The Narrative Advantage:
Gender and the Language of Crowdfunding

Andreea Gorbatai

Laura Nelson

Haas School of Business

Kellogg School of Management

UC Berkeley

Northwestern University

August 2015
Abstract
In this study, we set out to examine the role of language in the success of online fundraising—a
new form of entrepreneurial project financing. In particular, we evaluate the influence of
linguistic content on fundraising outcomes, above and beyond type of product or service offered.
Online fundraising settings pose an interesting empirical puzzle: women are systematically more
successful than men, an outcome contrary to offline gender inequality. We propose that this
outcome is partially explained by linguistic differences between men and women in terms of
language they use, and we test this mechanism using data from the online crowdfunding platform
Indiegogo. The results support our theory, suggesting a link between micro-level linguistic
choices and macro level outcomes: the institution of crowdfunding may reduce gender
inequalities in the fundraising arena by benefitting the communication style of women.

1
 

 INTRODUCTION
Economic and social arrangements in markets and organizations have been shown to
systematically disadvantage women across a wide range of outcomes ranging from hiring,
performance evaluations, rewards, and promotion in the labor market to financial support in the
capital market. Research has identified several mechanisms through which this inequality is
perpetuated, ranging from homophily, tokenism and structural constraints to negative
stereotyping and women’s own beliefs about their skill level and worth in the labor market.
In particular, research on financing, small business, and entrepreneurship has shown that
women are at a marked disadvantage compared against men with similar skill and experience
levels. This difference has been largely attributed to choice homophily among predominantly
male funders, and to the type of businesses that women start. In the venture capital industry,
male venture capitalists acts as gatekeepers; this results in less funding and mentorship for
female entrepreneurs. In other entrepreneurial ventures, women gravitate towards small business
ventures where they are often the sole employee instead of choosing scalable business projects.
In all these contexts, the long term outcomes – financing, and the terms of the financing
deals – are the results of many difficult to quantify factors resulting from the interaction between
the funder and the entrepreneur. It is thus difficult to isolate the effects that non-verbal behavior,
paralinguistic cues, contextual factors, and interactions between the entrepreneur and the funder
have on the final decision regarding funding. Some of these factors have been studied
experimentally in the laboratory (Brownlow and Zebrowitz 1990; Carney, Cuddy and Yap 2010;
Kramer 1977). The majority of these studies rely on evaluating the effect of these factors on the
audience, controlling for language content. But the language we use is intimately connected to

2
 

 how we think, and how others evaluate what we are saying. Moreover, language is connected
with socio-demographic characteristics of the speaker or writer, such as gender, age or
occupation.
In this study we aim to examine the effect of language on the success of crowdfunding
campaigns, and the relationship between linguistic content and gender. Online, text-based
campaigns are ideal for examining the effect of language content apart from characteristics of the
delivery medium, message sender, and audio-visual information because the message is
delivered to the potential donors via an information-poor, asynchronous text interface. In turning
our attention to the language used in crowdfunding campaigns, we examine four different
dimensions of language content in campaign descriptions: positive (sentiment) language,
vividness, inclusive language, and business language. We then suggest that three of these types
of linguistic content (positive sentiment, vividness and inclusive language) are both more likely
to be rewarded in crowdfunding campaigns, and more likely to be used by women, while the use
of language related to money is more likely to be penalized in the crowdfunding context and
more likely to be used by men. We then test and confirm our theory that language mediates the
relationship between gender and fundraising outcomes using data from the online crowdfunding
site Indiegogo.
Our findings indicate that gender-specific language partially mediates the success of
women in fundraising money through crowdfunding. This study identifies an economic
institution (crowdfunding) where female-specific linguistic patterns are preferred over malespecific patterns, leading to a reversal in gender inequality with respect to funding. Additionally,
this study contributes to economic sociology research on gender by quantifying the impact of
linguistic choices on fundraising outcomes. Lastly, this research contributes to research in
3
 

 computational sociology by employing topic models to refine the product classification of
crowdfunding campaigns and quantify crowdfunding campaign text along several content
dimensions using the Linguistic Inquiry Word Count dictionary (Tausczik and Pennebaker
2010).

THEORY AND HYPOTHESES
MONEY AND LANGUAGE
Financial investments represent a particular type of economic exchange driven by expectations
of future returns. Research interested in the antecedents of successful access to financial
resources have found that bank loans, one major category of funding, more often go to larger
businesses, more established businesses, and businesses with collateral, while small businesses
that do not own the collateral necessary to secure a bank loan often seek equity investors (Berger
and Udell 1998). Research on equity investors suggests that previous ties with funders increases
the chance of being funded (Shane and Cable 2002), and that equity investment is geographically
concentrated—it is difficult to get funded far away from an industry's center (Stuart and
Sorenson 2003). In whole, small businesses, businesses that are not in high-growth sectors such
as technology, and businesses away from major city centers are at a disadvantage when seeking
funding. As a result, the majority of new small businesses rely solely on personal funds and
personal debt, at times supplemented by funds from friends and family, rather than outside funds
from banks or investors (Coleman and Robb 2009).
More recently, many people and small organizations have turned to funding their
entrepreneurial ideas through donation-based crowdfunding sites. The global crowdfunding
market reached 1.1 million campaigns worldwide in 2012, and raised approximately 2.7 billion
4
 

