Nonprofit and Voluntary Sector
Quarterly
http://nvs.sagepub.com/
The Nature and Causes of the U-Shaped Charitable Giving Profile
Russell N. James III and Deanna L. Sharpe
Nonprofit and Voluntary Sector Quarterly 2007 36: 218
DOI: 10.1177/0899764006295993
The online version of this article can be found at:
http://nvs.sagepub.com/content/36/2/218
Published by:
http://www.sagepublications.com
On behalf of:
Association for Research on Nonprofit Organizations and Voluntary Action
Additional services and information for Nonprofit and Voluntary Sector Quarterly can be found at:
Email Alerts: http://nvs.sagepub.com/cgi/alerts
Subscriptions: http://nvs.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://nvs.sagepub.com/content/36/2/218.refs.html
>> Version of Record - May 24, 2007
What is This?
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
The Nature and Causes of the
U-Shaped Charitable Giving Profile
Russell N. James III
University of Georgia
Deanna L. Sharpe
University of Missouri–Columbia
The U-shaped income-giving profile, where those in the lower and higher income brackets give higher percentages of their income to charity, has been the subject of much dispute. Examining data from 16,442 households, the authors find clear evidence of a
U-shaped relationship. Previous findings contradicting the U-shaped profile are shown
to suffer from selection bias that systematically deflates reported lower-income giving
levels. Although the U-shaped profile is an appropriate descriptor, it does not reflect
typical household behavior. Instead, it is driven almost entirely by the 5% of households
that contribute one tenth or more of their after-tax income. Traditionally, the presence
of so many highly committed, low-income households has been attributed to religious
sect affiliation by the poor. The authors find an additional explanation in that these
highly committed, lower-income households are dramatically wealthier than other
members of their income classification, in part reflecting the presence of lower-income,
higher-asset, retirement-aged households.
Keywords: philanthropy; nonprofit; religion; charitable giving; consumer expenditure
survey
The United States is notable for its level of private charitable giving to nonprofit organizations. Individuals contributed more than $207 billion toward
nonprofit organizations in 2004 (AAFRC Trust For Philanthropy, 2005). Such
giving is integral to the operation of the American nonprofit sector.
Individual charitable contributions account for 20% of all nonprofit organization receipts in America (Brooks, 2004). Religious nonprofits are even
more reliant on charitable gifts, as donations represent 84% of total receipts
Nonprofit and Voluntary Sector Quarterly, vol. 36, no. 2, June 2007 218-238
DOI: 10.1177/0899764006295993
© 2007 Association for Research on Nonprofit Organizations and Voluntary Action
218
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Charitable Giving Profile
219
(Brooks, 2004). As a group, Americans are among the world’s most generous
contributors to nonprofit organizations. But this generosity is not evenly
distributed across individuals. One particularly fascinating, and disputed,
concept has been the U-shaped relationship between the percentage of
income contributed to philanthropy and household income levels. This
U shape refers to a phenomenon appearing in income-giving profiles
where those at the lowest and highest ends of the income spectrum give
the largest portions of their income to charitable causes, whereas those in
the middle give lower proportions. We attempt to provide convincing evidence of the existence of the relationship, uncover a serious methodological deficiency of previous challenges, and analyze the remarkable factors
that drive this phenomenon.
The concept of the U-shaped income-giving profile is not new to charitable
research. Clotfelter and Steuerle (1981), for example, graphed a U-shaped relationship using data from the 1975 Internal Revenue Service—Statistics of
Income. Auten, Clotfelter, and Schmalbeck (2000) find the same relationship with mean, but not median, giving levels using a 1991 to 1995 panel
from the Internal Revenue Service—Statistics of Income. The greater proportional giving at higher income levels is often accounted for by the presence of greater disposable income combined with an American culture of
philanthropy (Ostrower, 1995). Greater giving at lower income levels is
often explained by the impact of religious sect affiliation among the poor
(Iannaccone, 1988; Van Slyke & Brooks, 2005). Although still occasionally
referenced in the literature, the validity of this U-shaped relationship as a
legitimate descriptor is presently in question because of a series of challenges by Schervish and Havens (1995a, 1998, 2001), reviewed in more
detail later.
We examine the relationship using data from the U.S. Department of
Labor’s Consumer Expenditure Survey (CE). For many decades, the CE has
been successfully employed as a source of data for research on charitable
giving (Andreoni & Scholz, 1998; Bradley, Holden, & McClelland, 2005;
Brooks, 2002; Hrung, 2004; Reece, 1979; Reece & Zieschang, 1985).
Specifically, the CE has been employed as a source of information on household charitable giving and household income. For example, Andreoni and
Scholz (1998) employ the CE-reported levels of household charitable giving
and household income to estimate income elasticity, as do Reece (1979) and
Reece and Zieschang (1985).
DATA
We examine several years of data gathered from the CE. The 16,442 households include all complete income reporters finishing a final (5th quarter)
report during the 3-year period between the 2nd quarter of 1998 and the 1st
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
220
James, Sharpe
quarter of 2001. We convert all monetary figures to constant 2001 dollars
adjusted by the Consumer Price Index. Charitable giving reports are made
in response to five different questions in the CE:
1. During the past 12 months, how much were contributions to church or
other religious organizations, excluding parochial school expenses,
made by you (or any members of your [household])?
2. During the past 12 months, how much were contributions to charities,
such as United Way, Red Cross, etc., made by you (or any members of
your [household])?
3. During the past 12 months, how much were contributions to educational organizations made by you (or any members of your [household])?
