Electronic copy available at: http://ssrn.com/abstract=2234765 Electronic copy available at: http://ssrn.

com/abstract=2234765

!"#$%&'(%)(* !",-.)/, )%,-01
.2, %3(-4)!0 #& 5"#6,!. 7-!8,"0 )( 8)!80.-".,"



/9:;<= 8>??>@A<BC 7<DDC EF 7<C>@
Kevavtagter v.ive.. cboot Kevavtagter v.ive.. cboot
|virer.it, of ^ortb Carotiva |virer.it, of ^ortb Carotiva
C10; McCott 10 C10; McCott 1:21
Cbaet itt, ^C 2: Cbaet itt, ^C 2:
enkatUNC.edu Barry_BayusUNC.edu



ABS1RAC1

Lntrepreneurs are turning to crowdunding as a way to inance their creatie ideas. Crowdunding
inoles relatiely small contributions o many consumer-inestors oer a ixed time limit ,generally a ew
weeks,. In online crowdunding communities, potential donors can see the leel o support rom other
project backers as well as its timing beore making their own unding decisions, suggesting that social
information (i.e., others funding decisions) will play an important role in the ultimate success of a project.
1wo years o publicly aailable panel data on successully and unsuccessully unded projects listed on
Kic/.tarter is used to empirically study the role o social inormation in the dynamic behaior o project
backers. Building o the well-established social psychology theory around diusion o responsibility eects,
we show that additional backer support is negatiely related to its past backer support. Many potential
backers do not contribute to a project that has already receied a lot o support because they assume that
others will proide the necessary unding. Consistent with the deadline eect widely obsered in bargaining
and online auctions, we also show that the diusion o responsibility eects diminish as the project unding
cycle approaches its closing date. Moreoer, as the project deadline draws near we ind that project updates
tend to increase as the project creators make a inal plea or help to reach their unding goal. Reduced
diusion o responsibility eects, together with the positie inluence o project updates, lead to generally
increasing project support in the inal stages o unding. 1his is particularly the case or projects that
successully achiee their goals as they are more likely to hae an update in the last weeks o unding and
generate more excitement rom recent backers than projects that all short.

|Keywords: Innoation, Lntrepreneurship, Strategy, Marketing|




Acknowledgments: Comments rom participants in research workshops at the Uniersity o North Carolina
at Chapel lill and the Uniersity o Illinois helped to improe this paper. \e also thank Atul Nerkar, Page
Ouimet, and Ai Goldarb or helpul comments on an earlier drat.




March 16, 2013
Electronic copy available at: http://ssrn.com/abstract=2234765 Electronic copy available at: http://ssrn.com/abstract=2234765
2

GF )(."#%'!.)#(
An important barrier to innoation is the aailability o early-stage unding ,Cosh, et al. 2009,. Gien
the diiculties that new entures ace in attracting inancing rom angel inestors, banks and enture capital
unds, some entrepreneurs are tapping into large, online communities o consumer-inestors ,Lconomist
2010, Schwienbacher and Larralde 2012,. Called crowdfunding, this relatively new form of informal
enture inancing allows entrepreneurs to directly appeal to the general public (i.e., the crowd) for help in
getting their innoatie ideas o the ground. As deined by Bellelamme, et al. ,2012,, crowdunding inoles
an open call ,through the Internet, or the proision o inancial resources either in the orm o donation or
in exchange or some orm o reward in order to support initiaties or speciic purposes
1
.
Collecting small amounts o money rom a large number o people has a rich history in many
domains ,Ordanini, et al. 2011,. lor example, Mozart and Beethoen inanced concerts and new music
compositions with money rom interested patrons, the Statue o Liberty in New \ork was unded by small
donations rom the American and lrench people, a human rights organization is trying to raise money in
order to buy a communications satellite to proide Internet access to people in third world countries
,http:,,www.buythissatellite.com, accessed Noember 2, 2012,, and President Barak Obamas 2008 election
campaign raised most o its unds rom small donations oer the \eb ,lemer 2011,. 1oday, seeral hundred
global intermediaries with online platorms exist to match up consumer inestors with initiaties that they
wish to help und. Prominent examples in the popular press include the narratie moie project by Stee
Taylor that got almost 4,500 people to contribute nearly $350,000 and Scott Wilsons idea to create a
wristband that will conert an iPod nano into a watch raised oer >940,000 rom oer 13,500 indiiduals
,Adler 2011,. One o the largest crowdunded projects to date is Lric Migicoskys E-Paper \atch that
integrates with an Android or iPhone that receied donations totaling oer >10.2M rom well oer 65,000
backers. According to one industry report, crowdunding platorms raised almost >1.5B and successully
unded more than one million projects in 2011 ,Massolution 2012,. Gien the potential dollars inoled,
crowdunding has recently garnered attention rom policymakers and regulators as eidenced by the ]vv.tart
Ovr v.ive.. tartv. .ct ,JOBS Act, recently signed into U.S. law ,Chasan 2012,.
Crowdunding diers rom the traditional inancing o new entures in two important ways. lirst,
unding is proided by the relatiely small contributions o many indiiduals oer a ixed time limit ,generally
a ew weeks,. Second, potential donors can see the leel o support rom other project backers as well as its
timing beore making their own unding decisions, suggesting that social information (i.e., others funding
decisions, will play an important role in the ultimate success o a crowdunded project. Understanding these
eects is important because studies ind that social information can lead to non-rational behaviors. For
example, people oten choose music or downloading based on popularity not quality ,Salganik, et al. 2006,

1
1his is conceptually similar to crowdsourcing in which community members ,non-experts, propose new product and
serice ideas, as well as comment on and ote or the ideas o others ,Bayus 2013,.
3

and bidders tend to herd into online auctions with more bids een though this actiity is not a signal o higher
quality ,Simonsohn and Ariely 2008,. Many legal scholars and policy makers beliee that this kind o
irrational herding behaior increases the chances or raud in crowdunded projects since consumer
inestments are not protected by goernment regulations or oersight ,Bradord 2012, lazen 2012,. 1o date
howeer, there is ery little empirical research to deinitiely support any position.
In general, crowdfunding communities differ in terms of whether the funders primary motiation
or participating is the expectation o a inancial return. lor example, crowdunding communities like
ettaava and !efvvaer oer consumer inestors an interest in the enture in the orm o equity or some sort
o proit sharing agreement ,\ard and Ramachandran 2010, Agarwal, et al. 2011,. Other crowdunding
communities such as Pro.er and Zoa inole peer-to-peer lending in which it is expected that the original
principal is repaid, along with some ixed interest ,lerzenstein, et al. 2011, Zhang and Liu 2012,. Research
on these types o equity- and lending-based crowdunding communities inds eidence or herding behaior,
i.e., indiiduals want to contribute to projects that already hae a lot o support rom other community
members. Because consumer inestors in these communities expect a inancial return, herding behaior is a
rational way or indiiduals to reduce their own risk in the ace o uncertainty about the proposed new
entures on these websites. lollowing the literature on inormation cascades ,Bikhchandani, et al. 1992,,
these studies argue that an initiatie with a lot o community support signals that the project is o high quality.
Unlike the platorms in which participants expect some sort o inancial return, other crowdunding
communities inole no monetary compensation or participation. lor example, ]v.tCirivg and ot.v. rely on
altruistic motiations in which unders oluntarily donate their money with no expectations o any tangible
reward ,Burtch, et al. 2012, Smith, et al. 2012,. Research on these types o donation-based crowdunding
communities draw on the extensie literature inoling philanthropy and public goods ,Andreoni 2006,
Vesterlund 2006,. Because the consumption o public goods cannot be withheld rom non-contributors,
ree-riding is a potential issue in which contributions can be crowded-out by the prior unding decisions o
others ,Bergstrom, et al. 1986,. At the same time, there are seeral models based on social norms that predict
a positive effect of others funding decisions (Sugden 1984; Bernheim 1994). Depending on the perspective
taken, some donation-based crowdunding studies ind positie effects for other community members
unding decisions on contributions ,Smith, et al. 2012,, while others ind the opposite ,Burtch, et al. 2012,.
Unlike existing research that considers crowdunding communities with tangible inancial returns or
no tangible rewards at all, our interest is in crowdunding communities like Kic/.tarter and vaiegogo in which
project backers do receie tangible, but non-inancial, beneits or their inancial contributions. 1hese
rewards oten take the orm o tokens o appreciation ,thank-you message, artists autograph, mentioning the
crowdfunders name in the credits, tee-shirt, or the pre-purchasing o products or serices ,lemer 2011,.
Not surprisingly, qualitatie studies ind that rewards are one o the most important motiations or
participating in crowdunding communities ,de \itt 2012, Gerber, et al. 2012, Steinberg 2012,. Reward-
4

based crowdunding has the largest number o online platorms and is the astest growing orm o
crowdunding ,Massolution 2012,. With the exception of Mollicks (2012) cross-sectional study, ery little
research to date has considered reward-based crowdunding and none has examined the role o social
inormation.
|insert ligure 1 about here|
1he empirical setting or our study is one o the oldest and largest reward-based crowdunding
communities on the \eb. Since its launch in April 2009, Kic/.tarter has oer one million community
members who hae combined to pledge hundreds o millions o dollars to und creatie ideas in categories
like art, ilm and ideo, dance, design, and technology ,Ricker 2011,. Anecdotal cases studies suggest that
Kic/.tarter projects tend to get a lot o backer support in the irst and last weeks o their unding cycle
riends and amily tend to be early project supporters, while strangers, who make up the majority o
contributors, oten proide unding as a project nears its conclusion ,de \itt 2012, Steinberg 2012,. As an
example, consider the backer support oer time or Cody Webbs New CD (a music project supported by 89
backers that successully achieed unding o >12,01, shown in ligure 1. 1his bathtub shaped pattern o
backer support is sharply dierent than the generally increasing pattern associated with herding obsered with
equity or lending-based crowdunding ,e.g., Zhang and Liu 2012, or the decreasing pattern ound with
donation-based crowdunding ,Burtch, et al. 2012,. Although this pattern o backer support is well known by
reward-based crowdunding pundits ,de \itt 2012, Mod 2010, Steinberg 2012,, insights into its driers are
lacking.
1wo years o publicly aailable inormation on successully and unsuccessully unded Kic/.tarter
projects is used to empirically study the role o social inormation in the dynamic behaior o project backers.
1o explain the obsered bathtub shaped pattern o backer support oer time, we introduce a new theoretical
lens. Building o the well-established social psychology theory around diusion o responsibility eects
,lisher, et al. 2011,, we show that backer support or a reward-based crowdunding project is negatiely
related to its past backer supportmany potential backers do not contribute to a project that has already
receied a lot o support because they assume that others will proide the necessary unding. As a result,
initial project excitement is quickly ollowed by a strong downward trend in backer support. Consistent with
the deadline eect widely obsered in bargaining and online auctions ,Roth, et al. 1988, Ariely and Simonson
2003,, we also show that the diusion o responsibility eects diminish as the project unding cycle
approaches its closing date. Moreoer, as the project deadline draws near we ind that project updates tend to
increase as the project creators make a inal plea or help to reach their unding goal. Generally increasing
project support in the inal stages o unding comes rom reduced diusion o responsibility eects, together
with the positie inluence o project updates and inertia rom recent backers and exposure on the Kic/.tarter
web site. 1his is particularly the case or projects that successully achiee their goals as they are more likely
5

