: Babson College Center for Entrepreneurship.
Aldrich, Howard, Pat Ray Reese, and Paola Dubini. 1989. “Women on the Verge of a
Breakthrough:networking among Entrepreneurs in the United States and Italy.” Entrepreneurship &
Regional Development 1 (4): 339–56.
Babcock, Linda, and Sara Laschever. 2003. Women Don’t Ask: Negotiation and the Gender Divide.
Princeton, N.J: Princeton University Press.
Beckman, Christine M., and Damon J. Phillips. 2005. “Interorganizational Determinants of Promotion:
Client Leadership and the Attainment of Women Attorneys.” American Sociological Review 70 (4): 678–
701.
Berger, Joseph, H. Famit Fisek, Robert Z. Norman, and Morris Zelditch. 1977. Status Characteristics and
Social Interaction: An Expectation-States Approach. Elsevier Scientific Pub. Co.
Berger, Joseph, Morris Zelditch, Bo Anderson, and Martha Foschi, eds. 1989. “Status Characteristics,
Standards, and Attribution.” In Sociological Theories in Progress: New Formulations, 58–72. Sage
Publications.
Bernard, H. R., P. D. Killworth, and L. Sailer 1981 ‘‘Summary of research on informant accuracy in
network data and the reverse small world problem.’’ Connections, 4: 11–25.
Bird, Sharon R., and Stephen G. Sapp. 2004. “Understanding the Gender Gap in Small Business Success:
Urban and Rural Comparisons.” Gender and Society 18 (1): 5–28.
Booth, Alison, and Andrew Leigh. 2010. “Do Employers Discriminate by Gender? A Field Experiment in
Female-Dominated Occupations.” Economics Letters 107 (2): 236–38.
Botelho, T. and Abraham, M. Lack of Information or Bias? Unpacking the role of gender in evaluation.
Working Paper.
Braddock, Jomills Henry, and James M. McPartland. 1987. “How Minorities Continue to Be Excluded from
Equal Employment Opportunities: Research on Labor Market and Institutional Barriers.” Journal of Social
Issues 43 (1): 5–39.
Brass, Daniel J. 1985. “Men’s and Women’s Networks: A Study of Interaction Patterns and Influence in an
Organization.” The Academy of Management Journal 28 (2): 327–43.
Burstein, Paul, and Paula England. 1994. “Neoclassical Economists’ Theories of Discrimination.” In Equal
Employment Opportunity: Labor Market Discrimination and Public Policy, 59–69. Aldine de Gruyter.
Burt, Ronald. 1995. Structural Holes: The Social Structure of Competition. Cambridge, Mass.: Harvard
University Press.
———. 2007. “Secondhand Brokerage: Evidence on the Importance of Local Structure for Managers,
Bankers, and Analysts.” Academy of Management Journal 50 (1): 119–48.
Buttner, E. Holly, and Benson Rosen. 1988. “Bank Loan Officers’ Perceptions of the Characteristics of Men,
Women, and Successful Entrepreneurs.” Journal of Business Venturing 3 (3): 249–58.
Campbell, Karen E. 1988. “Gender Differences in Job-Related Networks.” Work and Occupations 15 (2):
179–200.
Castilla, Emilio J. 2008. “Gender, Race, and Meritocracy in Organizational Careers.” American Journal of
Sociology 113 (6): 1479–1526.
Castilla, Emilio J., and Stephen Benard. 2010. “The Paradox of Meritocracy in Organizations.”
Administrative Science Quarterly 55 (4): 543–76.
Cejka, Mary Ann, and Alice H. Eagly. 1999. “Gender-Stereotypic Images of Occupations Correspond to the
Sex Segregation of Employment.” Personality and Social Psychology Bulletin 25 (4): 413–23.
Clark, C. Robert, Samuel Clark, and Mattias K. Polborn. 2006. “Coordination and Status Influence.”
Rationality and Society 18 (3): 367–91.
Coleman, James S. 1988. “Social Capital in the Creation of Human Capital.” American Journal of Sociology
94 (S1): 95.
