The habit of giving.
Meer, Jonathan
I. INTRODUCTION
Habit formation is thought to exert great influence on behavior. It
has been proffered as a potential answer to questions as disparate as
the size of the equity premium (Abel 1990), optimal purchases of
insurance (Ben-Arab, Briys, and Schlesinger 1996), labor force
participation (Woittiez and Kapteyn 1998), the relationship between
savings and growth rates (Carroll, Overland, and Well 2000),
responsiveness to monetary policy (Fuhrer 2000), the importance of brand
loyalty (Gupta, Chintagunta, and Wittink 1997), and the existence of a
"gateway" effect between alcohol and illegal drug use (Pacula
1998). Yet there is scant discussion in the economics literature about
long-term habit forming. Most studies focus on shorter-term
intertemporal relationships, like changes in annual consumption (see,
inter alia, Naik and Moore 1996; Dynan 2002; Carrasco, Labeaga, and
Lopez-Salido 2005). The importance of early influences on later risk
taking (Malmendier and Nagel 2007) and motivations for purchasing
different goods (Portolese-Dias 2004) has been hypothesized. Yet there
is little direct evidence on the long-term impact of shocks to behavior
early in life, particularly in the way that preferences form and evolve.
One recent exception is the study of Bronnenberg, Dube, and Gentzkow
(2010), who use variation in consumers' previous states of
residence to show that early exposure to particular brands of package
goods affects purchasing behavior decades later.
Charities, in particular, care about building relationships with
their donors and expend a great deal of effort in the pursuit of small
gifts, with the expectation that they may lead to larger gifts in the
future. Universities seem to be convinced that this strategy is
effective, and with $8.7 billion raised from alumni in 2008, the stakes
are high (Council for Aid to Education 2009). The dean of alumni affairs
at Columbia University stated that "it isn't about the
dollars," and that the purpose of getting young alumni to donate is
to create a habit of giving (Durkin 2005). Fundraising professionals
agree--one consultant explained, "I would never say that a small
gift is not important because it's building that relationship. If
you don't build those relationships today, you may not have their
interest when the day comes that they can give those $101 million
donations" (Westmoreland 2008). This sentiment was echoed by an
expert on senior class giving at the Council for Advancement and Support
of Education, who explained that "the goal [of these programs] is
not to raise money, but to begin a pattern of behavior" (Ensign
2010).
There is scant evidence that this belief is justified. While a
number of studies have documented a positive correlation between giving
when young and giving when older (see Monks 2003; Turner, Meserve, and
Bowen 2001), this may be driven by a number of factors that have nothing
to do with building a relationship. This correlation may actually
represent spurious state dependence that arises from unobserved
heterogeneity--like the alumnus's affinity for the school. This is
contrasted with true state dependence, in which a donation in one period
affects preferences for donating in a later period.
While participation rates are a factor in university rankings, it
seems evident that development officials assume that habit-forming in
charitable contributions exists and justifies the pursuit of small gifts
when alumni are young, with the expectation that this will lead to
larger gifts in the future. These beliefs hinge on the idea that a habit
can form by the simple act of making a gift, and that the amount given
is possibly irrelevant. In essence, the proponents of this idea believe
that giving regularly when young will cause the individual to be in a
state of "focus" for giving when older--willing to make a
larger gift, perhaps because they are accustomed to giving to the
charity in each year. Standard models of habit forming, in which the
amount given in an earlier period affects a stock variable and the
individual incurs disutility from deviating from this level, do not
account for this phenomenon. Those models imply that individuals are in
the habit of giving a certain amount of money per year, not that they
are in the habit of giving in general.
This article proposes a simple model that predicts habit forming
from both the amount of giving and whether a gift is made, and uses a
unique data set to measure the relative importance of these effects. We
study alumni contributions to an anonymous private selective research
university, henceforth referred to as Anon U. The proprietary data
provided by Anon U contain detailed information about donations made by
alumni as well as a variety of their economic and demographic
characteristics. Using information about athletic performance and
solicitation by peers allows us to estimate measures of habit forming in
charitable giving untainted by unobserved heterogeneity.
As with all studies that focus on usefully unusual settings,
generalizability is an open question. It is quite possible that these
estimates cannot be directly applied to other types of charities, let
alone to the development of habits in brand preferences or drug use,
among other applications. However, the detail and scope of the data are
particularly well suited to answer questions about habits in this
particular context. At minimum, one expects that other universities
considering a fundraising program oriented toward young alumni may take
these findings as strongly suggestive. More importantly, the results
provide insight into the formation of preferences that more
representative data cannot, and future work using other distinct cases
can establish a case for generalizability.
Section II discusses prior work. Section III describes the Anon U
data set, while Section IV presents an empirical model of habit
formation in charitable giving and describes the identification
strategy. Section V presents the results, which show that persistence in
charitable giving is mostly driven by frequent giving when young, not
the amount of giving when young. Section VI concludes and provides
suggestions for future research.
II. PREVIOUS RESEARCH
Perhaps due to the difficulties of separating true state dependence
from spurious correlation, there is little research that directly
addresses persistence in donations. In a paper analyzing panel data from
income tax returns, Auten, Sieg, and Clotfelter (2002) note that
"habit formation is probably not very important in charitable
behavior," as they fail to find significantly positive
autocorrelations of donations over time. Monks (2003) mentions, in
passing, that "[i]dentifying young alumni who are more likely to
give and encouraging them to do so, even in modest dollar amounts, may
have significant lifetime giving effects." Turner, Meserve, and
Bowen (2001) concur, explaining that "participation rates are often
thought to be ... important precursors of giving patterns later in life.
In this regard, young alumni are sometimes encouraged to make token
gifts ... so that they may begin a habit of giving back." Lindahl
and Winship (1994), in an effort to identify large donors, model giving
to Northwestern University between 1988 and 1990 as a function of
earlier giving and other predictors. As their purpose is solely to
identify these large donors, they admit that causality is unclear and go
as far as to say that they "would not be at all surprised to find
that past giving had little or no 'true' effect on current
giving." Smith, Kehoe, and Cremer (1995) look at how a
household's "altruistic history" affects its probability
of making a donation to a local health clinic, using indicators for
donations to other charities in the previous year. They find that prior
donations to non-religious charities are associated with a higher
likelihood of donation, but this relationship is taken as a proxy for
attitudes toward altruism rather than a causal relationship. Finally,
Rosen and Sims (2011) use a series of cross-sections of the Survey of
Giving and Volunteering that include retrospective questions on
volunteering as a child. They find that childhood volunteering is
strongly correlated with contemporaneous giving and volunteering
behavior. They proxy for family attitudes toward altruism by including
volunteering by the respondent's parents when the respondent was a
child as a control and conclude that their results are "consistent
with the notion that altruistic behavior is habit forming." The
study is hampered, however, by the use of retrospective questions rather
than panel data and, importantly, an inability to examine the dimensions
on which habits form.
III. DATA
Our primary data source is the administrative archives of Anon
U's Development Office, which contain information on all alumni
donations from 1983 to 2009. The data are proprietary and sensitive, and
individuals' names were stripped from the records before being made
available to us. Our unit of observation is the individual. We define
giving when young as the log of the average gift made between graduation
and the end of the alumnus's fifth year since graduation, that is,
through the first major reunion. Frequent givers when young are those
who gave in each of the first 5 years after graduation, irrespective of
amount. Giving when older is defined in two ways: first, as the log of
the average gift made between the alumnus's 20th year since
graduation and 2009. Second, large gifts when older are defined using an
indicator equaling one if the alumnus was in the top 10% in his or her
class in total giving between the 20th year since graduation and 2009.
An alternate specification redefines both of these measures using the
gifts made between 15 years after graduation and 2009 for classes with
at least 20 years of data in the sample.
