Association for Information Systems AIS Electronic Library (AISeL) ICIS 2017 Proceedings Dec 10th, 12:00 AM Investing or Gambling? Empirical Evidence on the Role of the Lottery in Reward-based Crowdfunding Platforms Zuyin (Alvin) Zheng Temple University, zheng@temple.edu Jing Gong Temple University Fox School of Business and Management, gong@temple.edu Paul Pavlou Temple University, pavlou@temple.edu Follow this and additional works at: http://aisel.aisnet.org/icis2017 Zheng, Zuyin (Alvin); Gong, Jing; and Pavlou, Paul, "Investing or Gambling? Empirical Evidence on the Role of the Lottery in Reward-based Crowdfunding Platforms" (2017). ICIS 2017 Proceedings. 22. http://aisel.aisnet.org/icis2017/peer-to-peer/presentations/22 This material is brought to you by the International Conference on Information Systems (ICIS) at AIS Electronic Library (AISeL). It has been accepted for inclusion in ICIS 2017 Proceedings by an authorized administrator of AIS Electronic Library (AISeL). For more information, please contact elibrary@aisnet.org.
Investing or Gambling? Empirical Evidence on the Role of the Lottery in Reward-based Crowdfunding Platforms Completed Research Paper Zuyin (Alvin) Zheng Temple University 1801 N Broad St, Philadelphia, PA 19122 zheng@temple.edu Jing Gong Temple University 1801 N Broad St, Philadelphia, PA 19122 gong@temple.edu Paul A. Pavlou Temple University 1801 N Broad St, Philadelphia, PA 19122 pavlou@temple.edu Abstract Crowdfunding aims to collect small investments from a large crowd to support a project by fundraisers. Motivated by the observation that crowdfunding projects often attract a small number of backers, this study examines the role of the lottery a novel mechanism used to attract backers in crowdfunding platforms in three key crowdfunding outcomes (total number of backers, total money raised, probability of reaching the funding goal). Using data from a reward-based crowdfunding platform, we show that although the lottery does help attract a higher number of backers, it reduces the total money raised and the probability of reaching the funding goal. Our study contributes to the crowdfunding literature by both confirming the positive role of the lottery in backing engagement and demonstrating its overall negative role in crowdfunding success. Managerial implications for fundraisers and crowdfunding platforms are discussed. Keywords: Lottery, Crowdfunding, Reward-based Crowdfunding Platforms, Crowdfunding Outcomes, Intrinsic Motivation, Extrinsic Motivation Thirty Eighth International Conference on Information Systems, South Korea 2017 1
Introduction Instead of raising funds from a small group of wealthy investors as in traditional fundraising, crowdfunding platforms allow fundraisers to raise small amounts from the crowd (Mollick et al. 2014, Burtch et al. 2013). The power of the crowd has become a key feature of crowdfunding. However, in practice, the average number of backers per successful project is around 30 (Burtch et al. 2013, 2015), and this number is much smaller for projects that fail to reach their funding goals. Consequently, many crowdfunding platforms have designed various incentives to attract backers to help projects reach their funding goals. This study seeks to investigate the role of the lottery, a commonly-used approach in the literature to increase engagement (e.g., Reczek et al. 2014, Zhang et al. 2016), as an incentive for backers to contribute to reward-based crowdfunding projects. Specifically, we examine how the lottery in crowdfunding platforms affects three key project fundraising outcomes, i.e., the total number of backers, total money raised, and the probability to reach the funding goal. We also seek to uncover the underlying theoretical mechanisms by which backers decide to contribute on reward-based crowdfunding platforms given the presence of a lottery. While the lottery was shown to be effective in marketing (e.g., in the form of sweepstakes) at increasing consumer engagement (e.g., Reczek et al. 2014, Zhang et al. 2016), the lottery in crowdfunding platforms differs in several aspects. First, while the lottery often involves a large prize (e.g., a large sum of money), the prize of the lottery in crowdfunding platforms is typically the focal product of the crowdfunding project often a new product of relatively low or unknown value, which is likely to decrease the effectiveness of the lottery. Second, while in marketing promotions, consumers often automatically enter a lottery after purchasing a product without having to pay anything extra, the lottery in crowdfunding platforms is designed as an optional choice with an extra cost (i.e., the lottery price), which may prevent backers from entering the lottery in the first place. Third, in crowdfunding, no winner is selected if the project fails to reach its funding goal, implying that the winning probability of the lottery not only depends on the pre-specified odds of the lottery, but also on the probability that the project will reach its funding goal, thus further affecting the attractiveness of the lottery. Therefore, whether the lottery is effective in boosting crowdfunding success is an open empirical question, given these differences in the lottery design, such as the prize of the lottery, its cost, and the winning probability. Most importantly, the theoretical mechanisms by which the lottery may affect crowdfunding outcomes are not well understood. On the one hand, the lottery may lead to an incentivizing effect, in that the lottery may attract new backers who would otherwise not fund the project (non-backers) to become lottery backers. These marginal backers could be extrinsically motivated, but they are willing to pay less than the minimum amount to get the reward (minimum reward price), or intrinsically motivated to contribute, but they face a substantial effort associated with backing, which discourages them from doing so. The lottery serves as an extrinsic incentive for these backers, thus potentially increasing the total number of backers and the total money raised, thus increasing the probability of achieving the project s funding goal. On the other hand, the lottery may also have a cannibalizing effect since prospective backers may switch from other backing options to the lottery one. If this is the case, prospective