CROWDFUNDING CREATIVE IDEAS: THE DYNAMICS OF PROJECT BACKERS IN KICKSTARTER

CROWDFUNDING CREATIVE IDEAS: THE DYNAMICS OF PROJECT BACKERS IN KICKSTARTER Venkat Kuppuswamy Barry L. Bayus Kenan-Flagler Business School Kenan-Flagler Business School University of North Carolina University of North Carolina CB3490; McColl 4608 CB3490; McColl 4521 Chapel Hill, NC 27599 Chapel Hill, NC 27599 venkat@unc.edu Barry_Bayus@UNC.edu ABSTRACT Entrepreneurs are turning to crowdfunding as a way to finance their creative ideas. Crowdfunding involves relatively small contributions of many consumer-investors over a fixed time period (generally a few weeks). The purpose of this paper is to add to our empirical understanding of backer dynamics over the project funding cycle. Two years of publicly available data on projects listed on Kickstarter is used to establish that the typical pattern of project support is U-shaped in general, backers are more likely to contribute to a project in the first and last week as compared to the middle period of the funding cycle. We further establish that this U-shape pattern of support is pervasive across projects, including both successfully and unsuccessfully funded projects, those with large and small goals, and projects in different categories. We then empirically explore the dynamics associated with several factors, including collective attention effects from platform sorting options, the role of family and friends in supporting projects, the effects of social influence, and the role of project updates over the project funding cycle. [Keywords: Innovation, Entrepreneurship, Strategy, Marketing] Acknowledgments: Comments from participants in research workshops at the University of North Carolina at Chapel Hill, Duke, University of Illinois, University of Utah, DRUID 2013, Emory, HEC Paris, and University of California at Berkeley helped to improve this paper. We also thank Atul Nerkar, Page Ouimet, Avi Goldfarb, Tarun Kushwaha, Sri Venkataraman, Isaac Dinner, and Amin Sayedi for helpful comments on an earlier draft. Revised January 29, 2014 Electronic copy available at: http://ssrn.com/abstract=2234765

1. INTRODUCTION An important barrier to innovation is the availability of early-stage funding (Cosh, et al. 2009). Given the difficulties that new ventures face in attracting financing from angel investors, banks and venture capital funds, some entrepreneurs are tapping into large, online communities of consumer-investors (Economist 2010; Schwienbacher and Larralde 2012; Agrawal, et al. 2013b). Called crowdfunding, this relatively new form of informal venture financing allows entrepreneurs to directly appeal to the general public (i.e., the crowd ) for help in getting their innovative ideas off the ground. As defined by Belleflamme, et al. (2013), crowdfunding involves an open call (through the Internet) for the provision of financial resources either in the form of donation or in exchange for some form of reward in order to support initiatives for specific purposes 1. Crowdfunding, which typically involves collecting small amounts of money from a large number of people, is a new label for an activity that has a rich history in many domains (Ordanini, et al. 2011). For example, Mozart and Beethoven financed concerts and new music compositions with money from interested patrons, the Statue of Liberty in New York was funded by small donations from the American and French people, and President Barak Obama s 2008 election campaign raised most of its funds from small donations over the Web (Hemer 2011). Today, several hundred global intermediaries with online platforms exist to match up consumer investors with initiatives that they wish to help fund. Prominent examples in the popular press include the narrative movie project by Steve Taylor that got almost 4,500 people to donate nearly $350,000 and Scott Wilson s idea to create a wristband that will convert an ipod nano into a watch raised over $940,000 from over 13,500 individuals (Adler 2011). One of the largest crowdfunded projects to date is Eric Migicovsky s e-paper Watch that integrates with an Android or iphone even though its original goal was $100,000 the project eventually received contributions totaling over $10.2M in 37 days from well over 65,000 backers. According to one industry report, crowdfunding platforms raised almost $1.5B and successfully funded more than one million projects in 2012 (Massolution 2013). Given the potential dollars involved, 1 This is conceptually similar to crowdsourcing in which community members (non-experts) propose new product and service ideas, as well as comment on and vote for the ideas of others (Bayus 2013). 1 Electronic copy available at: http://ssrn.com/abstract=2234765