 dollars, of which 1.6 billion dollars in the United States (Massolution. 2013). To get funded
using traditional methods requires taking a financial risk via loans, or entering the often
unfriendly world of investors. Donation-based crowdfunding sites offer a completely different
institutional setting in which to obtain funding for new small businesses or products. Taking
appeals for funds directly to the public, over the impersonal medium of the internet, bypasses
many risks and difficulties inherent in traditional methods of fundraising.
Crowdfunding sites differ from traditional channels for business financing—such as bank
loans, venture capital, and microcredit groups (Anthony 2005), along three main dimensions: the
interaction medium, the types of investors, and the motivations for investing. These three aspects
combined suggest that entrepreneurs seeking funding through crowdfunding sites may be
evaluated differently than when seeking funding through a traditional channel, but little is known
about what leads to success in this setting.
The interaction medium in a crowdfunding setting is different from other investment
settings in that crowdfunding is computer-mediated and a-synchronous, such that potential
contributors make decisions in absence of face-to-face interaction or other visual cues. In
traditional investing settings entrepreneurs will make a pitch directly in front of potential
investors, and the investors have a chance to ask questions about the entrepreneur and her
product or business. Crowdfunding takes place online, absent of face-to-face interaction or other
visual cues, and absent of the need to be in the same location as your investor. Second, the types
of investors are different. Traditional investors are typically specialized in business and finance,
and many traditional forms of investment rely on one or a small number of individuals or
institutions providing large amounts of money. Projects pitched on crowdfunding sites,
alternatively, target the general population, not specialized individuals or institutions, and they
5
 

 seek small amounts of money from a large number of people. Many investors on crowdfunding
sites provide only $5 or $10 each, and few provide more than a few thousand dollars.
Third, the institutional difference between traditional financing and crowdfunding-based
financing alters the reasons why an individual might provide money to finance a project.
Traditional investors seeking to invest in a new product or business, for example, will typically
invest because they believe the project or business will succeed, that is, it will become profitable
and the investor will see a return on their investment. The investor does not need to like the
product or business or find it valuable—they must believe that there is a significant group of
people who will like it or find it valuable, and the business will thus bring in a profit.
On crowdfunding sites, alternatively, people invest in a product or business for reasons
other than financial returns. Individuals give money in order to support projects that they value
for social or value-related reasons—receiving symbolic gifts and messages in exchange for their
contribution to the writing of a book, recording of a music album, or launch of an educational
initiative —, and in order to receive products or services that will be available, if the proposed
project is successful. Similar to other forms of investment, contributors are not guaranteed the
success of the projects they invest in, but, in contrast to many other types of investment,
contributors are not eligible to receive their money back if the project fails to produce results.
Individuals contributing on crowdfunding sites thus must either personally like the project or
business, or they want to be part of a project that they think is valuable in some way.
The process of convincing investors to invest in a particular product, project, or business
on a crowdfunding site is thus much different than in a traditional setting. The lack of face-toface interaction alters how entrepreneurs build relationships with investors. In traditional

6
 

 investment settings visual and verbal cues are important, such as making eye contact, having an
erect posture, dressing appropriately, and having a strong voice. Instead, crowdfunding
campaigns must catch the attention of potential investors, depending much more on written
language rather than visual and verbal cues.
Prior research on language and health-related communication indicates that the way a
message is framed is consequential for decision making, holding constant the message itself
(Rothman and Salovey 1997). As crowdfunding campaigns rely largely on written words to
persuade people to fund a business campaign, and the literature shows that the way something is
written affects the messages' efficacy, we expect the language used to frame the business idea
behind a campaign to affect the ability to attract online donations.
Second, the type of language that will succeed in attracting money is different. While
specialized financial language is necessary to attract traditional investors, this type of language
can sound formulaic and dry to the general population and may be off-putting or confusing for
those donating money on crowdfunding sites. Furthermore, individuals contributing to donationbased crowdfunding sites are not doing so for financial gain, so financial language is unlikely to
convince them to donate. For these reasons we expect that proposals detailing monetary or
financial aspects of the project are less likely to succeed than campaigns that choose to focus on
other aspects of the business plan:
Hypothesis 1a. Crowdfunding proposal success decreases with the use of money-related
language.
Entrepreneurs seeking funding must get potential investors excited about their projects;
this is difficult to do in a traditional financing setting, but even more so in crowdfunding, where
7
 

 financial gain is not factored into the decision. One way to do this is to tell an exciting story
about the product or business, building excitement through narrative. Research focused on
business pitches in a traditional fund-raising environment suggests that entrepreneurs are
successful when they adequately address both the “what”—storytelling aspect of the business
idea—and the “how”—narrative aspect of it (Pollack, Rutherford and Nagy 2012: 918). The
interest in a vivid story is high among evaluators of movie pitches as well. For example, one of
the informants in Elsbach and Kramer’s (2003) study stated: “I think that magic is perhaps the
most important part of the pitch […] it’s a seduction, a promise of what lies ahead” (p. 296). This
would suggest that crowdfunding proposals are more successful when potential investors are able
to visualize the idea, and when the narrative is compelling. Rather than the dry language of
finance, crowdfunding pitches require colorful, vivid language:
Hypothesis 1b. Crowdfunding proposal success increase with the use of vivid language.
Aside from being personally excited by a campaign, individuals on crowdfunding sites
may contribute to a project because the idea resonates with them and they feel emotionally
invested in the cause. One way to get people to feel good about giving to a business idea is by
using positive language and appealing to people’s emotional side. Elsbach and Kramer’s (2003)
qualitative study of Hollywood pitches examined 28 videos of television and movie pitches and
found that ideas with great potential are those that generate positive emotions. Linguistic studies
have similarly shown that people who use more positive emotion terms are more likely to be
optimistic and confident, and are perceived as such (Tausczik and Pennebaker 2010).
Hypothesis 1c. Crowdfunding proposals success increases with positive emotion.