backers who would otherwise contribute the full amount for the reward in the absence of the lottery (referred to as rewardees) may opt for the lottery and instead become lottery backers. These backers may back a lower amount (as the lottery price is always lower than the minimum reward price) when the lottery is available versus when the lottery is absent. In addition, some potential backers who would otherwise back the project altruistically without getting a reward (referred to as donors) may opt for the lottery, thus becoming lottery backers instead. Taken together, if the cannibalizing effect of a lottery is strong, it may reduce the total money raised by a project, thus making it more difficult for a crowdfunding project to reach its funding goal. In sum, these arguments offer competing predictions for the role of the lottery in a project s fundraising success, calling for an empirical examination of both the role of the lottery in fundraising outcomes and also the underlying theoretical mechanisms. We collected a 4-year dataset from one of the largest crowdfunding platforms in China, which implemented the lottery feature on November 12, 2015. After this date, fundraisers could add an additional lottery choice by specifying the price to participate in the lottery, the winning probability, and the lottery prize (usually the focal reward of the crowdfunding project). Prospective backers can choose one of the three options: (1) backing the lottery with a chance of winning the reward, (2) backing as a donor without the reward, or (3) backing as a rewardee to receive the reward with certainty. Results show Thirty Eighth International Conference on Information Systems, South Korea 2017 2
that projects with a lottery attract, on average, 16% more backers than those without a lottery, providing evidence of the incentivizing effect of the lottery. When decomposing this effect, the lottery is associated with a 40% decrease in the number of rewardees and a 26% decrease in the number of donors, providing evidence of the cannibalizing effect. These two competing effects make a project with the lottery 11% less likely to reach its funding goal due to a 64% decrease in the total money raised. These findings are robust across different model specifications, including linear and Logit models with and without fundraiser fixed effects, Coarsened Exact Matching (CEM) to address fundraisers self-selection of the lottery at the observable level, plus falsification tests. This study contributes to several streams of literature crowdfunding, lottery, and prosocial behavior. First, our study extends crowdfunding literature by theorizing and empirically examining the effect of the lottery as an extrinsic incentive on crowdfunding outcomes and backer behavior. While the lottery does make backers more likely to fund a project, it also reduces the total money raised and reduces the chances of reaching the project s funding goal by reducing the average backing amount. Second, while lottery literature has shown the positive role of the lottery in consumer engagement (e.g., Kalra and Shi 2010, Zhang et al. 2016), our study is the first to show the negative role of the lottery in crowdfunding success. Third, as prosocial literature shows that extrinsic motivations may crowd out prosocial behaviors (e.g. Bénabou and Tirole 2006, Gneezy et al. 2011), we validate the crowding out effect of the lottery as an extrinsic incentive on inducing intrinsically-motivated donors to opt for the lottery when available, thus becoming lottery backers. Finally, this study also has managerial implications: Since the lottery has been an increasingly commonly-used incentive by crowdfunding platforms (e.g., JD Finance and SUNING Zhongchou), our results indicate that the lottery fails to help projects reach their funding goals, despite its incentivizing effect. Hence, fundraisers and platform managers should take into account the negative role of the lottery in backer contribution and project success. Related Literature The crowdfunding literature has examined many aspects that may play a role in crowdfunding outcomes, including project characteristics, such as project type, goal, and duration (Frydrych et al. 2014, Liu et al. 2014, Mollick 2014), fundraiser characteristics, such as experience and social capital (Giudici et al. 2013, Kim and Viswanathan 2016, Liu et al. 2015, Zvilichovsky et al. 2014), geographic location of fundraisers and backers (Lin and Viswanathan 2015), peer influence (Burtch et al. 2013), and privacy concerns (Burtch et al. 2015). We are particularly interested in: (1) the literature on backers motivation and incentives to contribute to crowdfunding projects, and (2) the literature on the lottery, as reviewed below. Backers Motivations to Contribute to Crowdfunding Projects Backers could either be (i) intrinsically motivated out of a desire to help fundraisers, whether altruistically or to feel good about themselves, or (ii) extrinsically motivated because of potential rewards (Allison et al. 2015). Backers motivations to participate in crowdfunding are heterogeneous across different contexts, such as reward-based crowdfunding (Kickstarter and Zhongchou), donation-based crowdfunding (GoFundMe), and peer-to-peer lending (Prosper.com) (e.g., Allison et al. 2015, Lin et al. 2015). First, backers can be intrinsically motivated, in which case they perceive backing a project to be exciting, or want to support a good cause, or may gain utility from the backing activity itself. Intrinsic motives, such as altruism, are primary drivers of backers behaviors in donation-based crowdfunding (Aitamurto 2011). Burtch et al. (2013), for example, showed that the cumulative contribution of prior backers negatively affected a backer s backing decisions, since backers favored contributing to those projects that had a stronger need for their support, implying that backers are altruistic. Similarly, Amara et al. (2014) showed that donors in donation-based projects who were motivated to support a good cause became attached to a social group and received enjoyment from the project outcome. Other intrinsic motivations include being part of a community of backers (Gerber et al. 2012) and self-affirmation obtained from the backing process (Hemer et al. 2011). In a more recent study by Bretschneider and Leimeister (2017), they proposed that backers can be prosocially motivated in reward-based crowdfunding. Thus, in rewardbased crowdfunding platforms, donors (i.e., those who back a small amount without receiving a reward) are mainly intrinsically motivated, as they back an amount without receiving an extrinsic reward in return. Thirty Eighth International Conference on Information Systems, South Korea 2017 3