The Development Office data also include information on academic
major and minor, extracurricular activities when the alumnus was an
undergraduate, several variables that can be considered as proxies for
affinity (such as payment of class dues), post-graduate education,
residence, whether he or she is married to another graduate of Anon U,
and location in a given year. Anon U's Registrar supplemented these
data with information on SAT scores, academic honors, ethnicity, type of
high school, summary evaluations made by the Admissions Office during
the application process, and college grade point average.
In addition, we have information regarding varsity athletic teams
on which the alumnus participated as an undergraduate, as well as the
team's conference finish in each year. This provides a valuable
source of exogenous variation. As discussed below, variables indicating
whether the varsity team on which the alumnus participated as an
undergraduate--if any--won its conference championship have a transitory
effect on giving. The data also contain information about the
volunteering activities of alumni. Variables indicating whether an
alumnus's former freshman year roommate is a solicitor in that year
also provide exogenous shocks to giving. Further discussion of these
measures and their use as instrumental variables is in Section IV; see
Meer and Rosen (2009) for a more complete discussion of the role of
athletics in alumni giving and Meer (2011) for more details on the
effects of peer influence on charitable giving.
Since we need to observe the first 5 years of an alumnus's
giving history, the oldest class that can be included in the sample is
the class of 1982, for which the first giving opportunity was 1983. This
limitation is not ideal, since the alumnus's giving histories do
not extend through the entirety of peak earnings years; members of the
class of 1982 are about 49 years old at the end of our sample. Moreover,
a relatively limited amount of data is available past the 20
years-since-graduation mark--the class of 1982 has 8 years of data
comprising their measure of giving when older. However, the richness of
our data should enable us to examine the mechanisms by which habits form
in charitable giving.
Freshman year roommate information was not recorded for the class
of 1983. Focusing on alumni from classes of 1982 to 1989, excluding
1983, the sample includes 7,324 alumni giving histories. Dropping those
with missing covariates and those who died prior to 2009 leaves 7,113
individuals; 71.3% of these individuals made gifts of any size between
their 20th year since graduation and 2009, with a mean positive average
gift, in 2009 dollars, of $2,039.14, and a median of $119.79. (1)
Examining their giving when young, in the first 5 years since
graduation, 80.1% made any gift, and 26.3% of individuals gave in each
of those first 5 years. This latter category is our definition of
frequent givers when young. The mean positive average gift in this
period in 2009 dollars, is $51.02, with a median of $25.37. It is clear
that giving is characterized by large outliers; in our estimates,
therefore, we take logs of the amount of giving. (2)
The raw data indicate that there is a relationship between giving
when young and giving when older. The correlation between the log of the
average gift in the first 5 years and the log of the average gifts from
the 20th year after graduation onward is 0.50. Among those who were not
frequent givers when young, 63.3% gave at least once when older, while
the giving rate when older is 93.5% among those who were frequent givers
when young. The mean gift when older, conditional on giving, for those
who were not frequent givers when young is $1,203.99 with a median of
$100.07, while for those who were frequent givers, the respective
figures are $3,626.26 and $189.76. However, one cannot ascribe a causal
relationship to these differences--unobserved affinity drives both
giving when young and giving when older.
Unfortunately, the data include no direct information on income,
which is clearly an important determinant of giving. However, for a
large subset of these alumni, 5,599 individuals, we have information
that is closely related to permanent income: field and occupation. The
start- and stop-dates for these variables are unreliable; we therefore
create a series of indicators for whether the alumnus was ever employed
in that field or at that occupation. We estimate the model with this
subsample, including the field and occupation data, in order to see
whether our results are sensitive to their inclusion. (3)
Table A1 contains summary statistics and definitions of the
variables used in this study, including field and occupation variables.
IV. MODEL AND EMPIRICAL APPROACH
A. Model
We begin with the outline of a model that allows for habit
formation based on both the amount given when young, in period 1 (gl),
and whether the individual gave when young.
Individuals can either be in a non-giving state when young or a
giving state. That is,
(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [R.sub.1] is resources when young, [m.sub.1] is fundraising
effort by the charity in the pursuit of small gifts (that is, in
convincing potential donors to make any gift at all, perhaps by making
potential donors aware of the charity and its needs), and [n.sub.1] is
fundraising effort by the charity in the pursuit of larger gifts (for
example, through more intense solicitation). [P.sub.1]([m.sub.1]) is
increasing in ml, so an individual is more likely to be in the giving
state if the charity pursues small gifts. Conditional on being in the
giving state, the individual maximizes a utility function that is
differentiable and increasing in each argument. Period 2 resources are
random, reflecting the university's uncertainty about which of its
alumni will have high incomes in the future. Once again, individuals can
be in either a non-giving state or a giving state
(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [R.sub.2] are realized resources in the second period,
[m.sub.2] is fundraising effort by the charity in the pursuit of small
gifts, and [n.sub.2] is fundraising effort by the charity in the pursuit
of large gifts. The probability of being in the giving state is now
determined by both fundraising effort on the extensive margin and
whether the individual was a giver in the first period, with the
probability increasing on both dimensions.
Assumption 1: The marginal utility from giving in both periods is
increasing in the charity's contemporaneous solicitation efforts.
Assumption 2: The discount factor between the two periods is very
small.
Assumption 3: The marginal utility from giving in period 2 is not
decreasing in period 1 giving.
The increasing marginal utility in Assumption 1 may stem from a
variety of sources, including benefits from additional recognition, warm
glow, or the alleviation of increased social pressure from additional
solicitation. (4) Assumption 2 prevents agents in the early period from
being forward looking, which is consistent with the time lapse between
the early and later periods in our data and allows the two functions to
be maximized separately. Assumption 3 requires that those who give
earlier not feel decreased satisfaction from continuing to give, which
seems reasonable.
To show that gifts in period 1 are increasing in resources and the
charity's efforts, we show that [u.sub.1] is supermodular in
[g.sub.1] and the parameters [R.sub.1] and [n.sub.1]. Given the
differentiability assumption, it is sufficient to show that the mixed
partial derivatives between the choice of gift and parameters are weakly
positive. Concavity assures this for financial resources, while
Assumption 1 assures it for solicitation intensity.
(3) [[partial derivative][u.sub.1]/([partial
derivative][g.sub.1][partial derivative][R.sub.1])] > 0
(4) [[partial derivative][u.sub.1]/([partial
derivative][g.sub.1][partial derivative][n.sub.1])] > 0
From Equations (3) and (4), and Topkis's theorem, we have
(5) [([partial derivative][g.sub.1]([R.sub.1], [n.sub.1]))/[partial
derivative][R.sub.1]] [greater than or equal to] 0
(6) [([partial derivative][g.sub.1]([R.sub.1], [n.sub.1]))/[partial
derivative][n.sub.1]] [greater than or equal to] 0
Gifts in period 1 are increasing in resources and the
charity's solicitation efforts in the pursuit of larger gifts.
Next, we show that E[[g.sub.2]] is weakly increasing in I([g.sub.1]
> 0), [R.sub.2], [n.sub.2], and [g.sub.1]. Note that [g.sub.2] = 0 in
the nongiving state and is constant with respect to I([g.sub.1] > 0)
in the giving state; I ([g.sub.1] > 0) only affects the probability
of being in the giving state. Since [P.sub.2]([m.sub.2], 1) >
[P.sub.2]([m.sub.2], 0), E[g.sub.2]] is weakly increasing in I([g.sub.1]
> 0). It is also the case that
(7) [[partial derivative][u.sub.2]/([partial
derivative][g.sub.2][partial derivative][R.sub.2])] > 0
(8) [[partial derivative][u.sub.2]/([partial
derivative][g.sub.2][partial derivative][n.sub.2])] > 0
(9) [[partial derivative][u.sub.2]/([partial
derivative][g.sub.2][partial derivative][g.sub.1])] > 0
The arguments behind Equations (7) and (8) are equivalent to those
behind Equations (3) and (4), while Assumption 3 implies Equation (9).