crowdfunding has recently garnered attention from policymakers and regulators as evidenced by the Jumpstart Our Business Startups Act (JOBS Act) recently signed into U.S. law (Stemler 2013). Due to the rapidly growing interest in this online form of venture financing, academic research is beginning to explore the drivers behind successfully crowdfunded projects. To date, the majority of empirical studies in this domain focus on identifying the project and entrepreneur characteristics associated with successfully funded projects (e.g., Mollick 2014). Our interest in this paper is to add to this research stream by explicitly considering the dynamics of project support over time. Because most crowdfunding campaigns last for only a few weeks, understanding project-level funding behaviors over time is important as we do not expect contributions to be uniform over the funding cycle. Like in online auction bidding (Ariely and Simonson 2003), contribution decisions by many consumer investors during the project funding cycle suggests that earlier decisions can dynamically impact later behaviors. Furthermore, as suggested by related research in online auction bidding (Ariely and Simonson 2003), the key drivers of contribution decisions may also vary over the beginning, middle and later stages of a crowdfunding campaign. Thus, the purpose of this paper is to add to our empirical understanding of crowdfunding by focusing on the number of project backers added to a project each day over its funding cycle. We refrain from formally developing and testing specific hypotheses because our empirical study is exploratory in nature. We believe this approach is appropriate for a nascent and evolving topic like online crowdfunding as no prior work on backer dynamics exists with which to guide our research. Instead, we hope that our empirical findings can be useful for future theory-building. The setting for our analysis is one of the oldest and largest crowdfunding platforms on the Web. Since its launch in April 2009, Kickstarter has several million community members who have combined to pledge hundreds of millions of dollars to fund creative ideas in categories like art, music, film and video, games, design, and technology (Ricker 2011). Using two years of publicly available information on successfully and unsuccessfully funded Kickstarter projects, we first establish that the typical pattern of project support is not uniform over its funding cycle. In general, backers are more likely to contribute to a project in the first and last week as compared to the middle period, and are much less likely to pledge once a 2 Electronic copy available at: http://ssrn.com/abstract=2234765

project reaches its goal. We further demonstrate that this U-shaped pattern of backer support is pervasive across crowdfunding projects including both successfully and unsuccessfully funded projects, those with large and small goals, and projects in different categories. We then empirically explore the dynamics associated with several factors that have been related to project-level funding success by other researchers, including collective attention effects from platform sorting options, the role of friends and family in supporting projects, the dynamic effects of social influence, and the role of project updates over the project funding cycle. We find no support for the idea that the majority of contributions come in at the beginning and end of a project because projects are most visible then due to the sorting options available on the platform (i.e., Kickstarter projects can be sorted based on whether they were Recently Launched or Ending Soon ). We do however, find that pledges from family members tend to occur in the first week after launch as well as just before the project ends. Moreover, most of the contributors at any point in the funding cycle are one-time backers that likely come from the entrepreneur s own social circle. With respect to the effects of social influence, we find strong evidence consistent with the goal-gradient hypothesis (Hull 1932; Kivetz, et al. 2006) project support generally increases monotonically as it approaches its end goal. For Kickstarter projects, potential backers are not affected by the number of contributors a project has garnered, but instead are influenced by how much of the goal has already been pledged. We also find that additional backer support is positively related to project updates, and updates are more likely to be posted during the first week and last three days as compared to the middle period of the funding cycle. Finally, project creators tend to use updates more aggressively as their project nears its goal. 2. PRIOR CROWDFUNDING RESEARCH As noted by many researchers (e.g., Mollick 2014), there is relatively small but growing literature dealing with crowdfunding. Tomczak and Brem (2013) provide a recent review of the general crowdfunding process, including definitions and its role in the financing of new ventures. In this section, we focus on summarizing the key findings from the empirical crowdfunding literature. In Table 1, we organize the main findings by type of crowdfunding platform based on the different models identified in Massolution (2013). Details for each of the studies noted in Table 1 are in the online Appendix A. 3

In general, crowdfunding platforms differ in terms of whether the contributor s primary motivation for participating is the expectation of a financial return. For example, crowdfunding communities like SellaBand and Wefunder offer consumer investors an interest in the venture in the form of equity or some sort of profit sharing agreement (Ward and Ramachandran 2010; Agrawal, et al. 2013a). Other crowdfunding platforms such as Prosper and Zopa involve peer-to-peer lending in which it is expected that the original principal is repaid, along with some fixed interest (Herzenstein, et al. 2011a; Zhang and Liu 2012). Some crowdfunding communities involve no monetary compensation for participation. For example, JustGiving and Spot.us rely on altruistic motivations in which funders voluntarily donate their money with no expectations of any tangible reward (Burtch, et al. 2013a; Smith, et al. 2012). Finally, project backers in crowdfunding communities like Kickstarter and Indiegogo receive non-financial rewards for their financial contributions (Mollick 2014). These rewards often take the form of tokens of appreciation (thank-you message, artist s autograph, mentioning the crowdfunder s name in the credits, tee-shirt) or the prepurchasing of products or services (Hemer 2011). [insert Table 1 about here] As indicated in Table 1, the empirical crowdfunding literature mainly focuses on the rational motivations behind the contribution decisions of investors, funders, and donors. While we cannot rule out the potential role of emotional motivations for contributions by believers (e.g., family and friends) that simply want to support the fundraiser (or want to feel the warm glow from making a contribution), very few studies take this perspective (Agrawal, et al. 2013a). Given the rational motivations of investors, funders, and donors, most of the existing crowdfunding literature studies the effects of quality signals on fundraising success. For example, researchers show that funding success is significantly related to project quality signals such as preparedness, narrative, and others contribution decisions as well as individual quality signals like personal characteristics, creditworthiness, and social networks (see Table 1). Some research also considers the effects of social norms, finding that the contribution amount for reward and donation-based crowdfunding projects is related to the size of others contributions (see Table 1). 4

