PaperlayerView exact revision on arXiv

Source-linked AI summary

The Hidden Cost of Accommodating Crowdfunder Privacy Preferences: A Randomized Field Experiment

Gordon Burtch, Anindya Ghose, Sunil Wattal

arXiv:1408.4194v1cs.SIcs.CY

TL;DR

Online crowdfunding makes transactions more visible, raising questions about how privacy controls affect contributor behavior. Using a randomized field experiment, the paper finds that delaying those controls increases conversion while lowering average contributions, with privacy priming and publicity effects helping explain the trade-off. The results inform the design of online crowdfunding platforms, but their generalizability beyond crowdfunding is uncertain.

  • Problem

    The paper examines how transaction-level information controls affect crowdfunders’ willingness to contribute and subsequent contribution amounts.

  • Method

    The authors employ a randomized field experiment to study information-control timing and observe contributors’ real transaction decisions.

  • Results

    Delaying information controls increases conversion rates while simultaneously lowering average dollar contributions.

  • Takeaways & Limitations

    The findings identify competing positive and negative effects of privacy mechanisms that matter for crowdfunding-platform design.

  • Takeaways & Limitations

    The results may not generalize to other, non-crowdfunding contexts.

Abstract

from arXiv · show

Online crowdfunding has received a great deal of attention from entrepreneurs and policymakers as a promising avenue to fostering entrepreneurship and innovation. A notable aspect of this shift from an offline to an online setting is that it brings increased visibility and traceability of transactions. Many crowdfunding platforms therefore provide mechanisms that enable a campaign contributor to conceal his or her identity or contribution amount from peers. We study the impact of these information (privacy) control mechanisms on crowdfunder behavior. Employing a randomized experiment at one of the largest online crowdfunding platforms, we find evidence of both positive (e.g., comfort) and negative (e.g., privacy priming) causal effects. We find that reducing access to information controls induces a net increase in fundraising, yet this outcome results from two competing influences: treatment increases willingness to engage with the platform (a 4.9% increase in the probability of contribution) and simultaneously decreases the average contribution (a $5.81 decline). This decline derives from a publicity effect, wherein contributors respond to a lack of privacy by tempering extreme contributions. We unravel the causal mechanisms that drive the results and discuss the implications of our findings for the design of online platforms.

1. Introduction

This paper examines how transaction-level privacy controls shape whether crowdfunders contribute and how much they give. A randomized field experiment finds that delaying these controls increases participation but lowers average contributions, with publicity tempering extreme giving.

  • Crowdfunding increases transaction visibility and traceability, prompting platforms to offer controls that conceal contributors’ identities or contribution amounts.
  • The paper asks whether information controls causally affect willingness to transact and subsequent contribution behavior.
  • The study uses a randomized field experiment on a leading crowdfunding platform, observing real decisions, including users who do not transact.
  • 4.9% increase in the probability of completing a transaction followed delayed presentation of information controls.
  • $5.81 decline in average campaign contributions occurred among users who transacted under the delayed-control treatment.
  • Participation gains offset lower average contributions, producing an immediate net benefit for the platform and motivating permanent adoption of the post-payment setup.
  • The treatment reduced contribution variance, especially for large contributions, as contributors responded to greater publicity by tempering extreme amounts.
  • The paper contributes evidence on the competing effects of privacy features on conversion and conditional contributions, and links these effects to altruistic motives and publicity.

2. Methods: Randomized Experiment

The paper uses a randomized field experiment that delays an information-control prompt until after payment, comparing it with presentation before payment. The design estimates effects on conversion and contribution amounts using campaign and time fixed effects with additional controls.

  • Platform and setting: The experiment was conducted on a large global reward-based crowdfunding platform serving millions of dollars in monthly contributions.The platform had more than 1 million users from over 200 countries.
  • Platform and setting: Contributors could conceal either their identity or contribution amount from peers, while organizers and the platform operator could still see both.The mechanism could not hide both pieces of information simultaneously.
  • Experimental design: The treatment delayed the information-control question from immediately before payment until after payment, mimicking partial removal of the mechanism.The control condition presented the question before payment; the treatment condition presented it after payment.
  • Experimental design: The timing manipulation could reduce privacy priming before payment but might also affect willingness to transact when privacy concerns arise during contribution.Because these effects could oppose one another, the treatment’s fundraising impact was not theoretically clear beforehand.
  • Econometric specification: The analysis estimates treatment effects with ordinary least squares, campaign fixed effects, time fixed effects, and campaign, user, browser, language, country, and device controls.The treatment coefficient captures the effect on conversion rates, while the specification indexes users, campaigns, and days.
  • Outcomes: The study evaluates conversion, conditional contribution, and unconditional contribution across treatment and control groups.The reported figures compare these outcomes between the two experimental conditions.

3. Results

Delaying the privacy-control prompt increased participation but reduced contributions among those who converted. The combined effect was a net increase in fundraising because the conversion gain outweighed the decline in average contribution.

