Time-Inconsistent Charitable Giving

Full text

Time-Inconsistent Charitable Giving*
James Andreoni
University of California, San Diego and NBER
Marta Serra-Garcia
University of California, San Diego and CESifo
November 14, 2016
Abstract
This paper examines the interaction between moral contradictions and time in char-
itable giving. Applying a simple theoretical framework to two longitudinal experiments
with actual charitable donations, we show that moral contradictions become the source of
a new kind of time inconsistency linked to a demand for flexibility, rather than the more
typical demand for commitment. This kind of time inconsistency coexists with the oppo-
site of kind of time inconsistency arising from temptation to give, which is exhibited by a
substantial minority of individuals. Our results reveal that time inconsistency is pervasive
and exhibits unique features in the charitable domain.
JEL classification: D64, D90, C91.
Keywords: prosocial behavior, charitable giving, pledging, intertemporal choice.
*Andreoni: University of California, San Diego, Department of Economics, 9500 Gilman Drive, La Jolla, CA
92093 (andreoni@ucsd.edu); Serra-Garcia: University of California, San Diego, Rady School of Management,
9500 Gilman Drive, La Jolla, CA 92093. We are grateful to Menusch Khadjavi, David Reiley, Charlie Sprenger,
and Bertil Tungodden for very helpful comments. This research was conducted under IRB #140762. We would
like to thank the National Science Foundation, grant SES-1427355, the Science of Philanthropy Initiative, the
John Templeton Foundation, and internal funds from UCSD for financial support.

1 Introduction
For many years research on charitable giving has focused on understanding how utility flows
from the act of giving. We have learned that giving can be intrinsically joyful (e.g., Andreoni
1989, 1990; Ribar and Wilhelm, 2002), enhance self-image (e.g., Benabou and Tirole, 2006),
improve social-image (e.g., Andreoni and Bernheim, 2009), and be a source of prestige (Har-
baugh, 1998). Giving can also have less clearly positive effects. It can create moral contra-
dictions for givers (e.g., Dana, Cain and Dawes, 2006; Dana, Weber and Kuang, 2007), and
can be manipulated by social pressure and empathy. Some individuals enjoy saying yes to a
fundraiser (Crumpler and Grossman, 2008; Andreoni and Rao, 2011; Chowdhury and Jeon,
2015), some dislike saying no, others will have a good or bad feeling just from being asked
(DellaVigna, List, Malmendier, 2015; Trachtman et al., 2015; Andreoni, Rao, and Trachtman,
2016).1
These avenues for utility have been studied from a static perspective thus far, where act of
giving occurred at a single point in time. In this paper we show that adding time to the decision
to give enhances the importance of moral issues in giving. This has three main implications.
First, if individuals do not enjoy giving, but do dislike saying no to a fundraiser, then introduc-
ing time in the giving process can generate additional giving: when a commitment to a future
gift is solicited, individuals who commit avoid saying no immediately, while delaying the cost
of giving. Second, pledges to give in the future are often made, but do not increase giving, since
most pledges are insincere. Third, and most importantly, moral contradictions become a new
source of dynamically inconsistent behavior, with features that are qualitatively different from
dynamic inconsistencies that have been well documented in standard consumption choices (for
a review, see, Frederick, Loewenstein and O’Donoghue, 2002). Dynamic inconsistencies that
arise with moral contradictions are tied to the desire for flexibility and, hence, unlike other
consumption choices, with a strict preference not to commit to choices made in advance.
In this paper, we conduct two longitudinal experiments to examine the impact of the intro-
duction of time in actual donation decisions. To the best of our knowledge, these are the first
1For reviews, see Andreoni (2006), List (2011), and Andreoni and Payne (2013).
1

experiments to provide a comprehensive study of the dynamics of charitable giving. We also
develop a stylized theoretical framework that captures the interaction between moral contra-
dictions and time in the giving process. We find empirical support for the three implications
above, all of which can be explained within our theoretical framework.
The theoretical framework we propose is based on the idea that charitable giving is a collec-
tion of decisions and transactions, each of which brings their own flows of utility, money and
production of charitable goods. At the beginning of the process, after the ask has been made,
the individual faces the decision whether to say yes or no to the fundraiser. Several recent
papers suggest that individuals may derive utility from saying yes (Crumpler and Grossman,
2008; Andreoni and Rao, 2011; Chowdhury and Jeon, 2015) and disutility when saying no
(DellaVigna, List, Malmendier, 2015; Trachtman et al., 2015; Andreoni, Rao, and Trachtman,
2016), such that some may be willing to incur costs to avoid the ask. The model thus assumes
that individuals derive utility from making a giving decision, which is separate from the util-
ity experienced when paying for the gift. Hence, we depart from standard models of giving,
which typically assume that the utility from giving is only generated by the act of giving (e.g.,
Andreoni, 1989, 1990).
This simple change yields several predictions. We consider first giving choices when gifts
are delayed, and individuals commit in advance to their giving decision. The first result is that
more individuals will give when the decision to give is made in advance, than when the decision
and payment of the gift both take place immediately. The increase in giving is generated by
individuals who dislike giving, that is, those for whom the utility generated when making the
gift is lower than its cost. They choose not to give when the gift is paid immediately, but choose
to give when the cost of giving is delayed, as giving is made more attractive due to simple
discounting of its cost, relative to the immediate cost of saying no. Second, these dynamically
inconsistent individuals exhibit a strict preference for flexibility. If there is a future opportunity
to revise their commitment, individuals choose to give in advance, but choose not to commit to
that choice, leaving for themselves room to revise their choice towards saying no later.
We next consider giving choices when individuals pledge to give in the future, where
pledges are non-binding. Pledges offer individuals an opportunity to delay the cost of say-
2

ing no. However, reneging may carry moral costs derived from breaking a promise or stated
intention (e.g., Charness and Dufwenberg, 2006; Serra-Garcia, van Damme and Potters, 2013).
This leads to three predictions. First, many will choose to pledge, but pledging will not increase
ultimate giving significantly, unless the moral costs of reneging on a pledge are high. Second,
offering both the opportunity to pledge and to give immediately are offered, will induce selec-
tion. Those types who choose to pledge are overwhelmingly those who are delaying the no,
and hence will renege frequently. Third, having obtained a pledge from individuals who intend
to renege, one may increase the “social pressure” to confirm the pledge and donate by thanking
individuals who pledge, and thus decrease reneging.2
In our first experiment we examine the effect of introducing time in the giving process,
while in the second we provide within-subjects evidence of dynamic inconsistency and its re-
lation with commitment demand. In the first experiment, over 690 undergraduate students
participated in a longitudinal experiment, that took place over two weeks, in which they were
asked make an actual donation of $5 to GiveDirectly, a charity providing cash transfers to indi-
viduals living in extreme poverty in Kenya and Uganda.3In the baseline treatment, donations
were made immediately in the first week of the study. In this treatment, 31% of subjects choose
to give. In contrast, when individuals were asked to commit in the first week to donate in the
second week of the study, donation rates increased to 45%, an increase of nearly 50% in giving.
This provides evidence that a significant fraction of individuals are time inconsistent or, more
precisely, violate time stationarity (Halevy, 2015).4
When pledging is introduced as the only form of giving, this results in a high rate of pledg-
ing, of over 65%. However, over 45% of individuals who pledge in the first week choose to
renege on their pledge in the second week of the study, leading to an ultimate level of giving
that is indistinguishable from that achieved when requesting immediate gifts. If donors are
2Thank-you notes are a frequently-used strategy by charities to cultivate repeat donors. Here, we consider their
effectiveness within a single fundraising campaign, to increase the likelihood of converting pledges into gifts.
3Details regarding the procedures of the experiment are provided below.
4Evidence of violations of time stationarity is also provided in two large-scale field experiments by Breman
(2011). Among employees who are already donating to charity, she finds that requests to increase donations are
significantly more effective when the increases are delayed, rather than immediate. These experiments however
do not provide evidence of within-subject dynamic inconsistency and commitment demand, two features that are
crucial in understanding the source of violations of time stationarity.
3

