Charitable giving and intermediation: a principal agent problem with hidden prices

Abstract

Donations are often made through charitable intermediaries that can fund themselves from these same donations. After intermediation, only a fraction of the amount donated may reach the intended beneficiary. The price of charitable output is therefore higher after intermediation than if donors donated directly toward the end cause. At the same time, this price is hidden from donors since they cannot verify how much intermediaries pass on. We show that while donors reduce their donation in intermediation itself, and also reduce their donation because they expect the price of charitable output to increase, both reactions are either fully or partly compensated by their ethical preferences for the recipient’s rights. Charitable output, therefore, can be a Giffen-good.

1 Introduction

Donations to charitable causes are often made through intermediaries (Marshall, 1978; Hansmann, 1987; Coffman, 2017). A donor who entrusts a donation to an intermediary has little or no ability to determine whether the intermediary has performed the service well, or whether she has performed any service at all. Donors who give to intermediaries in order to support a philanthropic cause are effectively purchasing charitable output at hidden, i.e. unobservable, prices: donors do not know how much of their donation reaches the intended beneficiary and, therefore, donors do not know how much charitable output their donation eventually purchases.¹ While intermediaries provide revenue reports and voluntary feedback on their activities, this information is essentially unregulated in the field. Given this lack of regulation, feedback and filings are adjusted to the needs of the intermediary: actual efficiency grossly differs from reported efficiency, costs are misreported, and revenue filings manipulated. To illustrate the extent of the problem, more than 50% of all revenue filings report maximal efficiency with zero fundraising costs. Charity watchdogs condense these erroneous filings into charity rankings which provide intermediaries with additional incentives to appear as efficient as their competitors. Because the actual price of charitable output is hidden, competition does not reduce this price (Wing and Hager, 2004; Baird et al., 2013) and instead, results in more manipulation.²

Our goal in this article is to study how donors react to this hidden price problem, which is central to intermediation. To do so, one must access donors’ beliefs about the price at which they expect to purchase charitable output. If charitable output were a standard economic good, and donors believed that intermediation increases the price of that good, then donors would decrease their consumption, and intermediation should reduce donations.

Charitable output — the amount of a donation that reaches the intended beneficiary — however, is a commodity with ethical dimensions. Consider a donation towards the building of schools. The charitable output from this donation does not merely improve the beneficiary’s material circumstances or her living standard, it provides the beneficiary with some degree of education and empowers her to look after her own interest through participation in the labor market and the political arena. In this example, charitable output secures a civil right; in cases where the donation purchases infrastructure such as the access to clean water, charitable output preserves life and health and secures the recipients’ human rights. If donors hold preferences over these ethical dimensions of charitable output, then their donations may not decrease with intermediation. Indeed, we show that for a given preference, donors can maximize their utility by donating the same or more when they expect intermediaries to reduce donors’ donation toward the end cause. The idea is that a certain critical amount of charitable output is needed to secure the recipient’s inalienable rights. If the amount intermediaries are expected to pass on falls short of this amount, donors increase their donation to maintain a critical output.

In order to understand the effect of intermediation, we therefore need, first, to control donors’ beliefs about the price of charitable output with and without intermediation, second, donors’ ethical preferences, and third, the length of the intermediation chain. Since these requirements are hard to meet using observational data, we conduct an experiment in collaboration with small local communities of disadvantaged Indigenous Australians. We design an intermediated donation game, in which individuals can donate toward health and education programs in these local communities via another participant. This intermediary may take any amount of the donation for herself, ostensibly to remunerate herself for the task, and then decide to pass on the remaining amount to the end cause. All aspects of the situation are commonly known to donors and intermediaries, with the exception of the amount passed on by the intermediary. In addition to the intermediated game, donors also donate in a standard real donation task (Eckel and Grossman, 1996) where, since we control the intermediation chain outside the laboratory, donors give directly to the same communities of disadvantaged Indigenous Australians. In the intermediated task, we elicit donors’ beliefs about the price of charitable output. In the standard donation task, each Dollar donated also reaches the end cause, and the price of charitable output is therefore known.

To observe how different sets of ethical preferences give rise to different responses to intermediation, we design a within-subject experiment and vary the order of the donation tasks to examine the robustness of our main results. The choice of the local charitable initiatives, which fund health and education programs, helps control the length of the intermediation chain outside the laboratory, and ensures that any amount passed on by the intermediary reaches the intended beneficiary as directly as possible. This information is disclosed to donors in the laboratory at the time of the experiment and is therefore common knowledge.

We elicit donors’ ethical preferences, which we expect govern their reaction to intermediation, by means of a standardized psychological moral judgment test. Donors may apply ethical criteria to derive the right course of action which, in our study, is their reaction to the introduction of the intermediary. Kohlberg (1984) summarizes extensive field research on the ethical criteria people actually use to guide their own actions. The psychological test (see, for instance, Lind, 2008) we use in the experiment elicits donors’ preferences over the criteria described in this body of research. The test elicits whether donors refer to each ethical criterion at all to derive the right course of action, and to what extent they are influenced by it. Donors may, for example, consult their social image, others’ expectations, or a social norm to determine the right course of action; they may ask whether that course of action helps maintain the social order and is in line with the law, whether it protects individual civil rights as stipulated by the social contract, or some universal standard such as human rights, individuals’ freedom of choice, their will, and dignity.³ Donors may simultaneously consider all criteria by various degrees or none at all.