Although brief, our review of the empirical crowdfunding literature identifies few studies that use panel data to study the dynamics of project funding behavior (see Appendix A), and none that consider backer dynamics over the project funding cycle. Our research seeks to add to our empirical understanding of crowdfunding by studying the dynamics of project support over the funding cycle. The setting for our study is Kickstarter, a large well-known reward-based crowdfunding platform. We study backer dynamics in a reward-based platform for two primary reasons: (1) qualitative studies document the importance of rewards in motivating participation in crowdfunding communities (de Witt 2012; Gerber, et al. 2012; Steinberg 2012), and (2) reward-based crowdfunding has the largest number of online platforms and is the fastest growing form of crowdfunding (Massolution 2013). 3. DATA In this section, we briefly discuss the empirical context of our study. Based in the U.S., Kickstarter is one of the world s largest crowdfunding platforms. By April 2012, Kickstarter had raised more than $200 million for 20,000 projects, or about 44 percent of those that sought financing on the site (Wortham 2012). According to their website, Kickstarter is focused on creative projects. We're a great way for artists, filmmakers, musicians, designers, writers, illustrators, explorers, curators, performers, and others to bring their projects, events, and dreams to life. Projects are grouped into thirteen broad categories: Art, Comics, Dance, Design, Fashion, Film and Video, Food, Games, Music, Photography, Publishing, Technology, and Theater. The website defines a project as something finite with a clear beginning and end. Someone can be held accountable to the framework of a project a project was either completed or it wasn t and there are definable expectations that everyone can agree to. Kickstarter does not accept projects created to solicit donations to causes, charity projects, or general business expenses. In order to participate, individuals must join the Kickstarter community (at no cost) by selecting an anonymous username. Like most online communities, information on demographics and personal characteristics are not collected (the Kickstarter community is a large, undefined crowd ). Community members can propose projects for funding, back a project by financially contributing (with a credit card via 5

Amazon), and/or comment on projects. Kickstarter projects can only be proposed by U.S. residents (for tax purposes); project contributors have no geographic restrictions. To use Kickstarter, an entrepreneur (called creator in Kickstarter) creates a webpage for the project on the platform explaining the purpose of the project and the specific deliverables that they aim to produce with the contributed funds. Along with an end date for the project funding cycle, the creator also indicates the funding goal of the project, i.e., the amount of money they require to execute the project as specified. Creators can communicate with their supporters by posting public updates that everyone can see. When a potential funder (called backer in Kickstarter) visits an active project s webpage, they are presented with all the project information initially posted by the creator. Moreover, potential backers are shown the current funding status of the project (i.e., the funds raised thus far), the ultimate funding goal, and the number of days remaining until the project funding cycle expires. A potential funder can also see a listing of the other backers that have contributed to the project, as well as the timing of these contributions 2. Specific donation amounts by backers are not publicly shown on the website. To help potential backers discover projects they want to support, Kickstarter has a number of search options. In particular, projects can be sorted based on the first week after their initial launch ( Recently Launched ), last week before the project funding closes ( Ending Soon ), or popularity (based on the number of backers recently added to a project). Occasionally, Kickstarter staff mentions an active project on their blog. There are two important features of Kickstarter that distinguish it from many other smaller crowdfunding platforms, as well as more traditional forms of entrepreneurial finance. The first is the all-ornothing aspect of fundraising on the platform. A project must be fully funded before its funding cycle concludes or no money pledged by any backer is transferred to the project creator. An over-ambitious funding goal may thus result in the fundraising effort falling short and consequently, raising no funds whatsoever. At the same time, once a project has reached its funding goal, it can continue to receive contributions until its deadline. As a result, funded projects can exceed their original funding goal. 2 Shortly after our data collection in March 2012, Kickstarter removed this information in their updated website design. 6

The second important feature of the Kickstarter model is the fact that individuals contributing to a project do not receive equity in the project in return for their funds. Specifically, backers do not receive any financial incentives, returns, or repayment in exchange for their contributions. Instead, project creators typically offer more modest rewards to contributors which vary by the level of contribution. According to the Kickstarter website, the four most common reward types are: (a) copies of the thing (e.g., the actual product, an assembled version of a DIY kit); (b) creative collaborations of various kinds (e.g., a backer might appear as a hero in the comic, or she may be painted into the mural); (c) creative experiences (e.g., a visit to the film set, a phone call from the author, dinner with the cast, a concert in the backer s backyard); and (d) creative mementos (e.g., photos sent from filming location, explicit thanks in the closing credits of the movie, etc.). Data for our study come from publicly available information on the Kickstarter web site. We extracted information on all backed projects posted on the platform from its inception in May 2009 through February 2012. We focus on projects with at least one backer since we are interested in the dynamics of backer behavior (projects with zero backers will not contribute any information to our analyses). To allow a time buffer for community activity around a project to stabilize, projects completed after 2011 are dropped from the analysis. In addition, projects started in 2009 are not used in the analysis because the look and feel of the web site underwent several revisions in the first few months after launch. We restrict our analyses to projects with a duration of at least 21 days to ensure an adequate length of time to examine backer behaviors during the early, middle, and late stages of the funding cycle. After cleaning the data for inaccuracies and incomplete information, two years of daily data on 14,704 projects that began on or after January 1, 2010 and concluded by December 31, 2011 is available for analysis purposes. [insert Table 2 about here] Descriptive statistics for these projects are reported in Table 2. The average project 3 has a goal of just over $9,900 but only receives a little more than $2,100 in pledged contributions 4. Projects tend to last for 3 One project had a goal of over $21M (Kickstarter s limit) to help reduce the national debt (http://www.kickstarter.com/projects/2116548608/help-erase-the-national-debt-of-the-usa?ref=search). This project only had 8 backers who pledged $180. 7

