1 Introduction

Charitable giving, public good provision, and crowdfunding all have one thing in common: agents give money to finance a nonprivate good. The main difference between the three lies in the nature of the good for which money is being collected. While the beneficiaries of charitable giving are typically other people and the beneficiaries of public goods are by definition everyone, the beneficiaries of many crowdfunding campaigns often include the contributors. Neither charitable giving nor public good provision mechanisms typically involve rewards for donors; by contrast, crowdfunding campaigns often involve nested reward schemes for different contributions.Footnote ¹

Regardless of the nature of the beneficiaries of a funding mechanism, the question arises as to how to describe to potential contributors the act of giving money. In public good games, it is common to refer to the money that is given as a ‘contribution,’ while in charitable giving settings, money given is mostly called a ‘donation.’ In this paper we explore whether this choice of wording matters for behaviour. While we do this in the context of a crowdfunding campaign,Footnote ² we believe that our results also have implications for other funding mechanisms.

Specifically, we implemented a crowdfunding campaign to finance one occurrence of an institute’s annual summer party. In previous years, a ‘donation box’ had been displayed in a prominent location during the party, which frequently led to shortfalls in financing. This time, a crowdfunding campaign was announced around 20 days in advance via personalized e-mails. The campaign offered a multitude of incentives to increase giving. Those incentives were available to all e-mail recipients alike and included rewards like vouchers for tournaments and games and matching for early gifts. Three e-mail reminders were sent. Our setting was a relatively small community consisting of an institute’s more than 500 affiliates and friends; expected attendance at the party was between 150 and 200 guests.Footnote ³

We implemented a subtle treatment manipulation in the wording of our e-mails, that is, we referred to either donations or contributions. In order to learn more about the mechanism driving giving behaviour in both conditions, we also varied non-binding suggestions, which were either €10 or €20. This was the second dimension of our 2 × 2 design. Additionally, we studied the responsiveness to other incentives offered (without experimental variation) depending on the frame. Specifically, we analysed gift levels relative to the reward thresholds and self-selection with respect to the timing of gifts. Regarding timing, early gifts were matched with a fixed amount offered by an anonymous sponsor, but later gifts were not matched.Footnote ⁴

While the term ‘donation’ has a clear meaning linked to charitable giving, ‘contribution’ has multiple meanings. Some of these meanings relate to charitable giving, but others are more related to duties. In Fig. 4 in the Appendix we present word association maps that show different meanings and their connections. They suggest that the act of donating is more self-oriented, while contributing invokes a notion of joint participation. On Google Trends, search terms combined with the word ‘donation’ mostly relate to charitable giving (blood, organ, plasma, Goodwill, Salvation Army, Red Cross, clothing), while those combined with ‘contribution’ mostly relate to individual accounts, savings, or insurance and ask questions about their regulation (see the lists in Table 9 in the Appendix). The search term ‘donation’ is approximately 20% more common than ‘contribution,’ and ‘charitable donation’ occurs 71% more often than ‘charitable contribution’ in Google searches.Footnote ⁵ While charities predominantly use the term ‘donation,’ they do employ the term ‘contribution’ in some instances (see Table 10 in the Appendix).

Relatedly, in the literature on charitable giving, Andreoni (Reference Andreoni1995) documented that framing the same task as implying a positive externality rather than a negative externality generates more giving. This result has been replicated several times by, among others, Sonnemans et al. (Reference Sonnemans, Schram and Offerman1998) and Park (Reference Park2000). This line of research concludes that positive frames are more successful at stimulating warm glow than negative frames.Footnote ⁶

In line with the above-mentioned research, we expected that the more unique meaning of the term ‘donation’ and its connotation with voluntary charitable giving would increase giving by intensifying warm glow when compared to the term ‘contribution.’ Given the different connotations, we also expected that gifts in the contribution frame would be more responsive to suggestions and rewards than those in the donation frame.

In order to further investigate the reasons for our treatment effects, we conducted an additional survey experiment on Amazon Mechanical Turk (MTurk), where we measured emotional responses to the two frames. In line with our conjecture, we found more positive emotional responses to the donation frame and also show that emotional responses are correlated with behaviour in a public good game—highlighting that the main result from our field experiment extends beyond the crowdfunding setting.

Our paper contributes to three strands of literature: the large literature on framing effects (including the papers cited above), the literature on emotions and economic decision-making, and the nascent literature on crowdfunding that emerged in the 2010s. Mollick (Reference Mollick2014) provides an early descriptive study of almost 50,000 crowdfunding projects, and Agrawal et al. (Reference Agrawal, Catalini and Goldfarb2014) provide an early overview of the basic economic principles governing the crowdfunding market. Strausz (Reference Strausz2017) contributes a formal economic model of crowdfunding highlighting the tension between its screening function in the presence of demand uncertainty and moral hazard. Belleflamme et al. (Reference Belleflamme, Omrani and Peitz2015) studied the economics of crowdfunding platforms and illustrate how externalities between crowdfunding projects lead to the natural emergence of platforms.