However, backers can also be extrinsically motivated by the financial returns from their backing behavior, such as receiving a cash payment, a product reward, or a compensatory service (Allison et al. 2015). Gerber et al. (2012) identified seeking rewards as the primary motive for backers, which was also found by Fisk et al. (2011) in the context of supporting musicians. Hemer et al. (2011) confirmed that interest in a reward is a key determinant of a backer s backing decision. The motivation of rewardees (i.e., those who receive a reward after backing a project) can be purely extrinsic or both extrinsic and intrinsic. On the one hand, rewardees could be purely extrinsically motivated, because they are interested in the reward itself and are willing to pay at least the minimum reward price to get the reward. On the other hand, rewardees can also be intrinsically motivated by altruistic reasons, such as a desire to help the fundraiser, and may simply appreciate the focal reward as a bonus. Given the fact that rewardees often back a larger amount than donors and receive the focal reward, we argue that extrinsic motivation is the primary motive for rewardees. In sum, donors are intrinsically motivated, while rewardees are primarily extrinsically motivated. Those who do not back a project at all (non-backers) are neither intrinsically motivated, as donors, nor extrinsically motivated to contribute beyond the minimum reward price for the focal reward, as rewardees. Also, there may be non-backers who are interested in the focal reward (extrinsically motivated), albeit with a willingness to back the project less than the cost of the reward (i.e., minimum reward price), and those who are intrinsically motivated to help, but choose not to back because of the substantial effort associated with the backing process, such as creating an account on the crowdfunding platform and filling in their personal and financial information (Table 1s). Table 1. Comparison of Backers Motivation Intrinsic Motivation Extrinsic Motivation Donors Yes No Rewardees Maybe Yes Non-Backers No No Lottery The lottery is often featured with 1) an attractive prize, 2) a disproportionally small price to participate in the lottery, and 3) a random selection process of the winners (i.e. winning based on a certain probability). The fact that the lottery provides participants with a chance to win a large-value prize with a disproportionally small cost makes the concept of the lottery appealing (Friedman and Savage 1948). Also, the low cost of a lottery makes it possible for potential backers to allocate their budget which is often constrained in practice to various products in which they are interested (Kwang 1965). In addition, because of innate cognitive biases, lottery participants tend to overestimate the probability of winning, either because they believe that they are luckier than others or because of innate optimism (Alloy and Abramson 1988, Goldsmith and Amir 2010). The lottery has been consistently shown in the marketing and charity literature to have a significant, positive role in consumer engagement, purchase, and donation decisions. For instance, a lottery is shown to be more effective than coupons in incentivizing purchases by creating an interactive customer experience (e.g., Zhang et al. 2016). The lottery, as an extrinsic incentive, also helps charities raise more money than voluntary donations by inducing a higher participation rate (Landry et al. 2006). Notably, donors attracted by the lottery are as likely as those donors who have donated before to make another charity donation (Landry et al. 2010). With an implicit assumption about the positive effect of the lottery, some studies focused on different determinants of the effectiveness of the lottery. For instance, a lottery with a larger-value prize is consistently shown to be more effective than with a small-value prize (Kalra and Shi 2010). A fixed-prize lottery induces significantly higher participation and raises more money compared to a revenuedependent prize (Dale 2004). Moreover, perceptions of the winning probability tend to be biased (Reczek et al. 2014): consumers who spend more time and money with the brand tend to believe that they have a higher likelihood of obtaining the lottery prize. In sum, the lottery has been shown to be effective in incentivizing people in marketing promotions and charity donations when the lottery is designed as an opportunity to win a large-value prize (e.g. a big prize or a car) without an explicit cost to participate. Thirty Eighth International Conference on Information Systems, South Korea 2017 4
Finally, the value of the lottery prize and the winning probability (both objective and perceived) are key determinants of the effectiveness of a lottery. Hypotheses Development Our theorization is based on two potential mechanisms (incentivizing effect and cannibalizing effect) by which the lottery may affect the decisions of prospective backers and determine key crowdfunding outcomes (total money raised and the probability of reaching the funding goal), as we hypothesize in detail below. The Role of the Lottery in the Total Number of Backers (H1) The introduction of the lottery in crowdfunding platforms provides prospective backers with a chance to win the reward (focal product) with a much lower backing amount (the lottery price) than what is typically required to obtain the focal reward, which is known as the disproportionate benefit (e.g. Friedman and Savage 1948). Winning a reward with a disproportionately low price may attract new backers who are interested in the focal reward of the project, but who are not willing to pay the minimum price to get the reward sans the lottery, thus motivating them to back the project as lottery backers (incentivizing effect). The incentivizing effect may thus increase the number of backings a project