Therefore, we see that
(10) [[DELTA]E[[[g.sub.2]([R.sub.2], [n.sub.2],
[g.sub.1])]/[DELTA]I([g.sub.1] > 0)] [greater than or equal to] 0
(11) [[partial derivative]E[[g.sub.2]([R.sub.2], [n.sub.2],
[g.sub.1])]/[partial derivative][g.sub.1]] [greater than or equal to] 0
Thus, giving in the second period will be higher both if the
individual gave at all when young and if the individual gave a larger
amount when young.
Do habits form as a result of the size of the gift, or from the act
of giving a gift when young irrespective of the size, or both? Merely
estimating a model of giving in period 2 as a function of giving in
period 1 and being a frequent giver in period 1 will not yield estimates
with a causal interpretation. After all, giving when both young and old
could be driven by some unobservable variables, such as affinity for the
school. Equations (12) to (14) show a specification that is consistent
with the model described above
(12)
[Y.sub.2i] = max(0, [[beta].sub.1][Y.sub.1i] +
[[beta].sub.2][D.sub.1i] + [X.sub.i][gamma] + [v.sub.2i])
[v.sub.2i] = [[mu].sub.i] + [[epsilon].sub.2i]
(13)
[Y.sub.1i] = max(0, [[phi].sub.1][Z.sub.1i] + [X.sub.i][delta] +
[v.sub.1i])
[v.sub.1i] = [[mu].sub.i] + [[epsilon].sub.1i]
(14)
P([D.sub.1i] = 1) = F([PHI].sub.1][Z.sub.1i] + [X.sub.i][lambda] +
[[eta].sub.1i]
[[eta].sub.1i] = [[mu].sub.i] + [[mu].sub.i] + [[omega].sub.1i]
[Y.sub.2i] is the log of the average gift in the 20th year after
graduation and onward, [Y.sub.1i] is the log of the average gift in the
first 5 years after graduation, [D.sub.1i] is an indicator for being a
frequent giver in period 1--making a gift in each of the first 5 years
after graduation--while [X.sub.i] is a vector of covariates described in
Table A1. (5) [[beta].sub.1] and [[beta].sub.2] represent true state
dependence, that is, the actual effect of giving behavior in period 1 on
giving in period 2. But spurious state dependence can be present as
well--note that the error term consists both of a period 2 specific
shock and a time-invariant effect. The latter, [[mu].sub.i], represents
unobserved affinity, which is related to giving behavior in both
periods. Higher levels of [[mu].sub.i] are associated with higher
[Y.sub.1i], a higher probability that Dig equals one, and higher
[Y.sub.2i], leading to spurious state dependence. Since [[mu].sub.i]
affects both [Y.sub.1i] and [D.sub.1i], estimating Equation (12) without
accounting for this correlation results in biased estimates.
B. Identification
An instrumental variables approach is required in order to identify
the causal effect of giving when young on giving when older. In
particular, we require a set of variables [Z.sub.1i] that plausibly
affect [Y.sub.1i] and [D.sub.1i], but are uncorrelated with [[mu].sub.i]
and [[epsilon].sub.2i]. The athletic performance and roommate
solicitation variables mentioned in Section II meet these requirements
and are described below.
Athletic Performance. We construct an indicator taking a value of 1
if the varsity athletic team on which the alumnus participated as an
undergraduate, if any, won its conference championship in any of the
first 5 years after graduation. Previous research has documented that
alumni who participated on an athletic team have an increased propensity
to donate in years in which their former team does particularly well,
fulfilling the relevance criterion (Meer and Rosen 2009). Lack of
correlation between athletic performance and [[epsilon].sub.2i] also
seems fairly evident. There is no reason to think that an alumnus's
former team's performance in the first few years after graduation
will be correlated with a shock to giving 15 to 20 years later.
The possibility of correlation between athletic performance and
[[mu].sub.i] is more worrisome. First, it is likely that there is
correlation between the performance of the alumnus's team while he
or she was an undergraduate and his or her affinity, as measured by
[[mu].sub.i. If there is also correlation between a team's
performance from year to year, this may lead to correlation between
[Z.sub.1i] and [[mu].sub.i]. We account for this possibility by
including, in X, a set of indicators taking a value of 1 if the
alumnus's team, if any, won a conference championship in his or her
freshman, sophomore, junior, or senior year. (6) A second possible
problem is that if affinity is a stock variable and shocks to it do not
dissipate from year to year, then the athletic performance variables may
have long-lasting effects and may not be excluded from Equation (12).
The long period of time between periods 1 and 2 in our data, though,
make this scenario very unlikely. For this mechanism to be operative,
one would have to believe that giving between 2002 and 2009 for an
alumnus who graduated in 1982 is substantially directly affected by
whether his or her former team won a conference championship in, say,
1985. (7) Evidence from Meer and Rosen (2009) also shows there is still
a strong effect of winning on giving in specifications with individual
fixed effects. As such, we do not believe that this problem afflicts our
results.
Slightly more than a third of individuals in our sample
participated in varsity athletics. Conditional on being an athlete,
57.6% of individuals' former teams won a championship during the
relevant period; this corresponds to an overall rate of 20.3%. Table 1
shows the coefficients for the team championship variable, drawn from
estimates of Equations (13) and (14) above and described in Section
III.C. Winning the conference championship in the first 5 years after
graduation is associated with approximately 14% (SE = 6.3%) higher
giving in the first 5 years after graduation and a 2.7 percentage points
(SE = 1.5 percentage points) higher likelihood of being a frequent giver
when young.
Roommate Solicitation. Turning to the solicitation variables, we
construct an indicator taking a value of 1 if the individual's
former freshman year roommate is a solicitor at any point in the first 5
years after graduation. (8) Approximately 30% of the sample satisfies
this definition. (9) Meer (2011) shows that the presence of such a
relationship affects giving in the years in which an individual has a
roommate solicitor. Correlation with the error terms when older seems
unlikely for same reason as athletics; there is no reason to believe
that having a former freshman year roommate who is a solicitor in the
first few years after graduation is correlated with shocks to giving 20
years later.
Correlation with unobserved affinity is potentially more
problematic. An individual's affinity for the school (and therefore
giving) is likely to be correlated with his or her roommate's; if
both have high affinity due to common experiences, there may be a
spurious correlation between an alumnus's giving and his or her
roommate's volunteering. Meer (2011) analyzes whether the
relationship between the alumnus's giving and his or her former
roommate's volunteering is because of joint affinity shocks.
Freshman roommates at Anon U are conditionally randomly assigned and
thus do not represent potentially endogenous sorting. However, common
shocks to affinity might still drive the relationship between one
alumnus's donations and another's volunteering as a solicitor.
Several diagnostics are used to evaluate this question. First, there is
no correlation between nonsolicitation types of volunteering, which
should also be related to unobserved affinity, and former
roommates' giving. Second, the timing of these personal
solicitations is such that a peer effect should only be observed toward
the end of the fiscal year; this is the case. Finally, and most
powerfully, fixed effects estimates, controlling for all time-invariant
affinity, show an increase in giving in years when an alumnus's
former freshman year roommate is a solicitor. Thus, the results of that
work strongly indicate that this peer influence on charitable giving is
not due to spurious correlation. Rather, it represents a transitory
shock to giving behavior. (l0)
Table 1, again drawing on the estimates described in Section III.C,
shows that having a former freshman year roommate who is a solicitor in
the first 5 years after graduation is associated with 7.1% (SE = 3.8%)
higher giving when young and 1.9 percentage points (SE = 0.89 percentage
points) higher likelihood of being a frequent giver. Coupled with the
results in first subsection of III.B, the instruments are jointly
significant in both equations separately and across both equations.
Other Potential Issues. There are several additional possible
concerns about these instruments. It is possible that [[beta].sub.1] and
[[beta].sub.2] are measuring a selection effect rather than habit
forming. It may be that the Development Office targets individuals in
period 2 who were large or frequent donors in period 1, and this
increased solicitation is responsible for the correlation between giving
in the two periods. We have no way of definitively proving or disproving
this hypothesis. However, it is important to note that at Anon U, all
alumni are solicited in essentially the same way, with the exception of
a small number of extremely large givers. This is especially true when
the alumni are young; indeed, an Anon U development officer explained
that this is a deliberate practice, precisely to foster the habit of
giving and avoid crowding out small donations with large gifts from a
few alumni within a class. Thus, this mechanism is unlikely to be
driving the estimates of [[beta].sub.1] and [[beta].sub.2]. Furthermore,
if this effect were actually driving the results, then estimates that
drop large givers would be very different from the main specifications.
Dropping the top 1% of givers in the young period has little qualitative
effect on the results presented in Section IV; we therefore conclude
that the possibility of increased solicitation based on earlier giving
is unlikely to be affecting the results.
C. Estimation
We estimate [Y.sub.1i] and [D.sub.1i] with, respectively, a Tobit
and a probit, as a function of [Z.sub.1i] and [X.sub.i]. These models
are shown in Equations (13) and (14), where F is the cumulative normal
distribution function. We estimate Equations (12), (13), and (14)
jointly using Roodman's (2009) conditional recursive mixed-process
estimator. Roodman's method is appropriate for models with clearly
defined stages in which endogenous variables appear as observed, that
is, not as latent variables. Given that the model above posits that
habits form through actual choices, and not an underlying desire to
give, the specification meets these criteria. With the assumption of
joint normality of the errors, Roodman shows that numerical estimation
provides consistent estimates of the parameters of interest,
[[beta].sub.1] and [[beta].sub.2]. If the amount given when young truly
has an effect on giving when older, then [[beta].sub.1] will be
positive; if focus mechanism is operative, then [[beta].sub.2] will be
positive. See Roodman (2009) for a full description of this approach, as
well as a guide to implementation.
V. RESULTS
A. Amount of Giving
We begin by examining whether habit forming has an effect on the
average gift given when older. Table 2 presents unconditional marginal
effects from a Tobit model without instrumental variables--that is,
estimating Equation (4) above. Column (1) shows marginal effects for the
average gift given from the alumnus's 20th year since graduation
through 2009. The elasticity of giving between young and old is about
0.30, meaning that a 10% increase in giving when young is associated
with a 3% increase in giving when old. Given the means of giving, this
implies that an approximately $4 increase in the average amount given
when young is associated with an approximately $44 increase in the
average amount given when older. Being a frequent giver when young is
associated with a statistically insignificant 5.0% higher average giving
when older.
As mentioned above, limiting the sample to the 20th year after
graduation and later leaves relatively few giving opportunities for each
alumnus. Taking the average gift from the 15th year after graduation
through 2009 for those who, by 2009, graduated more than 20 years
previously reduces the likelihood that the results are not being driven
by those alumni who give smaller gifts more frequently. Those results,
in Column (2), show similar results to Column (1). Including proxies for
income, like field and occupation, do not affect the results either (see
Column (3)). These variables are related to permanent income, which will
obviously be a driving factor in an individual's ability to give.
These uninstrumented results imply that universities' policies of
pursuing frequent small gifts when alumni are young in an effort to
create a habit may not pay large dividends. In essence, being a frequent
giver when young does not seem to exert influence on the amount given
when older; one possible explanation for this result is that the amount
given when young proxies for true affinity, while frequent givers either
fail to form habits, form habits that lead to small gifts, or give small
amounts often to avoid social pressure. Other explanations cannot be
discounted, and one cannot draw causal conclusions from these results,
as they do not correct for the fact that giving when both young and old
is likely to be driven by unobserved affinity. Results that account for
this endogeneity are presented in Table 3, calculated as per the
discussion in Section IV. The results are radically different from those
in Table 2. The results for the average gift from 20 years onward, in
Column (1), indicate that the amount given when young has little effect
on the amount given when old, though the estimates are not precise
(-0.14, SE = 0.18). On the other hand, the coefficient for the frequent
giver when young indicator is very large and significant (1.89, SE =
0.16). This implies that, ceteris paribus, an alumnus who gave
frequently when young gives, on average, 5.6 times more when older than
an alumnus who did not give. It is important to note that this holds the
amount given when young fixed--that is, if two alumni give the same
amount when young, but one gives in each year and the other does not,
the frequent giver is expected to give much more when older. These
results are consistent across specifications. Defining giving when older
as beginning in the 15th year after graduation, the elasticity of giving
when older with respect to giving when young is -0.17 (SE = 0.12), while
the frequent giver effect is 1.59 (SE = 0.16). Including field and
occupation variables yields similar results, with an elasticity of
giving of -0.011 (SE = 0.27) and a frequent giver effect of 1.95 (SE =
0.19). Without drawing too much inference from the exact magnitudes, it
seem evident that the pursuit of frequent gifts from young alumni, even
if the university suffers a small loss in the process, is justified.
These results suggest that the mere act of giving frequently affects
giving behavior in later years. (11)
It is also important to clarify the interpretation of the
instrumental variables estimates of [[beta].sub.1] and [[beta].sub.2].
They measure the average change in the size of a gift when older for
those whose giving patterns when younger were affected by a combination
of the presence of a solicitor who was a freshman year roommate and
their athletic team's performance. That this local average
treatment effect yields such dramatic results provides strong evidence
for habit formation, in that those who were induced into giving by
arguably exogenous variables representing social pressure and positive
affinity shocks increase their giving later in life.
We now turn to habit forming effects for large givers, whose
donations make up the bulk of the money raised in each year.
B. Class Leaders
Given that the university's desire is to cultivate large
givers, it stands to reason that we should examine the probability that
an alumnus is a large giver relative to his or her class as a function
of giving when young. To that end, we define a "class leader"
as being an individual whose gift when older is in the top 10% of his or
her class. (12) Column (1) of Table 4 presents uninstrumented results
for the probability of being a class leader, defining giving when old as
being the sum of gifts in the 20th year after graduation and onward.
(13) The coefficient on the log of giving when young (0.022, SE =
0.002), implies that a 10% increase in the size of the gift when young
increases the probability of being a class leader when older by 0.22
percentage points. While this result is statistically significant, it is
relatively small. The frequent giver effect in this specification is
significant at p = .09, but is even smaller in absolute value (-0.007,
SE = 0.004) and economically insignificant--in essence, a precisely
estimated 0. The results in Columns (2) and (3) are similar. The
conclusion drawn from these uninstrumented estimates is that giving when
young has little predictive power on an individual's likelihood of
being a big giver, relative to his or her class, when older.
Turning to the instrumented estimates, in Table 5, we find a
different story. The log of
giving when young has a positive and statistically significant
effect on the probability of being a class leader (0.12, SE = 0.018).
This effect seems quite large--a 10% increase in giving when young is
associated with an increase in the probability of being a class leader
of about 1 percentage point, with the baseline probability of being a
class leader being, by definition, 10%. The effect is similar for the 15
years onward specification in Column (2)and the field and occupation
specification in Column (3). Turning to the effect of being a frequent
giver when young, we see large and significant effects. In Column (1),
this effect is 0.21 (SE = 0.052). This is a very large effect given that
a randomly chosen alumnus has a 10% probability of being a class leader.
The coefficient's magnitude can be explained by the relatively
small number of frequent givers (14) and the local nature of the
marginal effect. Results in Columns (2) and (3) are of somewhat
different magnitudes, but similarly large and significant. Regardless,
it seems that once endogeneity is accounted for, frequent givers when
young are far more likely to be class leaders when older.
VI. CONCLUSIONS
Using a unique and relatively long panel of data on alumni giving,
we have examined how giving patterns when young affect giving when
older. The intuition of professional fundraisers, who believe that
building a habit of giving among young alumni leads to larger gifts when
older, seems justified. Because this case study examines behavior at a
single university, one must, of course, take great care in generalizing
the findings. The estimates in this article are based on those induced
into giving by the effects of freshman year roommate solicitors and
former athletic team success. It may be tempting for practitioners to
argue, based on these results, that any tactic which induces young
alumni to give should be used. Ensign (2010) documents fundraising
drives at two elite universities that used strong-arm techniques to
shame students into participating. While the use of the roommate
solicitor instrument indicates that giving induced by some social
pressure can result in large effects in the future, it seems unlikely
that those who are essentially bullied into giving will respond in the
same way. That said, the large magnitude of the effect of being a
frequent giver when young suggests that nonprofit organizations in
general and universities in particular should give serious consideration
to devoting additional resources to raising participation rates among
young potential donors. Even if the benefits are far in the future, the
effects are large enough to justify incurring some losses in the pursuit
of gifts in the present. (15) These results also have implications for
the accounting practices of charities, which are often required to
report fundraising expenses, with the ratio of donations to expenses
being used as a measure of the charity's efficiency. But in the
presence of substantial long-term habit formation, these ratios will
understate the true benefits of fundraising and perhaps even unfairly
penalize charities that focus on building relationships that lead to
large gifts in the future.
Habit forming also has implications for assessing the impact of the
charitable deduction in the personal income tax. Lowering the cost of
giving may induce much larger lifetime effects than those typically
estimated using short panels or cross-sectional data. The charitable
deduction is, of course, available to all who itemize. An important
topic for future research is to determine whether these sorts of
long-term effects can arise in older individuals. On the other hand,
while this article examines habit formation over a relatively long
period, there may also be shorter-term effects. For instance, giving in
one year may affect giving in the next by providing a reference amount
or simply the routine of giving a certain amount.
These results also have significance for models of habit forming in
other contexts. The findings in this article are not inconsistent with
those in Bronnenberg, Dube, and Gentzkow (2010), who find that exposure
to particular brands when younger affect purchasing behavior for
decades. Early experiences and habits that form through a
"focus" margin may have large impacts late in life and should
be considered in the design of models of behavior.
ABBREVIATION
Anon U: Anonymous University
TABLE A1
Variable Definitions and Summary Statistics
Variable Description
Gave20 Gave at all from the 20th year after
graduation on
Gave15 Gave at all from the 15th year after
graduation on
Gave5 Gave at all in the first 5 years after
graduation
Average20 Average of gifts, in 2009 dollars, from
the 20th year after
graduation on, conditional on giving
Average15 Average of gifts, in 2009 dollars, from
the 15th year after graduation on,
conditional on giving
Averages Average of gifts, in 2009 dollars, in
the first 5 years after graduation,
conditional on giving
Firsts 1 if the alumnus made gifts in each of
the first 5 years after graduation
Won20 1 if the alumnus's own former team won
the conference championship from the
20th year after graduation on
Solicitor20 l if the alumnus's freshman year
roommate was a solicitor from the
20th year after graduation on
Wonl5 1 if the alumnus's own former team won
the conference championship from the
20th year after graduation on
Solicitor15 1 if the alumnus's freshman year
roommate was a solicitor from the
20th year after graduation on
Won5 1 if the alumnus's own former team won
the conference championship in any of
the first 5 years after graduation
Solicitors 1 if the alumnus's freshman year
roommate was a solicitor in any of
the first 5 years after graduation
FreshmanRec 1 if the alumnus's team won the
conference championship during the
alumnus's freshman year
SophomoreRec 1 if the alumnus's team won the
conference championship during the
alumnus's sophomore year
JuniorRec 1 if the alumnus's team won the
conference championship during the
alumnus's junior year
SeniorRec 1 if the alumnus's team won the
conference championship during the
alumnus's senior year
Spouseisalum 1 if the spouse is an alumnus
Male 1 if the alumnus is male
Race/Ethnicity
White Omitted Category: 1 if the alumnus is
White
American 1 if the alumnus is a Native American
Black I if the alumnus is Black
Hispanic 1 if the alumnus is Hispanic
Asian 1 if the alumnus is Asian
Secondary Schooling
Public Omitted Category: 1 if the alumnus
attended public school
Boarding 1 if the alumnus attended boarding
school
Private 1 if the alumnus attended private school
School--Other 1 if the alumnus attended another type
of school
SATmath SAT math score. Scores prior to 1996
are adjusted to reflect recentering
of the scoring scale
SATverbal SAT verbal score. Scores prior to 1996
are adjusted to reflect recentering
of the scoring scale
Admissions Office "Nonacademic" Ranking
A Omitted Category: 1 if the alumnus
received the highest nonacademic
ranking from the admissions office
B 1 if the alumnus received the second
highest nonacademic ranking from the
admissions office
C I if the alumnus received the third
highest nonacademic ranking from the
admissions office
D 1 if the alumnus received the fourth
highest nonacademic ranking from the
admissions office
E 1 if the alumnus received the fifth
highest nonacademic ranking from the
admissions office
Admissions Office "Academic" Ranking
A Omitted Category: 1 if the alumnus
received the highest academic ranking
from the admissions office
B 1 if the alumnus received the second
highest academic ranking from the
admissions office
C 1 if the alumnus received the third
highest academic ranking from the
admissions office
D 1 if the alumnus received the fourth
highest academic ranking from the
admissions office
E 1 if the alumnus received the fifth
highest academic ranking from the
admissions office
Clubsport 1 if the alumnus played on a club team
Honors I if the alumnus graduated magna, summa,
or cum laude
GPA Alumnus's GPA
Greek 1 if the alumnus was a member of a
fraternity or sorority
Athlete 1 if the alumnus played a varsity sport
Major
Molbio Omitted Category: 1 if the alumnus
majored in Molecular Biology
Small Social Science 1 if the alumnus majored in
Anthropology, Urban Studies, or
Sociology.
English 1 if the alumnus majored in English
Economics 1 if the alumnus majored in Economics
Public Policy 1 if the alumnus majored in Public
Policy
Political Science 1 if the alumnus majored in Political
Science
Psychology 1 if the alumnus majored in Psychology
History l if the alumnus majored in History
MAE 1 if the alumnus majored in Mechanical
and Aerospace Engineering
EE/CS 1 if the alumnus majored in Electrical
Engineering or Computer Science
Arch & Civ 1 if the alumnus majored in Architecture
or Civil Engineering
Small Humanities 1 if the alumnus majored in Art, Art
History, Classics, East Asian Studies,
Linguistics, Music, Near Eastern
Studies, Philosophy, Religion, or
Languages and Literature departments
Small Engineering l if the alumnus majored in
"Engineering," Operations Research and
Financial Engineering, or Chemical
Engineering
Small Sciences 1 if the alumnus majored in Applied
Mathematics, Astrophysics,
Biochemistry, Biology, Chemistry,
Ecology and Evolutionary Biology,
Geology, Mathematics, Physics, or
Statistics
Minor
No Minor Omitted Category: 1 if the alumnus
African/African- received no minor 1 if the alumnus
American received a minor in African or
Studies African-American Studies
American Studies 1 if the alumnus received a minor in
American Studies
Theater 1 if the alumnus received a minor in
Theater
Public Policy 1 if the alumnus received a minor in
Public Policy
Other Engineering 1 if the alumnus received a minor in
Architecture, Basic Engineering,
Bioengineering, Electrical
Engineering, Geological Engineering,
Management, Materials Sciences, or
Robotics.
Other Sciences 1 if the alumnus received a minor in
Applied and Computational Mathematics,
Biophysics, Cognitive Studies,
Environmental Studies, Science in
Human Affairs, or Neuroscience.
Other Humanities I if the alumnus received a minor in a
humanities field
Teaching l if the alumnus received a teaching
certificate
UnivAward l if the alumnus received a university
service award
GradScholarship 1 if the alumnus received a graduate
scholarship from the university
AcadAward 1 if the alumnus received an academic
award
DeptAward 1 if the alumnus received a department
award
AthleteAward 1 if the alumnus received an athletic
award
MiscAward 1 if the alumnus received a
miscellaneous award
Magazine I if the alumnus receives the alumni
magazine
AC Mailable 1 if the alumnus is on the alumni
council mailing list
AG Mailable 1 if the alumnus is on the alumni giving
mailing list
AG Phonable 1 if the alumnus is on the alumni giving
call list
No Solicit 1 if the alumnus is on a no-solicit list
Reduce Solicit 1 if the alumnus is on a reduced
solicitation list
SP Participant 1 if the alumnus was a participant in
the senior class gift
No Dues 1 if the alumnus has never paid class
dues
Current Dues I if the alumnus is current on class
dues in 2009
Post Baccalaureate
Education
No Advanced Omitted Category: 1 if the alumnus has
no advanced degree
PhD 1 if the alumnus has a Ph.D. or
equivalent degree
Masters 1 if the alumnus has a masters
JD 1 if the alumnus has a JD
MD/DDS 1 if the alumnus has a medical degree
MBA I if the alumnus has an MBA
Field (b)
Arts 1 if the alumnus ever worked in the Arts
field
Agriculture 1 if the alumnus ever worked in the
Agriculture field
Architecture 1 if the alumnus ever worked in the
Architecture field
Pharmaceuticals 1 if the alumnus ever worked in the
Pharmaceuticals field
Communications 1 if the alumnus ever worked in the
Communications field
Consulting 1 if the alumnus ever worked in the
Consulting field
Education 1 if the alumnus ever worked in the
Education field
Finance 1 if the alumnus ever worked in the
Finance field
Health Care 1 if the alumnus ever worked in the
(Business/Industry) Health Care field
Hospitality 1 if the alumnus ever worked in the
Hospitality field
Information Technology 1 if the alumnus ever worked in the IT
field
Law 1 if the alumnus ever worked in the Law
field
Manufacturing 1 if the alumnus ever worked in the
Manufacturing field
Retail 1 if the alumnus ever worked in the
Retail field
Transportation 1 if the alumnus ever worked in the
Transportation field
Federal Government 1 if the alumnus ever worked for the
Federal Government
State Government 1 if the alumnus ever worked for a State
Government
Foreign Government 1 if the alumnus ever worked for a
Foreign Government
Nongovernmental Organization 1 if the alumnus ever worked in the NGO
field
Religion 1 if the alumnus ever worked in the
Religion field
Other 1 if the alumnus ever worked in another
field
Multilateral Organization 1 if the alumnus ever worked in the
Multilateral Organization field
Military 1 if the alumnus ever worked for the
Military
Occupation (b)
Government Worker 1 if the alumnus ever worked as a
government worker
Miscellaneous Worker 1 if the alumnus ever worked in some
miscellaneous occupation
Physician/Dentist 1 if the alumnus ever worked as a
physician or dentist
White Collar 1 if the alumnus ever worked in a white
collar occupation
Attorney 1 if the alumnus ever worked as an
attorney
Executive 1 if the alumnus ever worked as an
executive
Academic Worker 1 if the alumnus ever worked as an
academic
Standard
Variable Mean Deviation
Gave20 0.712 0.453
Gave15 0.793 0.404
Gave5 0.801 0.399
Average20 2,039.14 24,737.63
Average15 1,100.85 8,409.77
Averages 51.02 180.39
Firsts 0.263 0.440
Won20 0.226 0.418
Solicitor20 0.224 0.458
Wonl5 0.302 0.459
Solicitor15 0.248 0.432
Won5 0.203 0.402
Solicitors 0.299 0.458
FreshmanRec 0.0853 0.279
SophomoreRec 0.0734 0.261
JuniorRec 0.0713 0.257
SeniorRec 0.0703 0.256
Spouseisalum 0.147 0.354
Male 0.622 0.485
Race/Ethnicity
White 0.822 0.382
American 0.0027 0.0516
Black 0.0662 0.249
Hispanic 0.0420 0.201
Asian 0.0673 0.251
Secondary Schooling
Public 0.581 0.493
Boarding 0.138 0.345
Private 0.264 0.441
School--Other 0.0164 0.127
SATmath 695 76.9
SATverbal 694 77.2
Admissions Office "Nonacademic" Ranking
A 0.0195 0.138
B 0.606 0.489
C 0.364 0.481
D 0.0101 0.100
E -- --
Admissions Office "Academic" Ranking
A 0.137 0.383
B 0.422 0.493
C 0.285 0.451
D 0.154 0.361
E 0.0018 0.0427
Clubsport 0.141 0.348
Honors 0.454 0.498
GPA 3.19 0.457
Greek 0.723 0.445
Athlete 0.352 0.478
Major
Molbio 0.0291 0.168
Small Social Science 0.0204 0.141
English 0.108 0.310
Economics 0.0839 0.277
Public Policy 0.0547 0.227
Political Science 0.0952 0.293
Psychology 0.0430 0.203
History 0.126 0.331
MAE 0.0419 0.200
EE/CS 0.0727 0.259
Arch & Civ 0.0731 0.260
Small Humanities 0.111 0.315
Small Engineering 0.0208 0.143
Small Sciences 0.120 0.325
Minor
No Minor 0.811 0.391
African/African- 0.0231 0.150
American
Studies
American Studies 0.0245 0.154
Theater 0.0153 0.123
Public Policy 0.0401 0.196
Other Engineering 0.0180 0.132
Other Sciences 0.0188 0.136
Other Humanities 0.0468 0.211
Teaching 0.0080 0.0892
UnivAward 0.0143 0.119
GradScholarship 0.0522 0.222
AcadAward 0.182 0.386
DeptAward 0.129 0.336
AthleteAward 0.0335 0.180
MiscAward 0.0143 0.119
Magazine 0.933 0.250
AC Mailable 0.991 0.0944
AG Mailable 0.604 0.489
AG Phonable 0.873 0.333
No Solicit 0.0723 0.258
Reduce Solicit 0.204 0.403
SP Participant 0.547 0.498
No Dues 0.279 0.449
Current Dues 0.213 0.409
Post Baccalaureate
Education
No Advanced 0.559 0.497
PhD 0.0755 0.264
Masters 0.153 0.360
JD 0.109 0.312
MD/DDS 0.0620 0.241
MBA 0.110 0.313
Field (b)
Arts 0.0718 0.258
Agriculture 0.0030 0.0550
Architecture 0.0279 0.165
Pharmaceuticals 0.0293 0.169
Communications 0.106 0.308
Consulting 0.105 0.307
Education 0.138 0.345
Finance 0.201 0.401
Health Care 0.177 0.382
(Business/Industry)
Hospitality 0.0075 0.0863
Information Technology 0.123 0.329
Law 0.196 0.397
Manufacturing 0.0797 0.271
Retail 0.0243 0.154
Transportation 0.0102 0.100
Federal Government 0.0493 0.217
State Government 0.0334 0.179
Foreign Government 0.0038 0.0611
Nongovernmental Organization 0.0345 0.182
Religion 0.0102 0.100
Other 0.316 0.465
Multilateral Organization 0.0080 0.0499
Military 0.0080 0.0893
Occupation (b)
Government Worker 0.0113 0.105
Miscellaneous Worker 0.0802 0.272
Physician/Dentist 0.131 0.338
White Collar 0.312 0.463
Attorney 0.280 0.449
Executive 0.525 0.499
Academic Worker 0.0836 0.277
(a) Except where noted, figures are based on 7,113 observations on
alumni who graduated between 1982 and 1989, excluding the class of
1983. No alumni remaining in this sample received the lowest
nonacademic rating from the admissions office.
(b) Based on 5,599 observations with complete information on field and
occupation.
REFERENCES
Abel, A. "Asset Prices under Habit Formation and Catching Up
with the Joneses." The American Economic Review, 80((2)), 1990,
38-42.
Auten, G., H. Sieg, and C. Clotfelter. "Charitable Giving,
Income, and Taxes: An Analysis of Panel Data." American Economic
Review, 92(1), 2002, 371-82.
Ben-Arab, M., E. Briys, and H. Schlesinger. "Habit Formation
and the Demand for Insurance." Journal of Risk and Insurance,
63(1), 1996, 111-19.
Bronnenberg, B., J.-P. Dube, and M. Gentzkow. "The Evolution
of Brand Preferences: Evidence from Consumer Migration." NBER
Working Paper 16267, 2010.
Carrasco, R., J. Labeaga, and J. D. Lopez-Salido. "Consumption
and Habits: Evidence from Panel Data." The Economic Journal,
115(1), 2005, 144-65.
Carroll, C., J. Overland, and D. Weil. "Saving and Growth with
Habit Formation." The American Economic Review, 90(3), 2000,
341-55.
Council for Aid to Education. "Contributions to Colleges and
Universities up 6.2 Percent to $31.60 Billion." 2009. Accessed
February 25, 2009. http://www.
cae.org/contenffpdf/VSE_2008_Survey_Press_Rele ase_with Tables.pdf
Diamond, W., and S. Noble. "Defensive Responses to Charitable
Solicitations." Journal of Interactive Marketing 15(3), 2001, 2-12.
Durkin, E. "Senior Fund Collect; 75 Percent Gave in
'04." Columbia Spectator. February 14, 2005.
Dynan, K. "Habit Formation in Consumer Preferences: Evidence
from Panel Data." American Economic Review, 90(3), 2002, 391-406.
Ensign, R. "2 Ivy League Drives Shame Seniors Who Don't
Give." Chronicle of Higher Education. 2010. October 24, 2010.
http://chronicle.com/article/Students-at2-Ivy-League/125056/
Fuhrer, J. "Habit Formation in Consumption and Its
Implications for Monetary-Policy Models." American Economic Review,
90(3), 2000, 367-90.
Gupta, S., P. Chintagunta, and D. Wittink. "Household
Heterogeneity and State Dependence in a Model of Purchase Strings:
Empirical Results and Managerial Implications." International
Journal of Research in Marketing, 14(4), 1997, 341-57.
Lindahl, W., and C. Winship. "A Logit Model with Interactions
for Predicting Major Gift Donors." Research in Higher Education,
35(6), 1994, 729-43.
Malmendier, U., and S. Nagel. "Depression Babies: Do
Macroeconomic Experiences Affect Risk-Taking?" Working Paper,
University of California at Berkeley, 2007.
Meer, J. "Brother, Can You Spare a Dime? Peer Effects in
Charitable Solicitation." Journal of Public Economics 95(7-8),
2011,926-41.
Meer, J., and H. S. Rosen. "The Impact of Athletic Performance
on Alumni Giving: An Analysis of Micro Data." Economics of
Education Review, 28(3), 2009, 287-94.
Monks, J. "Patterns of Giving to One's Alma Mater among
Young Graduates from Selective Institutions." Economics of
Education Review 22, 2003, 121-30.
Naik, N., and M. Moore. "Habit Formation and Intertemporal
Substitution in Individual Food Consumption." Review of Economics
and Statistics, 78(2), 1996, 321-28.
Pacula, R. "Adolescent Alcohol and Marijuana Consumption: Is
There Really a Gateway Effect?" NBER Working Paper No. 6348, 1998.
Portolese-Dias, L. "Generational Buying Behaviors."
Dissertation, Argosy University, 2004.
Roodman, D. "Estimating Fully Observed Recursive Mixed-Process
Models with CMP." Working Paper No. 168, Center for Global
Development, 2009.
Rosen, H., and S. Sims. "Altruistic Behavior and Habit
Formation." Nonprofit Management and Leadership, 21(3), 2011,
235-53.
Smith, V., M. Kehoe, and M. Cremer. "The Private Provision of
Public Goods: Altruism and Voluntary Giving." Journal of Public
Economics, 58(1), 1995, 107-26.
Turner, S., L. Meserve, and W. Bowen. "Winning and Giving:
Football Results and Alumni Giving at Selective Private Colleges and
Universities." Social Science Quarterly, 82(4), 2001, 812-26.
Westmoreland, M. "U. Targets Seniors for Fundraising."
The Daily Princetonian. March 25, 2008.
Woittiez, I., and A. Kapteyn. "Social Interactions and Habit
Formation in a Model of Female Labour Supply." Journal of Public
Economics, 2(2), 1998, 185-205.
(1.) Examining the sum of gifts made from the 15th year since
graduation onward by these individuals, 79.3% made a gift of any size,
with a mean positive average gift of $1,100.85 and a median of $98.44.
(2.) A logarithmic transformation presents problems for
observations that take a value of 0; 66 individuals have an average gift
greater than 0 but less than or equal to $1.00 when young, along with 17
such observations when older. We set these equal to $1.01. Therefore,
observations for which there is no giving are associated with $1, whose
logarithm is 0.
(3.) Estimating the model without field and occupation covariates,
but using the field and occupation sample, shows no qualitative
differences with the results from the full sample.
(4.) For a discussion of social pressure in charitable giving, see
Meer (2011). It is also possible that increased solicitation results in
lower marginal utility. Diamond and Noble (2001), using results from a
small survey, find that donors may develop defense mechanisms in
response to frequent or aggressive solicitation. We assume that Equation
(3) holds in the region with which we are concerned.
(5.) Means of time-varying variables--specifically, the location
effects--for each alumnus are used in X. Graduating class-year effects
are also included, which account for any cohort-specific shocks during
those individuals' time at the university, as well as the number of
years since their graduation.
(6.) Including sets of indicators for second, third, and other
finishes in each undergraduate year does not affect the results,
alleviating concern about intertemporal correlation in team performance
introducing bias.
(7.) An indicator for the performance of the alumnus's team in
the 20th year after graduation and onward is also included in the model.
(8.) Start and stop dates for volunteers are not reliable prior to
1992. Therefore, this variable measures whether an individual's
freshman year roommate is listed as having been a solicitor in 1991 or
earlier. Clearly, for younger classes this is quite close or identical
to whether the alumnus's roommate was a solicitor in the first 5
years after graduation. For older classes, the variable is merely
measured with some noise.
(9.) A total of 6.8% of individuals had both a former freshman
roommate who was a solicitor and a former team that won a conference
championship in the first 5 years; thus, approximately 44% of
individuals fall into at least one category.
(10.) It is certainly possible that alumni who volunteer as
solicitors when young also do so when older, influencing their former
roommates' giving in both periods and introducing spurious
correlation. An indicator for whether the individual's freshman
year roommate is a solicitor in the later period is included to account
for this possibility.
(11.) An interesting additional approach is to limit the focus on
individuals who give in both the early and late period. There are 4,486
observations in this sample, and the results are not qualitatively
different from those using the full sample. The effect of the amount
given is relatively small and noisily estimated (0.35, SE = 1.55), while
the effect of being a frequent giver is large and significant (0.94, SE
= 0.19). The point estimates for frequent givers in Table 3 are larger,
but it is important to note that the comparison group in this case is
different.
(12.) The threshold for being a class leader varies by cohort from
an average gift of $8,375 to a maximum $19,573 for the 20th year and
onward. For the 15th year onward, the threshold varies from $5,158 to
$10,944.
(13.) Sample sizes differ slightly because of variable
collinearity.
(14.) Of the frequent givers when young (who comprise about 26% of
the sample), 19.2% are class leaders when older, while only 7.4% of
those who were not frequent givers when young are class leaders when
older.
(15.) This presumes, of course, that the costs of solicitation are
relatively low. While the addition of another solicitation for an
individual may be relatively inexpensive, setting up an entire program
for the aggressive solicitation of new donors may be much more costly.
JONATHAN MEER *
JONATHAN MEER, I am grateful to Nicole Arshan, B. Douglas Bernheim,
Kevin Cotter, William Hardt, Han Hong, Caroline Hoxby, Liam Morton,
Sriniketh Nagavarapu, Kyle Pubols, David Roodman, Harvey Rosen, Andres
Santos, Jeffrey Yellin, and seminar participants at Rice University, the
University of Houston, the American Economic Association, and Texas
A&M University. I am especially indebted to Edward Van Wesep. This
research was supported in part by the Stanford Institute for Economic
Policy Research and by the Koret Foundation.
Meer: Professor, Department of Economics, Texas A&M University,
College Station, TX 77843. Phone 650291-4925, Fax 650-291-4925, E-mail
jmeer@econmail.tamu.edu
doi: 10.1111/ecin.12010
TABLE 1
First Stage Estimates (a)
(1) (2)
Log Average Frequent
Giving when Giver when
Young Young
Freshman roommate was 0.0709 * 0.0189 **
solicitor in the first
5 years (0.0379) (0.0089)
Team championship in the 0.139 ** 0.0268 *
first 5 years (0.0634) (0.0150)
Joint probability of 0.0136 0.0128
significance in equation
Joint probability of 0.0272
significance across
equations
(a) Results in this table are based on 7,113 observations
on alumni graduating in the classes of 1982 through 1989,
excluding the class of 1983. Column (l) reports unconditional
marginal effects on the log of giving when young
based on results from the conditional recursive mixed
process estimator (Roodman 2009) corresponding to Column
(1) in Table 3. Column (2) reports marginal effects on the
probability of being a frequent giver when young, based on
results from a conditional recursive mixed-process estimator
corresponding to Column (1) in Table 3. Robust standard
errors are reported in parentheses. In addition to the variables
listed above, models include cohort effects, the covariates
listed in Table Al, and location effects (averaged over each
alumnus's post-graduation history). Full results are available
on request.
* Significant at the 10% level; ** significant at the 5%
level.
TABLE 2
Uninstrumented Estimates--Amount of Gifts (a)
(1) (2)
Twenty Years On Fifteen Years On
Log average giving when young 0.298 ** 0.343 **
(0.0217) (0.0173)
Frequent giver when young 0.0499 -0.0559
(0.0532) (0.0467)
(3)
Field and Occupation
Log average giving when young 0.303 **
(0.0253)
Frequent giver when young 0.0256
(0.0604)
(a) Columns (1) and (2) are based on 7,113 observations on alumni
graduating in the classes of 1982 through 1989, excluding the class of
1983. Column (3) is based on 5,599 observations with complete data on
field and occupations. This table reports unconditional marginal
effects on the amount of giving when older, generated by a Tobit model
without instrumenting for the log of giving when young and the
frequent giver when young indicator. Robust standard errors are
reported in parentheses. In addition to the variables listed above,
models include cohort effects, the covariates listed in Table Al, and
location effects (averaged over each alumnus's post-graduation
history). Column (3) includes the field and occupation variables
listed in Table Al. Full results are available on request.
* Significant at the 10% level; ** significant at the 5% level.
TABLE 3
Instrumented Estimates--Amount of Gift (a)
(1) (2)
Twenty Years On Fifteen Years On
Log average giving when young -0.135 -0.167
(0.183) (0.117)
Frequent giver when young 1.89 ** 1.59 **
(0.159) (0.156)
(3)
Field and Occupation
Log average giving when young -0.0108
(0.266)
Frequent giver when young 1.95 **
(0.187)
(a) Columns (1) and (2) are based on 7,113 observations on alumni
graduating in the classes of 1982 through 1989, excluding the class of
1983. Column (3) is based on 5,599 observations with complete data on
field and occupations. This table reports unconditional marginal
effects on the amount of giving when older based on results from a
conditional recursive mixed-process estimator (Roodman 2009),
instrumenting for the log of giving when young and the frequent giver
when young indicator using the won5 and solicitor5 variables described
in Table Al. Robust standard errors are reported in parentheses. In
addition to the variables listed above, models include cohort effects,
the covariates listed in Table Al, and location effects (averaged over
each alumnus's post-graduation history). Column (3) includes the field
and occupation variables listed in Table Al. Full results are
available on request.
* Significant at the 10% level; ** significant at the 5% level.
TABLE 4
Uninstrumented Estimates--Class Leaders (a)
(1) (2)
Twenty Years On Fifteen Years On
Log of giving when young 0.0215 ** 0.0206 **
(0.0021) (0.0021)
Frequent giver when young -0.0072 * -0.0051
(0.0039) (0.0032)
(3)
Field and Occupation
Log of giving when young 0.0252 **
(0.0026)
Frequent giver when young -0.0119 **
(0.0051)
(a) Columns (1) and (2) are based on 6,862 and 6,915 observations,
respectively, on alumni graduating in the classes of 1982 through
1989, excluding the class of 1983. Column (3) is based on 5,422
observations with complete data on field and occupations. This table
report marginal effects on the probability of being in the top 10% of
givers in one's class, generated by a probit model without
instrumenting for the log of giving when young and the frequent giver
when young indicator. Robust standard errors are reported in
parentheses. In addition to the variables listed above, models include
cohort effects, the covariates listed in Table Al, and location
effects (averaged over each alumnus's post-graduation history). Column
(3) includes the field and occupation variables listed in Table Al.
Full results are available on request.
* Significant at the 10% level; ** significant at the 5% level.
TABLE 5
Instrumented Estimates--Class Leaders (a)
(1) (2)
Twenty Years On Fifteen Years On
Log of giving when young 0.115 ** 0.136 **
(0.0180) (0.0232)
Frequent giver when young 0.212 ** 0.168 **
(0.0520) (0.0703)
(3)
Field and Occupation
Log of giving when young 0.108 **
(0.0322)
Frequent giver when young 0.255 **
(0.0646)
(a) Columns (1) and (2) are based on 6,862 and 6,915 observations,
respectively, on alumni graduating in the classes of 1982 through
1989, excluding the class of 1983. Column (3) is based on 5,422
observations with complete data on field and occupations. This table
reports unconditional marginal effects on the probability of being in
the top 10% of givers in one's class, based on results from a
conditional recursive mixed-process estimator (Roodman 2009),
instrumenting for the log of giving when young and the frequent giver
when young indicator using the won5 and solicitor5 variables described
in Table A1. Robust standard errors are reported in parentheses. In
addition to the variables listed above, models include cohort effects,
the covariates listed in Table Al, and location effects (averaged over
each alumnus's post-graduation history). Column (3) includes the field
and occupation variables listed in Table A1. Full results are
available on request.
* Significant at the 10% level; ** significant at the 5% level.