around six weeks; a relatively large proportion of backers support a project in the first or last week of its funding cycle. Almost eighty percent of the projects include a video. The average project offers more than seven reward categories as incentives for their donors and receives about $70 per backer. Creators generally post a couple of project updates. Over ninety percent of creators only propose a single Kickstarter project. [insert Figure 1 about here] There is a considerable amount of variance in the funding outcomes for Kickstarter projects. Figure 1 shows the distribution of project success: projects that reach their funding goal do so by a small margin (almost half of all the successful projects are within ten percent of their original funding goal), while projects that miss their targets do so by a large margin (almost half of all the unsuccessful projects achieved less than ten percent of their goal). [insert Table 3 about here] Descriptive statistics by project funding outcome are reported in Table 3. Although these statistics cannot be used to assess causality, this information is useful to better understand the nature of a reward-based crowdfunding community. While unsuccessful projects have a funding goal more than four times as large as successful projects ($14,686 compared to $3,486), these projects receive well less than half of the amount contributed to successful projects ($1,214 compared to $3,496). Successful projects tend to be shorter in duration. All projects receive a relatively large proportion of their backers in the first week, and successful projects also get a lot of support in the last week of their funding cycle. Not surprisingly, successfully funded projects have significantly more backers than unsuccessful projects, and add more backers each day. Successful projects tend to communicate more to the community and their backers by posting updates. 4. EMPIRICAL ANALYSES In this section, we attempt to shed some empirical light on the dynamics of backer behaviors in reward-based crowdfunding. To do this, we exploit the panel structure of the Kickstarter data to explore the relationship between the daily support a project receives and various explanatory and control variables. The key dependent variable in our analyses is BackersAdded, a count variable which is the number of backers 4 The largest funded project in our sample received a little over $95K. Since our data collection, several projects have received over $1M in funding. 8

project i receives in day t. Because the dependent measure is a non-negative integer, our empirical strategy is to estimate appropriate panel count models (Poisson and Negative Binomial). To account for any unobserved project heterogeneity (e.g., projects may differ in unobserved quality ), we estimate fixed-effects models. Essentially, fixed-effects models incorporate project specific intercept terms. Based on a Hausman type test (see Allison 2005), fixed-effects models are preferred over random effects models for the Kickstarter data. Importantly, a fixed effects model removes any unobserved time-invariant heterogeneity across projects and allows these unobserved differences to be correlated with the independent variables (and thus is less likely to be biased). Although time-invariant characteristics are controlled, estimation of the fixed effects models is accomplished using a conditional maximum likelihood estimator where all time-invariant project effects are conditioned out of the model using an individual s total count (Cameron and Trivedi 2009). Cluster-robust standard errors for the estimated coefficients are used for statistical tests due to dependence among the errors over time within a project. Because count models are non-linear, the magnitude, significance and direction of an interaction effect cannot be directly determined based only on its estimated coefficient (Ai and Norton 2003). In order to properly interpret interaction terms in a non-linear count model, we report the average marginal effects associated with the key explanatory variables (Karaca-Mandic, et al. 2012). However, information on the project specific intercept terms is required to calculate marginal effects. Thus, when considering interaction terms we take a random sample of five hundred projects and estimate an unconditional negative binomial count model with intercepts for each project. Based on simulation results reported by Allison and Waterman (1982), this approach gives good results. Although not shown here, the random sample we use closely matches the characteristics of the full set of projects. We note that other random samples produce the same conclusions as those discussed in this section. 4.1 The Dynamics of Project Support We begin by empirically exploring the dynamics of project support over its funding cycle. The average number of backers added to a project over its relative funding cycle is depicted in Figure 2. Consistent with the descriptive statistics in Tables 2 and 3, projects tend to get a lot of backer support in the 9