4. During the past 12 months, how much were political contributions
made by you (or any members of your [household])?
5. During the past 12 months, how much were other contributions made
by you (or any members of your [household])?
We label the amount reported in response to the first question as “religious
gifting,” the sum of the responses to the final four questions as “secular gifting,” and the sum of all responses as “total gifting.”
THE U-SHAPED GIVING PROFILE
Table 1 provides the details for each of our income increments. We examine
income-giving segments in $10,000 increments from less than $10,000 to
$100,000. To address issues of micronumerosity, we condense the upper income
categories into the $100,000 to $149,999 level and the $150,000 and more level.
This compression also helps to avoid distortion from topcoding where, to preserve participant privacy, each income component observation that falls outside
certain very high critical values is replaced with a topcoded value that represents the mean of the subset of all such outlying observations (U.S. Department
of Labor, Bureau of Labor Statistics, 2005). Average giving levels reflect average
gifts among all households in the category including nongiving households.
The income-giving profile among these 16,442 households displays the
expected U-shaped characteristics (see Figure 1). For comparison, we also display the profiles for the two subdivisions of religious and nonreligious charitable giving. Households at the lowest income level (less than $10,000) on
average give 4.55% of their after-tax income to charitable organizations. Those
in the $10,000 to $19,999 category give 2.37% of income. Those making $20,000
to $29,999 give 2.14% of income. The declining trend continues until we reach
those making $50,000 to $59,999 giving 1.36% of income. We later see an
increase as those making $100,000 to $150,000 give 1.51% of income, and
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Charitable Giving Profile
221
Table 1. Income Giving Profile for 16,442 Households
(1998-2001 Consumer Expenditure Survey)
After Tax
Income
Less than $10,000
$10,000-$19,999
$20,000-$29,999
$30,000-$39,999
$40,000-$49,999
$50,000-$59,999
$60,000-$69,999
$70,000-$79,999
$80,000-$89,999
$90,000-$99,999
$100,000-$149,999
$150,000+
% of Income
Gifted
% of Income
Gifted
(Nonreligious)
4.55
2.37
2.14
1.59
1.66
1.36
1.44
1.71
1.46
1.34
1.51
2.17
1.28
0.57
0.65
0.26
0.44
0.30
0.30
0.44
0.41
0.40
0.52
0.92
% of Income
Gifted
(Religious)
3.27
1.81
1.49
1.32
1.22
1.05
1.13
1.27
1.06
0.94
0.99
1.25
% of
Sample
% of
Donations
12.50
18.40
15.00
12.70
9.40
8.00
6.10
4.60
3.40
2.70
4.90
2.30
4.10
8.40
10.40
9.10
9.10
7.80
7.30
7.70
5.50
4.50
11.30
14.80
finally with those making more than $150,000 giving 2.17% of their after-tax
income. Graphing these income-giving profiles reveals the distinctive U
shape. More formally, a least squares regression of the quadratic,
Ci = α + β1Xi + β2X2i + εi
where Ci is each income increment’s mean giving percentage, Xi is each
income increment’s mean after-tax income, X2i is the square of the income
increment’s mean after-tax income, and εi is the error term, confirms the U
shape of our graph. β1 is negative (–3.56E-07) and significant at the 1% level
(p = .008); β2 is positive (1.34E-12) and significant at the 1% level (p = .006),
with R2 = .59. The resulting curve is roughly U or reverse J shaped, with a
minimum giving percentage at $132,282. Because the X2i term transforms the
equation from a linear to a quadratic form, the significance of the β2 term
also equals the results from the F test comparing the quadratic against the
linear equation form. Thus, the significance level of β2 demonstrates the likelihood that the quadratic, in this case the posited U shape, better fits our data
than a linear relationship.
As an alternative approach, we consider income percentiles, dividing all
households in the sample into 100 income percentiles based on after-tax
income levels. Our model then becomes,
Ci = α + β1Xi + β2X2i + εi
where Ci is the mean giving percentage for each income percentile, Xi is the
mean income level of each income percentile, X2i is the square of the mean
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
222
James, Sharpe
5.00%
4.50%
4.00%
3.50%
3.00%
2.50%
2.00%
1.50%
1.00%
0.50%
+
9
00
99
9
,0
50
$1
,0
00
-$
14
9,
99
9
$9
00
$1
$9
0,
00
0-
$8
000
0,
9,
99
9
9,
99
9
$7
00
0$8
0,
$7
9,
99
9
9,
$6
000
0,
$6
00
0-
$5
9,
99
9
99
9
0,
$5
00
0-
$4
9,
99
9
0,
$4
0,
00
0-
$3
9,
99
9
9,
$2
000
0,
$2
$3
$1
0-
00
0,
$1
un
de
r$
10
9,
,0
99
00
0.00%
Percentage of Income Gifted
Percentage of Income Gifted (Religious)
Percentage of Income Gifted (Non-Religious)
Figure 1. The U: Percentage of Income Gifted in Each Income Increment (Consumer
Expenditure Survey 1998-2001)
Note: N = 16,442.
income level of each income percentile, and εi is the error term. We drop the
lowest percentile because of negative average income and analyze the
remaining 99 observations. Again, Table 2 displays similar results, confirming that the U-shaped quadratic better fits the data than does a linear relationship. The β coefficients, significant at the .1% level, describe a roughly
U-shaped or reverse-J curve, with its lowest point at $151,590.