to hae an update in the last weeks o unding and generate more excitement rom recent backers than
projects that ultimately all short.
HF .2, .2,#",.)!-E &"-4,$#"8
In this section, the theoretical ramework that guides our empirical study is discussed. Our interest is
in reward-based crowdunding communities like Kic/.tarter that oer tangible, but non-inancial, beneits or
the inancial contributions o project backers. Because our primary interest is in the role o social
inormation, we deelop a ramework that has speciic predictions about the eects o past backers. 1hus, an
apparent explanation or the pattern in ligure 1 that projects simply exhaust the initial support rom riends
and amily is not helpul since there is no role or the eects o social inormation.
Gien this context, rational herding behaior due to uncertain project quality obsered in lending-
based crowdunding is unlikely or Kic/.tarter projects because there is no expectation o a inancial return.
\hile Burtch, et al. ,2012, and Smith, et al. ,2012, argue that donation-based crowdunding inoles the
proision and consumption o a public good, this is not the case with reward-based crowdunding.
Importantly, the creatie ideas posted on Kic/.tarter do not hae the properties o being non-excludable and
non-rialrous. Reciprocity ,Sugden 1984, and conormity ,Bernheim 1994, are also not expected to operate
in this enironment since donors are anonymous and speciic donation amounts are not isible. Moreoer, i
indiiduals care mostly about the end result ,i.e., proision o the public good,, then any crowding-out eects
o social inormation should not ary oer time. And, to the extent that indiiduals in the public goods
situation care only about the size o their donation and how it makes them eel, there is no role or social
inormation, i.e., the contributions of donors are unrelated in that one persons donation does not affect the
utility someone else receies rom giing ,Duncan 2004,.
Instead, we build o the well-established social psychology theory inoling the diusion o
responsibility ,lischer, et al. 2011,. Studies demonstrate that an individuals likelihood of helping decreases in
the actual or perceied presence o others ,Darley and Latane 1968, Latane and Darley 190, Garcia, et al.
2002,. Importantly, the diusion o responsibility is a robust phenomenon that occurs in many experimental
and ield situations. 1he original research program on this phenomenon was in response to the ery sad real-
lie case o Catherine Genoese who was raped and murdered in New \ork while seeral o her neighbors
looked on and did not report the attack to the police ,Latane and Nida 1981,. Literature reiews by Latane
and Nida ,1981, and lisher, et al. ,2011, show that this type o bystander eect operates in many dierse
situations, including non-emergencies ,e.g., answering the door, helping with a lat tire, leaing a tip,.
Moreoer, this eect occurs or nearly all age groups ,except or ery young children, as well as or both
genders ,Latane and Nida 1981,. A diusion o responsibility has also been obsered with donation behaior
,\iesenthal, et al. 1983,.
Latane and Darley ,190, propose a decision model that a bystander must go through beore
interening in a critical situation. lirst, bystanders need to notice the situation. Bystanders must then
6

interpret the situation as one in which action is necessary, and then deelop a eeling o personal
responsibility ,empathy,. Next, bystanders need to beliee they hae the skills and resources to help. linally,
they must decide to actually take action to help. Although presented as a linear sequence, this decision model
is meant to be iteratieat any point in this decision model, the bystander can cycle back to a preious
decision step. Bystanders can exhibit signs o discomort oer inaction i they ind it diicult to reach a
decision in any stage o this decision model. In addition, delayed responses will oten lead to inaction
altogetherthe longer bystanders wait to respond, the less likely they are to eer help. Based on anecdotal
reports, this general model seems to capture the key decisions made by backers in crowdunding communities
like Kic/.tarter ,de \itt 2012, Gerber, et al. 2012, Steinberg 2012,.
Latane and Darley ,190, identiy three dierent social psychological processes that can interere
with the completion o this decision sequence. 1he irst process is diusion o responsibility in which people
ail to help because they assume someone else will do so. In this case, the knowledge that others could
instead respond reduces their eelings o personal responsibility and thus, inhibits helping. Indiiduals tend
to subjectiely diide their own personal responsibility to help by the number o bystanders. 1his idea is
closely related to social loafing (a reduction in motivation and effort when individuals work collectively
compared with when they work individually, Karau and Williams 1993: 681). The diffusion of responsibility
predicts that the likelihood o helping is directly related to the size o the bystander group ,lorsyth, et al.
2000,. 1he second process is pluralistic ignorance ,or social inluence, in which people tend to rely on the
oert reactions o others when interpreting an ambiguous situation. In this case, indiiduals look or cues in
the enironment that can help them determine whether action is necessary. As noted by Cialdini ,2001: 100,,
we view other behavior as correct in a given situation to the degree that we see others performing it. A
strong bystander eect occurs when no one helps because eeryone beliees that no one else perceies an
emergency. 1he third process is audience inhibition ,or ealuation apprehension, in which people eel the
risk o embarrassment i the situation is misinterpreted. In other words, indiiduals are reluctant to help
because they are araid o making mistakes or acting in a way that might be negatiely ealuated by onlookers.
Gien the inherent characteristics o online crowdunding communities like Kic/.tarter ,creators, backers and
community members are or the most part anonymous, and the projects are not ambiguous in that they all
explicitly ask or inancial help,, the audience inhibition and pluralistic ignorance processes are not as releant
as the diusion o responsibility.
Lxtending the literature which ocuses on the bystander eect in ace-to-ace situations, more recent
studies ind eidence or the irtual diusion o responsibility in computer-mediated communication and
online communities. lor example, Barron and \echiam ,2002, show that the presence o others copied in a
private email communication reduces ones willingness to reply to a request for help. Markey (2000) shows
that the time it takes to receie help in online chat groups increases with group size. \echiam and Barron
,2003, ind that signiicantly more people that were emailed indiidually completed an online surey as


compared to a general request sent to members o a Listser. Voelpel, et al. ,2008, examine irtual diusion
o responsibility eects in a number o large online communities consisting o \ahoo!Groups members.
1hey show that the likelihood o responding to a help request and the quality o response is signiicantly
related to group size: small groups are more likely to respond and more likely to hae a high quality response
than larger groups. In all these studies, perceied group size is negatiely related to helping behaior.
1o date, the published literature has only considered cross-sectional ariation in group size to
demonstrate a diusion o responsibility eect. In the crowdunding context we study howeer, time-series
ariation in group size within a project is o prime interest. lelp in the orm o inancial support can come at
any point during a projects funding cycle beore it has reached its unding goal. Moreoer, perceptions about
the number o others that might proide unding will also ary oer time. lor each time period during the
unding cycle, we argue that potential Kic/.tarter backers use the list o publicly displayed backers already
supporting a project as an indicator o the size o the group that could proide the remaining unding. \hile
it is not expected that the existing backers will directly proide the necessary unding, a relatiely large
number o existing backers is suggestie that others will also eentually pledge their support. 1his approach
o using past project support to gauge uture support is consistent with related research on irtual
communities ,Voelpel, et al. 2008, as well as studies o inestment decisions in the enture capital community
,Gompers, et al. 1998,. It is also in line with recommendations on how to plan and manage a Kic/.tarter
campaign ,Mod 2010, de \itt 2012, Steinberg 2012,. Due to a diusion o responsibility, many prospectie
backers do not contribute to a project that has already receied a lot o support because they assume that
someone else will proide the remaining inancing. 1hus, we predict that project support at any time oer its
unding cycle is negatiely related to the leel o support it receied prior to that time. 1he ollowing
hypothesis summarizes these arguments.
2G1 1be ti/etibooa a reraraba.ea crorafvvaivg ro;ect receire. aaaitiovat bac/er .vort i. vegatiret, retatea to it.
a.t bac/er .vort.

According to Kic/.tarter statistics, a lot of backer support comes in the later stages of a projects
unding cycle. Matt laughey, a backer o more than 150 Kic/.tarter projects, sums it up this way ,Steinberg
2012: 149,:
once you a.. :0 ercevt of ,ovr fvvaivg, at av, oivt, ,ov bare a : ercevt cbavce of
reaching your goal. Theres a human psychology element where people go, yeah Itt /ic/ iv vore,
tbi. gv, i. .o cto.e. Ovt, a bavafvt of ro;ect. bare fivi.bea vv.vcce..fvtt, barivg reacbea :
percent or more of their funding. The people who are at like 60, 70 percent with a week to go, its
govva be OK!

1his kind o deadline eect in which a lot o action occurs as the end o an experience is approached
has been widely obsered in many contexts. lor example, last minute agreements are common in
negotiations ,Roth, et al. 1988, Ma and Manoe 1993, Zhou 2011, and a large number o bids are made near
the end o online auctions ,Ariely and Simonson 2003, Ockenels, et al. 2006,. \ebb and \eick ,199, cite
8

seeral unpublished papers that report deadline eects in college applications ,more applications are receied
right beore deadline dates,, trading on the New \ork Stock Lxchange ,trading olume systematically
increases two hours beore the closing bell,, and play calling in the National lootball League ,total plays
executed are highest in the second quarter right beore the hal time break and ourth quarter right beore the
end o the game,. Similar behaiors hae also been obsered in rats and pigeons that increase their eorts as
the expected end o a ixed reinorcement schedule approaches, een though this behaior does not increase
rewards ,lerster and Skinner 195,. As noted by Ariely and Simonson ,2003,, unlike the earlier stages,
decisions near a deadline are clearly consequential and oten irreersible.
In terms o the bystander decision model proposed by Latane and Darley ,190,, we argue that
perceptions about the number o others that might proide unding are also inluenced by the deadline eect.
lor a recently launched project, there is a lot o time or others to make contributions. Due to a diusion o
responsibility, potential backers eel less personal responsibility or a project in the early stages o its unding
cycle and thus are less likely to contribute. lor an ununded project about to end howeer there is a ery real
possibility that the necessary unding to reach its goal will not come rom others ,een though the project
may hae a high leel o past backer support,. In this case, potential backers hae lower expectations o how
much support a project will ultimately receie rom others, and consequently they are more likely to
contribute. 1hus, we expect that the diusion o responsibility eect is reduced when a deadline is present,
i.e., when a deadline to act is looming, people are more likely to help een when others are present. 1o
capture this idea, we propose that the diusion o responsibility eect in 2G is positiely moderated by time
in the unding cycle, i.e., it becomes less negatie in the later stages o the unding cycle. 1his discussion is
summarized in the ollowing hypothesis.

2H1 1be effect. of a.t bac/er .vort for a reraraba.ea crorafvvaivg ro;ect are o.itiret, voaeratea b, tive iv tbe
projects funding cycle so that tbe effect of a.t bac/er .vort i. te.. vegatire for ro;ect. iv tbe tater .tage. of
fvvaivg tbav tbo.e iv tbe eartier .tage. of fvvaivg.
IF %-.-
In this section, we briely discuss the empirical setting o our study. Based in the U.S., Kic/.tarter is
one o the worlds largest crowdfunding platforms. By April 2012, Kic/.tarter had raised more than >200
million or 20,000 projects, or about 44 percent o those that sought inancing on the site ,\ortham 2012,.
According to their website, Kic/.tarter is ocused on creatie projects. \e're a great way or artists,
ilmmakers, musicians, designers, writers, illustrators, explorers, curators, perormers, and others to bring
their projects, eents, and dreams to life. Projects are grouped into thirteen broad categories: Art, Comics,
Dance, Design, lashion, lilm and Video, lood, Games, Music, Photography, Publishing, 1echnology, and
1heater. 1he website deines a project as something finite with a clear beginning and end. Someone can be
held accountable to the ramework o a projecta project was either completed or it wasntand there are
9