first and last weeks of their funding cycle. A high level of initial project support in the first few days is followed by generally decreasing support over most of the first week. A pronounced lull in project activity occurs during the middle period of the funding cycle. As the project approaches its conclusion, there is an increase in contributions. To better understand the dynamics of project funding behaviors, we next turn to an econometric analysis of these data. [insert Figure 2 about here] We define binary variables to capture the first seven days (Day T, where T=1,,7) and last seven days (L LastDay, where L=1,,7) in the project funding cycle. Here the reference category is the middle period. Other analyses not shown here indicate that daily binary variables in the middle period (i.e., days not included in the first or last week) are not significantly different from each other. In addition, several timevarying variables that account for possible effects due to other project or situational factors are included in our analyses. As suggested by Table 3 and Figure 2, several projects in our sample exceed their original funding goal. To account for any differences in backer behaviors for these projects, we include PostFunded, defined to be one for each day a project has already been funded and zero otherwise. We control for competition among projects for backer support by including ActiveProjects (the number of Kickstarter projects across all categories 5 that are accepting pledges on day t in thousands) in our estimations. We also include MaxCompetingBackers (the maximum number of cumulative backers across all competing projects accepting pledges on day t) to control for any possible negative effects due to other projects that are receiving a lot of backer support. Finally, we control for the possibility that pledges concentrate on certain days by including separate dummy variables for day of week and account for any other unobserved time-varying effects by including month-year dummy variables. This framework with the dummy variables for time in the project funding cycle, along with control variables, is the basic econometric model used in most of our analyses. The results of estimating a conditional fixed-effects Poisson model that corrects for over dispersion and allows for cluster-robust standard errors (Wooldridge 1999) is in Table 4. While we do not report the estimation details for day of week and month-year to conserve space, we can make a few observations. First, 5 An alternate measure involving the number of competing projects in the same category as project i gives the same results across all our estimated models as those reported for ActiveProjects. 10

projects are more likely to receive contributions on weekdays compared to weekends, with activity increasing from Sunday to a peak on Wednesday; thereafter, activity decreases to its lowest point on Saturday. Monthyear fixed effects indicate that projects are less likely to add backers as we move from the beginning of the sample period (January 2010) to the end (December 2011). With one exception, the coefficient estimates for MaxCompetingBackers are insignificant in Table 4. These results do not strongly support the idea of a Blockbuster Effect in which a project with a large number of backers steals potential backers from other projects (Kickstarter 2012). Across all the models in Table 4, the coefficient estimate for PostFunded is negative and significant. This indicates that backer support drops off considerably once a project surpasses its goal. From Model 1, we find the effect of ActiveProjects on backer support is negative and significant. This is consistent with the idea of Kickstarter Fatigue as proposed by some industry followers in which potential backers are becoming weary due to an increasing number of projects asking for their financial contributions (Goninon 2013; Maxwell 2013; Nelson 2013). This particularly seems to be the case for projects with high goals (Model 4). Further, the results in Models 2 and 3 indicate that a large number of competing projects is associated with more backer support for projects that are eventually funded and less for unsuccessful projects. Together, these results suggest that there are limits to the financial support of backers. [insert Tables 4 and 5 about here] Strongly confirming a U-shaped pattern of backer support, the coefficient estimates for the Day and LastDay binary variables are generally significant and positive in all the models. Moreover, these variables are significantly decreasing in magnitude over the first week and significantly increasing in magnitude over the last week 6. The positive coefficient estimates indicate that backers are more likely to pledge in the first and last week as compared to the middle period in the project s funding cycle. Deadline effects in which a lot of action occurs as the end of an experience is approached has been widely observed in many contexts (Webb and Weick 1979; Ariely and Simonson 2003). Consistent with the idea of a deadline effect, almost two-thirds of the projects in our sample achieved their target goal in the last week of their funding cycle. More 6 The consistent drop in the coefficient estimate for the very last day comes from the fact that projects end at various times during the last day, i.e., many projects do not have a complete 24 hours of funding time in the last day. 11

persuasive, the significant and increasing coefficient estimates for the LastDay variables strongly suggest a deadline effect. Models 2 and 3 demonstrate that both successfully and unsuccessfully funded projects also exhibit a U-shaped pattern of backer support, as do projects with different goal targets (Models 4 and 5). It is interesting that successfully funded projects exhibit the same dip in activity during the middle period of the funding cycle as projects that do not meet their goal. Thus, even successful projects find it very difficult to maintain their initial momentum in continuing to get pledges over the entire funding cycle. In Table 5 we confirm that this U-shaped pattern of project support is pervasive across different project types (Art, Product Design, Film and Video, Games, Music, Technology). These results extend common thinking that only successfully funded projects exhibit a U-shaped pattern in project support over its funding cycle (de Witt 2012; Steinberg 2012) in fact, this U-shaped pattern is systematic and persistent across Kickstarter projects 7. 4.2 Inside the Dynamics of Project Support In this section, we examine several factors that have been related to project funding success by other researchers. We extend these prior studies by empirically exploring the dynamics associated with these factors. Specifically, we consider four questions: (1) Is the U-shaped pattern of project support due to collective attention effects, i.e., backers simply support projects that are easily found and most visible from using the available platform project sorting options (Qui 2013)? (2) What is the role of family and friends over the project funding cycle (Agrawal, et al. 2013a)? (3) What are the dynamic effects of social influence in supporting a project (Herzenstein, et al. 2011a; Zhang and Liu 2012; Agrawal, et al. 2013a)? (4) What is the role of project updates over the project funding cycle (Mollick 2014)? 4.2.1 Collective Attention Effects One interesting perspective that might account for the U-shaped pattern of backer support comes from research on the effects of consumers limited attention in the digital economy where information is abundant (Falkinger 2008; Wu and Huberman 2007; Hodas and Lerman 2013). The problem of collective attention is at the center of online communities and the spread of ideas a wealth of information creates a 7 Yan Budman, Director of Marketing at Indiegogo, reports a similar pattern of backer behavior for Indiegogo projects (Budman 2012). 12

