Scharf, K., & Smith, S. L. (2016). Relational altruism and giving in social groups. Journal of Public Economics, 141, 1-10. https://doi.org/10.1016/j.jpubeco.2016.06.001 Peer reviewed version License (if available): CC BY-NC-ND Link to published version (if available): 10.1016/j.jpubeco.2016.06.001 Link to publication record in Explore Bristol Research PDF-document This is the author accepted manuscript (AAM). The final published version (version of record) is available online via Elsevier at http://www.sciencedirect.com/science/article/pii/S0047272716300718. Please refer to any applicable terms of use of the publisher. University of Bristol - Explore Bristol Research General rights This document is made available in accordance with publisher policies. Please cite only the published version using the reference above. Full terms of use are available: http://www.bristol.ac.uk/red/research-policy/pure/user-guides/ebr-terms/ Relational Altruism and Giving in Social Groups Kimberley Scharf University of Warwick and CEPR Sarah Smith1 University of Bristol, IFS and CEPR May 2016 Abstract Much fundraising is done by individuals within existing social groups. Exploiting a unique dataset, we demonstrate (i) a positive relationship between social group size and the number of donations; (ii) a negative relationship between group size and the size of individual donations; (iii) no clear relationship between group size and the total amount raised. Free riding with respect to the activity being funded cannot explain the relationship between group size and donation size, since the number of social group members is only a subset of total contributors. Instead, the findings are consistent with the notion that giving in social groups is motivated by “relational altruism”. JEL: D64, Z1, H31 Keywords: Online giving; Fundraising; Social groups; Donations; Charity; Altruism Acknowledgements: Thanks to the editor and three anonymous referees, Luigi Guiso, Michael Price and participants at the Warwick Generosity and Wellbeing conference and the CESIfo Social Economics conference for very helpful comments and suggestions. 1 Corresponding author: [email protected] Department of Economics, University of Bristol, Bristol, BS8 1TN, United Kingdom. 1 1 Introduction Donations by individuals are an important source of income for charities – more than $200 billion is donated annually in the US and £10 billion in the UK – yet underlying individual motives for giving are not well understood. One possibility is that donors are motivated to give for altruistic reasons; that is, they care about the total amount of public good that is provided. Another is that donors give because they gain direct utility from the act of giving; that is, they experience a “warm glow” from giving (Andreoni, 1990). There is also interest in how donations may be determined in a “charity market” (List, 2011) where donors interact with other key players, including fundraisers and/or charities who are active in seeking donations through their own fundraising efforts. Yet another possibility is that donors are motivated by a personal relationship with a fundraiser. This may be important in practice given that charitable giving often takes place in social settings that are unrelated to the charitable activity, with the fundraising request coming not from a charity representative but from someone known to the donor. One example is individual-led fundraising, in which individuals engage in fundraising activities and ask their networks of friends, family and colleagues to make sponsorship donations to charity.2 Table 1 summarizes survey evidence on donors’ perceptions of the factors that were important determinants of how much they gave in response to an individual fundraiser, showing that the personal relationship between the donor and the fundraiser comes near the top – well above tax incentives, for example. In this paper, we propose the idea that, in social group settings, there may be a “relational altruism” motivation for giving. By this we mean a motivation to give that comes from a donor’s altruism towards a member of their social group who engages in individual fundraising activities and who experiences a warm glow from the amount of money that is raised from their fundraising effort. In practice, where fundraisers solicit donations from their existing social groups, the size of social groups – and therefore the size of the potential donor pool – is likely to vary. In our data, for example, where the number of Facebook friends is a plausible indicator of social group size, the 10th percentile of the distribution is 77 friends, while the median is 252 and the 90th percentile is 654 friends. Such variation may have implications for fundraising outcomes. We explore this conjecture theoretically and empirically by focusing on the relationship between the size of the fundraiser’s social group and donation behaviour in the individual fundraising context. 2 Individual-led fundraising is widespread in many countries, including the UK where 18% of donors report having sponsored family and friends for charity in the past year (source: Citizenship Survey, 2008‐09, Department for Communities and Local Government, 2009). 