  • Mechanisms: The treatment generated competing effects because information controls can increase comfort while privacy-related prompts can prime scrutiny concerns.The paper frames these as countervailing influences on willingness to participate.
  • Mechanisms: The contribution decline is consistent with a publicity effect in which users temper extreme contributions when privacy is less salient.The authors anticipated that privacy awareness would make users more willing to engage in very small or very large contributions.
  • Participation: 4.9% increase in the probability of conversion followed treatment, consistent with reduced privacy sensitivity.The result appears in the treatment analysis of campaign contribution probability.
  • Contribution amount: $5.81 decline in average contribution conditional on conversion followed treatment.The decline indicates that the treatment affected contribution amounts among contributors, not only participation.
  • Overall fundraising: $3.55 increase in average contribution per visitor resulted because increased conversion dominated reduced conditional contributions.The paper characterizes this as a net benefit for the platform operator in overall fundraising.

4. Supporting Analyses

Supporting analyses indicate that treatment reduced dispersion in contribution amounts, especially for large contributions, and that publicity was central to the treatment effect. Effects were stronger for sensitive campaign topics.

  • Contribution dispersion: 21% decrease in deviations from the campaign average followed treatment.The analysis used logged absolute deviation to express the effect in percentage terms.
  • Contribution dispersion: Treatment significantly decreased the variance of contribution amounts across conditions.Levene’s and Brown–Forsythe tests also produced p < 0.001.
  • Interpretation: Together, the analyses support publicity as a central mechanism underlying the treatment’s effect on contribution amounts.The evidence links reduced variance and topic sensitivity to increased perceptions of publicity.
  • Information hiding: Contributions in both distribution tails were significantly more likely to use information hiding, with larger contributions almost twice as likely.The difference between the tail coefficients was statistically significant: F(1, 3581) = 6.92, p < 0.01.
  • Topic sensitivity: The treatment effect was much stronger for sensitive campaign topics.The study classified politics, religion, education, and environment as potentially sensitive categories.

5. Additional Analyses and Alternative Explanations

The analyses test whether interface complexity, reduced payment effort, mobile sensitivity, or organizer self-contributions explain the treatment effects. The results generally argue against these alternative explanations.

  • Alternative explanations: The intervention removed a payment-process radio button, raising concern that simpler UI could explain increased conversion.The authors identify reduced interface complexity and payment effort as a potential confound.
  • Alternative explanations: Treatment did not significantly shorten payment visits, weakening the reduced-effort explanation.Visit duration differences were not statistically significant; the t-test used logged durations and excluded visits over 1,500 seconds.
  • Alternative explanations: No positive treatment moderation appeared among mobile users, contrary to the prediction that mobile users would be more sensitive to UI changes.The result was described as directly counter to a UI-complexity explanation.
  • Self-contribution: The significant decline in average contribution amounts, together with these analyses, makes UI complexity unlikely to explain the results.The authors then examine whether publicity reduced organizers’ self-contributions.
  • Self-contribution: No evidence indicated that treatment reduced campaign organizers’ probability of supporting their own campaigns.This makes organizer self-contribution an unlikely explanation for the observed contribution effects.

6. Robustness Checks

Robustness checks repeat the main analyses after excluding influential observations, using alternative estimators, and restricting attention to recently registered users. These analyses produce results consistent with the primary findings and show no significant heterogeneity across contribution-related campaign characteristics.

  • Outliers: Excluding the top 5% of contribution amounts or funding targets left the results generally unchanged.These checks address possible influence from outlier observations and unusually large campaigns.
  • Alternative estimators: Alternative Logit, Probit, fixed-effects Poisson, and negative-binomial estimators produced results consistent with the primary analyses.The supplementary models report marginal effects for the alternative specifications.
  • New-user subsample: Among users who registered within the prior 24 hours, the conditional contribution treatment effect was roughly equivalent.New users were expected to be less aware of alternative information-control conditions.
  • New-user subsample: The authors conclude that the results are not driven by subjects’ awareness of alternative conditions.This conclusion follows from the recently registered-user subsample analysis.
  • Heterogeneity: Treatment effects did not significantly differ between high- and low-spend campaign categories or across median contribution size.The interaction analyses found no significant differences for either campaign classification.

7. Manipulation Checks

Manipulation checks confirm that delaying access to information hiding substantially reduced its use. The treatment effect was broadly consistent across campaign categories and project targets, with limited moderation in Video and Web campaigns.