8
 

 Relatedly, individuals decide to donate to a project because to feel like they are helping a
great idea succeed—and may incidentally benefit from it too. For example, one evaluator in
Elsbach and Kramer’s study stated: “the best approach is sort of relational in that I have to feel
good and feel as if I’ve ‘won’ something as well” (Elsbach and Kramer 2003: 295) from
investing into an idea. Therefore language that will make potential investors feel connected to
the cause or the founders, and included as part of the venture will increase likelihood of success.
Hypothesis 1d. Crowdfunding proposal success increases with the use of inclusive
(relational) language.
GENDER AND LANGUAGE
Language use, however, is correlated with socio-demographics and geography. People with
different backgrounds and from different demographic groups have different communication
styles, and this often is exhibited through written language. In particular, the link between gender
and language has been extensively studied in linguistics. Research shows readers can often guess
the gender of an author through their language use, and machines can fairly easily correctly
assign a text a gender (Thomson and Murachver 2001), meaning gender is correlated with
language use in predictable ways. For example, several studies have shown that women and men
have substantively different writing styles, with women exhibiting more interpersonal sensitivity
(McMillan et al 1977), referencing emotional states more often (Mulac and Lundell 1994: 307),
and showing a higher level of social engagement compared to men. Thus, if language is
important in online funding platforms, gender will be important too, whether or not the gender of
the writer is made visible.

9
 

 In particular, research on gender and negotiation has found that – women are more
relational, express more emotion in their language (Kray and Thompson 2005), and they
approach social interactions as a “negotiation for closeness and connection” (Tannen 1990, cited
in Kray and Thompson 2005). Additionally, because women view morality through a care-based
perspective (Gilligan 1982), they are more likely to consider both parties’ interests in
negotiations and exchanges. Findings from these prior studies would suggest that:
Hypothesis 2a. Women use more inclusive language than men do.
Hypothesis 2b. Women use positive emotional language more than men do.
Prior research has found that men are more likely to use language related to money
(Newman et al. 2008). Additionally, as suggested above, women are generally more empathetic
to others’ needs, being able to position their idea as desirable from the perspective of its
ideational characteristics. Coupling this with the fact that women tend to use more perceptual
words (Newman et al. 2008)—such as “feel,” “taste,” or “delicious”—, we expect women are
better at telling a story that resonates with potential crowdfunding investors, and are less likely to
use money- or finance-related language than men are:
Hypothesis 2c. Women use more vivid language than men do.
Hypothesis 2d. Women use less language related to money than men do.

LANGUAGE, GENDER AND CROWDFUNDING SUCCESS
Prior research has found that gender socialization affects the type of jobs and projects men and
women choose (Barbulescu and Bidwell 2013)—for example, women identify less with
“masculine” jobs and expect to be less successful in them. Women-owned small businesses are

10
 

 also heavily concentrated in the service and retail sectors and are poorly represented in the
technology sector. These structural differences are often cited as the reason for the lack of
external funding for women-owned businesses (Anna, Chandler, Jansen, and Mero 1999).
However we propose that, even holding the type of project constant, women and men make
different linguistic choices in terms of framing these projects, and these linguistic choices
mediate the relationship between gender and crowdfunding success:
Hypothesis 3.Language mediates the relationship between gender and fund-raising
success.

DATA & METHODS
DATA
The data for this study was collected from the online crowdfunding platform Indiegogo.
Indiegogo is a successful peer-to-peer crowdfunding platform founded in 2008 with the initial
purpose of raising funds for film-making, and, soon after, for other artistic endeavors. As of
2014, Indiegogo hosts a wide variety of campaigns, from music and film to small business,
technology, and non-profit causes, allowing campaign founders to solicit money for an idea,
charity or start-up business. Indiegogo has raised a total of approximately 271 million dollars
since its founding, and, as of December 2014, over nine million people visit the site on a monthly
basis.
For the purposes of this analysis, we have elected to examine the relationship between
language used in the request for money, gender of founder-entrepreneur and of donor, and

11
 

 campaign success in the Technology and Small Business categories. Our choice was motivated
by the desire to constrain the type of fundraising market we are looking at to commercial
products and services. By way of comparison, other crowdfunding categories such as
Community, Education, and Health include donation requests for non-profit and humanitarian
causes; and the remaining categories pertain to sponsorship of artistic projects such as Music,
Film, Photography or Theatre.
Among these campaigns, we have selected campaigns in the English language, run by
solo entrepreneurs (teams of one), with campaign text longer than 50 words. This allows us to
compare and contrast the effect of language on campaign success. In order to assess the role of
gender in crowdfunding success, we have created a list including all participant names (founderentrepreneurs and donors-contributors) and used the Social Security database to probabilistically
assign gender based on recorded correlations between newborn gender and name choice.
The final dataset includes 9,943 campaigns from business and technology created
between February 2, 2010 and December 25, 2013. Of these, 66 percent were campaigns related
to small business projects, and 34 percent technology project campaigns. Women were solo
entrepreneurs on 35 percent of the small business campaigns, and on approximately 10 percent
of the technology campaigns.
Topic Models
In order to allay concerns that linguistic patterns identified are correlated with the nature of the
business projects proposed we have used a topic modeling algorithm to categorize campaigns
(Blei and Lafferty 2009). Topic modeling looks for words that co-occur in documents in rates
higher than expected if the words were distributed randomly. Words that have a relatively high
12
 

