Categorization and success of crowdfunding...

30
Categorization and success of crowdfunding projects Jonathan Sitruk SKEMA Business School, Strategy, Entrepreneurship & Economics Department 60 rue Dostoïevski, 06902 Valbonne, France Tel: 04 93 95 45 63 e-mail: [email protected] This article focuses on crowdfunding determinants of success by using a novel approach and methodology. We use categorization theory as our theoretical framework that claims that success of organizations is linked to inner and outer categorical specification and our results bring support to three contentions surrounding this theory. We first suggest that clearly belonging to a category is associated with success. We also show that this is a concave relationship between the success and the amount of categories a project belongs to consistent with the idea of spanning categories. We finally look at inner categorical proximity, thus addressing problems of optimal distinctiveness and find that project success is concavely related to the level of proximity. The study is based on over 15,000 projects in the Music category presented on Indiegogo from 2009 to early 2015. The use novel semantic analysis methods to measure proximity of projects with musical genres.

Transcript of Categorization and success of crowdfunding...

Page 1: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

CategorizationandsuccessofcrowdfundingprojectsJonathanSitruk

SKEMABusinessSchool,Strategy,Entrepreneurship&EconomicsDepartment

60rueDostoïevski,06902Valbonne,FranceTel:0493954563

e-mail:[email protected]

Thisarticlefocusesoncrowdfundingdeterminantsofsuccessbyusing

anovelapproachandmethodology.Weusecategorizationtheoryas

ourtheoretical frameworkthatclaimsthatsuccessoforganizations

is linkedtoinnerandoutercategoricalspecificationandourresults

bringsupport to threecontentionssurroundingthis theory.We first

suggest that clearly belonging to a category is associated with

success.Wealsoshowthatthisisaconcaverelationshipbetweenthe

successandtheamountofcategoriesaprojectbelongstoconsistent

with the idea of spanning categories. We finally look at inner

categorical proximity, thus addressing problems of optimal

distinctiveness and find that project success is concavely related to

the levelofproximity.Thestudy isbasedonover15,000projects in

theMusiccategorypresentedonIndiegogofrom2009toearly2015.

The use novel semantic analysis methods to measure proximity of

projectswithmusicalgenres.

Page 2: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

1 Introduction

Crowdfundingisafairlynewphenomenonandscholarsinthefieldofmanagement

and entrepreneurship have taken great interest in understanding its multiple

facets.Oneofthemostprominentfocusestakenistounderstanddeterminantsof

success of crowdfunding projects (Mollick, 2014). Researchers have taken

different theoretical foundations to explainwhat are success factors of different

crowdfundingprojects.Forinstance,Agrawaletal.(2014)andMollick(2014)take

itasanasymmetryofinformationproblemandexplainthatsignalssuchasuseof

videos,images,typeoffinancing,lengthofprojects,numberofprojectholdershelp

reduce the latter.Others suggest that crowdfunding success ismostlydue to the

amount of friends and family (i.e.: love money) investments in early stages of

projects that lead to a herding behavior by other funders. We take a different

perspectiveinthispaperbyusingcategorizationtheorytoexplainhowaprojects’

self-categorizationhasaninfluenceonthesuccessofaproject.

In the third point on Indiegogo’s tips to having a successful campaign they

recommendthat“themoreyoucandemonstrateyou’vethoughtthroughasmany

aspects of your campaign as possible […] the better your audience can visualize

where their money will go, which builds trust”. However, for the audience to

understandwheretheirmoneywillgo,theymustfirstclearlyunderstandwhatitis

thataprojectisproposingtodo.Weproposetoexplorethisideabyadaptingthe

categorizationliteratureasadeterminantofcrowdfundingsuccess.

Thus,whilescholarsapplyingsignalingtheoryincrowdfundinghavereliedonthe

idea that project holders try to communicate on the underlying quality of their

offering,wepropose to use categorization theorywhich offers funders away to

understandtheprojectsthatarepresentedtothem.Inotherwords,inorderfora

projecttohopetoreceivefunds,fundersfirsthavetounderstandwhattheproject

isabout.Thecontentionofourpaperisthatexploringcategorizationwillallowus

togiveprescriptiveadvicetocrowdfundingprojectholdersaswellascontributeto

management theory and issues of asymmetry of information by adopting a

different theoretical lens.Webelieve thatusingcategorization theorycanoffera

noveltakeonthedeterminantsofcrowdfundingsuccess.

Page 3: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Moreprecisely,categorizationtheoryhasbeenusedtodetermineventuresuccess

andinvestmentdecisions(Brewer,1991;Überbacher,2014;VergneandWry,2014).The

main premise of the theory is that individuals are cognitively limited and use

categories as amental help in their sense-makingprocess.As such, they see the

wayothers–letthembeindividuals,ventures,objects–fitintopredefinedboxes

orcategoriesthattheyalreadyunderstandandthattheycantheninterpreteasier

(Navis and Glynn, 2010). Furthermore, self-categorization allows individuals or

ventures to express their social identity (Brewer, 1991), vital to their survival.

Identity isan importantconstruct inmanagementstudiesandagoodcategorical

belonging will allow an entity to express both their similarity as well as their

difference,whichisknownasoptimaldistinctiveness.

Using novel data collected from Indiegogo, our objective is to contribute to this

debateandseehowcategorization influencescrowdfundingsuccess.Assuch,we

analyze how crowdfunding projects in the music category use subcategories to

betterexpresstheiridentity,makingiteasierforinvestorstounderstandwhatthe

project aims to do. We find that not mentioning a categorical belonging has a

negativeimpactonsuccess.Wealsolookatcategoryspanningaswellandreport

the results. In the next section we perform a literature review to explore

crowdfundinginmoredetail(definition,types,actors,theprocess),seehowsignal

theory has been applied to it. In section 3, we develop our hypotheses, while

section 4 and 5 deal with data collection and methods. We finally conclude in

section6.

2 TheoreticalBackground

2.1 Crowdfunding

2.1.1 Definitionofcrowdfunding

Lambert and Schwienbacher (2010, p.6) define crowdfunding as “an open call,

essentially throughthe Internet, for theprovisionof financialresourceseither in

theformofdonationorinexchangeforsomeformofrewardand/orvotingrights

in order to support initiatives for specific purposes”. Hemer (2011) complements

LambertandSchwienbacher(2010)definitioninaquasi-identicalfashion.

Page 4: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Althoughithasgrowninpopularityoverthepast5-10years,aswitnessedbythe

increased media coverage it has received, crowdfunding is actually an old

phenomenon.Classicexamplesofpre-Internetcrowdfundingarenumerous;such

asPulitzer’scrowdfundingcampaign(CFC)togetNewYorkerstodonatemoneyto

fundtheStatueofLiberty’spedestalbaseorevenMozart’spre-sellingofticketsto

financeconcerts(Hemer,2011;Niederer,2014).

Thus, crowdfunding has been revitalized, has seen substantial growth, and has

become increasingly popular in the advent of Web 2.0, the Internet, and “viral

networkingandmarketing”whichhasallowedmobilizinga“largenumberofusers

inspecificWebcommunitieswithinarelativelyshortperiodoftime”(Hemer,2011,

p.8). In addition, Schwienbacher & Larralde (2010) explain that the three main

characteristics ofWeb2.0provide the technological ability and a “participatory”

Internetthatmovesbeyondtheprevious“passive”oneinwhichusersweremerely

just observers. These main characteristics are: 1) collaboration permitting

combination of knowledge and resources, 2) openness permitting easymultiple

contributions,and3)easeofaccessincreasingparticipation(Leeetal.,2008).