definable expectations that everyone can agree to. Consequently, Kic/.tarter does not accept projects created
to solicit donations to causes, charity projects, or general business expenses.
In order to participate, indiiduals must join the Kic/.tarter community ,at no cost, by selecting an
anonymous username. Like most online communities, inormation on demographics and personal
characteristics are not collected ,the Kic/.tarter community is a large, undefined crowd). Community
members can propose projects or unding, back a project by inancially contributing ,with a credit card ia
Amazon,, and,or comment on projects. Kic/.tarter projects can only be proposed by U.S. residents ,or tax
purposes,, project contributors hae no geographic restrictions.
1o use Kic/.tarter, an entrepreneur (called creator in Kic/.tarter, creates a webpage or the project
on the platorm explaining the purpose o the project and the speciic delierables that they aim to produce
with the contributed unds. Along with an end date or the project unding cycle, the creator also indicates
the unding goal o the project, i.e., the amount o money they require to execute the project as speciied.
Creators can communicate with their supporters by posting priate updates as well as potential contributors
by posting public updates that eeryone can see.
\hen a potential donor (called backer in Kic/.tarter, isits an active projects webpage, they are
presented with all the project inormation initially posted by the creator. Moreoer, potential backers are
shown the current unding status o the project ,i.e., the unds raised thus ar,, the ultimate unding goal, and
the number o days remaining until the project unding cycle expires. A isitor can also see a listing o the
other backers that hae contributed to the project, as well as the timing o these contributions
2
. Speciic
donation amounts by backers are not publicly shown on the website. 1o help potential backers discoer
projects they want to support, Kic/.tarter has a number o search options. In particular, projects can be sorted
based on the irst week ater their initial launch (Recently Launched,, last week beore the project unding
closes (Ending Soon), or popularity ,based on the number o backers recently added to a project,.
Occasionally, Kic/.tarter sta mention an actie project on their blog.
1here are two important eatures o Kic/.tarter that distinguish it rom many other smaller
crowdunding platorms, as well as more traditional orms o entrepreneurial finance. The first is the all-or-
nothing aspect o undraising on the platorm. A project must be ully unded beore its unding cycle
concludes or no money pledged by any backer is transerred to the project creator. An oer-ambitious
unding goal may thus result in the undraising eort alling short and consequently, raising no unds
whatsoeer. At the same time, once a project has reached its unding goal, it can continue to receie
contributions until its deadline. As a result, unded projects can exceed their original unding goal. \e expect
that the diusion o responsibility and deadline eects described in 2G and 2H operate beore a project has

2
Shortly ater our data collection in March 2012, Kic/.tarter remoed this inormation in their updated website design.
10

reached its unding goalbacker motiation and behaior during the post-unded phase o a project may be
quite dierent since the project no longer needs inancial help.
1he second eature o the crowdunding model that diers rom traditional enture capital is the act
that indiiduals contributing to a project do not receie equity in the project in return or their unds.
Speciically, backers do not receie any inancial incenties, returns, or repayment in exchange or their
contributions. Instead, project creators typically oer more modest rewards to contributors which ary by
the leel o contribution. According to the Kic/.tarter website, the our most common reward types are: ,a,
copies o the thing ,e.g., the actual product, an assembled ersion o a DI\ kit,, ,b, creatie collaborations o
arious kinds ,e.g., a backer might appear as a hero in the comic, or she may be painted into the mural,, ,c,
creatie experiences ,e.g., a isit to the ilm set, a phone call rom the author, dinner with the cast, a concert
in the backers backyard); and (d, creatie mementos ,e.g., photos sent rom ilming location, explicit thanks
in the closing credits o the moie, etc.,.
Data or our study come rom publicly aailable inormation on the Kic/.tarter web site. \e extracted
inormation on all backed projects posted on the platorm rom its inception in May 2009 through lebruary
2012. \e ocus on projects with at least one backer since we are interested in the dynamics o backer
behaior ,projects with zero backers will not contribute any inormation to our analyses,. 1o allow a time
buer or community actiity around a project to stabilize, projects completed ater 2011 are dropped rom
the analysis. In addition, projects started in 2009 are not used in the analysis because the look and eel o the
web site underwent seeral reisions in the irst ew months ater launch. 1hus, two years o daily data on
25,058 projects with complete
3
inormation that began on or ater January 1, 2010 and concluded by
December 31, 2011 is aailable or analysis purposes.
|insert 1able 1 about here|
Descriptie statistics or these projects are reported in 1able 1. 1he aerage project
4
has a goal o
just oer >8,500 but only receies a little more than >4,600 in pledged contributions
5
. Projects tend to last or
almost six weeks, a relatiely large proportion o backers support a project in the irst or last week o its
unding cycle. Oer eighty percent o the projects include a ideo. 1he aerage project oers more than six
reward categories as incenties or their donors and receies almost >5 per backer. Creators generally post a
couple o public project updates and inrequently post priate updates to their backers. Oer ninety percent
o creators only propose a single Kic/.tarter project.
|insert ligure 2 about here|

3
lorty projects had incomplete or suspect inormation and were dropped rom our sample.
4
One project had a goal o oer >21M (Kickstarters limit) to help reduce the national debt
,http:,,www.kickstarter.com,projects,2116548608,help-erase-the-national-debt-o-the-usare~search,. 1his project
only had 8 backers who pledged >180.
5
1he largest unded project in our sample receied a little oer >942K. Since our data collection, seeral projects hae
receied oer >1M in unding.
11

1here is a considerable amount o ariance in the unding outcomes or Kic/.tarter projects. ligure 2
shows the distribution o project success: projects that reach their unding goal do so by a small margin
,almost hal o all the successul projects are within ten percent o their original unding goal,, while projects
that miss their targets do so by a large margin ,almost hal o all the unsuccessul projects achieed less than
ten percent o their goal,. Consistent with the earlier quote rom Matt laughey, there are ery ew
unsuccessully unded projects that reach between ity and ninety-nine percent o their goal.
Descriptie statistics by project unding outcome are reported in 1able 2. Although these statistics
cannot be used to assess causality, this inormation is useul to better understand the nature o a reward-based
crowdunding community. \hile unsuccessul projects hae a unding goal more than three times as large as
successul projects ,>14,580 compared to >4,26,, these projects receie well less than hal o the amount
contributed to successul projects ,>1,232 compared to >4,43,. Successul projects tend to be shorter in
duration. All projects receie a relatiely large proportion o their backers in the irst weekmultiplying the
total number o backers in the irst week by the aerage contribution per backer, and diiding by the goal
indicates that successul projects aerage almost thirty percent o their goal in the irst week ,as compared to
unsuccessul projects that achiee less than one percent,. Successul projects also get a lot o support in the
last week o their unding cycle. Not surprisingly, successully unded projects hae signiicantly more
backers than unsuccessul projects, and add more backers each day. Successul projects are generally more
likely to hae a ideo and a larger number o reward categories than unsuccessul projects. Successul
projects tend to be eatured more oten ,on the Kic/.tarter blog or Most Popular List, and communicate more
to the community and their backers by posting updates.
|insert 1able 2 and ligure 3 about here|
1he aerage number o backers added to a project oer its relatie unding cycle is depicted in ligure
3. Consistent with ligure 1 and the descriptie statistics in 1ables 1 and 2, projects tend to get a lot o backer
support in the irst and last weeks o their unding cycle. A high leel o initial project support in the irst ew
days is ollowed by generally decreasing support oer most o the unding cycle. As the project approaches
the end o its unding cycle, successul projects are likely to hae a sharp increase in backer support. 1o
better understand the dynamics o backer unding behaiors and the role o social inormation, we next turn
to an econometric analysis o these data.
JF .2, ,45)")!-E 0.'%3
In this section, we exploit the panel structure o the Kic/.tarter data to inestigate the relationship
between the additional backer support a project receies and its past backer support ,2G,, as well as the
moderation o this relationship by time in the project unding cycle ,2H,. Because 2G and 2H concern the
likelihood a project receies additional backer support, an appropriate dependent measure to test these
hypotheses is ac/er.aaeait, a binary ariable where a alue o one indicates that project i receied a new
12

contribution pledge on day t ,0 otherwise,. 1his measure is similar to that used by Simonshon and Ariely
,2008, to study online auctions and lerzenstein, et al. ,2011, in lending-based crowdunding.
1o examine 2G and 2H, we start with the ollowing basic model:
ac/er.aaeait ~ 1Pa.tac/ervortit - 2ir.t!ee/it - a.t!ee/it - 1Pa.tac/ervortit ir.t!ee/it
- :Pa.tac/ervortit a.t!ee/it - `it - Zi - it ,1,
lere, t~2,,Ti where 1i is the duration o project i and it is a random error term with mean zero and
constant ariance. 1he control ariables that ary oer projects and time are captured in `it, the ariables
that describe projects but do not ary oer time are in Zi. Cluster-robust standard errors or the estimated
coeicients are used or statistical tests due to dependence among the errors oer time within a project
,co,it, it)0). Later we will discuss our estimation approach. 1able 3 summarizes our ariables, their
deinitions, and reports descriptie statistics.
|insert 1able 3 about here|
Because projects hae dierent goals, the total number o backers required to reach a unding goal
will dier across projects. 1hus, to better acilitate comparisons across projects o dierent sizes,
Pa.tac/ervort is a relatie measure deined to be the ratio o the cumulatie number o backers supporting
project i beore day t to the total number o backers required to reach project is goal. \e estimate the
number o required backers as the goal diided by the aerage pledged contribution ,calculated as the ratio o
total unds pledged to the total number o backers obtained by the end o the project,. lere, total required
backers are the same as total backers actually obtained i a project exactly achiees its unding goal. Because
Pa.tac/ervort is highly skewed ,e.g., 1able 3,, its log transorm is used in the estimations. Notably, our
conclusions to be discussed are unchanged i we use the actual number o backers supporting a project
instead o this relatie measure. lrom equation ,1,, 2G is conirmed i 1 is negatie and signiicant.
Our second hypothesis predicts that the eect o past backer support will be positiely moderated by
project funding phase, and it will be weakest near the projects deadline. To test this hypothesis, we construct
multiplicatie interaction terms inoling past backer support and time in the project unding cycle. \e
diide the unding cycle into three time periods, and deine ir.t!ee/ to be one i a project is in the irst week
o its unding ,0 otherwise, and a.t!ee/ to be one i a project is in the last week beore its deadline ,0
otherwise,. lere, the reerence category is the projects middle time period during its funding cycle. Using
discrete categories allows or the possibility o non-monotonic eects during the unding cycle. Moreoer,
these weekly variables closely match the Recently Launched and Ending Soon search options already
aailable on the Kic/.tarter web site. \e note that our general results are robust to other discrete time periods
,i.e., separate daily binary ariables or irst and last seen days, separate weekly binary ariables or irst and
last two weeks, quintiles, deciles,. lrom equation ,1,, 2H is supported i the estimated coeicient o the
interaction term : inoling Pa.tac/ervort and a.t!ee/ is positie.
13