receives. If the lottery only encourages prospective backers to switch from backing as donors and rewardees (absent a lottery) to back as lottery backers, the total number of backings a project receives is not likely to change. Therefore, the effect of a lottery on the total number of backers depends on the extent to which the lottery may incentivize new backers who would otherwise not back the project. Furthermore, for those who are intrinsically motivated to contribute, but do not because of the donation cost, the lottery serves as a tangible incentive to offset the full cost of participation, thus incentivizing these backers to become lottery backers. Since the increase in the number of lottery backers is counted towards the total number of backers and displayed on the first page of a project, this helps projects gain an initial advantage. If the lottery does incentivize a large number of lottery backers, this initial advantage may help attract more backers eventually (Koning and Model 2015). We therefore argue that the lottery has a positive effect on the total number of backers a project may receive. Accordingly, we propose the following hypothesis for testing: H1 (Total Number of Backers): The lottery is positively associated with the total number of backers that a crowdfunding project attracts. The Role of the Lottery in the Total Number of Rewardees (H2) In reward-based crowdfunding platforms, prospective rewardees are primarily extrinsically motivated because of a project reward (Allison et al. 2015, Gerber et al. 2012, Hemer et al. 2011). When the lottery becomes available in a crowdfunding project, it provides prospective backers with a chance to win the reward (product) at a much lower price (the lottery price) than what is typically required to obtain the reward (the minimum reward price). Although the probability of winning the lottery prize is uncertain, the lottery may be attractive to those who are price-sensitive and willing to tolerate the risk of not winning the reward. In addition, if the lottery is available, prospective backers could allocate their limited budget, which they would otherwise spend on a single reward, to several projects in an attempt to win multiple rewards that interest them (Kwang 1965). Therefore, it is possible that some prospective backers may switch from backing as rewardees absent the lottery to backing as lottery backers when the lottery is available (cannibalizing effect). In sum, we propose: H2 (Total Number of Rewardees): The lottery is negatively associated with the total number of rewardees that a crowdfunding project attracts. The Role of the Lottery in the Total Number of Donors (H3) The second group of prospective backers consists of those who are mainly intrinsically motivated (interest in helping the fundraisers or being altruistic) and do not seek to obtain the focal reward. Before the introduction of the lottery, intrinsically-motivated donors who donated a small amount less than the minimum reward price did not receive any reward. When the lottery is available, those prospective Thirty Eighth International Conference on Information Systems, South Korea 2017 5
backers who are intrinsically motivated now have two options: back a small amount as donors without any reward (as before) or participate in the lottery as lottery backers and get a chance to win the reward. On the one hand, it is possible that intrinsically-motivated prospective backers remain donors and do not opt for the lottery, because their altruistic deeds ostensibly remain unblemished. On the other hand, cognitive evaluation theory (Deci and Ryan 1985, 2002) suggests that the way in which extrinsic rewards affect intrinsic motivation depends on personal interpretation. Specifically, intrinsic motivation decreases when people interpret the rewards as indicative of external control that decreases their sense of selfcontrol and competence. In reward-based crowdfunding, the lottery is offered as an optional choice, which is unlikely to affect prospective donors sense of self-control and competence. Hence, if prospective donors interpret the lottery as affirming their own competence and self-control over their prosocial donation behaviors, they may choose to participate in the lottery, becoming lottery backers instead of donors (cannibalizing effect), thereby reducing the number of donors. In sum, the lottery may cannibalize prospective donors and encourage them to back as lottery backers. Therefore, we propose that: H3 (Total Number of Donors): The lottery is negatively associated with the total number of donors that a crowdfunding project attracts. The Role of the Lottery in Crowdfunding Success (H4 & H5) As discussed earlier, on the one hand, the lottery may cannibalize potential rewardees who, absent of the lottery, would otherwise back a large amount for the reward. Since lottery backers always back a smaller amount than rewardees, this behavior may reduce the average contribution amount. The lottery may also cannibalize potential donors who would otherwise back a small amount without the reward. Although the lottery price is not necessarily lower than the backing amount of the donors, empirically, over 70% of the projects with a lottery set the lottery price as the lowest pre-specified backing amount. Thus, the lottery is also likely to decrease the average contribution amount due to a cannibalizing effect on prospective donors. On the other hand, the proposed incentivizing effect suggests that the lottery, as an extrinsic motivator, may incentivize non-backers who would otherwise not back at all to become lottery backers, thus increasing the total number of backers attracted to the project, as well as the total money raised. Therefore, the role of the lottery in project fundraising outcomes depends on which is the dominant effect. If the incentivizing effect dominates the cannibalizing effect, we would observe that the lottery has a positive effect on the total money raised and the probability of reaching a project s funding goal. This happens if the lottery incentivizes enough additional backers to offset the reduction in total raised money caused by the lower average contributions due the cannibalizing effect. In contrast, if the lottery fails to incentivize an adequate number of non-backers, the lottery will