Printer Friendly

Fundraising intermediaries inhibit quality-driven charitable donations.

I. INTRODUCTION

This article investigates how intermediary fundraisers--those that raise money that will then be sent to a charity--can limit donors' responsiveness to the quality of a charity. Charitable giving by individuals alone in the United States in 2014 comprised approximately 1.5% of the nominal gross domestic product (1) Frequently, this money is gifted through intermediary fundraisers. Rather than looking up a charity's address or website and sending them a donation directly, we often give to our school's campaign to raise money for United Way, buy cookies from a colleague's daughter, or donate in support of a friend walking 5 km for breast cancer research. In each case, the solicitation and gift is made by and to an intermediary rather than the charity itself. This investigation asks whether these intermediaries distract donors from the nuances of the charity to which they are potentially donating. Because of the added layer and complexity, does donation behavior no longer show sensitivity to the quality of the recipient charity?

Through a series of experiments, we show that intermediary fundraisers dramatically reduce differences in gifts across charities. When donations and elicitations are direct, charities in our sample receive substantially different financial support; high-overhead and unattractive charities receive fewer and smaller gifts. However, when donations are framed as going to a (meaningless) fundraising campaign (but ultimately to the charity), the charities receive indistinguishable donations; unattractive charities are supported as strongly as attractive charities.

There are many real-world examples where a donor may give more to a seemihgly lower quality charity with good reason. Perhaps she assumes the fundraiser vetted the charity or she simply wants to make the fundraiser happy. This article specifically rules out these hypotheses. We are not investigating whether donors glean information from or show altruism toward individual fundraisers. Our experiments strip away these features. The fundraising campaigns transparently provide no information about charity quality, and the fundraisers are anonymous strangers.

Rather, the investigation isolates a behavioral hypothesis: Simply adding a layer to the donation solicitation and processing inhibits the donor's ability to discriminate across charities; when money is raised through an intermediary fundraiser, the donors will show lower sensitivity to features of the charity, even features that would substantially change their donation were they donating directly to the charity. We will confirm this hypothesis, and show that it is the superfluity of information in an intermediary fundraiser context that clouds the judgment of the donor.

The majority of the literature on charitable giving concerns within-charity comparative statics--How do donations to charity A change when we go from policy X to policy Y? This article, on the other hand, illuminates across-charity giving. (2) This article does not investigate how an intermediary affects giving for a given charity, nor do we have any specific ex ante hypothesis of the effect intermediation may have within a firm. Rather, we focus on how an intermediary affects differences in donations across charities.

The results herein also provide evidence that a donation does not necessarily reveal the donor's preference. Many charities, rightfully or not, are criticized for large expenses. For example, in 2011, the American Breast Cancer Foundation (ABCF) spent $3.4 million on fundraising and $1.3 million on the programs and services it exists to deliver (ABCF 2012), and in 2010 the CEO of the American Cancer Society (ACS) was paid over $2.2 million in salary and benefits (Charity Watch 2012). While such numbers have raised critics' eyebrows, they have not necessarily deterred donors. In 2011, the ABCF tallied over $5 million in contributions while the ACS received $883 million. From these donations, one may infer the donors prefer the bundle of features that comprise these charities, including their overhead. This may not be the case. In our experiments, donors show a great sensitivity to charity characteristics, such as overhead, when the donation transaction is made simple and direct to the charity. When the transaction is made more complex by adding an intermediary fundraiser, even though they can still recall charity characteristics such as overhead with the same ease, donors show substantially less, if any, sensitivity to charity characteristics. In a world with intermediaries ubiquitous in fundraising, perhaps donor preferences are hardly ever truly revealed.

To test this hypothesis and potential mechanisms behind it, we employ multiple experimental treatments. In the baseline, we elicit charitable gifts for different charities in a very simple, direct setting--you donate money; charity X receives it--and we find that charities receive very differently sized gifts. In the intermediary treatment, we elicit gifts for the same charities but through an intermediary--you donate money to a fundraiser; the fundraiser donates to charity X--and we find the charities receive similarly sized gifts. A simple example comes from the second study that features two charities that differ only in the size of their overhead, 10% versus 40%. In the direct donation set-up, the 10% charity receives 76% higher donations, on average. However, when the donations are raised through an intermediary fundraiser, the two charities receive almost identical donations.

The convergence in gifts is true despite the fact that the fundraising intermediary is designed to be completely neutral: She has no special knowledge about the charity, makes no commission, never interacts with the donor, and does not know the donors' identities. In two follow-up studies, we find no evidence that the convergence in gifts is due to participants making their donation decisions based on features of the intermediary rather than the charity: Changing how hard the intermediary works as well as paying the intermediary a portion of the donation does not change the size of the donation. Intermediation is not shifting their focus onto the intermediary, away from the charity.

Across all these studies, the effects of charity characteristics on donation sizes are greatly dampened, or eliminated, in the indirect setting; the information is no longer affecting decisions. For this failure, we have two leading explanations. The indirect setting is affecting either the subjects' ability to acquire the information about the charities, or it is affecting their ability to process the information.

In our experimental contexts, intermediation does not lead to failures in information acquisition: Subjects show the same ability to recall features of the charity under intermediation, and subjects continue to report significantly different beliefs of the charities' quality ratings under intermediation. Including an intermediary does not hinder subject ability to differentiate between charities. Then why does this acquired information not map into donation behavior?

In the final study, we show the intermediation context inhibits donors' ability to process information. A necessary feature of intermediation, and fundraising, is added information, or complexity. Describing the fundraiser, its procedures, its goals, et cetera adds length to the narrative given to the potential donor. In this study, we lengthen the narrative given to the donors in the direct-donation setting by providing superfluous details about how the donations will be collected and sent in. In this setting, though they are framed as going directly to the charity, donations are indistinguishable across charities. Simply put, intermediaries in fundraising do not preclude acquiring information about charities, but the complexity provided by the nature of the transaction all but precludes using it.

Though there is no direct evidence on intermediation and charitable giving, evidence on how intermediation substantially affects decision-making is growing. Recent experiments in Psychology and Economics have highlighted the importance of indirectness in moral judgments. Fershtman and Gneezy (2001), Paharia et al. (2009), Hamman, Loewenstein and Weber (2010), Bartling and Fischbacher (2012), Coffman (2011), Drugov, Hamman, and Serra (2014), and Grossman and Oexel (2013) all provide evidence that perceived responsibility is attenuated for bad outcomes that occur as an indirect result of an action as opposed to bad outcomes that occur directly. Punishment, reciprocity, and guilt are all lower for misbehavior that hurts someone indirectly, through an agent or intermediary. Moreover, Coffman (2011) and Eisenkopf and Fischbacher (2011) show this is true for positive outcomes as well. In the study of Coffman (2011), subjects who donated mosquito nets to pregnant women in Kenya were rewarded dramatically less when the salience of the charity in the transaction was increased. When the rewarders were primed to think about the transaction as donor-charity-recipient rather than donor-recipient, rewards decreased greatly.