Correll, Shelley, and Stephen Benard. “Biased Estimators? Comparing Status and Statistical Theories of
Discrimination.” In Social Psychology of the Workplace, 23:89–116. Emerald Group Publishing.
Correll, Shelley J., Stephen Benard, and In Paik. 2007. “Getting a Job: Is There a Motherhood Penalty?”
The American Journal of Sociology 112 (5): 1297–1338.
Correll, Shelley, and Cecilia Ridgeway. 2003. “Expectation States Theory.” In Handbook of Social
Psychology, 29–51. Delamater, John. New York: Kluwer Academic/Plenum.
Davison, Heather K., and Michael J. Burke. 2000. “Sex Discrimination in Simulated Employment Contexts:
A Meta-Analytic Investigation.” Journal of Vocational Behavior 56 (2): 225–48.
Drentea, Patricia. 1998. “Consequences of Women’s Formal and Informal Job Search Methods for
Employment in Female-Dominated Jobs.” Gender and Society 12 (3): 321–38.
Eagly, Alice H. 2004. “Prejudice: Toward a More Inclusive Understanding.” In The Social Psychology of
Group Identity and Social Conflict: Theory, Application, and Practice, edited by A. H. Eagly, R. M.
Baron, and V. L. Hamilton, 45–64. APA Decade of Behavior Volumes. Washington, DC, US: American
Psychological Association.
Eagly, Alice H., and Steven J. Karau. 2002. “Role Congruity Theory of Prejudice toward Female Leaders.”
Psychological Review 109 (3): 573–98.
Elvira, Marta M., and Mary E. Graham. 2002. “Not Just a Formality: Pay System Formalization and SexRelated Earnings Effects.” Organization Science 13 (6): 601–17.
Emerson, Robert M. 1983. “Holistic Effects in Social Control Decision-Making.” Law & Society Review 17
(3): 425–55.
Emirbayer, Mustafa, and Jeff Goodwin. 1994. “Network Analysis, Culture, and the Problem of Agency.”
American Journal of Sociology 99 (6): 1411–54.
Fernandez-Mateo, Isabel, and Zella King. 2011. “Anticipatory Sorting and Gender Segregation in
Temporary Employment.” Management Science 57 (6): 989–1008.
Fernandez, Roberto M., and M. Lourdes Sosa. 2005. “Gendering the Job: Networks and Recruitment at a
Call Center.” American Journal of Sociology 111 (3): 859–904.
Foschi, M. 1989 “Status Characteristics, Standards, and Attributions.” In Sociological Theories in Progress:
New Formulations, edited by Joseph Berger, Morris Zelditch, and Bo Anderson, pp. 58–72. Sage.
Gorman, Elizabeth H. 2005. “Gender Stereotypes, Same-Gender Preferences, and Organizational Variation
in the Hiring of Women: Evidence from Law Firms.” American Sociological Review 70 (4): 702–28.
Granovetter, Mark. 1995. Getting a Job: A Study of Contacts and Careers. University of Chicago Press.
Gupta, Vishal K., and Nachiket M. Bhawe. 2007. “The Influence of Proactive Personality and Stereotype
Threat on Women’s Entrepreneurial Intentions.” Journal of Leadership & Organizational Studies (Baker
College) 13 (4): 73–85.
Gupta, Vishal K., Daniel B. Turban, S. Arzu Wasti, and Arijit Sikdar. 2009. “The Role of Gender
Stereotypes in Perceptions of Entrepreneurs and Intentions to Become an Entrepreneur.” Entrepreneurship
Theory and Practice 33 (2): 397–417.
Hagan, Oliver, Carol Rivchun, Donald L. Sexton, and Howard Aldrich, eds. 1989. “Networking Among
Women Entrepreneurs.” In Women-Owned Businesses, 103–32. Praeger.
Hallen, Benjamin L. 2008. “The Causes and Consequences of the Initial Network Positions of New
Organizations: From Whom Do Entrepreneurs Receive Investments?” Administrative Science Quarterly 53
(4): 685–718.