offered the choice to give immediately or pledge in the first week, we observe 21% choosing
to give immediately. There is evidence of selection among self-selected pledgers, who display
a significantly higher rate of reneging, of over 70%, relative to the treatment in which pledging
is the only form of giving. Further, when individuals who pledge are thanked for their pledge
the same day they pledge, and a week before they are asked to confirm their pledge, reneging
decreases and ultimate giving increases. These findings are all consistent with a significant
majority of our subjects exhibiting moral contradictions as described in our framework, where
individuals display a desire to avoid saying no to the fundraiser and, at the same time, to avoid
giving.
Our second experiment examines in detail the source of the violations of time stationar-
ity observed in our first between-subjects experiment. According to our framework, dynamic
inconsistency is a result of moral contradictions, which yields the unique prediction that indi-
viduals who are dynamically inconsistent – that is, who choose to donate in advance, but not
immediately – will not demand commitment, but will, in fact, have a preference for flexibil-
ity. This contrasts with dynamic inconsistency that is generated by temptation or self-control
problems (e.g., Laibson, 1997; Fudenberg and Levine, 2006; and Gul and Pesendorfer, 2007).
Dreber et al. (2016) show that individuals may face self-control problems in the social do-
main, where the short-run self is assumed to be altruistic, while the long-run self is assumed
to be selfish. Their model predicts the opposite of type of dynamic inconsistency than we doc-
umented in our first experiment. Individuals in their model are less likely to give when the
decision is made in advance, compared to when giving is immediate, and exhibit a preference
for commitment.5Our second experiment hence uses a within-subjects design that identifies
both dynamic inconsistency and commitment demand.
Our results reveal that 37% of the subjects are dynamically inconsistent, making different
donation decisions depending on whether these concern immediate or delayed gifts. Among
these, a majority (62%) exhibit the type of dynamic inconsistency documented in our first
5The predictions of Dreber et al. (2016) have found support using dictator games (see also, Kovarik, 2009),
and closely relate to experiments manipulating time pressure and cognitive load in social dilemmas (e.g., Rand,
Greene and Novak, 2012; Rand et al., 2014a,b), although other evidence in the latter settings is mixed (e.g.,
Kessler, Kivimaki, Niederle, 2016, and Recalde, Riedl, & Vesterlund, 2015).
4

experiment: they choose to give when the choice is made in advance, while they reverse their
decision, choosing not to give, when giving occurs immediately. Most of these subjects (65%)
exhibit a strict preference for flexibility, i.e., they do not wish to commit to the giving decision
made in advance. This reveals that the source of dynamic inconsistency for these subjects is
not temptation, but can be explained by moral contradictions experienced in the giving process.
At the same time, we find a significant portion of subjects (38%) who exhibit the opposite
type of dynamic inconsistency. These subjects choose not to give in advance, but give when the
gift is immediate. More than half (52%) of these subjects exhibit a strict preference for com-
mitment. This suggests that temptation to give is an important driver of dynamic inconsistency
for some of our subjects, in support of the model by Dreber et al. (2016).
This heterogeneity among givers is of extreme interest. It shows that introducing time in
the process of giving can uncover important dynamics, as well as substantial heterogeneity in
the motivations for giving. It also suggests the value in creating fundraising strategies that
allow donors to reveal their types, and for fundraisers to tailor solicitations to each type.6
In order to establish the first main set of facts resulting from the introduction of time to the
decision to give, we begin this paper by presenting our primary between-subjects experiment
in Sections 2 and 3. Section 4 provides a simple framework that captures our main findings,
including illustrating the dynamic inconsistency. In Sections 5 and 6 we present our within-
subjects experiment, where we test for individual heterogeneity in dynamic inconsistency and
relate dynamic inconsistency to commitment demand. Section 7 discusses how our ideas can
be used to deepen the understanding of the giving process.
2 Experiment 1: Putting Time Into Giving Money
Our Experiment 1 consisted of two sessions spread exactly one week apart, where the oppor-
tunity to donate $5 to charity was presented. We refer to the first and second sessions as the
week 1 and week 2 sessions, respectively.
In all treatments, the week 1 session opens with a scripted slide show about the charity
6See also Andreoni et al. (2016) for an exercise of this type related to labor supply.
5

GiveDirectly (www.GiveDirectly.org). The presentation, which lasts about 15 minutes, dis-
cusses the work the charity does by giving direct cash grants to poor households in Kenya
and other African nations. It also discusses the results of scientific evaluations of the program
(Haushofer and Shapiro, 2016) showing very high returns on investment as well as endorse-
ments from charity rating groups such as GiveWell. Importantly, the presentation also high-
lights that one of the co-founders and current officers of GiveDirectly is Professor Paul Niehaus
of the Department of Economics at the University of California, San Diego, where the study
was conducted. This, we expect, adds confidence to both our claims about the quality and effi-
cacy of the charity and our (true) promises that the donations would indeed go to GiveDirectly.
The presentation ends with an ask to give $5. The experimental treatments vary on when the
final decision will be required, and when financial transactions will occur.
2.1 Experimental Treatments
The baseline treatment is called Give-Now. Here, all steps take place at once, with the ask, the
decision to donate, and gift happening in week 1. In the second treatment, Give-Later, the ask
and the decision to donate occur in week 1, while the gift is delayed to week 2.
In the third treatment, Pledge, the ask takes place in week 1, at which point individuals can
choose to pledge a donation. If they do, they state an intention to give a week later, subject
to confirmation then. The Pledge treatment, therefore, allows subjects who intend to say no
an opportunity to postpone announcing this decision.7The introduction of the delay naturally
raises the question, whom does it benefit? Someone who is certain to give in the future might
wish for a more clear signal of her intentions, while someone who is sure to say no in the
future can use the delay to discount the uncomfortable feeling of saying no. Acknowledging
this heterogeneity, we added a treatment that allows individuals to self-select into the timing
they preferred. In the Pledge-or-Give-Now treatment, individuals could pledge in week 1 to
7The wording of the potential answers to the ask was always either “No,” or “Yes, I’d like to donate $5 X,”
where for Give-Now X was “today,” for Give-Later X was “next week,” and for Pledge is was “next week. Ask
me again next week and I will make my final decision.” To avoid confusion about the level of commitment to the
pledge, the word pledge was not used anywhere in Pledge treatment. However, the initial statement of an intention
to give may have been viewed by subjects as a promise (see Hanfling, 2008, for a philosophical argument), which
is our intention. The instructions are presented in Online Appendix A.
6