2.1.2 TypesofCrowdfunding

Severaltypesofcrowdfundingexistandhaveemergedovertime.Rankedinorder

of simplicity both regarding the level of legal complexity, and asymmetry of

information, these are: donation-, reward-, lending-, and equity-based

crowdfunding (Ahlers and Cumming, 2012; Hemer, 2011; Schwienbacher and

Larralde,2010).Otherstudieshavegivenawidertypologyofcrowdfundingtypes

(adding for example equity-like crowdfunding)while the crowdfunding industry

report shows the recent emergence of royalty crowdfunding as well as many

mixedmodels(Massolution,2012,2013). Inthisstudy,wewillonly focusonthe

formerfournamedcrowdfundingmodels.

As the name suggests, donation-based crowdfunding is one in which funders

donate their money to a crowdfunding campaign. This model is particularly

appropriatefornot-for-profitorganizationstogainaccesstodonors.Theycando

so directly (e.g.: the Red Cross website) or by using dedicated crowdfunding

platforms(e.g.:Crowdrise).Inreward-basedcrowdfunding,thecreatorprovidesa

Page 5: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

rewardor gift to the funder in exchange for their contribution. This canbe aT-

shirt,oracap,orevenanearlyversionoftheproductinpreorderingschemes(e.g.:

Kickstarter or Indiegogo – the platform from which we have acquired data).

Reward crowdfunding has increasingly been seen as a pre-sales crowdfunding

platform.Moneyislentandpaidbackwithinterestinlending-basedcrowdfunding

(e.g.:Prosper,Unilend).Itisinterestingtonotethattherearetwotypesoflending

platforms;somelendtoindividuals(peer-2-peer),whileotherslendtobusinesses

(peer-2-business). Finally, equity-based crowdfunding (e.g.: ASSOB, Wiseed,

Crowdcube)allowsfunderstoacquiresharesofaproject,generallyalsoobtaining

votingrightsintheventurecreated.

2.1.3 Theactorsincrowdfunding

Several established actors are present in crowdfunding (Ahlers and Cumming,

2012;Griffin,2012;Hemer,2011).Thecreators(alsoknownasthecrowdfunded,

founders, or entrepreneurs) are the people who launch their crowdfunding

campaigninthehopesoffundingtheiridea,whichcanbeavarietyofprojectslike

asong,anartproject,abusinessidea,orahumanitarianeffort.Thefunders(also

referredtoascrowdfunders,investors,orbackers)aretheindividualswhodecide

to fund the creator’s idea. They give money in exchange for a pre-established

return. The crowdfunding platforms are the virtual place where funders and

creatorsmeet;theyareaperfectillustrationof2-sidedmarkets.Thegovernment

isalsoseenasaninvolvedpartyespeciallyintheimplementationofthelegaland

regulatoryframeworkconcerningcrowdfunding(Belleflamme,Omrani,andPeitz,

2015;Bradford,2012;Griffin,2012;Hazen,2011;Mollick,2014).

2.1.4 Thecrowdfundingprocess

The principle is fairly simple (De Witt, 2012). Independent of the model used,

funders and creators meet on a crowdfunding platform. Creators propose their

projectsandask thecrowd(i.e.: the funders) to invest in theminexchange fora

predetermined return (which varies according to the project, platform, and

crowdfundingmodelused).Iffunderslikeaproject,theycandecidetoinvestinit

for one of the offered predetermined amounts. The individual contributions are

generallykeptbytheplatforminanescrowaccountuntilthetargetamountsetby

thecreatorhasbeenmetorexceeded(DeBuysereetal.,2012).Ifacreatorattracts

Page 6: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

sufficient backers, the campaign is a success and they will receive the money.

Otherwise, the campaign is a failure and contributions are returned to the

individualfunders.

2.1.5 Priorresearchincrowdfunding

Recent research attempting to explain the determinants of success in

crowdfunding has concentrated on the idea of asymmetry of information that

exists between founders and backers and has used signaling theory as a basis.

Consistentwiththistheory,apparentqualityofaprojectseemstobeanimportant

signalthatfundersobservetolowertheirasymmetryofinformationwithregards

proposed projects. Interestingly, what is considered to be a signal of quality by

fundersvariesdependingonthetypeofCFP.Itwouldappearthatfundersseemto

usemanysignalsemittedbycreatorsandprojectstoinferthequalityofaproject

inordertotryandcircumventtheaforementionedasymmetryofinformationwith

varyingdegreesofsuccess.

For instance, in peer-to-peer lending, group leaders serve as signals for other

investors to identifyquality investmentsbecause the latter trust that the former

perform the appropriate screening of the investments they are pursuing

(Hildebrandetal.,2013).Ontheotherhand,inequityCFPs,AhlersandCumming

(2012) attempt to shed light on this exact issue; “given different start-upswith

similarobservablecharacteristics,what leadssmall investors to invest incertain

start-ups andnot inothers?” (Ahlers andCumming, 2012:2).The authors claim

that although asymmetry of information in crowdfunding is quite important,

“small investors have seemingly been able to infer the quality of start-ups by

interpretingtheinformationprovidedontheplatform”(Ahlers&Cumming,2012,

p. 18). In staying close to signals of quality traditionally used in entrepreneurial

finance, theauthorsconclude that themainsignals fundersuse todetermine the

quality of a project are the number of board members, the entrepreneur’s

education, the network quality, the envisaged exit type (IPO or trade sale), the

provisionoffinancialforecast,andthebusinessage.Wefocushereafteronsignals

usedinrewardcrowdfundingplatform.

Page 7: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Mollick(2014)associatespreparednessofaprojectholderasanoverallsignalof

quality and shows its positive impact on project success. He measures this

preparednessbylookingatthreedifferentcriteria:whetherprojectshadavideo,

the speed of updates (i.e.: activity creators had on the platform), and spelling

errors.Second,theorganizationalformofprojectsmaybeanindicatorofquality.

Indeed,observingthatprojectsthataretaggedasnot-for-profitonCFPSaremore

successful,theysuggestthattheseprojectholdersmaybe“morepronetocommit

to high quality products or services” (Schwienbacher and Larralde, 2010: 10)

becausetheyput“lessemphasisonprofitmaking,theymaythereforefocusmore

on quality, which may be an important requirement for attracting donations”

(SchwienbacherandLarralde,2010:10).Third, social capitalhasbeen identified

asadeterminantofprojectsuccessonseveraldifferenttypesofCFPs(Agrawal,etal.,

2011; Colombo, Franzoni,&Rossi-Lamastra, 2014; Freedman& Jin, 2014;Mollick, 2014;

Zvilichovsky et al., 2013). However, the reasons that social capital increases one’s

chances of success are often interpreted quite differently. Some consider that

socialcapitalprovidesaccesstomorepotentialfundersandsignalquality(Mollick,

2014;ShaneandCable,2002),maybereflectingsituationsofreciprocity(Colomboet

al.,2014;Zvilichovskyetal.,2013)orevencharity(FreedmanandJin,2014).

Although many scholars have already discussed the importance of signaling

quality forproject success,wepropose toobservehowcategorization influences

crowdfunding success. Indeed, by self-categorizing themselves into specific

groups,venturescananswerthequestion“whatdowedo”(Martens,etal.,2007),

making it easier for investors tomake sense of them. Next section explains our

rationaleforusingcategorizationtheoryandpresentsourhypotheses.

2.2 Anoverviewofthecategorizationliterature

Socialpsychologysuggeststhatpeoplehavelimitedcognitiveabilitymeaningthat

theyareunabletounderstandeverythingthattheytrytoevaluatewithoutusing

some sense making tools (Brewer, 1991). Indeed, we do not relearn how to

distinguish the difference between an action movie, a smartphone, and a horse

every timeweencounter them inoureveryday lives.Oneof these sensemaking

tools is categorization (Brewer,1991)which essentially consists of fitting objects,

people,businesses intoboxes forwhichwehaveaclearunderstandingandclear

Page 8: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

prototype. This is more efficient and cognitively feasible given our limited

capacities.We thus recognize an actionmovie because wemay at the broadest

level categorize it as amovie (as opposed to a phone, or an animal).Within the

movie realm, wemay acknowledge that themoviewe arewatching produces a

higher level of stress and adrenaline than other types of movies (e.g.: comedy,

drama).Thesecategoriesallowustoclearly identify that thisparticularmovie is

an action movie. It is important to note how categories can occur in different

scopes (i.e.: global level vs. more precise) but our understanding relies on this

categorization.Cognitivelyspeaking,whenweseeanactionmovieandtrytomake

sense of it, we try to see where it fits, understand the “what is it” question

(Martens,etal.,2007)byseeinghowitcomparesto theprototypical idealof the

categorythatwearecomparingitto.

