Abstract
How does receiving public assistance affect an individual’s charitable giving and volunteering? Using the 1994 to 2005 Panel Study of Income Dynamics (PSID) and 2005 Center on Philanthropy Panel Study (COPPS) data, we use a series of comparison group and propensity score matching approaches to overcome sticky issues of selection bias and to explore this question. We find that neither current public assistance receipt nor the amount of public assistance income has any effect on an individual’s charitable contributions of time and money. Cumulative past public assistance appears to suppress charitable giving but not volunteerism.
Keywords
Much scholarship considers the factors that motivate one to donate time or money to charitable causes, but very little scholarship considers the public assistance recipients who appear least able to make such contributions. As the U.S. economy stresses more families with unemployment, causing more widespread access to the social safety net, one wonders whether those recipients only take from the state or whether they also give back to society. Polling data shows that recipients of public assistance and the poor share the same values as the rest of society. So it seems that there is no reason to believe they might behave differently in terms of their interest in volunteering time or donating money. Nevertheless, income is a major constraint on one’s ability to donate to charity, and it is income that public assistance recipients and other poor segments of our population lack.
Some research on charitable giving among welfare recipients suggests that welfare receipt tends to reduce the levels of charitable giving (e.g., Brooks, 2002, 2004); but more recent work shows that this reduction might be offset by higher levels of volunteering (Guo & Peck, 2009). One of the main challenges in this line of research is the problem of selection bias. That is, those who use public assistance are different—in ways that affect their outcomes of interest (i.e., donations of money and time)—from those otherwise, seemingly comparable individuals who do not use welfare. This selection mechanism is not well understood or modeled due to the kind of data commonly available.
Building on prior work, this article adds to the small but important literature on the role of public assistance in helping or hindering welfare recipients in making charitable contributions. We explore the issues of public assistance recipients’ donations of their time and (quite limited) money for charitable purposes. We use the 1994 through 2005 waves of the Panel Study of Income Dynamics (PSID) as well as the 2005 wave of the Center on Philanthropy Panel Study (COPPS), which includes data on charitable giving and volunteering as a module added to the PSID. We address the challenge of selection bias by using two methodological innovations. We first use discrete comparison groups to reduce the amount of bias in impact estimates from the selection process associated with public assistance use. We then introduce a propensity score analysis where we use propensity score matching to create a comparison group that further reduces selection bias to produce more reliable results. This approach is also useful as an early application of propensity score methods used in this field. In brief, we find that neither current public assistance receipt nor the amount of public assistance income has any effect on an individual’s charitable contributions of time and money; however, cumulative prior public assistance appears to suppress charitable activity.
This article is presented in four sections. It begins with discussion of the effects of both earned and public assistance income on people’s charitable behaviors. Second is the discussion of the methodology, including data source and measures, and the selection bias problem that challenges this research. We describe our use of discrete and propensity-score-based comparison groups as strategies to minimize selection bias in impact estimates. The third section presents and discusses the analysis results with respect to prior research and in consideration of methodological lessons. We conclude with a discussion of the study’s contributions and limitation, as well as implications for future research and policy.
Background
Prior research examines a variety of factors associated with individuals’ donations of money and time. Here we discuss findings from that research, in particular what we know about the relationships between public assistance use, as well as demographic traits and charitable activity.