give in week 2, or decide to give immediately, in week 1.
The second question raised by pledging is, can the extra time between the pledge and the
final decision can be used productively by the charity? Since fundraisers extoll the benefits of a
carefully designed thank-you letter,8we hypothesized that one way to increase the conversion
of pledges into donations may be to use gratitude. To explore this, in the Pledge and Pledge-
or-Give-Now treatments we sent thank-you notes via email to a randomly chosen subset of
subjects who pledged to give in the first week of the experiment. The e-mail was delivered by
5:00 p.m. on the same day of the session in week 1, seven days prior to having to confirm their
pledges.9
To examine how the thank-you letter may be working, we designed both a “strong” and
a “weak” version of the thank-you note. The weak thank-you note emphasized the impor-
tance of the pledge and thanked individuals for pledging. The strong thank-you note included
two manipulations shown elsewhere to enhance the identifiable victim effect and to strengthen
identity as a donor.10 We compare the effect of the weak versus strong thank-you note in the
Pledge-or-Give-Now treatment. In the Pledge treatment only weak thank-you notes were sent.
2.2 Procedures
A total of 692 students participated in Experiment 1, conducted at the UC San Diego Eco-
nomics Laboratory. There were 180 students in the Give-Now treatment, 179 in the Give-Later
treatment, 118 in the Pledge treatment and 215 in the Pledge-or-Give-Now treatment. We
purposely recruited more subjects in the latter treatment to have enough observations when
8See, for example, “Tips for thanking (and keeping) donors,” in the Chronicle of Philanthropy, December 22,
2015.
9All subjects received an email 24 hours prior to their week 2 session simply reminding them to attend.
10Specifically, in the weak thank-you note subjects were thanked for their participation and their decision to
pledge. They were told that their contribution would make an important difference in the life of the recipient
family. The note closed by stating that we looked forward to seeing them in a week when they could confirm their
pledge. The strong thank-you note had the same opening sentence. Instead of telling subjects about the general
importance of their donation, the text emphasized that the donation would go to a family in Kenya “like this one,”
and a picture of a family was shown. This reflects the importance of the identifiable victim, as shown by Small &
Loewenstein (2003). In addition, the weak note thanked them for their pledge, while the strong note thanked them
for “being a donor,” which is a framing device known by psychologists such as Bryan, Adams & Monin (2013)
and Walton & Banaji (2004), to increase the appeal to an individual’s identity as a donor and thereby increase
behavior in line with this identity.
7

examining the effect of the thank-you note on giving.
A concern in a longitudinal experiment is attrition. For this reason, the first set of sessions
had a higher show-up fee in week 2 than in week 1 ($6 in week 1 and $20 in week 2). The
attrition rate was low, 8%, and did not vary with the treatment, the decision subjects made in
week 1, or their individual characteristics. A detailed analysis of attrition is shown in Online
Appendix B.
A drawback of the differential show-up fees is that they could affect giving decisions and
contaminate our treatment effects, especially in Give-Now and Give-Later. Thus, we self-
replicated Give-Now and Give-Later treatments with equal show up fees ($15 each week). We
found no significant differences in the sessions with equal show up fees and those weighted
toward week 2.11 Given this, in what follows we pool the data, excluding those who did not
participate in both sessions.12
3 Results For Experiment 1
Our main objective for Experiment 1 is to examine the treatment effects on ultimate giving.
Subordinate to this, we also hope to see whether the treatments can reveal information about
preferences in week 1 that can be useful to the charity in the time before week 2 decisions.
3.1 Week 1 Decisions
Figure 1 presents our results for week 1. The first two bars show that introducing a delay
between the decision and the gift increases giving. In the Give-Now treatment, when decid-
ing and giving occur together and immediately, 30.9% of the subjects choose to donate. In
the Give-Later treatment, where only the payment of the gift is delayed, the giving rate rises
11First, attrition was 10.5% in the second set of sessions and not significantly different by show-up fee (χ2=
0.242,p= 0.623). Second, donation rates were 32.5% and 29.4% in Give-Now (χ2= 0.184,p= 0.668), and
43.8% and 46.5% in Give-Later (χ2= 0.114,p= 0.736), in the first and second set of sessions, respectively.
Third, individuals who did not participate in the week 2 session of the experiment did not behave differently in
week 1 than individuals who did. Their donation rates in week 1 are not significantly different (χ2= 0.369,
p= 0.544). Detailed results for all participants are shown in Online Appendix B.
12Online Appendix B also provides detailed results including all participants, and shows the conclusions from
the analysis remain unaltered.
8

significantly to 45.3% of the subjects (χ2= 7.104,p < 0.01). This nearly 50% increase in
the rate of giving resulted from a delay of just one week in payment. The data reveal a sig-
nificant violation of time stationarity in giving choices made by subjects across the Give-Now
and Give-Later treatments. Moreover, if this pattern remains in the within-subjects setting of
Experiment 2, it will point to dynamically inconsistent choices: when choosing consumption
for next week, the decision is different than the one that would have been made in a week for
consumption immediately.
It is important to note the subtle but real difference here with other consumption choices.
Receiving an object in the present but paying for it later is obviously better. But in our exper-
iment, no actual goods or dollars were exchanged for another week–only the decision is made
today.
Note: Error bars denote ±1 S.E.
Figure 1: Giving and Pledging in Week 1
9

The final two bars of Figure 1 show the results of the pledging treatments. In the Pledge
treatment, 65.5% of the subjects pledge to donate. In the Pledge-or-Give-Now treatment, the
total percentage of subjects who pledge or give immediately is 69.5%. These frequencies are
not significantly different from each other (χ2= 0.543,p= 0.461), but are significantly higher
than the frequency of giving in Give-Now (χ2= 31.860,p < 0.01;χ2= 53.690,p < 0.01,
respectively), and Give-Later (χ2= 10.63,p < 0.001;χ2= 21.36,p < 0.001, respectively).
While not significantly higher than the Pledge treatment, the Pledge-or-Give-Now shows the
greatest participation, with 48.2% of the subjects pledging to donate and 21.3% of the subjects
giving immediately.
3.2 Week 2 Decisions
Individuals who pledged in week 1 were asked in week 2 to confirm their donations or to
renege. Figure 2 graphs the frequency with which individuals renege on their pledges.13
Consider first the case with no thank-you notes. Here, 46.9% of the individuals renege in
the Pledge treatment. This fraction increases to 70.8% in the Pledge-or-Give-Now treatment
(χ2= 3.214,p= 0.073), confirming that the option to give immediately in the Pledge-or-
Give-Now treatment induces selection. In particular, those least likely to renege appear to have
selected giving in week 1, thus are not among those pledging to give, leading to a higher rate
of reneging.
What happens when a thank-you note follows a pledge? In the Pledge treatment, individuals
receiving a weak version of the thank-you note renege in 42.5% of the cases, compared to
46.9% when they do not receive a thank-you note. This drop in reneging is, however, small
and not significant (χ2= 0.138,p= 0.710). In the Pledge-or-Give-Now treatment, pooling
all conditions with thank-you notes and comparing rates of reneging with and without thank-
you notes, we find that the thank-you note reduces reneging by a very large and significant
22.9 percentage points (χ2= 3.798,p= 0.051). When comparing the strong and the weak
13In the Pledge treatment 72 subjects pledged to give in week 2. Among pledgers, approximately half (55.6%)
received the weak version of the thank-you note. In the Pledge-or-Give-Now treatment 95 subjects pledged to
give in week 2. Among them, 27.4% received the weak version of the thank-you note and 47.4% received the
strong version.
10

Note: Error bars denote ±1 S.E.
Figure 2: Reneging in week 2
versions of the thank-you notes, however, we find the strong thank-you performed slightly
better on average, with reneging about 5 percentage points below that for the subjects getting
the weak thank-you, but the difference in not significant (χ2= 0.734,p= 0.786). Thus,
while the identifiable victim and the identity manipulations may be effective on their own, it
would appear that their effects are largely overwritten when presented alongside a message of
gratitude.
The stronger effect of the thank-you note in Pledge-or-Give-Now further confirms the role
of self-selection in this treatment. It suggests that many of the individuals who pledged appear
to have done so with the intention of reneging, and for them the thank-you note resulted in a
significant reduction in reneging. This effect is especially striking in light of the fact that the
thank-you note came within a few hours of their pledges and a full seven days before subjects
returned to confirm them or renege.
11