Seeral time-arying ariables that control or possible eects due to other project or situational
actors are included in the analysis. As suggested by 1able 2, seeral projects in our sample exceed their
original unding goal. \hile we expect the negatie diusion o responsibility eect to be present with these
projects, the deadline eect is complicated by the act that all these projects achiee their goal beore the
unding cycle ends. Len though a project may be approaching its end date, all o the unding uncertainty
has been resoled since it has already met its goal. 1o consider this eect, we include Po.tvvaea, deined to
be one or each day a project has already been unded ,i.e., Pa.tac/ervort1,, and zero otherwise.
Communication eects between project creators and backers are incorporated into our analysis by including
inormation on project updates. 1here are two types o updates: ,a, backer-only updates, which are only
isible to those indiiduals who hae preiously contributed to the project and ,b, public updates, which are
visible to anyone visiting the projects webpage. \e deine Prirate|aate to be one i project i has an update
posted only to backers during day t ,0 otherwise, and Pvbtic|aate to be one i project i has posted an update
isible to eeryone during day t ,0 otherwise,. \e expect updates to hae a positie eect on the likelihood a
project adds a new backer. In addition, we include agac/er. ,deined to be the number o new backers
contributing to project i in day t1, to control or immediate word-o-mouth eects ,Duan, et al. 2009, and
general inertia in contribution behaior ,Burtch, et al. 2012,. Because agac/er. is highly skewed, its log
transorm is used in the estimations. Gien that potential backers can sort projects on the Kic/.tarter website
based on popularity ,in addition to recently launched and ending soon,, we include Povtar in our model.
lere, Povtar is deined to be one i a project is ranked in the top ity o all actie projects in terms o
backers added oer the prior week ,0 otherwise,. \e also include togPo.t ,deined to be one i
www.kickstarter.com,blog made speciic mention o project i on day t, 0 otherwise, to capture any positie
eects related to a mention on the Kic/.tarter blog. linally, we control or the possibility that pledges
concentrate on certain days by including separate dummy ariables or day o week and account or any other
unobsered time-arying eects by including month-year dummy ariables.
\e also include seeral time inariant control ariables related to project characteristics. lollowing
Mollick ,2012,, we control or dierences in project is Coat ,dollars, and Dvratiov ,length o the unding cycle
in days,. lere, the log transorm o Coat is used in our estimations due to its highly skewed alues and a
squared term is included to capture any non-linear eects ,Mollick 2012,. Because a high proportion o
successul projects hae a ideo ,de \itt 2012, Mollick 2012, Steinberg 2012,, we control or whether the
creator o project i included a ideo ,1iaeo~1 i yes, 0 otherwise,. In line with the experiences rom
successul Kic/.tarter projects ,de \itt 2012, Mod 2010, Steinberg 2012,, we include controls or the number
o reward categories ,ReraraCategorie., and whether the project creator had preiously proposed any other
Kic/.tarter projects ,Creatorerievce~1 i yes, 0 otherwise,. Among the hundreds o actie projects that are
aailable or unding, one o the most important screening mechanisms used by potential backers is the
project title ,de \itt 2012, Steinberg 2012,. Research inoling scientiic articles has shown that title length
14

and punctuation are related to subsequent downloads and citations ,Jacques and Sebire 2010, Jamali and
Nikzad 2011,. 1o account or any o these eects, we also include 1itteevgtb ,number o words in the title o
project i, and Cotov ,~1 i the title o project i has a colon, 0 otherwise, in our estimations. linally, we include
dummy ariables or the sel-reported category o project i ,see 1able 3,.
1hough our primary dependent measure is binary, we ollow the existing literature and estimate
equation ,1, as a linear probability model ,\ooldridge 2010, Goldarb and 1ucker 2011, Simcoe and
\aguespack 2011,. \e do this or two reasons. lirst, the estimated coeicients can easily be interpreted as a
change in probabilities. Second, and more important, interpretation o interaction terms in nonlinear models
is not straightorward because they are a cross-deriatie o the expected alue o the dependent ariable. As
Ai and Norton ,2003: 123, note, the magnitude of the interaction effect in nonlinear models does not equal
the marginal eect o the interaction term, can be o opposite sign, and its statistical signiicance is not
calculated by standard software. In line with \ooldridge ,2010,, we ind that there is little qualitatie
dierence in our results or the linear probability and logit speciications ,see appendix,. Moreoer, most o
our explanatory ariables are binary and the predicted probabilities or our model generally lie between zero
and one
6
, suggesting that there is little bias rom the bounded dependent ariable in our estimations
,\ooldridge 2010,.
JFG 5D9KLBL:<DC -:<KC@L@ MN 7<O;9D 79P<QLMD
\e begin with a nae pooled approach in which equation ,1, without the interaction terms is
estimated by ordinary least squares ,OLS, regression. 1he results are reported in 1able 4, Model 1. \hile we
do not report the estimation details or day o week, month-year, and category ixed eects to consere
space, we can make a ew obserations. lirst, projects are more likely to receie contributions on weekdays
compared to weekends, with Monday representing the peak and actiity decreasing steadily until lriday.
Second, we ind that the categories o Dance, Design, and lood hae a highest likelihood o contributions
while lashion, Publishing, and lilm & Video projects hae the lowest likelihood o actiity. No signiicant
trends are obsered with month-year ixed eects. Consistent with Mollicks (2012) cross-sectional analysis o
successul projects, we ind that the likelihood o an additional backer supporting a Kic/.tarter project is
signiicantly related to Coat, Dvratiov and 1iaeo. Speciically, projects with smaller goals, o shorter duration,
and haing a ideo are likely to garner additional backer support. In addition, projects with many reward
categories and haing a colon in their title tend to attract new backer pledges , as do projects rom creators
with prior experience. Not surprisingly, the negatie and signiicant coeicient estimates or Po.tvvaea imply
that backers are less likely to support a project ater it reaches its unding goal. In addition, eatured projects
,togPo.t, Povtar, ir.t!ee/, a.t!ee/,, as well as those with communications rom creators ,Prirate|aate,

6
Lliminating the ery ew obserations that were outside this range produced essentially the same estimates as reported
in our tables.
15

Pvbtic|aate,, are more likely to receie additional backer contributions. \ord-o-mouth and inertia eects
,agac/er..aaea, are also positiely related to the likelihood a project receies new backer support.
|insert 1able 4 about here|
More important, the results in 1able 4, Model 2 are not consistent with 2G: the coeicient estimate
or Pa.tac/ervort is positie and signiicant. \e note howeer that equation ,1, does not control or
possible unobsered heterogeneity across projects. In particular, unobsered dierences in project quality
not accounted or with the included control ariables are likely to be present and, importantly, related to the
past backer support a project receies. lor example, high quality projects are expected to receie backer
support oer their unding cycles ,and thus hae a high leel o past backer support, and are likely to hae
additional backer pledges in each time period. 1hus, the positie coeicient estimate or Pa.tac/ervort
may simply relect that high quality projects typically hae more backers than low quality projects. In this
case, the estimated coeicients or Pa.tac/ervort in 1able 4, Model 1 are biased due to an omitted
ariable. \e next turn to a model speciication that accounts or unobsered heterogeneity across projects.
JFH -OOM>:=L:R NMD ':MS@9DQ9T 5DMU9O= 29=9DMR9:9L=C
\e extend our basic model ,1, by decomposing the error term into two components:
ac/er.aaeait ~ 1Pa.tac/ervortit - 2ir.t!ee/it - a.t!ee/it - 1Pa.tac/ervortit ir.t!ee/it
- :Pa.tac/ervortit a.t!ee/it - `it - Zi - i - it ,2,
lere, i represents all dierences between projects ,such as quality, that are stable oer time and not
accounted or by Zi. Based on a lausman type test ,see Allison 2005,, ixed-eects models are preerred
oer random eects models or the Kic/.tarter data. lere, the unobsered project-leel dierences i are
permitted to be correlated with the independent ariables. Importantly, a ixed-eects model remoes any
unobsered, time-inariant heterogeneity across projects since the i are dierenced out o equation ,2,.
Although the time-inariant characteristics are controlled, a ixed-eects estimation approach will not
produce estimates or these ariables.
\e irst consider 2G. 1he results rom an OLS ixed-eects regression estimation o
equation ,2, without the interaction terms are reported in 1able 4, Model 2. \e now ind strong support or
2G ,the coeicient estimate or Pa.tac/ervort is negatie and signiicant,. As expected, the positie
coeicient estimates or ir.t!ee/ and a.t!ee/ ,which are signiicantly larger than the middle period which
is the reerence category, capture the general bathtub shaped pattern o backer support oer the unding cycle
,ligure 3,. 1hese results clearly indicate the need to account or unobsered project heterogeneity in order to
understand the role o social inormation in the dynamic behaior o backers in crowdunding communities.
Ater accounting or unobsered project heterogeneity, we ind strong eidence or a diusion o
responsibility among Kic/.tarter backers. Assuming that others will proide the necessary unding, many
potential backers do not contribute to a project that has already receied a lot o support. Lerything else
16

being equal, there is a strong negatie eect o past backer pledges on additional project support. Clearly,
this result is not consistent with the simple idea that projects exhaust the initial support rom riends and
amilyan explanation that suggests no role or social inormation.
1o urther explore this result, seeral robustness analyses are also reported in 1able 4. Because
backer dynamics or projects with low goals may be dierent than projects with higher goals ,e.g., a small
group o riends and amily may be able to completely und projects with low goals,, the eects o Coat are
consider in Model 3. Not surprisingly, the signiicant negatie coeicient estimate or the interaction term
inoling Pa.tac/ervort and Coat indicates that the diusion o responsibility is een stronger or projects
with larger goals. Although relatiely ew Kic/.tarter creators link their projects to a lacebook account, we
expect that the diusion o responsibility among the projects that do will be stronger or those showing many
lacebook riends ,potential backers,. 1he negatie and signiicant coeicient estimate or the interaction
term inoling Pa.tac/ervort and rieva. ,deined to the number o lacebook riends reported on the
project page, in Model 4 conirms this idea. linally, we demonstrate that a strong diusion o responsibility
exists or dierent backer types in Models 5, 6, and . lere, immediate backers are indiiduals who join
Kic/.tarter and back a single project in the same day, delayed backers are indiiduals who join Kic/.tarter and
back a single project in some other day, and serial backers are indiiduals who hae backed more than one
Kic/.tarter project. \e expect that riends and amily members are part o the immediate backer group. 1he
signiicant negatie coeicient estimates or Pa.tac/ervort or each o these backer types proides
eidence or 2G irrespectie o backer experience. 1hus, all backer types seem to be ulnerable to the
negatie eects associated with a diusion o responsibility.
|insert 1able 5 about here|
1he results rom an OLS ixed-eects regression estimation o the complete model in equation ,2,
are reported in 1able 5, Model 1. 1he signiicant positie coeicient estimate or interaction o
Pa.tac/ervort and a.t!ee/ proides strong support or 2H. As expected, Model 2 demonstrates that a
diusion o responsibility continues to hold ater a project is unded ,the sum o the coeicient estimates or
Pa.tac/ervort and Pa.tac/ervort Po.tvvaea is signiicantly dierent than zero, l~.14, p0.01,, but
there is no deadline eect once a project reaches its goal ,the sum o the coeicient estimates or
Pa.tac/ervort a.t!ee/ and Pa.tac/ervort a.t!ee/ Po.tvvaea is negatie and signiicantly
dierent than zero, l~5.86, p0.05,. \e urther show that Models 3, 4, and 5 conirm 2H or the dierent
backer types ,immediate, delayed, and serial backers,. 1hese results are all consistent with a strong deadline
eect as the project unding cycle approaches its closing dateas the project deadline looms, people are
more willing to help by proiding unding een when others are present.
1aken together, the results in 1ables 4 and 5 lead to a better understanding o the bathtub shaped
pattern o backer support oer time depicted in ligure 3. Initial project excitement rom being tabbed as
Recently Launched ,ir.t!ee/, is quickly ollowed by a strong downward trend in backer support due to a
1

