Are Goodwill Ambassadors Good for Business? The Impact of Celebrities on International Organization Fundraising

Abstract

Many international organizations (IOs) rely on voluntary contributions from member states and private actors to fund their operations. Donations from individuals are a significant and increasing income source for these IOs, who rely on marketing strategies such as celebrity endorsement, in the form of Goodwill Ambassadors, to help raise funds. Little is known, however, about the effectiveness of this strategy in the context of IOs although intuition from literatures in marketing and psychology suggests that celebrity endorsement should be effective. We conduct a survey experiment to investigate the effectiveness of Goodwill Ambassadors and, contrary to expectations, find no average effect of celebrity endorsement on donations to, and interest in, IOs and only limited effects among certain sub-groups. We speculate that the context of IOs makes it harder to generate the type of connection between celebrity and cause necessary to make endorsement effective and suggest that further investigation is needed.

Introduction

United Nations (UN) agencies rely almost exclusively on voluntary contributions to fund global development and humanitarian operations. For some, a significant part of these contributions comes from individuals, whose collective contributions can add up to hundreds of millions of dollars annually. ¹ One way in which many international organizations (IOs) raise awareness and funds is by recruiting celebrities – so-called Goodwill Ambassadors – who volunteer their time and effort to raise the agency’s profile. ² In many ways, Goodwill Ambassadors are comparable to brand or product endorsers who appear on behalf of the organization or talk about it on social media, using their name recognition to promote the organization to the general population.

Despite the popularity of this fundraising strategy among many IOs, we know little about its effectiveness. Although many factors identified in scholarship on celebrity endorsements might hold (cf. Bergkvist and Zhou, 2016; Erdogan, 1999; Halder, Pradhan and Roy Chaudhuri, 2021), there are contextual differences that make celebrity endorsement of IOs important to examine specifically. Unlike product endorsement, donors do not receive products in exchange for their “purchase,” but rather the knowledge of having contributed to a worthy cause. Additionally, IOs are further removed from donors’ daily lives than domestic charitable organizations, making donors unlikely to personally see the effectiveness of their donations.

We conduct what we believe is the first survey experiment to directly test the effect of celebrity endorsement on donations to, and interest in, IOs. We randomly assigned respondents to read a brief description of UNICEF’s work on one of two issues, highlighted by either a celebrity endorser or non-celebrity expert. Respondents then answered questions about their willingness to donate and interest in UNICEF. Counter to much of the extant literature and our pre-registered expectations, we find little support for the effectiveness of celebrity endorsement. We find no significant average treatment effect (ATE) (substantively or statistically) of Goodwill Ambassador endorsement; however, we do find that donors sharing an ethnicity with the endorser donate more in the treatment condition and that although men donate, on average, significantly less than women, treatment renders that difference insignificant. We speculate that, in the case of IOs, celebrities are less likely to have a personal connection to the causes they promote, weakening the tie between endorser and brand. We further conjecture that any “warm-glow” feeling that people get from donating to a charity (Andreoni, 1989) may not be enhanced by celebrity endorsement. These findings suggest that IOs should carefully evaluate the potential impact of celebrity endorsers and closely consider which populations they seek to influence.

Celebrity endorsement and IOs

Psychology and communications scholars have long studied sources of celebrity influence (Giles, 2002; Brown, 2018). This literature argues that people can form so-called parasocial relationships with celebrities, where individuals develop a pseudo-relationship with a celebrity through media consumption. These relationships can evolve into celebrity identification, where individuals start to see celebrities as role models for their own behavior (Brown, 2018), and it is generally through this identification mechanism that celebrities are expected to be influential. In fact, IOs recruit celebrities as Goodwill Ambassadors, in part, because of their perceived ability to influence their followers and act as role models, thus raising both awareness and funds for the organization. ³

Research in psychology and marketing suggests additional mechanisms of influence, such as the “match-up” hypothesis (Kamins, 1990), where an endorsement is only effective if a stimulus matches the endorser well (e.g., attractive celebrities endorsing beauty-enhancing products), and associative learning (Till, 1998), where repeated exposure to celebrity and product together creates an association between the two in the minds of individuals. Consequently, celebrity status can be a persuasive force on consumers (Kumru and Vesterlund, 2010).