While most existing empirical studies of crowdfunding make use of observational data (for example, Meer, Reference Meer2014, Argo et al., Reference Argo, Klinowski and Krishnamurti2020), there are also a small number of experiments on crowdfunding. Cason and Zubrickas (Reference Cason and Zubrickas2017, Reference Cason and Tabarrok2020) and Cason, Tabarrok, and Zubrickas (Reference Cason and Tabarrok2020) conducted laboratory experiments in which they explored different incentive schemes such as bonuses for early contributions. Similarly, in a web-based experiment, Ansink et al. (Reference Ansink, Koetse, Bouma, Hauck and van Soest2017) tested the effects of seed money and the impact of the attraction effect. In a field experiment, Burtch et al. (Reference Burtch, Ghose and Wattal2015) studied the effects of privacy. Castillo et al. (Reference Castillo, Petrie and Wardell2017) show how including a feature to ask friends on Facebook for additional donations increased giving on a crowdfunding platform. Our study is the first to document a substantial framing effect in crowdfunding campaigns.

In addition to the literature on framing, the literature on nudging often investigates the effects of small changes in the design of choice architecture or in the precise choice of wording. A recent meta study of the effects of nudging in the context of tax collection can be found in Antinyan and Asatryan (Reference Antinyan and Asatryan2020). They show that emphasizing deterrence is more effective than emphasizing tax morale. In contrast to standard nudging interventions, we should also note that our variation is extremely minimal, akin to a one-word nudge. In addition, the terms that we used—‘donation’ and ‘contribution’—have a similar meaning, do not change the information structure, and do not affect the choice architecture.

As the effect that we document appears to be mediated through associated emotions, our paper also makes a contribution to the literature on emotions and economic decision-making. The role of emotions for contribution games was documented earlier by Drouvelis and Grosskopf (Reference Drouvelis and Grosskopf2016), who show that anger reduces contributions in a public good game with punishments. While they manipulated emotions before the actual decision through video clips screened to subjects,Footnote ⁷ the emotional response in our subjects was triggered simply by different wordings in the instructions and measured in our online survey experiment. Emotions have also been shown to be relevant in other contribution contexts, such as tax compliance (Enachescu et al., Reference Enachescu, Olsen, Kogler, Zeelenberg and Breugelmans2019) and pro-environmental donations (Ibanez, Moureau, and Roussel, Reference Ibanez, Moureau and Roussel2017), with positive emotions being associated with better compliance.Footnote ⁸

In light of these studies, it is perhaps not very surprising that emotions also matter for contributions in a crowdfunding campaign. What appears non-trivial is that different emotions can be triggered through the slightest change in wording.

The remainder of the paper proceeds as follows. In Sect. 2 we present the basic idea and our hypotheses. In Sect. 3 we describe the design and implementation of the crowdfunding campaign, followed by the results from the field experiment in Sect. 4. Sect. 5 presents the additional experiment on MTurk measuring emotional responses, and Sect. 6 concludes.

2 The basic idea and hypotheses

3 Design and implementation of the crowdfunding campaign

Each year one of the departments of the research institute is responsible for the organization of a summer party. The fields represented at the institute include sociology, political science, law, and economics. In 2016, the department of economics was responsible for the organization and financing of the summer party.Footnote ⁹ As usual, almost 550 employees, guests, and affiliated researchers were invited. Around half were employed as researchers (including PhD candidates and student research assistants), one quarter worked in administration, and the final quarter was made up of guests, affiliated researchers, alumni, and friends. The party usually involves free drinks and a barbecue or alternatives financed through monetary contributions, a salad and cake buffet organized through in-kind contributions, live music, and an entertainment program with games and humorous speeches.

Instead of employing a donation box, which in previous years had led to shortfalls in financing, this time the invitation e-mail announced a crowdfunding campaign to take place before the summer party. More specifically, there were four different versions of e-mails sent out 20 days before the party. A 2 × 2 design involved one treatment pair with a variation in wording and one pair with two different suggestions regarding the gift amounts. The e-mail recipients were asked to either contribute or donate money and/or make a pledge to a potluck buffet of salads and cakes (a buffet pledge).Footnote ¹⁰ In addition, suggestions were introduced in the first e-mail with the following sentence: ‘If the average monetary donation (contribution) is €20 <€10>, we need 100 < 200 > participants in the campaign to cover the expected costs.’ The same sentence was repeated in the final reminder e-mail. This formulation mirrors the variations in Adena, Huck, and Rasul (Reference Adena, Huck and Rasul2014). The total amount collected to date was posted and updated once a day on the institute’s intranet as well as communicated via reminder e-mails over the course of the campaign.