Finally, we examine a third model where Xi represents the income
percentile rank itself (from 1 to 99), X2i represents the square of the income
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
223
12
−3.55E-07
.0058
1.34E-12
.0089
0.0333
< .0001
.59
$132,282
n
β1 (income M)
p
β2 (income squared M)
p
Intercept
p
R2
Minimum point
(0.0043)
(4.04E-13)
(9.90E-08)
Original
10K
Increments
(0.0023)
(2.60E-13)
< .0001
1.01E-12
.0002
0.0307
< .0001
.21
$151,590
(6.20E-8)
99
−3.05E-07
Percentile
Increments
β2 (income
percentile
squared)
β1 (income
percentile)
Table 2. Least Squares Regressions (1998-2001 Consumer Expenditure Survey)
< .0001
0.043939
< .0001
.40
67.5th percentile
< .0001
6.71E-06
99
−0.0009
Percentile
Increments
(0.003171)
(1.42E-06)
(0.00015)
224
James, Sharpe
percentile rank, Ci is the mean giving percentage for each income percentile,
and εi is the error term. Again, Table 2 reports a similar pattern. β1 is negative and significant at the .1% level (p < .0001); β2 is positive and significant
at the .1% level (p < .0001), with R2 = .40 for the regression. This third
approach generates another U-shaped or reverse-J curve, with a minimum
point at the 67th income percentile in our sample.
We avoid using the reverse-J descriptor in favor of a U shape based on
results from other surveys, such as the Survey of Consumer Finances (SCF),
that specifically oversample very high income households. These demonstrate a rising percentage giving level at the very highest income levels,
levels that are not strongly represented in our data set. On the same basis, we
similarly avoid modeling the data as a power relationship with a rapidly
then slowly diminishing income-segment gifting percentage, although such
a model does generate a higher R2 than the U-form quadratic in some of our
permutations.
In sum, we see confirmation of the U-shaped income-giving profile, most
strongly in the left side of the U associated with lower income levels, demonstrated in several approaches. The pattern can be ascertained by visual
inspection of fixed income increments, by least squares regression of fixed
income increments, by least squares regression of income percentiles using
mean income levels and its square as our independent variables, and by least
squares regression of income percentiles using the percentile ranking and its
square as our independent variables.
CHALLENGES TO THE U-SHAPED GIVING CURVE
Schervish and Havens (1995a, 1998, 2001) have published a series of challenges to the existence of the U-shaped giving curve in similar income-giving
profiles. They explain that “the relationship between household income and
percentage of income contributed to charity is described by neither a reverseJ nor a U-shaped curve. Rather the relationship is essentially flat with, if anything a slight upturn as income rises” (Schervish & Havens, 1998, p. 421).
Although agreeing that the U shape appears if the analysis is restricted to
donating households, the authors contend that it does not appear in an
analysis of the general population (Schervish & Havens, 1995b).
We begin with Schervish and Havens’s (2001) article to uncover the source
of the dissonance between their conclusion and the results of our present
study. The results reported there are generated by an amalgamation of both
the 1995 SCF and the 1996 General Social Survey (GSS) that is used to generate an income-giving profile. The published information does not describe the
exact form of the amalgamation in such a way as to be precisely replicable, so
we review the two data sets separately. The SCF omits contributions of less
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Charitable Giving Profile
225
than $500, essentially counting them as $0. To understand the impact of this
approach for low-income households, consider an individual working a fulltime job at minimum wage who gives to charity at 2.5 times the average
national rate (estimated at 1.9% of personal income; AAFRC Trust For
Philanthropy, 2005). The SCF counts this individual as a nongiver.
Conversely, a person making $275,000 who gives at less than one tenth of the
national rate is counted as a donor. This survey methodology clearly makes
the SCF a poor instrument when trying to make charitable giving estimates
regarding lower income households. This is important because it is precisely
this lower income portion of the U-shaped curve that is being challenged.
The authors attempt to ameliorate this weakness in the SCF by commingling
its results with those from the 1996 GSS.
The 1996 GSS includes a subsample of fewer than 1,500 individuals who
were asked about their charitable giving. Actual income levels are not
reported, but rather individuals select a predefined income increment level.
To calculate giving averages as a percentage of household income, we estimate income levels using the midpoints of each GSS income increment converted to the increment levels used in the 2001 article. The accompanying
table (Table 3) shows that the income-giving table for the 1996 GSS does follow the distinctive U-shaped curve, except at one increment. Before giving
too much weight to this single nonconforming income increment, we note
that these subsamples contain relatively few observations—roughly 6% of
the size of the comparable income increment samples reported in our present study—and thus wider variation is not unexpected. Nevertheless, given
that 6 of 7 income categories do follow the characteristic U shape, it is at first
difficult to ascertain how the GSS data could be viewed as invalidating the
U-shaped income-giving profile hypothesis.