Factors that make celebrity endorsement most effective can largely be split into two groups: features of the celebrities themselves (source effects) and features of the connection between celebrity and brand (fit effects) (Bergkvist and Zhou, 2016). Source effects include credibility, attractiveness, and likeability. Fit effects focus on the match between celebrity and brand, either as a main effect or a moderator of source effects (Bergkvist and Zhou, 2016) and can also refer to fit between celebrity and consumer, such as shared ethnicity (Deshpande and Stayman, 1994).

Being perceived as a credible sponsor of a brand (a good fit) is perhaps of paramount importance (cf. Halder, Pradhan and Roy Chaudhuri, 2021; Ohanian, 1991; Wheeler, 2009; Park, 2017; Ilicic and Webster, 2013). For example, Ilicic and Baxter (2014) note that the late actor Christopher Reeve’s collaboration with the American Paralysis Association following his horse-riding accident led to a doubling of the association’s revenue within a few years and Wheeler (2009) finds that Michael J. Fox, who suffers from Parkinson’s, outperforms Harrison Ford as an endorser of a (fictitious) Parkinson’s disease-related organization.

Celebrity endorsement is also commonly used by non-profit organizations to promote charitable giving (Wheeler, 2009), although celebrity effectiveness in charity advertisement has received less scholarly attention (Ilicic and Baxter, 2014). Important contextual differences exist between endorsement for charities and product branding (Wymer and Drollinger, 2015); e.g., people donate to charities for various reasons, such as awareness of need, altruism, values, a “warm-glow” effect, and simply being asked (c.f. Bekkers and Wiepking, 2011; Andreoni, 1989). A celebrity endorser of a charity must persuade potential donors that the charity is important and needs help and that their contribution can make a difference (Wymer and Drollinger, 2015). Therefore, a celebrity’s ability to credibly advertise the cause – e.g., through a personal connection or experience, or continued engagement with the charity – is likely to matter more. Thus, it is notable that much of the extant experimental literature uses either fictional celebrities (Park, 2017) or fictional charities (Wheeler, 2009; Wymer and Drollinger, 2015), making it harder to directly test fit perceptions. Exceptions are Karlan and List (2012) and Ilicic and Baxter (2014), but both use little-known nonprofits.

By contrast, we use a well-known IO, UNICEF, which is, in many ways, generally representative of UN agencies in terms of both its funding structure and its use of celebrity endorsements. Operational activities of UN development and humanitarian agencies are almost entirely funded through voluntary contributions (Goetz and Patz, 2017; Graham, 2015). These contributions largely come from member states, who can (and do) earmark their contributions for specific recipients or purposes, but significant sums also come from private donors.

Several organizations therefore have fundraising strategies targeted toward individuals. UNICEF appeals directly to individuals for donations, e.g., through their Inspired Gifts concept. ⁴ Other UN agencies have similar strategies. For example, certain other mandate-driven UN funds and programs, such as UNHCR and the World Food Programme (WFP), list exactly what a specified donation amount will purchase, thus making contributions more concrete to individual donors; the WFP is working to build its individual donor base after the success of its Share the Meal app, and UN Development Programme, in addition to soliciting general-purpose donations from individuals, also appeals for funding from individuals through a recently developed crowdfunding initiative. Thus, contributions from individuals constitute both an important and growing source of revenue for a large number of UN agencies.

Fourteen UN agencies make use of Goodwill Ambassadors to help increase awareness and increase revenue. ⁵ These are celebrities who generally work with the organization for multiple years, make appearances across the globe on its behalf, and highlight its causes on social media and other venues. The organizations often have designated staff to manage Goodwill Ambassadors and some have developed handbooks to regulate how the organization is promoted through this mechanism (cf. UNICEF, 2010; UNFPA, 2021).

To maximize the ecological validity of our experiment, we use a real celebrity endorser (the singer Shakira) alongside UNICEF as a representative organization. Shakira was selected due to her global recognition and popularity and her extended tenure as a UNICEF Goodwill Ambassador. She is thus likely to be as credible an endorser as possible for an IO. ⁶

Due to the dearth of research on Goodwill Ambassador effectiveness, the experiment was designed to maximize the potential for observing possible impacts of endorsement. In addition to the ecological validity reasons above, we also selected an organization that helps children and is widely known for its good work, i.e., one that is unlikely to be perceived as controversial by respondents. Furthermore, the outcome questions were about hypothetical donations or willingness to learn, which have low activation barriers. Thus, if celebrity endorsement is even somewhat effective, our design should capture it.

