Related research and conceptual framework

Charitable giving is an active part of behavioral and experimental economics research and has flourished in the past ten years, with the focus on fundraising recently reaching a fever-pitch (Andreoni, Reference Andreoni2016). This paper contributes to the debate over the advisability for public economists to help fundraisers increase donations (Auerbach et al., Reference Auerbach, Chetty, Feldstein and Saez2013), as well as how reliable behavioral insights are in context. In particular, we focus on how donations to charities are influenced by reference points suggested to individuals through donation menus. While the pertaining marketing and business literature is that of reference pricing and scale design, for behavioral and experimental economics it is that of the anchoring effect (Tversky & Kahneman, Reference Tversky and Kahneman1974; for a comprehensive literature review, see Furnham & Boo, Reference Furnham and Boo2011).

In charitable fundraising, asking for a specific amount to be donated (‘anchoring’ the donor to a set sum) is a common yet not fully disentangled matter. The first systematic assessment of donation menus was done by Weyant and Smith (Reference Weyant and Smith1987). Their work became seminal, even if it was, in some respects, inconclusive due to low response rates. Their field experiments, using direct mailing and controlling for different menu scales serving as exogenous reference points, showed a significant increase in donations.

More recently, three field experiments have taken further steps toward understanding the impacts and limits of the ‘power of asking’. First, De Bruyn and Prokopec (Reference De Bruyn and Prokopec2013) ran a large field experiment on donation menus. Manipulating the first amount on menus and the rate of increase of the other amounts, they found that anchoring had a positive effect on donations, but the donation likelihood was, in some instances, negatively affected. Second, Edwards and List (Reference Edwards and List2014) ran a field experiment in which they prompted prospective donors with a $20 suggestion. This amount turned out to be lower than the average baseline donation, which translated into an increase in the number of subjects donating, as well as a strong treatment effect with donations moving toward the suggested amount. The authors also hypothesized that suggesting higher donation amounts might reduce the likelihood of donation, as is supported by the findings of Warwick (Reference Warwick2003) and Weyant and Smith (Reference Weyant and Smith1987). Third, Goswami and Urminsky (Reference Goswami and Urminsky2016) investigated how defaults (preselecting one option in the menu) affect donations. They found three main effects: a lower-bar effect, in which defaulting a low amount increases the donation likelihood; a scale-back effect, in which low defaults reduce average donations; and a default-distraction effect, in which the mere presence of defaults reduces the influence of external cues, such as positive charity information.

Finally, we would like to clarify that, since the publication of Richard Thaler and Cass Sunstein's book Nudge: Improving Decisions about Health, Wealth and Happiness (Thaler & Sunstein, Reference Thaler and Sunstein2008), the design and contextualization of options has been given the byname of ‘choice architecture’. Therefore, much of the previous experimentation would be classified today as a nudge, but it was not originally published under this term. Authors published their findings using terms such as the effects of appeals scales or price scales, attraction effects and reference price tests. Nevertheless, as the nudge approach has become an established tool for policy-makers and practitioners, it would be desirable to include past findings and present research under the current and accepted umbrella of ‘nudging’.

RCT design and participants

In June 2016, driven by the necessity of counteracting the ‘summer drought’ of donations, a large charitable organization supporting cancer research in Italy decided to implement the principles of choice architecture to nudge over 150,000 of its existing donors. The charity explained that the average donation made to the organization was close to €19 and that their aim was to increase both the number of donations and the average amount donated during the summer months in order to increase the total amount collected over the period. The nudge was embedded in a standard paper-based campaign, where each donor received an envelope containing a two-page letter (in Appendix A, an anonymized version of the letter is provided).

For the purpose of the RCT, the second page, hereafter referred as the ‘donation sheet’, consisted of a pre-cut piece of A4 paper divided into two sections: on the left was the call to action and a donation menu (for the menu treatments), while on the right was the postal payment slip. At the back of the donation sheet were illustrated past projects and results funded by donations, as well as a reminder of the other channels for sending donations (namely credit card, direct bank transfer and recurring monthly donations via direct debit orders).

The research team, in close collaboration with the charity, manipulated the donation sheet. The scope of the intervention was to study the impacts of different appeals and menus on donations. This was particularly meaningful for the charity since they had never included any sort of donation menu in past direct mailing campaigns.

In the control treatment, the donation sheet, consistent with past fundraising campaigns, included a plain recall to the main letter and three pictures of people who had benefited in the past from the charity's work and survived cancer, together with a short quote by one of the beneficiaries and his/her first name. The ‘baseline treatment’ (BT) involved 39,909 donors, who were randomly drawn from the pool of trial subjects. The photographs and statements (included and identical in all treatments) did not link the individual to a specific project or initiative of the charity, but rather were general accounts of people who had fought and survived cancer thanks to the charity.