How then, using these underlying data sets, do we get to a conclusion
that challenges the U-shaped giving profile? The answer appears to lie in the
introduction of a selection bias. In each of Schervish and Havens’s (1995a,
1998, 2001) articles challenging the U-shaped income-giving profile, the
authors introduce a selection process that omits all observations where survey information was provided by those not considered “heads” of households. The 1995 SCF explains, “The person referred to as the ‘head’ in this
codebook is either the male in a mixed-sex couple or the older individual in
a same-sex couple.” Some legitimately disfavor this methodology of omitting all households where the data were provided by married women or
other non-heads on purely a priori theoretical grounds (Presser, 1998). More
specific to the current analysis, we find that for the particular relationship
under examination, this selection process dramatically alters the resulting
abridged data sets, creating a systematic bias. Table 4 shows the impact of
this selection process. Examining $10,000 income increments up to the maximum bounded increment from the 1996 GSS, we report the impact on the
data from the SCF, from the GSS, and from both.1
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
226
James, Sharpe
Table 3. Income-Giving Profile From 1996 General Social Survey
Income
Increment
Less than $10,000
$10,000-$19,999
$20,000-$29,999
$30,000-$39,999
$40,000-$49,999
$50,000-$59,999
$60,000-$74,999
Estimated Average
Income
$5,510.07
$14,787.50
$24,667.09
$34,848.48
$45,000.00
$55,000.00
$72,500.00
Average
Gifting
$85.63
$167.81
$356.87
$488.45
$542.00
$599.39
$1,096.11
n
Percentage of
Estimated
Income Gifted
149
200
199
198
138
119
107
1.55
1.13a
1.45
1.40
1.20
1.09
1.51
a. Nonfitting.
This selection process changes average giving levels by as much as 50%.
More importantly, the selection process masks the U-shaped profile found in
preceding and subsequent research by dramatically reducing average giving
levels of lower income segments and at the same time increasing average
giving levels of middle-range income segments. We briefly postpone discussion regarding why this selection process creates such a strong bias until
after examining the sources of the U-shaped giving profile. Nevertheless, it
is clear from the descriptive statistics that excluding these households creates an appreciable change in the data.
In addition to the head of household selection issue, the Schervish and
Havens (1995a) article suffers from an additional reporting concern. In each
of the income increments, the authors use the increment midpoint as an estimator of average income. However, in the lowest category, under $7,000,
rather than using a $3,500 midpoint to match the approach used in the other
categories, the authors instead employ a $5,000 “midpoint” as an estimate.
Clearly, the lowest income category is being subjected to a bias not used for
the other categories. This curious increase in estimated income results in a
cut to the lowest category’s percentage of income given to charity by 30%. If
we treat the lowest income category as the others and calculate using a
$3,500 midpoint, a least squares regression confirms the resulting U-shaped
curve.
Specifically, the quadratic form,
Ci = α + β1Xi + β2X2i + εi,
where Ci is each income increment’s estimated mean giving percentage, Xi is
each income increment’s reported income, X2i is the square of the income
increment’s reported income, and εi is the error term, generates a negative
β1 (–2.9E-07) significant at the 1% level (p = .009), a positive β2 (2.86E-12)
significant at the 1% level (p = .006), with R2 = .28 for the regression. The
resulting curve is U shaped with a minimum point at $51,245 of income.
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
227
SCF
Original
Giving
786.75
150.52
363.75
510.98
824.75
1071.20
1667.97
Income
Increments
Less than $10,000
$10,000-$19,999
$20,000-$29,999
$30,000-$39,999
$40,000-$49,999
$50,000-$59,999
$60,000-$74,999
335.87
137.53
372.76
603.45
959.47
1238.04
2196.22
SCF Head
of Household
Selected Giving
(450.88)
(12.99)
9.01
92.48
134.71
166.84
528.25
∆ From
Selection
(57)
(9)
2
18
16
16
32
% ∆ in
Giving
85.63
167.81
356.87
488.45
542.00
599.39
1096.11
GSS
Original
Giving
90.46
160.86
324.91
456.89
646.21
730.32
1362.44
GSS Head of
Household
Selected Giving
4.82
(6.94)
(31.97)
(31.56)
104.21
130.93
266.33
∆ From
Selection
Table 4. Selection Bias From Exclusion of Non–Head of Household Respondents
in Survey of Consumer Finances (SCF 1995) and General Social Survey (GSS 1996)
6
(4)
(9)
(6)
19
22
24
% ∆ in
Giving
(51)
(6)
(3)
6
17
18
29
Combined
% ∆ in
Giving From
Selection
228
James, Sharpe
Analyzing the goodness of fit using an F test indicates that we can reject the
linear model in favor of the U-shaped second-order polynomial, as the
resulting F statistic is 9.051 (p = .006). In addition, the second order polynomial cannot be rejected in favor of the posited third-order polynomial—
adding the cube of estimated income—that was plotted in the 1995 article, as
the comparison generates an F statistic of 0.002 (p = .964). Thus, we must conclude that even the head of household–screened data underlying the 1995
article actually support the hypothesis of a U-shaped income-giving profile.
THE LINEAR PARTICIPATION PROFILE
If, then, we can successfully support the presence of a U-shaped incomegiving profile in American philanthropy, should we take from this profile
that only the poor and, to a lesser extent, the wealthy are generous, whereas
everyone in between is stingy? The answer is no. The U-shaped graph suggests a deceptively simple relationship that, on further analysis, is more
unusual than we might expect. Schervish and Havens (1995a) accurately
point out the inappropriate characterization of this relationship in popular
articles identifying low-income households as typically generous and the
middle class or wealthy as typically stingy. We see a clear demonstration of
the inappropriateness of such characterizations of typical generosity by
examining the charitable participation rates in each income increment. The
proportion of households engaging in charitable giving generally increases
as the income increments rise (see Figure 2). Households with more income
are much more likely to be charitable givers, and households with less
income are much less likely to be charitable givers.
THE COMMITTED FEW
But how then does this participation trend reconcile with our U-shaped
income-giving profile? How can lower income households be less likely to
make charitable gifts and still give a higher proportion of total income to
charity? Rather than reflecting typical household behavior within a segment,
the U shape results from a small segment of highly committed donors. Let
us define a committed donor household as one that gives 10% or more of
its after-tax income to charitable causes. In our sample, these committed
donors represent only about 5% of all households. However, as Figure 3
reflects, if we remove these households from our sample, the U shape completely disappears from the income-giving profile for the remaining 95% of
households.