 co-occurrence are grouped together into topics, with each topic represented as a weighted
distribution over all unique words in the corpus. The content of each topic is related to the top
weighted words per topic. Each document is then represented as a weighted distribution over
every topic. The top weighted topics for each document indicates what that document is about.
Topic modeling has been used in Political Science to determine what issues are salient in
political texts, and is often used to analyze how focal issues change over time or how focal issues
differ between politicians.
The majority of campaigns on Indiegogo include a written description of the proposed
project or product. Each campaign in our dataset is one document, and is represented as a
weighted distribution over 10 topics. For examples of the highest weighted terms for each topic,
which suggests what that topic is about, see Table 1.
----------------------Insert Table 1 here
-----------------------We are interested in analyzing how language affects how much money a campaign raises
regardless of the type of project being proposed. We include the topic weights in our models to
hold constant the nature of the proposed campaign, isolating word use as key variables.
Dependent Variables
Money Raised. In our analysis, we focus on the amount of money raised by a campaign as a
dependent variable for testing Hypotheses 1 and 3. We have chosen money raised as a measure
of success because many Indiegogo campaigns have “flexible” terms – entrepreneurs receive all
the money regardless whether they have met their fundraising target or not.

13
 

 Content. In order to evaluate Hypotheses 2 a-d, we have sought to operationalize inclusive-,
emotional-, financial- and vivid language and use them as dependent variables. To operationalize
these concepts, we have employed the Linguistic Inquiry Word Count (LIWC) dictionary
(Tausczik and Pennebaker 2010). We compared this dictionary against campaign texts, and
scored the relative frequencies of words in each category as a ratio to total words in that category
in a campaign text to the total length of the campaign text. The linguistic content categories
chosen were: positive emotion, perception (to measure the vividness of the campaign
description), money (to measure the profit-orientation), and inclusiveness (to measure relational
orientation). Examples of words from each category can be seen in Table 2.
----------------------Insert Table 2 here
-----------------------Independent Variables
Composition. In order to address concerns that we may be capturing holistic qualities of the text
that are correlated with the topics of interest, we control for three compositional characteristics:
lexical diversity, readability, and concreteness. Lexical diversity is widely accepted as a measure
of writing and speech proficiency which may affect investor decision and be correlated with our
content categories. The lexical diversity index is measured as a type-token ratio by comparing
the number of unique, different words (types) to the total words used (tokens). Readability
estimates the years of education needed to understand a piece of writing. To calculate readability
we used McLaughlin’s SMOG formula (McLaughlin 1969), which is a function of the average
sentence length and the number of polysyllable words (words with three or more syllables) in
each sentence. In the context of crowdfunding we may expect simpler language (more

14
 

 readability, lower reading grade) is correlated with broader appeal from a wide range of
investors, but also with certain content—for example, business vocabulary may be lower in
readability than language related to emotion.
Concreteness measures the extent to which the language in a text is specific as opposed
to general. Both storytelling and business explanations may include both concrete and abstract
language, but we would expect business language to be more concrete. More abstract language
may allow more potential investors to connect with an idea, or it may be too vague to convince.
The concreteness score is calculated using the lexical resource WordNet (Princeton University
2010). Among other things, WordNet organizes English nouns and verbs hierarchically through
hypernyms and hyponyms. Each sense of each noun has set paths of hypernyms to reach 25 root
(general) words, and each verb the equivalent to 9 root verbs. For example, the word "furniture"
has this path to its root word, "entity": furniture -> furnishing -> instrumentality -> artifact ->
whole -> object ->physical_entity -> entity. Compare this to chair: chair -> seat -> furniture->
furnishing -> instrumentality -> artifact -> whole -> object ->physical entity -> entity.
The length of the path to the root word is equivalent to a word's number of hypernyms—
the more hypernyms a word has the more specific it is. “Furniture” has 7 hypernyms while
“chair” has 9, making “chair” more specific than “furniture.” We calculated the number of
hypernyms for each noun and verb in the text. The specificity score is an average over all of the
nouns and verbs in a text. Since the maximum depth (concreteness) encountered in the text was
slightly over 9.6, we have transformed this into a fraction between zero and one, where one is the
most concrete campaign text in the dataset.

15
 

 Controls – Campaign Attributes. As additional controls few included the campaign start date;
campaign duration (log transformed, in days); the number of campaign images, log- transformed
(ln (image+1)); campaign video (binary measure, yes or no); the length of the campaign text
(log-transformed); campaign goal (sum of money requested, log-transformed): and currency type
(pound or dollar). Descriptive characteristics of our dataset are presented in Table 3.
----------------------Insert Table 3 here
-----------------------ANALYTICAL STRATEGIES
Our analytical approach consists in three types of models. We first test Hypotheses H1 a-d and
H3 using an ordinary least squares estimator, with log-transformed money raised as a dependent
variable, fixed effects for campaign type, and robust standard errors. Second, we test Hypotheses
H2 a-d using the proportion of words in a campaign text from each type of linguistic category as
dependent variable, and a fractional logit estimator, implemented via STATA as a generalized
linear model with logistic link and robust standard errors.
Lastly, in order to allay concerns that men- and women-ran campaigns differ significantly
not only in campaign gains, but also in terms of other campaign characteristics as well, we
construct a matched sample of campaigns using coarsened exact matching (CEM), with
entrepreneur gender (female) as treatment and male-run campaigns as control (Iacus, King and
Porro 2008). We apply the CEM algorithm using maximal information, and generate strata with
different numbers of treated and control units; to compensate for the differential strata sizes,
CEM returns weights to be used in subsequent analyses. We then rerun the first two estimators

16
 

 explained above with CEM weights on the subsample of matched data consisting in 4,180
campaigns out of the total 8,793 campaigns in the sample.