have a negative effect on the total money raised and the probability of reaching a project s funding goal. Taken together, since we cannot ex ante specify which effect (incentivizing or cannibalizing) is stronger, we hypothesize a set of competing hypotheses: H4 (Total Money Raised): The lottery is a) positively or b) negatively associated with the total money raised by a crowdfunding project. H5 (Project Success): The lottery is a) positively or b) negatively associated with the probability of reaching the funding goal of a crowdfunding project (i.e. project success). Table 2. A Summary of Hypotheses by which the Lottery Shapes Crowdfunding Outcomes Proposed Effects Lottery Number of Backers (H1) Number of Rewardees (H2) Number of Donors (H3) Total Money (H4) Project Success (H5) Incentivizing Positive No Change No Change Positive Positive Cannibalizing No Change Negative Negative Negative Negative Thirty Eighth International Conference on Information Systems, South Korea 2017 6
Empirical Analyses Data and Research Context We collected our data from Zhongchou (www.zhongchou.com), one of the largest reward-based crowdfunding platforms in China, launched in 2013. Similar to other reward-based crowdfunding platforms such as Kickstarter, Zhongchou allows fundraisers to pre-specify the backing amount needed to receive different combinations of the focal reward and a thank-you note when initiating a project (Figure 1). Previously, prospective backers could either back a pre-specified amount for a reward (as rewardees) or back a pre-specified amount without receiving the reward (as donors). Rewardees who back a prespecified reward price receive a reward if the project successfully reaches its funding goal. We define the lowest pre-specified price to get a reward (e.g., 36 RMB in Figure 1A) for a project as the minimum reward price. To facilitate a backer s engagement with the platform and attract more backers to contribute, the platform introduced a lottery feature on November 12, 2015, allowing fundraisers to set up a lottery for their project. When posting a project with a lottery, fundraisers need to specify the backing amount needed for backers to participate in the lottery (termed lottery price, e.g., 2 RMB in Figure 1B), 1 the winning probability (e.g., one out of twenty in Figure 1B), and the winning prize (which is typically the focal reward of the project), all of which are pre-specified and remain unchanged throughout the duration of a project. In sum, for projects with a lottery, potential backers have three ways of backing the project: (1) back the pre-specified (large) amount for the reward as rewardees; (2) back a pre-specified small amount without receiving the reward as donors; or (3) back the lottery as lottery backers. In total, we collected data on 15,252 projects spanning from March 2013 to September 2016, among which 6,033 projects were initiated after the introduction of the lottery feature. We excluded projects that had not completed the fundraising process by the time we collected the data and projects that did not receive any backing at all. Among projects posted after the introduction of the lottery feature on November 12, 2015, 37.71% (2,275) projects adopted the lottery (treatment group), while 62.29% (3,758) projects did not (control group). 1 Although fundraisers can specify the lottery price as any positive amount, in practice, over 50% of the projects with a lottery set the lottery price at 1 RMB, while 93% of the projects with a lottery set the lottery price at less than 10 RMB. Thirty Eighth International Conference on Information Systems, South Korea 2017 7
Figure 1. Backing Choices for Backers: Projects with a Lottery versus without a Lottery Model Specification We first performed our analysis at the project level by investigating how the lottery affects the proposed crowdfunding outcomes (the total number of rewardees, donors, and backers that a project receives and the total money raised and the probability of project success). We log-transformed all non-categorical variables to reflect percentage changes. This was deemed appropriate, primarily because these noncategorical variables vary widely, and understanding these effects in percentage terms is more useful (Keene 1995, Burtch et al. 2013). The definition of each variable is shown in Table 3 and the summary statistics are reported in Table 4. Table 3. Variable Definition Variable Definition Log No. of Total Backers Log total number backers a project received Log No. of Donors Log total number of backers who back for no reward Log No. of Rewardees Log total number of backers who back for the focal reward with certainty Log No. of Non-Rewardees Log total number of donors and lottery backers Log Total Money Total money raised for a project Lottery 1-a project used lottery, 0-otherwise Lottery Introduction 1-lottery feature is introduced and available on the platform, 0-otherwise Log Goal The funding goal of a project, in RMB Log Min. Reward Price Minimum price to get a reward, in RMB Project Duration Fundraising duration, in days Thirty Eighth International Conference on Information Systems, South Korea 2017 8
Video Log No. of Pictures Log Text Length Log Fundraiser Back Exp. Log Fundraiser Raise Exp. Project Type Project Post Time Project Province 1-a project has a video describing the project, 0-otherwise Log number of projects a project used to describe the project Log total number of Chinese characters in the text description Log total number of projects a fundraiser has backed before Log total number of projects a fundraiser has raised fund for previously 1-Public Cause, 2-Agriculture, 3-Press, 4-Entertainment, 5-Art, 6-Other The date a project is posted The location (province) of the project In Model 1, we examined the impact of the lottery on a set of non-categorical outcomes as follows: Y = + + + + [1] In Model 1, j indicates a project. Y is one of the five non-categorical dependent variables, including the log-transformed number of rewardees, number of donors, number of non-rewardees, total number backers, and total money raised. is a binary variable indicating whether the project has a lottery (1=Yes, 0=No). is a vector of observable project-level characteristics (Mollick 2014, Frydrych et al. 2014), such as the funding goal, duration, project