The source of our U shape, then, is not the behavior of 95% of households
but the substantial impact of the committed 5%. It is the unequal distribution
of these highly committed donors that produces our U-shaped profile.
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Charitable Giving Profile
229
80%
70%
60%
50%
40%
30%
20%
10%
+
9
$1
50
,0
00
99
9
-$
14
9,
9
99
9,
00
,0
00
$1
0,
00
0-
$9
9,
99
$8
0$9
00
$7
0-
0,
$8
0,
00
00
$7
99
9
9
9,
9,
99
$6
0-
9,
$5
$6
00
0,
0,
0-
000
$5
99
9
9
9
99
9,
$4
99
9
$4
0,
0-
$3
9,
99
9
9,
0,
00
$2
0$3
0,
00
$1
0$2
00
0,
$1
un
de
r$
10
9,
,0
99
00
0%
Percentage of Households Making Charitable Gifts
Figure 2. Percentage of Households Making Charitable Gifts (Consumer Expenditure
Survey 1998-2001)
Note: N = 16,442.
As Table 5 reflects, although lower income households are less likely to
participate in charitable giving, they are nevertheless more likely to be committed donors—giving 10% or more of after-tax income. The participation
and commitment trends operate in an almost precisely contrary fashion. Yet
even given the fact that lower income households are less likely to be charitable donors, they are still more likely to be committed charitable donors.
Figure 4 shows how the proportion of households classified as committed
donors drops as income rises over a wide range of income segments.
Our results reinforce and extend Auten et al. (2000), who find that when
only the top 5% of itemizing charitable givers are examined in each income
category, a very strong U-shaped profile appears, with the highest percentages of income being given at the low and high ends. The present study adds
strength to Auten et al., whose results were based exclusively on itemizing
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
230
James, Sharpe
5.0%
4.5%
4.0%
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
0.5%
+
00
,0
50
$1
00
,0
00
-$
14
9,
9,
$9
0-
99
99
9
9
9
9
99
9,
$8
0-
00
0,
$9
$1
9
99
9,
$7
0$8
0,
00
9
99
9,
$6
0$7
0,
00
9
99
9,
$5
0$6
0,
00
9
99
9,
$4
00,
$5
00
00
9
99
$4
0,
0-
$3
9,
99
00
$2
9,
99
00
$3
0,
0-
9,
$2
0,
000
0,
$1
un
de
r$
$1
10
,0
00
9
0.0%
All Households
Excluding Highly Committed Households
(Giving 10+% of Income)
Figure 3. Percentage of Income Gifted in Each Income Increment (Consumer Expenditure
Survey 1998-2001)
Note: All households N = 16,442; households excluding 10% or more donors N = 15,607.
taxpayers. As fewer than 10% of taxpayers in the bottom two income quintiles are itemizers, we might expect that selection bias would cause relatively
high-proportion donors to be overrepresented in this lower income group
(e.g., a low-income individual giving only 2% of income would not have reason to itemize—and hence would be excluded from the sample—whereas a
high-income individual giving 2% would have reason to itemize). This standard deduction–based selection could, by itself, explain the lower income
results obtained in Auten et al. However, such a selection argument does not
apply to the CE data used here.
The dramatic influence of a small, committed core of charitable givers is
consistent with the “small world” hypothesis indicating that a large portion
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
231
Note: N = 16,442.
Less than $10,000
$10,000-$19,999
$20,000-$29,999
$30,000-$39,999
$40,000-$49,999
$50,000-$59,999
$60,000-$69,999
$70,000-$79,999
$80,000-$89,999
$90,000-$99,999
$100,000-$150,000
$150,000+
All
N
Household
Income
34
39
45
49
56
58
60
65
68
67
68
75
50
16,442
Percentage of
Households
Making Any
Charitable Gift
11.56
6.40
4.83
4.25
3.61
2.58
2.31
3.03
2.86
2.02
2.86
2.97
5.08
16,442
Percentage of
All Households
Giving 10% or
More in
Charitable
Gifts
34.39
16.44
10.82
8.65
6.50
4.47
3.83
4.66
4.20
3.03
4.23
3.97
10.16
8,215
Percentage of
Donor
Households
Giving 10% or
More in
Charitable
Gifts
4.55
2.37
2.14
1.59
1.66
1.36
1.44
1.71
1.46
1.34
1.51
2.17
1.75
16,442
Percentage of
Income Gifted
(Overall)
Table 5. Households Making Charitable Gifts (1998-2001 Consumer Expenditure Survey)
0.85
0.93
0.92
0.98
1.00
0.98
1.06
1.01
1.08
1.05
1.03
1.22
1.03
15,607
Percentage of
Income Gifted
(Excluding 10%
or More
Givers)
232
James, Sharpe
35%
30%
25%
20%
15%
10%
5%
00
99
$1
50
,0
9,
14
-$
+
9
9
99
9
00
$1
00
,0
0,
00
0-
$9
9,
99
9,
99
$9
$8
0,
00
0-
$8
9,
99
$7
0-
9,
00
$6
00,
$7
9
9
9
99
9
9,
00
0,
$6
0,
00
0-
$5
9,
99
$4
$5
00
0-
9,
$3
00,
$4
00
0,
$3
99
9
9
99
99
9,
$2
0-
9,
0,
00
$1
0$2
00
0,
$1
un
de
r$
10
,0
00
9
0%
Proportion of All Households Giving 10+% to Charity
Proportion of Donor Households Giving 10+% to Charity
Figure 4. Households Giving 10% or More of After-Tax Income to Charitable Organizations
(Consumer Expenditure Survey 1998-2001)
of prosocial behaviors are located within a small proportion of the population (Kochen, 1989). Reed and Selbee (2001) note a similar committed core in
that 9% of Canadian adults account for 80% of volunteering, whereas 18% of
the population are responsible for 80% of all charitable gifts. Given the
importance of a small, committed donor household segment in our sample,
it is useful to consider the factors that are associated with such a high level
of charitable participation.