Table 1: Ultimate giving by treatment
Treatment Donation rate Std. Error
Give-Now 0.309 0.036
Give-Later 0.453 0.039
Pledge 0.364 0.046
Pledge+Without thank-you 0.348 0.068
Pledge+Thank-you:Weak 0.376 0.062
Pledge-or-Give-Now 0.437 0.035
Pledge-or-Give-Now+Without thank-you 0.354 0.068
Pledge-or-Give-Now+Thank-you: Weak 0.454 0.068
Pledge-or-Give-Now+Thank-you: Strong 0.470 0.052
3.3 Ultimate giving
Table 1 presents the rate of giving by treatment, and Table 2 presents the results of the re-
gression analysis of the treatment effects.14 Column (1) of Table 2 shows the results of a probit
regression, ignoring the presence of thank-you notes. To examine the effect of thank-you notes,
we conduct a placebo test by first assigning those who did not pledge to a thank-you condition
with a probability equal to that of their counterparts who did pledge. We then examine the
effect of the thank-you conditions using a weighted probit regression. This analysis is shown
in column (2) of Table 2.15
The regressions in Table 2 reconfirm that separating the timing of giving from the decision
to give (Give-Later) significantly raises the donation rate, leading to a 14 percentage point
increase (p≤0.001), when a one week delay in the gift is introduced. This predicts an increase
in donations of almost 50% relative to Give-Now.
Table 2 also shows that pledging alone has a small effect on giving, of less than 5 percentage
points, that is not statistically significant. For pledging to show a measurable positive and
significant effect on giving in our experiment, it must be accompanied by both the option to
give now, in week 1, and a thank-you note of either kind. When these conditions are both met,
14Our analysis of the treatment effects in Table 2 reports p-values that are uncorrected for multiple hypothesis
testing (e.g., List et al., 2016). However, since all p-values for significant differences are below 0.001, correcting
p-values leaves our conclusions unchanged.
15An alternative approach is to randomly assign a share of the individuals who did not pledge to each thank-you
condition, and use bootstrapping. Results remain qualitatively similar with this approach.
12

Table 2: Treatment effects on ultimate giving
(1) (2)
Donation
Give-Later 0.145*** 0.144***
(0.041) (0.041)
Pledge 0.057
(0.049)
Pledge+Without thank-you 0.042
(0.056)
Pledge+Thank-you: weak 0.069
(0.066)
Pledge-or-Give-Now 0.129***
(0.028)
Pledge-or-Give-Now+Without thank-you 0.050
(0.056)
Pledge-or-Give-Now+Thank-you: weak 0.147***
(0.036)
Pledge-or-Give-Now+Thank-you: strong 0.161***
(0.033)
Observations 631 631
Pseudo R-squared 0.011 0.0133
Note: This table presents the average marginal effects (calculated at the means of all variables)
from probit regressions on ultimate giving decisions. Column (1) presents the marginal effect from
simple probit regressions on the treatment. Column (2) presents results from weighted probit regres-
sions, whereby individuals who did not pledge in Pledge and Pledge-or-Give-Now are assigned to
both the no thank-you and the thank-you conditions, and weighted correspondingly. The variables
Give-Later, Pledge, Pledge+Without thank-you, Pledge+Thank-you, Pledge-or-Give-Now, Pledge-
or-Give-Now+Without thank-you, Pledge-or-Give-Now+Thank-you are dummies that take value
one in the corresponding treatment or treatment+thank you condition, zero otherwise. Robust stan-
dard errors, clustered at the session level, were used in each individual regression. ***,**,* indi-
cates significance at the 1%, 5%, and 10% levels, respectively.
giving rises by 14.7 (weak) to 16.1 (strong) percentage points, a statistically and economically
significant increase.
4 A Simple Framework for Analysis
Experiment 1 yields six main findings: (1) Individuals are more likely to give when the gift is
paid later, as in Give-Later versus Give-Now, which suggests dynamic inconsistency in giving;
(2) Far more intentions to give are expressed in the treatments that allow pledges as compared
to those that require immediate giving, as in Pledge versus Give-Later; (3) Reneging in the
Pledge treatment is frequent, with final giving below Give-Later but above Give-Now; (4) The
13

fraction of subjects who either pledge or give now in Pledge-or-Give-Now equals the fraction
of those who pledge in the Pledge treatment; (5) For subjects in the two pledging treatments, a
larger fraction of those pledging in Pledge-or-Give-Now end up reneging in week 2 than those
pledging in Pledge; (6) Thank-you notes sent immediately after a pledge significantly increase
the likelihood that a pledge is converted to a donation in Pledge-or-Give-Now.
These six findings cannot be explained by models of giving that assume that the utility of
giving all flows from a single event (such as Andreoni 1989, 1990). In this section we write
down a simple and parsimonious model of the giving process that allows us to identity several
core components of the full process of giving. The model, though simple, neatly accounts for
all six findings from the experiment just presented. Having said that, we also hasten to add
that we view the model as incomplete, leaving flexibility for future research to flesh out the
foundations of the components we put forth as well as to discover new ones.
4.1 Give Now
Consider a situation in which someone has been asked to make a charitable gift of a particular
size, to which she can say yes or no. Since giving is measured in dollars, normalize utility
across individuals so that the utility cost of paying for the gift is the same across people. Call
this utility value g.16 We assume the act of giving to charity also brings some (perhaps altruistic)
utility, which we will call α.17 This is the most typical formulation in the literature.
Now we generalize this model slightly to include the social utility surrounding being asked
to give. It is well established that people dislike saying no to reasonable requests to give,18
even while they may appreciate being asked.19 Assuming the charity is seen by all as providing
some benefits to society, we would expect saying yes to the ask brings social benefits, while
saying no brings social costs. Notice, however, that saying yes and no are mutually exclusive.
16This is equivalent to normalizing utility by the marginal utility of money, as is commonly done in the Public
Economics literature.
17We do not rule out that α≤0for some, as in the case of a donor being unsure about the legitimacy of the
solicitation, although in our study we generally regard α≥0.
18See Dana, Cain, and Dawes (2006), Dana, Weber, and Kuang (2007), DellaVigna, List, and Malmendier
(2012), Andreoni and Rao (2011), and Andreoni, Rao, and Trachtman (2016).
19See Crumpler and Grossman (2008), Andreoni and Rao (2011), and Chowdhury and Jeon (2014).
14

As such, the decision between the two will depend on the difference in their values rather than
their absolute amounts. To make this point clear, suppose we separated the utility from saying
yes, call it sy>0, from the disutility of saying no, −sn<0. Then a person will say yes if
Zy+sy> Zn−sn, where the Z’s represent other utility consequences of the decision. Since
“yes” and “no” are mutually exclusive, normalize the utility of yes to be 0 and the utility of no
then to be the net values of their utilities, −n=−sn−sy<0. Then −nwill be sufficient for
modeling both “yes” and “no” choices.
We begin with the baseline case in which the decision to give and the act of giving are made
simultaneously and both occur in the present. Since the utility cost for paying for the donation
is fixed at −g, saying yes yields a net utility of α−g. Saying no yields −n < 0. Combining
these, an individual says yes to the request to give if
α−g≥ −n.
Rearranging this we can define a critical level of generosity αN(n)≡g−n, such that an
individual will give now if α≥αN(n). Note that, if asked, an individual may decide to give
even if giving yields a disutility, α−g < 0, as long as this disutility is smaller than that of
saying no. This point will be important later.
4.2 Give Later
Now we take the model above and put time between the two moments of decision and payment.
As above, the person is asked at time t= 1 (period 1) to commit to giving, but the payment
for the gift comes in the future, t= 2 (period 2). With δ≤1as the discount factor, then an
individual says yes in t= 1 if the discounted utility of giving later exceeds the utility of saying
no:
δ(α−g)≥ −n.
15