Charitable Activity and Its Correlates
Personal characteristics
Several key personal characteristics are associated with charitable activity—including age, education, marital status, family structure, religion, health, race and ethnicity, and residential location—each of which is summarized below. To begin, research documents that age may be “the variable most consistently related to giving” (Clotfelter, 1997, p. 17), with donations increasing with one’s age (e.g., Arai, 2004; Brooks, 2002; Van Slyke & Brooks, 2005; Wilson & Musick, 1997). Education is also consistently and positively correlated with charitable giving (e.g., Feldstein & Clotfelter, 1976; Jencks, 1987; Morgan, Dye, & Hybels, 1977; Van Slyke & Brooks, 2005; Steinberg & Wilhelm, 2005) and volunteering (Choi, 2003; McPherson & Rotolo, 1996). Marital status, in contrast, has a less clear association with charitable giving, but research shows it matters. Some have found a positive relationship (e.g., Andreoni, Brown, & Rischall, 2003; Choi, 2003; Jencks, 1987; Rooney, Mesch, Chin, & Steinberg, 2005; Van Slyke & Brooks, 2005; Wilson, 2000), explained by the added resources and social connections that married people have, but others have found no difference in the volunteering that married and unmarried people do, controlling for other factors (Fischer, Mueller, & Vooper, 1991; Herzog & Morgan, 1993; Musick & Wilson, 2008). Parental status also affects charitable activity: The presence of very young children might suppress volunteering, while parents of older children might volunteer in activities related to their children, such as at schools (Day & Devlin, 1996). Another clear link exists between religious affiliation and practice and charitable activity (Wilson & Musick, 1997), with being Christian (e.g., Van Slyke & Brooks, 2005) and more frequent service attendance (e.g., Steinberg & Wilhelm, 2005) being associated with greater giving of both time and money. Health status also influences volunteer activity, with poor health reducing one’s ability to volunteer (Caputo, 1997) and the amount of time one volunteers (Gallagher, 1994; Wilson & Musick, 1997), particularly among older adults (Choi, 2003). Some research finds that race and ethnicity affect charitable activity, with African American and Hispanic families being less likely to donate or volunteer, or to donate less money or fewer hours, than Whites (Hodgkinson & Weitzman, 1996); but others find no effect (e.g., O’Neill & Roberts, 2000; Rooney et al., 2005; R. Steinberg & Wilhelm, 2005). Finally, residential location is associated with charitable activity, with those in rural areas more likely to volunteer than city and suburban dwellers (Musick & Wilson, 2008).
Income
In addition to these personal traits, two additional correlates of charitable activity demand attention: income and public policy. Most studies of the relationship between income and charity use regression models to focus on the income elasticity of giving. 1 Findings show that income elasticity is statistically significant and positive (see Clotfelter & Steuerle, 1981; McClelland & Brooks, 2004; R. Steinberg, 1990). Income is a powerful predictor, and therefore it is important to control for income in any analysis of the determinants of charitable giving. The relationship between income and volunteering is less clear. In general, the relationship is positive, with those donating money being more likely also to volunteer their time, but it is unclear whether this holds across the whole income distribution (Musick & Wilson, 2008). Furthermore, some evidence suggests that volunteering actually can increase an individual’s income (Day & Devlin, 1998).
Public policy
Substantial research focuses on the effect of public policy on charitable giving. This research falls into two related streams. The first stream examines the relationship between government spending and charitable giving (e.g., Abrams & Schmitz, 1984; Brooks, 2000; Kingma, 1989; Roberts, 1984; Schiff, 1985). Roberts (1984), Kingma (1989), and Brooks (2000) conclude that government spending tends to displace private charitable giving. On the other hand, Schiff (1985) finds that the effect of government spending on charitable giving depends on the type of government spending: Local funding displaces private donations to the nonprofit sector, whereas state funding leverages private donations. With few exceptions (e.g., Day & Devlin, 1996; Menchik & Weisbrod, 1987), relatively little research considers the relationship between government spending and volunteer work. Day and Devlin (1996) find that the level of government spending influences the decision to volunteer but does not influence the number of hours volunteered. They also find that the nature of this relationship varies with the particular types of government expenditures (see also Menchik & Weisbrod, 1987).
The second stream of policy research considers incentives designed to encourage charitable giving and volunteering. Relevant policies include the deduction of charitable donations from personal income, the tax-exempt status of various investments, and similar policies where charitable donations reduce one’s tax burden (e.g., Clotfelter, 1985; Karlan & List, 2007; Peloza & Steel, 2005; Randolph, 1995). Changes in such policies lead to change in the “price of giving” (i.e., what it costs for a household to donate an additional dollar to charity), which in turn can lead to change in the level of charitable giving: The level of giving decreases as the price of giving increases. The price of giving depends on whether a household itemizes deductions in computing personal income tax. Because households are permitted to itemize charitable deductions on their federal and most state personal income tax returns, it costs less than a dollar for each dollar donated. No such advantage exists for those who do not itemize. Therefore, tax itemizers have incentives to give a higher level than non-itemizers. We have used tax itemization as a proxy for the reduced price of giving and find that tax itemizers donate more money than those who do not itemize on their tax returns, all else equal (Guo & Peck, 2009).