A. Related Literature, on Within-Firm Predictions

Though this article makes no specific hypotheses about how intermediation affects giving to a given charity, there has been work done that may speak to what may be found. Many studies have shown why having an intermediary fundraiser might work to the advantage of a specific charity. In separate door-to-door field campaigns, DellaVigna, Malmendier, and List (2012) find evidence that a face-to-face interaction with a fundraiser might induce giving through social pressure, and Landry et al. (2006) find personality attributes of solicitors (self-efficacy, self-confidence, and performance motivation) and physical attractiveness of female solicitors predict donations. (3) Also, generally, the people doing the fundraising have been selected into fundraising for the charity. This likely means they have done some vetting of this charity, and they are publicly supporting it. This has been shown to increase donations in two ways. First, Potters, Sefton, and Vesterlund (2005) (testing the theory of Vesterlund 2003) find larger donations to a public good after others have revealed the quality of the charity. Second, Kessler (2011) finds more and larger donations from people who have been exposed to someone who signals support for the charity (in particular by wearing a pin). Kessler interprets this increase as a result of expectations about the signaler's donation, separate from general notion of charity quality.

There have also been studies on the effects of having potential donors think about individuals when making their donation decision. Real and hypothetical donations have been shown to be higher when a picture or description of an individual is shown versus a picture or description of a group, a pair of people, or a description of an individual along with statistics (Kogut and Ritov 2005; Small, Loewenstein, and Slovic 2007; and Vastfjall, Peters, and Slovic 2010). Though this article makes no predictions about how a fundraiser might affect giving within a charity, these studies provide evidence that if donors are thinking about an individual (a fundraiser) rather than an organization (a charity), this may increase overall gifts. However, it should be noted that intermediation could also decrease donations. For example, if donors thought there was loss--either from leakage or increased overhead--they might very reasonably decrease their donations.

II. STUDY 1: THREE CHARITIES--INTERMEDIATION AND DIRECT DONATIONS

A. Design

We elicited real charitable donations in an online environment. Students from the Ohio State University subject pool were sent an email message inviting them to participate in a survey (using ORSEE recruitment software [Greiner 2004]). (4) The invitation only said the survey would take less than 5 minutes, and if they completed the entire survey, their name would be entered into a drawing for a "large cash prize."

If a student clicked on the link, they were taken to a five-screen survey. (5-6) The first screen was a consent form, which contained no information as to what the experiment was about. On the second screen, subjects were informed that for every 75 students who completed the survey, there would be a drawing for $80 to take place on the following Wednesday. They were also told they would have the chance to donate some of this money to a charity if they won the drawing. The subject was assigned to one of three charities, and one of two donation procedures--a three by two design. Subjects were not made aware of the other charities or donation procedure. The experiment took place over 3 weeks during July 2011.

Three Charities. Students were randomly assigned to one of three charities. The charities were chosen with the intention that there would be differences in baseline donations. The ACS was chosen with the expectation that it would receive more donations than the other two; it is well-known and has a sympathetic cause. The Archaeological Conservancy (AC) was chosen since it was a charity the students would not know well and whose cause might not induce as much sympathy in the average student. (7,8) The third charity was the Community Foundation of Southeast Michigan (CFSM). Not only are most of the subjects not aware of the CFSM and not from Michigan, the students school's arch-rival, the University of Michigan, is in Southeast Michigan. The latter two charities were chosen with the expectation they would get significantly less donations than the ACS from our subject population.

Direct Donation Treatment. In the baseline treatment, subjects were asked how much of the $80, if they won the drawing, would they like to donate to charity (they were shown one of the three charities above): "How much of the $80 would you like to donate to the charity?" They were told the donation would be done immediately after the drawing, and it would be made anonymously.

Intermediary Donation Treatment. The Intermediary treatment was designed to be as similar as possible to the baseline with one change--money was being collected by a student to be donated to one of the three charities. The inspiration for this donation procedure is a workplace fundraising campaign: Money goes to an individual or office in your workplace, who then donates to the charity. In our study, the subjects were told a student from the same subject pool would be coming to the lab next week to go through the data, do the drawing to determine the winner, sum up the donations from the winners, and donate the money to the charity anonymously. They were told the student was "randomly chosen," and, "was not chosen based on knowledge of charities, nor has he or she been given the name of the charity or told what task they will be doing. Also, this person is not making money based on your donations, nor will we reveal your decision to him or her. Your email address will be clipped from the data set before we pass it along." Their decision to donate was the same: If they won the $80, how much would they like to donate? In this treatment, the donation was framed as, "How much would you like to donate to the student's fundraising cause?"

On the final, and fifth, screen, subjects were asked two unincentivized questions. First, "What was the name of the charity?" Second, "Did you look up the charity?" Subjects were not able to go back to a previous screen at any point during the survey. (9)

B. Results

We analyzed the data from the 315 students who completed the entire survey and only started the survey exactly once. (10) In total, 93% of the students who agreed to the consent form finished the short survey. (11)

Result 1--Convergence of Average Gift Size: The charities receive significantly and substantially different donations when solicited directly, but the difference significantly decreases when solicited indirectly.

As illustrated in the top row of Figure 1, as expected, the ACS received higher donations than both of the other charities in the direct elicitation (Wilcoxon rank-sum, two-tailed test comparing ACS to AC and CFSM pooled. p < .01). These differences were reasonably substantial as well. The ACS received an average donation decision of $26.1 versus $14.2 and $15.5 for the AC and CFSM. We want to avoid context-specific level effects in our analysis though, so we will rather focus on across-treatment differences. In the Intermediary treatment, the across-charity donation differences found in the Direct treatment cannot be replicated. As the bottom row of Figure 1 shows, the gift sizes and distributions to all three charities are now very similar. Wilcoxon rank-sum two-tailed tests cannot reject equality (P = -4).

Recall that ex ante, however, the hypothesis was not that the difference in donation sizes would disappear, but rather, that the difference would decrease. To compare the differences in donations across the treatments, we employ difference-in-differences (DD) analyses. In the analysis for every treatment, we will obtain DD p values in two ways. First, we use a nonparametric approach. We bootstrap the data to test whether (Difference of donation means in direct elicitation)--(Difference of donation means in indirect elicitation) is nonzero. 12 Second, we estimate the DD using a standard regression DD approach. We derive DD estimates of both the effect on donation size as well as probability of donation.

The first column of Table 1 shows the bootstrapped DD p values for this study. The difference in donation sizes between the preferred-charity and the less-preferred charities is significantly different in the indirect gift case. The DD regression estimates, the second coefficient reported in each column of Tables 2 and 3, give a sense of the direction and magnitude of the change. In every column, the more preferred charity (ACS, in this treatment) in the direct elicitation is the constant; the other coefficients indicate differences from this cell. A negative value for the first coefficient indicates the less-preferred charity (the non-ACS charities for this treatment) received smaller donations in the direct elicitation. A positive coefficient on the second term (the interaction term) indicates that the difference in donations across charities found in the direct donation treatment decreases in the fundraiser treatment. The interaction term is marginally significantly positive for Model I of Table 2, indicating support for our hypothesis that donation differences across charities decrease when solicited indirectly. It is worth noting the size of this effect as well: In the direct donation setting, the less-preferred charities received more than $ 11 less, on average, but in the indirect donation setting, this difference was reduced by over $9 on average.