Most commonly, the presence of the lower-income, highly charitable
households that drive the left side of our U-shaped profile has been
explained as the result of conservative religious sect affiliation by the poor.
Although the “sect effect” is clearly a factor—lower income individuals are
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Charitable Giving Profile
233
more likely to be sect members, and sect members are more likely to give a
high proportion of their income to their church (Iannaccone, 1988)—we find
an additional, powerful factor at work here.
Our lower-income committed (those giving 10% or more of after-tax
income to charity) donor households are in fact much wealthier than other
households in their same income segments. (We here use liquid assets,
defined as savings accounts, checking accounts, and marketable securities,
as a proxy for wealth.) This wealth disparity is true for committed donors
across the income spectrum. Committed donors in our sample are generally
2 to 3 times wealthier than other households in their income segment. But
within the lowest income segments, this disparity widens. Committed
donors in the lower income segments hold 4 to 17 times more liquid wealth
than other households within their same income group. The seemingly
extraordinary commitment of low-income households giving 10% or more
of their income to charitable causes becomes somewhat less compelling after
accounting for this dramatic wealth difference.
It appears that the prevalence of these low-income, high-asset committed
donor households may be explained in part by the influence of retired
households. Retired households living off of accumulated assets would be
much more likely to have lower income and higher assets than comparable
working households. In fact, highly committed retirement-aged households
in our sample on average hold about 2.7 times more liquid assets than highly
committed, non-retirement-aged households within the same income segment. Clearly, retirement-aged households are an important segment of the
committed donor subsample. In fact, those aged 65 and older are found in
43% of committed donor households but in only 23% of other households.
Most interestingly for our examination of low-income committed donors,
we find that retirement-aged households are more prevalent in the lower
income ranks of committed donors than in the higher-income committed
donor segments. For example, retirement-aged households (those containing members aged 65 and older) constitute 48% of all committed donors
earning less than $30,000 in our sample but only 15% of committed donors
earning $80,000 or more. Thus, it appears that the strong presence of retirees
with lower income and higher assets is a substantial factor in explaining the
existence of our U-shaped income-giving profile. Savoie and Havens (1998)
find similar characteristics of low-income, high-giving donors in the 1995
SCF, where such donors are more likely to be older, wealthier, married, and
retired. The sect-affiliation explanation remains a viable component of this
low-income, high-giving phenomenon, but it appears that we should supplement this standard explanation with the undeniable impact of lowincome, high-asset retired households.
A second distinction of the highly committed donor group may help to
enlighten our understanding of the selection bias consequence in the previously referenced research. Although highly committed households are only
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
234
James, Sharpe
slightly more likely to be married than other households (58% vs. 53%), this
ratio appears to vary systematically with the income increment under examination. Table 6 shows how, at the lowest income level, more than twice
as many committed-donor households are married as compared to other
households, whereas at the highest income level, the proportion of committeddonor households that are married is slightly less than other households.
This trend directly relates to the selection bias issue because eliminating
observations provided by non–heads of households disproportionately
affects married households. (A single-person household or single-parentwith-minor-children household cannot be eliminated in this selection
process.) The argument can be made that randomly eliminating married
households is acceptable because the same process is applied to all income
increments. Unfortunately, the proportion of committed households that are
married as compared to the proportion of noncommitted households that
are married varies systematically with each income increment. Thus, in the
lowest income segment, randomly removing married households will eliminate highly committed households at a rate 2.2 times greater than noncommitted households. But at higher income levels, the identical process will
eliminate households from both committed and noncommitted segments at
about the same rate. This reality helps to provide a partial explanation for
the dramatic biasing effects of the selection process used in the previous
research challenging the U-shaped income-giving profile. For example, we
find an even stronger biasing trend in the previously mentioned SCF data. In
that data set, randomly removing married households will eliminate highly
committed households at a rate 3.8 times faster than other households in the
lowest income segment but will eliminate highly committed households at a
lower rate than other households in the $60,000 to $100,000 income range.
LIMITATIONS OF THE DATA
The CE collects information on household income and expenditures
through quarterly interviews. Households participate for 5 consecutive
quarters. The charitable giving information is requested in the 5th quarter.
Thus, the typical household in our data set has participated in the survey for
15 months. The U.S. government uses the CE as the basis for the Consumer Price
Index—the most widely used measure of inflation. The CE uses a “national
probability sample of households designed to represent the total noninstitutional civilian population” (U.S. Department of Labor, Bureau of Labor Statistics,
2005, p. 7). The design of the CE, both in collecting data through quarterly inperson interviews and in selecting a nationally representative sample, can lead
to concerns over reported charitable giving levels and reported income levels.