Rearranging again yields the critical value αL(n)≡g−n/δ such that those with α≥αL(n)
will give later. Since δ < 1, then αL(n)< αN(n)for all n. That is, anyone who would give
now will also give later, but many who would say no to giving now will instead be willing to
say yes to giving later.
4.3 Dynamically-Inconsistent Giving
An interesting implication of our first two predictions is that they raise the possibility of
dynamically-inconsistent giving (result 1). Begin with those who say yes when asked in period
1 to give in period 2. Suppose in period 2 we set aside their earlier choice and ask them if they
are willing to give now. Assuming their tastes for giving now are the same in period 2 as in
period 1, then our data suggests that a significant sum of those saying yes in Give-Later would
change their preferences when asked again in period 2 if they would like to give now.
What kind of preferences would display such inconsistency? They would need to satisfy
αN(n)> α > αL(n).
Substituting in the definitions of αNand αLand rearranging, we see
−n>α−g > −n/δ. (1)
Equation (1) reveals that dynamically-inconsistent types are those who are made worse off by
having been asked, that is α−g < 0. As a result, their discounted utility, δ(α−g)is higher than
their undiscounted utility. Stated differently, while allowing people to give later may encourage
more giving, it encourages it among those whose utility from giving is already negative.
Dynamic inconsistency arises because individuals face a moral contradiction in choosing
whether to give or not: they do not desire to give, but also dislike saying no to the fundraiser.
If these types are sophisticated, they will not demand commitment to their advance choices.
The reason is that they anticipate their desire to say no to giving when asked again in period
2 and, hence, are better off by not demanding commitment. Thus, dynamic inconsistency is
16

directly linked to demand for flexibility in giving decisions, in contrast to the standard link
between dynamic inconsistency and demand for commitment generated by quasi-hyperbolic
preferences (such as in Laibson, 1997; O’Donoghue and Rabin, 1999) and temptation models
(Fudenberg and Levine, 2006; Gul and Pesendorfer, 2007). We will examine this difference in
detail in our second experiment.
4.4 Pledging
Next we put time between the ask and the final decision to give by allowing donors to submit
non-binding pledges. Pledges are common in fundraising campaigns in the United States but
have received little study by economists.20 When giving occurs through pledges the decision
to give can be split into two decisions separated by time. First is the decision to pledge or say
no to the ask. Second, for those who pledge, there is the later decision to honor the pledge or
renege.
Since pledging allows a non-giver to postpone saying no today, why should we not expect
all non-givers to simply pledge and renege? Research shows that breaking pledges may entail
costs akin to those from lying (see Ellingsen and Johannesson, 2004; Gneezy, 2005; Charness
and Dufwenberg, 2006, and Serra-Garcia et al., 2013). Therefore, having pledged, breaking
the pledge could have a utility loss of `≥0, in addition to the cost of saying no. Then in period
2 the full cost of reneging is n+`.
An individual who pledged will confirm the pledge if the utility from confirming is greater
than from reneging: α−g≥ −n−`. This leads to a critical αC(n+`)≡g−(n+`), where
those with α≥αC(n+`)confirm the pledges.
When deciding to pledge, therefore, the potential donor compares three options. First is to
pledge with the intention of giving. A person will choose this if α−g > −n−`. Second is to
20Notable exceptions include Breman (2011), who finds an increase in giving from pledging. She revisits those
who agreed to increase their monthly automatic payment to the charity a year after the pledge and found that
virtually no one had reversed or reduced the increased monthly payment. Other studies of pledges are often more
like our Give-Later treatment in that the are binding commitments to give, such as Zellermayer (1996), finds the
result is sensitive to whether the pledge is framed as coming from currently held money (a negative effect), or
from money earned in the future (positive effect). The psychologists Meyvis, Bennett and Oppenheimer (2010)
find that those more capable abstract thinking are also more willing to give in the future.
17

pledge with the intention of reneging, implying α−g < −n−`and −n < −δ(n+`). Third
is simply to say no immediately. This happens if α−g < −n−`and −n > −δ(n+`).
We can easily see from this analysis that the utility of pledging lies between the utility of
Give-Now and Give-Later treatments, depending on the value of `. If `= 0 then there is no
penalty for pledging, and any subject will pledge. Those for whom α−g > −nwill do so with
the intention of giving, while the rest will have the intention of reneging. Thus, the situation
reverts to that in Give-Now, the difference being that it is now played in period 2 rather than
period 1. Suppose instead that reneging on a pledge is extremely costly, then any subject who
pledges also gives. Specifically, if ` > n(1 −δ)/δ, the environment mirrors that of Give-Later
in which pledges are binding.
This now allows us two conclusions. First, the number of pledges in the Pledge treatment
will be at least as high as the number of donors in Give-Later, and the lower the costs of lying, `,
the greater the number of insincere pledges. Second, the number of final donors in the Pledge
treatment will be bounded below by the Give-Now treatment, and bounded above by Give-
Later. The lower the costs `, the closer final donations come to match the level in Give-Now.
These predictions for pledging are consistent with our experimental data, results 2, and 3.
4.5 Pledge or Give Now
Recall, a person who donates in Give-Now satisfies α−g > −n. Notice this inequality can
be satisfied when both sides are negative as well as when the left side is positive. If we ask a
donor in the Give-Now treatment to instead give later, her utility would become δ(α−g). Thus
only if she has α−g < 0would she agree; if α−g > 0she would prefer to give now.
Suppose that we offer subjects this choice in period 1: pledge to give next period or give
now. The option to give now cannot make anyone worse off, and can only make them better off
if they choose to give now. Thus, offering the choice to pledge or give now will not increase
the total number of people willing to express an intention to give (by pledging or giving now)
in period 1 above what we saw in the Pledge treatment, nor will it affect the ultimate number of
donations collected. All it will affect is timing for those with high values for α, removing them
from the pool of those pledging, and in doing so, will increase the average likelihood that a
18

pledge will be reneged. Again, all of these predictions are in line with our experimental results
4 and 5.
4.6 Gratitude
Seeing how significantly pledges reduce the number of people saying no immediately, it is
natural to ask if the time between the pledge and the final giving decision can be used con-
structively to increase final giving. This leads to a tactic often used by fundraisers: showing
gratitude.
Within our model, a thank-you note between the time of the pledge and the time for the
donation could have two effects. First, the thank-you note could increase α. This would happen
if gratitude increases the pledger’s support or attachment to the organization. A second, but not
mutually exclusive, effect of gratitude could be to increase nor `. After being thanked a person
may feel greater disutility from reneging, such as additional guilt from telling a lie. Either or
both of these effects could lead to less reneging and are consistent with our sixth finding of
our experiment: The thank-you note significantly increases the conversion pledges into actual
donations.21
5 Experiment 2: Dynamic Inconsistency and Commitment
Perhaps the most interesting result of our study so far is the prediction of dynamic inconsistency
in donations. In particular, more people agree now to give in the future than agree now to give
now. If this effect were shown within subjects we would call this dynamically inconsistent
behavior.
21The effect a thank-you note always differs across the Pledge and Pledge-or-Give-Now treatments due to
selection effects. It is larger in the Pledge-or-Give-Now treatment because thank-you notes are sent to subjects
with a significantly higher likelihood of reneging. Hence, a significant effect of thank-you notes will be detected
more often in the Pledge-or-Give-Now treatment, than in the Pledge treatment, as is the case in our experiment.
To illustrate this, suppose all subjects who gave in week 1 in the Pledge-or-Give-Now would have been forced to
pledge and confirm in week 2 (like in the Pledge treatment). Then, the frequency of reneging among subjects who
pledge and do not receive a thank-you note would have been 47.2%, while the frequency of reneging among those
who receive a thank-you note would have been 32.4%. This difference in reneging is not statistically significant
(χ2-test, p-value=0.11), and much smaller in magnitude than that observed in the Pledge-or-Give-Now treatment.
19