poverty of attention (Simon 1971: 40). According to Wu and Huberman (2007), there are two key drivers of attention dynamics in large groups. Focusing on the spread of new stories in digg.com, they argue that there is a positive reinforcement effect where attention to a story initially increases because the first few people that like it further pass on this information to others. This growth in attention is offset by a negative effect due to the fact that the novelty of a story decays over time. The dynamics of collective attention associated with these assumptions results in a log-normal curve of activity around a news story. Applying this model to crowdfunding, the dynamics of novelty and attention would generate an initially increasing pattern of backer support followed by a drop in backer contributions a pattern that is not observed in Kickstarter (e.g., see Figure 2). This collective attention framework is extended by Hogas and Lerman (2013) who find little evidence that the novelty of news stories decays over time (i.e., older stories are just as appealing as newer stories). Instead, they argue that people pay more attention to recent stories because they are easy to find and more visible. This idea is consistent with Nelson (2013) who suggests that the majority of pledges to a crowdfunding campaign come at the beginning and end of a project because projects are most visible then. In an online environment, there are often several web site features and sorting options that lower search costs making projects more visible (Bakos 1997). In the case of Kickstarter, projects can be sorted based on whether they were Recently Launched (the first week after a project s initial launch) and Ending Soon (the last week before a project s funding closes). Thus, the collective attention argument is that the significant coefficient estimates for the first and last week daily variables in Tables 4 and 5 are due to the use of these sorting options in Kickstarter potential backers simply support projects that are most visible from using these sorting options. If this were true however, we would also expect that the positive effects of the first and last weeks will be accentuated when there are more potential backers that can use these options to view the projects. To examine this idea, we incorporate information on daily traffic to Kickstarter over time in our basic econometric model (from Quantcast.com, KickstarterTraffic is the number of unique visitors to the Kickstarter web site on day t in 13

hundred thousands) 8. If visibility is a plausible explanation for the U-shaped pattern of project support, the marginal effects of time in the funding cycle should be increasing as traffic to the web site increases. Because potential backers can also sort projects on the Kickstarter website based on popularity, we also include Popular (=1 if a project is ranked in the top fifty of all active projects in terms of backers added over the prior week, 0 otherwise) as a control variable 9. [insert Table 6 and Figure 3 about here] Results from estimating a conditional fixed-effects Poisson model are in Table 6, Model 1. As expected, the coefficient estimate for KickstarterTraffic is positive and significantly related to project support. The results from estimating an unconditional negative binomial model with the random sample 10 of projects is in Models 2 and 3. The estimation results from the random sample in Model 2 are generally consistent with those from the full sample in Model 1; estimation results involving the traffic interaction terms are in Model 3. Because we cannot directly interpret the interaction effects from the coefficient estimates, we do not report these terms in Model 3. So as not to clutter the plots, the average marginal effects of the interactions between KickstarterTraffic and a single day in the first and last week are shown in Figure 3. This same pattern of marginal effects is obtained for the other days in the funding cycle. In general, we see a slight upward but insignificant trend in project support as Kickstarter traffic increases (the difference in marginal effects for the distinct traffic levels is significant at the p<0.05 level for the first day, but insignificant thereafter and during the last week). Thus, the greater project support observed in the first and last week does not seem to be due to higher project visibility associated with the Recently Launched and Ending Soon sorting options available with Kickstarter. 8 Website traffic data is unavailable for 115 dates in our two year sample window. Due to missing data, in our estimations with KickstarterTraffic we exclude 367 projects and 56,189 project-day observations (out of our total sample of 14,704 projects and 653,820 project-day observations). In order to retain as much information as possible, we do not include KickstarterTraffic in all our subsequent models. We note that none of our conclusions change if KickstarterTraffic is included. 9 We also find that a specific mention of project i on day t by the Kickstarter blog has a generally positive effect on added backers. However, we do not incorporate this variable into our analyses since the incidence of this event is extremely small in our random sample. 10 Seven projects were not included in the analysis due to missing traffic data. 14