Hanson, S., and G. Pratt. 1991. “Job Search and the Occupational Segregation of Women.” Annals of the
Association of American Geographers 81 (2): 229.
Heilman, Madeline E. 1983. “Sex Bias in Work Settings: The Lack of Fit Model.” Research in
Organizational Behavior 5: 269–98.
Hundley, Greg. 2001. “Why Women Earn Less Than Men in Self-Employment.” Journal of Labor Research
22 (4): 817–29.
Hurlbert, Jeanne S., Valerie A. Haines, and John J. Beggs. 2000. “Core Networks and Tie Activation: What
Kinds of Routine Networks Allocate Resources in Nonroutine Situations?” American Sociological Review 65
(4): 598–618.
Ibarra, Herminia. 1997. “Paving an Alternative Route: Gender Differences in Managerial Networks.” Social
Psychology Quarterly 60 (1): 91–102.
Ibarra, Herminia, Martin Kilduff, and Wenpin Tsai. 2005. “Zooming in and out: Connecting Individuals and
Collectivities at the Frontiers of Organizational Network Research.” Organization Science 16 (4): 359–71.
Jensen, Michael. 2006. “Should We Stay or Should We Go? Accountability, Status Anxiety, and Client
Defections.” Administrative Science Quarterly 51 (1): 97–128.
Kanter, Rosabeth. 1977. Men and Women of the Corporation. Basic Books.
Katz, Jerome, and Pamela M. Williams. 1997. “Gender, Self-Employment and Weak-Tie Networking
through Formal Organizations.” Entrepreneurship & Regional Development 9 (3): 183–98.
Kleinbaum, Adam M., Toby E. Stuart, and Michael L. Tushman. 2013. “Discretion Within Constraint:
Homophily and Structure in a Formal Organization.” Organization Science 24 (5): 1316–36.
Kmec, Julie A. 2008. “The Process of Sex Segregation in a Gender-Typed Field: The Case of Male Nurses.”
Sociological Perspectives 51 (2): 259–79.
Lerner, Jennifer S., and Philip E. Tetlock. 1999. “Accounting for the Effects of Accountability.”
Psychological Bulletin 125 (2): 255–75.
Lincoln, James R., and Jon Miller. 1979. “Work and Friendship Ties in Organizations: A Comparative
Analysis of Relation Networks.” Administrative Science Quarterly 24 (2): 181–99.
Lin, Nan. 2002. Social Capital: A Theory of Social Structure and Action. Cambridge; New York: Cambridge
University Press.
Marin, Alexandra. 2012. “Don’t Mention It: Why People Don’t Share Job Information, When They Do, and
Why It Matters.” Social Networks 34 (2): 181–92.
Marsden, Peter V. 1987. “Core Discussion Networks of Americans.” American Sociological Review 52 (1):
122–31.
McDonald, Steve. 2011. “What’s in the ‘old Boys’ Network? Accessing Social Capital in Gendered and
Racialized Networks.” Social Networks 33 (4): 317–30.
McGuire, Gail M. 2000. “Gender, Race, Ethnicity, and Networks The Factors Affecting the Status of
Employees’ Network Members.” Work and Occupations 27 (4): 501–24.
———. 2002. “Gender, Race, and the Shadow Structure: A Study of Informal Networks and Inequality in a
Work Organization.” Gender and Society 16 (3): 303–22.
Mencken, F. Carson, and Idee Winfield. 2000. “Job Search and Sex Segregation: Does Sex of Social Contact
Matter?” Sex Roles 42 (9/10): 847–64.
Merrett, Christopher D., and John J. Gruidl. 2000. “Small Business Ownership in Illinois: The Effect of
Gender and Location on Entrepreneurial Success.” Professional Geographer 52 (3): 425.
Moore, Dorothy Perrin, and E. Holly Buttner. 1997. Women Entrepreneurs: Moving Beyond the Glass
Ceiling. First Edition. SAGE Publications, Inc.
Moore, Gwen. 1988. “Women in Elite Positions: Insiders or Outsiders?” Sociological Forum 3 (4): 566.