Notice that the dynamic inconsistency generated in our theoretical framework does not rely
on a non-exponential discounting model (such as in Laibson, 1997; O’Donoghue and Rabin,
1999), but rather stems from altering the timing of the social and economic payoffs of giving.
This source of dynamic inconsistency is systematically different from inconsistencies resulting
from temptation or present bias. A potential donor might be aware that his giving decision in
advance is generated by picking the least of two bads, and be well aware that when giving is no
longer delayed his decision would change. He would thus value flexibility over commitment.22
Dynamic inconsistency could also be of a different kind if in period 1 a subject chooses
not to give in period 2, but when period 2 arrives she prefers to give. This version of dynamic
inconsistency has been suggested by Dreber et al. (2016), and is based on the assumption that
giving is tempting. In their paper subjects play dictator games with other subjects. The dictator
makes decisions today about payments at different points in the future. They observe giving is
highest when the gift is made in the present. This finding is consistent with giving being tempt-
ing.23,24 While our between-subjects design does not point to a predominance of this effect, it
also does not allow us to identify if a minority of subjects show dynamic inconsistency from
temptation to give. Furthermore, our simple model does not consider this type of preference.
Crafting the model to include temptation is trivial, but requires one extra degree of freedom.
22In a recent study Andreoni et al. (2016) find that people switch their decisions about how to allocate goods
fairly between two people depending on the context of that decision. When the decision is ex ante to a partial
realization of uncertainty, subjects favor ex ante notions of fairness. After the partial realization of uncertainty,
subjects prefer to adopt an ex post stance on fairness, despite the fact that from the ex ante perspective this change
is very unfair. Moreover, subjects rejected opportunities to commit to a fairness perspective, but rather preferred
the flexibility for their fairness criterion to fit the context.
23This hypothesis relies on results found by Rand and coauthors (2012, 2014a, 2014b), who showed in various
games that “fast decisions” also tended to be more altruistic. Recent work by Recalde, Riedl, and Vesterlund
(2015), however, shows that errors are also correlated with time, and that earlier decisions are also more prone to
errors, which undermines the inference that faster decisions are more representative of true preferences. Kessler,
Kivimaki, and Niederle (2016), also use reaction times to infer base motives, and have mixed conclusions. That
giving declines with time was shown by Meyvis, Bennett and Oppenheimer (2010) but attributed to a different
source. They hypothesized that the present is more concrete while the future is more abstract. When the payment
was framed as coming from currently held assets, pledges resulted in reduced giving, while opposite effect was
found when the payment was framed as taken from future income. They found increased giving was associated
with psychological measures of abstract reasoning.
24Note that an alternative model of temptation could be to assume that it is tempting to be selfish and, hence,
that it is the long-run self who wishes to be altruistic. However, such a model would predict individuals do not
avoid the ask, contrary to what has been documented in DellaVigna, List and Malmendier (2015) and Andreoni,
Rao and Trachtman (2016). Such a model would also predict commitment demand, a finding for which we do not
find support in Experiment 2.
20

For example, we could add a variable βt, where βt>0if the gift is transacted at the same
time the decision is made, and βt= 0 otherwise. With sufficient heterogeneity in the size of
β—low temptation for some and high temptation for others (as was observed by Dreber et al.,
2016)—then this approach would allow for the existence of two opposite types of dynamic
inconsistency. How would a subject with such temptation preferences feel about commitment?
Since temptation rather than social pressure is the motivation, the decision maker in these mod-
els has only her future self to answer to. Since this future self wishes the present self did not
yield to temptation, commitment to the advance choice not to give is desirable.
In sum, depending on whether the donor feels she suffers from a bias that needs to be
controlled or whether she is responding to changing social and economic incentives, different
degrees of commitment demand for the two kinds of dynamically-inconsistent giving may
arise. Next we describe our second experiment designed to answer all three of the questions
raised: Is there dynamic inconsistency within subjects? If so, is the pattern (Donate, Not
Donate) as the model of Section 4, is it (Not Donate, Donate) as suggested by a temptation
to give, or do we see two different types of dynamic inconsistency? Finally, when there is
dynamic inconsistency, is there also a demand for commitment, and does it differ for the two
types?
5.1 Experimental Design
We designed a longitudinal experiment following the same structure and features as the Give-
Now and Give-Later treatments. In week 1 subjects made a choice like that in Give-Later: a
decision today about a gift paid in week 2. In week 2 they returned to the lab to make a second
decision like that in Give-Now: a decision now about a gift paid now. In week 2, after both
choices are made, the experimenter randomly selected either the Give-Later (period 1 decision)
or the Give-Now (period 2 decision) to carry out.
We added one more step to the above in order to measure commitment demand. Knowing
that either their week 1 or week 2 choice would be implemented at random, each subject was
offered a probabilistic commitment device in week 1 (as in Augenblick, Niederle and Sprenger,
2015). In particular, after making her period 1 decision, the subject was offered the choice of
21

which week’s decision would be more likely to be implemented. Let pbe the probability week
1’s choice is implemented. Then the subject chooses pfrom the set p∈ {0.9,0.1}. If the
subject prefers week 1’s outcome, she chooses the high pof 0.9, while if she prefers week 2’s
decision (which she has yet to make) she would choose the low pof 0.1. In order differentiate
strict preferences for the earlier or later decision from a simple demand for randomization, we
also allowed the subject to indicate indifference between high and low probabilities. If they
did, their pwas selected as either high or low with the flip of a coin, effectively creating a
p= 0.5.
There were 183 participants in this experiment. Subjects were paid a show-up fee of $15 in
both week 1 and week 2 of the experiment. Twenty students (11%) did not participate in week
2. As before, attrition was unrelated to decisions made in week 1 (χ2= 0.750,p= 0.386).
6 Results from Experiment 2
The share of subjects who decide to give in period 1 is 47.9%, a share that is similar to the
45.3% we observed in Give-Later (χ2-test, p= 0.644). The share of individuals who decide
to give in period 2 is 39.3%, a portion that is higher than the 30.9% in Give-Now, but not
significantly different (χ2-test, p= 0.113). This reveals dynamically inconsistent behavior
is also present within subjects, though to a somewhat smaller extent than would have been
predicted by the between subjects data, perhaps owing to concerns about appearing consistent
(Falk and Zimmerman, 2016).
At the individual level, we find that 36.8% of the individuals are dynamically inconsistent
by making different donation decisions in week 1 and week 2. Out of these, 62% choose
(Donate, Not Donate) and 38% choose (Not Donate, Donate). The former type of dynamic
inconsistency is significantly more frequent than the later (McNemar’s test, p= 0.07). Thus,
while both types exist, those apparently driven by social costs of saying no outnumber those
who are hypothesized to face temptation to give in the present.
What about commitment demand? On aggregate, 35.4% of the individuals choose not to
commit, 27.0% are indifferent, and 38.9% exhibit a strict preference for commitment. Figure
22