4.2.2 The Role of Family, Friends, and Followers Even though there are relatively few empirical studies, it is generally acknowledged that financial support from family and friends is an important source of early stage funding for new ventures (Cumming and Johan 2009; Agrawal, et al. 2013a). Agrawal, et al. (2013a) empirically show the importance of friends and family investors in the SellaBand crowdfunding community. This is consistent with the general belief among crowdfunding pundits who argue that successful projects create a critical mass of early funding from the people in their close social circles (de Witt 2012; Steinberg 2012). [insert Table 7 about here] To explore this idea, we examine the timing of Kickstarter pledges from direct family relatives in our smaller random sample of projects. Given the anonymous nature of members in the Kickstarter community, we rely on usernames to construct an indicator variable, Family, for whether project i on day t was supported by a backer that has the same last name as the project creator (0 otherwise). Family is manually coded for creators that have distinguishable first and last names in our random sample (N=351 projects). In this case, our dependent variable, Family, is binary and thus, a panel logit model is estimated. Results for a conditional fixed-effects logit model are in Table 7, Model 1. Due to the conditional estimation approach used for a fixed-effects model, projects that have Family=0 or 1 for all t are eliminated from the estimation. Consequently, we also report in Model 2 the estimation results for a pooled model that corrects for any dependence among errors over time within a project, but does not include fixed project effects. In both cases, the estimated coefficients for the first five days are significant suggesting that contributions from family members are more likely to occur in the early stage of a project as compared to the middle period. There is also some evidence that family members tend to pledge in the last two days of the project funding cycle. Together, these results suggest that family members are most likely to support a Kickstarter project in the first week after launch, as well as just before it ends. [insert Tables 8 and 9 about here] As observed by several creators, the pool of backers for reward-based crowdfunding projects is not predominately provided by the community (Dushnitsky and Marom 2013). While serial backers who have 15

contributed to multiple projects are important, the vast majority of contributors only support a single project: over 70% of all Kickstarter backers in our sample only pledge to one project and 95% of these backers joined the community and pledged in the same day (see Table 8). This distribution of backer experience is very different than that for other forms of crowdfunding where serial backers with prior funding experience are the primary investors (Agrawal, et al. 2013a). Moreover, this suggests that Kickstarter project creators attract most of their funding by mobilizing their own social network of friends (who are directly known by the project creator) and followers (who indirectly know the project creator from social media connections). Although we do not have explicit information on whether a project creator knows any of their backers, we explore the contribution dynamics of friends and followers in Kickstarter using a similar approach as Agrawal, et al. (2013a). We consider several proxies: (1) one-time backers who only fund a single project, (2) first-time backers who pledge to a focal project before contributing to any other, and (3) project creators that have a lot of Facebook friends 11. The results of estimating conditional fixed-effects Poisson models are in Table 9. Based on the positive and significant estimated coefficients for the Day and LastDay binary variables, we again find strong evidence for a U-shaped pattern of project support. One-time and firsttime backers are more likely to pledge in the first and last week as compared to the middle period in a project s funding cycle. The dynamics of project support for creators with a lot of Facebook friends also follows the same familiar U-shaped pattern. These results can be compared to those for serial backers in Model 4. No matter which proxy we use to account for friends and followers, there seems to be little suggestion that these backers pledge at different times in the funding cycle than others. Thus, we provide some evidence that family, friends, and followers are important supporters in the early and late stages of a Kickstarter project. However, this is not really surprising in light of the fact that most of the contributors at any point in the funding cycle are one-time backers that likely come from the project creator s own social circle. 11 The sample for this analysis is based on a median split of Kickstarter creators who linked their project to a Facebook account. 16

4.2.3 The Effects of Social Influence As noted by several researchers in lending and donation-based settings (Herzenstein, et al. 2011a; Zhang and Liu 2012; Burtch, et al. 2013a; Agrawal, et al. 2013a), an important factor that can influence the behavior of backers in crowdfunding communities is information on others prior funding decisions. In particular, the level of financial support for each project as well as its timing is publicly visible on most platforms. Thus, we consider the evidence for herding behavior in Kickstarter projects. Here, herding is defined as the tendency to contribute to crowdfunding projects that already have a lot of prior support. The literature on rational herding and information cascades argues that an initiative with a lot of community support signals that the project is of high quality (Devenow and Welch 1996). In line with other crowdfunding studies (e.g., Herzenstein, et al. 2011a; Zhang and Liu 2012; Burtch, et al. 2013a), we measure herding momentum with TotalBackers (the cumulative number backers supporting project i up to day t) and include its lagged value as an explanatory variable for BackersAdded in our basic econometric model. The results from estimating a conditional fixed-effects Poisson model with the full sample of projects are in Table 10, Model 1. In contrast to studies of crowdfunding communities that expect a financial return (Herzenstein, et al. 2011a; Zhang and Liu 2012), the coefficient estimate for Lag TotalBackers is insignificant providing no support for herding based on the number of previous project backers. While Kickstarter projects do seem to vary in quality (Table 1), this result suggests that Kickstarter backers do not use others pledging decisions to infer project quality. Whereas the quality of an investment opportunity in equity and lending-based crowdfunding communities is somewhat objective since it directly relates to the expected financial return, the perceived value of a reward-based project is based more on whether a potential backer believes the project creator and their proposed endeavor is compelling. This is consistent with advice from crowdfunding experts who emphasize the importance of having a persuasive video story as part of the project description (de Witt 2012; Steinberg 2012). According to Nano Whitman, a successful Kickstarter creator (Steinberg 2012: 93), the video is key. Whatever you write and all those gifts are secondary. It s the video that makes people feel they want to give something to you The insignificant coefficient estimate for Lag TotalBackers is also contrary to findings in a donation-based crowdfunding community (Burtch, et al. 2013a). 17