The area of public policy that this article scrutinizes is the social safety net and its role in inhibiting or encouraging charitable activity of individuals who receive income support from government. This arena has been little studied, with our recent work (Guo & Peck, 2009; Peck, D’Attoma, Camillo, & Guo, 2012) and that of Brooks (2004, 2002) being the most direct tests of these policy effects. Brooks’ (2002) pioneering work finds that public assistance income functions differently from earned income in stimulating charitable behavior: Whereas a 10% increase in earned income is associated with an 8% rise in charitable giving, the same percentage increase in welfare income was associated with a 1.4% drop in charitable giving (see also Brooks, 2004).
While Brooks’ research considers charitable giving only in terms of money, our work considers the donation of both money and time and concludes that receiving public assistance, all else equal, may suppress money donations but may actually increase one’s donations of time for charitable purposes (Guo & Peck, 2009), at least among some subsets of this heterogeneous population (Peck et al., 2012). The evidence supports the observation that charitable contributions by the poor are more “likely to take a nonpecuniary form” (Brooks, 2002, p. 111). Although some have documented a positive correlation between donating money and volunteering (e.g., Hodgkinson & Weitzman, 1996; Van Slyke & Brooks, 2005), it may be that those with low incomes consider that donations of money and donations of time are substitutes rather than complements. Jencks (1987) and Duncan (1999), for example, find that when an individual’s likelihood of giving money declines, his or her contribution of time increases, on average. Reliance on public assistance represents a severe income constraint that may rule out charitable giving, but, if giving and volunteering are substitutes, then volunteering would be more likely.
Prior research also highlights evidence of and explanations for public policy’s effects on volunteering. For example, the G.I. Bill’s educational provisions have been documented to increase veterans’ voluntary participation in civic organizations and politics (Mettler, 2002). This potential positive association between public assistance receipt and volunteering may also be interpreted from the perspective of human capital and career development (e.g., Menchik & Weisbrod, 1987). Combined with job search (or other traditional welfare-to-work activities), volunteering can help welfare recipients gain social and economic independence by empowering them in terms of increasing self-efficacy and developing critical awareness (Cohen, 2009) and by and by providing opportunities for learning-specific work-related skills necessary to succeed in regular employment (Herr, Wagner, & Halpern, 1996). In addition, welfare recipients can also benefit from the signaling effect of volunteering: Their volunteer experience can signal the presence of certain skills or abilities to future employers (Schiff, 1990; Ziemek, 2006).
Research Question
We are particularly interested in the role of public assistance use in encouraging or inhibiting charitable activity, and so we simply ask: How does public assistance use affect charitable activity? Given uncertainty in findings from prior research, we do not pose a directional hypothesis but note that public assistance might have a positive, negative, or no effect on charitable activity; and it is our charge with this research to provide evidence on the effect’s directionality and magnitude.
Methodology
This section details data source and measures along with the analytic challenges we face and the responding approach we follow to answer the research question.
Data and Measures
We use the COPPS supplement to the PSID. In particular, we merge some variables from the 1994 through 2005 panels of the PSID to the 2005 COPPS supplement. The COPPS collects data on the amount of money and number of hours donated to several charitable purposes: religious, combined funds, basic needs, health, education, youth and family services, the arts, neighborhoods, the environment, and international aid. In addition to these key indicators, the data set contains information about people’s use of public assistance, which allows us to test the extent to which public assistance use contributes to people’s charitable activity, all else equal. Our sample in particular is the 7,822 households that comprise the 2005 COPPS.