In line with the charity's specifications, two menus were designed and the suggested amounts were discussed and agreed upon by both the charity and the research team.

The first treatment (low or T1), involved 39,890 subjects, and it was identical to BT, except for the addition of a list of three suggested donation amounts.Footnote ¹ The suggested amounts were €25, €50 and €100. These numbers were also recalled in a box on the postal payment slip. The amounts, as stated above, were chosen in agreement with the charity, following their indications and requirements. The lower bound menu option (€25) was chosen because it was above the average past donation (€18.91) and a round figure.

The second treatment (high or T2) involved 39,890 subjects and was identical in all aspects to T1, except for higher suggested donation amounts: €50, €100 and €200. In this case, the lower bound menu option was chosen to be double the lower bound option in T1.

Additionally, the charity expressed the desire to test whether a different approach to payment methods could improve redemption rates. As stated above, on the back of the postal payment slip was fine-printed a reminder of the other possible donation methods (namely credit card, direct bank transfer and recurring monthly donations via direct debit orders). However, some methods were omitted: SisalPay (paying the postal slip at newsagents, coffee bars and tobacconists using the infrastructure of a betting and gaming company), YouPay (an app and web portal for paying the postal slip using a credit card) and banking and home banking options using postal slips. In addition, potential donors were not saliently reminded that, in order to donate using the postal payment slip, one had to physically go to the nearest post office.

For the purpose of responding to the specific request of the charity's practitioners, we drew from the Behavioural Insights Team (BIT) nudging approach. The BIT was established in 2010 by the UK government to apply the lessons of behavioral economics and psychology to public policy-making, but has since then moved toward a broader social purpose, helping charities, non-profits and international organizations. The current framework used by the BIT is the EAST framework (make it Easy, Attractive, Social and Timely). This method stems from the behavioral literature, but has been perfected through the BIT's own field experiments and experience. Of the four principles for influencing behavior, we chose the ‘make it easy’ one: small, seemingly trivial details that reduce a task's friction or costs can make the difference between doing something or postponing it, perhaps indefinitely. Conversely, even minor increases in the effort required can be detrimental. The BIT has done extensive research on donation channels (BIT, 2013), also testing default options and prompted choices. Related to our work, interesting results stemmed from another BIT line of research. They used a simple SMS reminder providing instructions on how to declare a business in order to increase company registration in Mexico. They found that adding trivial but salient information on how to register increased declaration rates by 24% (BIT, 2017).

Given the charity's needs and considering the relevant literature, we designed and tested a message that could increase the ease and convenience of donating. This translated into a third treatment. An additional group, consisting or 39,890 donors, was reached by a donation sheet identical to the one of BT, but with a salient reminder on how to donate to the charity. We called this treatment the ‘Easy Payment’ (T3), and it was trivially, yet saliently, stated that in order to use the postal payment slip one had to go to a post office. In addition, all of the available donation methods were listed.

Table 1 summarizes the attributes of each single treatment (in Appendix B, an anonymized version of the donation sheet is provided).

Table 1.

Treatment attributes.

The entire sample consisted of 159,579 subjects, of whom 8081 responded to the call for a summer donation (5.06% overall response rate). Of these, 4143 (51.27%) were female, 3859 (47.75%) were male and the remaining 79 (0.98%) were donors registered as families. Across treatments, the seniority (i.e., how many years the subject had been a benefactor of the charity) was normally distributed, and the average years as a donor was 14.5, with the most senior donors having donated for over 44 years.

Six months after the launch of the campaign, the data were reported back to the research team. The data set included not only the donation amount, but also the response time (i.e., how long it took for the donor to donate after receiving the letter), the number of years the donor has been donating to the charity, the number of past donations, the cumulative sum donated to the charity, the donor's gender and the method of payment.

Behavioral predictions

Our approach to predicting behavior in the RCT assumes that people donate for a variety of reasons (e.g., loyalty to friends or family, support for a social cause, tax benefits), but these can be nudged in order to redirect biases or remove obstacles to action, thus improving the levels of giving (for a comprehensive review on nudges and donor behavior, see Parbhoo et al., Reference Parbhoo, Davis, Reynolds, Piyush, Trewn and Welch2018).

In line with the previous literature, since both donation menus suggest higher amounts than the average donation to the charity (€18.91), we expect to observe a decrease in redemption rates, but an increase in average amounts donated.