We also implemented some additional incentives that were equal for all versions of e-mails and aimed at making participation in the campaign more attractive. First, we offered various nested rewards by levels of gifts, with thresholds at €5, €10, €20, €30, and €100. The rewards included vouchers for participation in tournaments and games, and a rare book for the highest gifts. A buffet pledge was valued at €10, the average price that the organizers would have had to pay to a professional caterer for a cake or salad, and added to the monetary gift when determining the reward. Second, we offered a fixed match of €5 by an anonymous sponsor for early gifts; this was not counted towards the reward. In addition, it was announced that any surplus money would be donated to a refugee project (see Appendix D for details of the mailing). In addition to the first e-mail, three reminders were sent. The e-mails were sent in English,Footnote ¹¹ since a large proportion of the institute’s staff is international and has little or no command of the local language.

In the donation treatment, the word ‘donation’ appeared 19 times in the first e-mail, once in the first (short) reminder, twice in the second reminder, and four times in the third reminder, whereas the word ‘contribution’ was never used. Each time the e-mail was sent, all previous e-mail communications were appended such that with the third reminder the total word count of ‘donation’ was 26. The contribution treatment involved the same number of instances of the word ‘contribution’ and no use of the word ‘donation.’

We implemented block (strata) randomization based on the available individual characteristics, which in turn were based on membership in email lists such as ‘female,’ ‘postdocs,’ ‘PhD students,’ and those for different departments or different administrative divisions, amongst others.Footnote ¹² More specifically, we sorted the data according to the following dummy variables and in the following order: professor, female, data management unit, press and communication unit, doctoral students, postdocs, units IV, I, II, III, V, administration, secretaries, IT unit, student research assistants, and library. Next, in each consecutive group of four individuals (our blocks), we assigned one of the four experimental treatments at random.Footnote ¹³ We applied the block randomization in order to increase balance and subsequently precision. All variables used for the randomization and mean comparisons between different treatments can be seen in Table 7 in the Appendix. The given sample size of 545 individuals allows us to detect a standardized effect size (Cohen’s d) of 0.24 with alpha equal to 0.05 and power equal to 0.8 in a simple randomized experiment. By applying block randomization, we additionally increased power and therefore efficiency by reducing the residual variance.Footnote ¹⁴

By choosing personalized e-mails, we aimed to reduce spillovers between treatments. We cannot rule out that recipients discussed the party with one another. But since the differences between e-mails were rather subtle, they likely went unnoticed, and no one mentioned to us that they had become aware of the variation. If there was some awareness about treatment differences, for which we do not have any evidence, our results would constitute the lower bound of the true treatment effects.Footnote ¹⁵

Before proceeding to our results in the next section, we want to briefly reflect on how our setup compares to other funding mechanisms. In Table 1 we compare typical crowdfunding, typical public good games, and typical fundraising environments to our own experiment. While our setup does tick all the boxes for crowdfunding, it is closely related to both fundraising and public good games such that we would expect our results to also speak to other realms.

Table 1

Differences between crowdfunding, public goods, fundraising, and this experiment

	Public goods	Fundraising	Crowdfunding	This experiment
Beneficiaries	Everyone	Other people (everyone for certain charitable goals)	Contributors (other people in donation-based form)	Contributors (everyone at the institute)
Goods or services in return for payment	No	Typically no, but can include a lottery or small gifts	Typically yes, rewards possible	Yes, rewards included
Threshold	Typically no, but can be spelled out	Typically no, but can be spelled out	Typically yes, usually provided but not always binding (for example, JustGiving, betterplace.org)	Yes, implicitly spelled out; not binding but effectively affecting the amount of good provided
Visibility of amounts collected so far	Typically no, but can be spelled out	Typically no, but can be spelled out	Yes, usually provided	Yes, provided in reminder e-mails and updated once a day on the intranet

5 An additional experiment on MTurk measuring emotional responses

In order to parse out the mechanism behind the differences in behaviour in our two different frames, we conducted an additional survey experiment with 985 participants on the MTurk platform.Footnote ²⁰ Subjects were placed in an artefactual situation in which they were asked, depending on the treatment, to ‘donate’ or to ‘contribute’ to a public good. We subsequently measured their feelings using the Geneva Emotional Wheel (GEW).Footnote ²¹ The GEW measures 20 different emotions that are organized on a circle. The two main dimensions of the circle reflect the extent to which emotions are aligned with feelings of being in control (the vertical axis) and the positivity or negativity of emotions (the horizontal axis).