3 Hypotheses:

Past research has explained that category adherence is crucial in order for

venturestoacquiretheresourcesthattheyneedtosurvive(Zuckerman,1999).Once

again,audiences(i.e.: investors)whoevaluateproducers(i.e.:projectholders)do

so with limited cognitive abilities and rely on categorization in order to make

senseoforganizationsmoreeasily(Brewer,1991;Vergne,2012).Onceaninvestor

clearlyunderstandswhatcategoryaprojectclaimsbelongingto,theycanevaluate

whether this project correspond to what is expected of this category. Indeed,

"categories convey the cultural 'codes' that are associated with belonging to a

particularcategory"(Vergne&Wry,2014,p.63).Iftheprojectmeetstheexpectations,

it has a higher chance of being financed than if it does not. Also, if the project

cannotbepositionedwithinanycategory,investorswillbetroubledandunableto

evaluateitwithregardthatcategory’sstandards(Brewer,1991).Letusillustrate

byobservingourcontext.

Thequantityofprojectscreatedoncrowdfundingplatformsisvastandinvestors

donothavethetimetothoroughlyreadthrougheachprojectanddeducewhatthe

project intends to do with the funds they collect. Investors therefore rely on

categories to understand for instance that a particular project aims to create a

thrillermovieandnotaromanticone.Clearlyexplainingthatoneintendstofilma

thrillermakesinvestorsunderstandwhatkindofmovieitwillbeandthisclaimed

Page 9: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

category membership to the thriller genre should warrant a tailored project

descriptionthatisreflectiveofit.Therefore,wecontendthatprojectholderswho

claim their membership to a category in their project description will be more

likely to be understood by potential funders and thus have more successful

crowdfundingcampaigns.Forthesereasons,wepropose:

H1:Crowdfundedprojectsthatexpresscategoricalmembershipintheir

descriptionaremorelikelytobesuccessfulthanthosethatdonot.

Although expressing one’s categorical membership is important for venture

success, spanning has traditionally been viewed as having a negative effect on

organizational success. Spanning is the idea of belonging to more than one

categoryat thesame levelofanalysis.Somescholarshaveclaimedthatspanning

will make ventures unable to be as specialized as their single category

counterparts (Zuckerman,1999)given that theyarebydefinitiondiversifying into

two different categories. Other scholars have contended that spanning creates

confusion for investors who, as the amount of category belonging increases,

become increasingly unaware of what a venture does (Zuckerman, 1999). In line

withcategorizationliterature,asaventureincreasestheamountofcategoriesitis

in, theprototypesagainstwhichan investorhastoevaluatetheventure increase

anditbecomesincreasinglyproblematictodoso.

Illustrating with an example from the context at hand, if a project claims

membershiptomanydifferentmusiccategories in itsdescription,therewillbea

lackofclaritywithregardsfunderswhowillincreasinglyhavetroubleinterpreting

what it isaprojectdoes (Brewer,1991). Inotherwords, theywillnotknow from

which categorical prototype to draw from and thismisunderstandingwillmake

funders less likelyto invest intheprojectathandbecausetheywillbeunableto

evaluateit.Furthermore,belongingtovariousmusiccategorieswillalsomakethe

members of thatmusical categoryquestion theproject’s true belonging. To sum

up, crowdfunded projects that claim membership to several categories are less

likely to succeed than those who belong to fewer ones. For these reasons we

propose:

Page 10: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

H2:Thereisanegativerelationshipbetweenacrowdfundingproject’s

successandthenumberofcategoriesitbelongsto.

Thenumberofcategoriesthataprojectbelongstoisnottheonlydeterminantofa

project success. Indeed, within a given category, studies have suggested that

organizationsmustalsomeetcertainconditionsinordertobesuccessful.Indeed,

they must be relatively different from others as well as the prototype of this

category in order to standout in order for investors to see the organization as

distinctive(DeClercqandVoronov,2009;NavisandGlynn,2011).Thisdistinctivenessis

important as it allows organizations to express their individuation. Phrased

differently, they must not be too similar to the prototype of the category they

belong to or they run the risk of being seen as identical to anymember of that

given category (Navis andGlynn, 2011). Being too distinctive is risky as well and

reflectsthepointthatwediscussedintheprevioushypothesis.Ifanorganization

isfartoodissimilartotheprototypeandotherswithintheircategory,theywillbe

seenasnotrespectingtherulesandnormsofthelatterandthusreceiveanegative

discount (Zuckerman,1999). Furthermore, if a category is far too distinctive, they

maynotevenbeunderstoodasbeingpartofagivencategory,whichshouldalso

carrythesamenegativeconsequences.Organizationsmustthusbalancebetween

standingoutand fitting inagiven category inorder to increase their chancesof

success.

Thisdualityoffittinginandstandingoutiswhattheliteraturereferstoasoptimal

distinctiveness(Brewer,1991),legitimatedistinctiveness(NavisandGlynn,2011),or

institutional and innovative legitimacy (De Clercq and Voronov, 2009). Thus, if an

organizationcorrectlymanages thisdualityby following therulesandnormsset

byacategorywhilebeingdifferentfromothermembersandtheprototypeofthe

same category, the audience will understand it as an original member of that

category. This should in turn increase the organization’s chances of survival

through resource procurement (Durand and Jourdan, 2012; Überbacher, 2014).

Illustratingthisinthecrowdfundingcontext,projectsthatclaimmembershiptoa

categorymustbesimilartothatcategorytobeconsideredasamemberbutalso

individuatedenoughfromit,inordertohaveachanceofsecuringfinancing.Thus,

Page 11: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

there shouldbe a curvilinear relationshipwith regard thedegreeof proximity a

projecthastoacategoryanditslevelofsuccess.Forthisreason,wepropose:

H3:Thereisaninverted-Urelationshipbetweenacrowdfundingproject’s

successanditsproximitytothecategoriesitbelongsto.

Inthenextsectionwediscussdatacollectionandourempiricalsetting

4 DataCollection:

The data collected are from the crowdfunding platform (CFP) Indiegogo. The

platformwas launched in2008and isnowthesecond largestrewardCFP in the

U.S.,afterKickstarter.Projectsaredividedinto22categoriessuchasmusic,dance,

smallbusiness,film,technology,andanimalsandacceptsprojectsfromallaround

theworld,writtenindifferentlanguages,andaskingforoneof5currencies(USD,

EU,GBP,AUD,CAD).Theplatformisavailablein4languagesandacceptsprojects

fromverified non-for-profits, individuals, groups, and companies to name a few.

Projectsnormallylastupto60or120daysdependingonthetypeoffinancingthat

hasbeenchosen:All-or-Nothing(AON)vs.Keep-it-All(KIA).Projectholderscreate

theirprojectandexplainwhatperks(giftsorproducts),ifany,theyarewillingto

giveforthedifferentfundinglevelstheypropose.

Each project ismade up of 4 pages: Story,Updates,Comments, andFunders and

sometimesafifthonenamedGallery.TheStorypageissimilartothehomepageof

anywebsiteandiswheremostinformationaboutaprojectisfound(Annex1.1).A

project holder will provide information such as the title of their campaign; the

category;thecity&country;anyexternallinksthattheywanttoprovidesuchas

socialnetworks(Facebook,Twitter,YouTube,andLinkedIn),personalwebsitesor

blogs,orevencommunity specificwebsites (like IMDB formovies); theproject’s

amountgoal&percentagecompletion;timeleft;andperksthattheypropose;and

projectleaders’Indiegogoprofilelinks.