  • Information-hiding manipulation: Information hiding appeared in approximately 47% of control contributions versus approximately 21% of treatment contributions.The authors present these rates as evidence that the intervention produced the intended downward shift in hiding behavior.
  • Campaign-category heterogeneity: No campaign-category moderation appeared except in Video and Web, where the treatment effect was attenuated (β = 0.106, p < 0.01).The authors attribute the attenuation to that category’s comparatively low baseline hiding rate.
  • Campaign-category heterogeneity: Video and Web campaigns had a control hiding rate of 0.33, compared with 0.48 for all other categories.The next-lowest rate was 0.41 in Theatre.
  • Project-target heterogeneity: For project target size, the treatment effect was β = -0.266 (p < 0.001), while the interaction was extremely small (B = 6.46e-10, p < 0.01).Replacing project goal with its log made the interaction completely insignificant.
  • Overall interpretation: The authors characterize the treatment effect as generalizable and not heavily dependent on campaign type.They also found no evidence of attenuation when users arrived after an anonymous contributor.

8. Managerial Implications

The findings imply that information controls create a tension between contributor recognition and privacy or publicity concerns. Platform design should therefore account for campaign context and consider supplemental ways to reassure or recognize contributors.

  • Platform design: Information controls should be designed carefully, with greater attention for platforms hosting sensitive or potentially controversial campaigns.The authors suggest that such campaigns may generate greater awareness and use of information-control features.
  • Context dependence: Platform design should be context-dependent because transparency, reputation, recognition, and campaign characteristics shape the relevance of privacy controls.The authors caution that the results’ applicability depends heavily on contextual factors.
  • Recognition and privacy: Recognition and privacy are inherently in tension, motivating supplemental approaches to mitigate privacy priming around information controls.Suggested approaches include privacy seals and reassurance alongside control prompts.
  • Recognition alternatives: Organizers could separate recognition from transaction visibility by rewarding participation through naming rights, website thanks, or effort-based contributions.These approaches may allow contributors to maintain obscurity while still receiving recognition.
  • Platform design: Information-related prompts can severely affect conversion rates and platform contributions by shaping users’ privacy perceptions.The authors connect this result to prior evidence that privacy-risk perceptions are driven by available cues.
  • Identity and profiles: More than one third of contributions involved information hiding, indicating that many crowdfunders value their user profiles.The authors use this observation to complicate the view that concerned users can simply adopt pseudonyms.

9. Conclusion

The study examines how transaction-level privacy controls affect online crowdfunder behavior, finding that reducing access increases conversion but lowers average contributions. The authors attribute these effects to competing comfort and privacy-priming or publicity mechanisms, while emphasizing contextual limits on generalization.

  • 9. Conclusion: Transaction-level information controls create a tension between users’ comfort or security and privacy priming during crowdfunding contributions.Online transactions increase visibility and traceability, while prompts can prime privacy concerns and withholding controls can reduce comfort.
  • 9. Conclusion: Reduced privacy makes contributors perceive greater publicity, leading them to temper extreme contribution amounts.The publicity effect provides empirical evidence that visibility can influence contribution behavior.
  • 9. Conclusion: Treatment effects may depend on mechanism wording, information-hiding granularity, and user-interface positioning.The authors identify alternative privacy-control designs, including discretized contribution ranges, as areas for future study.
  • 9. Conclusion: The findings may not generalize to ordinary purchases or contexts where social capital and reputation are less pronounced.The authors also note that fixed transaction amounts cannot offset participation declines through larger contributions.
  • 9. Conclusion: Large-scale in vivo randomized experiments can estimate online user-behavior treatment effects while addressing endogeneity concerns.The authors argue that randomized experiments are broadly applicable and useful for causal inference in complex online settings.

SUPPLEMENTARY APPENDIX

The supplementary appendix reports additional model specifications, robustness checks, and alternative-explanation analyses for conversion and contribution outcomes.

  • SUPPLEMENTARY APPENDIX: The appendix includes an alternative-explanation test examining mobile interaction effects on conversion.The analysis uses a linear probability model with campaign fixed effects.
  • SUPPLEMENTARY APPENDIX: Robustness checks separately examine conversion rate and conditional contribution outcomes.The reported specifications include conversion and contribution models with estimator and sample notes.

PROBIT LOGIT

The appendix presents probit, logit, contribution, and duration analyses using marginal effects, robustness specifications, and campaign-category comparisons. Supplementary figures compare logged visit duration across treatment conditions and device types.

  • PROBIT LOGIT: 0.075*** and 0.077*** are reported as model estimates under alternative specifications.The table notes probabilistic marginal-effect estimates and clustered or bootstrap inference across specifications.
  • PROBIT LOGIT: 500.66 and 417.67 are reported as outcome values for the compared specifications.The supplied table values are accompanied by a sample size of 112.
  • PROBIT LOGIT: The appendix reports robustness checks for conditional and unconditional contribution, recent joiners, and campaign categories.These analyses use contribution outcomes, recent-joiner samples, and campaign-category comparisons.
  • PROBIT LOGIT: Treatment and mobile groups exhibit systematically shorter visit durations, although the difference is not statistically significant.Figures S1 and S2 show clustered histograms of logged visit duration by treatment condition and device type.
Loading 1408.4194v1…