From the raw COPPS data, we compute two dependent variables: the Amount Donated and the Hours Volunteered. The amount donated is simply the sum of households’ contributions made in 2004 to a variety of charitable sources and is measured in dollars. 2 Likewise, the number of hours volunteered is the sum of the head of households’ time donated to a variety of secular and religious charitable organizations in 2004. Our descriptive statistics include binary versions of these variables as well as the percent of the population that donates money or time at all.
Our specific measures of income combine specific sources of 2004 household income as reported by survey respondents. Earned Income is the sum of the household head and spouse’s labor income (e.g., wage, salaries, etc.). Our prior work operationalized “public assistance use” as receiving cash public assistance and/or food stamp benefits (Guo & Peck, 2009), and we use that same approach here: Annual Public Assistance Income is the sum of the household head and spouse income from cash public assistance (Temporary Assistance to Needy Families [TANF]), General Assistance (GA), and food stamps (Supplemental Nutrition Assistance Program [SNAP]). We feel justified in using a broader, inclusive conception and definition of public assistance (not just cash assistance), because substantial overlap exists between cash assistance and food stamps use. While some individuals receive only food stamps and not cash assistance, they may still be considered as at least somewhat reliant on the state for support and would therefore fit the same theoretical model as we have conceived it. Moreover, including food stamp recipients also increases the number of observations in our public assistance recipient subsample, thereby increasing our ability to detect effects, if they exist. In brief, our definition of public assistance receipt includes those who receive cash public assistance transfers or near-cash (food stamps) assistance; here, we do not consider those who are enrolled in Medicaid as “recipients” but instead focus on the primary cash- (and near-cash-) based supports as they are most commonly seen as the base social safety net programs. We compute one additional public assistance variable: Prior Public Assistance Use identifies the number of years between 1993 and 2002 (with a total of seven possible) that a household had income from cash public assistance (TANF and GA) or food stamps.
We also report the number of years of Prior Public Housing Residence between 1993 and 2002, and we report Current Public Housing Residence as whether the household reports living in public housing in 2004. We include these public housing variables as a way to control for other measures of disadvantage that might be associated with both welfare use and the ability to donate one’s money or time to charitable causes.
To capture the possible effect of the price of giving, we include a binary variable for Tax Itemizer, which equals 1 if the household head itemized deductions in his or her tax return (and 0 otherwise). While the price of giving one dollar to charity is one dollar for taxpayers who do not itemize deductions in their tax returns, for itemizers the price of giving is less than a dollar, as they receive a “rebate” equal in value to their deductible contributions times the applicable marginal tax rate.
The general household characteristics are straightforward and capture the traits that prior research identifies as associated with charitable activity: age, sex, number of children, marital status, race, ethnicity, education, religiosity, urban/rural residence, and health status. Our unit of analysis is the head of household, and some characteristics of the household more broadly are associated with those individuals. Table 1 presents the variables list and summary statistics for the overall sample as well as those who received public assistance in the most recent year, and those who did not.
Variable Descriptives, Overall and by Welfare Status, in Overall Sample.
Note. Rs refers to recipients. Data are weighted and therefore can be interpreted as nationally representative.
Statistically significantly different: *p < .10. **p < .05. ***p < .01.
Analytic Methods
Both dependent measures—Amount Donated and Hours Volunteered—are continuous and include relatively large numbers of observed zeroes. To take into account the fact that giving and volunteering are truncated at zero (i.e., one cannot make a negative donation of money or time), we follow prior literature on charitable giving to estimate our models using Tobit, a censored regression technique (Brooks, 2002; Van Slyke & Brooks, 2005). It should be noted that, while Tobit accounts for the censoring of the dependent variables, it is not robust to either non-normality or heteroskedasiticity, an error structure that charitable giving data tend to have (Rooney, Steinberg, & Schervish, 2001, 2004; K. Steinberg, Rooney, & Chin, 2002). However, a recent study by Wilhelm (2006) provides support for using Tobit with this particular data set, arguing that despite rejection of the underlying assumptions of Tobit at higher levels of statistical significance, the Tobit estimates are numerically close to more robust methods.