The Updates page allows project holders to discuss relevant updates to their

projectduringorevenafteracampaign(Annex1.2).TheCommentspageprovides

one with access to the amount of comments, the comment text, the comment

Page 12: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

author,andtheirlink(Annex1.3).ThisisthesamefortheFunderspagebutwith

theamountthattheyhavedecidedtofund(Annex1.4).Inisimportanttonotethat

theamountpledgedaswellasthepersonnamecanbeanonymous,anddoesnot

necessarilyhavealinkasexplainedabove).BothCommentsandFunderspagesare

sequentiallyordered ina reverse chronologicalmanner (i.e.: last amount funded

appearingfirst).

4.1. DataCollectionTimeframe

The data were collected by means of a web crawler from December 2014 to

February 2015 from the Indiegogo platform. We collected both ongoing and

finishedprojectsandhavealltheprojectsstartingin2009anduntilthebeginning

of 2015. The datawere collected in the following fashion.We first locally saved

every project URLs (3 pages per project: story, comments, and funders), all

partnershipURLs,andallindividualURLs.Thereasonfordoingthisistwofold.On

theonehandandasamatterof courtesy, it allowsus to respect theCFPbynot

sendingtoomanyrequestswhileacquiringthedata.Onthesecondhand,itallows

ustohaveeasyaccesstothefiles,havethembackedupandhavethemsavedina

way thatwe can access the data. Indeed, because Indiegogo constantlymodifies

andupdatestheirwebsite,thedatacrawlerthatweuse(explainedbelow)would

havetobeconstantlyupdatedwhichwouldcausedelayindatagathering.

4.2EmpiricalSetting

We gather information on more than 400,000 finished and active Indiegogo

projects. Inorder to testourhypotheses,we focusonover15,000projects from

theMusiccategory.Thereasonwedothisistwofold.First,researchhasbeendone

oncategorizationinmusic(MontautiandWezel,2014)givingusmoreconfidence

tousethistestedcontext.Second,andinpartbecauseofourfirstpoint,usingthe

context of music allows us to easily produce independently generated music

subcategories(i.e.:genres)thatwecanthenusetoassignprojectsto.Thisallows

limiting endogeneity problemswhen creating the dictionary.We develop on the

way we build our musical genre dictionary in the following sections. We could

haveusedtheSmallBusinessorHealthcategoriesbutthiswouldhaverequiredthe

additionalworkof searching for the subcategories that exist in these categories.

Indeed, had we used our text analysis methods to create these categories, we

Page 13: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

would run the risk of endogeneity and searching for these subcategories online

maybemorecomplicatedthanalreadystandardizedandstudiedonessuchas in

themusicalfield.

To create our constructs,we analyze the descriptions ofmusic projects through

two vector space modeling (VSM) approaches called Word2Vec (W2V) and

Doc2Vec(D2V).Thenextsectionexplainswhattheseprocessesareandwhatdata

isextracted.Inalaststep,thedataisusedtorunoureconometrictestsinStata.

4.2. VectorSpaceModeling

4.2.1. Introduction

UsingthedatacollectedfromourcrawlerandtheprogramminglanguagePython,

we performed vector space modeling (VSM) approaches called word to vector

(Word2Vec)anddocumenttovector(Doc2Vec)onthedescriptionof16,000Music

projects.We focusonmusicprojects inorder to test ourhypotheses concerning

categorybelongingandevolution.

VSMisseenasrobustinInformationRetrievalTheory(Manningetal.,2009)andis

thebackboneofmodern-daysearchengines(TurneyandPantel,2010).Word2Vec

and Doc2Vec are currently used by technology startups in the field of real-time

information retrieval from social media websites (i.e.: Twitter or Facebook) to

determine attributes such as people’s reaction (positive or negative) to events,

words,orproducts.ToimplementtheseVSM,weusewhatiscommonlyreferred

to as machine or deep learning where the aim is to teach the computer to

recognize the cooccurrence of words by going over a corpus, derive their

likelihoodofoccurrenceand,inthecaseofWord2Vec&Doc2Vec,determinetheir

meaning. The following sectionswill explain the process inmore detail and the

datathatwerederivedfromit.

4.2.2. Process

Before explaining themechanisms ofWord2Vec andDoc2Vec, it is important to

understanditsorigins.Vectorspacemodelingrepresentseachwordusedwithina

compilationoftextsinahigh-dimensionalspace.Thenaturallanguageprocessing

environment calls this compilation of texts a “corpus”. In our study, this corpus

Page 14: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

consists of all themusic project descriptions from Indiegogo andwill further be

discussed in section 1.2.3. In traditional VSM like Term Frequency-Inverse

Document Frequency (TF-IDF) or Latent Dirichlet Allocation (LDA), the

dimensionsofawordthathasbeenvectorizedcorrespondtoallofthewordsthat

areinthecorpus.

Thismeansthatthemachinewillobservethecooccurrenceofeachwordthrough

theuseofthe“bag-of-words”naturallanguageprocessingtechnique(Turneyand

Pantel,2010)andconvertwords to theircorrespondingvectorspacedimension.

More precisely, VSM would add all the words that are in our corpus of 16000

projects toadictionaryandgive the cooccurrencematrixof all thewordsof the

matrix with one another (cooccurrence value between -1 and 1). The method

computes the maximum average log probabilities of each pair of co-occurring

words.Forexample,theword“jazz”wouldbevectorizedinadimensionequalto

theamountofwords thatexist in thecreateddictionary.Although thishasbeen

used quite a lot in the past (Turney and Pantel, 2010), the traditional approach

requires large computer processing power and does not necessarily render the

mostappropriateresults.Indeed,vectorizingawordwithrespecttoallthewords

thatoccurinacorpusmayproducecooccurrencesthatarenotlogicallysignificant.

Forinstance,theword“jazz”mayoccurfrequentlywiththeword“computer”but

this cooccurrence does not teach us much. Word2Vec and Doc2Vec attempt to

solvebothoftheseissuesandoffersaddedbenefits.

In Word2Vec and Doc2Vec, the “bag-of-words” approach is also used but it is

incorporatedwithin the framework. Contrary to the traditional VSM approaches

thatruncooccurrencemeasuresofallwordswitheachother,W2Vonlycompares

cooccurrenceofwordswiththewordsthatprecede(T-1,2,3...)andfollow(T+1,2,

3…)theminalltheirappearancesacrossthecorpus.Thus,thecomputationisno

longerbasedonallwordsbyrather itbecomesbasedonawindowofwords(or

trainingcontext)precedingand trailing thecentralwordonwhichoneobserves

themaximumaveragelogprobabilitiesofeachpairofco-occurringwords.

!"#$

%&'(#)))))))))))'*+,-.#/+&))))))+(#'(#

!"#01$

!"#02$

!"#32$

!"#31$

Figure 1: The Skip-gram model architecture. The training objective is to learn word vector representationsthat are good at predicting the nearby words.

In this paper we present several extensions of the original Skip-gram model. We show that sub-sampling of frequent words during training results in a significant speedup (around 2x - 10x), andimproves accuracy of the representations of less frequent words. In addition, we present a simpli-fied variant of Noise Contrastive Estimation (NCE) [4] for training the Skip-grammodel that resultsin faster training and better vector representations for frequent words, compared to more complexhierarchical softmax that was used in the prior work [8].

Word representations are limited by their inability to represent idiomatic phrases that are not com-positions of the individual words. For example, “Boston Globe” is a newspaper, and so it is not anatural combination of the meanings of “Boston” and “Globe”. Therefore, using vectors to repre-sent the whole phrases makes the Skip-gram model considerably more expressive. Other techniquesthat aim to represent meaning of sentences by composing the word vectors, such as the recursiveautoencoders [15], would also benefit from using phrase vectors instead of the word vectors.