Result 2--Convergence of Probability of Donating: The charities have significantly and substantially different donation likelihoods when solicited directly, but this difference significantly decreases when solicited indirectly.

Perhaps the most striking difference between the bottom and top row in Figure 1 is that the large spikes in the $0-$4.99 bin for the AC and CFSM in the top row disappear in the bottom row. The statistics support the visual; in the direct treatment, the ACS received gifts from 82% of subjects while the AC and the CFSM only received gifts from 66% and 69% of subjects (two-tailed test of proportions comparing ACS to AC and CFSM pooled, p = .04). However, in the fundraiser treatment, the differences disappear; ACS received gifts from 85%, AC from 88%, and CFSM from 87% (two-tailed test of proportions comparing ACS to AC and CFSM, p = .98).

We see similar results for the binary outcome DD analyses. (13) The bootstrapped DD p value (Table 1, first column, bottom row) indicates a significant change in donation likelihood in the intermediary condition. The regressions show why: The ACS receives significantly more nonzero donations than the less-preferred charities (first term in Table 3, Model VI), but this difference decreases significantly in the intermediary fundraiser setting (second term. Model VI). Again, the point estimate on the interaction term is roughly equal and opposite in sign to the point estimates for the less-preferred charity dummy.

Result 3--Subjects are equally likely to recall the name of the charities across the Direct and Intermediary treatments; results 1 and 2 are not driven by a very basic confusion of who is receiving the donation.

To begin to understand why subjects make similar donations to very different charities in the Intermediary treatment, we consider if this could be driven by subjects being less likely to acquire information in the Intermediary treatment, resulting in not knowing to whom they are donating. This is not the case. Overall, the percentage of subjects who correctly recalled the name of the charity is roughly equal across treatments, decreasing from 59% to 54% from the Direct treatment to the Intermediary treatment (p = .4, two-sided test of proportions). (14) The results are not different for any of the charities across treatment either: 69% and 61% for ACS, 54% and 54% for the AC, and 52% and 47% for the CFSM in the Direct treatment and the Intermediary treatment, respectively. Although all three are directionally negative, none are significant; p = .35, .99, and .63 for ACS, AC, and CFSM with two-sided tests of proportions. Not only are they insignificant, the sizes may indicate their lack of consequence. Compared to a 17% increase in nonzero gifts to the AC and CFSM in the Intermediary treatment, a less than 5% decrease in recall is relatively unremarkable.

III. STUDY 2: TWO CHARITIES--EXOGENOUSLY VARYING ATTRACTIVENESS FOR A SIMPLE CHARITY

A. Design

This study is a second proof of the concept, addressing some potential alternative hypotheses from the Three-Charities study. The purpose of this study is threefold.

First, we want to make the attractiveness of the charities a very easy thing to evaluate. We do this in two ways. One, we make the charity a very simple, transparent transaction--giving money directly to a poor African household. Two, we make salient one very important attribute of this charity--the overhead (10% or 40%). It might have been true in the Three-Charities study that the participants were updating their beliefs about the quality of the charity when it had an intermediary. Although this would be unfounded in the Three-Charities study, this is a reasonable heuristic in general. The ease of evaluability of the charity here should, at the very least, attenuate any updating that may occur; the subject should be able to easily discern the attractiveness of the charity herself and be more confident in her beliefs. Also, with evaluability of the charities being quite easy in these treatments, we assume subjects will be likely to be able to recall features of the charity ex post; hence, this is a good test if whether (1) intermediation is distracting donors so they never acquire information about the features of charities in the first place or (2) if it is simply that these features, although known by the donors, are not factoring into their donation decisions.

Second, we want to make the attractiveness of the charities objective and exogenously varied. We relied on our intuitions about subject preferences in the first study. There might have been large heterogeneity, and there might have been a lot of subject uncertainty about their own preferences toward these charities. In this study, increasing the overhead is equivalent to a transfer from a poor African family to a charity and its employees. We reasonably assume subjects would not prefer this transfer.

Third, we want to replicate the original results with a new subject pool and different charities.

This set of treatments was very similar in structure to Study 1. We elicit real donation decisions from an online study pool. This time the subject pool was the Yale SOM eLab. Very much like Amazon Mechanical Turk, this is a standing subject pool who periodically log in to the website to see in what studies they can participate. All studies in the eLab offer an Amazon gift card of known size as payment to a winner from a drawing of known probability. This study was advertised as a "5 Minute Study" with a $65 Amazon gift card going to 1/50 participants. (15) As before, we asked the participants, if they won the drawing, would they like to donate some of their winnings to a cause. If they clicked on the survey, they agreed to the consent form on the eLab website and had to correctly respond to a captcha before being redirected to our survey.

Subjects in any treatment of this study would be donating to "a very poor household in Kenya," one that has "been verified as poor by an international nonprofit who has certified their house is made of mud, wood, and/or grass, a reliable indicator of extreme poverty." (16) The money will be sent directly to the family: "They will receive it by text message (they have been given a sim card; receiving money and paying for goods via text is relatively common in Kenya)."

Varying Charity Overhead. There were two treatments varying charity attractiveness. We use overhead in the transaction as a proxy for attractiveness. Subjects are randomized into one of two treatments, either 10% overhead or 40% overhead. They are told, "For every $1 you send, {10, 40j cents will go towards various transaction and overhead costs (e.g., employee wages at the nonprofit, data collection by the nonprofit, PayPal fees, text messaging money fees, etc.)."

Direct Versus Intermediary. As before, subjects are randomized into either a "direct" treatment or an "intermediary" treatment. In the direct treatment, subjects are randomized into one of the charity treatments and asked, if they win the drawing, how much they would like to donate. Subjects randomized into the "intermediary" treatment are provided a narrative akin to that in the Three-Charities study: A student from the subject pool was recruited to come in at some point next week. They will press a button to run the drawing, they will sum up the donations from the winners, and they will donate them anonymously. As before, the intermediary is designed to be completely neutral with respect to altruism, "The student we recruited was randomly chosen from the email list. The student does not know what they will be doing when they come in next week. They will not make money based on your donation, nor will they have the opportunity to keep any of the money you send them. They are being paid a flat fee for their help. They will not be shown your email address; emails will be replaced by random numbers for identifiers." After participants make the donation decision, they answer a few more questions. We ask them to guess how much they think the average donation is. If they are within $5 of the empirical average, and they win the drawing, they will earn an extra $5. We then ask them to guess what everyone guessed was the average, also for $5 for accuracy. Without incentives they were asked, why did you donate what you did, who is the final recipient of your money, had you donated $10 how much would be received, and several demographic questions. For the final wave of subjects (N = 59), we also asked them to guess how this charity was rated on a scale from "1 meaning 'Not Good: Should not donate'" to "9 meaning 'Very Good: Worthy of donation'" by Yale students who looked through "all the information we could find for this charity." They were paid $5 if they were within 0.5 of the actual average rating if they won the drawing.

B. Results

For the Two-Charities study, we analyzed the data from the participants who took the survey only once and completed it and there were 251 subjects in total. Subject characteristics for all studies conducted with the eLab subject pool can be found in the Appendix.