type, whether the description included a video (Liu et al. 2014), starting week, minimum reward price, and the length of the project description. We controlled for the starting week of the project (using dummy variables) to account for unobserved time-variant factors, such as seasonality and platform-wide changes, which may affect the number of backers and project success. is a vector of characteristics of the fundraiser for project j (several different projects may be posted by the same fundraiser), including the fundraiser fixed effect to account for unobserved time-invariant fundraiser characteristics, the total number of projects that the fundraiser had backed before, and the total number of projects the fundraiser had posted before on the crowdfunding platform, which are proxies for the fundraiser s experience and social capital on the platform (Giudici et al. 2013, Mollick 2014). Similarly, we next used a Logit model to examine the role of the lottery in project success: ( )= + + + [2] where is a binary variable indicating whether project j reached its funding goal. Similar to Model 1, our main focus is on the coefficient of. We used the same set of control variables as in Model 1. A major concern with our identification strategy is that the adoption of the lottery for a project may be self-selected by the fundraiser and may thus be endogenous. The comparison may be misleading if projects with a lottery differ systematically from projects without a lottery in terms of project type, goal, duration, project quality, and fundraisers desire and effort to reach the funding goal. That is, endogeneity issues arise if a) the treatment and control groups are imbalanced in terms of observable characteristics, and b) the treatment itself may be confounded by unobserved project quality and fundraiser effort. Following Blackwell et al. (2009), we used Coarsened Exact Matching (CEM) (e.g. Bapna et al. 2016, Subramanian and Overby 2016) to account for fundraisers self-selection for the lottery adoption based on observed characteristics. Meanwhile, we used a set of falsification tests to rule out alternative explanations, such as project quality, fundraiser effort, and the marketing effort of the platform. Table 4. Summary Statistics Variable Observations Mean Std. Dev. Min Max Log # Backers 6,033 2.588 1.501 0.693 8.463 Log # Rewardees 6,033 1.777 1.599 0 7.977 Log # Donors 6,033 1.647 1.426 0 7.571 Log # Non-Rewardees 6,033 2.039 1.420 0 7.571 Log Money 6,033 5.801 2.867 0.693 13.028 Success 6,033 0.268 0.443 0 1 Lottery 6,033 0.377 0.485 0 1 Log Goal 6,033 9.287 1.593 0.693 13.816 Log Min. Reward Price 5,861 4.170 0.975 2.398 8.517 Thirty Eighth International Conference on Information Systems, South Korea 2017 9
Project Duration 6,033 38.263 22.121 10 90 Video 6,033 0.218 0.413 0 1 Log No. of Pictures 6,033 2.681 0.719 0 5.220 Log Text Length 6,033 7.664 0.557 5.670 9.692 Log Fundraiser Back Exp. 6,033 0.322 0.806 0 6.766 Log Fundraiser Raise Exp. 6,033 0.345 0.784 0 4.970 Results Main Model: Coarsened Exact Matching To estimate the effect of the lottery on crowdfunding outcomes, we compared the outcome variables of projects with a lottery (treatment group) to projects without lotteries (control group). Ideally, we want the treatment and control groups to be statistically identical (by randomly assigning some projects with a lottery) except for the treatment. In our empirical setting, however, fundraisers decide whether to adopt a lottery for a crowdfunding project based on project characteristics and their own preferences. The selfselection of lottery adoption may result in differences in observable characteristics between the treatment group and the control group. CEM is designed to match observations in the control group to observations in the treatment group by balancing observable characteristics, thus mimicking an ideal counterfactual for projects with a lottery (Blackwell et al. 2009). Specifically, CEM was conducted in two steps: First, projects were stratified based on observable characteristics, including project type (exact match), funding goal, duration, fundraisers funding and backing experience, minimum reward price, description length, number of pictures used to describe the project, whether there is a video to describe the project (exact match), project posting week, and project location (exact match). After this, each stratum contains projects that are identical (exact match) or similar (approximate match) on the basis of these characteristics. Second, we discarded strata containing only treated or untreated projects, and we renormalized weights of observations in the remaining strata to place equal weight on treated and untreated units in each stratum. In sum, applying CEM resulted in 1,947 projects (86% of all treated) in the treatment group that were matched with 2,760 projects in the control group. We adopted the L 1 statistic, introduced by Iacus et al. (2012), to check the performance of CEM in reducing the imbalance between the treatment and the control groups. L 1 measures the difference in frequency between the multidimensional histogram of all covariates in the treatment group and the same in the control group, with a value of 0 indicating a perfect balance and 1 indicating a complete imbalance (Blackwell et al. 2009). Most of the L 1 values after CEM matching are close to 0 in the mean and each quantile and smaller than the overall L 1 value before matching. This indicates that our matching reduced the imbalances between the treatment group and the control group. Table 5. The Effect of Lottery on Crowdfunding Outcomes: Coarsened Exact Matching (1) (2) (3) (4) (5) (6) VARIABLES OLS OLS OLS OLS OLS Logit DV Ln Ln Ln Ln (# Non- Ln (#Backers) (#Rewardees) (#Donors) Rewardees) ($Money) Success Changes in Y 2 21% -39% -23% 93% -61% -9% Lottery 0.194*** -0.501*** -0.261*** 0.659*** -0.940*** -0.666*** (0.0494) (0.0739) (0.0470) (0.0470) (0.0964) (0.1110) Log Goal 0.104*** 0.067** 0.116*** 0.079*** 0.244*** -0.306*** (0.0187) (0.0275) (0.0178) (0.0178) (0.0366) (0.0418) Log Min. Reward Price -0.141*** -0.365*** 0.008 0.018-0.117* -0.179*** (0.0307) (0.0437) (0.0292) (0.0292) (0.0598) (0.0675) 2 In log-linear models, when the coefficient is negative or large, the coefficient is no longer an approximate estimate of the percentage change in the dependent variable y. Instead, we must calculate the percentage change in y using the following formula: ln( + y) ln( )=β. The percentage change is: =e 1. This applies to all Tables. Thirty Eighth International Conference on Information Systems, South Korea 2017 10