———. 1990. “Structural Determinants of Men’s and Women’s Personal Networks.” American Sociological
Review 55 (5): 726–35.
Neumark, David, Roy J. Bank, and Kyle D. Van Nort. 1996. “Sex Discrimination in Restaurant Hiring: An
Audit Study.” The Quarterly Journal of Economics 111 (3): 915–41.
Petersen, Trond, and Laurie A. Morgan. 1995. “Separate and Unequal: Occupation-Establishment Sex
Segregation and the Gender Wage Gap.” The American Journal of Sociology 101 (2): 329–65.
Podolny, Joel M. 1993. “A Status-Based Model of Market Competition.” The American Journal of Sociology
98 (4): 829–72.
Podolny, Joel Marc. 2005. Status Signals: A Sociological Study of Market Competition. Princeton University
Press.
Podolny, Joel M., and James N. Baron. 1997. “Resources and Relationships: Social Networks and Mobility
in the Workplace.” American Sociological Review 62 (5): 673–93.
Quintane, E., and A. M. Kleinbaum 2011 ‘‘Matter over mind? E-mail data and the measurement of social
networks.’’ Connections, 31: 22–46.
Renzulli, Linda. 1998. “Small Business Owners, Their Networks, and the Process of Resource Acquisition.”
Master’s thesis, Department of Sociology. University of North Carolina at Chapel Hill.
Renzulli, Linda A., & Howard Aldrich. “Who Can You Turn to? Tie Activation Within Core Business
Discussion Networks.” Social Forces 84, no. 1 (2005): 323–341.
Renzulli, Linda A., Howard Aldrich, and James Moody. 2000. “Family Matters: Gender, Networks, and
Entrepreneurial Outcomes.” Social Forces 79 (2): 523–46.
Riach, Peter A., and Judith Rich. 2006. “An Experimental Investigation of Sexual Discrimination in Hiring
in the English Labor Market.” Advances in Economic Analysis & Policy 5 (2).
Richardson, John G., and Pierre Bourdieu, eds. 1986. “The Forms of Social Capital.” In Handbook of
Theory and Research for the Sociology of Education, 241–58. Greenwood Publishing Group, Incorporated.
Ridgeway, Cecilia L. 2011. Framed by Gender: How Gender Inequality Persists in the Modern World.
Oxford University Press.
Ridgeway, Cecilia L., and Shelley J. Cornell. 2006. “Consensus and the Creation of Status Beliefs.” Social
Forces 85 (1): 431–53.
Ridgeway, Cecilia L., and Shelley J. Correll. 2004. “Unpacking the Gender System: A Theoretical
Perspective on Gender Beliefs and Social Relations.” Gender and Society 18 (4): 510–31.
Ridgeway, Cecilia L., and Lynn Smith-Lovin. 1999. “The Gender System and Interaction.” Annual Review
of Sociology 25: 191–216.
Robb, Alicia, and Susan Coleman. 2009. Characteristics of New Firms: A Comparison by Gender. SSRN
Scholarly Paper ID 1352601. Rochester, NY: Social Science Research Network.
Rosenfeld, Rachel A. 2002. “What Do We Learn about Difference from the Scholarship on Gender?” Social
Forces 81 (1): 1–24.
Rubineau, Brian, and Roberto M. Fernandez. (Forthcoming). “How do Labor Market Networks Work?,” in
Emerging Trends in the Social and Behavioral Sciences (eds.) Robert Scott and Stephen Kosslyn, Hoboken,
NJ: John Wiley and Sons.
Ruef, Martin, Howard E. Aldrich, and Nancy M. Carter. 2003. “The Structure of Founding Teams:
Homophily, Strong Ties, and Isolation among U.S. Entrepreneurs.” American Sociological Review 68 (2):
195–222.
Salancik, Gerald R., and Jeffrey Pfeffer. 1978. “Uncertainty, Secrecy, and the Choice of Similar Others.”
Social Psychology 41 (3): 246–55.
Shane, Scott, and Toby Stuart. 2002. “Organizational Endowments and the Performance of University
Start-Ups.” Management Science 48 (1): 154–70.