Result 4--With a simple charity and a salient attribute on which to evaluate, there remains a convergence in gift size and likelihood between attractive and unattractive charities, going from the Direct treatment to the Intermediary treatment.

Facilitating evaluation of the charities did not eliminate the convergence observed in the Three-Charities study. Figure 2 presents histograms of donations. As in the Three-Charities study, the distributions of donations are significantly different in the Direct condition (p < .01, two-sided rank sum), but they are indistinguishable in the Intermediary condition (p = .25, two-sided rank sum). Again, the likelihood of donating is different in the Direct condition (p < .01, two-sided test of proportions) but statistically indistinguishable in the Intermediary condition (p = .37, two-sided test of proportions). Even with a very simple, transparent charity, and a very salient attribute on which to evaluate, intermediation is enough to eliminate significant differences in donations.

Result 4 is corroborated with a DD approach. As shown in Table 1, the bootstrap DD p values are significant for both donation size (.04) and likelihood (<.01). The regression DD estimates agree. The interaction term in Model II of Table 2, the donation size, as well as Model VII of Table 3, the donation likelihood, both show a significant reduction in the difference in gifts across attractive and unattractive charities in the Intermediary treatment. Once again, simply adding an intermediary to the charitable transaction significantly decreases the amount to which donors discriminate between charities.

Result 5--In the Two-Charities study, the Intermediary treatment had no effect on subject ability to recall charity overhead.

Intermediation had no significant effect on the likelihood of a subject recalling how much of her donation would go toward charity overhead. In this study, following their donation decision, subjects were asked, "If you had chosen to donate $10, how much money would have been received?" Overall, 51% correctly stated that $9 or $6 would be received in their respective treatment in the Direct condition, and 48% answered correctly in the Intermediary treatment (p = .63, two-sided rank sum). The numbers were also pretty similar within charity: 51% and 44% recalled the cost was 10% in the Direct and Intermediary treatments respectively (p = .45, two-sided rank sum), and 52% and 53% recalled the cost was 40% across treatments (p = .90, two-sided rank sum). In this set of experiments, the attractive and unattractive charities are differentiated only by their overhead: hence, a subject's ability to recall overhead is a sufficient measure of acquiring the information necessary to distinguish the charities. Intermediation does not measurably affect information acquisition.

Result 6--Intermediation does not eliminate subjects' belief that the low cost charity is rated more highly than the high cost charity.

Though we only have beliefs of charity ratings for the last 59 subjects, some interesting results emerge. The average belief of average charity rating goes up from the Direct treatment to the Intermediary treatment for both charities, significantly so for the low cost charity, from 6.0 to 7.3 (p = .03, two-sided rank sum), and directionally so for the high cost charity 4.6-5.7 (p = .29, two-sided rank sum). Beliefs of the quality of the low-cost charity are also higher than those for the high cost charity, 6.0 versus 4.6 in the Direct treatment (p= .13, two-sided rank sum) and 7.3 versus 5.7 in the Intermediary treatment (p = .02, two-sided rank sum).

That the ratings go down as we increase overhead is a nice robustness check of our instrument for beliefs of charity quality. That the ratings go down for the high-cost charity in the Intermediary treatment is quite provocative. Not only does intermediation not change the ability to recall the charity (Results 3 and 5), it does not change the ability to discriminate between charities. The subjects show they have all the information they need to discriminate between charities: Charity B is high cost and is expected to be rated as lower quality by informed judges. There is a large disconnect, however, between these data and their donation behavior. The information they have measurably acquired about the charities, that they substantially rely on to make direct donation decisions, does not seem to be in their mind when they are making their indirect donation decisions.

IV. STUDY 3: MATH-FOR-CHARITY

A. Design

In the previous two studies, the intermediary (1) saw how much was donated by many, albeit anonymous, individuals and (2) donated the money, albeit anonymously, to the charities. Either one of these features might encourage more giving and conceivably drive some convergence under intermediation. Dana, Cain, and Dawes (2006) show some altruism can be driven by the desire to meet the expectations of an anonymous other. Further, the participants can expect the intermediary knows the group from whom these donations came; thus, even if there is not an individual reputation to support, there might be a group reputation the subjects care about (e.g., Bernhard, Fehr, and Fischbacher 2006). Finally, it is conceivable that the participants expect that the intermediary will receive some utility for donating money, so through an altruistic channel, the participant will make some nonzero donation so that the intermediary may do the same.

To control all these concerns, Math-for-Charity is designed not to let the intermediary see any of the donations or make any donations. The inspiration for the narrative in this study comes from "walk for a cause"-type events. In these events, typically a group of people walk a set distance, maybe 5 km, and raise donations beforehand in support of their effort. The effort is not intrinsically beneficial to the charity; the walk does not directly produce anything, other than the donations it raises. It is a peculiar but seemingly increasingly popular mode of charity fundraising in the United States. Though this study is inspired by such events, there are many features intentionally not included; we are not directly testing this institution.

The set up for Math-for-Charity is the same as the baseline Two-Charities study: Participants in the Yale SOM eLab click on a link to participate in a 5-minute study with a 1/50 chance of a $65 Amazon gift card. They are given the opportunity to donate to a poor household in Kenya, and they do so with either 10% or 40% overhead. This time, however, they make the donations "in support of [a] student's effort." As in the baseline Two-Charities study, a student from the subject pool has agreed to come in to the lab the following week. This student does not know why they are coming in. The student will be adding up five 2-digit numbers, arranged horizontally, with scrap paper but no calculator. They will solve a fixed number of problems, be paid a fixed fee (the size of which is not reported), and will be sent home. The participant doing the online study is told they can make a donation to the poor Kenyan household "in support of the student's effort."

To test the hypothesis that the donors are more responsive to features of the intermediary, we vary the number of math problems the intermediary solves. Each online study participant is randomly assigned to an intermediary. One intermediary will be doing 150 math problems, and the other will be doing 15 math problems. They are told that "In previous studies, students, on average, solved about 10 problems every 5 minutes, so our best guess is that they will be solving math problems for {between 5 and 10 minutes, a little over an hour}." (17) The online study participants do not know about the other intermediary.

B. Results

Recall Math-for-Charity was designed to eliminate two features of intermediation in both previous studies: (1) the intermediaries saw how much was donated, even though they were anonymous and (2) the intermediaries donated the money, albeit anonymously, to the charities. Eliminating these features, and the potential confounds outlined in Section II.B, does not eliminate the convergence in donations under intermediation. A total of 126 subjects participated in Math-for-Charity once and completed the entire survey. We find no difference in donations for how many math problems the intermediary solves (see the Appendix), so we present the pooled data in Figure 3. Figure 3 shows the histograms for donations for both the low cost and the high cost charity. The distributions are indistinguishable (p = .53, two-sided rank sum).