Our key findings are as follows. Average donations decline in the presence of intermediation itself, by roughly 16.5%. Donors’ ethical preferences to guide their actions by universal standards such as human rights, individuals’ freedom of choice, will, and dignity counterbalance this decline, and in 61% of all cases, are strong enough to fully compensate for it. Since this mechanism limits the decline of charitable output with intermediation and refers to inalienable rights of the intended beneficiary, we call it output-orientation. Next to intermediation itself, donations also decrease in the amount of money donors expect intermediaries to divert from the recipient, that is, in the price of charitable output, a mechanism which we call price-orientation. This effect is counterbalanced by a second channel of output-orientation. Donors’ ethical preferences to guide their actions by civil rights and the democratic social contract limit the decline of their donations in the expected price of charitable output by an average of 26%. Finally, we also investigate a mechanism related to donation-orientation. Donors may not care about charitable output, only derive utility from their own donation (Andreoni, 1990), and may therefore not respond to intermediation at all. These donors would be likely to guide their actions by ethical criteria such as their own social image, social norms and others’ expectations. While we observe donors who retain their donation in the face of intermediation, we do not observe them being influenced by these ethical criteria. In fact, we observe that the ethical criteria that we expected would bring about donation-orientation, instead magnify donors’ sensitivity to the price of charitable output.

We thus present the first study of intermediation in charitable giving with hidden prices, and show that donors’ reaction to intermediation depends on their ethical preferences. To date, intermediation has been conceptualized as a simple increase in the (known) price of charitable output (Gneezy et al., 2014; Coffman, 2017) or, for instance, as a transaction cost (Huck and Rasul, 2010). In these studies, donors are assumed to hold no beliefs about how intermediation affects the price of charitable output. Donors are asked to assume some (hypothetical) price while they are, at the same time, being correctly informed that any donation they make is merely fed into the intermediation chain outside the laboratory. The price of charitable output stipulated in these studies is the price the next intermediary outside the laboratory promises. Donors’ beliefs about this price remain uncontrolled as do their ethical preferences. These studies, understandably, report that charitable giving decreases as the hypothetical price of charitable output increases. There are also several insightful studies of intermediated giving in dictator games (see Coffman, 2011; Bartling and Fischbacher, 2012; Di Falco et al., 2020); however, these do not have charitable real-world end causes⁴ and do not control for the expected price of charitable output or donors’ ethical preferences.

2 Donor responses to the presence of an intermediary: a conceptual framework

Suppose that a donor who donates to an intermediary, derives utility from her own wealth, w_i, and from her donation, d_i. If the donor also derives utility from the amount which actually reaches the intended beneficiary — the intermediary’s charitable output $Z$ —, her utility is:

The intermediary’s charitable output $Z$ depends on the amount donated $di$ and on the intermediary’s productivity, x, of converting donations into charitable output. The more productive the intermediary, the higher her output $Z$⁠, i.e. $Z′x>0$⁠. Any amount the intermediary does not pass on to the intended beneficiary decreases the intermediary’s productivity since she converts donations into lesser output. As intermediary productivity decreases, each dollar of donation purchases lesser charitable output, and donors’ consumption of charitable output becomes more expensive relative to their noncharitable consumption. If charitable output is a normal good, then donors reduce their charitable consumption in response to this relative price increase. We denote a donor’s optimal bundle of noncharitable and charitable consumption before a price increase $(wi-di)*, Z(di,x)*$ and after the price increase$(wi-di)**, Z(di,x)**$⁠. For a normal good, we have $(wi-di)**>(wi-di)*$ and $Z(di,x)**<Z(di,x)*$⁠: i.e. donors increase their noncharitable consumption and decrease their consumption of charitable output. We describe donors whose donation decreases in their expectation about how much intermediation increases the price of charitable output, as being price-oriented.

Yet, the price of charitable output may not be a donor’s foremost concern in the face of intermediation. Consider a donor who seeks to secure some minimally acceptable living standard (such that donations should at least finance access to clean water), or some minimally acceptable degree of education (such that recipients learn at least to read and write, for instance). Her main concern is an ethical goal, and where charitable output fails to secure that goal, she receives zero utility from her donation. Suppose that, before intermediation, donations could purchase that minimally acceptable amount of charitable output which we denote by $Qi$ such that $Z(di,x)*>Qi$⁠. If intermediation increases the price of charitable output such that the donation can no longer purchase $Qi$⁠, then the donor’s utility from the current donation drops to zero, and she has an incentive to increase her donation until the donation purchases $Qi$ again. The donor reduces her noncharitable consumption $(wi-di)**<(wi-di)*$ to keep the quantity of charitable output constant, i.e. $Z(di,x)**=Z(di,x)*$⁠. Within the environment of $Qi$⁠, charitable output shows Giffen-behavior. We refer to the mechanism by which donations, driven by donors’ ethical preferences for civil or human rights, increase in intermediation output-orientation. The donor’s reaction to intermediation will also depend on others’ behavior. Where the donor expects that the overall donations exceed the minimally acceptable amount $Qi$⁠, she free-rides, and her donation decreases in the amount she expects to reach the intended beneficiary, again driven by her ethical preferences for civil or human rights.