The paper tests two pre-registered hypotheses. The former tests whether respondents receiving the celebrity treatment are, on average, more willing to donate to the organization or the topic they saw discussed, whether they donate more money to the organization, or whether they are more willing to learn more about the organization. The latter hypothesis tests the same dependent variables, but additionally tests whether respondents who are co-ethnics of the celebrity endorser are more affected by the treatment than others.

Research design

To answer our research question, we conducted an online survey experiment through Amazon’s Mechanical Turk (MTurk) in January 2019, among 1200 US residents, of which 1121 completed the survey. ⁷ The sample size was determined based on a combination of resource constraints and power analyses that indicated 93% power to detect differences of $5 and 0.05 in the continuous and binary dependent variables, respectively, at the 5% significance level.

All US-based MTurk workers were eligible to participate, creating a diverse sample across various age, gender, education, and regional groups. It excludes those without internet access, who are also unlikely to be donors to IOs, and very wealthy individuals who, if involved in charitable giving, are perhaps also less likely to be influenced by celebrity endorsement. Thus, the target population of the “ordinary citizens” who might sometimes donate to charitable causes is reasonably represented in our sample. MTurk surveys have generally been found to replicate the sign of effects from nationally representative samples in the United States (e.g., Mullinix et al., 2015; Coppock, 2018) despite concerns that these samples are not always representative (Berinsky, Huber and Lenz, 2012; Huff and Tingley, 2015).

The treatment involved respondents reading a brief description of UNICEF and its work on one of two issues and then reading about either a celebrity Goodwill Ambassador (treatment) or a non-celebrity expert (control) highlighting the issue and its importance. To ensure respondents noticed the endorser, we included a small picture for both conditions. See the first part of the Appendix for vignette and outcome question details. We chose a text-based treatment for several reasons. First, to ensure comparability, the content in treatment and control conditions had to be as similar as possible, which was only possible by using text. Second, IOs such as UNICEF often do use text-based messaging from their Goodwill Ambassadors in efforts to increase awareness and donations so this mimics a strategy that is already in use by the actual organizations (e.g., UNFPA (2021)). Further, text-based prompts have been shown to have significant effects in a wide variety of situations (including on persuasion and attitudes), in many cases comparable to audio and visual cues (see, e.g., Chaiken and Eagly (1976), Corston and Colman (1997), Wittenberg et al. (2021) and Yadav et al. (2011)).

A secondary treatment dimension included respondents reading about either Girls’ Education or Displaced Children to provide more generalizability to the findings. Thus, each respondent was randomly assigned to one of four treatment conditions. Appendix Table A2 reports the number of respondents in each condition. Respondents answered demographic questions before the intervention, and outcome questions regarding interest in UNICEF and willingness to donate to the organization or the issue they had read about after.

We chose UNICEF for three reasons beyond its general representativeness. First, it is almost universally known, including among our respondents (1060/1121 already knew UNICEF). This ensures any treatment effect is not driven by the message increasing awareness of the IO itself. Second, UNICEF deals with children’s issues, a policy domain with fairly universal appeal. Thus, any lack of a treatment effect is unlikely due to respondents being strongly averse to the organization’s mandate. Third, UNICEF is widely known for its use of celebrity Goodwill Ambassadors. Consequently, a UNICEF celebrity endorsement is unlikely to be surprising to respondents.

Shakira was selected as our Goodwill Ambassador for several reasons. First, she is well-recognized across various age groups, genders, and ethnicities. Second, her Latin roots allow us to analyze potential heterogeneous effects of ethnic identity. Third, she has been a UNICEF Goodwill Ambassador since 2003, often endorsing the issues used in the experiment. ⁸

Including a UNICEF expert as the control gives us a human endorser in both conditions without which we cannot disentangle the effect of a celebrity Goodwill Ambassador from that of having a human at all. Second, this comparison is more meaningful given our interest in the effectiveness of a celebrity endorser versus someone in-house.