Smith, Edward Bishop, Tanya Menon, and Leigh Thompson. 2011. “Status Differences in the Cognitive
Activation of Social Networks.” Organization Science 23 (1): 67–82.
Smith, Sandra S. 2000. “MOBILIZING SOCIAL RESOURCES: Race, Ethnic, and Gender Differences in
Social Capital and Persisting Wage Inequalities.” Sociological Quarterly 41 (4): 509–37.
Smith, Sandra Susan. 2005. “‘Don’t Put My Name on It’: Social Capital Activation and Job‐Finding
Assistance among the Black Urban Poor.” American Journal of Sociology 111 (1): 1–57.
Sørensen, Jesper B., and Amanda J. Sharkey. 2014. “Entrepreneurship as a Mobility Process.” American
Sociological Review 79 (2): 328–49.
Srivastava, Sameer B. 2014. “Social Capital Activation, Uncertainty, and Organizational Restructuring.”
Working Paper.
Stewart, Alex. 1990. “The Bigman Metaphor for Entrepreneurship: A ‘Library Tale’ with Morals on
Alternatives for Further Research.” Organization Science 1 (2): 143–59.
Stuart, Toby E., and Olav Sorenson. 2007. “Strategic Networks and Entrepreneurial Ventures.” Strategic
Entrepreneurship Journal 1 (3-4): 211–27.
Swim, J. K., and L. J. Sanna. 1996. “He’s Skilled, She’s Lucky: A Meta-Analysis of Observers’ Attributions
for Women’s and Men’s Successes and Failures.” Personality and Social Psychology Bulletin 22 (5): 507–19.
Tetlock, Philip. 1992. “The Impact of Accountability on Judgment and Choice: Toward a Social
Contingency Model.” In Advances in Experimental Social Psychology, Volume 25, edited by Mark P.
Zanna, 1 edition, 331–76. San Diego; London: Emerald publishing.
Thebaud, Sarah. 2010. “Gender and Entrepreneurship as a Career Choice: Do Self-Assessments of Ability
Matter?” Social Psychology Quarterly 73 (3): 288–304.
Trost, Jan E. 1986. “Statistically Nonrepresentative Stratified Sampling: A Sampling Technique for
Qualitative Studies.” Qualitative Sociology 9 (1): 54.
Turco, Catherine J. 2010. “Cultural Foundations of Tokenism: Evidence from the Leveraged Buyout
Industry.” American Sociological Review 75 (6): 894–913.
Valian, Virginia. 1999. Why So Slow? The Advancement of Women. The MIT Press.
Wagner, David G., and Joseph Berger. 1993. “Status Characteristics Theory: The Growth of a Program.” In
Theoretical Research Programs: Studies in the Growth of Theory, edited by J. Berger, M. Zelditch, and Jr.,
23–63. Stanford University Press: Stanford University Press.
Webster, Murray, Jr., and Stuart J. Hysom. 1998. “Creating Status Characteristics.” American Sociological
Review 63 (3): 351–78.
Williams, Christine L. 1992. “The Glass Escalator: Hidden Advantages for Men in the ‘Female’ Professions.”
Social Problems 39 (3): 253–67.
Williams, Joan. 1999. Unbending Gender: Why Family and Work Conflict and What To Do About It.
Oxford University Press.
Zuckerman, Ezra W., Tai-Young Kim, Kalinda Ukanwa, and James von Rittmann. 2003. “Robust Identities
or Nonentities? Typecasting in the Feature-Film Labor Market.” The American Journal of Sociology 108
(5): 1018–74.
Zuckerman, Ezra W., and Stoyan V. Sgourev. 2006. “Peer Capitalism: Parallel Relationships in the U.S.
Economy.” American Journal of Sociology 111 (5): 1327–66.