3 illustrates the choice of pin week 1 for each of the four possible dynamic paths. Here we see
a clear difference between the two categories of dynamic inconsistency. Subject who choose
(Donate, Not Donate) overwhelmingly prefer not to commit: 64.9% strictly favor flexibility
while only 18.9% choose to commit. This shows that there exists a clear preference to maintain
flexibility for giving to change in period 2. In contrast, among those who choose (Not Donate,
Donate) we see far more people selecting commitment: 52.2% select commitment while 34.8%
strictly prefer not to commit. This aligns with the prediction that commitment demand is used
to overcome temptation to give.25
Note: The error bars denote ±1 S.E.
Figure 3: Dynamic inconsistency and commitment demand
Looking at the two groups that were time consistent, we have no prediction for commitment
for either group. Indeed, for (Donate, Donate) the commitment choices appear as if they were
made at random, while (Not Donate, Not Donate) appears to favor commitment.26
25See Online Appendix B for detailed data that supports Figure 3.
26Dreber et al. (2016) predict that strict, rather than probabilistic, commitment can lead to commitment demand
even by those who would otherwise resist donations in both periods. The reason is that the commitment saves the
person the cognitive resources spent on exerting self-control. While in our experiment a decision is required in
23

Table 3 provides support for these results using a multinomial probit regression that relates
commitment choice with the type of dynamic preferences expressed. As can be seen in column
(1), individuals who choose to (Donate, Not Donate) are significantly more likely to choose no
commitment than subjects in other treatments. Column (2), by contrast, shows that that those
who choose (Not Donate, Donate) are marginally significantly more likely to commit than to
express indifference. For dynamically consistent subjects, column (3) confirms that individuals
who consistently donate over time do not exhibit a preference for flexibility or commitment,
while column (4) reveals that individuals who consistently say no exhibit a marginally lower
preference for flexibility, though they do not significantly prefer commitment, relative to indif-
ference.
Table 3: Dynamic inconsistency and commitment demand
(1) (2) (3) (4)
(Week 1 donation decision, Week 2 donation decision)
Dynamic inconsistency Dynamic consistency
(Donate, not donate) (Not donate, donate) (Donate, donate) (Not donate, not donate)
Flexibility: 0.259*** 0.108 -0.110 -0.257*
p= 0.1for week 1 decision (0.098) (0.103) (0.090) (0.138)
Commitment: -0.040 0.141* -0.117 0.016
p= 0.9for week 1 decision (0.117) (0.080) (0.097) (0.154)
Observations 163
Note: This table presents the marginal effects (calculated at the means of all variables) from a multinomial probit regression
relating patterns of dynamic choice to commitment choice. Robust standard errors, clustered at the session level, are shown
in parentheses. *** p<0.01, ** p<0.05, * p<0.1
The central finding of our second experiment is that there is heterogeneity surrounding
dynamic inconsistency. Most dynamically inconsistent individuals are well described by the
simple model we proposed, in which individuals agree to give later but not in the present. Since
they may want the decisions they make to match the social payoffs at the time of the decision,
they do not see these inconsistencies as problematic and thus do not choose commitment. A
significant minority, however, also appear to adopt an opposite stance: They choose not to give
both periods regardless of commitment, one can imagine an interpretation their model that would cause the same
effects here. If this is the case, then the fraction of subjects who satisfy their model would be under-estimated,
and should include some time-consistent non-givers who nonetheless demand commitment.
24

in the future but cannot resist giving in the present. Those who are sophisticated about their
self-control problem opt to commit to their period 1 choice to not give in period 2.
7 Discussion and Conclusion
Charitable giving can often be a complex social interaction with potentially many opportunities
for social as well as economic payoffs to accrue. Seemingly minor changes in the timing or
emphasis of the different components of the giving process can have significant effects on
ultimate giving. A primary contribution of this paper is to demonstrate this fact about giving
and point to the implications this has for dynamic behavior in the charitable domain. Most
importantly, we document a new kind of dynamic inconsistency that is linked with demand for
flexibility, in contrast to traditional sources of dynamic inconsistency which are linked with
demand for commitment.
Understanding the socially complex process of charitable giving is necessary to answer
many important questions about institutional design, the effect of tax policy on giving, and
the welfare effects of both private and public efforts to affect giving. Our work illustrates
the existence of flows of utility from different aspects of the decision, and hence suggests
important ways in which these aspects can be manipulated to engineer outcomes. This finding
should be of interest to fundraisers, economic theorists, and policy makers, as it suggests a
new architecture for building models of donors, fundraising, and of tax policy toward giving.
Consider the four examples below.
First, technology has made it easier to focus on “micro-donations” such as asking for the
$1 gift at the checkout counter, or with purchases on Amazon.com. Dozens of companies have
emerged in an attempts to disrupt the timing of saying yes and feeling the pain of paying,
such as Google’s OneToday app. A second example is the charitable sector’s reaction to the
concentration of wealth around the world. “High net worth donors” are often engaged in long
term planning for legacy gifts, and are becoming the primary focus of fundraisers. These givers
are deeply concerned with timing of the announcements of commitments to give, timing of the
actual payments, longevity of naming rights, and other social and temporal aspects of the gift.
25

A third example is the rise of donor advised funds (DAFs) as a vehicle for de-linking the
tax benefits of giving from the actual time of making the gift. DAFs act like reverse-401k funds
for donations—the tax benefits of giving are realized when the money is set aside for charity,
while the donations can be made at any future date, almost without restriction. Thus, taxpayers
are loaning the tax benefit to the donor, at zero interest, until the money is, at some uncertain
point, productively used by a charity. These funds present unique challenges for fundraisers
and policy makers (see Andreoni, 2016).
Fourth, for the extremely wealthy, there is the “giving pledge.” This pledge was started
by Bill Gates and Warren Buffett in 2010, to encourage the world’s wealthiest people “to give
the majority of their net worth to philanthropy, either during their lifetime or upon their death”
(givingpledge.org). This pledge challenges the traditional notion that the wealthy feel less
rich by giving their money away, and raises the question of whether there is an asset value in
past giving especially when such giving can be truly transformative. Stated differently, do the
wealthy feel less rich by giving money away or are these gifts capitalized into their psychic
notions of wealth?
In sum, more complete and nuanced models of giving, such as those pointed to here, will
be needed to address the ever more complex world of charitable giving.
26