Concerns in reported charitable giving levels stem in part from the lack of
oversampling of very wealthy households. A significant proportion of total
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
235
$70,222.84
$73,032.56
$77,913.71
$59,167.48
$159,506.66
$56,635.91
$89,645.48
$203,556.52
$86,840.94
$229,733.22
$292,539.74
$938,323.45
$99,984.24
835
After-Tax Income
Less than $10,000
$10,000-$19,999
$20,000-$29,999
$30,000-$39,999
$40,000-$49,999
$50,000-$59,999
$60,000-$69,999
$70,000-$79,999
$80,000-$89,999
$90,000-$99,999
$100,000-$150,000
$150,000+
Overall
n
Note: N = 16,442.
Committed
Donor
Households:
Liquid Assets
$4,119.69
$13,311.46
$17,927.73
$22,929.32
$36,072.80
$33,237.93
$37,861.69
$48,327.35
$74,310.73
$81,513.49
$93,049.64
$276,115.62
$35,241.90
15,607
Other
Households:
Liquid Assets
34
46
63
70
75
82
87
83
88
100
91
73
56
835
Committed Donor
Households:
Percentage
Married
16
31
47
54
63
72
75
80
81
87
89
89
53
15,607
Other
Households:
Percentage
Married
Table 6. Comparing Committed Donor Households (Giving 10% or More of After-Tax Income to Charity)
to Other Households (1998-2001 Consumer Expenditure Survey)
2.2:1
1.5:1
1.3:1
1.3:1
1.2:1
1.1:1
1.2:1
1.0:1
1.1:1
1.1:1
1.0:1
0.8:1
1.1:1
Ratio of %
Married
Committed
Donors to %
Married Other
Households
236
James, Sharpe
charitable giving dollars are given by the wealthiest in the nation. For
example, in the 2005 Giving USA, a single gift represented approximately
2% of the entire nation’s individual charitable giving for the year. As a data
source to estimate aggregate charitable giving, the CE is therefore lacking.
Other data sources, such as the tax returns of itemizing taxpayers (on which
the Giving USA estimates are largely based), are superior instruments for
this goal. With the absence of strong information from the very wealthy, CE
aggregate giving levels will not project to the full national totals. We rely on
the CE in this case largely because the households that are of most interest
are those in the lower and moderate income levels. Itemized tax return data,
for example, are of little use in examining these households, as more than
90% of taxpayers in the first two income quintiles choose the standard
deduction (Congressional Budget Office, 2002).
In addition, the data collection methodology may lead to underreporting
of income. Although the households in this study have been questioned
about 18 different categories of income for 5 consecutive quarters, some
household members may remain hesitant to fully disclose income. Our limitation to complete income reporters eliminates those households unwilling
to make any income disclosures but does not eliminate those choosing to make
partially accurate disclosures. One alternative approach is to measure household expenditures, rather than income. For lower-income, higher-asset, retirement-aged households, we would expect this measurement to return very
different results. Expenditures should exceed income for these households,
reflecting the normal process of asset consumption in retirement. As we posit
these households to be a major driver of the income-giving profile shape, this
alternative measure would give different results. In fact, using household
expenditures instead of income with the present data still produces U-shaped
coefficients in the quadratic but dramatically increases the relative size of the
standard errors.2
Although the income data-collection process for the CE is subject to various
failings from the veracity and memory of reporting households, it appears to
be a stronger methodology than that used in some other studies containing
nontruncated charitable data from representative low- and moderate-income
households. Income levels in the Giving and Volunteering in the United States
surveys used in Schervish and Havens’s (1995a, 1998) articles came from
one question asked one time over the phone (Hodgkinson & Weitzman, 1990,
p. 284). Respondents did not provide dollar amounts but rather selected an
income increment category. The 1996 GSS used in Schervish and Havens’s
(2001) article contained one question regarding household income and one
question regarding respondent income (Davis, Smith, & Marsden, 2005).
Again, respondents did not provide dollar amounts but rather selected an
income increment category. For all of its faults, the CE remains as “the only
U.S. government survey that relates expenditures of consumers in the United
States to demographic characteristics” (Paulin & Ferraro, 1994, p. 23).
Therefore, it continues to be a useful source of information.
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
Charitable Giving Profile
237
CONCLUSION
The U-shaped income-giving profile, where the average percentage of
household income given to charity is higher in the lowest and the highest
income segments, is a valid description of charitable behavior as reflected in
an analysis of this national data set of 16,442 households. The U shape is confirmed by visual inspection and by least squares regression of a quadratic
model for both common income segments and whole sample income percentiles. Previous findings contradicting the U-shaped profile are shown to
suffer from selection bias that systematically deflates lower-income giving
levels. Although the U-shaped income-giving profile is a valid description,
it does not reflect typical household behavior within these income segments.
Rather, the U shape is driven by the highly committed 5% of households that
give at least 10% of their household after-tax income to charitable causes.
These committed few include a high percentage of retirement-aged individuals, and as a group, the committed few are dramatically wealthier than
other households within the same income segments. In addition to the traditional “sect effect” explanation for low-income donors who give a high
percentage of income, it is important to recognize a wealth effect as a major
contributing factor in this behavior.
Notes
1. A similar effect from the selection process exists in the 1989 Survey of Consumer Finances
data used in Schervish and Havens’s (1998) article, where the lowest income category’s average
gifts are reduced, whereas midrange income levels are increased by as much as 34%.
2. Using expenditure percentiles as observation points, and increment ratios of charitable giving
to expenditures as the dependent variable in an ordinary least squares regression produces an intercept of 0.07869 (0.00424), an expenditure coefficient of –6.07054E-7 (5.412171E-7), and an expenditure
squared coefficient of 2.01969E-11 (1.114E-11). Replacing the original 12 income levels with identical
expenditure levels and using the ratio of charitable giving to expenditures as the dependent variable
produces an intercept of 0.01902 (0.00163), an expenditure coefficient of –2.29197E-8 (3.769127E-8),
and an expenditure squared coefficient of 3.15115E-13 (1.57295E-13).
References
AAFRC Trust For Philanthropy. (2005). Giving USA 2005: The annual report on philanthropy for the
year 2005. Glenview, IL: Author.
Andreoni, J., & Scholz, J. K. (1998). An econometric analysis of charitable giving with interdependent preferences. Economic Inquiry, 36, 410-428.
Auten, G. E., Clotfelter, C. T., & Schmalbeck, R. L. (2000). Taxes and philanthropy among
the wealthy. In J. Slemrod (Ed.), Does atlas shrug? The economic consequences of taxing the rich
(pp. 392-424). Cambridge, MA: Harvard University Press.
Bradley, R., Holden, S., & McClelland, R. (2005). A robust estimation of the effects of taxation on
charitable contributions. Contemporary Economic Policy, 23(4), 545-554.
Brooks, A. C. (2002). Welfare receipt and private charity. Public Budgeting and Financing, 22(3),
101-114.
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012
238
James, Sharpe
Brooks, A. C. (2004). The effects of public policy on private charity. Administration & Society, 36,
166-185.
Clotfelter, C. T., & Steuerle, C. E. (1981). Charitable contributions. In H. Aaron & J. Pechman
(Eds.), How taxes affect economic behavior (pp. 403-437). Washington, DC: Brookings
Institution.
Congressional Budget Office. (2002, December). Effects of allowing nonitemizers to deduct charitable contributions. Washington, DC: Author.
Davis, J. A., Smith, T. W., & Marsden, P. V. (2005). General social surveys, 1972-2004 (cumulative
file) [Computer file]. Chicago: National Opinion Research Center.
Hodgkinson, V. A., & Weitzman, M. S. (1990). Giving and volunteering in the United States:
Findings from a national survey. Washington, DC: Independent Sector.
Hrung, W. B. (2004). After-life consumption and charitable giving. American Journal of Economics
and Sociology, 63(3), 731-745.
Iannaccone, L. (1988). A formal model of church and sect. American Journal of Sociology,
94(Suppl.), S241-S268.
Kochen, M. (1989). The small world. Norwood, NJ: Ablex.
Ostrower, F. (1995). Why the wealthy give: The culture of elite philanthropy. Princeton, NJ: Princeton
University Press.
Paulin, G. D., & Ferraro, D. L. (1994). Imputing income in the Consumer Expenditure Survey.
Monthly Labor Review, 117(12), 23-32.
Presser, H. B. (1998). Decapitating the US Census Bureau’s “head of household”: Feminist mobilization in the 1970s. Feminist Economics, 4(3), 145-158.
Reece, W. S. (1979). Charitable contributions: New evidence on household behavior. American
Economic Review, 69, 142-151.
Reece, W. S., & Zieschang, K. (1985). Consistent estimation of the impact of tax deductibility on
the level of charitable contributions. Econometrica, 53(2), 271-294.
Reed, P. B., & Selbee, L. K. (2001). The civic core in Canada: Disproportionality in charitable
giving, volunteering, and civic participation. Nonprofit and Voluntary Sector Quarterly, 30(4),
761-780.
Savoie, A. J., & Havens, J. J. (1998, November). The high giving poor: Who are the low income people
who make high contributions? Presented at the 1998 annual meeting of the Association for
Research on Nonprofit Organizations and Voluntary Action, Seattle, WA.
Schervish, P. G., & Havens, J. J. (1995a). Do the poor pay more? Is the U-shaped curve correct?
Nonprofit and Voluntary Sector Quarterly, 24(1), 79-90.
Schervish, P. G., & Havens, J. J. (1995b). Explaining the curve in the U-shaped curve. Voluntas,
6, 203–225.
Schervish, P. G., & Havens, J. J. (1998). Money and magnanimity: New findings on the distribution of income, wealth, and philanthropy. Nonprofit Management & Leadership, 8(4), 421-434.
Schervish, P. G., & Havens, J. J. (2001). Wealth and the commonwealth: New findings on wherewithal and philanthropy. Nonprofit and Voluntary Sector Quarterly, 30(1), 5-25.
U.S. Department of Labor, Bureau of Labor Statistics. (2005). Consumer Expenditure Survey, 2001:
Interview survey and detailed expenditure files, ICPSR release [Computer file]. Washington, DC:
U.S. Department of Labor, Bureau of Labor Statistics [Producer]. Ann Arbor, MI: InterUniversity Consortium for Political and Social Research [Distributor].
Van Slyke, D. M., & Brooks, A. C. (2005). Why do people give? New evidence and strategies for
nonprofit managers. American Review of Public Administration, 35(3), 199-222.
Russell N. James III, JD, PhD, is an assistant professor in the Department of Housing & Consumer
Economics at the University of Georgia. His research interests include charitable giving, nonprofit management, and affordable housing.
Deanna L. Sharpe, PhD, is an associate professor in the Department of Personal Financial Planning at
the University of Missouri–Columbia. Her research interests include consumer expenditure patterns and
later-life economic issues and policy.
Downloaded from nvs.sagepub.com at University of Missouri-Columbia on February 15, 2012