2 At first sight, the idea that group size should affect the behaviour of individual donors might appear to be a straightforward implication of non-cooperative models of giving (as developed by, among others, Warr, 1983, and Bergstrom, Blume and Varian, 1986), in which individuals make private contributions towards a pure public good and that predict that, as the number of contributors to a public good increases, the size of contributions will fall. Indeed, this is the relationship we observe in our data – in larger social groups, we observe more donations, but each donor tends to give less. However, the crucial difference between this case and our setting is that the size of the social group we observe does not correspond to the total number of contributors to the public good provided by the charity. For the charities in our sample, the amount of money raised by any individual fundraiser typically makes an insignificant contribution towards the overall level of charitable provision. The most popular charity for which people fundraise is Cancer Research UK for whom the £100 million received in donations each year completely dwarves the amount raised by any individual fundraiser. In our analysis, we show that social group size is negatively correlated with contribution size across different fundraisers who are raising money for the same charity. Since these fundraisers face the same number of total contributors, we need a different explanation to explain the negative relationship between social group size and contributions in our data. Our preferred explanation is that, if donors are motivated by relational altruism – i.e. if a donor cares about the fundraiser and the fundraiser cares about how much they raise – then the amount of money raised by the individual fundraiser becomes a “local” public good. In this case, as mentioned above, contributions fall with social group size, but the free-riding behavior that generates this outcome stems from altruism that the donor has towards the fundraiser who cares about how much they raise, as opposed to stemming from altruism on the part of the donor towards the charitable cause. There is circumstantial evidence that individual fundraisers do care about how much they raise – for example, the fact that the majority set aspirational fundraising targets that are notional (unrelated to specific funding needs) and non-binding (donations are passed to the charity even if the target is not met). It is therefore plausible that fundraising success is a local public good to the social group and thus a public good for which incentives will vary locally with social group size. Our empirical analysis exploits a unique dataset of individual-led fundraising activity that links the donations made to individual fundraisers’ online fundraising pages to an observable proxy for the size of the fundraisers’ social group. This is a very rich dataset comprising all donations made to more than 35,000 individual fundraising pages; importantly, our data also contain 3 information about the number of Facebook friends of the fundraisers. While the number of Facebook friends cannot be taken to be the universe of the fundraiser’s entire social group, we find evidence to support the idea that it is a meaningful proxy for social group size, and we show that the number of Facebook friends is positively correlated with the number of donations that the fundraiser elicits. Our key finding is that there is a negative correlation between the number of Facebook friends and the size of donations, across almost the entire distribution of friends. Our estimates also indicate that the magnitude is economically significant – moving from the 10th percentile of the distribution of Facebook friends (77 friends) to the 50th percentile (252 friends), predicted donation size declines by more than 10%; moving from the 10th percentile to the 90th percentile (654 friends), predicted donation size declines by nearly 20%. This result is robust to including controls for demographic characteristics of the fundraiser that might plausibly be correlated with social group size and to including both charity- and event fixed effects – i.e. we find that donations are smaller for fundraisers participating in the same fundraising event, raising money for the same charity, but who have larger social groups. We also show that the negative correlation between the number of Facebook friends and donation size holds when we run separate regressions by the order of the donation on the page (i.e. for the first, second, third donation on each page and so on) and we also show that it holds for the size of the maximum donation to a page. We interpret these findings as indicating that a smaller mean donation size cannot simply be attributable to a negative marginal effect from the additional donations that are made in larger groups, but that larger group size tends to be associated more generally with smaller donations by all group members. Our paper builds on the existing literature on the relationship between group size and private contributions to public goods, but adds to it along several