Using the Direct treatment data from the Two-Charities study as the comparison, we can use a DD to test for convergence in gifts. As Tables 1-3 show, both the size and likelihood of a gift across the attractive and unattractive charity significantly decreases when an intermediary is present, and again, the point estimates suggest the differences are almost wholly eliminated. Even when the intermediary does not see the donations and does not make the donations herself--the donation is only made "in support of [her] effort"--the intermediary is sufficient to make donors no longer distinguish between the two very different charities we have in our set up. Changing the overhead by a factor of 4 no longer has any bearing on the size or probability of donation. Not only is this interesting insofar as it controls for potential counter-hypotheses, but it also speaks directly to charitable transactions that occur regularly, specifically charity walks. There are many reasons why you may donate when your friend walks 5 km for a charity, and many of those may shift how you discriminate (or do not discriminate) across charities. These results suggest your donation may not respond significantly to the identity of the charity simply because your donation is not going directly to the charity but rather is framed as "support of your friend's effort."

V. STUDY 4: INTERMEDIARY-OVERHEAD

A. Design

In the Intermediary-Overhead study, we test whether donors ignore overhead costs in the previous two studies because they are not direct and immediate costs. In those studies, the costs are applied to the second transaction--intermediary to charity. Hence, in this study, we place some costs on the first transaction between donor and intermediary. As the actual charity's overhead is 10%, we keep that constant. We simply add overhead by taking out a percentage of the donation to pay the intermediary's wage, either 5% or 30%. Otherwise, this study is identical to the previous two Two-Charities treatments. The exact wording is, "For every $ 1 you send to the student, 10 cents will go towards various transaction and overhead costs (e.g., employee wages at the nonprofit, data collection by the nonprofit, PayPal fees, text messaging money fees, etc.), and {5, 30} cents will go to pay the student's wage." Hence, the total overhead is either 15% or 40%. Other treatments varied overhead between 10% and 40%. We could not maintain that exact difference here without having 0% intermediary overhead in one of the cells. Out of worry that there may be a discontinuity at 0%, we slightly change the overall overhead to 15% to give the intermediary 5% overhead.

B. Results

Recall Intermediary-Overhead was designed to make the overhead costs more direct and proximate to the donor. Instead of framing the costs as a loss in the intermediary-to-charity transaction, in this study, some of the costs were framed as wages for the intermediary; hence the costs were shifted toward the donor-to-intermediary transaction. This study is designed to test whether the donors are ignoring the charity overhead in previous studies because the costs are not associated with a transaction the donors are directly involved in. This does not seem to be the case.

A total of 126 subjects took the survey once and completed it. Figure 4 shows the distribution of donations between the low-cost and high-cost charities are not substantially different, and this is validated through the DD analyses. Using the Direct treatment from the Two-Charities study, the DD analyses confirm the convergences. Tables 1-3 show the large differences in donation size and likelihood dramatically and significantly decrease when the intermediary is introduced, even if she comes with large overhead herself.

VI. STUDY 5: INFORMATION OVERLOAD

A. Design

Adding an intermediary to a transaction also mechanically adds more information for the potential donor to take in. Not only is there a description of the final recipient of the money, there is a description of the intermediary fundraiser and how it works. Information Overload investigates whether the additional information and complexity provided by the "fundraising campaigns" drive the convergence in donations.

To do so, this study varies whether or not there is an intermediary fundraiser but holds (more) constant how much information the subject has to read across the two conditions. This study uses the same set-up as the previous three Two-Charities treatments. The full texts used in this study can be found in the Appendix. On the donation screen, the direct treatment has 368 words, the intermediary treatment has 365, and 354 of these words are preserved, in order.

B. Results

The Information Overload study matches the amount of information given to the potential donors in the Direct and Intermediary treatments. In other studies, naturally describing the process and nature of the intermediary fundraiser required more words. This treatment eliminates this organic difference.

Result 7--Information overload induces convergence in gifts across charities.

Tables 1-3 show that the convergence found in previous studies is completely eliminated. This is because there is never any divergence. There are no longer any differences in size or likelihood of gifts in the Direct treatment. The average gift size in the Direct treatment is $17.3 to the low-cost charity and $17.1 to the high-cost charity (p = .6, two-tailed rank-sum). Further, in the Direct treatment, 70% make a nonzero gift to the low-cost charity, and 64% to the high-cost charity (p = .5, two-tailed test of proportions). Simply adding information, however superfluous, to the Direct treatment was enough to make a charity with four times the overhead receive the same gift size and likelihood as its low-cost counterpart. Hence, these results suggest, when an intermediary fundraiser adds a great deal to the information that is provided to the potential donor, this is sufficient to hamper decision-making.

VII. DISCUSSION

In this article, we show that intermediaries can make very different charities receive surprisingly similar donations. This holds even though the intermediaries in our experiments should not reasonably change a donor's decision of how much to donate. The intermediaries across our contexts are randomly chosen, do not have any special knowledge or support of our charities, do not make money based on the donations, do not see donor identifiers, make the donations anonymously, and in one treatment do not see any donation decisions or make the donation to the charities. Despite this, intermediation greatly reduces the extent to which subjects discriminate between charities in their donations. We are able to show that intermediation does not affect subject ability to recall the name of the charity, the overhead of the charity, or the ability to evaluate the overall quality of the charity; however, intermediation precludes the use of these charity features when determining how much to give. If we consider the quality of the charity to be its price, as discussed in Section III, what this means is that donors are almost completely ignoring the price of a good for which they are paying. We hypothesized that the elasticity of charity attractiveness would go down under intermediation, that (echarity I direct treatment) > (echarity I fundraiser treatment). This is supported in our studies. In the final study, we provide a possible explanation for the above results: Settings that have intermediary fundraisers provide a lot more information to the potential donor. As stated before, this superfluity of information does not inhibit donors' acquisition of information about the charities; however, it precludes the donors' use of any information about the charities.

Intermediaries are seemingly involved in almost every charitable transaction to some degree. The results herein suggest the pervasiveness of intermediaries might dramatically change the bundle of charities that are financially supported, and this distortion might result in a sub-optimal bundle of charities according to donor preferences. If we take behavior in the Direct treatments to be revealing donor preferences, 60% do not prefer to support the high overhead charity in the Two-Charities study. Many of these donors point out the charity is "too costly" in an unincentivized post-donation question. However, in the Intermediary treatment, the rate of size zero gifts drops to 39%. That is, 61% no longer behave as if this charity is too costly to support even though the cost remains the same. If the Intermediary> treatment is indicative of what might happen in a world full of fundraisers, many unattractive charities might be surprisingly well supported.

The results herein show that fundraisers can greatly reduce the differences in donations received by very attractive and very unattractive charities; however, we find this is driven by the added burden of information that the fundraisers in our experiments provided. This leads to the natural question, do fundraisers always, or even usually, cause information overload such as that found in this experiment? Intermediaries, and fundraisers, by their nature, add layers and information, complicating the donation process. Whether this is always, or usually, sufficient to muddy donor decision-making to the extent seen in this paper is an open question.

APPENDIX

YALE ELAB SUBJECT CHARACTERISTICS

TABLE A1

Descriptive Statistics for the Pooled Data for All the Studies
Involving the Yale eLab

                         Average

Female                     65%
Age                      38 years
Has high school degree    99.9%
Has college degree         79%
Has advanced degree        19%
USA                        91%


MATH-FOR-CHARITY--DOES THE NUMBER OF PROBLEMS MATTER?