Project Duration -0.004*** -0.006*** -0.001-0.001-0.009*** -0.016*** (0.0012) (0.0017) (0.0012) (0.0012) (0.0024) (0.0031) Video 0.140** 0.143* 0.154*** 0.130** 0.203* 0.233* (0.0613) (0.0850) (0.0584) (0.0584) (0.1196) (0.1308) Log No. of Pictures -0.002 0.035-0.071-0.068 0.228** -0.038 (0.0463) (0.0656) (0.0441) (0.0441) (0.0904) (0.1007) Log Text Length 0.506*** 0.561*** 0.357*** 0.457*** 1.108*** 1.399*** (0.0820) (0.1155) (0.0780) (0.0780) (0.1599) (0.1742) Log Fundraiser Back Exp. 0.113*** 0.133*** 0.093*** 0.087*** 0.193*** 0.239*** (0.0346) (0.0508) (0.0330) (0.0330) (0.0676) (0.0673) Log Fundraiser Raise Exp. -0.003 0.085-0.138*** -0.090** 0.150* 0.127 (0.0442) (0.0720) (0.0421) (0.0421) (0.0862) (0.0879) Constant -0.735-1.208-0.812-1.354** -3.613** -6.712*** (0.7222) (1.0049) (0.6874) (0.6876) (1.4088) (1.5056) Observations 4,707 4,707 4,707 4,707 4,707 4,707 (Pseudo) R-square 0.22 0.27 0.23 0.23 0.22 0.20 Project Type FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes City FE Yes Yes Yes Yes Yes Yes Note: Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Table 5 reports the estimation results of all crowdfunding outcomes based on the matched sample. Notably, projects with a lottery attract, on average, 21% more backers than those without a lottery (Column 1). This means that the lottery, in alignment with the crowdfunding platform s expectation and the prior literature (e.g. Zhang et al. 2016), does increase backer engagement (thus supporting H1), which provides evidence of the incentivizing effect of the lottery. When decomposing the overall increase, we find that projects with a lottery receive, on average, 39% fewer rewardees (Column 2) (thus supporting H2), 23% fewer donors (Column 3) (supporting H3), and 93% more non-rewardees (total number of donors and lottery backers, Column 4) 3 than projects without a lottery. In terms of total money raised and project success, projects with a lottery raise 61% less money than those without a lottery (Column 5) (supporting H4b), which makes them 9% (coefficient: -0.666) less likely to reach their funding goal (Column 6) (supporting H5b). The lottery thus does not incentivize an adequate number of non-backers to offset the decrease in the average contribution due to the cannibalizing effect, making it more difficult for fundraisers who use a lottery to reach their funding goals. Fundraiser Fixed Effects Model The CEM procedure helps alleviate concerns about the self-selection of fundraisers lottery adoption by balancing observable characteristics. Still, there may be other unobserved factors, such as unobserved fundraiser characteristics, which affect both the decision to use the lottery and the crowdfunding outcomes. To address these potential confounding factors, we considered a subset of fundraisers who had posted at least two projects, and we used a fundraiser fixed effects model to account for potential unobserved fundraiser time-invariant factors. Table 6 shows that all coefficients are consistent in terms of their sign and statistical significance, but they differ slightly in magnitude from the coefficients estimated with the CEM results (Table 2). This supports the role of the lottery in crowdfunding outcomes. The main reason for the difference in magnitude lies in the sample for the fixed effects model being constrained to fundraisers who posted at least two projects, 3 We included the number of non-rewardees as a dependent variable for the following reasons. First, non-rewardees have smaller backing amounts than rewardees, which helps explain the decrease in the total money raised by a project. Second, we used log-transformed outcome variables, which captures the percentage change between projects with a lottery and projects without a lottery. As projects without a lottery by default have zero lottery backers, which makes the percentage change from projects without a lottery to projects with a lottery infinite, the inclusion of nonrewardees as an additional dependent variable eliminates this concern. Thirty Eighth International Conference on Information Systems, South Korea 2017 11
and a large number of fundraisers who had only posted one project or who had the same outcome for all their projects was omitted from the fundraiser fixed effects model. Table 6. The Role of the Lottery in Crowdfunding Outcomes: Fundraiser Fixed Effects (1) (2) (3) (4) (5) (6) VARIABLES OLS OLS OLS OLS OLS Logit DV Ln (# Ln (# Ln (# Ln (# Non- Ln ($ Backers) Rewardees) Donors) Rewardees) Money) Success 4 Changes in Y 42% -23% -22% 132% -43% -- Lottery 0.349*** -0.267* -0.250* 0.840*** -0.567** -0.906** (0.1254) (0.1378) (0.1288) (0.1230) (0.2398) (0.3588) Observations 1527 1527 1527 1527 1527 1527 (Pseudo) R-square 0.48 0.69 0.42 0.49 0.48 -- Project Type FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes City FE Yes Yes Yes Yes Yes Yes Note: Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Summary of Main Results Table 7 summarizes the hypotheses testing along with the underlying theoretical mechanisms. After addressing the potential self-selection of the lottery adoption at the observable level with a CEM model and the unobservable level with both a fundraiser fixed-effects model, the results consistently show that while the lottery helps attract more backers, the lottery also decreases the total money raised by a project and its probability of success (reaching the funding goal). The result that the lottery increases the total number of backers indicates that the lottery in crowdfunding does incentivize new backers who would otherwise not back the project at all. When decomposing the increase in the total number of backers, we find that the lottery decreases the total number of both rewardees and donors. Because the lottery price is, by definition, lower than the minimum reward price to back as a rewardee, the average contribution of the lottery backers tends to be much lower than that of the rewardees. Thus, projects with a lottery raise less money and are less likely to reach their