Finally, donors may not care about charitable output, its price, and intermediation at all, for instance, if they donate to maintain their social image by signaling their own kindness and generosity to others. These donors only derive utility from their own donation $di$ (Andreoni, 1990; Minardi and Evren, 2017) because this donation itself affects their social image, is subject to a social norm, or to others’ expectations. Such donors simply maintain their donation under intermediation, they neither react to their beliefs about the price of charitable output nor to intermediation. Similarly, donors who, in deference to the authority of the church, consider it their duty to contribute to the collection or donors who, in deference to the authority of the law, contribute a fixed share of their wealth where no welfare state publicly provides for the disadvantaged and do not contribute where the law stipulates such a public provision, derive utility from donating, rather than from charitable output. We describe this mechanism in which donors donate because of social image concerns, social norms, others’ expectations, or to retain the social order and therefore simply maintain their donation under intermediation as donation-orientation.⁵

Summing up, the three donor responses to intermediation can be obtained assuming that donors have a utility function as described at the outset, and assuming a specific level $Qi$ of charitable output for each donor response below which said utility takes a value of zero.⁶ We obtain pure price-orientation if this $Qi$ is equal to zero and pure output-orientation if this $Qi$ is some $Qi>0$⁠. For the case of output-oriented compensators, for instance, who make up for the decrease in others’ donations, we have $Qi$ equal to overall donations before intermediation, i.e. $Qi=∑i=1nZdi,x$⁠. We obtain pure donation-orientation if this $Qi$ is equal to the nominal value of the donation (the price of charitable output $pZdi,x$ times the quantity of charitable output consumed), i.e. $Qi=pZdi,x·Zdi,x$⁠: donors receive zero utility if they donate too little to maintain their social image$.$⁷

Therefore, to identify price-, output-, and donation-orientation experimentally, in addition to donors’ beliefs about how much charitable output the intermediary converts into donations, we must access donors’ ethical preferences — that is, which ethical criteria donors prefer to apply to derive the right course of action. For this purpose, we draw upon an inventory of ethical criteria provided by Kohlberg (1984), using extensive data from the field, which consists of six classes of moral argumentation. In Kohlberg classes 1 and 2, individuals consider an action right if it is not punished by some authority, or if it is rewarded by that authority. In Kohlberg class 3, individuals deem an action right if it complies with a social norm, with others’ expectations, or if it assists the individual in maintaining her social image. In Kohlberg class 4, a course of action is right if it defers to authority, law, and duty to maintain the status quo and a fixed social order. In Kohlberg class 5, a course of action is right if it derives from a democratic social contract, that is, if the action respects that all individuals hold equal civil rights. In Kohlberg class 6, an action is right if it is derived from some universal principle such as the notion of human rights, the respect for the individual will, life, and human dignity. The resulting ethical preferences over these criteria are considered to be universally applicable parts of an individual’s decision-making toolbox. They only depend on the situational context in so far, as this context specifies the social norm or individuals’ beliefs where an individual uses these criteria to solve ethical problems.⁸ Importantly, in our context, the inventory is useful as it includes all ethical criteria relevant to charitable giving.

Lind (2008) provides a standardized moral judgment test to elicit donors’ preferences over all six categories of ethical criteria.⁹ We do not assume that donors only apply the criteria they prefer most. Instead, we use donors’ complete set of preferences over all six categories to model how they react to intermediation, and to their own beliefs about the price of charitable output. Supplementary Appendix A summarizes these motivations for ethical behavior. Chlaß and Riener (2015), Chlaß and Moffatt (2017), Chlaß et al., (2019), and Chlaß (2021) provide evidence that the ethical preferences issuing from the test yield consistent results across various economic setups such as simple distribution games, constant sum games, and auctions and do therefore appear to be stable across different contexts.

In Table 1, we present our hypotheses on how donors’ ethical preferences allow us to discriminate between price-, output-, and donation-orientation. Note that our hypotheses do not make use of Kohlberg classes 1 and 2 since our experiment does not include punishment or reward options. Empirically, however, and as explained in the results section, we include donors’ preferences over all six Kohlberg classes and test for their relevance in our context.

3 Experimental design

3.1 Overview

Subjects participate in four tasks. Instructions for each task were handed out at the beginning of each and can be found in Supplementary Appendices B and C. The first is a real-effort task where subjects earn money which can be used in the subsequent tasks. The second is a standard real-donation task in which donors donate directly to a local community of Indigenous Australians. The third is an intermediated donation task in which donors donate to the same local community of Indigenous Australians but this time, through another subject in the laboratory called ‘a Charitable Institution’. Subjects are informed that only one of the two donation tasks will be paid out. We next elicit subjects’ incentivized beliefs about the amount intermediaries pass on. Finally, subjects complete a pen-and-paper standardized psychological moral judgment test and an exit survey collecting a wide range of demographic information. All tasks are further detailed below. In addition to the setup described above, we compare two frames of the intermediation task between-subjects; one frame describing the amount the intermediary can take as remuneration and to cover administrative cost, the other as acting corruptly (Abbink and Ellman, 2010; Dyson, 2013). In additional sessions, we vary the sequence of tasks, first introducing the intermediated donation task and afterwards, the standard donation task to examine robustness to order effects.¹⁰

