The Role of Provision Points in Online Crowdfunding

Abstract

Extending recent work on market mechanisms in new fintech offerings, we explore the implications of a key mechanism in online crowdfunding-the use of a provision point. Under a provision point mechanism (otherwise known as all-or-nothing or fixed fundraising scheme), the fundraiser, typically an entrepreneur, only receives funds pledged toward his or her campaign if a preregistered fundraising target is met, rather than keeping everything that is raised. Provision points may weaken contributors' reliance on prior capital accumulation for judging a project's potential for success, by eliminating their concerns about a partial fundraising outcome and by signaling the project or entrepreneur's quality. Yet, provision points may also induce attention to prior capital accumulation, because the materialization of one person's contribution depends explicitly on sufficient contributions from others (a network effect). We assess this tension empirically, leveraging proprietary data from a leading crowdfunding platform that allows entrepreneurs to opt into a provision point. We consider the effects of prior capital accumulation on visitors' conversion and contribution decisions, and the moderating influence of a provision point. We find that provision points weaken the association between prior capital accumulation and visitor contribution, implying a reduction in potential herd behavior.

Key words and phrases:

• crowdfunding
• fintech
• fundraising
• market mechanisms
• mechanism design
• provision point mechanism
• social proof

Notes

1. While crowdfunding campaigns do not necessarily produce public goods, they do share many characteristics with public goods [20, 39]. Specifically, crowdfunding projects often require a nontrivial up-front fixed cost before the first unit of outcome can be produced. Such a fixed cost is shared by contributors in a nonexclusive (every contributor benefits from it) and nonrival (its benefits does not decrease with the number of contributors) manner, akin to private provision of public goods.

2. The vast majority of campaign organizers in our sample executed just one campaign. Therefore, the campaign fixed effects are in essence also a campaign-organizer fixed effect. This implies that our fixed-effect framework accounts not only for time invariant features of a campaign that may influence selection for PPM use but also time invariant features of campaign organizers as well, such as awareness of the PPM feature or self-confidence.

3. In the presence of a PPM, a partial funding outcome can also impose costs on contributors. Although a campaign employing a PPM that ultimately fails to reach its target will refund all contributions, the contributors cannot use the money elsewhere until the fundraising process completes. This liquidity cost is arguably smaller than the cost associated with not getting one’s money back should a partial fundraising outcome take place, under a thresholdless campaign.

4. The high conversion rate we observe here is not surprising, for two reasons. First, ample work notes that a substantial portion of contributions arrive from friends and family members of the entrepreneur [2]. Second, what we characterize here as a visit is in fact a set of prefiltered visitors, in some sense. This is because we observe individuals who in many instances have already elected to click on the brief campaign description on the initial campaign listings page.

5. We thank the anonymous reviewer for pointing us to this practical evidence of selection related to provision point use.

6. We have explored the robustness of our findings to the use of alternative increments, including a vector of dummies reflecting 2 percent increments, and another reflecting 10 percent increments. In both cases, we find consistent results, which are available from the authors upon request.

7. Given the size of our sample and the statistical significance of our estimates, it is worth noting two points that lead us to believe the findings are meaningful. First, although our overall sample is large, the support for specific dummies that we estimate in our models varies considerably, with the majority of the percentage-raised dummies having relatively few associated observations. This is particularly true in our subsequent regressions that consider the dynamics of the relationships in question, wherein we split the data into duration terciles. In many instances, we have just a few hundred observations supporting each percentage-raised dummy. Second, the practical (economic) significance of the estimates is rather large. Given that we estimate a linear probability model, the coefficients can be interpreted directly as shifts in the probability of conversion between stages of capital accumulation and zero fundraising. Thus, shifts on the order of 2–8 percent in the probability of conversion are observed here, which is particularly large when we consider the scale of the platform in question (i.e., Quantcast estimates that the platform we study now regularly receives upward of 4 million visitors each month).

8. We assessed whether the addition of the interaction terms produced significant improvements in model fit, based on information criteria. The AIC in the naive model (column 1) is 129,256.7, whereas the AIC of the interaction model is 129,209.8, a great deal lower. Similarly, the negative log-likelihood of the naive model is 64,587.33, whereas that of the interaction model is 64,540.92. A decline in the AIC of at least 2 is typically sufficient justification to prefer an alternative model. Here, we observe a decline of more than 30, supporting a preference for the interaction model.

9. We assessed robustness of the estimation to outliers in the dependent variable by excluding observations in the top and bottom decile of the distribution, that is, contributions of less than $10 or greater than $125, and then repeating the estimation. These results are reported in Table A1 of the Appendix.

10. We also performed a matching analysis employing propensity score matching (PSM) with a first-stage logistic regression determining the propensity to receive a PPM. The results of these conversion and contribution models are reported in the Appendix, in Table A2 and Table A3. We observe similar results.

11. The only covariates for which we do not enforce exact matches are the project dollar goal, the project duration in days, and the duration of the visit. The former two covariates are not of particular concern, because they are subsumed by the campaign fixed effect in our regression. We performed a t-test on log(Visit Duration), comparing the mean weighted value between PPM and non-PPM campaigns following the matching process, and determined that they are not statistically significant at conventional thresholds (p = 0.071).

12. For the sake of exploration, we also conducted a set of cross-sectional regressions, estimating the campaign-level, direct relationship between PPM use and fundraising outcomes, in terms of dollars raised. Controlling for campaign goal, duration, number of rewards and campaign category, we too observe a significant positive relationship between PPM use and success. However, we caution against reading deeply into this result, given that PPM use, duration, goal and reward-setup are all endogenously determined by an entrepreneur, and quite likely to be spuriously associated with fundraising outcomes. A more detailed analysis is reported by Cumming et al. [15], who similarly report a positive relationship between PPM use and overall fundraising outcomes.

Additional information

Notes on contributors

Gordon Burtch

Gordon Burtch ([email protected]; corresponding author) is an assistant professor and Jim & Mary Lawrence Fellow in the Information and Decision Sciences Department at the Carlson School of Management, University of Minnesota. He is also a consulting researcher with Microsoft Research. He received his Ph.D. from Temple University. His research employs econometric analyses and field experimentation to explore the drivers and consequences of individuals’ participation in online social contexts. His work has been published in Management Science, Information Systems Research, and MIS Quarterly, among others. His work has received several best-paper awards.

Yili Hong

Yili Hong ([email protected]) is an assistant professor, codirector of the Digital Society Initiative, and Ph.D. Program Coordinator in the Department of Information Systems at the W.P. Carey School of Business of Arizona State University. He obtained his Ph.D. in management information systems from Temple University. He is an external research scientist for a number of companies, including Freelancer, fits.me, Extole, Yamibuy, Meishi, and Picmonic, among others. His research interests are in the areas of the sharing economy, online platforms, and user-generated content. He has published in Management Science, Information Systems Research, MIS Quarterly, Journal of Consumer Psychology, and Journal of the Association for Information Systems. His research has received a number of awards.

De Liu

De Liu ([email protected]) is an associate professor of information and decision sciences, and 3M Fellow in Business Analytics at the Carlson School of Management, University of Minnesota. He received his Ph.D. from the University of Texas at Austin. His research interests include digital auctions, gamification, crowdsourcing, crowdfunding, and social commerce. His research has appeared in leading journals such as MIS Quarterly, Information Systems Research, Journal of Marketing, Journal of Market Research, and others. He serves as an associate editor for Information Systems Research and the Journal of Organizational Computing and Electronic Commerce, as well as academic director for the MS in Business Analytics Program and Ph.D. coordinator for the Department of Information and Decision Sciences at the Carlson School.