As Table 1 makes obvious, important differences exist in the characteristics of those who have received public assistance and those who have not. The fact that there is a significant difference between the contributions of public assistance recipients and non-recipients indicates that those who receive public assistance differ from non-recipients along observed characteristics implies that they differ as well along unobserved characteristics (e.g., motivation, generosity, parents’ charitable giving, upbringing). To deal with this, our prior work examined impacts estimated by making three distinct comparisons between our population of interest and a discretely defined similar comparison group. Our attempt was to reduce the amount of bias in impact estimates that derived from selection processes, primarily the selection process associated with welfare use. In the conclusion to that article, we proposed additional methods as a robustness check of our results. In turn, this article updates prior work with a new wave of data and contributes a propensity score analysis where we use a vector of characteristics to predict welfare use. The resulting propensity score is a scalar that embodies the likelihood of using welfare, both among those who did use welfare and among those who did not. Comparing the charitable activity outcomes for those with similar propensity scores reduces bias relative to making comparisons between less-similar cases. We detail this process next.
Propensity score matching
As observed in Peck (2007), the literature on propensity scores has experienced a recent growth spurt as both non-experimental and experimental evaluations make use of new methodologies to advance evaluation theory, practice, and scholarship (e.g., Luellen, Shadish, & Clark, 2005; Thoemmes, 2009). Early use of propensity scores was in “observational” (non-randomized) studies, those needing to compensate for a lack of a true randomized control group (Rosenbaum & Rubin, 1983). Furthermore, a growing body of “replication” studies compares the results of experiments with other non-experimental impact estimates and suggests the conditions under which other designs and analyses, including propensity score matching, can recreate experimental impacts (e.g., Bifulco, 2012; Cook, Shadish, & Wong, 2008). While they are more hopeful than pessimistic that propensity-score-matched comparison groups can generate unbiased impact estimates, some uncertainty remains, suggesting caution while methodological kinks continue to be worked out. That said, certainly propensity score matching has the potential to reduce bias over coarser methods.
In brief, a propensity score is a number that represents a combination of traits that influence an individual’s likelihood of receiving some treatment or belonging to some group. Rather than controlling for distinct individual characteristics one or two at a time, the propensity score allows a researcher to control for a large combination of traits, commonly used to reduce selection bias as a threat to internal validity. The score can also represent more complex subgroup membership such as being likely to use certain types of program services, being likely to become a long-term public assistance recipient, or being likely to respond to certain types of (financial or behavioral) incentives, for example.
Computing the propensity score is straightforward. In the process of computing a logistic (or probit, or linear probability) model, the modeled, predicted probabilities are saved as the propensity scores for those both in the treatment group and not. In most applications, this allows us to match treatment group cases with likely counterparts from an untreated group. In our application, we use a logistic model to predict public assistance use. Once created, the propensity score is used to match “treated” individuals with their closest untreated counterparts. We then use these matched cases to examine the role of public assistance use in determining charitable giving of money or time. Specifically, the model that we use to predict propensity to use public assistance in 2004 includes the following:
Household (head) characteristics: age, sex, number of children, age of youngest child, marital status, single-female-headed household, race, ethnicity, years of education, high school degree, urban residence, health status;
Economic indicators: amount of labor income, hours worked, spouse’s hours worked, whether working, whether retired, whether disabled; and
Prior/other public assistance use: years of prior public assistance use, current public housing residence, years of prior public housing residence.
With the propensity score estimated, we match each of those who had received public assistance in 2004 with someone who did not receive assistance. After testing a wide variety of matching approaches, we chose the approach that resulted in the best balance between our treatment and comparison groups. This involved a three-to-one comparison-to-treatment match, using the nearest neighbor algorithm and replacing each comparison case in the sample for possible future re-match, if appropriate. Our resulting analytic sample includes the 750 welfare recipients and a comparison group of 912 non-recipients, where 427 of those 912 are duplicated in instances where they are the best match for a treatment case. The computed propensity scores ranged from 0.002 to 0.994 with a mean of 0.354. The scores among the 6,180 cases 3 who had not received assistance in 2004 ranged from 0.000 to 0.968, with a mean of 0.069. With that distribution of scores, the matching process involves assessing the extent to which the resulting matches are “good” ones. The resulting 1,662 cases that we use in our analysis have very similar propensity scores, and their characteristics are summarized in Table 2.