In addition to the analysis done in Section IV.B. the data from Math-for-Charity can investigate an additional hypothesis: Do the donors respond to features of the intermediary? In Math-for-Charity, we varied the number of math problems the intermediary would solve, either 15 or 150. The potential donors were told this in addition to our expectation of how long it would take the student to do that many math problems, either "between 5 and 10 minutes" or "a little over an hour." If the donor was focusing on features on the intermediary, rather than the charity, we might find higher donations in the 150 problems treatment. We do not.

Figure A1 shows the histograms of donations, with the low-cost treatments on the left, the high-cost treatments on the right, 15 problems on top, and 150 problems on the bottom. The comparison for this analysis is to compare the top row to the bottom row, each side at a time and pooled. There is no evidence of a difference in any comparison. The pooled twosided rank sum (comparing all the data in the 15 problems versus all the data in the 150 problems treatment) yields a pvalue of .45. The p values are .82 and .36 for the low-cost and high-cost comparisons, respectively.

As with all null results, it is hard to conclusively say the hypothesis is incorrect. It is possible the treatment was too weak or subtle; perhaps the thought of doing any math problems sounds terribly onerous and worthy of support. Nonetheless, when we changed the amount of labor the intermediary has to do, by a factor of 10, we find no differences in donations. This perhaps suggests focusing on features of the intermediary, rather than the charity, is driving the convergence in donations we see under intermediation.

WORDING USED, BY STUDY AND TREATMENT

THREE-CHARITIES--DIRECT

(Screen 2 of 5)

In this survey, you will have the opportunity to donate to a charity (The Archaeological Conservancy).

For every 75 students who fill out this online survey, one student's name will be drawn who will win $80. Once you complete all five screens of questions, your name will be entered into this drawing.

(Screen 3 of 5)

On the next screen, you will have the opportunity to donate to the charity in the case that you win the $80 drawing. The drawing will happen (next Wednesday's date and time inserted).

(Screen 4 of 5)

Money will be donated online immediately after the drawing on Wednesday to the charity (The Archaeological Conservancy).

Imagine you win the $80 drawing.

How much of the $80 would you like to donate to the charity? (in dollars, no need to put a dollar sign)

THREE-CHARITIES--INTERMEDIARY

(Screen 2 of 5)

One Ohio State University student from the same subject pool from which you were recruited was randomly selected (and has agreed) to come in to Arps Hall next Wednesday afternoon, July 13.

They will be raising and anonymously donating money for charity. For 30 minutes on next Wednesday (July 13), this person will be going through the answers to these surveys, summing up the donations, and anonymously donating money online to a charity (The Community Foundation of Southeast Michigan). They do not know this yet. They will be told everything when they arrive.

For every 75 students who fill out this online survey, one student's name will be drawn who will win $80. Once you complete all five screens of questions, your name will be entered into this drawing.

(Screen 3 of 5)

The student who will be donating money to charity next Wednesday was randomly chosen.

He or she was not chosen based on knowledge of charities, nor has he or she been given the name of the charity or told what task they will be doing.

Also, this person is not making money based on your donations, nor will we reveal your decision to him or her. Your email address will be clipped from the data set before we pass it along.

On the next screen, you will have the opportunity to donate to this person's cause if you win the $80 drawing.

(Screen 4 of 5)

You can donate as much as you want to the student's fundraising effort.

The student will go online and donate all the money he/she raised immediately after he/she is done on Wednesday to the charity (The Community Foundation of Southeast Michigan).

Imagine you win the $80 drawing. We're going to offer you the chance to donate at least some of that, which the student will put together next Wednesday for his/her anonymous donation.

How much would you like to donate to the student's fundraising cause? (in dollars, out of the $80 you might win)

TWO-CHARITIES--DIRECT

For every 50 people that complete this survey, there will be a drawing for $65 (paid in the form of an Amazon gift card).

If you win that drawing, you will have the opportunity to donate some amount of your winnings.

If you win the $65, you can donate some to a very poor household in Kenya.

They have been verified as poor by an international nonprofit who has certified their house is made of mud, wood, and/or grass, a reliable indicator of extreme poverty.

You can donate money directly to this family. They will receive it by text message (they have been given a sim card; receiving money and paying for goods via text is relatively common in Kenya).

For every $1 you send. 10 cents will go towards various transaction and overhead costs (e.g. employee wages at the nonprofit, data collection by the nonprofit, PayPal fees, text messaging money fees, etc.).

How much would you like to donate? Enter a number between 0 and 65. Put 0 for no donation.

TWO-CHARITIES-INTERMEDIARY

For every 50 people that complete this survey, there will be a drawing for $65 (paid in the form of an Amazon gift card).

If you win that drawing, you will have the opportunity to donate some amount of your winnings.

If you win the $65, you can donate some to a student fundraising campaign.

Here is how that campaign works.

Although not everyone in the eLab participant pool is a Yale student, many are. We got a Yale student from the eLab pool to agree to come to our office next week for 1 hour. This student will be running the drawings to determine who wins the $65 (they would not get to choose; they will be pressing buttons on a computer that will randomly choose). They will also be running the student fundraising campaign. What this means is that they will go through all the donation decisions from the winners, sum them up, and he/she will go online and donate all of this money to a very poor household in Kenya. The household has been verified as poor by an international nonprofit who has certified their house is made of mud, wood, and/or grass, a reliable indicator of extreme poverty.

You can donate money to the student fundraising campaign. If you do so, the student who is coming in will donate to this family. The family will receive it by text message (they have been given a sim card; receiving money and paying for goods via text is relatively common in Kenya). For every $1 the student sends, 10 cents will go toward various transaction and overhead costs (e.g., employee wages at the nonprofit, data collection by the nonprofit, PayPal fees, text messaging money fees, etc.).

The student we recruited was randomly chosen from the email list. The student does not know what they will be doing when they come in next week. They will not make money based on your donation, nor will they have the opportunity to keep any of the money you send them. They are being paid a fiat fee for their help. They will not be shown your email address; e-mails will be replaced by random numbers for identifiers.

How much would you like to donate to the student fundraising campaign? Enter a number between 0 and 65. Put 0 for no donation.

INFORMATION OVERLOAD-DIRECT AND INDIRECT

Here, unformatted text is found in both scripts from Complexity, bold text is only found in the direct treatment, and italic text in parentheses is found only in the indirect treatment.

For every 50 people who complete this survey, there will be a drawing for $65 (paid in the form of an Amazon gift card).

If you win that drawing, you will have the opportunity to donate some amount of your winnings. If you win the $65, you can donate some to a very poor household in Kenya (student fundraising campaign).

Here is how that donation (campaign) will work.

Although not everyone in the eLab participant pool is a Yale student, many are. We got a Yale student from the eLab pool to agree to come to our office next week for 1 hour. This student will be running the drawings to determine who wins the $65 (they would not get to choose; they will be pressing buttons on a computer that will randomly choose). They will also be taking care of the donations (running the student fundraising campaign). What this means is that they will go through all the donation decisions from the winners, sum them up, and he/she will go online and donate all this money to a very poor household in Kenya. The household has been verified as poor by an international nonprofit who has certified their house is made of mud, wood, and/or grass, a reliable indicator of extreme poverty.