The extension from word based to phrase based models is relatively simple. First we identify a largenumber of phrases using a data-driven approach, and then we treat the phrases as individual tokensduring the training. To evaluate the quality of the phrase vectors, we developed a test set of analogi-cal reasoning tasks that contains both words and phrases. A typical analogy pair from our test set is“Montreal”:“Montreal Canadiens”::“Toronto”:“TorontoMaple Leafs”. It is considered to have beenanswered correctly if the nearest representation to vec(“Montreal Canadiens”) - vec(“Montreal”) +vec(“Toronto”) is vec(“Toronto Maple Leafs”).

Finally, we describe another interesting property of the Skip-gram model. We found that simplevector addition can often produce meaningful results. For example, vec(“Russia”) + vec(“river”) isclose to vec(“Volga River”), and vec(“Germany”) + vec(“capital”) is close to vec(“Berlin”). Thiscompositionality suggests that a non-obvious degree of language understanding can be obtained byusing basic mathematical operations on the word vector representations.

2 The Skip-gram Model

The training objective of the Skip-gram model is to find word representations that are useful forpredicting the surrounding words in a sentence or a document. More formally, given a sequence oftraining wordsw1, w2, w3, . . . , wT , the objective of the Skip-grammodel is to maximize the averagelog probability

1

T

T!

t=1

!

−c≤j≤c,j ̸=0

log p(wt+j |wt) (1)

where c is the size of the training context (which can be a function of the center word wt). Largerc results in more training examples and thus can lead to a higher accuracy, at the expense of the

2

Page 15: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

WhereT is theamountof trainingwords,w is theword, andc is the sizeof the

trainingcontextwhichweherechoosetobeafunctionofthecenteredwordwt,i.e.

thewindow previously discussed. Formore information on themethod refer to

(Mikolovetal.,2013).

The advantage is to observe theword in a context and see, for example, all the

wordsthatdirectlyprecedeorfollowtheword“jazz”.InWord2VecandDoc2Vec,

this context is a user modifiable window. The user then also decides of the

dimensionalitythattheywanttogiveeachword,thoughmoststudiesagreetouse

above100dimensions tohave consistent results but nomore than300because

cooccurrencesareonlymarginallyimprovedafterthisthreshold(Řehůřek,2011).

Forthesereasons,wechoseadimensionalityof300andawindowof10(Řehůřek,

2011).ThusthemachinelooksforallthecooccurrencesofwordTwithwordsT-

1/10andT+1/10untilreachingatmost300words.

Word2VecandDoc2Vecgreatly increasecomputationalspeed,allowforahigher

context-specificity, and provide other interesting benefits. Using Doc2Vec, a

derivative ofWord2Vec, allows us to 1) capture proximity ofwordswithwords

like any other traditional VSM, but also allows to 2) vectorize and capture

proximityofwholesentences(i.e.:crowdfundingprojectdescriptions)withother

sentences,andmostinterestinglyforus3)comparewholesentenceswithspecific

words.Forexample,weareabletoseetheproximityoftheword“jazz”withthe

16000 music projects. These improvements over traditional VSM approaches

explain why startups in the field of natural language processing have adopted

Word2VecandDoc2Vecandbyextensionwhywechosetouseit.

4.2.3. D2VappliedtotheIndiegogodata

We will explain the natural language processing technique known as “bag-of-

words” as well as the application of the Doc2Vector analysis performed on the

16,000Indiegogoprojectsrelatedtomusic.

We first had to merge all of our 16,000 descriptions into one single .csv file.

Effectively,thisfilebecomesourcorpus,andeachcellofthecsvisaninstanceofa

music project description.Unlike the traditionalVSMapproachdescribed above,

wedo not remove anywords other thanpunctuation andmake all letters small

Page 16: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Caps. Indeed, given that Word2Vec and Doc2Vec contextualize words, it seems

importanttokeepallthewordsofeachdescription.

WethenapplyDoc2Veconthecorpusandcreatethesentencevectors.Inessence,

wevectorizeeachwordfromeachprojectdescriptionaswellaseachproject.Once

again, to follow what is done in the field, we choose a window of 10 and a

dimensionalityof300(Řehůřek,2011).

Wesubsequentlybuildadictionaryofmusicalgenresindependentofourcorpus.

The point of building an independent dictionary is important as it allows us to

ensure that it is in no way influenced by the corpus itself, thus limiting

endogeneity. We use several online sources that map musical genres (Lamere,

2012) to create our musical genre dictionary. The musical map was designed

duringthe“RethinkMusicHacker’sWeekend”weremusichackersgettogetherto

“not just rethinkmusic, but to rebuild it” (Lamere, 2012). The author, or rather

hacker,designedamapthatincludesover1000musicgenresgeneratedthrougha

similarity indexing “based upon artist overlap” (Lamere, 2012) of terms used to

describe theirmusic genre. These terms come fromTheEchoNest1, theworld’s

leading music intelligence company, used by companies like Sirius, Spotify, or

Twitter to build better musical content for customers. From these sources, we

build a dictionary of 68 most widely known musical genres. Each genre is

composedofatleastthenameofthemusicalgenre(e.g.:Motown)andmayhave

additionalsubgenresorwordsthatareverycloselyrelatedto it (e.g.:soul,RnB).

For a complete list of themusical genres and associated words, please refer to

appendixA.Ofthese68musicalgenres,18donotoccuronceinourcorpusleaving

uswith50musicalgenres.

In the following step, we take each word of a musical genre and measure the

vectorial proximity to eachproject descriptionwith theD2V sentence andword

method.Wethenreporttheaverageproximityofeachpitchtoeachmusicalgenre.

Letusillustrate.Forexample,inourdictionary,themusicalgenre“Punk”ismade

upofthewords“punk”,“poppunk”,“emo”,“screamo”,“posthardcore”,“emocore”.

1http://the.echonest.com/company/

Page 17: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Wethenseetheproximityofeachpitchtoeachoftheserelatedwordsandreport

theaverageof all proximitymeasuresas themeasureofproximityof eachpitch

withthe“Punk”musicgenre.Thismeasureofproximitywillserveasourmeasure

of proximity of a project to a musical category and we use it to create the

independentvariableofthethirdhypothesis(i.e.:“CategoryProximity”discussed

inbelowintheMethodssection).

Wecreateasecondvariablethatmeasuresaproject’sclaimedmembershiplevelto

amusicalcategorybycountingtheamountoftimesthewordsofamusiccategory

areused inaprojectdescription.Forexampleproject25usesonceaword from

themusiccategory“classical”and11timesacombinationofwordsfromthemusic

category “opera”. These will allow us to create the independent variable

reportGenre(discussedintheMethodssectionbelow)

Finally,wecreateadatabasebymergingthetwoproxymeasuresofproximitytoa

musical genrewith all information collected from Indiegogo (project name, date

end,amountgoal,category,amountraised,externalwebsites).Thisdatabaseisthe

one thatwe use for our statistical analysis thatwewill explain in the following

section.

5 Methods

5.1 Dependentvariable

Tobeconsistentwithpriorresearchincrowdfunding,weuseadummyvariableas

the dependent variable to measure success of a venture on the crowdfunding

platform.Wedefinesuccessofacrowdfundedprojectas1ifithasachievedatleast

100%ofitsfundinggoal,0otherwise.Thismeasurehasbeenusedinseveralother

studies working on crowdfunding platforms (Belleflamme et al., 2015; Mollick,

2014).WedorealizethatIndiegogoisparticularbecauseofthreedifferentfunding

decisions (Flexible Funding, Fixed Funding, and InDemand) may influence this

variableandweincludeitasacontrolvariable(discussedlater).

5.2 Independentvariables

ClaimedCategoricalMembership (reportedGenre).We use a dummy variable that

we call reportedGenre in order to measure category membership claim. This

Page 18: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

variable reports whether a project has reported at least one musical category

membership(asdiscussedin4.2.2.)givingitavalueof1ifitreported,0otherwise.