Table 1 summarizes the main variables. The main (continuous) dependent variable is Amount Donated (to UNICEF), where respondents were asked to (hypothetically) split $100 between themselves and UNICEF. A further robustness check to ensure that any treatment effect was not simply due to the hypothetical nature of this question was that half the respondents were randomly told that one person’s split would be randomly selected and implemented. Though respondents in this group donated less, on average, there was no difference in the treatment effect between both groups, as shown in Appendix Table A15. The remaining dependent variables are binary. Donate to cause and Donate to UNICEF indicate whether the respondent was likely to donate to the cause they had read about or to UNICEF in general, while Learn about UNICEF measures interest in learning more about UNICEF at the end of the survey. Another post-treatment measure asked if respondents had previously heard of UNICEF. Virtually all had and this variable is not used in the analyses. Given the research design, respondents should be well-balanced on demographic factors; Appendix Table A3 nonetheless presents confirmatory balance tests. Appendix Table A4 also presents more detailed summary statistics for all four dependent variables by treatment group.

Table 1. Descriptive statistics

Results

We use linear regressions with the coefficient on Goodwill Ambassador representing the ATE. Figure 1 summarizes our main results across all four dependent variables with two specifications for each, with and without covariates. The main feature across all eight models is the lack of a significant treatment effect, indicating that, counter to the expectations of our first hypothesis, an endorsement from Shakira does not significantly affect a respondent’s donation behavior toward, or interest in, UNICEF. (See Appendix Table A9 for similar results when separating the treatment into the two causes.) This lack of significance is robust to various control variables. ⁹

Figure 1. Celebrity endorsement and attitudes toward UNICEF.

Note: This plot summarizes the average treatment effect of a Goodwill Ambassador on four dependent variables, explained above. For each pair of regressions, the top line is the baseline result while the second adds covariates. The thicker bars around the point estimates indicate the 90% confidence intervals, while the thinner extensions denote the 95% intervals. See Table A5 in the Appendix for full results.

The lack of significance is unlikely explained by other factors. First, it is unlikely due to sample size as the ATE on the main dependent variable is substantively very small, at $1 out of a possible $100, indicating no meaningful difference in donation choices between the two groups. The same holds true for the other (binary) dependent variables where the coefficient is no larger than 0.02. ¹⁰

Second, to guard against lack of attention driving any null findings, we asked respondents, post-treatment (so the analysis can only be treated descriptively), if they had seen any celebrities in the study and, if they said yes, whom they had seen. Only 54 of 568 treated respondents did not recall a celebrity and only 12 of the remainder could not recall her name. Thus, only 10% of respondents could not accurately identify Shakira. As a robustness check, we re-ran our analyses restricting our sample to those who did pay attention. The results are virtually unchanged (see Appendix Table A10).

Third, we ensure that the lack of ATE is not due to Shakira being disliked and, therefore, ineffective. Post-treatment, respondents rated their like for several celebrities, including Shakira, on a 0–10 Likert scale, with 5 being neutral. Approximately 17% of the sample rated Shakira less than neutral, which is comparable to other celebrities on the list. Our results are robust to dropping these respondents from the analysis (see Appendix Table A11).

Lastly, it does not seem that UNICEF is already so well-known and well-liked that there is no room for celebrity endorsement to add value. As Table 1 shows, the average donation is under $38 (out of a potential $100) and almost 20% of our respondents chose to donate nothing. Thus, there is ample room for the Goodwill Ambassador to increase donations. Furthermore, given UNICEF’s extensive use of Goodwill Ambassadors to promote its causes, the organization itself likely believes that celebrities bring added value. Similarly, this finding is not driven by respondents who would never donate to the UNICEF or the given issue more generally (see Appendix Table A12).

Heterogeneous treatment effects

Next, we test the second hypothesis of whether shared ethnicity (Hispanic, in this case) with a celebrity Goodwill Ambassador can make respondents react differently. Results displayed in Figure 2 show limited evidence of this. The specifications and variables are the same as before, the only difference being an interaction between the treatment, Goodwill Ambassador, and dummy variable Hispanic. Although only significant at the 10% level, and only in one specification, we do find that Hispanics in the treatment group donate, on average, $10 more than others (left-most panel).

Figure 2. Celebrity endorsement, ethnicity, and attitudes toward UNICEF.