Experiments were conducted with university students at the MonLEE lab at Monash University, Australia. Subjects were invited to participate in the experiment using ORSEE, an opt-in web-based recruitment system (Greiner, 2015). Subjects were only invited if they had not previously participated in any similar experiment but were otherwise randomly chosen from all fields of study for all sessions. In each session, subjects were randomly seated at visually isolated cabins. Subjects were asked to leave no information which could be used to identify them and to make no contact with any other participant in the experiment. Thirteen sessions were run with a total of 178 participants. Sessions ran for approximately an hour and were conducted on computer using z-Tree software (Fischbacher, 2007). Subjects were paid a $5 show-up fee, $5 for completing the moral judgment test, plus their earnings from the experiment (consisting of the real-effort task, either the standard or intermediated donation task, and the belief elicitation task).¹¹ Average earnings from the experiment were $27 and ranged from $17 to $36.

3.2 Task 1: the real-effort task

Subjects are randomly assigned either Role A or Role B. These roles later translate into the donor and the intermediary role. Each subject completes a simple task counting the frequency of Ones or Fives in a box of numbers. The tasks are similar across both roles but different in terms of the box of numbers subjects see. Subjects are informed that each role faces a different set of questions, and that they will be allocated a payment for answering the questions, which they can use for decision making in tasks 2 and 3 of the experiment. All subjects receive $10 compensation for completing the real-effort task and are only informed of their own earnings. All subjects successfully completed the task.

The real-effort task serves two purposes: it allows donors an opportunity to give money they have earned and mimics a real-life income from which donations are made. It also compensates intermediaries for performing their role — as they would be in real life applications, e.g. through payments to CEOs and employees. In absence of any compensation, intermediaries may justify taking money from the donor’s donation as payment for their task.

3.3 Task 2: the real-donation task

After completing the real-effort task, all subjects are given instructions for the standard real-donation task in which they can choose to allocate whole dollar portions of their real-effort task earnings between themselves and a disadvantaged recipient (Eckel and Grossman, 1996), in this case, local communities of disadvantaged Indigenous Australians. In the standard real-donation task, the price of charitable output is known: each dollar donated reaches the intended recipient and, therefore, each dollar is converted into one dollar of charitable output. Participants’ donations are anonymous.

3.4 Task 3: the intermediation task

In the intermediated donation task, subjects learn whether they are donors or intermediaries. Role A from the real-effort task translates into the donor role, Role B into the intermediary role. For ease of understanding, intermediaries were referred to as ‘Charitable Institutions’. Each donor is randomly matched with one intermediary. The donor decides how much of her earnings from the real effort task to donate, i.e. d_i, to the disadvantaged recipient through the intermediary. The intermediary decides what percentage of the donor’s donation to keep for herself and what percentage to pass on to the recipient. The recipient are the same local communities of Indigenous Australians as before. Both donors and intermediaries make their decisions simultaneously. Intermediaries do not know the amount donated by their matched donors prior to making their decisions. Donors receive no information about what amount is taken by the intermediary before or after making their donation decision. Since subjects also have no information about the possible earnings of the other role in the real effort task, neither donors nor intermediaries know the earnings of the other players when making their decision. Otherwise, donors might expect intermediaries to equalize payoffs between both roles which would affect donors’ beliefs about how much of their donation ultimately reaches the recipients.

The intermediation task was presented in one of two different frames: either the amount taken by the intermediary was framed as corrupt behavior (cheating) or as the intermediary charging administrative cost (inefficiency). These appear to be the two main empirical reasons for intermediaries to redirect donations from the recipient, and being subject to ethical preferences themselves, may confound donors’ reaction to intermediation with hidden prices. Inefficiency involves little ethical wrongdoing, and in many countries, may not even be stigmatized by a social norm. In the absence of a strong ethical imperative, intermediaries may not consider there being much wrong with funding an overly large administration with donations. Donors may therefore expect high prices of charitable output and may be less inclined to compensate these prices. Corruption, on the other hand, may be considered wrong by many ethical criteria: social norms, the law, and even by individual principles of conscience. Donors may wish to shield the intended recipient from the effects of others’ unethical or criminal actions. Note, however, that both mechanisms may easily work in the opposite direction: donors may wish to compensate inefficiency since it involves no ethical wrongdoing but may not wish to condone and fund corruption. Since it is unclear whether donors’ ethical preferences over the reasons for diverting money counteract or confound donors’ reaction to intermediation with hidden prices, we explicitly control for these reasons and preferences. There is, however, no variation in beliefs or donations across the frames described, and output-, price-, and donation-orientation persist even in the corruption frame. We therefore pool the data.

3.5 Task 4: belief elicitation task and moral judgment task

After the intermediation task, beliefs are elicited. Donors are asked what average percentage of the donors’ donations intermediaries passed on to the recipient, and what average percentage of their earnings the other donors in the room donated. Intermediaries are asked what percentage of their earnings the donor they are matched with donated, and what average percentage the other intermediaries in the room chose to pass on. The closer these revealed beliefs are to the actual average percentage from each session, the higher subjects’ payoffs from the belief elicitation task (Selten, 1998).