funding goal. Table 7. Summary of Hypotheses Testing and Underlying Theoretical Mechanisms Hypotheses Supported Mechanism H1 (No. of Backers): The lottery is positively associated with the total number of backers that a crowdfunding project attracts. H2 (No. of Rewardees): The lottery is negatively associated with the total number of rewardees that a crowdfunding project attracts. H3 (No. of Donors): The lottery is negatively associated with the total number of donors that a crowdfunding project attracts. H4b (Total Money): The lottery is negatively associated with the total money raised by a crowdfunding project. H5b (Success): The lottery is negatively associated with the probability of reaching the funding goal of a crowdfunding project. Yes Yes Yes Yes Yes Cannibalizing Incentivizing Cannibalizing Cannibalizing & Incentivizing Cannibalizing & Incentivizing 4 Given the attractive properties of linear probability models and that the marginal effect of Logit models with fixed effects is practically inapplicable, we also conducted a linear probability model as suggested by Horrace and Oaxaca (2006). Again, the effect of the lottery on a project s probability to reach the funding goal is similar to those reported in Tables 2 and 3 (the marginal effect of the LPM is -10%). Thirty Eighth International Conference on Information Systems, South Korea 2017 12
Robustness Checks: Alternative Plausible Explanations Unobserved Project Quality A potential concern is that projects with a lottery may differ from those without a lottery in terms of their quality, which is unobserved in our context. Suppose lower (or higher) quality projects are more likely to use the lottery to attract backers. Then projects with a lottery should have fewer (or more) backers, including rewardees, donors, and non-rewardees, because of their lower (or higher) quality. Our main results showed that projects with a lottery receive fewer rewardees and donors, but more non-rewardees. Therefore, project quality cannot fully explain the effect of the lottery, and it is unlikely to be an alternative driver of the results. To further rule out the concern of unobserved project quality, we conducted two additional analyses by examining whether projects with a lottery receive more (or fewer) likes and shares. On the focal platform, backers who viewed the projects could like the projects with a simple click or share the projects on their social network sites. If a project is of higher quality, it should also receive more likes and shares on social media. We conducted similar analyses to the main analyses in prior sections using both the number of likes and the number of shares a project receives as the dependent variable. Results indicate there is no significant effect of the lottery on either the number of likes or the number of social media shares. Therefore, it is unlikely that unobserved quality is a confounding factor when estimating the effect of the lottery on crowdfunding outcomes. Unobserved Fundraiser Effort Another alternative explanation of the effect of the lottery is unobserved fundraiser effort. It is possible that fundraisers who utilized a lottery in their projects have a stronger desire to raise money and thereby spend more effort, which may affect the crowdfunding outcomes. Although the fundraiser fixed effects model addresses this concern to some extent, concerns may remain, as a) fundraiser fixed effects could not fully account for time-variant fundraiser effort, and b) only a modest number of fundraisers appear twice or more to raise funds. Since the crowdfunding platform provides a very structured description for fundraisers to describe projects, making a video description is unstructured and relatively more time consuming, thus requiring more effort (Liu et al. 2014) compared to other activities, such as filling out a structured text description. Meanwhile, updating the status of the project during the fundraising period requires consistent time and effort and is thus a good proxy for fundraiser effort. We tested whether projects with a lottery differ from those without a lottery in the total number of updates and probability that a project has a video description. If fundraisers in the treatment group tend to spend more effort, then projects with a lottery should have more updates and be more likely to have a video description. And we do not observe a statistically significant association between lottery adoption and whether a project has a video description or whether the log total number of updates is affected. Therefore, we find that it is unlikely that fundraiser effort is a key driver of lottery adoption. Omitted Crowdfunding Platform Marketing Effort Another factor that may affect crowdfunding outcomes is marketing effort, such as advertisements, placements on front-page, and ranking effects, which are observable to backers, but unobservable in our data. Since the lottery is a feature introduced by the crowdfunding platform to incentivize backer engagement, it is possible that the crowdfunding platform may prioritize projects with a lottery by assigning a higher ranking in backers search results, displaying those projects on its front page, or promoting them using display ads, meaning that the observed lottery effect may be driven by such unobserved marketing efforts. Similar to the argument for project quality, if the platform intentionally exerts more marketing effort on projects with a lottery, we should observe similar (either positive or negative) effects on all types of backers. However, as shown earlier, the lottery has a negative effect on the number of rewardees and donors, but a positive effect on the number of non-rewardees. Therefore, the findings that we observe are unlikely to be driven by unobserved platform marketing effort. To further explore this alternative explanation, we ran another falsification test on the number of new backers a project receives. If the platform puts more marketing effort into projects with a lottery, those projects should receive more new backers who are without prior experience with the platform and are most likely to be affected by the platform s marketing effort. Again, there is no significant difference in the number of Thirty Eighth International Conference on Information Systems, South Korea 2017 13