Figure 1. Structure of groups
Group A
Group B
Figure 2A. Distribution of Entrepreneur-members to Occupation Categories based on Bureau of Labor Statistics
Standard Occupational Classification (SOC) Codes
Business and Financial Specialists
Construction and Trades
Office and Administrative Support
Retail and Services Sales
Arts, Design, Entertainment, and Media
Legal
Computer
Healthcare Practitioners
Community and Social Services
Building and Grounds Maintenance
Print and Production
Personal Care and Service
Installation and Repair
Architecture and Engineering
Food Preperation and Service
Advertising, Marketing, and Promotions
Healthcare Support
Transportation and Material Moving
Protective Services
Education and Training
0
50
100
150
200
250
300
350
Figure 2B. Gender Composition of Occupation Categories based on Bureau of Labor Statistics Standard Occupational
Classification (SOC) Codes
0.9
0.8
0.7
0.6
0.5
0.4
Male‐dominated
Female‐
0.3
0.2
0.1
0
TABLE 1. Replacement Events by Type
Male to Male
Male to Female
Female to Female
Female to Male
Total
Number of Unique
Persons
376
176
132
148
Number of
Replacements
188
88
66
74
% of Total
Replacement Events
45.19%
21.25%
15.87%
17.79%
832
416
100%
TABLE 2. Comparing Participation and Referral Behavior of Replacement Sample to Study Population, Yearly Averages
Mean
Study Population
s.d.
Min
23.58
19.68
Replacement Sample
s.d.
Min
Max
Mean
Max
0.00
196.00
23.35
20.29
0.00
196.00
9.73
15.06
0.00
0.00
73.00
188.00
7.37
15.99
11.20
15.20
0.00
0.00
73.00
151.00
20.30
8.87
16.44
30,410.57
0.00
0.00
0.00
0.00
264.00
85.00
248.00
789,598.40
22.42
7.07
15.35
9,935.09
18.97
8.23
15.89
31,500.74
0.00
0.00
0.00
0.00
264.00
73.00
248.00
661,666.00
15.62
2.89
4.89
0.00
0.15
0.00
75.00
16.83
52.00
12.75
2.17
4.42
13.36
2.12
5.31
0.00
0.15
0.00
75.00
14.41
52.00
Referrals Received
Total Referrals Received
Direct Use Referrals Received
7.69
Third-party Referrals Received
15.90
Referrals Given
Total Referrals Given
24.18
Direct Use Referrals Given
8.20
Third-party Referrals Given
15.98
Dollars Generated
13,045.00
Participation in Group
Additional Meetings
14.49
Tenure
3.03
Absences
4.04
N
2,310
832
TABLE 3. Basic Descriptive Statistics for Key Variables by Gender, First Year in Dataa
Female Members
Mean
Total Referrals Received
20.22
Direct Use Referrals Received
7.94
Third-party Referrals Received 12.28
Total Referrals Given
23.28
Dollars Generated
7,528.50
Additional Meetings
15.25
Tenure
1.41
Absences
3.84
First Year Observed
2.36
s.d.
17.99
10.74
11.80
18.77
18,235.82
15.87
1.75
4.84
1.40
Min
0.00
0.00
0.00
0.00
0.00
0.00
0.15
0.00
1.00
Male Members
Max
137.00
68.00
98.00
160.00
257,744.00
75.00
12.35
40.00
5.00
Mean
24.74
8.23
16.51
22.40
8,072.73
10.95
1.56
4.41
2.47
s.d.
22.84
10.52
17.85
20.83
19,517.00
12.91
2.04
5.55
1.46
Min
0.00
0.00
0.00
0.00
0.00
0.00
0.15
0.00
1.00
Max
196.00
73.00
188.00
205.00
389,108.00
74.00
12.83
46.00
5.00
persons
822
1488
a
Comparisons robust for each nth year in the data and by calendar year. Use first year because captures all members.