important dimensions. Models of non-cooperative private provision of public goods that are based on collective consumption motives predict that individual donations are negatively related to the total number of contributors, i.e. there is free-riding, and that individual contributions approach zero as the number of contributors become very large.3 This result has been tested in a laboratory setting (Isaac and Walker, 1988; Isaac, Walker and Williams, 1994) and in a real world setting (Zhang and Zhu, 2011). In our case, however, the groups we look at are primarily social in nature; their primary purpose for interaction is not charitable activity. Second, as mentioned above, the members of the social group are only a subset of the total number of potential contributors to the public good, implying 3 This result is dampened if there are impurely altruistic motives for giving (Andreoni, 1990). 4 that any group size effect on public good provision will be a “local” one, specific to the amount of public good funded by a subset of contributors. Thus, the situation that we study is of how socially-determined divisions of the universe of private contributors to a public good, along lines that are not directly relevant to the nature of the public good, can determine donation outcomes. The model of relational altruism-based giving that we propose here is novel, and generates predictions that are supported by our evidence. There is little theoretical literature focusing on donations in these social group contexts. Exceptions are Benabou and Tirole (2006) who consider the case where people make contributions out of concerns for reputation or status and Scharf (2014) who focuses on the effect of social interactions on giving decisions in a setting where social links are only relevant for information transmission and social learning. However, numerous empirical studies support the presence of social effects on giving. Among other things, donations have been found to be sensitive to: whether or not giving is publicly observable (Soetevent, 2005); social information and norms (Frey and Meier, 2004; Shang and Croson, 2008); social pressure (DellaVigna, List and Malmendier, 2012); and peer effects in solicitations and donations (Meer, 2011; Smith et al 2015). Many of these social effects are likely to interact with social group size, yet, to date, the sensitivity of donations to the size of social groups has not been explored. This paper extends this literature by presenting new evidence on the relationship between social group size and donations. The plan of the rest of the paper is as follows. The next section discusses the individual fundraising context in more detail. Section 3 introduces the idea of relational altruism and discusses mechanisms through which social group size may affect donation size. Section 4 presents the main results on the relationships between social group size and donation outcomes, while Section 5 extends the analysis to explore alternative explanations for the observed relationships. Section 6 concludes with a discussion of the findings. 2 The individual fundraising context Alongside traditional fundraising activities, which involve a direct approach from a charity to potential donors, the past decade has witnessed a huge growth in individual-led online fundraising. Since 2001, more than 1.3 million individual fundraisers have raised in excess of £1.5 billion in the UK through online fundraising via the leading website. The way this type of fundraising activity works is as follows: individual fundraisers choose a charity for which to raise money and a fundraising event. This may be an activity they do together with other fundraisers, such as running a marathon, or something that they choose to do 5 independently, such as shaving their head. In practice, running events (particularly marathons) are the most common type of activity in our sample (39.5% pages), followed by walking (14.8%) and cycling (11.4%). Other specified sporting events include parachuting (2.3%), swimming (1.8%) and triathlon (1.6%). Non-sporting activities include memorials (3.9%) and appeals (0.5%). There are also a small number of anniversaries, including weddings and birthdays (0.3%) where individuals choose to let people make donations to a charity in lieu of receiving gifts. There is also a substantial category of “other” activities (24.0%). Having decided on an activity and a charity, the fundraiser then sets up a personalized fundraising page on an online fundraising platform that hosts multiple pages by many fundraisers. The platform allows donors to make donations online and passes the money directly to the charity. Donors arrive at the page and choose whether and how much to give and whether to reveal the amount and their identity. In practice, 7% of donations in our sample are made anonymously. Donations are made at different points in time – typically over several weeks – and all are listed on the page, with the most recent first. Donors can see who has previously given and the amounts and have been shown to respond to this information in how much they choose to give (Smith et al., 2015). Almost all donors belong to the