diusion o responsibility ;Pa.tac/ervort,. As a project approaches its end date, a positie deadline eect
begins to reduce the negatie diusion o responsibility eects. 1he positie exposure rom being tabbed as
Ending Soon (a.t!ee/,, project support near the end o the unding cycle can come rom public and
priate updates, as well as rom immediate word-o-mouth and inertia eects related to recent backers
,agac/er..aaea,. 1his is especially the case or successully unded projectsas shown in 1able 2,
successul projects hae signiicantly more public and priate updates than unsuccessul projects, and add
signiicantly more backers each day ,and during the last week,. \e can urther demonstrate that successully
unded projects are more likely than unsuccessul projects to hae a public update in the later stages o its
unding cycle.
|insert 1able 6 about here|
1o do this, we estimate the ollowing model separately or public and priate updates:
;Pvbtic or Prirate)|aateit ~ 1ir.t!ee/it - 2a.t!ee/it - vvaeai ir.t!ee/it - 1vvaeai a.t!ee/it
- `it - Zi - i - it ,3,
lere, a time-inariant indicator or whether project i was eentually unded is included ,vvaeai~1 i yes, 0
otherwise,. Because equations ,2, and ,3, are recursie, independently estimating equation ,3, gies unbiased
coeicient estimates but is less eicient ,coeicient t-statistics are conseratie,. 1he results rom an OLS
ixed-eects regression estimation o equation ,3, are reported in 1able 6. 1he signiicant and positie
coeicient estimates or a.t!ee/ vvaea indicate that successul projects are more likely to hae a public
and,or priate update in the later stages o the unding cycle than unsuccessul projects. 1hus, we ind
strong eidence that the creators o successul projects use both public and priate updates as the end o their
unding cycle approaches to encourage potential backers to make a contribution.
VF %)0!'00)#( -(% )45E)!-.)#(0
Gien the diiculties that new entures ace in attracting inancing rom traditional sources, many
entrepreneurs are directly appealing to the general public or help through large, online communities o
consumer-inestors. 1hese crowdunding campaigns inole relatiely small contributions o many
indiiduals oer a ixed time limit ,generally a ew weeks,. More important, potential donors can see the leel
o support rom other project backers as well as its timing beore making their own unding decisions,
suggesting that social information (i.e., others funding decisions) will play an important role in the ultimate
success o a crowdunded project. Understanding these eects is important because studies ind that social
inormation can lead to irrational herding behaiors.
1he empirical setting or our study is Kic/.tarter, one o the oldest and largest reward-based
crowdunding communities on the \eb. 1wo years o publicly aailable inormation on successully and
unsuccessully unded projects is used to empirically study the role o social inormation in the dynamic
behaior o project backers. In sharp contrast to the herding patterns ound in equity and lending-based
crowdunding studies ,Agarwal, et al. 2011, Zhang and Liu 2012, and crowding-out ound in donation-based
18

crowdunding ,Burtch, et al. 2012,, backer behaior in reward-based crowdunding is bathtub shaped ,see
ligure 3,. Kic/.tarter projects typically get a lot o backer support in the irst and last weeks o their unding
cycle. A high leel o initial project interest in the irst ew days is quickly ollowed by decreasing support
oer most o the unding cycle. As the project nears the end o its unding cycle, successul projects are able
to generate renewed excitement which results in a rise in backer support.
\e propose a new theoretical ramework in which to better understand this obsered backer
behaior and then empirically test it with a panel data set inoling daily inormation on oer 25,000
successul and unsuccessul Kic/.tarter projects in the 2010-2011 time period. 1aken together, our analyses
conirm many o the same time-inariant project characteristics that are associated with successul
crowdunded projects suggested by others ,de \itt 2012, Mollick 2012, Steinberg 2012,. \e extend these
preious indings by exploiting the panel structure o our Kic/.tarter data to consider the dynamics o backer
unding behaiors and the role o social inormation. Ater controlling or all these time-ariant project
characteristics, as well as any unobsered project dierences in quality, we ind strong eidence or our
hypotheses. Building o the well-established social psychology theory around the diusion o responsibility
eects ,lisher, et al. 2011,, we ind that backer support is negatiely related to its past backer support. Due
to a diusion o responsibility, many potential backers do not contribute to a project that has already receied
a lot o support because they assume that others will proide the necessary unding. Consistent with the
deadline eect widely obsered in bargaining and online auctions ,Roth, et al. 1988, Ariely and Simonson
2003,, we also ind that the diusion o responsibility eects diminish as the project unding cycle
approaches its closing date.
Given the strong diffusion of responsibility effects we observe in the early stages of a projects
unding cycle, it is interesting to speculate on the general importance o a speciic project goal and deadline
,i.e., a proision point mechanism, see Bagnoli and Lipman 1989,. Our indings proide one possible
explanation for the pervasive use of all-or-nothing funding mechanismswithout an explicit deadline, the
negatie diusion o responsibility eects inherent in crowdunding may dominate and many ,i not most,
projects would permanently lose momentum ater a short period o time. luture research might consider a
thorough comparison o crowdunding platorms and their unding mechanisms ,e.g., vaigogo allows both
ixed unding with reunds i the project goals are not met as well as lexible unding in which the project
creator receies all the pledges,.
VFG )B?KLO<=LM:@ NMD 5DMU9O= !D9<=MD@
\ith ery ew notable exceptions

, the ast majority o creatie ideas on Kic/.tarter inole relatiely

modest amounts o money ,rom 1able 1, the aerage project has a goal o >8,500 and rom 1able 2,
successul projects on aerage receie less than >,000,. Almost by deinition, crowdunding inoles small

As o December 29, 2012 only seenteen Kic/.tarter projects had raised oer >1M ,Shontell and Dickey 2012,.
19

contributions rom many people ,rom 1able 1, the aerage contribution is >5 and rom 1able 2, no more
than 100 backers support the aerage project). As emphasized in the how to crowdfund books (e.g., de
\itt 2012, Steinberg 2012,, setting appropriate unding goals is paramount to haing a successul project.
Many project ailures set unreasonable unding targets gien the scope o their creatie idea ,rom 1able 2,
the aerage goal or unsuccessul projects is almost three times the goal or successul projects,. Our
statistical analyses conirm that projects with smaller goals, o shorter duration, and haing a ideo are likely
to garner additional backer support, as are projects with many reward categories and haing a colon in the
title. Not surprisingly, projects eatured on Kic/.tarter (blog, Most Popular, Recently Launched, Ending
Soon) also are more likely to receive additional backer contributions. In general, potential backers are much
less interested in supporting a project ater it reaches its unding goal.
Very ew Kic/.tarter projects achiee at least ity percent o their goal and are not eentually unded
,ligure 2,. 1his suggests that the Kic/.tarter community is generally sympathetic to the creators plea or
helpKic/.tarter reports an oerall success rate o almost 45. At the same time, our econometric results
highlight the diicult challenge acing project creators. Inherent to the Kic/.tarter crowdunding model is a
strong diusion o responsibility eect, i.e., additional backer contributions are negatiely related to the
backer support a project has already receied ,1able 4,. 1hus, the initial excitement around a new project is
quickly ollowed by a sharp drop in support as potential backers assume that others will proide the requested
unding. Project creators need to oercome this diusion o responsibility eect in order to reach their
unding target. While it is important to attract contributors throughout a projects funding cycle, backer
support oer time is typically bathtub shaped ,ligure 3,. 1his means that projects are usually not that actie
in the middle period o their unding cycle but become much more dynamic in the later stages. Somewhat
osetting the diusion o responsibility eect, we do ind eidence or a deadline eect, i.e., the diusion o
responsibility eect is reduced as a project nears the end o its unding cycle ,1able 4,. \hile it is true that
successul projects are more likely to receie additional backer support rom Kic/.tarter related actions ,e.g.,
blog post, being categorized as Most Popular or Ending Soon), these positive effects are not directly
under the control o a project creator. Instead, project creators need to be proactie by posting public and
priate updates. \e ind that successully unded projects are likely to hae a public and,or priate update as
the end o their unding cycle approaches ,1able 6,.
1his last point about project updates is particularly important since it represents one o the driers
that can be directly inluenced by a project creator. Clearly, our analysis o project updates is ery basic.
1hus, additional research on how product creators might eectiely use project updates to oercome any
diusion o responsibility eects is warranted. lurther, data limitations preent us rom considering the
potential role o other social and online media in generating buzz around a crowdunding project. Seeral
guidebooks stress the importance o crating an online marketing campaign and its importance in generating
excitement and backer support throughout the unding cycle ,de \itt 2012, Steinberg 2012,. At this point
20

howeer, only anecdotal case studies o success stories and their marketing eorts exist, suggesting that
urther research would be helpul in careully sorting out the arious possible eects.
VFH )B?KLO<=LM:@ NMD 5MKLOC 4<;9D@
\ith the passing o the JOBS Act by the US Congress in 2012, policy makers are persuaded by the
potential o crowdunding to help und small businesses and entrepreneurs with creatie ideas. At the same
time howeer, many pundits hae sounded alarms at the potential or raud associated with unregulated
inesting behaior, especially by consumer-investors (the crowd). Of particular concern is that project
creators with unreasonable ideas will still ind unding rom nae inestors due to herding behaior ,i.e.,
consumers will simply support projects that others hae already supported,. In contrast to this perspectie,
we ind no eidence or irrational herding in the Kic/.tarter community. Instead, we demonstrate that a strong
diusion o responsibility eect exists, i.e., others previous funding decisions are vegatiret, related to the
likelihood an additional backer will support a project ,1able 4,. \hile this diusion o responsibility eect
diminishes oer the unding cycle, the oerall eect associated with past backers remains negatie. Potential
backers in this reward-based crowdunding community seem to be cautious in pledging to ongoing projects,
preerring to let others proide the requested unding. lurther research is needed to determine i our
indings generalize to equity-based crowdunding communities.
Another possible ear is that project creators propose projects, receie the unding rom consumer-
inestors, and then neer complete the project or delier the promised rewards. Interestingly, Mollick ,2012,
inds that that most Kic/.tarter project creators in the 1echnology and Design categories make serious eorts
to ulill their obligations, but the majority delier the promised rewards later than expected. Although
explicitly disclaimed by Kic/.tarter, many consumer-inestors beliee that the web site is essentially an online
retail storeront in which project creators are pre-selling products. lor example, consider the case o Neil
Singh ,Markowitz 2013,. Ater a searching the web or iPad stands, Singh came across the lanlree iPad
Accessory ,http:,,www.kickstarter.com,projects,831303939,hanree-ipad-accessory-use-the-ipad-hands-
ree,. According to Singh,
aiavt /vor av,tbivg abovt Kic/.tarter. ra. a t,icat bac/er ti/e av,ove et.e. cave acro..
tbi. iPaa .tava. 1o ve, it too/ea ti/e a coot tbivg ,ov covta bv,. f ,ov gire ve 0, tt .eva
,ov ove of tbev. aiavt ao av, ave aitigevce. aiavt tbiv/ baa to. v vot ivre.tivg. v vot
aoivg tbe .ave .ort of tbivg. a otevtiat .barebotaer rovta ao. v ;v.t bv,ivg a roavct.

Like Singh, a total o 440 backers liked this product and pledged an aerage o >80 per person ,a total o oer
>35, 000 was raised, well oer the target goal o >10,000,. Beore listing his idea, Seth Quest, the project
creator, had neer started a business and had neer manuactured a product. Being unable to complete all the
required product design and manuacturing planning, Quest posted an update almost eight months ater the
unding cycle ended explaining that the project had ailed. Although Quest initially oered to proide
21

reunds, all o his Kic/.tarter money was gone and he was eentually orced into personal bankruptcy
,Markowitz 2013,.
1o explore the idea o new product pre-orders and the possible eects o product rewards on backer
behaior, we split our sample and estimate separate models or 1echnology, Design and Games projects
categories that typically oer tangible new products as rewards or contributions ,Steinberg 2012, see also
Mollick 2012,
8
and projects in any other category. 1he results rom an OLS ixed-eects estimation o
equation ,2, is reported in 1able . Consistent with our preious results, the coeicient estimates or
Pa.tac/ervort and its interaction with a.t!ee/ are in agreement with the diusion o responsibility ,2G,
and deadline eects ,2H,, respectiely. loweer, we ind that backer behaior in these two subsamples
diers once a project reaches its unding goal. Speciically, the coeicient estimate o the interaction term
inoling Pa.tac/ervort and Po.tvvaea is positie and signiicant or 1echnology, Design and Games
projects while it is negatie and signiicant or other projects. 1hus, there is strong eidence o herding
behaior among backers or successully unded projects in the 1echnology, Design and Games categories
,the sum o the coeicient estimates or Pa.tac/ervort, Pa.tac/ervort Po.tvvaea and Pa.tac/ervort
a.t!ee/ is positie and signiicantly dierent than zero, l~9.19, p0.01,.
|insert 1able about here|
1hese results indicate that unded 1echnology, Design, and Games projects with a lot o backer
support are likely to get een more contributors beore their unding cycle ends. Although our data do not
allow us to determine i this represents rational or irrational herding behaior, it is interesting to speculate that
this herding behaior or successully unded projects may be responsible or the product rewards deliery
delays reported by Mollick ,2012,. Successully unded projects may actually be suering rom their own
success! 1his will especially be the case or projects that dramatically exceed their original goal. lor example,
Eric Migicovskys Pebble L-Paper \atch, the largest unded Kic/.tarter project to date, receied almost 69,000
backerswell oer the 1,000 backers expected with their original goal o >100,000. \hile the promised
watch is now shipping more than three months ater its promised deliery date, all o their backers ,including
the authors, will not receie their watch or seeral more months.
Additional studies are needed to more completely understand the possible herding behaior o
consumer-inestors in crowdunding communities. lor example, in light o our results in 1able an obious
recommendation might be to establish a restriction on the number o additional new backers that can support
a project once it reaches its goal. lere, the thinking would be that small startup businesses with limited
resources and connections should not be stressed beyond their real capabilities. Such a restriction howeer,
could also change the dynamics o backer behaior so that any diusion o responsibility eects is
minimized. In this case, potential backers may be een more interested in contributing to projects with a lot

8
\e obtain parallel results or each o these three product categories separately.
22

o past backer support because the project ,and its product rewards, may not be aailable once it meets its
target. 1hus, herding may be more prealent i project backers are limited. lurther research should tackle
these topics more deeply by considering backer motiations and behaior in supporting projects that hae
already reached their goal.

",&,",(!,0

Adler, C. (2011), How Kickstarter Became a Lab for Daring Prototypes and Ingenious Products, !irea,
March 18, http:,,www.wired.com,magazine,2011,03,_kickstarter,all,1 ,accessed May 3, 2012,.

Agarwal, A., C. Catalini, and A. Goldarb ,2011), The Geography of Crowdfunding, aailable at SSRN:
http:,,ssrn.com,abstract~1035.

Ai, C. and E. Norton (2003), Interaction Terms in Logit and Probit Models, covovic. etter., 80,1,, 123-
129.

Allison, P. ,2005,, iea ffect. Regre..iov Metboa. for ovgitvaivat Data |.ivg ., Cary, NC: SAS Institute.

Andreoni, J. (2006), Philanthropy, in S. Kolm and J. \thier ,eds.,, avaboo/ of tbe covovic. of Cirivg,
.ttrvi.v, ava Recirocit,, North-lolland: Llseier, 1201-1269.

Ariely, D. and I. Simonson (2003), Buying, Bidding, Playing, or Competing Value Assessment and Decision
Dynamics in Online Auctions, ]ovrvat of Cov.vver P.,cbotog,, 13,1&2,, 113-123.

Bagnoli, M. and B. Lipman ,1989,, Proision o Public Goods: lully Implementing the Core through Priate
Contributions, Rerier of covovic tvaie., 56 ,4,, 583-601.

Barron, G. and E. Yechiam (2002), Private E-mail Requests and the Diffusion of Responsibility, Covvter.
iv vvav ebarior, 18, 50-520.

Bayus, B. ,2013,, Crowdsourcing New Product Ideas Over Time: An Analysis of the Dell IdeaStorm
Community, Mavagevevt cievce, 59 ,January,, 2013, 226-244.

Bellelamme, P., 1. Lambert and A. Schwienbacher ,2012), Crowdfunding: 1apping the Right Crowd,
aailable at SSRN: http:,,ssrn.com,abstract~183683.

Bergstrom, T., L. Blume, H. Varian (1986), On the Private Provision of Public Goods, ]ovrvat of Pvbtic
covovic., 29,1,, 25-49.

Bernheim, D. (1994), A Theory of Conformity, ]ovrvat of Potiticat covov,, 102,5,, 841-84.

Bikhchandani, S., D. Hirshleifer, and I. Welch (1992), A Theory of Fads, Fashion, Custom, and Cultural-
Change as Informational Cascades, ]ovrvat of Potiticat covov,, 100,5,, 45-486.

Bradord, C. (2012), Crowdunding and the lederal Securities Laws, Cotvvbia v.ive.. ar Rerier, 1, 1-150.

23

Burtch, G., A. Ghose, and S. Wattal (2012), An Empirical Examination of the Antecedents and
Consequences o Inestment Patterns in Crowd-Funded Markets, available at SSRN:
http:,,ssrn.com,abstract~1928168, orthcoming in vforvatiov ,.tev. Re.earcb.

Chasan, E. (2012), Crowd-Funding Industry Braces for Regulation, !att treet ]ovrvat, April 5,
http:,,online.wsj.com,article,SB1000142405202303302504532620030411034.html ,accessed April 16,
2012,.

Cialdini, R. ,2001,, vftvevce: cievce ava Practice, Needham leights, IL: Allyn and Bacon.

Cosh, A., D. Cumming, and A. Hughes (2009), Outside Entrepreneurial Capital, 1be covovic ]ovrvat,
119,540,, 1494-1533.

Darley, J. and B. Latane (1968), Bystander Intervention in Emergencies: Diffusion of Responsibility, ]ovrvat
of Per.ovatit, ava ociat P.,cbotog,, 8,4,, 3-383.

de \itt, N. ,2012,, A Kickstarters Guide to Kickstarter, http:,,kickstarterguide.com,iles,2012,0,A-
Kickstarters-Guide.pd ,accessed Noember 15, 2012,.

Duan, \., B. Gu, and A. \hinston ,2009,, Informational Cascades and Software Adoption on the Internet:
An Empirical Investigation, M Qvartert,, II,1,, 23-48.

Duncan, B. (2004), A Theory of Impact Philanthropy, ]ovrvat of Pvbtic covovic., 88,9-10,, 2159-2180.

Economist (2010), Putting Your Money Where Your Mouse Is, September 2,
http:,,www.economist.com,node,16909869story_id~16909869 ,accessed March 10, 2012,.

lerster, C. and B. Skinner ,195,, cbeavte. of Reivforcevevt, New \ork: Appleton-Century-Crots.

lisher, P., J. Kruegar, 1. Greitemeyer, C. Vogrincic, A. Kastenmuller, Dieter lrey, M. leene, M. \icher, and
M. Kambacher (2011), The Bystander-Lect: A Meta-Analytic Reiew on Bystander Interention in
Dangerous and Non-Dangerous Emergencies, P.,cbotogicat vttetiv, 13,4,, 51-53.

Forsyth, D., L. Zyzniewski, and C. Giammanco (2002), Responsibility Diffusion in Cooperatie
Collections, Per.ovatit, ava ociat P.,cbotog, vttetiv, 28,1,, 54-65.

Garcia, S., K. \eaver, G. Moskowitz, and J. Darley (2002), Crowded Minds: The Implicit Bystander Effect,
]ovrvat of Per.ovatit, ava ociat P.,cbotog,, 83,4,, 843-853.

Gerber, L., J. lui, and P. Kuo (2012), Crowdunding: \hy People are Motiated to Post and lund Projects
on Crowdfunding Platforms, Design, Influence, and Social Technologies Workshop at Computer Supported
Cooperatie \ork, http:,,distworkshop.iles.wordpress.com,2012,01,dist2012_submission_11.pd
,accessed October , 2012,.

Goldfarb, A. and C. Tucker (2011), Privacy Regulation and Online Advertising, Mavagevevt cievce, 5,1,,
5-1.

Gompers, P. J. Lerner, M. Blair, and T. Hellmann (1998), What Drives Venture Capital Fund Raising,
roo/ivg. Paer. ov covovic .ctirit,. Microecovovic. , ,1998,,149-204.

24

Hazen, T. (2012), Social Networks and the Law: Crowdfunding or Fraudfunding? Social Networks and the
Securities Laws\hy the Specially 1ailored Lxemption Must be Conditioned on Meaningful Disclosure,
^ortb Carotiva ar Rerier, 90, 135-180.

Herzenstein, M., U. Dholakia, and R. Andrews (2011), Strategic Herding Behaviors in Peer-to-Peer Loan
Auctions, ]ovrvat of vteractire Mar/etivg, 25,1,, 2-36.

Hemer, J. (2011), A Snapshot on Crowdfunding, Fraunhofer ISI Working Paper,
http:,,isi.raunhoer.de,isimedia,docs,p,arbpap_unternehmen_region,ap_r2_2011.pd\SLSSIONID~1
beb093ebe9de0c0b8d0b9dbcda95 ,accessed lebruary 23, 2012,.

Jacques, 1. and N. Sebire (2010), The Impact of Article Titles on Citation Hits: An Analysis of General and
Specialist Medical Journals, ]ovrvat of tbe Ro,at ociet, of Meaicive bort Reort., 1,2,, 1-5.

Jamali, H. and M. Nikzad (2011), Article Title Type and Its Relation with the Number o Downloads and
Citations, cievtovetric., 88,2,, 653-661.

Karau, S. and K. \illiams (1993), Social Loafing: A Meta-Analytic Review and Theoretical Integration,
]ovrvat of Per.ovatit, ava ociat P.,cbotog,, 65,4,, 681-06.

Latane, B. and J. Darley ,190,, The Unresponsive Bystander: Why Doesnt He Help? New \ork: Appleton-Cnetury-
Crots.

Latane, B. and S. Nida (1981), Ten Years of Research on Group Size and Helping, P.,cbotogicat vttetiv,
89,2,, 308-324.

Ma, C. and M. Manoe (1993), Bargaining with Deadlines and Imperfect Player Control, covovetrica, 61,6,,
1313-1339.

Markey, P. (2000), Bystander Intervention in Computer-Mediated Communication, Covvter. iv vvav
ebarior, 16, 183-188.

Markowitz, E. (2013), When Kickstarter Investors Want Their Money Back, vc., Jan 10,
http:,,www.inc.com,eric-markowitz,when-kickstarter-inestors-want-their-money-back.html ,accessed
January 15, 2013,.

Massolution (2012), Crowdfunding Industry Report, May, http:,,www.marketwire.com,press-
release,crowdunding-platorms-raise-15-billion-successully-und-one-million-campaigns-2011-1654020.htm
,accessed May 8, 2012,.

Mod, C. (2010), Kickstartup: Successful Fund Raising with Kickstarter.com,
http:,,craigmod.com,journal,kickstartup, ,accessed August 15, 2012,.

Mollick, E. (2012), The Dynamics of Crowdfunding: Determinants of Success and Failure, available at
SSRN: http:,,ssrn.com,abstract~2088298.

Ockenfels, A., D. Riley, and A. Sadrieh (2006), Online Auctions, in 1. lendershott ,ed.,, avaboo/ iv
vforvatiov ,.tev., North lolland: Llseier Science, 51-628.

Ordanini, A., L. Miceli, M. Pizzetti, and A. Parasuraman (2011), Crowd-lunding: 1ransorming Customers
into Investors through Innovative Service Platforms, ]ovrvat of errice Mavagevevt, 22,4,, 443-40.

25

Ricker, 1. ,2011), Kickstarted: How One Company is Revolutionizing Product Development, 1be 1erge,
December 20, http:,,www.theerge.com,2011,12,20,2644358,kickstarter-success-product-deelopment-
reolution ,accessed April 16, 2012,.

Roth, A., J. Murnighan, and F. Schoumaker (1988), The Deadline Effect in Bargaining: Some Experimental
Evidence, .vericav covovic Rerier, 8,4,, 806-823.

Salganik, M., P. Dodds, and D. Watts (2006), Experimental Study of Inequality and Unpredictability in an
Artificial Cultural Market, cievce, 311,854,, 854-856.

Schwienbacher, A. and B. Larralde (2012), Crowdfunding of Entrepreneurial Ventures, in D. Cumming
,ed.,, 1be Ofora avaboo/ of vtrerevevriat ivavce, Oxord, UK: Oxord Uniersity Press.

Shontell, A. and M. Dickey (2012), 17 New Ideas That Are Actually Worth Millions, v.ive.. v.iaer,
http:,,www.businessinsider.com,1-million-dollar-ideas-on-kickstarter-2012-12op~1 ,accessed lebruary
18, 2012,.

Simcoe, T. and D. Waguespack (2011), Status, Quality, and Attention: Whats in a (Missing) Name?
Mavagevevt cievce, 5,2,, 24-290.

Simonsohn, U. and D. Ariely (2008), When Rational Sellers Face Nonrational Buyers: Lidence rom
Herding on eBay, Mavagevevt cievce, 54,9,, 1624-163.

Smith, S., F. Windmeijer, and E. Wright (2012), The Effect of Social Information on Charitable Donations:
Evidence from the (Running) Field, Uniersity o Bristol \orking Paper,
http:,,www.bristol.ac.uk,cmpo,eents,2012,charitablegiing,smith.pd ,accessed May 3, 2012,.

Steinberg, D. ,2012,, 1be Kic/.tarter avaboo/, Philadelphia, PA: Quirk Books.

Sugden, R. (1984), Reciprocity: The Supply of Public Goods through Voluntary Contributions, 1be
covovic ]ovrvat, 94,36,, 2-8.

Vesterlund, L. (2006), Why Do People Give? in W. Powell and R. Steinberg (eds.), 1be ^ovrofit ector: .
Re.earcb avaboo/, New laen, C1: \ale Uniersity Press, 568-590.

Voepel, S., R. Eckhoff, and J. Forster (2008), David against Goliath? Group Size and Bystander Effects in
Virtual Knowledge Sharing, vvav Retatiov., 61,2,, 21-295.

Ward, C. and V. Ramachandran (2010), Crowdunding the Next lit: Microunding Online Lxperience
Goods, Uniersity o Massachusetts \orking Paper, http:,,people.cs.umass.edu,~wallach,workshops,
nips2010css,papers,ward.pd ,accessed January 16, 2012,.

Webb, E. and K. Weick (1979), Unobtrusive Measures in Organizational Theory: A Reminder,
.avivi.tratire cievce Qvartert,, 24,4,, 650-659.

Wiesenthal, D., D. Austrom, and I. Silverman (1983), Diffusion of Responsibility in Charitable Donations,
a.ic ava .tiea ociat P.,cbotog,, 4,1,, 1-2.

\ooldridge, J ,2010,, covovetric .vat,.i. of Cro.. ectiov ava Pavet Data, Cambridge, MA: MI1 Press.

26

Wortham, J. (2012), Start-Ups Look to the Crowd, ^er Yor/ 1ive., April 29,
http:,,www.nytimes.com,2012,04,30,technology,kickstarter-sets-o-inancing-rush-or-a-watch-not-yet-
made.html_r~1 ,accessed April 29, 2012,.

Yechiam, E. and G. Barron (2003), Learning to Ignore Online Help Requests, Covvtatiovat ava
Matbevaticat Orgaviatiov 1beor,, 9, 32-339.

Zhang, J. and P. Liu (2012), Rational Herding in Microloan Markets, Mavagevevt cievce, 58,5,, 892-912.

Zhou, J. (2011), 1he 1iming o Out-o-Court Settlements Reisited: 1heory and Cross-Sectional Lidence
rom 1exas since 1988, aailable at SSRN: http:,,ssrn.com,abstract~190543.
2

.<SK9 G
Summary Statistics or All Kic/.tarter Projects 2010-2011
,N~25,058,


/<DL<SK9


49<: 0=TF %9QF 4L: 4<W

lunded ,1~yes,
Goal ,>,
1otal Amount Pledged ,>,
Percent o Goal lunded
Duration ,days,
1otal Number o Backers
1otal Number o Backers in lirst \eek
1otal Number o Backers in Last \eek
Number o Backers Added per Day
Aerage Contribution per Backer ,>,
Number o Reward Categories
Maximum Reward Claimed ,>,
las Video ,1~yes,
Number o \ords in 1itle
1itle has Colon ,1~yes,
Number o Blog Posts by Kic/.tarter
Number o Days on 1op 50 Most Popular List
1otal Number o Public Updates
1otal Number o Priate Updates Only to Backers
Proposed by a Serial Creator ,1~yes,

0.63
8509.83
4633.1
93.22
41.3
62.59
19.31
15.93
1.51
3.64
.89
504.3
0.81
6.3
0.21
0.02
1.41
2.65
0.44
0.08


0.48
139093.10
1359.15
19.94
18.52
189.54
69.3
58.68
8.52
0.08
3.50
964.50
0.39
2.8
0.40
0.15
4.96
4.09
1.28
0.26


0
1
1
0
1
1
1
1
0
1
1
1
0
1
0
0
0
0
0
0


1
2144836
94258
15454.20
93
13364
4683
3535
2212
169.6
31
10000
1
19
1
4
5
83
49
1



28

.<SK9 H
Mean Values or All Kic/.tarter Projects 2010-2011 by lunding Outcome
,N~25,058,



&>:TL:R E9Q9K -OPL9Q9T





/<DL<SK9


XGY
ZG[[\ MN *M<K
X(]^_I[`Y

XHY
G[[\ MN *M<K
X(]G_^GHY

XIY
aG[[\ MN *M<K
X(]GI_bI^Y



=c@=<=L@=LO@
XGYcXHY XGYcXIY XHYcXIY

Goal ,>,
1otal Amount Pledged ,>,
Percent o Goal lunded
Duration ,days,
1otal Number o Backers
1otal Number o Backers in lirst \eek
1otal Number o Backers in Last \eek
Number o Backers Added per Day
Aerage Contribution per Backer ,>,
Number o Reward Categories
Maximum Reward Claimed ,>,
las Video ,1~yes,
Number o \ords in 1itle
1itle has Colon ,1~yes,
Number o Blog Posts by Kic/.tarter
Number o Days on 1op 50 Most Popular List
1otal Number o Public Updates
1otal Number o Priate Updates Only to Backers
Proposed by a Serial Creator ,1~yes,


1458.01
1232.15
15.85
44.26
1.56
.12
2.5
0.40
59.35
.5
25.0
0.
6.31
0.20
0.01
0.15
1.61
0.23
0.06


426.40
442.84
100.26
40.25
42.84
11.18
14.11
1.06
118.68
8.05
608.6
0.9
6.38
0.20
0.01
0.52
2.43
0.3
0.09


4950.20
6905.2
144.28
39.59
95.61
28.64
25.14
2.41
.02
8.16
618.50
0.83
6.40
0.21
0.03
2.39
3.39
0.59
0.08


3.54
-50.50
-459.3
11.68
-53.50
-36.30
-4.28
-13.23
-31.14
-9.04
-30.02
-9.90
-2.04
-0.85
-6.62
-23.53
-26.45
-16.85
-4.50


3.9
-2.32
-51.16
20.23
-29.31
-24.04
-25.18
-110.34
-10.92
-10.23
-2.52
-8.62
-2.16
-3.16
-12.60
-3.99
-30.89
-20.06
-6.30


8.32
-.56
-20.68
8.2
-13.85
-15.46
-6.81
-51.95
23.19
-1.15
3.2
1.32
-0.08
-2.19
-6.2
-20.12
-3.6
-4.13
-1.68
signiicant at 0.05 leel, signiicant at 0.01 leel, signiicant at 0.001 leel

29

.<SK9 I
Variable Deinitions and Summary Statistics
,N~ 1,011,01 Project-Days,



/<DL<SK9

%9NL:L=LM: 49<: 0=TF %9QF 4L: 4<W

ac/er.aaea


Pa.tac/ervort


vvaivg1ive


ir.t!ee/


a.t!ee/


Prirate|aate


Pvbtic|aate


agac/er.

Povtar


~1 i project i receied a new contribution pledge on day t
,0~otherwise,

Ratio o the cumulatie number o backers supporting project i beore
day t to the total number o backers required to reach project is goal

Ratio o the cumulatie number o days that hae elapsed or project i
up to day t to the length o project is funding cycle in days

~1 i project i is in the irst week o its unding cycle on day t ,0
otherwise,

~1 i project i is in the inal week o its unding cycle on day t ,0
otherwise,

~1 i project i has posted an update isible to only prior backers during
day t ,0 otherwise,

~1 i project i has posted an update isible to eeryone during day t ,0
otherwise,

Number o new backers contributing to project i in day t1

~1 i project i is ranked in the top ity o all actie projects in terms o
backers added oer the prior week ,0 otherwise,

0.39


0.49


0.52


0.15


0.1


0.01


0.06


1.49

0.04

0.49


1.08


0.28


0.36


0.38


0.10


0.23


8.25

0.18

0


0


0.02


0


0


0


0


0

0

1


154.54


1


1


1


1


1


2212

1


cont on next page



30

.<ble 3 (cont)
Variable Deinitions and Summary Statistics
,N~ 1,011,01 Project-Days,



/<DL<SK9

%9NL:L=LM: 49<: 0=TF %9QF 4L: 4<W

togPo.t

Coat

Dvratiov

1iaeo

ReraraCategorie.

Creatorerievce


1itteevgtb

Cotov

Pro;ect Categor,

~1 i the Kic/.tarter blog mentioned project i on day t ,0 otherwise,

Goal or project i ,>,

Length o the unding cycle or project i ,days,

~1 i project i has a ideo ,0 otherwise,

Number o reward categories or project i

~1 i the creator or project i created at least one preious Kic/.tarter
project ,0 otherwise,

Number o words in the title o project i

~1 i there is a colon in the title o project i ,0 otherwise,

Dummy ariables or the category in which project i is classiied: Art,
Comics, Dance, Design, lashion, lilm & Video, lood, Games, Music,
Photography, Publishing, 1echnology, 1heater

0.00

10685.02

49.8

0.81

8.04

0.0


6.41

0.21

-



0.02

201512.60

20.66

0.39

3.53

0.25


2.90

0.41

-



0

1

2

0

1

0


1

0

0

1

2144836

93

1

31

1


19

1

1




31

.<SK9 J
Panel OLS Lstimation Results or 1esting 2G
,cluster-robust standard errors, t-statistics in parentheses,

/<DL<SK9


4MT9K G


4MT9K H


4MT9K I


4MT9K J
X!M::9O=9T =M
&<O9SMM;Y

4MT9K V
X)BB9TL<=9
7<O;9D@Y


4MT9K d
X%9K<C9T
7<O;9D@Y

4MT9K `
X09DL<K
7<O;9D@Y


ln,Pa.tac/ervort,
ln,Pa.tac/ervort, ln,Coat,
ln,Pa.tac/ervort, ln,rieva.,

.LB9c/<DCL:R !M:=DMK /<DL<SK9@
ir.t!ee/
a.t!ee/
Po.tvvaea
Povtar
togPo.t
Prirate|aate
Pvbtic|aate
ln,agac/er.,

.LB9c):Q<DL<:= !M:=DMK /<DL<SK9@
ln,Coat,
ln,Coat,
2
Dvratiov
1iaeo
ReraraCategorie.
Creatorerievce
1itteevgtb
Cotov
Pro;ect Categor,

Constant

&LW9T ,NN9O=@
Day-o-\eek
Month-\ear
Project


0.08 ,116.9,




0.16 ,105.02,
0.02 ,13.40,
-0.11 ,-45.3,
0.00 ,0.11,
0.13 ,6.60,
0.12 ,25.64,
0.15 ,0.60,
0.20 ,161.63,


0.12 ,12.89,
-0.00 ,-5.10,
-0.00 ,-59.44,
0.02 ,11.6,
0.00 ,16.36,
0.01 ,3.81,
-0.00 ,-1.66,
0.01 ,4.2,
yes

-0.25 ,-6.4,


yes
yes
no

-0.03 ,-19.66,




0.08 ,41.88,
0.08 ,56.29,
-0.14 ,-53.0,
0.04 ,12.2,
0.08 ,4.4,
0.10 ,23.58,
0.14 ,65.48,
0.13 ,123.80,












0.28 ,24.9,


yes
yes
yes


-0.01 ,-1.26,
-0.00 ,-2.48,



0.08 ,42.25,
0.08 ,56.25,
-0.14 ,-53.36,
0.04 ,12.81,
0.08 ,4.4,
0.10 ,23.59,
0.14 ,65.46,
0.13 ,123.9,












0.28 ,25.24,


yes
yes
yes

-0.03 ,-3.35,

-0.00 ,-2.1,


0.0 ,11.60,
0.09 ,18.58,
-0.13 ,-15.83,
0.06 ,4.9,
0.03 ,0.56,
0.12 ,8.06,
0.15 ,22.8,
0.13 ,38.50,












0.30 ,10.15,


yes
yes
yes

-0.02 ,-15.38,




0.06 ,35.22,
0.0 ,4.58,
-0.14 ,-58.41,
0.06 ,1.1,
0.05 ,2.36,
0.10 ,21.82,
0.13 ,62.59,
0.13 ,132.86,












0.26 ,24.0,


yes
yes
yes

-0.00 ,-4.29,




0.01 ,10.82,
0.04 ,46.68,
-0.04 ,-2.05,
0.0 ,20.32,
0.10 ,5.11,
0.05 ,13.66,
0.06 ,33.9,
0.05 ,69.69,












0.09 ,16.63,


yes
yes
yes


-0.04 ,-30.05,




0.03 ,23.,
0.0 ,63.05,
-0.0 ,-3.54,
0.0 ,19.21,
0.16 ,8.69,
0.0 ,16.93,
0.09 ,45.96,
0.08 ,90.53,












0.19 ,23.00,


yes
yes
yes
(XMS@9DQ<=LM:@Y
(X?DMU9O=@Y
-TU "
H
1,011,01
25,050
0.30
1,011,01
25,050
0.35
1,011,01
25,050
0.35
96,514
2,40
0.35
1,011,01
25,050
0.33
1,011,01
25,050
0.1
1,011,01
25,050
0.31
signiicant at 0.05 leel, signiicant at 0.01 leel, signiicant at 0.001 leel
32

.<SK9 V
Panel OLS Lstimation Results or 1esting 2H
,cluster-robust standard errors, t-statistics in parentheses,


/<DL<SK9


4MT9K G


4MT9K H


4MT9K I
X)BB9TL<=9
7<O;9D@Y


4MT9K J
X%9K<C9T
7<O;9D@Y

4MT9K V
X09DL<K
7<O;9D@Y


ln,Pa.tac/ervort,
ln,Paetac/ervort, ir.t!ee/
ln,Paetac/ervort, a.t!ee/
ln,Pa.tac/ervort, Po.tvvaea
ln,Pa.tac/ervort, ir.t!ee/ Po.tvvaea
ln,Pa.tac/ervort, a.t!ee/ Po.tvvaea

.LB9c/<DCL:R !M:=DMK /<DL<SK9@
ir.t!ee/ Po.tvvaea
a.t!ee/ Po.tvvaea
ir.t!ee/
a.t!ee/
Po.tvvaea
Povtar
togPo.t
Prirate|aate
Pvbtic|aate
ln,agac/er.,

Constant

&LW9T ,NN9O=@
Day-o-\eek
Month-\ear
Project


-0.04 ,-21.14,
-0.01 ,-10.84,
0.03 ,48.39,







0.05 ,18.94,
0.12 ,63.41,
-0.16 ,-62.2,
0.05 ,13.4,
0.08 ,4.68,
0.10 ,22.93,
0.14 ,64.41,
0.13 ,122.18,

0.33 ,28.82,


yes
yes
yes

-0.04 ,-23.5,
-0.01 ,-10.06,
0.04 ,53.0,
0.01 ,1.22,
-0.05 ,-3.15,
-0.06 ,-8.49,


0.0 ,.33,
-0.0 ,-16.60,
0.05 ,16.4,
0.16 ,64.95,
-0.14 ,-41.08,
0.04 ,13.09,
0.08 ,4.65,
0.10 ,22.92,
0.14 ,64.3,
0.13 ,122.32,

0.32 ,2.91,


yes
yes
yes

-0.03 ,-1.80,
-0.00 ,-5.88,
0.03 ,48.29,







0.05 ,1.60,
0.10 ,54.89,
-0.1 ,-65.46,
0.06 ,1.,
0.05 ,2.30,
0.10 ,21.26,
0.13 ,61.63,
0.13 ,130.8,

0.30 ,28.12,


yes
yes
yes

-0.01 ,-.49,
-0.00 ,-3.10,
0.02 ,42.8,







0.01 ,3.01,
0.06 ,50.11,
-0.05 ,-33.55,
0.0 ,20.64,
0.10 ,5.0,
0.05 ,13.30,
0.05 ,33.33,
0.05 ,68.52,

0.11 ,20.02,


yes
yes
yes

-0.04 ,-33.13,
-0.00 ,-2.01,
0.02 ,4.45,







0.03 ,11.8,
0.10 ,66.33,
-0.09 ,-44.50,
0.0 ,19.53,
0.16 ,8.61,
0.0 ,16.48,
0.09 ,45.18,
0.08 ,88.89,

0.22 ,26.41,


yes
yes
yes
(XMS@9DQ<=LM:@Y
(X?DMU9O=@Y
-TU "
H

1,011,01
25,050
0.35
1,011,01
25,050
0.35
1,011,01
25,050
0.33
1,011,01
25,050
0.1
1,011,01
25,050
0.31
signiicant at 0.05 leel, signiicant at 0.01 leel, signiicant at 0.001 leel
33

.<SK9 d
Panel OLS Lstimation Results or Project Updates
,cluster-robust standard errors, t-statistics in parentheses,


/<DL<SK9

4MT9K G
X5>SKLO '?T<=9@Y

4MT9K H
X5DLQ<=9 '?T<=9@Y

ir.t!ee/
a.t!ee/
ir.t!ee/ vvaea
a.t!ee/ vvaea

.LB9c/<DCL:R !M:=DMK /<DL<SK9@
Po.tvvaea
Povtar
togPo.t
ln,agac/er.,
ln,Pa.tac/ervort,

Constant

&LW9T ,NN9O=@
Day-o-\eek
Month-\ear
Project


0.02 ,16.50,
0.00 ,0.01,
-0.02 ,-12.48,
0.04 ,2.0,


-0.03 ,-23.83,
0.02 ,8.83,
0.06 ,3.12,
0.03 ,50.65,
0.01 ,8.05,

0.0 ,12.2,


yes
yes
yes

0.00 ,4.11,
0.01 ,14.46,
-0.00 ,-0.56,
0.01 ,.06,


-0.00 ,-1.00,
0.00 ,3.39,
0.01 ,1.20,
0.01 ,24.96,
0.00 ,8.16,

0.01 ,5.15,


yes
yes
yes
(XMS@9DQ<=LM:@Y
(X?DMU9O=@Y
-TU "
H
1,011,01
25,050
0.11
1,011,01
25,050
0.06
signiicant at 0.05 leel, signiicant at 0.01 leel, signiicant at 0.001 leel


34

.<SK9 `
Panel OLS Lstimation Results by Project 1ype
,cluster-robust standard errors, t-statistics in parentheses,


/<DL<SK9


4MT9K G
X.9OP:MKMRC_
%9@LR:_ *<B9@Y

4MT9K H
X(M= .9OP:MKMRC_
%9@LR:_ *<B9@Y

ln,Pa.tac/ervort,
ln,Pa.tac/ervort, Po.tvvaea
ln,Pa.tac/ervort, ir.t!ee/
ln,Pa.tac/ervort, a.t!ee/

.LB9c/<DCL:R !M:=DMK /<DL<SK9@
Po.tvvaea
ir.t!ee/
a.t!ee/
Povtar
togPo.t
Prirate|aate
Pvbtic|aate
ln,agac/er.,

Constant

&LW9T ,NN9O=@
Day-o-\eek
Month-\ear
Project


-0.04 ,-6.21,
0.06 ,5.09,
-0.01 ,-3.49,
0.01 ,6.2,


-0.04 ,-4.00,
0.05 ,5.0,
0.06 ,9.9,
0.05 ,5.53,
-0.01 ,-0.34,
0.06 ,4.55,
0.09 ,15.58,
0.09 ,2.92,

0.24 ,.23,


yes
yes
yes

-0.04 ,-20.45,
-0.11 ,-8.59,
-0.01 ,-11.35,
0.04 ,48.8,


-0.16 ,-56.89,
0.05 ,16.4,
0.13 ,63.48,
0.04 ,11.93,
0.09 , 4.94,
0.10 ,22.4,
0.14 ,63.22,
0.13 ,118.98,

0.35 ,28.68,


yes
yes
yes
(XMS@9DQ<=LM:@Y
(X?DMU9O=@Y
-TU "
H

80,19
2,00
0.46
931,522
23,043
0.34
signiicant at 0.05 leel, signiicant at 0.01 leel, signiicant at 0.001 leel


35

&LR>D9 G
1he Dynamics o Backer Support or Cody \ebb's New CD
,http:,,www.kickstarter.com,projects,codywebb,cody-webbs-new-cd,




&LR>D9 H
Distribution o Project lunding Outcomes
,N~25,058,


36


&LR>D9 I
Aerage Number o Backers Added to a Project in Any Day
,N~25,058,





3

-55,(%)e

.<SK9 -G
Panel Logit Lstimation Results
,cluster-robust standard errors, t-statistics in parentheses,


/<DL<SK9


4MT9K G


4MT9K H


4MT9K I


ln,Pa.tac/ervort,
ln,Paetac/ervort, ir.t!ee/
ln,Paetac/ervort, a.t!ee/

.LB9c/<DCL:R !M:=DMK /<DL<SK9@
ir.t!ee/
a.t!ee/
Po.tvvaea
Povtar
togPo.t
Prirate|aate
Pvbtic|aate
ln,agac/er.,

.LB9c):Q<DL<:= !M:=DMK /<DL<SK9@
ln,Coat,
ln,Coat,
2
Dvratiov
1iaeo
ReraraCategorie.
Creatorerievce
1itteevgtb
Cotov
Pro;ect Categor,

&LW9T ,NN9O=@
Day-o-\eek
Month-\ear
Project


0.58 ,152.31,




1.0 ,123.31,
0.09 ,13.33,
-0.6 ,-59.45,
0.84 ,34.69,
1.3 ,.63,
0.0 ,24.59,
0.91 ,0.81,
1.08 ,183.40,


0.59 ,15.26,
-0.10 ,-4.11,
-0.01 ,-60.45,
0.16 ,13.94,
0.02 ,18.40,
0.0 ,4.48,
-0.00 ,-1.55,
0.05 ,5.42,
yes


yes
yes
no


-0.18 ,-18.24,




0.51 ,43.30,
0.55 ,61.80,
-0.83 ,-60.53,
0.66 ,24.32,
0.96 ,5.5,
0.66 ,23.24,
0.89 ,6.69,
0.0 ,130.60,













yes
yes
yes


-0.14 ,-13.11,
-0.15 ,-21.8,
0.30 ,36.81,


0.19 ,11.63,
0.6 ,68.25,
-1.0 ,-2.93,
0.0 ,25.64,
0.96 ,5.54,
0.66 ,22.96,
0.88 ,66.83,
0.69 ,128.44,













yes
yes
yes

(XMS@9DQ<=LM:@Y
(X?DMU9O=@Y
!PLc0f><D9T

1,011,01
25,050
123,8.98
96,908
23,663
49,309.58
96,908
23,663
51,063.32
signiicant at 0.05 leel, signiicant at 0.01 leel, signiicant at 0.001 leel