Treatment and Comparison Group Characteristics, After Matching.
Note. Rs refers to recipients.
Statistically significantly different: *p < .10. **p < .05. ***p < .01.
One way to consider balance is to examine the characteristics of the treatment group to counterparts in the matched comparison group. As Table 1 made clear, of the 25 variables used, 23 of them statistically significantly varied (22 of those at the p < .01 level) between the public assistance recipients and non-recipients. In contrast, with our propensity-score-matched sample, only two of the variables differ between the two groups, as shown in Table 2; one of them—public assistance income—differs between the two groups by definition, and the other—prior time on public assistance—is both an offshoot of the main definition and relatively small in magnitude (2 vs. 1.7 years), a point we will return to in discussing. In brief, this suggests that those traits most associated with public assistance use are balanced between treatment and comparison cases in our analysis sample.
With this well-matched sample, we argue that our Tobit analysis of the effect of public assistance use on charitable activity results in minimally biased estimates of policy impact. Our equations are specified as follows:
where Y is our outcome of interest (dollars donated or hours volunteered);
We examine the results comparing a propensity score-matched comparison group to the results from a simple comparison group approach, where we identify comparison cases based on discrete features. Whereas the propensity score-matched sample relies on a single continuous variable as described above, the other comparison groups are created on discrete traits: in this application, we compare current public assistance recipients to all non-recipients; we compare current public assistance recipients to a comparison group of poor households (operationalized as having income falling below 130% of the federal poverty line); and we compare current public assistance recipients with those who had received public assistance in at least one prior year (back to 1993) but not in the current year (see Guo & Peck, 2009). Basically, the progression of our analyses moves from most to least selection bias, with the raw comparison of recipients with non-recipients embodying the most selection bias and the comparison of propensity score-matched recipients and non-recipients embodying the least. The next section reports and discusses the results of these analyses.
Findings and Discussion
In general, fewer variables are statistically significant in the restricted samples of public assistance recipients and similar counterparts than are observed in research examining a wide cross-section of the population. This is to be expected because we aim to isolate the characteristic of interest—welfare receipt—as the possible driver in variation in charitable activity. 4
In prior research, we argued that the comparison groups of poor and of prior recipients served to reduce selection that would bias estimates of the effect of welfare on charitable activity. The first three sets of columns in Tables 3 and 4 are updated versions that report the analysis results after adding the extra year of data available in the 2005 COPPS data. This time-extended analysis is indecisive on the effect of current public assistance on money donations: the estimated effect (relative to non-recipients) is −$196; the estimated effect (relative to the poor) is +$119; and the estimated impact (relative to prior recipients) is no different from 0. The first two of these, opposite in sign, are statistically significant. While the comparable estimated effects on hours donated are negative, none is statistically significant. The tables also show that the greater the number of prior years of public assistance use one has, the lower one’s donations to charity (at least for the comparisons with non-recipients and the poor). Although these comparisons aim to create a contrast of relevance for the analysis, they remain relatively crude. We cannot guarantee that other factors are not influencing the impact estimates.
Effect of Public Assistance Use on Amount Donated.
Note. aPoor is operationalized as having income falling below 130% of the federal poverty line.
The parameter estimates and marginal effects show slightly different levels of statistical significance because the standard deviation vary across different parts of the Tobit distribution.
All the equations control for prior public housing residence (years), current public housing residence, age of household head, age squared, sex of head (female), number of children, marital status (married), African American, Hispanic, education (≤ high school), education (years), Catholic, Jewish, Protestant, other religion, church attendance, location (urban), health status (in good health).
Statistically significant: *p < .10. **p < .05. ***p < .01.
Effect of Public Assistance Use on Hours Volunteered.
Note. aPoor is operationalized as having income falling below 130% of the federal poverty line.
The parameter estimates and marginal effects show slightly different levels of statistical significance because the standard deviation vary across different parts of the Tobit distribution.