The family will receive it by text message (they have been given a sim card; receiving money and paying for goods via text is relatively common in Kenya). For every $1 the student sends, 10 cents will go toward various transaction and overhead costs (e.g., employee wages at the nonprofit, data collection by the nonprofit, PayPal fees, text messaging money fees. etc.).

The student we recruited was randomly chosen from the email list. The student does not know what they will be doing when they come in next week. They will not make money based on your donation, nor will they have the opportunity to keep any of the money you send them. They are being paid a flat fee for their help. They will not be shown your email address; emails will be replaced by random numbers for identifiers.

How much would you like to donate to the poor African family (student fundraising campaign)? Enter a number between 0 and 65. Put 0 for no donation.

ABBREVIATIONS

ABCF:   American Breast Cancer Foundation
AC:     Archaeological Conservancy
ACS:    American Cancer Society
CFSM:   Community Foundation of Southeast Michigan
DD:     Difference-in-Differences
OLS:    Ordinary Least Squares


doi: 10.1111/ecin.12379

Online Early publication July 25, 2016

REFERENCES

Ai. C., and E. C. Norton. "Interaction Terms in Logit and Probit Models." Economics Letters, 80, 2003, 123-29.

American Breast Cancer Foundation. Financial Statements. Baltimore. MD: The American Breast Cancer Foundation, Inc., 2012. Accessed August 6, 2012. http:// abcf.org/wp-content/uploads/2012/02/AmericanBreast-Cancer-Foundation-Financial-Statement-3-31-1 l.pdfhttp://abcf.org/wp-content/uploads/2012/ 02/American-Breast-Cancer- Foundation-FinancialStatement-3-31-1 l.pdf.

Bartling, B., and U. Fischbacher. "Shifting the Blame: On Delegation and Responsibility." Review of Economic Studies. 79(1), 2012, 67-87.

Bernhard, H., E. Fehr, and U. Fischbacher. "Group Affiliation and Altruistic Norm Enforcement." American Economic Review, 96(2), 2006, 217-21.

Charity Watch. Charity Rating Guide & Watchdog Report. Chicago. IL: Charity Watch, 2012.

Coffman, L. "Intermediation Reduces Punishment (and Reward)." American Economic Journal: Microeconomics, 3(4), 2011, 77-106.

Dana, J., D. M. Cain, and R. M. Dawes. "What You Don't Know Won't Hurt Me: Costly (but Quiet) Exit in Dictator Games." Organizational Behavior and Human Decision Processes, 100, 2006, 193-201.

DellaVigna, S., U. Malmendier, and J. List. "Testing for Altruism and Social Pressure in Charitable Giving." Quarterly Journal of Economics, 127(1), 2012, 1-56.

Drugov, M., J. Hamman, and D. Serra. "Intermediaries in Corruption: An Experiment." Experimental Economics, 17(1), 2014, 78-99.

Eisenkopf, G. and U. Fischbacher. "Doing Well by Doing Good--or Doing Better by Delegating?" Working Paper No. 63, Thurgauer Wirtschaftsinstitut, Universitat Konstanz, 2011.

Fershtman, C., and U. Gneezy. "Strategic Delegation: An Experiment." The RAND Journal of Economics, 32(2), 2001, 352-68.

Giving USA Foundation. Giving USA 2014: Highlights. Indianapolis, IN: Giving USA Foundation, 2015.

Greiner, B. "An Online Recruitment System for Economic Experiments," in Forschung und wissenschaftliches Reclinen 2003. GWDG Bericht63, edited by K. Kremer and V. Macho. Gottingen, Germany: Gesellschaft faur Wissenschaftliche Datenverarbeitung, 2004, 79-93.

Grossman, Z., and R. Oexel. "Shifting the Blame to a Powerless Intermediary." Experimental Economics. 16, 2013, 306-12.

Hamman, J., G. Loewenstein, and R. A. Weber. "Self-interest through Delegation: An Alternative Rationale for the Principal-Agent Relationship." American Economic Review, 100(4), 2010, 1826-46.

Kessler, J. "Signals of Support and Public Good Provision." Working Paper, The Wharton School of the University of Pennsylvania, 2011.

Kogut, T., and I. Ritov. "The 'Identified Victim' Effect: An Identified Group, or Just a Single Individual?" Journal of Behavioral Decision Making, 18(3), 2005, 157-67.

Landry, C. E., A. Lange, J. A. List, M. K. Price, and N. G. Rupp. "Toward an Understanding of the Economics of Charity: Evidence from a Field Experiment." Quarterly Journal of Economics, 121(2), 2006, 747-82.

Niederle, M., and L. Vesterlund. "Do Women Shy Away from Competition? Do Men Compete Too Much?" Quarterly Journal of Economics, 122(3),2007, 1067-101.

Null, C. "Warm Glow, Information, and Inefficient Charitable Giving." Journal of Public Economics, 95, 2011, 455-65.

Paharia, N., K. Kassam, J. Greene, and M. Bazerman. "Dirty Work, Clean Hands: The Moral Psychology of Indirect Agency." Organizational Behavior and Human Decision Processes, 109(2), 2009, 134-41.

Potters, J., M. Sefton, and L. Vesterlund. "After You: Endogenous Sequencing in Voluntary Contribution Games." Journal of Public Economics, 89(8), 2005, 1399-419.

Rose-Ackerman, S. "Charitable Giving and 'Excessive' Fundraising." Quarterly Journal of Economics, 97(2), 1982, 193-212.

Small, D. A., G. Loewenstein, and P. Slovic. "Sympathy and Callousness: The Impact of Deliberative Thought on Donations to Identifiable and Statistical Victims." Organizational Behavior and Human Decision Processes, 102(2), 2007, 143-53.

Van Diepen. M. "Dynamics and Competition in Charitable Giving." 2009. http://repub.eur.nl/res/pub/14526/ EPS2009159MKT9058921888vanDiepen.pdf

Vastfjall, D., E. Peters, and P. Slovic. "Compassion Fatigue: Donations and Affect are Greatest for a Single Child in Need." Unpublished Manuscript, University of Oregon. 2010.

Vesterlund, L. "The Informational Value of Sequential Fundraising." Journal of Public Economics, 87(3-4), 2003, 627-57.

World Bank. "United States: Data." Washington, DC. 2015. Accessed January 29, 2016. http://data.worldbank .org/country/united-stateshttp://data. worldbank.org/ country/united-states.

LUCAS C. COFFMAN

* This research is indebted to advice and feedback from Katherine Coffman, Max Bazerman, Daylian Cain, Eugene Caruso, Zoe Chance, Shane Frederick, John Kagel, Judd Kessler, Cade Massey, George Newman, Muriel Niederle, Alvin Roth, Aditya Sunderam, Lise Vesterlund and participants at Bazerman Non-Lab. CMU SDS, Colgate, Cornell, Notre Dame, NYU, SITE Experimental session, and ESA Tucson meetings.