Subsequently, subjects complete a Moral Judgment Test (Lind, 2008), which elicits their preferences over Kohlberg’s taxonomy of ethical criteria described in Section 2.¹² The test introduces two vignettes. The first portrays workers who break into a factory in order to steal evidence that their management was listening in on them; the other a doctor who assists a terminally ill patient to commit suicide upon that patient’s request. Subjects first submit their opinion on whether they deem the respective protagonist’s behavior to be right or wrong. Afterwards, the test lists an inventory of 24 arguments (12 after each vignette, two arguments per vignette pertaining to the same Kohlberg class from Supplementary Appendix A) and asks subjects on a 9-point scale, how much they would agree/disagree to use each argument in order to judge whether the protagonist’s behavior was right (6 of 12 arguments per vignette) or wrong (another six arguments per vignette). Each argument refers to an ethical criterion belonging to one Kohlberg class such that four test items refer to the same class. We obtain subjects’ preference over each Kohlberg class as the average rank given over the four arguments pertaining to this class, adjusted for subjects’ personal use of the ranking scale (the difference between the maximal and minimal rank a subject ever ticks in the entire test). The test purposely does not refer to the donation experiment and is designed such that in a sample of subjects who do not necessarily give their real opinions in the test but try, for instance, to answer in what they deem a socially acceptable or appropriate way, the distribution of the scores obtained is not biased (Lind, 2008).¹³ Previous research finds that Kohlberg classes 3 and 6 scores from the test explain dictator game giving (Chlaß and Moffatt, 2017), that Kohlberg classes 5 and 6 scores explain lying and sabotaging aversion in competitive setups (Chlaß and Riener, 2015) or overbidding in common value auctions (Chlaß, 2021). Chlaß et al. (2019) show that Kohlberg class 5 scores can be used as an instrument for purely procedural preferences.

Finally, subjects complete an exit survey to collect relevant demographics which might affect donations, beliefs, and moral judgment. The payoff-relevant task was then randomly selected with equal probability, and donations from the entire session in this task were donated to small local groups of Indigenous Australians which provide health and education programs in disadvantaged Indigenous communities.¹⁴ A volunteer from the participants was asked to help make the aggregate donation to the association via their website in front of all subjects.

3.6 Research questions

We investigate whether donors’ reactions to intermediation are:

• price-oriented in that donations significantly decrease in the price of charitable output per se, that is, increase in the share donors expect intermediaries to pass on.

• output-oriented in that donations decrease significantly less in the intermediation task and/or in the price of charitable output, the stronger donors’ preferences to derive the right course of action from the social contract, civil and human rights (Kohlberg classes 5 and 6).

• donation-oriented in that donations significantly increase in donors’ preference to derive the right course of action from their social image, others’ expectations, a social norm, or a fixed social order (Kohlberg classes 3 and 4), regardless of whether there is intermediation or not.

4 Results

4.1 Descriptives: donations, beliefs, and ethical preferences

In the standard real donation task, the average donation by all 150 participants is $2.73 out of $10 (27%). Forty participants (27%) give zero. This is consistent with the existing literature. For instance, Eckel and Grossman (1996) report that subjects in the laboratory give on average 31% of the pie in a real donation experiment, while 27% of subjects give zero. Using an interval regression of the amount donated, we find no significant difference between donors (Role A) and intermediaries (Role B) in terms of the average donation in the real donation task, suggesting that subjects were effectively randomized into roles (p¹⁵ = 0.67).

Table 2 summarizes the number of subjects who give (and the average donation amount) in each task by role and frame. The average amount donated by the 75 donors in the intermediation task is $2.25. This is significantly less (p = 0.02) than the average of $2.83 these same donors donated in the real donation task. Figure 1 uses violin plots (Hintze and Nelson, 1998) to compare the distribution of donations across both tasks, expressed in terms of shares of the pie. Shares donated in the intermediation task are indeed smaller than in the real donation task. Table 3 disaggregates the data and shows that among 61 donors who donate in at least one of the two tasks, 16.4% give more in the intermediation task (an average of $2.80 compared with $1.20 in the real donation task), 42.6% give less ($1.85 compared with $4.12), and 41% give the same positive amount ($3.72).

These differences in donors’ reactions to intermediation are not merely due to different beliefs about how much intermediaries pass on to the intended recipient. Donors who give more, or less, hold similar beliefs about how much intermediaries pass on and therefore expect the same price increase for charitable output under intermediation. To be precise, donors expect intermediaries to pass on an average half of their donation (51.44%) and an average charitable output¹⁶ of $1.38 in the intermediation task.

Donors who give more in the intermediation task expect the same amount of charitable output with and without intermediation (interval regression of expected whole Dollars, p = 0.66), suggesting they may have aimed at keeping that output constant. Donors who give less or the same amount with intermediation, also expected to purchase lesser output. Table 3 shows that these patterns can be found in both the inefficiency and the corruption frame. Descriptive statistics therefore suggest that donors show genuinely different reactions to the introduction of an intermediary; that mechanisms of price-, output-, and donation-orientation might indeed be at work, outweighing each other differently for each donor.