For Kickstarter projects which generally involve private rather than public goods, others contributions do not seem to crowd-out pledges from potential backers. [insert Table 10 about here] As noted earlier, Figure 1 highlights an interesting phenomenon in Kickstarter that also involves social influence within the community. Commonly known as the Kickstarter Effect, as a project nears its goal there can be a flurry of activity that pushes it over its target (Galinsky 2010; Nelson 2013). Matt Haughey, a backer of more than 150 Kickstarter projects, sums it up this way (Steinberg 2012: 149): once you pass 50 percent of your funding, at any point, you have a 95 percent chance of reaching your goal Only a handful of projects have finished unsuccessfully having reached 85 percent or more of their funding. The people who are at like 60, 70 percent with a week to go, it s gonna be OK! Clearly, there is solid empirical support for this notion from the strong U-shaped pattern of project support over the project funding cycle (e.g., see Table 4). While much of the research studying the reasons for goal pursuit has emphasized individuals and their personal goals, this work can be used to understand the motivations for individuals to contribute to the shared goals of a group (Fishbach, et al. 2011). When group identification is relatively weak (as in crowdfunding communities with anonymous members), research finds that individuals decide to pursue a shared group goal if they believe the goal is worthwhile (Fishbach, et al. 2011). Here, others prior contributions can positively influence the assessment of goal value. In the case of crowdfunding, whether or not a project is deemed worthy of support depends on how much of the goal has already been pledged. Backers want the project to succeed, and projects closer to their target goal are more likely to reach their funding objective. Given that a project is considered to be worthwhile, the Kickstarter Effect further suggests an acceleration in funding activity as a project nears its goal. Such an increase in motivation and effort to reach a goal as it is approached has been found in humans and other animals (Liberman and Forster 2008; Toure- Tillery and Fishbach 2011). Rats run faster through a maze as they get closer to food (Hull 1932), people increase their purchases as they approach rewards from loyalty cards (Kivetz, et al. 2006), and groups of donors contribute more to charitable campaigns close to reaching their goals (Fishbach, et al. 2011; Cryder, et al. 2013). More formally, the goal-gradient hypothesis is that motivation to reach a goal increases 18

monotonically with proximity to the desired end state (Hull 1932). One key reason for goal-gradient behavior is that the perceived impact of later stage decisions tends to increase over the course of goal pursuit (Toure- Tillery and Fishbach 2011). For example, the marginal impact of a $100 contribution to a project that is halfway towards its goal of $1000 is much less than the marginal impact of the same contribution if this project has already achieved 90% of its goal. As discussed by Cryder, et al. (2013), perceived impact is an important rationale for prosocial acts like crowdfunding. Even in situations when there are no financial rewards, backers still perceive that later stage funding decisions close to the goal have more impact and thus they are even more likely to make a donation when the target is in sight. To examine whether goal-gradient behavior exists for Kickstarter projects, we define a set of discrete binary variables for quintiles of PercentGoal and include their lagged values as explanatory variables (here Lag PercentGoal < 20% is the reference category). The results from estimating a conditional fixed-effects Poisson model is in Table 10, Model 2. Consistent with the goal-gradient hypothesis, the estimated coefficients for the binary Lag PercentGoal variables are significant and monotonically increasing in magnitude up to 80% of goal. The decrease in project support when it is very close to its goal may be due to a drop in motivation as potential backers view the target to be at hand (Fishbach, et al. 2011). We also note that similar decreases in motivation and effort very near the goal is also reported in the original Hull rat experiments as well as the loyalty card reward experiments by Kivetz and his colleagues (see Figures 1 and 3 in Kivetz, et al. 2006). Thus, we do find empirical evidence for positive effects of social influence linked to how much has already been pledged to the project goal. [insert Figure 4 about here] We also consider whether goal-gradient behavior varies over the project funding cycle by including interaction terms involving the funding time binary variables and the PercentGoal categories. Estimation results from the random sample are reported in Table 10, Model 3. Because we cannot directly interpret the interaction effects from the coefficient estimates, we do not report these terms in Model 3. So as not to clutter the plots, the average marginal effects for a single day in the first and last week are shown in Figure 4. This same pattern of marginal effects is obtained for the other days in the funding cycle. While we see 19