This dummy is used in hypothesis 1 to measure whether expressing category

membershipdeterminessuccess.

Amount of Categorical Memberships (genreCount, genreCount2). To measure the

impact of the amount of categorical membership, we first use a continuous

variablethatwecallgenreCount,whichcountstheamountofcategoriesaproject

claimsmembershipto.Wenoticethattheaverageamountofgenresreportedbya

project is1.75withaminimumof0andamaximumof22.Then,measuring the

curvilinear relationship between success and the intensity of categorical

membership is fairly straightforward. We use the squared term of the variable

genreCount,whichwecallgenreCount2.Weusethisvariablewithhypothesis2.

GlobalCategoryProximity(globalProximityandglobalProximity2).Tomeasurethe

ideaofoptimaldistinctivenesswillbedonebyseeingthecurvilinearrelationship

thatexistsbetweenaprojectandtheproximitythatithastothemusicalgenresit

claimsmembership to.We first create the variableglobalProximity,which is the

weighted normalized average proximity of a project to any project it claims

membership to (i.e.: where reportedGenre equals 1). We then create

globalProximity2 that is globalProximity squared and use these variables in our

probitregression.Weusebothvariablesinourthirdhypothesis.

5.3 Controlvariables

We control for various variables that have been used in previous studies as

standardcontrols(Belleflammeetal.,2015;Mollick,2014).Theseare:

Project Financing (financeDum).We control for the type of financing that the

projectdecidestousewiththedummyvariable financeDum.Firstly,wedropthe

financing decision InDemand because this concerns projects that have already

beensuccessfulbutare still asking for fundsafter theendof the campaign.This

categorywasrecentlyintroduced(January2015)forprojectsthatweresuccessful

anddonotserveourpurposehereandcouldpotentiallybiasourresults.Wethus

drop 39 observations. We then control for the two remaining financing types:

FlexibleFundingandFixedFunding.Intheformer,theprojectholderswillgetany

amount ofmoney collected regardless ofwhether they reach themonetary goal

Page 19: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

thattheyhaveset.Inthelatter,projectholderswillonlyreceivethefundingifthey

reachorexceedthemonetarygoalsetatthebeginningoftheircampaign.

ProjectDuration(duration).Thelengthoftime(days)aprojectranfor.Indiegogo

recommendsthatprojectholderstohaveprojectsrunningatleast30daysbutno

longerthan60days.

Activity(activity).Theamountoftimes(integer)theprojectholderupdatedtheir

projectduringthecampaign.Moreactiveprojectsaremorelikelytosucceed.

URLLinks(logLinks).Thelogamountoftheexternalwebsitelinksprovidedbythe

projectholder.Morelinksmayshowthattheprojectismoreactive.

Comments (comments). The amount of comments (integer) funders gave on the

project’s dedicated thumbnail. More comments may show that the project is

receivingattention.

Photos(photos).Adummyvariablethatreportswhetheraprojectprovidesphotos

inthededicatedphotothumbnail.

Videos(videoDum). Adummyvariable that reportswhetheraprojectprovidesa

video.HavingavideoishighlyencouragedbyIndiegogoandhasbeenshowedto

haveastrongpositiveimpactonprojectsuccess.

PitchLength(lengthLog).Theloglengthofthepitchdescription.Pastresearchhas

explainedthatthelengthofapitchisimportantasiftheyaretooshort,itwillbe

hardforfunderstodeterminethequalityandiftheyaretoolong,theinformation

maybecomeburdensome.

ProjectGoal(goal).Thelogdollargoalsetbythefounder(integer).

Rewards (rewards). A dummy variable reporting whether a project provides

rewardstofunders.Notprovidinganyrewardsmaydeterfundersfromfinancinga

project.

Sponsor(sponsor).Adummyvariablereportingwhetheraprojecthasacompany

orassociationasasupporter.Thepartnershipisclearlymentionedontheproject’s

pagewith a link to the partner’s profile. On the sponsor’s page, one can see the

projectsthatthispartnerhashelpedfundedorsponsored.Itisfairtosuggestthat

thismayinfluencefunders’willingnesstofinance.

Verified (verified). A dummy variable that reports whether a project has been

verifiedasanon-profitorganization.Asseenintheliteraturereview,pastresearch

hasshownthatbeinganon-profithasapositiveeffectonsuccess.

Page 20: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Category Membership Dummy. Dummy variables that control for a project’s

categorical belonging.Wedonot report these variables to save space given that

theyare43ofthem.

CategoryProximity.Variablesthatcontrolforprojects’proximitytoallcategories.

Wedonotreportthesevariablestosavespacegiventhattheyare43ofthem.

YearDummy. A seriesof yeardummies to control for temporal variation.Wedo

notreportthesevariablestosavespace.

5.4 Model

We use a probit regression on our dependent variable success to test our

hypotheses.

5.5 Results

Table 1 presents the correlation results of our dependent variables (success)

independent variables (reportGenre, genreCount, globalProximity), and control

variables (financeDum,duration,activity, comments,photos, videoDum, lengthLog,

goal,rewards,sponsor,verified).

[InsertTable1]

We observe that most variables are positively correlated to our dependent

variable except for financing choice, project duration who are negatively

correlated. This is consistent with prior research (Belleflamme et al., 2015).

Correlationsarelowwithourmostofourcontrols;commentsandactivityhavethe

highest correlations with 0.23 and 0.19 respectively. There is a relatively high

correlationbetweengenreCount andglobalProximitywhich is normal because in

order to create the latter, we see the global proximity of a project to all the

categoriesitbelongsto.Otherinterestingfairlyhighcorrelationsareofthelength

of the project with our independent variables. This does not seem odd as the

longer a project description is, the more details a project holder can disclose.

Overall, the correlations that we observe are consistent with what we would

expectfromtheliterature.

Page 21: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Table2showstheresultsofourfirsthypothesisbyusingthedummyreportGenre.

Wefirstrunthemodelwithoutanyofourexplanatoryvariables(model1).Wecan

seethatjustlikeinpreviousstudies,thelevelofactivity,usingavideo,pickingAll-

or-NothingfinancingasopposedtoKeep-it-All,theamountofcomments,andthe

durationoftheprojectsallpositivelyinfluenceaproject’soutcome(Belleflammeet

al., 2015). In the secondmodel,we see that reportingagenrehasapositivebut

insignificantimpactonprojectsuccess,thusnotsupportinghypothesis1.

[InsertTable2]

Table 3 pursues our second hypothesis inwhichwe expect thatmentioning too

many categorybelongingswill cause aproject tobe confusing and thuswarrant

lessbacking,thussuggestinganegativerelationship.Werunasimilarregression

asabovewherethefirstmodelonlyshowsthecontrols. Inthesecondmodel,we

addthenumberofcategoriesaprojectmentionstotestourhypothesis.Finally,in

thethirdmodeladdsthesquaredtermofthisexplanatoryvariabletoseewhether

the curvilinear relationship we expect from theory appears. We find that

genreCount2 significantly (at the 10% level) and negatively influences project

successthusprovidingsupporttoourthirdhypothesis.

[InsertTable3]

We finally hypothesized that projects would be more successful if they are

simultaneously close to the category that they claimmembership to but not too

close as to lose all distinctivenesswith regards othermembers of this category.

Table4showssupportforthishypothesisasweseethatourindependentvariable

has a coefficient that is positive, while its square is clearly negative. Both

coefficients are significant at the 1% level, thus providing support to our third

hypothesis.

[InsertTable4]

Page 22: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

6 Conclusion

New ventures have a hard time securing financing. Entrepreneurship scholars

have explored this liability of newness and have found that new ventures often

have difficulty showing the quality of their venture (Stinchcombe, 1965).