Turning to donors’ ethical preferences, Figure 2(a) plots donors’ donations in the intermediation task against their beliefs about the share intermediaries pass on, along with a blue kernel regression line. Bubbles represent donors and are sized according to donors’ Kohlberg class 5 scores: The larger the score, the larger the bubble and lighter its color. We observe a strong increase of donations in the share donors believe intermediaries to pass on, and observe that in particular lighter shaded large bubbles fall below the kernel regression line. That is, donors with stronger preferences for Kohlberg class 5 seem to increase their donations less as the price of charitable output decreases and vice versa, to decrease them less as this price increases. Figure 2(b) shows that such a link does not exist for Kohlberg class 6: light-shaded large and dark-shaded small bubbles seem to fall to an equal extent below and above the regression line. Preferences for Kohlberg class 6 do not seem to interact with donor beliefs.

4.2 Price-orientation

Appending the data from the real donation and the intermediation task, we construct a panel data set which includes two observations for each donor. Donations being interval-censored, we perform interval regressions of donations on the key determinants of price-orientation: (i) a dummy for the intermediation task and (ii) the share donors expect intermediaries to pass on to the intended recipients. To account for the correlation structure of the data, we cluster errors at the individual level and include session fixed effects. We also control for demographics including age, gender, religion, religiosity, nationality, fields of study, ethnicity and socio-economic status. In order to avoid omitted variable bias and, at the same time, keep the estimator as precise (efficient) as possible, all regression specifications are tested downwards, removing variables that are insignificant and without impact on the goodness of fit, unless they are of theoretical interest.¹⁷ Columns 3 and 4 in Table 4 show the results of our two final specifications, with and without additional controls. Donations decrease in the expected price of charitable output — donors donate more, the more they believe intermediaries will pass on (the lower the expected price of charitable output). The effect is highly significant (5.279, p = 0.001) and robust to the inclusion of demographics (4.548, p = 0.001). Table 5 shows that the price of charitable output continues to be highly significant after controlling for the complete set of donors’ ethical preferences (Kohlberg classes 1, 2, 3, 4, 5, and 6, whereby 1 and 2 are insignificant and dropped), including their interactions with the intermediation task, and the share intermediaries are expected to pass on.¹⁸ Since the share intermediaries are expected to pass on has a robust stand-alone effect, we conclude that donations do indeed react to how intermediation changes the price of charitable output: we observe price-orientation.

It is noteworthy that in Table 4, donations significantly decrease in the intermediation task dummy itself (−0.698, p = 0.016), even after controlling for donors’ beliefs about the share intermediaries pass on in column three (−3.521, p = 0.000) and for donors’ demographics in column four (−3.155, p = 0.000). This implies that intermediation itself makes charitable output costlier to the donor, beyond the share of the donation taken by the intermediary. This additional cost could be a notion of trust betrayal as in Bohnet and Zeckhauser (2004) where individuals are more reluctant to take risks when a given outcome is due to another player being trustworthy or not, rather than due to chance.

4.3 Output-orientation

Next, we analyze whether donors’ preferences to derive the right course of action by means of the social contract and the civil rights granted therein (Kohlberg class 5) or by means of universal principles such as human rights, human dignity, and respect for the individual will (Kohlberg class 6), moderate the (negative) effects of intermediation. In Section 2, we hypothesized that these ethical preferences could reverse the link between donations and the price of charitable output. Table 5 details the corresponding interaction effects of Kohlberg classes 5 and 6 with the intermediation task, and with the share intermediaries are expected to pass on. We analyze the ethical preferences operating on the intermediation task (Columns 1 and 2) separately from those operating on donors’ beliefs (Columns 3 and 4) to limit collinearity since they will require different controls in Supplementary Appendix D.4.

Two effects moderate the decline of donations due to intermediation. First, Kohlberg class 6 significantly interacts with the intermediation task dummy (1.370, p = 0.014), moderating the extent to which donations decrease in intermediation itself. The effect is robust to the inclusion of demographics (1.443, p = 0.011). In fact, there are 46 donors (61%) whose ethical preferences for Kohlberg class 6 are strong enough to overcompensate the negative main effect of the intermediation task. Figure 3(a) presents a histogram of the total interaction effect of Kohlberg class 6 divided by the main effect for the intermediation task in our sample. This share ranges from 0 to a maximum of 4.37. Consider a donor for whom this share is 4.37. This donor’s preferences for Kohlberg class 6 can compensate 4.37 times the trust betrayal cost (i.e. it is 4.37 times as high as the coefficient of the intermediation task). To show the distribution of these compensatory shares in our sample, we superimpose a violin plot on the top right of the graph. The violin plot also shows the median share (1.16) and the interquartile range (the lower end of the thick black line shows the first quartile of 0.68 and the upper end depicts the third quartile of 1.53).