References
[1] Andreoni, J (1989). Giving with Impure Altruism: Applications to Charity and Ricardian
Equivalence. Journal of Political Economy, v. 97, December, 1447-58.
[2] Andreoni, J. (1990). Impure Altruism and Donations to Public Goods: A Theory of
Warm-Glow Giving. Economic Journal, v. 100, June, 464-477.
[3] Andreoni, J. (2006). Philanthropy. Handbook of Giving, Reciprocity and Altruism, L.-A.
Gerard-Varet, Serge-Christophe Kolm and Jean Mercier Ythier, Editors, Elsivier/North-
Holland.
[4] Andreoni, J. (2016). Who Benefits from Donor Advised Funds. Unpublished working
paper, 2016.
[5] Andreoni, J., Aydin, D., Barton, B., Bernheim, B. D., & Naecker, J. (2016). When Fair
Isn’t Fair: Sophisticated Time Inconsistency in Social Preferences. Available at SSRN
2763318.
[6] Andreoni, J., & Bernheim, D. B. (2009). Social image and the 50-50 norm: A theoretical
and experimental analysis of audience effects. Econometrica 77, no. 5: 1607-1636.
[7] Andreoni, J., Callen, M., Jaffar, K.H., Khan, M.Y., & Sprenger, C.D. (2016). Using Pref-
erence Estimates to Customize Incentives: An Application to Polio Vaccination Drives
in Pakistan. Working paper.
[8] Andreoni, J., & Payne, A.A. (2013) Charitable Giving. Handbook of Public Economics,
Volume 5, (Amsterdam: North Holland), Alan Auerbach, Raj Chetty, Martin Feldstein,
Emmanuel Saez, editors, 1–50.
[9] Andreoni, J., & Rao, J.M. (2011). The power of asking: How communication affects
selfishness, empathy, and altruism. Journal of Public Economics 95, no. 7: 513-520.
[10] Andreoni, J., Rao, J.M., & Trachtman, H. (2016). Avoiding the ask: a field experiment
on altruism, empathy, and charitable giving. Journal of Political Economy, forthcoming.
27

[11] Augenblick, N., Niederle, M., & Sprenger, C. (2015). Working over time: Dynamic
inconsistency in real effort tasks. Quarterly Journal of Economics 130 (3), 1067–1115.
[12] B´
enabou, R., & Tirole, J. (2006). Incentives and prosocial behavior. American Economic
Review 96, no. 5: 1652-1678.
[13] Breman, A. (2011). Give more tomorrow: Two field experiments on altruism and in-
tertemporal choice. Journal of Public Economics, 95(11), 1349–1357.
[14] Bryan, C.J., Adams, G. S. & Monin, B. (2013). When cheating would make you a
cheater: Implicating the self prevents unethical behavior. Journal of Experimental Psy-
chology: General 142 (4), 1001–1005.
[15] Charness, G., & Dufwenberg, M. (2006). Promises and partnership. Econometrica,
74(6), 1579–1601.
[16] Chowdhury, S. M., & Jeon, J-Y. (2014). Impure altruism or inequality aversion? An
experimental investigation based on income effects. Journal of Public Economics 118,
143–150.
[17] Crumpler, H., & Grossman, P.J. (2008). An experimental test of warm glow giving.
Journal of Public Economics, 92 (5), 1011–1021.
[18] Dana, J., Cain, D. M., & Dawes, R. M. (2006). What you don’t know won’t hurt me:
Costly (but quiet) exit in dictator games. Organizational Behavior and Human Decision
Processes, 100(2), 193-201.
[19] Dana, J., Weber, R. & Kuang, J.X. (2007). Exploiting moral wiggle room: experiments
demonstrating an illusory preference for fairness. Economic Theory 33 (1), 67–80.
[20] DellaVigna, S., List, J. A., & Malmendier, U. (2012). Testing for altruism and social
pressure in charitable giving. Quarterly Journal of Economics, 127(1), 1–56.
[21] Dreber, A., Fudenberg, D., Levine, D. K., & Rand, D. G. (2016). Self-Control, Social
Preferences and the Effect of Delayed Payments. Working paper.
28

[22] Ellingsen, T., & Johannesson, M. (2004). Promises, Threats and Fairness. Economic
Journal, 114 (495), 397–420.
[23] Falk. A., & Zimmerman, F. (2016). Information Processing and Commitment. Working
paper.
[24] Frederick, S., Loewenstein, G., & O’Donoghue, T. (2002). Time discounting and time
preference: A critical review. Journal of Economic Literature, 351–401.
[25] Fudenberg, D., & Levine, D. (2006). A dual-self model of self-control. American Eco-
nomic Review 96 (5), 1449–1476.
[26] Gneezy, U. (2005). Deception: The role of consequences. American Economic Review
95 (1), 384–394.
[27] Gul, F. & Pesendorfer, W. (2001). Temptation and self-Control. Econometrica 69 (6),
1403–35.
[28] Halevy, Y. (2015). Time consistency: Stationarity and Time Invariance. Econometrica
83 (1), 335–352.
[29] Hanfling, O. (2008). How we trust one another. Philosophy, 83(02), 161–177.
[30] Harbaugh, W. T. (1998). The prestige motive for making charitable transfers. American
Economic Review 88 (2), 277–282.
[31] Haushofer, J. & Shapiro, J. (2016). The short-term impact of unconditional cash trans-
fers to the poor: Experimental Evidence from Kenya. Quarterly Journal of Economics,
forthcoming.
[32] Kessler, J., Kivimaki, H. & Niederle, M. (2016). Generosity Over Time. Unpublished
manuscript.
[33] Kovarik, J. (2009). Giving it now or later: Altruism and discounting. Economics Letters,
102 (3), 152–154.
29

[34] Laibson, D. (1997). Golden eggs and hyperbolic discounting. The Quarterly Journal of
Economics, 443-477.
[35] List, J.A. (2011). The market for charitable giving. Journal of Economic Perspectives
25 (2), 157–180.
[36] List, J.A., A.M. Shaikh, & Y. Xu (2016). Multiple Hypothesis Testing in Experimental
Economics. NBER Working Paper 21875.
[37] Meyvis, T., A. Bennett & D.M. Oppenheimer (2010). Precommitment to Charity. The
Science of Giving: Experimental Approaches to the Study of Charity. D. M. Oppen-
heimer & C.Y. Olivola (Eds.). New York: Taylor and Francis.
[38] O’Donoghue, T., & Rabin, M. (1999). Doing it now or later. American Economic Review
89 (1), 103–124.
[39] Rand, D. G., Greene, J. D., & Nowak, M. A. (2012). Spontaneous giving and calculated
greed. Nature, 489(7416), 427–430.
[40] Rand, D. G., & Kraft-Todd, G. T. (2014a). Reflection does not undermine self-interested
prosociality. Frontiers in behavioral neuroscience, 8.
[41] Rand, D. G., Peysakhovich, A., Kraft-Todd, G. T., Newman, G. E., Wurzbacher, O.,
Nowak, M. A., & Greene, J. D. (2014b). Social heuristics shape intuitive cooperation.
Nature communications 5, 3677.
[42] Recalde, M. P., Riedl, A., & Vesterlund, L. (2015). Error prone inference from response
time: The case of intuitive generosity in public-good games. University of Pittsburgh
Department of Economics WP, 15-004.
[43] Ribar, D. C., & Wilhelm, M. O. (2002). Altruistic and joy–of–giving motivations in
charitable behavior. Journal of Political Economy, 110 (2), 425–457.
[44] Serra-Garcia, M., van Damme, E., & Potters, J. (2013). Lying about what you know or
about what you do. Journal of the European Economic Association,11(5), 1204–122.
30

[45] Small, D. A., & Loewenstein, G. (2003). Helping a victim or helping the victim: Altru-
ism and identifiability. Journal of Risk and Uncertainty, 26 (1), 5-16.
[46] Trachtman, H., Steinkruger, A., Wood, M., Wooster, A., Andreoni, J., Murphy, J.J., &
Rao, J.M. (2015). Fair Weather Avoidance: Unpacking Costs and Benefits in a Replica-
tion of “Avoiding the Ask”. Journal of the Economic Science Association, forthcoming.
[47] Walton, G.M. & Banaji, M. R. (2004). Being what you say: The effect of essentialist
linguistic labels on preferences. Social Cognition 22 (2), 193–213.
[48] Zellermayer, O. (1996). The pain of paying (Doctoral dissertation, Carnegie Mellon
University).
31