Crowdfundingprojectsarealsovictimof thesamesituationandonecouldargue

thatitisactuallyaccentuatedbythefactthatentrepreneursandfinancersdonot

directly communicate to each other and are potentially scattered all across the

globe.Pastresearchincrowdfundinghasattemptedtoexplainthedeterminantsof

successofprojectspresentedontheseplatforms(Agrawaletal.,2011;Colomboetal.,

2014;FreedmanandJin,2014;Mollick,2014;Zvilichovskyetal.,2013).Wedecidetotakea

novelapproach,slightlydeviatingfromsignalingbyobservinghowcategorization

influences crowdfunding success.When ventures self-categorize themselves into

specific groups, they allow investors to better comprehend what the venture

actuallydoeswhichiscognitivelyimportantinordertoreceivefinancing(Brewer,

1991).

Ourmain findingsprovide support to the contention that crowdfundingprojects

are more successful if they clearly manifest their categorical membership

(hypothesis 1). This is consistentwith categorization theory and belonging to a

categorywillmake funders comprehend projectsmore easily and thus increase

theirprobabilityoffunding(Brewer,1991).However,wealsonotethatifaproject

holderclaimsmembershiptotoomanycategories,itwillnotfavortheirchancesof

success.Doingsoonlyconfusesfundersandtheyeventuallyarenotabletograsp

whattheventuredoesthusdecreasingthechancesofaprojectreceivingfunding

(hypothesis2)(Zuckerman,1999).

Furthermore, crowdfunding projects must not only focus on the amount of

categoriestheybelongtobutalsobeawareofthelevelofdistinctivenessthatthey

havewithin their categories. Indeed, we find support for the idea that within a

givencategory, it isbettertobeneithertoosimilarnortoodifferentfromothers

withinacategory(hypothesis3).Theorysupportsthiscontention(Brewer,1991;

DeClercqandVoronov,2009;NavisandGlynn,2011)becauseontheonehand,ifa

projectthatistoofarfromacategory,fundersmaynotunderstandthattheproject

belongstoagivencategoryorwillfindthatitisfartoodistinctfromthenormsof

Page 23: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

this category. On the other hand being too close to a category runs the risk of

losingone’sindividuationoruniquenessandinvestorswillnotseehowtheproject

isanydifferenttowhatalreadyexists.

Thepaperisnotimmunetocertainlimitationsthatweexplorehereafter.First,we

believethatdatawerecollected in themostcomprehensivewaypossiblebutwe

cannot guarantee that there is no selection bias. Second, themethodsweuse to

evaluateprojectandmusiccategoryproximityhave,asfarasweareaware,rarely

ifnotneverbeenusedinmanagementresearch.Thisisalimitationbecauseithas

yet tobeseenasa reliablemethod.However, the fact that thecomputerscience

research field and practitioners in the field of data mining and computer deep

learningareusingthesemethodsmakesusconfidentastoitsvalidity.

Thispapercontributesbothtotheoryandpractice.First,ourresultscontributeto

thegrowingfieldofcrowdfundingbyprovidingalternativeconstructsdetermining

crowdfunding success. Second, we advance the field of management studies by

using novel textual analysis that goes beyond the traditional word count and

spelling mistake. We believe that much research could benefit from such

alternative methods. Third, we contribute to the categorization literature by

measuring optimal distinctiveness ofmore than 16000 projects spanning across

43musiccategories.

Finally,weprovideinterestinginsightstoentrepreneurswhowanttofinancetheir

projects. Indeed,theyshouldfirstpayparticularattentiontoclearlyexplaintheir

raisond’êtrebyself-categorizinginthecategorythatcorrespondstothem,second

be aware that if they span multiple categories it will eventually confuse their

potentialfunders,andthirdmakesuretostandoutwithregardsothermembersof

the(few)categoriestheybelongto.

Page 24: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

7 References:

Agrawal A, Catalini C, Goldfarb A. 2011. THE GEOGRAPHY OF CROWDFUNDING.

Working Paper 16820. Available at:

http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:No+Title#0

.

AgrawalA,CataliniC,GoldfarbA.2014.Somesimpleeconomicsofcrowdfunding.

Innovaion Policy and the Economy 14(April): 1–46. Available at:

http://www.nber.org/papers/w19133.

AhlersG,CummingD.2012.Signalinginequitycrowdfunding.AvailableatSSRN :

49.Availableat:http://www.ssrn.com/abstract=2161587.

Belleflamme P, Omrani N, Peitz M. 2015. The economics of crowdfunding

platforms.InformationEconomicsandPolicy33:11–28.

BradfordC.2012.Crowdfundingandthefederalsecuritieslaws.Colum.Bus.L.Rev.

2012:150.Availableat:http://ssrn.com/abstract=1916184.

BrewerM.1991.THESOCIALSELF -ONBEINGTHESAMEANDDIFFERENTAT

THESAMETIME.PersonalityandSocialPsychologyBulletin17(5):475–482.

DeBuysereK, GajdaO,KleverlaanR,MaromD. 2012.Aframework foreuropean

crowdfunding. www.crowdfundingframework.eu A FRAMEWORK FOR

EUROPEAN. Available at:

https://d21buns5ku92am.cloudfront.net/26522/documents/17930-

1351284179-FRAMEWORK_EU_CROWDFUNDING.pdf.

De Clercq D, Voronov M. 2009. The Role of Domination in Newcomers’

Legitimation as Entrepreneurs. Organization 16(6): 799–827. Available at:

http://org.sagepub.com/content/16/6/799.abstract.

ColomboMG, FranzoniC,Rossi-LamastraC. 2014. Internal social capital and the

attraction of early contributions in crowdfunding. Entrepreneurship: Theory

andPractice:75–101.

DurandR,JourdanJ.2012.JulesorJim:Alternativeconformitytominoritylogics.

AcademyofManagementJournal55(6):1295–1315.

FreedmanS, JinGZ. 2014.The signaling value of online social networks: lessons

from peer-to-peer lending. Available at

http://kuafu.umd.edu/~ginger/research/freedman_jin_networks_submitted_01

2014.pdf:1–53.Availableat:http://www.nber.org/papers/w19820.

Page 25: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Griffin Z. 2012. CROWDFUNDING: FLEECING THE AMERICAN MASSES. Case

WesternReserveJournalofLaw,Technology&…2(2011):1–46.Availableat:

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2030001.

Hazen TL. 2011. Crowdfunding or Fraudfunding ? Social Networks , and the

SecuritiesLaws–WhyanySpeciallyTailoredExemptionShouldbeConditioned

onMeaningful Disclosure as well as for charitable solicitations . 2 These and

similarfundraisingendeavorsareknownas,17.

Hemer J. 2011. A snapshot on crowdfunding. Available at:

http://www.econstor.eu/handle/10419/52302.

Hildebrand T, Puri M, Rocholl J. 2013. Adverse incentives in crowdfunding.

Available at SSRN … (April). Available at:

https://www.funginstitute.berkeley.edu/sites/default/files/Rocholl_Hildebra

nd_Puri_2013-04-15_AdverseIncentivesInCrowdfunding.pdf.

LambertT,SchwienbacherA.2010.Anempiricalanalysisofcrowdfunding.Social

Science Research Network : 1–23. Available at:

http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:An+Empiri

cal+Analysis+of+Crowdfunding#0.

Lamere P. 2012. Map of Music Style. Available at:

http://musicmachinery.com/2012/04/22/map-of-music-styles/.

Lee SH, DeWester D, Park SR. 2008. Web 2.0 and opportunities for small

businesses.ServiceBusiness2(4):335–345.

ManningC,RaghavanP,SchützeH.2009.AnIntroductiontoInformationRetrieval

InformationRetrieval.CambridgeUP.

MartensML, JenningsJE, JenningsPD.2007.Dothestoriestheytellgetthemthe

money they need? The role of entrepreneurial narratives in resource

acquisition.AcademyofManagementJournal50(5):1107–1132.