fundraisers’ existing social groups of friends, family and work colleagues and come to the fundraising page following a solicitation from the fundraiser. In a survey of more than 19,000 sponsors,4 84% of those that were asked for a donation had been asked by a family member (of whom 87% said that they always gave when asked); 96% had been asked by a friend (67% always gave); 89% had been asked by a colleague (48% always gave) and 70% had been asked by a charity representative (only 9% always gave). This indicates that most people who donate to a page will do so because they are asked by someone they know personally. Fundraisers can contact members of their social networks directly by email and/or via Facebook. The fundraising platform offers an option for fundraisers to share the page directly to Facebook; fundraisers can also post the web address for their fundraising page into their Facebook status. Sharing the fundraising page directly to Facebook allows the fundraising platform to capture information on the number of Facebook friends that the fundraiser has at the time of sharing. However, it is important to note that this is the only piece of information that is available from Facebook – there is no information on whether donors respond to the link, on the identity of Facebook friends or on whether donations are made by people who are Facebook friends of the 4 See Payne, Scharf and Smith (2012). 6 fundraiser. However, even without this other information, the number of Facebook friends is of interest since it provides a measure of the size of the social group of the fundraiser and the potential number of contributors to the fundraising page. While there is considerable debate over the extent to which Facebook friends constitute a genuine social group, in this case, since fundraisers share their fundraising page with an explicit aim of soliciting donations, the number of Facebook friends is arguably a plausible measure of the number of potential contributors to the fundraising page. In practice, the number of Facebook friends – and hence the number of potential contributors to a page – varies widely across individual fundraisers. Our sample for analysis comprises 35,571 fundraisers.5 The mean number of Facebook friends in this sample is 332, but there is considerable variation as shown in Table 2. The 10th percentile is 77, while the 90th percentile is 654. Figure 1 compares the (mean) number of Facebook friends among fundraisers in our sample with the (mean) number of Facebook friends in the wider population. For the youngest age group (aged 18 to 34), the number of Facebook friends in our sample (not shown) is broadly representative of the population. This implies that these individuals do not only link their fundraising page to a Facebook page when they have an above-average number of friends. Older fundraisers exhibit a comparatively higher degree of selection with respect to the number of their Facebook friends – this may be selection into fundraising or into linking to Facebook. As a robustness check, we look at whether the results hold for different age sub-groups, and we show that they do. Table 2 provides summary information from the fundraising pages in our sample. We have all information that is publicly available on the page, including the name of the charity, the number of donations, the total amount raised, whether or not there is a fundraising target, and the value of the target. A typical page has nine donations and raises just over £130 in total. Most pages have a target, typically £300. We also have information on all the donations made online to the pages, including the date the donation was made, the amount given and the name of the donor where available (just over 7% of donations are made anonymously). 3 Relational altruism In this section, we discuss whether the predictions of the standard model of non-cooperative private giving – in which individuals make private contributions to a collective good in an 5 This is after some cleaning. We remove 3,817 pages where we cannot identify the charity registration number for England and Wales. We also drop 30 pages with zero friends and 364 with zero amounts donated. 7 uncoordinated fashion – can be used to interpret the observed relationship between group size and donation levels in our data; we then propose an alternative conceptual framework for thinking about donations to individuals’ fundraising pages. In the basic model of non-cooperative private contributions by privately motivated individuals (Warr, 1983; Bergstrom, Blume and Varian, 1984), each individual donation decision only accounts for the individual valuation of the contributor, neglecting the effect that donations have on other contributors; as a consequence, the level of donations is lower than the efficient level, and the more so the smaller is the individual valuation in relation to the total valuation of all contributors – i.e. the large is the number of contributors, (cid:1). Accordingly (as shown in Appendix A), the basic framework predicts that as (cid:1) becomes large, the size of individual contributions goes to zero; that is, “free-riding” increases with group