This expectation stems from the robustness of the anchoring and reference dependence effects. Fraser et al. (Reference Fraser, Hite and Sauer1988), in a seminal field experiment, observed that response magnitudes can be improved by employing a moderately large but acceptable anchor point request. According to the theory of cognitive reference points, the higher the reference, the greater the donation amount. However, Smith and Berger (Reference Smith and Berger1996) proposed, but did not test, that the magnitude of the anchor would work in different directions: greater suggested donations would yield higher average gifts, while smaller suggested donations would generate higher redemption rates. De Bruyn and Prokopec (Reference De Bruyn and Prokopec2013) reported the results of the first field experiment on the adverse consequences of the anchoring effects on donation likelihood: although anchors increased donation amounts, redemption rates dropped, negatively affecting overall donations.

There is a common saying among fundraisers: ask for too little and you will get what you asked for, ask for too much and you will get nothing (Andreoni, Reference Andreoni2016). This phenomenon is related to the ‘power of the ask’, which has been investigated by Andreoni and Rao (Reference Andreoni and Rao2011). Their research aimed to reconcile prior laboratory findings of subjects exhibiting significant altruism and inequality aversion with real-world behavior, where great levels of inequality are voluntarily counteracted by minimal and insufficient levels of altruism (i.e., charitable giving). The intuition of the authors was that, since giving in the real world occurs in a complex context of social interaction and not in an impersonal setting, the mediation of social cues should be brought back into the laboratory. Social cues and incentives can potentially conceal the altruistic potential of individuals and transform it into real acts of kindness or, alternatively, into selfish behavior.

Using a dictator game and manipulating the possibility of sending a message to the counterpart, Andreoni and Rao (2010) discovered that if no communication – a typical social cue – was allowed, 15% of the endowment was donated, while anytime the recipient spoke, at least 24% of the endowment was passed on. The ‘power of the ask’ experiment also showed that asking for more generally induces greater donations, but with limits. Subjects who asked for more than an equal division of the pie (50% of the endowment) were punished: ‘selfish requests’ led to a mean donation rate that was less than half of the mean in the no-communication condition.

Thanks to the information included in the charity's database, we could also extend the research to some of the individual characteristics of donors. The additional information related to the number of years an individual was an active donor and the total number of donations made during this time. From this information, we inferred the average yearly donation frequency for each donor (number of donations over years as a donor). Generally speaking, donors with a longer history as patrons have been retained by the charity because of overall satisfaction, identification with the cause, a high level of commitment and trust (Sargeant, Reference Sargeant2008). The implication is that mature donors have already given many times in the past and have a stronger internal reference point serving as an anchor. Therefore, we can expect that subjects with a longer donation history would not be affected by the nudge as much as junior donors. In addition, Shang and Croson (Reference Shang and Croson2009) have suggested that social information is more likely to have an effect in ambiguous circumstances, which would be more relevant to inexperienced donors.

These issues have already been addressed in past studies. For example, Desmet and Feinberg (Reference Desmet and Feinberg2003), De Bruyn and Prokopec (Reference De Bruyn and Prokopec2013) and Lee and Feinberg (Reference Lee and Feinberg2013) looked into donation menus including both internal and external reference points. In particular, Desmet and Feinberg (Reference Desmet and Feinberg2003), retracing the prior literature, hypothesized that donations are based on an internal reference point in relation to the donation menu proposed (external reference point). For example, if a subject intends to donate €20 to the charity (internal reference point) and he or she receives an appeal for €40, they may take one of four possible courses of action in relation to their internal reference point: donate more, donate less, do not donate at all or gift exactly €20.

However, extrapolating internal reference points is not a straightforward exercise in field experiments. Contrary to what happens in laboratory experiments, where beliefs and preferences can be elicited using standardized methods, RCTs cannot rely on the same tools (we could not ask donors to state their gifting intentions before opening the envelope). The previous literature points out that the closest proxy to internal reference points is past behavior. It could be argued that past behavior does not necessarily reflect internal reference points, since they could have been influenced by various factors that might shifted them from the true internal reference point. Furthermore, other previous studies (Rooney et al., Reference Rooney, Steinberg and Schervish2004) have found that donors cannot recall exactly how much they contributed to a cause in the past. In our case, we used past average donation as a reasonable proxy for the internal reference point. We also argue that donation frequency might have an effect on how donation menus perform as nudges.

To sum up, the main purpose of this study was to validate the effects on donation likelihood and average amounts donated of two different manipulations of the donation menu (low and high) of a direct mailing fundraising campaign. An additional goal was to test whether individual donor habits act as moderating factors on nudges. Moreover, we investigated whether formulating trivial yet salient supplementary information on how to donate would increase donation likelihood.