FINDINGS
Our first hypotheses propose a relationship between linguistic content and crowdfunding
campaign success. Table 4, model 1 represents a baseline model assessing the relationship
between campaign characteristics and money raised. Model 2 includes the linguistic terms we are
interested in. The results confirm our Hypotheses 1a, c, and d indicating that business language is
negatively correlated with money raised, while positive emotion and inclusive language are
positively correlated with fundraising success. We fail to find support for Hypothesis 1 b, linking
vivid language to campaign success.
----------------------Insert Tables 4, 5 and 6 here
-----------------------Table 5 presents the relationship between gender of campaign founders and language
used, as estimated using fractional logit estimators. These results confirm our Hypotheses 2 a-d,
indicating that women are more likely to use language related to positive emotion, vividness, and
inclusion, and less likely to use money/ business related terms. Table 6 models 1-4 tests these
relationships using the CEM algorithm, with gender (female) as treatment. These models suggest
effects consistent with those presented in Table 5.
Lastly, we turn to Hypothesis 3 to examine the relationship between the entrepreneur’s
gender, language and money raised, and the mediating role of language, in Table 4 models 3 and

17
 

 4, and Table 6 models 5 and 6. The former results suggest that the effect of entrepreneur’s
gender (female) on money raised by the crowdfunding campaign is slightly diminished when
considering linguistic attributes; a multiple mediation analysis reveals this effect to be around
10.4 percent. However, the mediation effect of language is stronger in the matched sample
results presented in Table 4, accounting for 15-20 percent of the effect of gender on money
raised.
ROBUSTNESS CHECKS
In order to evaluate the robustness of our findings we have examined several alternative
specifications. We have considered controlling for the pre-defined campaign types (small
business and technology) instead of the ten types resulting from the topic model classification.
We have also examined only the subset of campaigns that raised at least one dollar (under the
assumption that campaigns that failed to raise any money may have not been seen). Lastly, we
have examined only campaigns that did not feature a video, in order to eliminate paralinguistic
cues, body language, and visual and auditory gender cues, and found results consistent with our
theory. When matching this latter subset of campaigns on all other characteristics with gender as
the treatment variable, we find that linguistic content (positive emotion, inclusiveness, vividness,
and business language) fully mediates the relationship between gender and money raised by the
crowdfunding campaign.
Prior research has suggested that women are less likely than men to benefit from
investments due to the fact that many gatekeepers in the venture capital industry are male, and
they are prone to in-group bias. More recent research on crowdfunding campaign outcomes
indicates that women perform better than men in crowdfunding campaigns due to the female

18
 

 investors (Greenberg and Mollick 2014). Using our data, we have examined patterns of male
and female donations (ratio of male and female donations and total money donated to total
donations and total money donated) and have found that the linguistic characteristics proposed
have similar impacts on male and female donors. Thus, although women are more likely than
men to donate to other women’s campaigns, language partially accounts the positive relationship
between female entrepreneurs and female donors. Conversely, the negative relationship between
female entrepreneurs and donations by men becomes stronger when we control for linguistic
characteristics, suggesting that both male and female donors respond with the same valence to
the linguistic content our theory focuses on.
Another study examining the relationship between gender and crowdfunding has
proposed that men and women donors are motivated by different benefits from crowdfunding,
such that women are more likely to donate for altruistic reasons, while men sometimes donate in
order to receive attractive products and perquisites (Marom, Robb, and Sade 2014). We are
currently in the process of manually classifying the types of benefits awarded to investors for
every level of participation in the campaigns we are studying, such that we will be able to
ascertain whether participants requested a significant benefit in exchange for their investment or
not.

DISCUSSION
Our results are robust, but they are not without limitations. One of these limitations the lack of a
quality variable in our dataset, such that we cannot account for the fact that some campaigns may
receive more funding than others because they offer a more interesting product, or a more
19
 

 credible business idea. We have sought to overcome this problem in two main ways. First, we
have controlled for composition-level characteristics of campaign text such as level of
abstractness, lexical diversity and readability. Second, we have used coarsened exact matching
on campaign characteristics with gender as a treatment variable in order to restrict the analysis to
a region of common support in the data.
In terms of external validity, crowdfunding campaigns represent an investment context
different from traditional settings. First, investors are not experts, but rather individuals from the
general population who are making investment decisions without an expectation of financial
returns. In this sense, crowdfunding is more similar to investment in non-profit causes, where
individuals contribute for non-material reasons, but may nevertheless evaluate ideas in order to
estimate their soundness or odds of success. Second, traditional investment agents (such as angel
investors or bank loan officers) make investments with the purpose of diversifying the risk: this
consideration is less important in crowdfunding, where interest in the idea itself or in the
promised products and services may drive investment decisions. However, prior research on
business and movie pitches suggests that the linguistic dimensions considered (positive emotion,
vividness, inclusiveness, and business language) are important in traditional settings as well.
While in these settings entrepreneurs’ business plans and financial forecasts may be subject to
close scrutiny, the pitch itself often serves to estimate the entrepreneur’s vision and ability to
inspire—both dimensions that may be correlated with linguistic content of the type proposed in
this study. In future research we plan to examine linguistic content characteristics of successful
and failed social entrepreneurship business ideas, in order to evaluate the role that these choices
play in an offline, centralized investment decision.

20
 

 CONCLUSION
In this study we identify an institutional setting (crowdfunding platforms) that benefits female
communication patterns and contributes to the reduction of gender inequality in the economic
sphere. Prior research on organizations and negotiations (for a review, see Kray and Thompson
2005) is replete with social mechanisms accounting for why women have worse outcomes in the
economic sphere than men, from receiving funding and support as independent entrepreneurs
(Wu and Chua 2012), to being hired and receiving raises and promotions within organizations.
Many of these studies key in on differences in self-confidence (Hargittai and Schafer 2006),
social capital, risk behaviors, and gender discrimination or stereotyping; and insignificant
differences in skill or actual performance (Hargittai and Schafer 2006; Kalleberg and Leicht
1991). This research suggests that women are less like to ask for favor, and more likely to be
modest in appraising their skills (Hargittai and Schafer 2006).
However, most these field studies are predicated on interactional gender categorization,
which is inherent in face-to-face interactions such as interviewing, negotiating, presenting a
pitch, or working together. Our study suggests that online environments may mitigate
interactional gender categorization because the mediated (virtual) environment allows
participants independence from their physical persona and can downplay the prominence of
gender as a salient social category.
Our theory also suggests scope conditions on gender categorization, such that women and
men may be evaluated differently eliciting free-form descriptions because of systematic
differences in the linguistic choices of men and women. This contributes to a new research
stream that is bringing computational methods, sociolinguistics, and sentiment analysis into

21
 

 sociology and organization theory to inform our understanding of social, cultural, and
organizational practices.

22
 

  

Table 1. Topic models for campaign categories (Latent Dirichlet Allocation, 10 topics)
Energy

Campaigns

Family

Small Business

Education

energy
electr
water
product
system
power
engin
manufactur
technolog
light
design
develop
patent
vehicle
research
test
car

folk
explain
trust
campaign
contributor
describe
cant
remind
nois
ask
introduce
break
doesn’t
tell
contribut
entir
real

children
dream
busi
love
would
family
donat
kid
life
community
school
mother
home
live
know
abl
give

busi
company
servic
loan
family
invest
provid
job
money
local
pay
month
market
small
income
property
owner

book
student
product
busi
publish
social
market
com
women
communiti
compani
global
educ
service
young
event
entrepreneur

Art
game
artist
music
site
video
youtub
art
com
advertis
websit
record
platform
studio
creativ
busi
fan
entertain

Perks
shirt
websit
print
donat
cloth
item
shop
store
com
thank
card
onlin
design
plus
page
busi
inventori

Phone apps
app
user
devic
softwar
applic
develop
mobil
data
android
system
phone
server
file
iphon
control
access
featur

Devices
3D
print
machin
batteri
product
prototype
design
charg
color
devic
manufactur
price
materi
case
power
ship
part

Food & Beauty
food
anim
product
bar, natur
restaur
room
bodi
line
paint
love
open
organ
made
beauty
local
car
man

23
 

 Table 2. Categories of language content, and sample words (LIWC dictionaries).
Money
bank
bargain
bought
dollar
donate
fee
sale
own
invest

Inclusive
and
along
both
include
inside
open
with
plus
we

Vividness
aroma
audible
bright
delicious
delicate
harmony
rotten
heavy
picture

24
 

Positive emotion
accept
admire
charm
enthusiastic
laugh
love
splendid
glamorous
adore

 Table 3. Descriptive statistics.
Mean
1.548
0.011
0.030
0.025
0.030
9.336
19308
3.778
0.959
0.032
0.557
0.746
11.221
6.008

Variable
Money (ln)
Vividness
Pos emo
Business
Inclusive
Goal (ln)
Start date
Duration (ln)
Images (ln)
Video
Lexical div
Concrete
Readability
Words (ln)

1
2
3
4
5
6
7
8
9
10
11
12
13
14

Money (ln)
Vividness
Pos emo
Business
Inclusive
Goal (ln)
Start date
Duration
(ln)
Images (ln)
Video
Lexical div
Concrete
Readability
Words (ln)

Std. Dev.
2.610
0.010
0.013
0.017
0.025
1.699
246
0.617
0.528
0.176
0.096
0.034
2.262
0.718

1

2

3

4

5

6

0.08
0.08
-0.16
0.04
-0.04
0.08

0.08
-0.17
-0.04
-0.05
0.03

-0.03
0.13
-0.07
0.02

0.04
-0.02
-0.03

0.06
-0.04

0.07

-0.07
0.18
0.17
-0.17
0.06
0.03
0.26

0.00
0.10
0.03
-0.02
0.10
-0.07
0.06

-0.01
0.01
0.02
0.02
-0.05
-0.09
-0.02

-0.01
-0.11
-0.06
0.13
0.03
-0.01
-0.19

0.01
-0.02
0.02
0.01
-0.02
0.02
-0.01

0.21
0.02
0.04
-0.17
-0.04
0.15
0.21

 

25
 

Min
0.000
0.000
0.000
0.000
0.000
6.215
18306
0.000
0.000
0.000
0.226
0.627
3.000
3.850

Max
14.223
0.145
0.111
0.146
0.154
21.416
19680
6.346
3.829
1.000
0.923
0.961
31.983
8.808

7

8

9

10

11

12

13

-0.24
0.09
0.05
-0.06
0.03
0.01
0.07

-0.02
-0.02
-0.05
-0.01
0.04
0.05

0.06
-0.21
0.05
0.03
0.28

-0.05
-0.02
0.01
0.08

0.06
-0.08
-0.88

-0.14
-0.02

0.16

   

 Table 4. Language and money raised by campaign (Hypotheses 1 a-d).
Money raised
Model 1