b
indicates whether differences in values for male and female entrepreneurs are statistically significant based on two-sided t-tests
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001
t-testb
***
***
***
*
Table 4. Negative Binomial Regressions Predicting Total Referrals Received per Year
Model 1
Female
Total Referrals Given
Additional Meetings
Tenure
Absences
Dollars Generated (in 000's)
Constant
Fixed Effects
Year
Group
Occupation
Wald Chi-squared
DF
Persons
Person-years
Model 2
Model 3
-0.251 ***
(0.032)
0.010 ***
(0.001)
0.004 ***
(0.001)
0.010
(0.006)
-0.006
(0.003)
0.000
(0.000)
-0.149
(0.023)
0.013
(0.001)
0.003
(0.001)
0.030
(0.004)
-0.007
(0.002)
0.001
0.000
2.656 ***
(0.110)
2.735 ***
(0.118)
2.991 ***
(0.051)
Yes
Yes
No
Yes
Yes
No
Yes
No
Yes
485.320
45
2,310
5,588
548.460
46
2,310
5,588
1479.670
30
2,310
5,588
0.009
(0.001)
0.003
(0.001)
0.011
(0.006)
-0.006
(0.003)
0.000
(0.000)
***
**
*
*
Notes: Standard errors, in parentheses, are clustered by individual member. Models including
occupation fixed effects use BLS SOC major codes. Results are robust to more fine-grained
occupational fiexed effects.
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001
***
***
***
***
**
Table 5. OLS Regressions Predicting Difference in Total Referrals Received by Replacer Relative to Leaver
Model 1
Replacement Type
Male-to-Female
Tenure
Absences
Dollars Generated (in 000's)
Controls for Exiter
Total Referrals Given
Additional Meetings
Tenure
Absences
Dollars Generated (in 000's)
Constant
Model 4
-5.257 *
(2.631)
-1.511
(3.157)
1.377
(2.724)
-6.832 *
(2.719)
-1.062
(3.329)
1.633
(2.813)
0.310 ***
(0.055)
0.161
(0.085)
3.150 *
(1.520)
-0.104
(0.206)
-0.023
(0.000)
0.307 ***
(0.055)
0.170 *
(0.085)
2.962
(1.516)
-0.117
(0.206)
-0.019
(0.050)
0.302 ***
(0.056)
0.213 *
(0.090)
2.041
(1.585)
-0.117
(0.213)
-0.028
(0.051)
0.287 ***
(0.058)
0.219 *
(0.092)
1.972
(1.620)
-0.126
(0.217)
-0.017
(0.052)
-0.264 **
(0.080)
-0.177
(0.124)
-0.944
(0.520)
0.395 *
(0.187)
0.145 **
(0.000)
-0.251 **
(0.080)
-0.202
(0.125)
-0.870
(0.521)
0.387 *
(0.188)
0.150 **
(0.053)
-0.224 **
(0.083)
-0.199
(0.130)
-0.864
(0.538)
0.297
(0.193)
0.141 **
(0.054)
-0.204 *
(0.085)
-0.191
(0.135)
-0.967
(0.548)
0.299
(0.195)
0.150 **
(0.054)
Female-to-Male
Additional Meetings
Model 3
-5.305 *
(2.462)
-3.639
(2.661)
0.437
(2.478)
Female-to-Female
Controls for Replacer
Total Referrals Given
Model 2
-6.186
(10.132)
-0.545
(10.419)
4.507
(11.455)
21.596
(14.280)
Yes
No
No
Yes
No
No
Yes
Yes
No
Yes
No
Yes
0.235
416
0.242
416
0.244
416
0.250
416
Fixed Effects
Group
Occupation - SOC Major
Occupation - SOC Minor
Adj R-squared
Observations
Notes: Standard errors in parentheses.