Coffinan: Department of Economics, Ohio State University, Columbus, OH 43210. Phone 865-387-2084, Fax 614 292-3906, E-mail coffman.155@osu.edu

(1.) Total donations in 2014 are estimated to have been approximately $358 billion and giving by individuals was 72% of the sum (Giving USA Foundation 2015), and nominal GDP totaled $17.42 trillion (World Bank 2015).

(2.) On across-charity giving, see Rose-Ackerman (1982) for a model of fundraising and across-charity giving, and Van Diepen (2009) and Null (2011) for experimental analyses.

(3.) All but self-confidence are positively correlated with giving; self-confidence in the solicitor decreases giving.

(4.) All 6,501 students in the subject pool received an email, in a random order over the 3 weeks.

(5.) We used a Qualtrics survey platform.

(6.) At the top of the screen, it said, "Screen 1 of 5" giving the subjects an additional indication of the length.

(7.) Based in New Mexico, The Archaeological Conservancy acquires and preserves archaeological sites in the United States.

(8.) The first two charities were pre-tested in the laboratory with 30 subjects' $10 show-up fees, 15 per charity. Though not significant, the ACS distribution of gift sizes first-order stochastically dominated gifts received by the AC, with the exception of one $10 gift to the AC.

(9.) Many subjects' IP or e-mail address appears twice in the data; some second attempts quit on the second screen (the screen which indicates the name of the charity). These students are removed for the analysis.

(10.) When an e-mail address or IP address appeared more than once, every entry containing that address was excluded.

(11.) The 25 students who started the survey but did not finish were balanced across the direct (N = 12) and indirect (N = 13) treatments. Ten respondents agreed to the consent form but were not assigned to a treatment. We assume this was a software glitch; there are no data for these subjects.

(12.) That is. we measure at what percentile the observed difference-in-differences of donation averages lies within a bootstrapped distribution of mean DD. This distribution is produced by 10,000 iterations of drawing a data sample for each treatment cell of equal size from the actual data, with replacement. For example, if there were 50 donations in the direct setting for the preferred charity, we would draw 50 data points with replacement from this cell, and we would do so for all four cells. We then compute the DD for the four drawn cells, do this 10,000 times, and observe where our actual DD lies in this distribution.

(13.) We use the Ai & Norton (2003) probit interaction corrections for both the coefficient and standard error and report these in Models IV-VI.

(14.) We consider subjects who wrote down the word "Cancer," "Archeology" or "Archaeological," or "Michigan" to have recalled the charity name in their respective treatment.

(15.) The expected payout from a survey at the Yale eLab is generally slightly less than $1.

(16.) The nonprofit who does this verification as well as the actual transaction of money to poor households in Kenya (with 10% overhead) is Give Directly--www.givedirectly.org. Since overhead was varied in the study, we did not want the participants to know Give Directly's actual overhead, so they were not told the actual charity's name.

(17.) This claim comes from the results in Niederle and Vesterlund (2007).

TABLE 1

Bootstrap Difference-in-Differences p values

Treatment              Three-Charities   Two-Charities

Donation (a)                 .04              .04
Donation>0 dummy (b)         .04             <.01

Treatment              Math-for-Charity   Intermediary-Overhead

Donation (a)                 .03                   .02
Donation>0 dummy (b)         <.01                  .06

Treatment              Complexity

Donation (a)              .45
Donation>0 dummy (b)      .30

Notes: p Values are from a nonparametric bootstrap difference-in-
differences approach as explained in footnote 12.

(a) Top row uses size of donation as outcome.

(b) Bottom row uses a dummy that equals 1 if any donation was made.

TABLE 2

Difference-in-Differences Ordinary Least Squares (OLS) Regressions on
Amount Donated

                             OLS DV = Donation ($)

                        Three-Charities   Two-Charities
Treatment:                    (I)             (II)

Less-preferred               -11.2            -6.3
  charity                  (3.9) ***        (3.0) **
Less-preferred                9.1              7.5
  charity *                 (5.1) *          (4.2) *
Fundraiser dummy             -1.8             -1.7
                             (4.4)            (2.9)
Constant                     26.1             13.3
  (preferred charity)      (3.5) ***        (2.1) ***
Observations                  315              251

                                  OLS DV = Donation ($)

                        Math-for-Charity   Intermediary-Overhead
Treatment:                    (HI)                 (IV)

Less-preferred                -6.3                 -6.3
  charity                   (2.6) **             (2.9) **
Less-preferred                6.4                   7.7
  charity *                 (3.7) *               (4.1) *
Fundraiser dummy              -4.2                 -2.4
                             (2.7)                 (2.9)
Constant                      13.3                 13.3
  (preferred charity)      (1.9) ***             (2.1) ***
Observations                  254                   256

                        OLS DV = Donation ($)

                             Complexity
Treatment:                       (V)

Less-preferred                  -0.2
  charity                       (3.0)
Less-preferred                   0.5
  charity *                     (4.4)
Fundraiser dummy                -4.9
                                (3.1)
Constant                        17.3
  (preferred charity)         (2.1) ***
Observations                     304

Note: Robust standard errors in parentheses.

* p< .1; ** p < .05; *** p < .01.

TABLE 3 Difference-in-Differences Probit Regressions on Likelihood of
Donating

                            Probita DV = Donate {0,1}

                        Three-Charities   Two-Charities
Treatment                    (VI)             (VII)

Less-preferred               -0.12            -0.65
  charity                  (0.06) **       (0.23) ***
Less-preferred               0.17             0.33
  charity * (a)            (0.09) *        (0.12) ***
Fundraiser dummy             0.03             -0.31
                            (0.08)           (0.22)
Constant                     0.79             0.39
  (preferred charity)     (0.02) ***         (0.16)
Observations                  315              251

                               Probita DV = Donate {0,1}

                        Math-for-Charity   Intermediary-Overhead
Treatment                    (VIII)                (IX)

Less-preferred               -0.65                 -0.65
  charity                  (0.23) ***           (0.23) ***
Less-preferred                0.35                 0.20
  charity * (a)            (0.12) ***             (0.12)
Fundraiser dummy             -0.52                 -0.08
                           (0.23) **              (0.23)
Constant                      0.39                 0.39
  (preferred charity)      (0.16) **              (0.16)
Observations                  254                   256

                        Probita DV = Donate {0,1}

                               Complexity
Treatment                          (X)

Less-preferred                    -0.15
  charity                        (0.21)
Less-preferred                    0.06
  charity * (a)                  (0.11)
Fundraiser dummy                  -0.18
                                 (0.21)
Constant                          0.52
  (preferred charity)          (0.15) ***
Observations                       304

Note: Robust standard errors in parentheses.

(a) Probit interaction estimates and standard errors corrected as per
Ai and Norton (2003).

* p < .1; ** p < .05; *** p< .01.
COPYRIGHT 2017 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Coffman, Lucas C.
Publication:Economic Inquiry
Article Type:Abstract
Geographic Code:1USA
Date:Jan 1, 2017
Words:10869
Previous Article:Category reporting in charitable giving: an experimental analysis.
Next Article:Does option theory hold for major league baseball contracts?
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters