Electronic copy available at: http://ssrn.com/abstract=2524910

1

Community Impact on Crowdfunding Performance

Yael Inbar Ohad Barzilay

ABSTRACT

Many digital platforms, regardless of their business domain, follow the common practice of

incorporating social and community features in order to increase their user engagement and expand

their online community. Although this practice is advocated by the literature and clearly makes sense,

its implications are not well understood. In this research, we aimed to close this literature gap,

providing a theoretical framework and empirical evidence regarding the impact of the online

community on platform performance. As a testbed, we studied crowdfunding platforms, that is,

designated websites aimed at enabling entrepreneurs to raise money over the Internet. We used

comprehensive data collected from Kickstarter, the largest crowdfunding platform established to date.

We theorized that online platforms, such as Kickstarter, consist not of a single community but rather a

hierarchy of multiple, partially competing communities. These communities vary considerably with

respect to the interests of their members, their platform participation patterns, and their impact on

platform performance. Our suggested framework incorporates the notion of fluidity of online

communities; that is, online users and digital communities evolve and change over time. As the

interests of the online user change, so does the membership of her immediate community. The

proposed framework allows us to identify such community changes and, consequently, to better

identify pivotal members of online communities and predict their lifetime value as potential backers.

Empirically, we validated our theory by studying the participation patterns of over 6.3 million

Kickstarter users, who have supported more than 150 thousand crowdfunding campaigns over more

than 5 years. We demonstrated the growth of the different community types and estimated their

different impacts on crowdfunding performance over time. Interestingly, we found that some

communities, despite high participation rates, had negative impacts on crowdfunding campaign

success. We discuss managerial and practical implications of our theory and findings.

Electronic copy available at: http://ssrn.com/abstract=2524910

2

INTRODUCTION

Crowdfunding enables entrepreneurs to raise money from a large base of supporters over the Internet

(Belleflamme et al. 2011). In its early days, crowdfunding used social media such as mailing lists or

online social networks to achieve its goals. Today, with the maturing of Web 2.0 technologies and the

success of crowdsourcing (Giudici et al. 2012; Kleemann et al. 2008), there are dedicated

crowdfunding platforms, such as Kickstarter.com, which bring together project owners and potential

backers and facilitate information flow and transactions. Each crowdfunding campaign is centered on

a webpage that describes the project, which usually contains an embedded video, and that keeps track

of the fundraising process.

Crowdfunding was originally positioned as a new funding technology. Its value proposition included

providing entrepreneurs with affordable off-the-shelf tools to manage their fundraising campaign, to

communicate with their backers, and to facilitate money transfer. On the demand side of the market,

crowdfunding technology mitigates part of the perceived risk to backers by the enforcement of the all-

or-nothing policy (Hemer 2011): A minimum project-financing goal is set and a limited time period is

given for achieving said goal. The sum is transferred to the project owner only if the targeted amount

is pledged within the given period. If the amount is not reached, the project is considered unsuccessful

and the backers (funders) pay nothing.

Online users have been shown to react to weak signals that document the subtle actions of others

(Umyarov et al. 2013). On crowdfunding platforms, by choosing to fund a campaign, users reveal

their preferences (interests) as well as their consumption decisions (the funded product or service),

and this information may impact the decisions of other users (Faraj and Johnson 2011; Miller et al.

2009). At an aggregate level, the visibility of users' actions may affect other users’ decisions via

mechanisms of herding (Li and Wu 2014; Zhang and Liu 2012) and observational learning (Kim and

Viswanathan 2013).

As crowdfunding platforms have gained popularity and embraced social features, additional dynamics

have emerged. Ward and Ramachandran (2010) suggested that peer effects, and not network

3

externalities, influence consumption. Funders are looking for social interactions in crowdfunding

platforms (Gerber et al. 2012), and their participation in those platforms represents their belonging to

a community with similar interests. Crowdfunding platforms also manifest reciprocal behavior (either

direct or indirect) between users in terms of mutual backings (Zvilichovsky et al. 2013).

Platform owners may wish to leverage on their user community. Recent studies suggest that those

who are more socially involved in an online community built around a website are more likely to

consume the content of the site (Oestreicher-Singer and Zalmanson 2013). This correlation is

explained via mechanisms of developing a deeper sense of commitment (Bateman et al. 2011) and

perceived ownership (Preece and Shneiderman 2009).

It is not surprising then, that Kickstarter, like many other popular websites, refers to itself as a

community. In its official blog, it states: “A Kickstarter project does more than raise money. It builds

community around your work.” 1 Furthermore, Kickstarter highlights the opportunities that stem from

repeat backers and owners and interconnected communities.2 The company publishes and enforces

“Community Guidelines” governing behavior on the platform3 and holds meetups that take place in

major US cities designed to promote offline connections among its users.4

Crowd-based platforms leveraging on their community (i.e., existing users) introduce a unique

understudied tension. On the one hand, these platforms are positioned as allowing entrepreneurs to

raise money from 'the crowd', the flat and egalitarian mass of the population (Arazy et al. 2014); on

the other hand, online community performance and survival is dependent on a highly motivated and

engaged core of users, which is relatively small (Borgatti and Everett 2000; Kuk 2006). In the context

of crowdfunding, this core may not be large enough or self-sufficient enough (Yang and Wei 2009) to

finance campaigns at scale. The crowdfunding entrepreneur thus faces a dilemma – whether to focus

her efforts on drawing the community pivots to finance her campaign or to address potential backers

1 https://www.kickstarter.com/learn, visited April 29, 2014 2 http://www.kickstarter.com/blog/familiar-faces-interconnected-communities, visited April 29, 2014 3 http://www.kickstarter.com/blog/community-guidelines-update, visited April 29, 2014 4 http://www.kickstarter.com/blog/july-30th-kickstarter-everywhere, visited April 29, 2014

4

outside Kickstarter (e.g., her social network, media).

The literature provides no empirical evidence regarding the magnitude or the impact of the

crowdfunding community on campaign performance; in addition, the community effect on platform

performance at large has not been characterized. Although entrepreneurs acknowledge the importance

of community to the fundraising process (Gerber et al. 2012), it is not clear who core members of a

community are. Unlike other crowd-based platforms, Kickstarter lacks formal hierarchies such as in

Wikipedia (Arazy et al. 2014; Chen et al. 2010; Choi et al. 2010; Sundin 2011) and rating systems

such as in Stack Exchange (Barzilay et al. 2013).

This study addresses the gaps in the literature and advances our understanding of the dynamics of

online communities in general and of crowdfunding platforms in particular. We conducted a

comprehensive empirical study using a unique dataset comprising more than 150,000 crowdfunding

campaigns, conducted using the Kickstarter platform, the largest crowdfunding platform to date. We

estimated the extent to which the online community built around Kickstarter affected the fundraising

performance.

We found that Kickstarter's online community is actually composed of a hierarchy of multiple,

partially-overlapping, competing communities whose members manifest different patterns of

behavior. We differentiate among three community types: (1) ad hoc communities centered around a

single campaign or project, (2) communities of interest centered around a specific category, and (3)

the platform-centered community, whose members are Kickstarter enthusiasts, interested in

crowdfunding and innovation per se. We used typology suggested by Porter (2004) as a prism to

examine the characteristics of each of these community types.

Our rich dataset allowed us to take the novel approach of associating users with different communities

over time. The first backing by a user associated her with the ad hoc community around that

campaign. Subsequent backings (of different campaigns) may manifest a change in user's interests or

preferences and may associate her with a different type of community depending on her choices.

Through these associations, we incorporate an ecological mindset (Lin and Lin 2006; Wang et al.

5

2012), acknowledging the fluidity of online communities and the dynamics of community boundaries

and of user attention (Faraj et al. 2011).Surprisingly, we found that platform attachment acted as a

two-edged sword. We found that users who funded multiple campaigns of different categories

supported more campaigns than category-centered community members did. However, their support

missed the community added value associated with the support of members of the other community

types. Furthermore, this trend increased with users who focused first on one category and only after

some time supported campaigns in other categories. In contrast, we found that backing by category-

based community members had a positive impact on campaign success. In our analysis we controlled

for platform age and the effect of herding on the different community types in addition to other

controls advocated by the literature.

We discuss our results in the light of marketing theories, and compare our community-centered

empirical approach to the alternative social network analysis paradigm. We suggest that our

framework and findings may be applicable on other digital platforms as well.

CROWDFUNDING AND KICKSTARTER

Crowdfunding platforms attract millions of people around the world5 who have become involved in

this new process either as entrepreneurs who seek to secure funds for their ventures or as

investors/backers who wish to contribute money to campaigns of their choosing (with or without

receiving a reward). Crowdfunding can be based on one of several methods of raising money from the

crowd: equity purchase, loan, donation, and pre-ordering/reward based (Ahlers et al. 2012;

Belleflamme et al. 2011; Ingram and Teigland 2013).

Kickstarter, on which we focus in this study, is a reward-based platform that follows the "all or

nothing" business model. A campaign is successful upon achievement of the campaign financing

target in the allocated timeframe. We define success using this metric in our study rather than basing

success on execution or commercial success in later stages. As a for-profit company, Kickstarter

5 http://www.crowdsourcing.org/editorial/2013cf-the-crowdfunding-industry-report/25107 (visited April 21st, 2014)

6

receives 5% of funds collected for each successfully funded campaign.6

Kickstarter divides all campaigns into 13 categories7: Art, Comics, Dance, Design, Fashion, Film and

Video, Food, Games, Music, Photography, Publishing, Technology, and Theater. Currently, the most

popular category8 (in terms of number of projects) is Film and Video (24% of projects), and the

second most popular is Music (20%). The least popular category is Dance, with only 1963 (1.3%)

projects. Surprisingly, this was also the most successful category in our dataset, with a success rate of

70.4%. Another successful category is Theater, with a 64.12% success rate. The most unsuccessful

category is Fashion, with a success rate of only 29.4%.

Research in the domain of crowdfunding include analysis of the motivation to participate in

crowdfunding from the points of view of the backers and the campaign owners (Belleflamme et al.

2011; Schwienbacher and Larralde 2012), funders’ decision-making processes on whether to support

a campaign (Agrawal et al. 2011; Burtch et al. 2013; Kuppuswamy and Bayus 2013), and the key

success factors of crowdfunding campaigns (Mollick 2014).

THEORY AND HYPOTHESES

Emergence and Growth of Online Communities

Online communities are aggregations of individuals or business partners who interact around a shared

interest, where interaction is at least partially supported and/or mediated by technology and guided by

protocols or norms (Porter 2004). Members of an online community can be geographically dispersed

individuals who have no previous acquaintance with each other but who share common interests

(Faraj and Johnson 2011). Many digital networks that facilitate online activity of users do not meet

the usual definition of a community as they do not involve direct interactions among individuals

(Sundararajan et al. 2013). Nonetheless, users on crowdfunding platforms like Kickstarter emulate a

community establishment by leveraging on the co-backing relations of campaigns inducing a quasi-

social network (Provost et al. 2009).

6 http://www.kickstarter.com/hello, visited April 29, 2014 7 On June 11 2014, Kickstarter introduced two additional categories. In order to avoid potential bias we excluded from our dataset campaigns, which started after the new categories were introduced. 8 https://www.kickstarter.com/help/stats, retrieved on June 13, 2014

7

The success of an online platform and, in turn, an online community depends on its ability to retain

active participants over time (Arguello et al. 2006), the community’s sociability and usability (Preece

2001), and strong activity of its pivotal subgroup or subgroups (Maloney-Krichmar and Preece 2005).

Although most of the participants of online platforms are one-time users (for example, 68% of

newcomers to Usenet groups drop out after contributing a single post (Arguello et al. 2006)), the

remainder compensate for its small portion size by active and dominant participation patterns (Kuk

2006). As Kickstarter maintains a successful track record of successfully funded projects, we expected

to find similar dynamics happening on Kickstarter as well. We hypothesized:

H1a: Kickstarter performance relies on large portion of one-time contributors and on significant

participation by a smaller group of repeat backers.

One motivation to participate in an online community is a user’s structural embeddedness in the

community (Wasko and Faraj 2005). Ren et al. (2012) found evidence for such attachment by

strengthening either group identity or interpersonal bonds (with a stronger effect for group identity).

Bateman et al. (2011) showed that users’ behaviors on content sites are directly linked to their

commitment levels: As users increase their engagement with the site, they develop a deeper sense of

commitment to the website. Such socialization processes catalyze engagement around future foci and

drive future participation (Feld 1981).

We assumed that repeat Kickstarter backers form a community, induced by a quasi-social network

(Provost et al. 2009); we expected that over time these users would increase their community

embeddedness and platform attachment and, in turn, increase their participation in funding new

campaigns (Feld 1981). We hypothesized:

H1b: The participation ratio of repeat backers in funding new campaigns increases over time.

Another central component in our research design drew on the concept of fluidity of online

communities. Fluidity is the constant change of boundaries, norms, participants, artifacts, interactions,

and foci (Faraj et al. 2011). An online community evolves over time (Lin et al. 2008; Palla et al. 2007)

as do users and their attentions. Considering the fluidity concept, we expected that the backing

8

patterns of some Kickstarter users would evolve over time. Once a new user has joined the platform

and supported one of the campaigns, she may explore new opportunities offered by the platform.

Furthermore, some users may then fund campaigns that are substantially different from the first

campaign they backed, whereas other users may focus on supporting campaigns of the same area of

interest (e.g., music, food, technology). These considerably different backing patterns are likely to be

a manifestation of the user's change of attention or interest and may be amplified by the highly

flexible or permeable boundaries of the online community (Faraj et al. 2011). We hypothesized:

H2a: Over time, many repeat backers back only campaigns in a specific category. The percentage

of these backers within the community increases over time.

H2b: Over time, an increasing number of repeat backers back campaigns in multiple categories.

The percentage of these backers within the community also increases over time.

Community Hierarchy

Although members' participation and contributions are crucial for sustaining a successful community

(Butler 2001), patterns of use vary (Butler et al. 2013). Several studies in this field have attempted to

classify users of online communities by participation level. Different social and other roles of users of

online communities have also been identified (Kim 2000; Welser et al. 2007). Li and Bernoff (2011)

developed a ladder-type model to create profiles of online behavior, and Preece and Schneiderman

(2009) proposed a ‘Reader to Leader’ framework to examine participation levels. A recent study

(Oestreicher-Singer and Zalmanson 2013) showed that users’ willingness to pay for premium services

on a content website is strongly associated with their level of community participation.

Users' participation in online communities is unevenly distributed, and many communities include a

core subset of members who play key roles in sustaining the entire group (Faraj and Johnson 2011),

effectively creating their own self-sufficient sub-community (Yang and Wei 2009). Online

communities are characterized by subgroups with particular focuses (Faraj et al. 2011), and focused

and non-focused users exhibit different behavior patterns (Ung and Dalle 2010). Similarly, we

assumed that Kickstarter users could be classified into subgroups and that behavior patterns of these

users would differ.

9

However, we considered these subgroups as communities on their own. We hypothesized that the

aggregations of the different archetypes of repeat Kickstarter backers (single backing; multiple

backings in single category; multiple backings in multiple categories) correspond to distinct online

communities whose populations, interactions, and motivations vary considerably from one another.

We used a foci-driven granular perspective to identify communities centered around campaigns,

categories, and the platform. Figure 1 illustrates this idea. Backers who supported a single campaign

are grouped together in an ad hoc campaign-centered community with other backers who supported

only this campaign. Backers who supported multiple campaigns of a single category are group

together in a category-centered community, and backers who supported campaigns of multiple

categories belong to a platform-centered community.

Figure 1. An Illustration of Community Hierarchy.

Users are associated with communities based on their backing pattern.

A typology proposed by Porter (2004) emphasizes five main elements (the five P’s) of virtual

communities: (1) purpose (content of interaction), (2) place (extent of technology mediation of

interaction) – the levels of virtuality of the interaction, (3) platform (design of interaction), (4)

population (pattern of interactions), and (5) profit model (return on interaction) – whether a virtual

community creates tangible economic value. By examining Kickstarter backers through the lens of

Porter's typology we showed that backers' archetypes varied significantly from one another with

respect to the main elements. Therefore, we suggest that Kickstarter should not be considered as a

11

single community but rather as a set of multiple, hierarchical, partially competing communities.

Further, we qualitatively examined the community types using Porter's (2004) typology.

Project-Centered Communities

Project-centered communities are ad hoc communities centered around a campaign or a project9.

Members of these communities are focused on a specific project; many times, it is a physical product,

situated in the offline world. In some cases, members know the campaign owner outside Kickstarter

or heard about the project via their social networks. Project-based community members consider

Kickstarter to be a facilitating technology for the funding process. We found that the vast majority

(95%) created their Kickstarter account specifically to back a certain project.

The Kickstarter platform provides various means of communication for project-based communities.

Backers may post to the campaign page and leave comments on updates of the campaign owners.

Community members use these mechanisms not only to contact the project owner but also to conduct

conversations and send direct messages to other backers using the @username twitter convention.

Campaign owners use the campaign webpage to update the backers during the campaign and also

after it has ended. Some campaign owners use these features to attract backers to their subsequent

campaigns. In some cases the campaign is meant to finance an event happening in the physical world

(such as a show, concert, or exhibition) in which the members of the community meet in person.

Category-Centered Communities

Category-centered communities are communities of interest (Armstrong and Hagel 2000) centered

around a broad theme or topic – in our study, one of the 13 predetermined Kickstarter campaign

categories. The members of these communities are Mavens (Gladwell 2000), and their support of a

campaign signals to their followers that the campaign in question was appealing not only in and of

itself but also in a broader context of other campaigns in its category. These communities are

sometimes extensions of virtual communities outside Kickstarter. For example, gamers maintain user

9 In some cases, an entrepreneur may conduct subsequent campaigns to finance further stages of developments

or future versions of the product. We consider the backers of the subsequent campaigns to be part of the same

project-based community.

11

groups, forums, and websites that update the community regarding trending Game campaigns.

Technology blogs and magazines review interesting campaigns regularly.

The Kickstarter platform provides easy-to-navigate listings of all active campaigns in each category

and recently enhanced its search mechanism to enable focus on campaigns of a certain category or

sub-category.

Platform-Centered Communities

Platform-centered community members are Kickstarter backers who have supported more than one

campaign in at least two different campaign categories. This criterion differentiates them from the

category-centered members. Platform-centered community members' interests are broad and eclectic,

and members may be considered as innovators and early adaptors (Moore 1991; Rogers 2010). Their

interests are aligned with what the Kickstarter platform has to offer – a hub for creativity and

entrepreneurship (Boudreau and Lakhani 2013). They maintain active community membership over

longer periods than do members of the other community types.

The Kickstarter platform provides means for platform-level communication and interaction:

Kickstarter maintains a blog, a Facebook page, and a newsletter to provide the community members

with updates regarding the platform and to promote featured projects. Featured projects ('staff picks',

'trending now') are also shown on the site homepage. We note that these platform-level

communications are moderated by the platform owners and do not promote asynchronous

conversations among the users.

Kickstarter facilitates additional channels for information flow via the design of user profile pages. It

was found that crowdfunding users browse the profile pages of other users before deciding on their

own backing (Burtch et al. 2013). Kickstarter highlights the campaigns that were backed by each user.

Furthermore, if some of the backed campaigns are still live (their funding period has not yet ended)

this fact is emphasized.

We hypothesized:

H3a: Backers who belong to different community types have different participation patterns.

12

Using the fluidity perspective, we argue that Kickstarter users evolve over time. This may occur

within a community, or it may result in a member leaving one community to join another (virtual)

community residing on the same platform. This fluidity manifests changes in users' focus, interest,

and attention.

We expected that over time, some category-centered members would begin to back campaigns outside

their category of interest. We consider this to be a manifestation of changes in interest and attention;

hence we associated these backers with a new community type – the category-diverged community.

Figure 2 depicts transitions between communities as a state automaton10.

Figure 2. User Fluidity across Communities over Time.

10 The criteria by which we classified the communities may seem simplistic. For instance, one may note there

are no backward arrows in the diagram, suggesting that once a user becomes a platform-centered community

member, she can no longer 'go back' to being a category-centered community member. Indeed, for robustness,

we considered several alternative ('relaxed') models; for example, we put thresholds on the minimum number of

backings that triggered a category change. These models yielded similar results.

13

One notable element is that there are two distinct communities, namely platform-centered and

category-diverged. We expected members of these communities to vary considerably in their

participation patterns; hence, we hypothesized that:

H3b: Category-diverged community members exhibit behaviors different from category-centered

community members and from platform-centered community members.

The Impact of Kickstarter Communities on Platform Performance

Next, we sought to estimate the impact of Kickstarter communities on its performance. We expected

that the impacts of the four community types (the original three types and the fluid category-diverged

type) would be different. We analyzed whether one type of community dominates the others. We

expected the difference would be driven by the following two factors: (1) the heterogeneity and

diversity of the community members and their interest, and (2) the persistence of backing behavior,

namely, whether they change their community type from category-centered to category-diverged.

There is no consensus among researchers regarding the effect of user heterogeneity and diversity on

community performance. Members who belong to multiple communities are subject to conflicting

forces. On the one hand, they enjoy the overlap benefits (i.e., diversity of choice, interest, and

stimulation). On the other hand, each member has only limited resources (time, attention, and money)

and may have trouble maintaining active membership in multiple communities over time.

López and Butler (2013) questioned the viability of designs for local online communities that focus

narrowly on single topics, goals, and audiences. Raban et al. (2010) studied the effect of diversity of

an initial seed on the long-term sustainability of online chat channels and found that channels with

more diverse populations were more likely to survive than those with more homogeneous populations.

Similarly, a recent study by Zhu et al. (2013) concluded that membership overlap (members who

belong to multiple communities at the same time) is associated with the survival of the focal

community. Yoganarasimhan (2012) suggested that high clustering around the initiating node,

implying that users belong to a close-knit community, is associated with low performance outcomes.

14

In contrast, other studies (Wang and Kraut 2012) have shown that lower content diversity is

associated with a larger and more connected group of followers. Gu et al. (2014) argue that the desire

to make good quality decisions should steer people away from homophily and toward heterophily.

Wang et al. (2012) showed that sharing members with other groups reduces growth rate, suggesting

that membership overlap puts competitive pressure on online groups.

We drew on the theoretical framework of membership overlap (Wang et al. 2012; Zhu et al. 2013) and

adapted it to the settings of our theoretical framework. In the lens of membership overlap, category-

centered community members do not divide their resources and are able to identify value and quality

in their area of interest. Therefore, their support is more indicative of campaign success than that of

platform enthusiasts. We hypothesized:

H4a: Campaigns supported by a higher percentage of category-centered community members

will have a higher likelihood of raising their stated goal.

We expected that this correlation stemmed from causal relation; that is, the category-centered

community support drives the success. In order to show this, we needed to eliminate alternative

explanations and to resolve potential endogeneity issues. Therefore, we studied the dynamics along

the fundraising period, and investigated the interplay with the different community types over the

course of the campaign. Specifically, we focused on eliminating the herding effect from our

estimations. We elaborate on this aspect in the identification strategy section.

We hypothesized:

H4b: The support of category-centered community members ("homogeneous") has a greater

positive impact on campaign success than the support of platform-centered community members

("heterogeneous", "overlapping community membership").

H4c: The support of persistent community members (project-centered, category-centered) has a

greater positive impact on campaign success than the support of platform-diverged community

members ("non-persistent").

We suggest that H4b and H4c are driven by the "community added value" of the backing; getting

15

support from users who belong to cohesive communities may result in successive support from their

fellow members. This idea draws on a recent marketing theory regarding the network value of

products. It was shown that an increase of the sales of one product in a product network propagates to

an increase of the sales of the products linked to it (Oestreicher-Singer et al. 2013). Hence, the

economic impact of a sale should also account for these future sales. Although Kickstarter

communities do not have on-platform explicit links between their members, we argue that category-

centered communities maintain such links off platform.

DATA COLLECTION AND DESCRIPTION

Kickstarter does not provide an API nor does it provide access to a directory of past campaigns and

users. Furthermore, its web interface does not allow for exhaustive searches. Therefore, to build our

datasets, we developed two dedicated Web crawlers – one for comprehensive, static historical data,

and the other for monitoring the dynamics of live projects.

The first crawler implemented a recursive “breadth-first search” algorithm (Leiserson et al. 2001) that

traversed the links from each campaign page to its backers’ pages and from each backer page to the

pages of the campaigns she backed and created. Crawling was initiated using a publically available

seed comprising 45,000 campaigns (Pi 2012). Recursive iterations from campaigns to backers, and

back to campaigns, were performed until the number of newly discovered campaigns per iteration

converged. The following data was collected by the crawler:

Campaign data: campaign owner, financing goal, financing duration, campaign creator profile,

profiles of all backers (funders), detailed reward levels and reward selections, use of a video,

amount of money pledged, comments, updates, location, category, and sub-category .

User data: personal data (name, location, date account was opened, number of Facebook friends),

campaign ownership data (links to and number of all campaigns created by owner), and campaign

funding data (links to and number of all projects backed by the user).

The data presented in this paper is a result of an exhaustive crawling that ended on August 23, 2014.

The dataset contains the details of 6,632,241 backers and 157,661 completed campaigns, covering

16

97% of the campaigns completed on Kickstarter by that date. Table 1 displays descriptive statistics of

the main campaign attributes that were incorporated into our models.

On June 11, 2014 Kickstarter introduced two new campaign categories11. In order to avoid potential

analysis bias we removed from our dataset all the campaigns that were launched at or after that date.

Hence, our final dataset comprised 146,386 campaigns, to which a total of 6,352,395 backers had

pledged USD 1,168,477,721.

Table 1. Descriptive Statistics – Project Attributes

Variable Min-Max Mean/Probability Std. dev

Goal (USD) 0.01-100,000,000 22,758.06 439,652.41

Duration (Days) 1-91 35.26 14.16

IsSuccessful (Goal Achieved) 0/1 0.45 0.50

Level of Funding Achieved

(Raised/Goal) 0-41,535 2.03 147.85

Number of Weeks since the Owner

Joined the Platform 0-263.14 28.37 39.02

Has Video 0/1 0.81 0.39

Num. of Backers 0-105857 107.15 866.16

Num. of Reward Levels 0-227 8.76 5.48

Limits on Number of Backers in one or

more reward category 0/1 0.59 0.49

Has FB Friends in profile 0/1 0.59 0.49

Owner HadCreated Previous Projects 0/1 0.12 0.33

Owner HadBacked Other Projects 0/1 0.42 0.49

Num. of Comments 0-176467 31.79 973.72

Total Money Pledged 0-10,266,845 8,135.47 68,879.65

In some estimations we split the campaigns according to their size (measured by total number of

backers). We found that the subset of campaigns that had less than 10 backers exhibited no obvious

pattern of distribution of the different community types, unlike other subsets of campaigns. Hence, for

these estimations we show here only campaigns with more than 10 backers12.

For privacy reasons, Kickstarter does not reveal the timing of the backings nor the amount pledged by

each backer. The backers’ list of each campaign is updated only after 10 new backings have been

11 https://www.kickstarter.com/blog/introducing-two-new-categories-journalism-and-crafts (visited at November 4th, 2014) 12 Moreover, as Kickstarter reveals the campaign backers' profiles only when the number of backers reaches 10 (or when the

campaign is over), the visible list of backers of such campaigns remained empty until the last day of the campaign and thus

the metrics do not represent the dynamics that were present when the campaign was alive.

17

made. Furthermore, the list is shuffled on every update. Therefore, merely from observing site

listings, it is impossible to deduce the order in which backers pledged to a particular campaign or the

timing of a particular pledge. In order to overcome this deficiency we developed a second, dynamic

crawler, run in parallel on multiple machines. Every 10.2 hours on average, this crawler recorded a

snapshot comprised of all the data associated with all live campaigns as detailed above. The crawler

was initiated on September 12, 2013 and has been run constantly since. The data presented in this

paper comprise daily snapshots of 37,161 completed campaigns between September12 2013 and

August 23, 2014. The total number of snapshots was 3,348,511.

METHODOLOGY

Our main methodological challenge in this study was related to studying the community impact on

campaign success. Specifically, we wished to distinguish the effects of the different communities

(platform-centered, category-centered, project-centered and category-diverged) on campaign

dynamics. However, relying on observational data makes it difficult to untangle the underlying

forces.

Our central tool for the analysis was the estimation of logistic regression models for predicting the

probability of campaign success. In the following paragraphs, we summarize our empirical approach,

the issues we addressed, and describe our econometric model.

Using the number of backers of each community type in order to predict campaign success is

obviously endogenous; the number of backers is directly linked with campaign success. The more

backers a campaign has – it is more likely to succeed. Hence, instead of using the absolute number of

backers of each community, we evaluated the percentage of each community type among the total

number of campaign backers. This is a zero-sum game; an increase of one percent in the share of

some community type must be compensated via decrease of one percent in the share of the other

community types (combined). Hence, this regression allowed us to estimate the impact of community

composition on the success likelihood of a campaign and also to measure the relative impacts of the

different communities.

18

One may note that the interpretation of the regression odds ratio as "all-else-equal" contradicts the

"zero-sum game" nature of using relative community shares. In addition, having the shares of all four

types as explained variables results in multicollinearity. Hence, we left one of the types out (category-

diverged ratio). This is a common practice used to estimate the impact of composition (e.g.,

Francalanci and Galal (1998) used it to estimate the impact of worker composition on productivity).

Backer perspective. In order to provide empirical evidence for our theory that Kickstarter is

composed of multiple communities, we performed several independent sample t-tests to show that

users who are members of different communities manifest differences in behavior patterns that are

statistically significant.

As our original data set is campaign driven, we compiled a complimentary view of the data in which

the backer is the main point of analysis. We tracked the behavior of each backer from the time she

joined the platform: the number of campaign backed by her, the categories and sizes of the backed

campaigns, the number of days since her last backing (as a proxy for user liveness), her backing

period, the backing frequency during her activity period on the platform, and the success rate of the

campaigns backed by this user.

For backers of campaigns within our dynamic data set we were also able to calculate the timing of the

backings (in days since the campaign started and as a fraction of the funding period), the state of the

supported campaigns at the time of the backing (in terms of the percentage of the funding goal raised),

and the percentage of the backer's total backings that were made before the campaigns backed by her

reached 100% of the funding goal.

Furthermore, we also studied the change of user behavior over time. For users classified as category-

diverged we compared their behavior before and after "being diverged" using paired sample t-tests.

We showed that the changes in behavior were statistically significant.

Identification Challenges

Incorporating campaign dynamics and herding. Another challenge of estimating the community

impact on campaign performance is related to the campaign state at the time when backings were

performed and the presence of herding effect.

19

When using the percentage of community members in order to predict campaign success we

encountered an endogeneity challenge. The naïve logistic regression cannot distinguish whether

community support drove campaign success or whether campaign success attracted the community

members to support the campaign13. Moreover, even when the campaign backing occurred before the

campaign met its goal, it might be the case that a pledge was made as part of herding dynamic that

began before the pledge and/or that the community member was reacting to the herding. To avoid this

issue, only community members who made pledges before the herding effect began can be

considered.

In order to empirically identify such campaign tipping points, we examined the distribution of all

campaigns, according to their percent of goal raised at the campaign end day. We found that there are

almost no campaigns that raised more than 40% but less than 100% of their goal. Put differently, more

than 97% of campaigns able to raise 40% of their target amount succeeded in meeting their goal. This

may suggest that by the time 40% of a campaign goal is reached, herding is already in action.

Therefore, we based our regression models on snapshots of the campaign states at the following levels

of funding (percentage of goal raised): 5%, 10%, 20%, 30%, 40%, 100%, and final state14. The

multiple models allowed us to identify the dynamics of the different communities and their impacts on

the campaign funding process.

Controlling for platform age. In order to accurately measure the forces that affect the performance

of a mature digital platform, such as Kickstarter, one needs to realize that these forces evolve over

time (Faraj et al. 2011). Hence, the dynamics measured today may be different from those three years

ago. Therefore, our empirical challenges were to identify the trends that correlated with the platform

maturity process and to control for these changes over time in our estimations.

In order to identify the trends evident on the Kickstarter platform we leveraged upon our panel data

obtained at monthly resolution. We found that the Kickstarter online community grew over time and

13 When a campaign reaches its funding goal, the funding process continues until the campaign duration set beforehand is over. In such cases, the campaigns eventually raise over 100% of the funding goal. 14 For unsuccessful campaigns (i.e., those that did not reach the funding goal), the 100% snapshot is similar to

their final state.

21

took increasing part in financing new campaigns. This fact made it difficult to estimate the impact of

community support on campaign success while considering the entire 5 years of the platform

existence (April 2009 – June 2014).

We addressed this concern in two ways: (1) we estimated the community impact on campaign success

only for a recent, limited time interval, in which the community’s relative growth (compared to first-

time backers) was stagnant and (2) we used the fact that community growth was monotonic. When

considering the entire data set in our logistic regression, we controlled for the month in which the

campaign was launched as a proxy for platform maturity.

The Econometric Model

We used a logistic regression model to predict the probability of campaign success given the ratios of

backers from the four community types among the total campaign backers. We controlled for

campaign characteristics, owner attributes, and platform age at campaign launch day. The control

variables were consistent with the main models estimated for predicting Kickstarter campaign success

in previous studies (Burtch et al. 2013; Marom and Sade 2013; Mollick 2014; Zvilichovsky et al.

2013). We used the variable isSuccessful to represent campaign success, defined as attainment of the

funding target within the specific timeframe. The variable has the value of 1 if a campaign achieved

this target.

Formally, we estimated the following:

𝑉(𝑖𝑠𝑆𝑢𝑐𝑐𝑒𝑠𝑓𝑢𝑙) = 𝛼 + 𝛼1𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦 + ∑ 𝛽𝑗𝐶𝑎𝑚𝑝𝑎𝑖𝑔𝑛 𝐶𝑎𝑡𝑒𝑔𝑜𝑟𝑦 𝑗 +𝐽𝑗=1

∑ 𝛾𝑗𝐶𝑎𝑚𝑝𝑎𝑖𝑔𝑛𝐶ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠𝑗 + ∑ 𝛿 𝑗𝑂𝑤𝑛𝑒𝑟𝐴𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑠𝑗𝑃𝑗=1 +𝐾

𝑗=1

∑ 𝜂 𝑗𝐶𝑜𝑚𝑚𝑢𝑛𝑖𝑡𝑦𝑇𝑦𝑝𝑒𝑅𝑎𝑡𝑖𝑜𝑠𝑗𝑀𝑗=1 + 𝜖

Where:

PlatformMaturity represents the Kickstarter platform age in months at campaign launch. We

observed that community dynamics evolved over time as previously described (Faraj et al. 2011),

and thus we normalized for platform age.

CampaignCategoryj are dummy variables representing 12 of the 13 Kickstarter categories. This

was controlled for because backing patterns may vary across different categories.

21

CampaignCharacteristicsj include the following campaign attributes:

o Goal is the funding target (in USD) that the campaign owner sought to raise. This variable is

logged in our model due to high variance. Other currencies were converted to dollars using

the exchange rate at the time of the campaign.

o Duration is the number of days allocated as funding period, after which the success of the

campaign is determined.

o NumRewardCategories is the number of reward levels offered by the campaign owner. It

was suggested that this number is correlated with campaign sophistication level and hence

may affect the propensity of potential backer to support the campaign.

o HasLimitedCategory indicates whether at least one reward level is limited to a certain

number of backers. This indicator may serve as a measure of the sense of exclusiveness the

owner is attempting to build around backing the campaign.

o HasVideo indicates whether the campaign description incudes a video. This measure was

considered in previous studies as a measure of campaign quality (Mollick 2014)

OwnerAttributesj include the following owner attributes:

o HasFBFriends is a dummy variable indicating whether the campaign description includes a

direct report from Facebook of the number of the owner’s Facebook friends. This indicator

may serve as a control for the owner's social capital outside the platform.

o OwnersPastCampaignsInfoj includes the variables that indicate whether the owner had

created a campaign on Kickstarter in the past (HadCreated) and whether she had backed

other campaigns before launching the current campaign (HadBacked). These measures were

previously found to have a significant impact on the funding success (Zvilichovsky et al.

2013).

CommunityTypeRatioj is the percentage of a community-type subgroup relative to total campaign

backers: ProjectCenteredPercent, CategoryCenteredPercent, PlatformCenteredPercent, and

CategoryDivergedPercent. In order to avoid multicollinearity we left one of the types out

(category-diverged ratio).

22

The conditional probability of campaign success is thus: 𝑒𝑣

1+𝑒𝑣

RESULTS

We quantitatively estimated the extent to which Kickstarter really is a community. Namely, we

explored the community features of the platform and estimated the magnitude and impact of these

features on campaign performance. Specifically, we considered the impact of repeat backers and their

participation patterns on the success of crowdfunding campaigns.

The Growth of Kickstarter Communities

The official Kickstarter statistics15 report that 6.3 million individuals have made 15.8 million pledges

since Kickstarter’s inception. This is approximately 2.5 pledges per backer on average. However,

pledges on Kickstarter do not distribute uniformly; the one-time backers behave differently than

repeat backers. Kickstarter's repeat backers account for 30% of all backers (1,904,375 backers).

Repeat backers have supported 5.9 campaigns on average and are in charge of 72% of the pledges on

the platform.

15 https://www.kickstarter.com/help/stats, 13 June 2014

23

Figure 3 displays the mean percentages of new vs. repeat backers across all campaigns per month

since the establishment of Kickstarter (April 2009). We considered all campaigns launched in each

month, and for every campaign we calculated the percentage of new and repeat backers relative to the

total number of campaign backers. Over time, as the platform has matured, new backers have

accounted for a diminishing percentage of all backers, whereas repeat backers have accounted for an

increasing percentage of all backers. Interestingly, we observed that in the most recent months, the

majority of the backings were performed by existing users. This fact is particularly salient considering

that new backers are constantly joining the platform. This trend may be interpreted as reflecting

Kickstarter’s transition from a crowdfunding technology, focused on mitigating concerns of potential

backers, to a crowdfunding community, a reoccurring meeting place for members of shared-interest

communities.

Figure 3. Mean Percentage of New Backers and Repeat Backers on Kickstarter over Time.

We associated Kickstarter backers with sub-communities based on their interests. We associated first-

time backers with an ad hoc project-based community, which was centered on a particular campaign

(or series of campaigns of the same owner). We divided the repeat backers into platform-centered and

category-centered based on whether they supported campaigns in multiple Kickstarter categories or

not.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 4 7 11 13 16 19 22 25 28 31 34 37 41 43 46 49 52 55 58 61

Mea

n P

erce

nta

ge o

f To

tal C

amp

aign

B

acke

rs

Platform Age (Months)

Repeat Backers

New Backers

24

Figure 4 displays the proportions of the three community types out of all Kickstarter backers, over

time. Project-centered type members comprise the highest share of all Kickstarter backers (71% on

June 2014). It is evident that this subgroup proportion has steadily decreased since the platform’s

early days. The percentages of the other two subgroups, platform-centered and category-centered

communities, increased over time (21% and 8% of all Kickstarter backers, respectively, at the end of

the data collection period).

Figure 4. Percent of Community Members Types out of All Kickstarter Backers over Time.

The Fluidity of Kickstarter Communities

We showed that backing behavior was not arbitrary, but rather a manifestation of user interests and

attentions. Therefore, different backing behaviors are correlated with other characteristics of user

participation and interactions with the online platform. Furthermore, we found that over time some

users changed their attention. In particular, we focused on users who began as category-centered

members who subsequently began to support campaigns in other categories. We refer these users as

category-diverged community members. Members of this quasi-community demonstrated behavior

that was significantly different from that of either category-centered members or platform-centered

members.

Table 2 summarizes the differences among the four backer types. It shows that our theory for

considering these communities as distinct is anchored in empirical data. We performed independent

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 4 7 11 13 16 19 22 25 28 31 34 37 41 43 46 49 52 55 58

Per

cen

t o

f A

ll K

icks

tart

er B

acke

rs

Platform Age (Months)

Platform-Centered Category-Centered Project-Centered

25

sample t-tests to compare category-centered and platform-centered community members, as well as to

compare category-centered and community-diverged members before they were diverged. We also

compared platform-centered and community-diverged members after their diversion. In addition, we

conducted paired sample t-tests to compare category-diverged members' behavior before and after

they diverged. All results are statistically significant and support hypotheses H3a and H3b.

Platform-centered members supported significantly more campaigns than category-centered members,

and category-diverged members supported more campaigns than platform-centered members. These

differences may be interpreted as due to loss of interest of category-centered members after 3.2

backings on average. On the other hand, the category-diverged members performed their third

backing in a different category, which renewed their interest, and might have exposed them to new

stimulations. This stimulation encouraged them to persist for another 8.4 additional backings, many

more than the average number of backings of platform-centered members.

Category-diverged members were also active for longer times than other types of members. We used

the metric of "Days Since The Last Campaign Backed" as a proxy for user 'liveness' (Fader et al.

2010); when this number is smaller, the community is more likely to have live members.

We also observed that the average personal success rate (ratio of successful campaigns of the total

campaigns supported) is high across all community type members (86%-92%). This confirms the

official Kickstarter reports stating that most of the money pledged by Kickstarter users (88%) was on

successful campaigns.

26

Table 2. Comparison of Different Types of Backers16

Community

Metric type

(std. dev.)

Project-

Centered

Category-

Centered

Platform-

Centered

t-test

P value

Category-Diverged

(Category-centered members who later

supported campaigns outside of the category)

Number (Percent)

4,459,096

backers

(70.2%)

523,256

backers

(8.2%)

1,058,691

backers

(16.7%)

- 310,352 backers

(4.9%)

Days Since The Last

Campaign Backed

509

(375)

385

(309)

281

(265) ***

198

(215)

While Being

Category-

Centered

While Being

Platform-

Centered

t-test

P Value

Num. of Backings 1.02

(0.136)

3.26

(3.84)

5.46

(13.28) ***

2.45

(3.63)

8.42

(20.76) ***

Personal Success

Rate

86.7%

(33.8%)

90.7%

(25.84%)

89.3%

(25.07%) ***

91.8%

(23.4%)

90.1%

(20.44%) ***

Period Active

in Days -

301

(250)

400

(319) ***

146

(169)

347

(289) ***

Days before First

Backing as Type

2.9117

(104.17)

179.35

(235.85)

245.47

(264.33) ***

113.04

(174.25)

138.11

(184.85) ***

***- significant at the 0.001 level

16 Measured on June 10th, 2014 17 Measured on a random sample of 841,159 backers.

27

The data shown in Figures 3 and 4 were also analyzed from a campaign perspective. We considered

all campaigns launched in each month, and for every campaign we calculated the percentage of the

four community types relative to the total number of campaign backers. Figure 5 displays the

participation of the four Kickstarter community types in campaigns over time. As the platform

matured, project-centered backers, the vast majority of whom are single-time contributors, accounted

for a deceasing share of all campaign backers; these backers accounted for less than 50% in the last

few months analyzed. Platform-centered backers accounted for an increasing proportion of the total

campaign backers, reaching 30%. The percentage of category-centered backers reached a peak of

9.3% on month no. 38 and accounted for 6.8% in the last month analyzed. The category-diverged

subgroup gradually increased and was 13.9% of backers in the last month analyzed.

Figure 5. Mean Percent of Community Members Types of Total Campaign Backers over Time.

The Kickstarter ecosystem exposes its members to diverse stimulations, different from their original

interest (i.e., different from the category of their first funded campaign). A significant number of

campaign supporters are existing Kickstarter backers who have never supported a campaign of the

same category. Moreover, for most campaign categories this trend increased over time.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 4 7 11 13 16 19 22 25 28 31 34 37 41 43 46 49 52 55 58 61

Mea

n P

erce

nt

of

Tota

l Cam

pai

gn

Bac

kers

Platform Age (Months)

Platform-Centered Category-Diverged Category-Centered Project-Centered

28

Figure 6 displays the mean percent of serial backers who back the given campaign's category for the

first time. This measure represents the share of backers who are new to the category but not to the

platform. The figure shows a major difference between the Games category (black) and other

categories. Games is the only category where the percentage of repeat Kickstarter backers who are

backing this category for the first time decreased over time. This suggests that Games backers either

come from within the community (Games category-centered community members) or are exogenous

to Kickstarter (other virtual or physical Games communities).

Figure 6. Mean Percentage of Serial Backers who were Backing the Given Campaign Category

for the First Time as a Function of Platform Age.

The Impact of Kickstarter Communities on Campaign Performance

We have shown that Kickstarter communities play a significant part in the crowdfunding process. We

wished to identify the impact of the different communities on campaign performance and campaign

dynamics. Specifically, we were interested in the effect of community support on the likelihood of

campaign success. We wished to identify those community members who behave as early adopters

and drive campaign success and those community members who support campaigns only after

financing goals are met.

0%

5%

10%

15%

20%

25%

30%

35%

40%

1 4 7 11 13 16 19 22 25 28 31 34 37 41 43 46 49 52 55 58 61

Mea

n P

erce

nt

of

Seri

al B

acke

rs W

ho

Bac

k Th

is

Cat

ego

ry f

or

the

Firs

t Ti

me

Platform Age (Months)

Art

Comics

Dance

Design

Fashion

Film & Video

Food

Music

Photography

Publishing

Technology

Theater

Games

29

One of the dynamics that may occur during the funding period of successful campaigns is herding (Li

and Wu 2014; Zhang and Liu 2012). After receiving some support, a campaign may reach a tipping

point, which in turn, provides a positive signal to other backers to support the campaign (Kim and

Viswanathan 2013). We found evidence for two such tipping points during the course of the funding

period: the first came after raising some initial portion of the financing goal and the second after

meeting the financing goal.

Figure 7 shows the distribution of all campaigns as a function of the percent of goal raised at the

campaign end day. Almost no campaigns raised more than 40% but less than 100%. Indeed, more

than 97% of the campaigns that raised 40% of their target amount succeeded in meeting the funding

goal. Another spike in the histogram occurs after the 100%; after the campaign has been reached its

goal, the feasibility of the project becomes concrete, mitigating the concerns of some potential

backers who then pledge. The two tipping points may be driven by different mechanisms; the first

tipping point may be a result of on-platform observational and social learning (Kim and Viswanathan

2013). The second may be driven by exogenous sources such as media coverage.

Figure 7. The Distribution of All Campaigns based on Percent of Goal Raised at the Campaign

End Day.

In order to estimate the impact of different communities on campaign success, we considered the

timing of the backing and the potential effect of herding. In order to do so, we again used the

campaign perspective. For every campaign in our 'dynamic' dataset, we calculated the percentages of

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

Nu

mb

er o

f C

amp

aign

s

Percent of Campaign Goal Funded

31

backer types at various points during the funding period. Using this set, which spans 10 months,

allowed us also to control for platform maturity, and in particular the change in sub-community

proportions over time. Table 3 and Figure 8 detail our results. We found that, on average, project-

centered community members (mostly first timers) comprised of 38.7% of early supporters (backing

the project before 5% of its target goal was reached). Their portion of the funders population

increased as the campaign progressed to 52.5% when the campaign goal was met. This implies that

project-centered members are positively affected by the first herding wave, to a greater extent than are

the other types of backers.

Platform-centered and category-diverged member shares decreased over time, which implies that

these sub-types are less prone to herding than the other types. From a community perspective, this

may suggest that the signaling within these two communities is weaker than within the other

communities.

The second tipping point (after reaching 100%) had opposite effects on project-centered and category-

diverged community members, which associates project-centered with the early adopters, and

category-diverged with the late majority (Moore 1991; Rogers 2010).

Table 3. Mean Percentage of Campaign Backers Types over the Levels of the Percentage of

Funding Goal Raised

Percentage

Backer Raised

Type

5% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Final

Project-Centered 38.7 40.1 44.1 45.7 46.7 47.8 48.2 48.2 48.8 49.8 52.5 48.9

Category-Centered 7.2 7.1 7.2 7.2 7.1 7.2 7.1 7.2 7.0 7.1 6.8 6.9

Platform-Centered 34.2 34.6 32.7 32.0 31.8 31.0 30.7 30.7 30.4 29.8 28.4 29.9

Category-Diverged 19.8 18.2 16.1 15.1 14.4 14.0 13.9 14.0 13.8 13.4 12.3 14.3

31

Figure 8. Mean Percentages of Campaign Backer Types as a Function of Percentage of Funding

Goal Raised.

We wished to further study the impact of the different communities on campaign dynamics by

considering the quality of the campaign. If some communities are better than others in identifying

good campaigns (campaigns that will become successful), we expected that this fact would also be

manifested in their dynamics. We repeated the measurements described above (results summarized in

Table 3, Figure 8) after division of the dataset into successful and unsuccessful campaigns.

0%

20%

40%

60%

80%

100%

5% 11% 21% 31% 41% 51% 61% 71% 81% 91% 111%

Mea

n P

erce

nta

ge o

f C

amp

aign

Bac

kers

Percent Money Raised

Project-Centered

Platform-Centered

Category-Centered Who TurnedPlatform-Centered

Category-Centered

32

0%

10%

20%

30%

40%

50%

60%

5% 11% 21% 31% 41% 51% 61%

Mea

n P

erce

nt

of

Pro

ject

-C

ente

red

Bac

kers

ou

t o

f To

tal

Cam

pai

gn B

acke

rs

Percent of Funding Goal Raised

Project-Centered

Unsuccessful Campaigns Successful Campaigns

0%

5%

10%

15%

20%

25%

5% 11% 21% 31% 41% 51% 61%

Mea

n P

erce

nt

of

Cat

ego

ry-

Div

erge

d B

acke

rs o

ut

of

Tota

l C

amp

aign

Bac

kers

Percent of Funding Goal Raised

Category-Diverged

Unsuccessful Campaigns Successful Campaigns

Percentages of backers who were platform-centered community members or category-centered

members did not differ between successful and unsuccessful campaigns. The percentage of project-

centered community members, who increased in proportion for all campaigns during the funding

period, was lower for unsuccessful campaigns than for successful campaigns. The opposite was the

case for category-diverged community members; the percentage of category-diverged backers was

higher for unsuccessful campaigns than for successful campaigns. Figure 9 shows platform-centered

and category-diverged behavior side by side. The difference may be because project-centered

community members are better able to evaluate the quality and potential of a crowdfunding campaign

in a certain domain than are backers from other communities. Similarly, category-diverged members

have lower competence to judge a successful campaign than do project-centered backers. This also

confirms our theory: category-diverged members, by definition, are outside the comfort zone of their

original interest.

Figure 9. Ratio of Project-Centered and Category-Centered Community Members of Total

Campaign Backers during Fundraising Period.

33

Table 4 compares the backing patterns of different community types. Our data show that most of the

backings on Kickstarter are done after the campaign goal has reached. We observed that after

category-diverged members began to support campaigns outside of their original category they had a

significant tendency to support campaigns after the campaign goal had been reached. This implies that

the impact of category-diverged members on campaign success decreases after they have diverged

from their original category.

Table 4. Comparison of Backing Patterns of Different Community Members

Community Type Project-

Centered

Category-

Centered

Platform-

Centered

t-test

P Value

(Category

vs.

Platform)

Category-Diverged

While

Being

Category-

Centered

While

Being

Platform-

Centered

t-test

P Value

N (# of backers) 1,390,295 230,245 612,754 46,403 46,403

Backing Day

(absolute) 17.22 16.8 15.9 0.00*** 16.2 16.5 0.00***

Backing Day

(percent) 52% 51.8% 48.3% 0.00*** 48.8% 48.7% 0.613

Money

Raised

when

Backed (std. dev.)

All

Campaigns

228%

(586%)

356%

(756%)

287%

(549%) 0.00***

410%

(817%)

337%

(532%) 0.00***

Successful

Campaigns

Only

260%

(623%)

393%

(789%)

318%

(574%) 0.00***

461%

(862%)

374%

(557%) 0.00***

Unsuccessful

Campaigns

Only

20.7%

(49.98%)

22.6%

(38.1%)

23.3%

(49.2%) 0.007***

23.7%

(50.29%)

21.1%

(27.79%) 0.009***

Backings Before Goal

Reached 54.8% 45.3% 47.03% 0.00*** 42.12% 37.31% 0.00***

**- significant at the 0.05 level; ***- significant at the 0.01 level

In order to measure the impact of Kickstarter communities on campaign performance, we estimated a

logistic regression model to predict the probability of campaign success given the ratios of the four

community types among the total campaign backers. In order to normalize for the effect of herding we

repeated the regression considering campaigns at key stages of their fundraising period: after 5% was

collected, 10%, 20%, 30%, 40%, and 100%. As we showed earlier, there almost no unsuccessful

campaigns that raised more than 40% of their target goal. We also conducted the regression after the

fundraising period has ended.

34

We controlled for campaign characteristics, owner attributes, and platform age at campaign launch

day. Our model’s control variables are consistent with the main models estimated for predicting

Kickstarter campaign success in previous studies (Burtch et al. 2013; Marom and Sade 2013; Mollick

2014; Zvilichovsky et al. 2013). Table 5 summarizes our results.

35

Table 5. Binary Logistic Regression Model: Predicting the Successful Funding of a Crowdfunding

Campaign on Kickstarter incorporating Subgroup Share Variables at Different Levels of Percentage of

Money Raised (Category-Diverged Subgroup Excluded)

5% 10% 20% 30% 40%

Last/

100% Final State

Exp(B)

(S.E.)

Exp(B)

(S.E.)

Exp(B)

(S.E.)

Exp(B)

(S.E.)

Exp(B)

(S.E.)

Exp(B)

(S.E.)

Exp(B)

(S.E.)

Logged Goal 0.339***

(0.193)

0.466***

(0.117)

0.569***

(0.095)

0.555***

(0.095)

0.585***

(0.105)

0.194***

(0.042)

0.131***

(0.020)

Original

Duration

0.982**

(0.007)

0.990**

(0.005)

0.988***

(0.004)

0.997

(0.004)

0.991

(0.005)

0.985***

(0.002)

0.988***

(0.001)

Has FB

Friends

0.705***

(0.132)

0.761***

(0.085)

0.819***

(0.076)

0.940

(0.080)

0.996

(0.094)

0.850***

(0.039)

0.856***

(0.018)

Platform

Maturity

0.991

(0.024)

0.993

(0.015)

0.994

(0.014)

0.973

(0.014)

0.968

(0.017)

0.991

(0.007)

0.998***

(0.001)

Weeks On

Platform

1.001

(0.001)

1.002***

(0.001)

1.002***

(0.001)

1.002**

(0.001)

1.002***

(0.001)

1.003***

(0.000)

1.004***

(0.000)

Had

Created

1.280

(0.193)

1.122

(0.117)

0.959

(0.102)

0.813**

(0.102)

0.757**

(0.115)

1.259***

(0.051)

1.294***

(0.029)

Had

Backed

1.079

(0.142)

1.130

(0.093)

1.150

(0.081)

1.097

(0.087)

0.975

(0.103)

1.520***

(0.040)

1.457***

(0.018)

Has

Video

3.853***

(0.478)

1.374

(0.210)

2.093***

(0.153)

1.874***

(0.148)

1.728***

(0.160)

1.706***

(0.061)

1.533***

(0.027)

Num

Reward

Categories

1.013

(0.011)

1.014**

(0.007)

1.011

(0.006)

1.013

(0.007)

1.011

(0.008)

1.048***

(0.003)

1.074***

(0.002)

Has

Limited

Category

1.229

(0.185)

1.293**

(0.109)

1.073

(0.092)

1.076

(0.096)

0.939

(0.114)

1.158***

(0.043)

1.032***

(0.019)

Project-

Centered

Percent

1.004

(0.006)

1.005

(0.004)

1.014***

(0.004)

1.013***

(0.004)

1.014***

(0.004)

1.015***

(0.002)

1.023***

(0.001)

Platform-

Centered

Percent

1.016**

(0.008)

1.011**

(0.005)

1.024***

(0.005)

1.014***

(0.005)

1.019***

(0.007)

1.010***

(0.003)

1.013***

(0.002)

Category-

Centered

Percent

1.034***

(0.011)

1.027***

(0.007)

1.030***

(0.007)

1.022***

(0.007)

1.026***

(0.009)

1.041***

(0.004)

1.046***

(0.002)

Category-

Diverged

Percent

- - - - - - -

Category

Controls Yes Yes Yes Yes Yes Yes Yes

Constant 8.659

(1.823)

6.012

(1.119) 2.2 (0.948)

21.4***

(0.986)

44.5***

(1.136)

140.3***

(0.476)

251.6***

(0.136)

Observations 1561 2972 4156 4700 4952 17460 89362

Log

likelihood: 1602 3750 5046 4715 3747 19714

88719

Cox &Snell

R-Square: 0.107 0.102 0.104 0.103 0.084 0.200

0.222

Nagelkerke

R-Square: 0.157 0.137 0.142 0.153 0.148 0.270

0.312

36

The results should be interpreted as follows. We omitted the category-diverged proportion from the

input variables, hence the odds ratio of one of the other three community types should indicate

whether an increase in the proportion of that community type at the expense of the category-diverged

members increased the success likelihood of the campaign. Because we applied this consideration for

each of the three types we can also deduce the relative impacts of the three types to each other and not

only with respect to category-diverged members.

The results show that category-centered community members have the highest positive effect on

campaign success, and that category-diverged community members have a negative effect (with

respect to the alternative of replacing them with backers of other communities). These results may be

explained via theory of membership overlap: It has been suggested that members belonging to

multiple communities are subject to conflicting forces. On the one hand, these members enjoy the

benefits of overlap (i.e., diversity of choice, interest, and stimulation). However, each member has

only limited resources (time, attention, and money) and may struggle to maintain active membership

in multiple communities over time. Category-centered community members do not divide their

resources among multiple communities and are able to identify value and quality in their area of

interest. Therefore, they are a more indicative index than platform enthusiasts of campaign success.

From a marketing perspective, financing a crowdfunding campaign relies on having a few mavericks,

mavens and social connectors as the product early adopters (Gladwell 2000). Our results suggest that

category-diverged members has a negative impact on campaign success and performance comparing

to all other types because they do not play any of the mentioned roles.

DISCUSSION AND IMPLICATIONS

Social Network Perspective

Crowdfunding dynamics may be considered a case study of broader online phenomena such as

information diffusion, social learning, and herding. Previous studies have investigated these

phenomena either where all users are equally exposed to the same information or in the context of

online networks where explicit associations between the users exist and information channels may be

37

identified. In contrast, we study these phenomena from a community perspective, which differ from

either these approaches. We used community association as a proxy for information flow and peer

effect and were able to show that these dynamics vary among different community types. Our

approach may be particularly useful in the context where explicit associations between the users do

exist but cannot be observed (for example, they exist outside of the platform).

From a social network analysis perspective, the category-centered and platform-centered community

members may be considered as weak ties (Granovetter et al. 1983) or long ties (Susarla et al. 2012) of

the project-centered community members. Such ties are known to be effective brokers of information

(Easley and Kleinberg 2010) and to drive network diffusion. Our proposed framework of community

hierarchy provides an additional, somewhat complementary, qualitative interpretation of the concept

of long ties. We argue that these ties are not only 'long' but also 'tall'; that is, they diffuse the

information to a community of different granularity types, whose members have some broader interest

than the specific campaign in question. The cohesiveness and user visibility levels on these

communities determine their "community added-value" and in turn their impact on campaign success.

Digital Platforms and Two-Sided Markets

Our findings may also be used to better predict a customer's lifetime value (Venkatesan and Kumar

2004). Our results suggest that platform owners should be cautious when they leverage upon their

existing users via cross-sale offers. Diverging users from their original interest may have

counterproductive implications and negative impacts associated with their future platform

participation. In that sense, crowdfunding platforms may incorporate conversion strategies that were

found to be effective on e-commerce sites (Moe and Fader 2004).

Furthermore, many commercial platforms wish to leverage upon their community to increase their

revenues. Kickstarter, however, is part of an emerging platform family, in which this trend is

amplified. Kickstarter is a two-sided market (Eisenmann et al. 2006; Rochet and Tirole 2003). Its

users consist of campaign owners and backers, and the crowdfunding platform serves as an

intermediary between them (Zvilichovsky et al. 2013). As such, not only platform owners are

incentivized to leverage upon their community, but also campaign owners wish to nurture

38

communities of their own. Such dynamics may be found on other digital platforms such as YouTube

(uploading and watching movies), Airbnb (hosting and renting), and eBay (buying and selling).

Community hierarchy may be found on these platforms as well. For example, YouTube channel

subscribers may be classified as an ad hoc community, category-based communities are centered

around YouTube categories such as Music and Sport, and platform enthusiasts are those who

subscribe to channels on different categories.

LIMITATIONS

This study aspires to contribute to the literature in the domain of online communities. However,

supporting a crowdfunding campaign involves real money rather than other types of utility used on

other online communities such as time, effort, or knowledge. Although in economic research, money

may be regarded as yet another form of resource (Marshall 2004), there may be differences in user

perceptions regarding community contributions that involve real money. Hence the findings of this

research may not be directly applicable to online communities outside the domain of crowdfunding.

Nevertheless, studies have shown that commercial websites, such as e-commerce and paid content

platforms, may generate online communities as well (Oestreicher-Singer and Zalmanson 2013).

Like other observational studies, this research faced data limitations and identification challenges.

However, we utilized the large data set to increase our confidence in the reported results.

SUMMARY AND CONCLUSION

In this paper, we introduced a community-oriented estimation approach and used it to study the

community features apparent in Kickstarter and their impact on campaign performance. We explored

the growth of Kickstarter communities that were platform centered, category centered, and campaign

centered and investigated the growing dominance of repeat backers in Kickstarter project financing.

We found that the category-based community is a strategic pivotal subgroup in the overall Kickstarter

community. We suggest that drawing this type of member to support a campaign will increase its

likelihood of success, via signaling to their peers either on Kickstarter or outside. We also investigated

the impact of user fluidity on her backing patterns and consequently on her contribution to campaign

39

success. We found that such users, after detaching from their community, increase their platform

participation, however, their backing lacks community added value and their impact on campaign

success decreases.

How far can we generalize the results of this research? The study described articulates and examines

empirically some of the mechanisms that underlie social behavior in digital spheres to the point where

behavior becomes more predictable.

41

References

Agrawal, A. K., Catalini, C., and Goldfarb, A. 2011. The Geography of Crowdfunding. Working

Paper 16820, National Bureau of Economic Research

Ahlers, G., Cumming, D., Günther, C., and Schweizer, D. 2012. Signaling in Equity Crowdfunding.

Available at SSRN 2161587

Arazy, O., Nov, O., and Ortega, F. 2014. The [Wikipedia] World is Not Flat: On the Organizational

Structure of Online Production Communities. Proceedings of the 22nd European Conference

on Information Systems (ECIS’2014)

Arguello, J., Butler, B. S., Joyce, E., Kraut, R., Ling, K. S., Rosé, C., and Wang, X. 2006. Talk to me:

foundations for successful individual-group interactions in online communities. Proceedings

of the SIGCHI conference on Human Factors in computing systems 959–968.

Armstrong, A., and Hagel, J. 2000. The real value of online communities. Knowledge and

communities, 85-95.

Barzilay, O., Treude, C., and Zagalsky, A. 2013. Facilitating Crowd Sourced Software Engineering

via Stack Overflow. in Finding Source Code on the Web for Remix and Reuse 289–308.

Springer, New York.

Bateman, P. J., Gray, P. H., and Butler, B. S. 2011. Research Note—The Impact of Community

Commitment on Participation in Online Communities. Information Systems Research 22(4)

841–854.

Belleflamme, P., Lambert, T., and Schwienbacher, A. 2011. Crowdfunding: Tapping the right crowd.

International Conference of the French Finance Association (AFFI) 11–13.

Borgatti, S. P., and Everett, M. G. 2000. Models of core/periphery structures. Social Networks 21(4)

375–395.

Boudreau, K. J., and Lakhani, K. R. 2013. Using the crowd as an innovation partner. Harvard

Business Review 91(4) 60–69, 140.

Brandtzaeg, Petter Bae, and Jan Heim. “User Loyalty and Online Communities: Why Members of

Online Communities Are Not Faithful.” In Proceedings of the 2nd International Conference

on INtelligent TEchnologies for Interactive enterTAINment, 11:1–11:10. INTETAIN ’08.

ICST, Brussels, Belgium, 2007. http://dl.acm.org/citation.cfm?id=1363200.1363215.

Burtch, G., Ghose, A., and Wattal, S. 2013. An Empirical Examination of the Antecedents and

Consequences of Contribution Patterns in Crowd-Funded Markets. Information Systems

Research 24(3) 499–519.

Butler, B., Kiesler, S., and Kraut, R. 2013. Community Effort in Online Groups: Who Does the.

Leadership at a distance: Research in technologically-supported work 171.

Butler, B. S. 2001. Membership size, communication activity, and sustainability: A resource-based

model of online social structures. Information systems research 12(4) 346–362.

Chen, J., Ren, Y., and Riedl, J. 2010. The effects of diversity on group productivity and member

withdrawal in online volunteer groups. Proceedings of the SIGCHI Conference on Human

Factors in Computing Systems 821–830.

Choi, B., Alexander, K., Kraut, R. E., and Levine, J. M. 2010. Socialization tactics in wikipedia and

their effects. Proceedings of the 2010 ACM conference on Computer supported cooperative

work 107–116.

Easley, D. and Kleinberg, J. 2010. Networks, crowds, and markets: Reasoning about a highly

connected world. Cambridge University Press.

Eisenmann, T., Parker, G., and Van Alstyne, M. W. 2006. Strategies for two-sided markets. Harvard

business review 84(10) 92.

41

Faraj, S., Jarvenpaa, S. L., and Majchrzak, A. 2011. Knowledge collaboration in online communities.

Organization Science 22(5) 1224–1239.

Faraj, S., and Johnson, S. L. 2011. Network exchange patterns in online communities. Organization

Science 22(6) 1464–1480.

Feld, S. L. 1981. The Focused Organization of Social Ties. American Journal of Sociology 86(5)

1015–1035.

Gerber, E. M., Hui, J. S., and Kuo, P.-Y. 2012. Crowdfunding: Why people are motivated to post and

fund projects on crowdfunding platforms. Proceedings of the International Workshop on

Design, Influence, and Social Technologies: Techniques, Impacts and Ethics

Giudici, G., Nava, R., Rossi Lamastra, C., and Verecondo, C. 2012. Crowdfunding: The New Frontier

for Financing Entrepreneurship? Available at SSRN 2157429

Gladwell, M. 2000. The tipping point: How little things can make a big difference, Little,Brown and

Co., New York.

Granovetter, M. 1983. The strength of weak ties: A network theory revisited. Sociological

theory, 1(1), 201-233.

Gu, B., Konana, P., Raghunathan, R., and Chen, H. M. 2014. Research Note—The Allure of

Homophily in Social Media: Evidence from Investor Responses on Virtual Communities.

Information Systems Research 25(3) 604–617.

Ingram, C., and Teigland, R. 2013. Crowdfunding Among IT Entrepreneurs in Sweden: A Qualitative

Study of the Funding Ecosystem and ICT Entrepreneurs’ Adoption of Crowdfunding.

Available at SSRN 2289134

Kim, A. J. 2000. Community building on the web: Secret strategies for successful online communities,

Addison-Wesley Longman Publishing Co., Inc., New York.

Kim, K., and Viswanathan, S. 2013. The Signals in the Noise: The Role of Reputable Investors in a

Crowdfunding Market. Available at SSRN 2258243

Kleemann, F., Voß, G. G., and Rieder, K. 2008. Un (der) paid Innovators: The Commercial

Utilization of Consumer Work through Crowdsourcing. Science, Technology & Innovation

Studies 4(1) 5–26.

Kuk, G. 2006. Strategic interaction and knowledge sharing in the KDE developer mailing list.

Management Science 52(7) 1031–1042.

Kuppuswamy, V., and Bayus, B. L. 2013. Crowdfunding Creative Ideas: the Dynamics of Projects

Backers in Kickstarter. Working Paper

Leiserson, C. E., Rivest, R. L., Stein, C., and Cormen, T. H. 2001. Introduction to algorithms, MIT

press.

Li, C., and Bernoff, J. 2011. Groundswell: Winning in a world transformed by social technologies,

Harvard Business Press.

Lin, S., and Lin, F.-R. 2006. Towards an Ecological Perspective on the Evolution of Online

Communities of Practice. Proceedings of the 39th Annual Hawaii International Conference

on System Sciences, 2006. HICSS ’06 (Vol. 6) 134a–134a.

Lin, Y.-R., Chi, Y., Zhu, S., Sundaram, H., and Tseng, B. L. 2008. Facetnet: a framework for

analyzing communities and their evolutions in dynamic networks. Proceedings of the 17th

international conference on World Wide Web 685–694.

López, C. A., and Butler, B. S. 2013. Consequences of content diversity for online public spaces for

local communities. Proceedings of the 2013 conference on Computer supported cooperative

work 673–682.

Maloney-Krichmar, D., and Preece, J. 2005. A multilevel analysis of sociability, usability, and

community dynamics in an online health community. ACM Transactions on Computer-

Human Interaction (TOCHI) 12(2) 201–232.

42

Marom, D., and Sade, O. 2013. Are the Life and Death of a Young Start-Up Indeed in the Power of

the Tongue? Lessons from Online Crowdfunding Pitches. Available at SSRN 2255707.

Marshall, A. 2004. Principles of economics. Digireads.com Publishing.

Miller, K. D., Fabian, F., and Lin, S. 2009. Strategies for online communities. Strategic Management

Journal 30(3) 305–322.

Moe, W. W., and Fader, P. S. 2004. Dynamic Conversion Behavior at E-Commerce Sites.

Management Science 50(3) 326–335.

Mollick, E. 2014. The dynamics of crowdfunding: An exploratory study. Journal of Business

Venturing 29(1) 1–16.

Moore, G. A. 1991. Crossing the Chasm: Marketing and Selling Technology Products to Mainstream

Customers, Harper-Collins, New York.

Oestreicher-Singer, G., and Zalmanson, L. 2013. Content or Community? A Digital Business Strategy

for Content Providers in the Social Age. Management Information Systems Quarterly 37(2)

591–616.

Oestreicher-Singer, G., Libai, B., Sivan, L., Carmi, E., and Yassin, O. 2013. The Network Value of

Products. Journal of Marketing, 77(3), 1-14.

Palla, G., Barabási, A.-L., and Vicsek, T. 2007. Quantifying social group evolution. Nature 446(7136)

664–667.

Pi, J. 2012. AppsBlogger. http://www.appsblogger.com/kickstarter-infographic Visited: May 1, 2013

Porter, C. E. 2004. A typology of virtual communities: A multi‐disciplinary foundation for future

research. Journal of Computer‐Mediated Communication 10(1) Article 3.

Preece, J. 2001. Sociability and usability in online communities: Determining and measuring success.

Behaviour & Information Technology 20(5) 347–356.

Preece, J., and Shneiderman, B. 2009. The reader-to-leader framework: Motivating technology-

mediated social participation. AIS Transactions on Human-Computer Interaction 1(1) 13–32.

Provost, F., Dalessandro, B., Hook, R., Zhang, X., and Murray, A. 2009. Audience selection for on-

line brand advertising: privacy-friendly social network targeting. Proceedings of the 15th

ACM SIGKDD international conference on Knowledge discovery and data mining 707–716.

Raban, D. R., Moldovan, M., and Jones, Q. 2010. An empirical study of critical mass and online

community survival. Proceedings of the 2010 ACM conference on Computer supported

cooperative work 71–80.

Ren, Y., Harper, F. M., Drenner, S., Terveen, L., Kiesler, S., Riedl, J., and Kraut, R. E. 2012. Building

Member Attachment in Online Communities: Applying Theories of Group Identity and

Interpersonal Bonds. Mis Quarterly 36(3)

Rochet, J.-C., and Tirole, J. 2003. Platform Competition in Two-Sided Markets. Journal of the

European Economic Association 1(4) 990–1029.

Rogers, E. M. 2010. Diffusion of Innovations, 4th Edition, Simon and Schuster.

Schwienbacher, A., and Larralde, B. 2012. Crowdfunding of small entrepreneurial ventures. The

Oxford Handbook of Enrepreneurial Finanace, D. Cumming (ed.), (Vols. 1-Book, 1-Section)

369–391.

Sundararajan, A., Provost, F., Oestreicher-Singer, G., and Aral, S. 2013. Research Commentary-

Information in Digital, Economic, and Social Networks. Information Systems Research 24(4)

883–905.

Sundin, O. 2011. Janitors of knowledge: constructing knowledge in the everyday life of Wikipedia

editors. Journal of Documentation 67(5) 840–862.

Susarla, A., Oh, J. H., & Tan, Y. 2012. Social networks and the diffusion of user-generated content:

Evidence from YouTube. Information Systems Research, 23(1), 23-41

43

Umyarov, A., Bapna, R., Ramaprasad, J., and Shmueli, G. 2013. “One-Way Mirrors and Weak-

Signaling in Online Dating: A Randomized Field Experiment,” in International Conference

on Information Systems, Milan, Italy.

Ung, H., and Dalle, J.-M. 2010. Project management in the Wikipedia community. Proceedings of the

6th International Symposium on Wikis and Open Collaboration 13.

Venkatesan, R., and Kumar, V. 2004. A Customer Lifetime Value Framework for Customer Selection

and Resource Allocation Strategy. Journal of Marketing 68(4) 106–125.

Wang, X., Butler, B. S., and Ren, Y. 2012. The Impact of Membership Overlap on Growth: An

Ecological Competition View of Online Groups. Organization Science 24(2) 414–431.

Wang, Y.-C., and Kraut, R. 2012. Twitter and the development of an audience: those who stay on

topic thrive! Proceedings of the 2012 ACM annual conference on Human Factors in

Computing Systems 1515–1518.

Ward, C., and Ramachandran, V. 2010. Crowdfunding the next hit: Microfunding online experience

goods. Workshop on Computational Social Science and the Wisdom of Crowds

Wasko, M. M., and Faraj, S. 2005. Why should I share? Examining social capital and knowledge

contribution in electronic networks of practice. MIS quarterly 35–57.

Welser, H. T., Gleave, E., Fisher, D., and Smith, M. 2007. Visualizing the signatures of social roles in

online discussion groups. Journal of social structure 8(2) 1–32.

Yang, J., and Wei, X. 2009. Seeking and Offering Expertise Across Categories: A Sustainable

Mechanism Works for Baidu Knows. ICWSM

Yoganarasimhan, H. 2012. Impact of social network structure on content propagation: A study using

YouTube data. Quantitative Marketing and Economics 10(1) 111–150.

Zhang, J., and Liu, P. 2012. Rational herding in microloan markets. Management science 58(5) 892–

912.

Zhu, H., Kraut, R., and Kittur, A. 2013. The Impact of Membership Overlap on the Survival of Online

Communities. Proceedings of the 32nd annual ACM conference on Human factors in

computing systems (pp. 281-290). ACM.

Zvilichovsky, D., Inbar Y., and Barzilay O. 2013. “Playing Both Sides of the Market: Success and

Reciprocity on Crowdfunding Platforms.” In Proceedings of the 34th International

Conference on Information Systems,