Money raised
Model 2

Money raised
Model4
Female
0.461***
(0.064)
Business
-11.229***
-11.015***
(1.498)
(1.493)
Vividness
1.839
1.261
(2.645)
(2.636)
Inclusive
2.916*
2.634*
(1.225)
(1.219)
Positive emotion
11.544***
10.464***
(1.906)
(1.902)
Goal (ln)
-0.125***
-0.124***
-0.110***
-0.111***
(0.015)
(0.015)
(0.015)
(0.015)
Start date
0.000***
0.000***
0.000***
0.000***
0.000
0.000
0.000
0.000
Duration (ln)
-0.234***
-0.236***
-0.229***
-0.231***
(0.044)
(0.044)
(0.044)
(0.044)
Images (ln)
0.438***
0.419***
0.427***
0.411***
(0.062)
(0.062)
(0.062)
(0.061)
Video
2.071***
2.021***
2.074***
2.028***
(0.197)
(0.195)
(0.196)
(0.195)
Pounds (baseline $)
-0.230**
-0.230**
-0.211**
-0.212**
(0.071)
(0.071)
(0.070)
(0.070)
Text length (ln)
1.625***
1.547***
1.583***
1.512***
(0.077)
(0.078)
(0.077)
(0.078)
Readability
-0.013
-0.008
-0.014
-0.009
(0.011)
(0.011)
(0.011)
(0.011)
Lexical diversity
6.317***
6.006***
6.163***
5.881***
(0.550)
(0.553)
(0.548)
(0.550)
Concreteness
2.711***
3.323***
2.686***
3.264***
(0.779)
(0.782)
(0.777)
(0.780)
R squared
15.2
16.1
15.9
16.6
Deg freedom
19
23
20
24
Note: Constant term and control for product type omitted from the analysis. N=8,793. ~ p<0.1, *
p<0.05, ** p<0.01, *** p<0.001

26
 

Money raised
Model 3
0.500***
(0.064)

 Table 5. Gender and language (fractional logits).
Vividness

Positive
Business
Inclusive
emotion
Model 1
Model 2
Model 3
Model 4
Female
0.069***
0.078***
-0.030*
0.029***
(0.020)
(0.011)
(0.015)
(0.008)
Goal (ln)
-0.013*
-0.004
0.001
0.017***
(0.006)
(0.003)
(0.004)
(0.002)
Start date
0.000
0.000*
0.000
0.000
(0.000)
(0.000)
(0.000)
(0.000)
Duration (ln)
0.021
0.004
-0.004
-0.006
(0.016)
(0.009)
(0.011)
(0.006)
Images (ln)
0.089***
0.021*
-0.050***
-0.005
(0.016)
(0.009)
(0.014)
(0.007)
Video
0.071~
0.057*
-0.077~
0.069***
(0.043)
(0.025)
(0.040)
(0.019)
Pounds (baseline $)
-0.014
0.027~
0.029
0.007
(0.024)
(0.014)
(0.018)
(0.010)
Text length (ln)
0.166***
0.038*
-0.235***
-0.039**
(0.034)
(0.017)
(0.022)
(0.012)
Readability
-0.022***
-0.014***
0.007*
0.009***
(0.006)
(0.003)
(0.003)
(0.002)
Lexical diversity
0.966***
0.326**
-0.749***
-0.230**
(0.246)
(0.118)
(0.157)
(0.088)
Concreteness
1.309***
-1.171***
0.707***
-0.265*
(0.330)
(0.165)
(0.210)
(0.126)
Chi-squared
875.8
666.7
2502.4
280.8
Deg freedom
20
20
20
20
Note: Constant term and control for product type omitted from the analysis. N=8,793. ~ p<0.1, *
p<0.05, ** p<0.01, *** p<0.001

27
 

 Table 6. Coarse Exact Matching – Gender, Language and Success.

Female

Vividness
Model 1
0.054~
(0.030)

Pos.
emotion
Model 2
0.100***
(0.014)

Business Inclusive
Model 3 Model 4
-0.044*
0.025*
(0.021)
(0.011)

Vividness
Positive
emotion

14.299***
(2.808)
-19.461***
(2.303)
4.694**
(1.810)

Business
Inclusiveness
Lexical
diversity
Readability
Concreteness

0.383
(0.301)
-0.009
(0.012)
2.505***
(0.593)
30.88

0.03
(0.097)
-0.018***
(0.004)
-1.106***
(0.242)
84.35

0.689***
-0.046
(0.151)
(0.080)
0.002
0.015***
(0.005)
(0.003)
0.922*
-0.158
(0.372)
(0.181)
33.79
34.39

Chi-squared
R-squared
Deg freedom
4
4
4
4
Note: N=4,810. ~ p<0.1, * p<0.05, ** p<0.01, *** p<0.001

28
 

Money received
Model 5
Model 6
0.461***
0.539***
(0.074)
(0.074)
10.230**
(3.710)

-3.560***
(0.438)
0.095***
(0.017)
4.849***
(1.100)

-3.852***
(0.441)
0.090***
(0.017)
4.168***
(1.104)

0.0595
8

0.0326
4

  