Second, Kohlberg class 5 significantly interacts with the share donors expect intermediaries to pass on (−1.917, p = 0.023), the effect being robust to the inclusion of demographics (−1.899, p = 0.019). That is, the lower the expected price of charitable output, the more do donations decrease in this price. Vice versa, the higher the expected price, the smaller the decline of donations in the price of charitable output. In fact, the interaction reduces the total decline of donations in the expected price of charitable output by an average of 26%. Figure 3(b) shows a histogram of the total interaction effect of Kohlberg class 5 divided by the total effect of the share intermediaries are expected to pass on in absolute terms. This share ranges from 0 to 1.39. Consider the donor whose compensatory share is 1.39. Her preferences for Kohlberg class 5 can compensate 1.39 times the total effect of her belief about the share intermediaries pass on. The superimposed violin plot shows the median share (0.20) and the interquartile range (first quartile of 0.10 and third quartile of 0.40). Note that neither Kohlberg class 5 nor 6 have a stand-alone effect on donations: donors seem to consult these ethical criteria solely to determine the reaction to intermediation but not to decide on the donation itself. To conclude, we observe two channels that work against the decline of donations in intermediation: two channels of output-orientation that moderate donors’ price-orientation. These channels are robust to including all other Kohlberg classes as well as their interactions with the intermediation task and the expected price of charitable output, see Supplementary Appendix Tables A9 and A10.

4.4 Donation-orientation

Finally, we analyze whether there are donors who are only concerned with the donation itself, and do not consider charitable output at all. These donors would refer to ethical criteria which imply the same donation with and without intermediation and would apply to the donation itself rather than to charitable output. The relevant ethical criteria should therefore have a significant stand-alone effect in our specifications. Classical examples for such ethical criteria are social image concerns, social norms, and others’ expectations: donors donate to signal their own generosity, because they are expected to, and in order to be taken for a good person (Kohlberg class 3). Interestingly, we find that Kohlberg class 3 has no significant stand-alone effect on donations (−0.121, p = 0.789) before or after controlling for demographics (−0.258, p = 0.566). We therefore do not observe pure donation-orientation. Instead, donors’ preferences to consult these criteria make donations even more sensitive to intermediation. The respective interaction effect of Kohlberg class 3 with the price of charitable output is significant (2.036, p = 0.014) and robust to the inclusion of demographics (2.019, p = 0.014). We conclude that donors who do not respond to intermediation do so out of pure output-orientation (in which case charitable output is a borderline Giffen-good) or out of a mixture of price- and output-orientation.

It is possible, however, that two donation situations in which the intermediary is expected to pass on different shares, are also situations where the social norm or social expectations about giving differ. In this case, the price of charitable output would convey two different norms or expectations about how much donors should give, and donors who would normally only consider their donation, look towards the price of charitable output to infer the relevant social norm and what is expected of them. If the price is low, then one might be expected to give a lot and not be greedy. In this case, Kohlberg class 3 would increase the extent to which the price of charitable output matters, a scenario which is particularly likely where this price results from the strategic actions of another agent as we discussed in footnote 7. It is noteworthy that donor and intermediary beliefs do not correlate with their Kohlberg scores. In Section 2, we also hypothesized that donors who donate in deference to some authority such as the law or the church or to retain a fixed social order (Kohlberg class 4) would be concerned with the donation itself rather than the amount of charitable output and hence be donation-oriented. Yet, Kohlberg class 4 has no stand-alone effect on donations either, neither before (−0.261, p = 0.611) nor after controlling for demographics (−0.333, p  = 0.520). Instead, we observe a weak interaction with the intermediation task dummy (−0.708, p  = 0.063), and a somewhat stronger one after controlling for demographics (−0.761, p = 0.047), which we interpret such that donors think it is against the law that intermediaries can divert money for any reason. Religiosity in turn has a direct stand-alone effect on donations, aside from Kohlberg class 4 which implies that religious donors draw their motivation to give from the contents of their religion itself, rather than in deference to the institutions behind it.

4.5 Robustness of price-, output-, and donation-orientation

In Supplementary Appendix D, we report that price-, output-, and donation-orientation exist across different contexts within our experiment, in particular, across the respective frames of corruption and inefficiency in the intermediation task, the order of standard donation and intermediation task, and across donor and intermediary role. We also investigate whether the effects of donors’ ethical preferences can be attributed to omitted demographic information. In all cases, our main findings remain robust.

5 Contributions and limitations

We observe that intermediation with hidden prices makes donors consider ethical motives, which do not play a role when donors donate directly to the recipient. The mere fact that an intermediary is introduced who can divert donations at her own discretion, gives rise to ethical dimensions that donors did not attribute to charitable output before. Output-orientation therefore mainly emerges with intermediation; charitable output is unlikely to display Giffen or borderline Giffen behavior if price increases do not result from the immediate actions of another agent but instead, from exogenous shocks such as inflation or war.

In the field, donations are often observed by a wide audience whereas in our experiment, they remain anonymous even from the intermediary. We therefore expect that in the field, donors who put store by social norms and expectations will — contrary to our experiment — also be concerned with the donation itself, and not only with the expected price of charitable output.

Similarly, donations in the field may pass an entire chain of intermediation while we focus on the causal effect of a single intermediary. Since we find that not only donors but also intermediaries show compensatory efforts, a longer chain has more agents to compensate price increases. The agents early on (i.e. donors, early intermediaries) may expect agents further down the chain to exert the relevant compensatory efforts and consequently free-ride more often than in our setup.¹⁹ Donors could, however, also simply expect less charitable output with more intermediation, and hence exert more compensatory effort than we observe. Note that the mechanisms described here require a proper understanding of the effects a single intermediary has, which our study provides.