We implemented one-shot public good games with staggered rewards at a number of thresholds as in our summer party crowdfunding campaign. Subjects played in groups of five, and each subject had an endowment of US$2 from which they could choose how much to donate or contribute to a group account. Payments into the group reaching a threshold of $5 were doubled and shared equally among all subjects; payments below that threshold were not doubled but still shared equally. Payments that exceeded certain thresholds were met with a symbolic reward and an individual rebate. Specifically, at $0.20 subjects received a downloadable certificate called the ‘bronze contributor/donor recognition award’; at $0.40 they received a ‘silver award,’ at $0.80 a ‘gold award,’ and at $1.60 a ‘platinum award.’ Additionally, they were offered a rebate of $0.05 for payments above $0.40, $0.10 for payments above $0.80, and $0.20 for payments above $1.60. Notice that none of these rebates affects the equilibrium prediction of zero payments for selfish rational agents. Each subject received a baseline payment of $0.50, independent of the game outcome. After their choice, subjects were asked to quantify how strongly they experienced the 20 different emotions that feed into the GEW (see Appendix E for detailed instructions).

Average payments into the group account were close to $1.10 under both frames, with almost identically appearing distributions and no treatment effect, as documented in Table 6, Column I. There is however a difference in groups’ abilities to meet the $5 threshold that triggered group payments to be doubled. Under the donation frame, 82.28% of groups reached that threshold, compared to 74.17% under the contribution frame, with higher resulting payouts for the donation frame. While those differences are not statistically significant at conventional levels, we did find a significant treatment effect when examining emotions, and once we explore how emotions map onto payments we will see why there were no effects on choices in the MTurk setting.

Our measurements of emotional responses are presented in condensed form in Fig. 1 and in more detail in Table 11 in the Appendix. In Fig. 1, which shows the GEW, all emotion variables are standardized with mean zero and standard deviation equal to one, chosen because of stark differences on the scale between different emotions. It is easy to see that the contribution frame is associated with more negative feelings than the donation frame: the two frames are roughly two standard deviations apart across the entire left side of the GEW. In terms of positive emotions, the two frames generated much more similar responses, though donations are associated with stronger feelings of ‘love’ and ‘compassion.’

These results appear to be in line with what word maps and Google Trends had suggested: as the term ‘contribution’ implies far less voluntary sentiment and is more reflective of an obligation, it also evokes more negative emotional responses.

Fig. 1

‘Donations’ versus ‘contributions’ on the GEW. (Note: All emotion variables are standardized with mean zero and standard deviation equal to one. The dashed line presents the deviation of the mean in the donation treatment from the overall mean (in terms of standard deviations). The solid line presents the deviation of the mean in the contribution treatment from the overall mean)

In a second step, we completed two regression exercises. In Table 6, Column II, we regress a simple index capturing negative emotions (the sum of negative emotional responses, standardized) on the treatment, while in Column III we regress payments into the group account (standardized) on the negative emotion index and a treatment dummy. An interesting pattern emerges. As is to be expected from inspecting Fig. 1, we found a strong treatment effect on the negative emotion index. We also found that negative emotions significantly reduced contributions, in line with the findings of Drouvelis and Grosskopf (Reference Drouvelis and Grosskopf2016), discussed above. However, this relationship is significantly attenuated under the contribution frame. It appears that the subjects in our MTurk treatment tried to keep their negative emotions in check more so when we shocked them upwards in our contribution frame. It is this attenuation that leads to the absence of a treatment effect on average payments into the group account in this particular setting.

Table 6

Results of the MTurk experiment

Outcome:	Gifts, standardized	Sum of negative emotions, standardized	Gifts, standardized
	I	II	III
‘Contribution’	0.008 (0.064)	0.207*** (0.063)	0.051 (0.063)
Sum of negative emotions, standardized			-0.294*** (0.049)
‘Contribution’ x sum of negative emotions, standardized			0.190*** (0.065)
Observations	985	985	985
R 2	0.000	0.011	0.036

Note: Robust errors; ^* p < 0.10, ^** p < 0.05, ^*** p < 0.01

Our results suggest that economic decision-makers are influenced by their emotions but are not slaves to them. Many individuals participate in MTurk first and foremost to earn money. Still, they are prone to emotional responses that depend on the framing of their choice environment. But it appears that they are able to exert some control over the transmission of emotions onto choice. We presume that the strength of such attenuation is moderated by financial need and largely absent in our field experiment.