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001
Table 6. OLS Regressions Predicting Difference in Direct Use versus Third-party
Referrals Received by Replacer Relative to Leaver
Model 1: Difference in
Direct Use Referrals
Replacement Type
Male-to-Female
Female-to-Female
Female-to-Male
Controls for Replacer
Total Referrals Given
Additional Meetings
Tenure
Absences
Dollars Generated (in 000's)
Controls for Exiter
Total Referrals Given
Additional Meetings
Tenure
Absences
Dollars Generated (in 000's)
Constant
Fixed Effects
Group
Occupation
Adj R-squared
Observations
o
Model 2: Difference in
Third-party Referrals
-0.005
(1.251)
0.445
(1.295)
0.107
(1.501)
-5.231 *
(2.221)
-1.610
(2.665)
0.934
(2.299)
0.066 *
(0.027)
0.092
(0.043)
0.080
(0.754)
0.064
(0.101)
-0.035
(0.024)
0.237 ***
(0.048)
0.122
(0.076)
1.946
(1.338)
-0.181
(0.185)
0.007
(0.043)
-0.024
(0.039)
-0.062
(0.062)
(0.554) *
(0.256)
0.120
(0.092)
0.022
(0.026)
-0.201 *
(0.070)
-0.137
(0.109)
-0.309
(0.454)
0.178
(0.163)
0.119 **
(0.045)
3.590 *
5.447
0.865
(9.668)
Yes
Yes
Yes
Yes
0.1739
416
0.193
416
Notes: Standard errors in parentheses. Occupation fixed effects for SOC major codes used.
Results robust to more fine-grained occupation fixed effects.
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001
Table 7. OLS Regressions Predicting Difference in Third-party
Referrals Received based on Difference in Direct Use Referrals
Received by Replacer Relative to Leaver
Replacement Type
Male-to-Female
Female-to-Female
Female-to-Male
Screening Effect
Direct Referrals Received
MFxDirect
FFxDirect
FMxDirect
Controls for Replacer
Total Referrals Given
Additional Meetings
Tenure
Absences
Dollars Generated (in 000's)
Controls for Exiter
Total Referrals Given
Additional Meetings
Tenure
Absences
Dollars Generated (in 000's)
Constant
Fixed Effects
Group
Occupation
Adj R-squared
Observations
-5.772 *
(2.531)
-0.918
(3.108)
0.573
(2.932)
0.106
(0.141)
0.124
(0.258)
-0.154
(0.287)
-0.008
(0.274)
0.234 ***
(0.049)
0.129
(0.077)
1.742
(1.362)
-0.168
(0.181)
0.009
(0.044)
-0.202 **
(0.071)
-0.129
(0.110)
-0.335
(0.461)
0.193
(0.165)
0.121 **
(0.046)
-0.438
(9.791)
Yes
Yes
0.1875
416
Notes: Standard errors in parentheses. Occupation fixed effects for
SOC major codes used. Results robust to more fine-grained
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001
TABLE 8. Comparing Third-party Referrals Received by Leavers Involved in
a Replacement Event, by Replacement Type
Third-party Referrals
Mean
s.d.
Min
Max
Male-to-Male
Male-to-Female
Female-to-Male
Female-to-Female
15.93
17.19
13.74
13.41
12.96
14.63
11.04
12.81
0.00
0.00
0.00
0.00
69.00
83.00
53.00
97.00
Figure 3A. Predicted Difference in Third-party Referrals Received by Replacer Relative
to Leaver in Male to Female Replacements versus Male to Male replacements by Gender
Composition of Occupation
Difference in Number Third-party Referrals
-10
-5
0
5
10
_
0
1
Male-dominated Occupation = 1
Male-to-Male
Male-to-Female
Error bars represent 95% confidence intervals
Figure 3B. Predicted Difference in Third-party Referrals Received by Replacer Relative
to Leaver in Female to Male Replacements versus Female to Female replacements by
Gender Composition of Occupation
Difference in Number Third-party Referrals
-10
-5
0
5
10
_
0
1
Female-dominated Occupation = 1
Female-to-Female
Error bars represent 95% confidence intervals
Female-to-Male
Figure 4. Predicted Overall Gender Difference in Third-party Referrals Received by
Gender Composition of Occupation
Number Third-party Referrals Received
5
10
15
20
_
< 30%
30 - 40%
40 - 50%
50 - 60%
Percent Male in Occupation
M
60 - 70%
> 70%
F
Error bars represent 95% confidence intervals
Figure 5. Predicted Overall Gender Difference in Direct Use Referrals Received by
Gender Composition of Occupation
Number Direct Use Referrals Received
4
6
8
10
12
_
< 30%
30 - 40%
40 - 50%
50 - 60%
Percent Male in Occupation
M
Error bars represent 95% confidence intervals
F
60 - 70%
> 70%
Download