REFERENCES
Anna, A. L., G. N. Chandler, E. Jansen and N. P. Mero. 1999. “Women Business Owners in
Traditional and Non-Traditional Industries.” Journal of Business Venturing 15(3): 279303.
Anthony, Denise. 2005. "Cooperation in microcredit borrowing groups: identity, sanctions, and
reciprocity in the production of collective goods." American Sociological Review
70(3):496-515.
Barbulescu, Roxana, and Matthew Bidwell. 2013. "Do women choose different jobs from men?
Mechanisms of application segregation in the market for managerial workers.”
Organization Science 24(3):737-56.
Berger, Allen N. and Gregory Udell. “The Economics of Small Business Finance: The Roles of
Private Equity and Debt Markets in the Financial Growth Cycle.” Journal of Banking and
Finance 22(6-8):613-673.
Blei, David M. and John D. Lafferty. 2009. “Topic Models.” Pp. 71-94 in Srivastava, Ashok N.
and Mehran Sahami (eds), Text Mining: Classification, Clustering, and Applications.
Boca Ration, FL: Chapman & Hall/CRC.
Brownlow, Sheila, and Leslie A Zebrowitz. 1990. "Facial appearance, gender, and credibility in
television commercials." Journal of Nonverbal Behavior 14(1):51-60.
Carney, Dana R, Amy JC Cuddy, and Andy J Yap. 2010. "Power posing brief nonverbal displays
affect neuroendocrine levels and risk tolerance." Psychological Science 21(10):1363-68.
Coleman, Susan and Alicia Robb. 2009. “A Comparison of New Firm Financing by Gender:
Evidence from the Kaufmann Firm Survey Data.” Small Business Economics 33(4):387411.

29
 

 Elsbach, Kimberly D. and Roderick M. Kramer. 2003. “Assessing Creativity in Hollywood Pitch
Meetings: Evidence for a Dual-Process Model of Creativity Judgments.” The Academy of
Management Journal 46(3):283-301.
Gilligan, Carol. 1982. In a Different Voice: Psychological Theory and Women's Development.
Boston: Harvard University Press.
Greenberg, Jason, and Ethan R Mollick. 2014. "Leaning In or Leaning On? Gender, Homophily,
and Activism in Crowdfunding." Retrieved on December 31, 2014 from
http://papers.ssrn.com/sol3/Papers.cfm?abstract_id=2462254
Hargittai, Eszter and Steven Shafer. 2006. “Differences in Actual and Perceived Online Skills:
The Role of Gender.” Social Science Quarterly 87(2): 432- 448.
Iacus, Stefano M., Gary King and Giuseppe Porro. 2008. \Matching for Causal Inference
Without Balance Checking.". http://gking.harvard.edu/_les/abs/cem-abs.shtml.
Kalleberg, Arne L. & Kevin T. Leicht. 1991. “Gender and Organizational Performance:
Determinants of Small Business Survival and Success.” Academy of Management
Journal 34(1): 136-161.
Kramer, Cheris. 1977. "Perceptions of female and male speech." Language and Speech
20(2):151-61.
Kray, Laura J. and Leigh Thompson. 2005. “Gender Stereotypes and Negotiation Performance:
An Examination of Theory and Research. In Staw, Barry and Roderick M. Kramers (eds)
Research in Organizational Behavior Series, 26:103-182. Greenwich, CT: JAI Press.
Marom, Dan, Alicia Robb, and Orly Sade. 2014. "Gender Dynamics in Crowdfunding
(Kickstarter): Evidence on Entrepreneurs, Investors, Deals and Taste Based
Discrimination." NBER working papers.

30
 

 Massolution. 2013. 2013CF- Crowdfunding Industry Report. Retrieved from
http://www.crowdsourcing.org/editorial/2013cf-the-crowdfunding-industry-report/ on
December 29, 2014
McMillan, Julie R., A. Kay Clifton, Diane McGrath, and Wanda S. Gale. 1977. “Women's
Language: Uncertainty or Interpersonal Sensitivity and Emotionality.” Sex Roles 3(6):
545-599.
McLaughlin, G. Harry. 1969. “SMOG Grading – A New Readability Formula.” Journal of
Reading 12(8):639-646.
Mulac, Anthony and Torborg Louisa Lundell. 1994. “Effects of Gender-Linked Language
Differences in Adults’ Written Discourse: Multivariate Tests of Language Effects.
Language and Communication 14(3):299-309.
Newman, Matthew L., Carla J. Groom, Lori D. Handleman, and Janes W. Pennebaker. 2008.
“Gender Differences in Language Use: An Analysis of 14,000 Text Samples.” Discourse
Processes 45:211-236.
Pollack, Jeffrey M., Matthew W. Rutherford, and Brian G. Nagy. 2012. “Preparedness and
Cognitive Legitimacy as Antecedents of New Venture Funding in Televised Business
Pitches.” Entrepreneurship Theory and Practice 36(5):915-939.
Princeton University. 2010. “About WordNet.” Word Net. Princeton University.
http://wordnet.princeton.edu.
Rothman, Alexander J. and Peter Salovey. 1997. “Shaping Perceptions to Motivate Healthy
Behavior: The Role of Message Framing.” Psychological Bulletin 121(1):3-19.
Shane, Scott and Daniel Cable. 2002. “Network Ties, Reputation, and the Financing of New
Ventures.” Management Science 48(3):364-381.

31
 

 Stuart, Toby and Olav Sorenson. 2003. “The Geography of Opportunity: Spatial Heterogeneity
gin Founding Rates and the Performance of Biotechnology Firms.” Research Policy
32:229-253.
Tannen, Deborah. 1990. You Just Don't Understand: Women and Men in Conversation. New
York: William Morrow & Company.
Tausczik, Yla R., and James W Pennebaker. 2010. "The psychological meaning of words: LIWC
and computerized text analysis methods." Journal of language and social psychology
29(1):24-54.
Thomson, Rob and Tamar Murachver. 2001. “Predicting Gender from Electronic Discourse.
British Journal of Social Psychology 40(2):193-208.
Wu, Zhenyu and Jess H. Chua. 2012. “Second-Order Gender Effects: The Case of U.S. Small
Business Borrowing Cost.” Entrepreneurship Theory and Practice 36(3):443-463.

 
 
 
 

32