size. Moreover, if individual contributions are sufficiently elastic with respect to (cid:1), total contributions will also be decreasing in the number of contributors. In our empirical analysis we find a negative correlation between the number of Facebook friends of fundraisers and the size of donations and little correlation between social group size and total contributions to the page. At first sight, it would seem that this finding fits the predictions of the basic model. Upon closer inspection, however, this conclusion is unwarranted. To see this, partition individuals in the economy so that (cid:1) = (cid:1)(cid:1) +(cid:1)(cid:2), where (cid:1)(cid:1) is the number of Facebook friends of a fundraiser, and (cid:1)(cid:2) is the number of other donors not belonging to the group of friends. In our data, we observe (cid:1)(cid:1) but not (cid:1)(cid:2) – and thus not (cid:1) – and there is no reason to think that the total number of contributors to the public good should be correlated to the number of people that are Facebook friends of a particular fundraiser. Indeed, in our empirical analysis we include charity fixed effects and so estimate a negative relationship between donation amounts and (cid:1)(cid:1), conditioning on (cid:1). In this case, we need something other than the basic model to explain the correlations that we see in our data. An alternative specification that could explain the negative correlation is one where fundraisers experience “warm glow” from the donations they raise, and where the members of the fundraisers’ social group are altruistic towards the fundraiser, which makes fundraising success a “local” public good to the social group – and thus a public good for which incentives will vary locally with (cid:1)(cid:1). Formally, if (cid:2)(cid:1) is average donation size within the social group and fundraising success within the social group is measured by (cid:1)(cid:1)(cid:2)(cid:1), we can model relational altruism in terms of an additional positive effect associated with (cid:1)(cid:1)(cid:2)(cid:1) (in addition to the standard collective 8 consumption effect) that is restricted to group members and that reflects the group members’ concern for the warm glow experienced by their friend from fundraising success. According to this specification, even though there is still no reason to assume that the total number of contributors to the given cause is related to the number of Facebook friends, the total number of contributors to (cid:1)(cid:1)(cid:2)(cid:1) will be related to the number of Facebook friends, and thus, through this channel, we can expect a negative correlation between (cid:2)(cid:1) and (cid:1)(cid:1), and possibly, a negative correlation between (cid:1)(cid:1)(cid:2)(cid:1) and (cid:1)(cid:1). The notion of relational altruism, as we have described above, is silent on an almost ubiquitous aspect of individual fundraising – the fundraiser engages in some kind of “arduous” activity; running a marathon, walking or swimming some distance. One way in which we can rationalize this feature being consistent with relational altruism is that it may be used by the fundraiser as a way of signalling their valuation when this valuation for the cause in question is private information. Engaging in a costly action may then serve, in a separating equilibrium, as an informative signal to their friends about the fundraisers’ valuation; and doing so by undertaking an arduous activity (e.g. running a marathon), rather than by making a larger visible donation to the cause, may be more informative (i.e., would make it easier to support a separating equilibrium) if the fundraiser’s friends cannot fully observe his/her fundraiser’s income level – which would affect the “effective” cost to the fundraiser of a given observable donation but may not affect the cost of arduous activity.6, 7 In sum, although we can rule out standard free-riding behaviour, relational altruism from the donor to the fundraiser (where the donor cares directly about the fundraiser and/or how much the fundraiser raises) provides one plausible channel for donations being lower in larger social groups. In the rest of the paper, we explore this relationship empirically. 6 In other words, the cost to the fundraiser of engaging in an arduous activity may be comparatively less affected by income than other visible costs may be. The comparative signaling strength of donations of money could be improved by making one’s income more observable, but then this could prompt unhelpful interpersonal comparisons that might interfere with relational altruism. 7 An alternative mechanism that could account for this is the fundraiser’s reliance on relational warm glow in the presence of donation indivisibilities and loss aversion. If the fundraiser is loss averse, then incurring a cost would bring the fundraiser into the loss domain, implying higher marginal relational altruistic effects. It can be shown, that if there is a minimum level of donation, by incurring a cost of a certain size, the fundraiser can trigger giving by friends that would not otherwise give, and in this way, obtain a higher net payoff than by not incurring the cost. 9