Note: This figure summarizes results for heterogeneous treatment effects of celebrity Goodwill Ambassadors on those who share an ethnicity with the celebrity. The four dependent variables and each specification are the same as presented in earlier results. The thicker bars around the point estimates indicate the 90% confidence intervals, while the thinner extensions denote the 95% intervals. See Table A6 in the Appendix for full results.

The base term ensures this effect is driven by Hispanic respondents in our sample being more responsive to the Goodwill Ambassador rather than them being more charitable than non-Hispanics. Table A7 (Appendix) re-weights our data to account for the lower proportion of Hispanic respondents in our sample than the national average, which strengthens the findings. Exploratory analysis finds no evidence of other minority groups being more responsive to the treatment, suggesting that our findings are not likely due to non-majority groups being more prone to celebrity influence. ^11,12

Finally, we consider possible heterogeneous treatment effects by gender based on the vast literature that finds gender differences in donation behavior (cf. Piper and Schnepf, 2008; Wheeler, 2009; Mesch et al., 2016; Dale et al., 2018) and finds that men and women respond to different types of cues (cf. Meyers-Levy and Maheswaran, 1991; Klaus and Bailey, 2008; Wymer and Drollinger, 2015). These results, summarized in Figure 3, were not pre-registered so they should be taken as exploratory. We include them nonetheless as the celebrity in our experiment is a woman and widely considered attractive, which may affect men and women differently.

Figure 3. Celebrity endorsement, gender, and attitudes toward UNICEF.

Note: This table summarizes results for heterogeneous treatment effects of celebrity Goodwill Ambassadors by gender. The four dependent variables and each specification are the same as presented in earlier results. The thicker bars around the point estimates indicate the 90% confidence intervals, while the thinner extensions denote the 95% intervals. See Table A8 in the Appendix for full results.

Two results stand out. First, in line with existing literature, men have less interest in UNICEF and are less likely to donate, indicated by negative coefficients across all specifications, most of which are statistically significant and substantively meaningful. In the first sub-figure, men donate, on average, approximately $8 less than women in the control condition. Second, for Amount donated, we find a positive ATE among men, indicated by the positive significant interaction coefficient. Men in the treatment condition donate, on average, almost $7 more than men in the control condition, making their average donation statistically indistinguishable from women’s. ¹³ In other words, the Goodwill Ambassador endorsement attenuates the difference between men and women. As before, however, these differences do not extend to the other dependent variables.

Overall, our findings suggest a limited effect of a celebrity Goodwill Ambassador, even in a “most likely” scenario where the celebrity is well-known and well-liked, the causes and organization are credible and non-controversial, the donation does not involve giving up actual money, and the intervention is very similar to how people are likely to encounter it in the real world.

Conclusion

IOs are predicting a massive shortfall in humanitarian funding (OCHA, 2023), making it important to analyze fundraising effectiveness. In this paper, we examine a common IO fundraising tool, namely celebrity Goodwill Ambassadors. Contrary to our expectations and much extant literature, we do not find clear effects of celebrity endorsement.

Why might this be? A potential explanation is that respondents in the treatment condition “overcorrect” and donate less when they see Shakira as the endorser, thinking that this must be a trick to get them to donate (Kang and Herr, 2006). Another possibility is that the perceived expertise of the UNICEF Director General, whose picture respondents in the control condition saw, is sufficient to make her as effective as Shakira (Wymer and Drollinger, 2015). If this is the case, it is unlikely that many organization officials would command the same perception, meaning that Goodwill Ambassadors may still be effective in general. Lastly, the plight of the children in the vignettes may seem farther removed from the everyday lives of potential donors in the United States than a domestic charity appeal would. The hundreds of millions of dollars donated each year by individuals through the US National Committee for UNICEF, however, suggest otherwise. Further, an observational study in the UK showed no material differences in the types of individuals who donate nationally vs internationally (Micklewright and Schnepf, 2009).

These findings are not without limitations. Different endorsers, organizations, or modes of appeal could have different effects, and our findings only speak to a certain type of organization. Specifically, the parameters of our experiment were chosen to maximize the likelihood of finding a treatment effect within an IO that is representative of a broad range of IOs that rely on voluntary contributions, fund-raise from individuals using text-based appeals, and focus primarily on humanitarian and development issues that many around the world care about. Despite this, we find limited effects, suggesting that further research is needed to determine the overall effectiveness of celebrity endorsement in the context of IOs.