Finally, one might argue that concerns for human and civil rights, human dignity, and freedom of choice limit our findings to rather specific real-world end causes. Note, however, that real-world end causes such as health, education but also more generally, abating hunger, death rates, and building infrastructure in developing countries raise substantial philanthropic effort to which the ethical criteria above apply immediately. We therefore expect that Giffen-good or borderline Giffen-good behavior of charitable output are regular phenomena outside the laboratory. Similarly, note that the core point of the study, that prices are hidden, cannot be remedied by supposed quality signals such as third-party ratings of competing intermediaries in the field: these ratings depend on unverified and unverifiable intermediary communications themselves. Along these lines, we do not expect experience to matter either. On the one hand, no feedback on payoffs or behaviour can be obtained with hidden prices; on the other hand, the ethical preferences underneath the mechanisms we explore do not seem to crowd out if the same experimental task is repeated (Chlaß and Moffatt, 2017; Chlaß, 2021).

6 Conclusion

We investigate how the existence of an intermediary and donors’ expectations about the behavior of this intermediary affect charitable giving when not just the actions of the intermediary, but also the consequences of these actions are hidden from donors. We compare giving in a standard real donation experiment with an intermediated donation game where donors donate through an intermediary played by another participant in the experiment. We collaborate with small, local recipient groups, such that upon leaving the experiment all donations are transferred to these intended groups, to avoid any additional field intermediators who might have uncontrolled effects on donors’ expectations about how much of a donation ultimately reaches the recipient.

Our main idea is that donors who donate through an intermediary, seek to purchase charitable output, that charitable output is a commodity with ethical dimensions, and that donors’ reaction to intermediation must therefore depend on donors’ ethical preferences. Starting with the reference case of a standard economic model that an increase in the price of a commodity will reduce the consumption of that commodity relative to other goods, we posit a mechanism of price-orientation. If donors expect intermediaries to take from their donation, then donors expect a lesser amount to reach the intended beneficiary, and therefore expect a higher price of charitable output. They purchase less charitable output relative to other goods as its price increases and therefore reduce their donation. We posit, second, a mechanism of output-orientation. If donors consult certain ethical principles such as human rights, human dignity, and the individual’s freedom to choose, or the individual’s civil rights when opting for a course of action, they may wish to maintain a critical level of charitable output, irrespective of its cost relative to other goods. Since by this reasoning, a certain amount of charitable output is linked to some inalienable right of the intended beneficiary, donors must increase their donation to purchase this amount when its price increases, turning charitable output into a Giffen-good. Output-orientation therefore weakens the extent to which donors consider the price increase of charitable output due to intermediation. We posit, third, a mechanism of donation-orientation that leads donors to maintain their donation under intermediation. Within this mechanism, in addition to price or the recipient’s rights, donors may consider how generous others believe them to be or how much others expect them to donate when making a donation; these ethical preferences apply to the donation itself rather than to charitable output.

We observe that donors significantly reduce their donations in the amount which they expect intermediaries to divert from the intended beneficiary: i.e., we observe price-orientation. This decline is, however, significantly moderated by donors’ ethical considerations for others’ civil rights as granted by the democratic social contract: The more donors refer to the latter, the less they decrease their donations in the amounts diverted. We therefore also observe output-orientation which compensates, on average, 26% of the price effect. Interestingly, donations decrease in intermediation per se even after controlling for donors’ beliefs on intermediary behavior. We interpret this result as a cost of potential trust betrayal. In most cases, this decline in intermediation is fully overcompensated by donors’ considerations of others’ human rights, their dignity, and freedom to choose. Surprisingly, we do not observe donation-orientation: in fact, donors’ preferences to consider others’ expectations, social norms, and their own social image magnify the effect of price. This may be because in the specific case of intermediation with hidden prices, the expected price of charitable output are donors’ expectations about others’ behavior, and price itself may therefore describe a social norm.

Our findings help inform the mixed results in the literature on the impact of charity effectiveness. For instance, Null (2011), Butera and Horn (2017), and Karlan and Wood (2017) find that overall, donors are unresponsive to charity effectiveness, whereas Adena et al. (2019) report that introducing quality certification increases giving by new, but not existing donors. We show that donors’ reaction to intermediation depends on the ethical considerations they refer to, as well as their beliefs regarding the ultimate cost of charitable output. Donors may, in particular, donate more when the price of giving goes up (Adena and Huck, 2022). The increase in giving in such cases is not due to idiosyncrasies of the data; rather, a genuine reaction of donors, which needs to be accounted for in the charitable giving literature.

Footnotes

Supplementary material

Supplementary material is available on the OUP website. These are the data and replication files and the online appendix.

Funding

This work was supported by the Science of Philanthropy Initiative; and Australian Research Council Grants [DP1094676, DP140101900].

Acknowledgments

We thank Philip Grossman, Stephen Knowles, Sherry Li, seminar participants at Monash University, the editors Maja Adena, Michael Drouvelis, Steffen Huck, and three anonymous referees for their input.

References