Massolution. 2012. Crowdfunding Industry Report (May). Available at:

http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Crowdfund

ing+industry+report#2.

Massolution. 2013. 2013CF The Crowdfunding Industry Report. Available at:

http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:No+Title#0

.

MikolovT,ChenK,CorradoG,DeanJ.2013.DistributedRepresentationsofWords

andPhrasesandtheirCompositionality.Nips:1–9.

Page 26: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

MollickE.2014.Thedynamicsofcrowdfunding:Anexploratorystudy. Journalof

Business Venturing. The Author 29(1): 1–16. Available at:

http://linkinghub.elsevier.com/retrieve/pii/S088390261300058X.

MontautiM,WezelFC.2014.CategoryRecombinationandEntrepreneurialInertia

in the Market for Electronic Music, 1978-2011. InAcademy ofManagement

Proceedings,163.

Navis C, Glynn MA. 2010. How New Market Categories Emerge. Administrative

Science Quarterly 55(3): 439–471. Available at:

http://asq.sagepub.com/lookup/doi/10.2189/asqu.2010.55.3.439.

Navis C, GlynnMANN. 2011. Legitimate Distinctiveness and the Entrepreneurial

Identity :InfluenceonInvestorJudgmentsofNewVenturePlausibility36(3):

479–499.

NiedererP.2014.HowSmallBusinessesCan !EmbraceCrowdfunding !forSuccess !

Řehůřek R. 2011. Subspace Tracking for Latent Semantic Analysis. In ECIR’11

Proceedings of the 33rd European conference on Advances in information

retrieval: 289–300. Available at:

http://link.springer.com/chapter/10.1007/978-3-642-20161-

5_29\nhttp://link.springer.com/10.1007/978-3-642-20161-5_29.

Schwienbacher A, Larralde B. 2010. CROWDFUNDING OF SMALL

ENTREPRENEURIALVENTURES,2010.

Shane S, Cable D. 2002. Network ties, reputation, and the financing of new

ventures. Management Science. Available at:

http://pubsonline.informs.org/doi/abs/10.1287/mnsc.48.3.364.7731.

Stinchcombe AL. 1965. Social structure and organisations. In Handbook of

Organizations:142–193.

Turney PD, Pantel P. 2010. From frequency tomeaning: Vector spacemodels of

semantics.JournalofArtificialIntelligenceResearch37:141–188.

Überbacher F. 2014. Legitimation of New Ventures: A Review and Research

Programme.JournalofManagementStudies51(4):667–698.

Vergne J-P, Wry T. 2014. Categorizing Categorization Research: Review,

Integration,andFutureDirections.JournalofManagementStudies51(1):56–

94.Availableat:http://doi.wiley.com/10.1111/joms.12044.

VergneJP.2012.Stigmatizedcategoriesandpublicdisapprovaloforganizations:A

mixed-methods study of the global arms industry, 1996-2007. Academy of

Page 27: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

ManagementJournal55(5):1027–1052.

DeWittN.2012.AKickstarter’sGuidetoKickstarter-Howtosuccessfullyfundyour

creativeproject.

Zuckerman E. 1999. The Categorical Imperative Securities Analysts and the

IllegitimacyDiscount.AmericanJournalofSociology.

ZvilichovskyD,InbarY,BarzilayO.2013.PLAYINGBOTHSIDESOFTHEMARKET :

SUCCESSANDRECIPROCITYONCROWDFUNDING.

8 TablesTable1

Table2Duetoitslargehorizontalsize,thefulltableisnotpresentedherebutcanbeviewedatthefollowinglink:http://tinyurl.com/zksh7fc

variable abbreviation task operationalizationsuccess success DV binomial:hasprojectreached100%ofgoal?ClaimedCategoricalMembership reportedGenre IV binomial:hasprojectreportedatleast1musicalgenre?AmountofCategoricalMemberships genreCount IV int:countscategoriesaprojectclaimsmembershiptoGlobalCategoryProximity globalProximityIV weightednormalizedavg.proximityofprojectProjectFinancing financeDum C binomial-0:All-or-Nothing,1:Keep-it-AllProjectDuration duration C int:projectlengthindaysActivity activity C int:amountofupdatesbyprojectholderURLLinks logLinks C logamountoftheproject'sexternalwebsitelinksprovidedComments comments C int:amountofcommentsprojectreceivedPhotos photos C binomial:doesprojectusephotos?Videos videoDum C binomial:doesprojectusevideos?PitchLength lengthLog C loglengthofthepitchdescriptionRewards rewards C binomial:doesprojectproviderewards?Sponsor sponsor C binomial:isaprojectsponsored(inapartneship)?Verified verified C binomial:isaprojectaverifiednon-profit?CategoryMembershipDummy C vectorsofcategorymembership-controlsforcategoryCategoryProximity C vectorsofcategoryproximity-controlsforproximityYearDummy C vectorsofyears-controlsforyearfixedeffectDV:dependentvariable;IV:independentvariable;C:control

Page 28: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Table3

Table4

VARIABLES (1) (2)reportedGenre 0.09**

(0.04)activity 0.04*** 0.04***

(0.00) (0.00)photos -0.05* -0.05*

(0.03) (0.03)videoDum 0.28*** 0.28***

(0.03) (0.03)financeDum -0.72*** -0.72***

(0.06) (0.06)lengthLog 0.16*** 0.15***

(0.02) (0.02)logLinks 0.05** 0.05**

(0.02) (0.02)duration -0.01*** -0.01***

(0.00) (0.00)sponsor 0.21** 0.22**

(0.10) (0.10)verified -0.04 -0.03

(0.06) (0.06)rewards 0.46*** 0.46***

(0.06) (0.06)Constant -1.07*** -1.06***

(0.20) (0.20)

Observations 15,474 15,474Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

VARIABLES (1) (2)reportedGenre 0.09**

(0.04)activity 0.04*** 0.04***

(0.00) (0.00)photos -0.05* -0.05*

(0.03) (0.03)videoDum 0.28*** 0.28***

(0.03) (0.03)financeDum -0.72*** -0.72***

(0.06) (0.06)lengthLog 0.16*** 0.15***

(0.02) (0.02)logLinks 0.05** 0.05**

(0.02) (0.02)duration -0.01*** -0.01***

(0.00) (0.00)sponsor 0.21** 0.22**

(0.10) (0.10)verified -0.04 -0.03

(0.06) (0.06)rewards 0.46*** 0.46***

(0.06) (0.06)Constant -1.07*** -1.06***

(0.20) (0.20)

Observations 15,474 15,474Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Page 29: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

Table5

VARIABLES (1) (2) (3)

globalProximity 0.02 1.07***(0.10) (0.38)

globalProximity2 -1.52***(0.53)

activity 0.03* 0.03* 0.01(0.01) (0.01) (0.02)

photos 0.04*** 0.04*** 0.04***(0.00) (0.00) (0.00)

videoDum -0.01*** -0.01*** -0.01***(0.00) (0.00) (0.00)

financeDum 0.26*** 0.26*** 0.26***(0.03) (0.03) (0.03)

lengthLog -0.72*** -0.72*** -0.72***(0.06) (0.06) (0.06)

logLinks 0.12*** 0.11*** 0.11***(0.03) (0.03) (0.03)

duration -0.01*** -0.01*** -0.01***(0.00) (0.00) (0.00)

sponsor 0.05** 0.05** 0.05**(0.02) (0.02) (0.02)

verified -0.06 -0.06 -0.07(0.06) (0.06) (0.06)

rewards 0.39*** 0.39*** 0.39***(0.06) (0.06) (0.06)

Constant -0.67* -0.66* -0.88**(0.37) (0.37) (0.38)

Observations 15,474 15,474 15,474Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Page 30: Categorization and success of crowdfunding projectsunice.fr/laboratoires/gredeg/contenus-riches/... · success of crowdfunding projects (Mollick, 2014). Researchers have taken different

30