All the equations control for prior public housing residence (years), current public housing residence, age of household head, age squared, sex of head (female), number of children, marital status (married), African American, Hispanic, education (≤ high school), education (years), Catholic, Jewish, Protestant, other religion, church attendance, location (urban), health status (in good health).
Statistically significant: *p < .10. **p < .05. ***p < .01.
Our propensity score model generates a well-matched sample in the characteristics that matter to selection into public assistance. As a result, we put greater confidence in these results as minimally influenced by selection. The right-most columns of Tables 3 and 4 lead to the conclusion that current welfare receipt has an insignificant effect on donating both money and time; and added prior receipt suppresses money donations. None of these indicators appears to influence volunteer hours, suggesting that welfare receipt, either currently or in the past, has no effect on volunteering. We can not support the hypothesis that donations of money and donations of time are substitutes. There is no evidence that public assistance recipients volunteer any more than their non-recipient counterparts in this sample as was hinted in prior research. Each additional year of prior public assistance use is associated with $80 less in donations, while holding the other variables in the model constant. Because there is no statistically significant effect of current public assistance use, we conclude that the effect of public assistance is generated by duration of assistance, not simply exposure.
Given that we are comparing methods, we also present here the average value of propensity scores by discretely defined comparison group. As Table 5 shows, the “treatment” group of public assistance recipients has a statistically significantly different average propensity score from two of the three discretely defined subgroups of non-recipients and prior recipients, though recipients’ average score is not statistically different from that of poor households. Likewise, the matched propensity score sample differs from these two groups as well. The treatment group and propensity score-matched sample have statistically similar propensity scores (both averaging 0.450). This suggests that, at least according to our model of the predictors of public assistance use in this sample, the best matched comparison is between public assistance recipients and their counterparts in the propensity-score-matched sample subgroup.
Average Propensity Score, by Discretely Defined Comparison Groups.
Note. aBeing poor is operationalized as having income falling below 130% of the federal poverty line.
Statistically significant difference (p < .05) between subgroup and:
welfare recipients
non-recipients
those in poverty
prior recipients
propensity score-matched (three-to-one, with replacement) group.
The comparison of the treatment group—public assistance recipients—to the discretely defined subgroups certainly holds face validity, and one can understand the ways in which selection bias may remain in force. For example, recipients are quite different from non-recipients (as shown in Table 1), but their characteristics are much more similar when compared with those who are poor or those who have received public assistance in the past (analysis not shown; see Guo & Peck, 2009). That said, the characteristics of the propensity-score-matched sample are the closest to those in the treatment group, suggesting that the results of the analysis that uses that subgroup for comparison are the least biased of our estimates.
While all of the analyses show no effect of current public assistance use on charitable activity, the point estimates vary in the magnitude of the effect of prior years of public assistance receipt on amount donated to charity. The comparison of current recipients with non-recipients shows the largest negative effect, and this is the estimate that is most biased in the sense of selection bias: $380 annually per added prior year on assistance. The other discrete comparisons—with poor households and with prior recipients—have smaller estimated impacts: −$54 and −$59 per added prior year on assistance; whereas the propensity-score-matched sample suggests that the effect of an additional year of prior assistance has the effect of lowering charitable donations $80 per year, ceteris paribus. The range of the effects that attempt to minimize selection bias (−$80 to −$54) is relatively small, compared with the estimated effect on the whole sample (−$380), which we believe unequivocally to be biased. Given that all the estimated effects share the same sign, we can be somewhat more confident in the finding that longer-term prior assistance receipt suppresses charitable donations.
Nevertheless, it is important to note an alternative explanation for the effect of prior public assistance use: The increasing prior years of public assistance use might suggest household vulnerability and long-term depletion of resources. That is, this “prior public assistance use” variable will necessarily pick up the effects of other sources of vulnerability. It would not be surprising if a chronically vulnerable household gave less either because they experienced a depletion of wealth or because they had a fear of future vulnerability and more need to retain assets. This, however, has nothing to do with government assistance. Stated more succinctly, our analysis in no way eliminates the hypothesis that household vulnerability would have both a positive effect on the amount of aid received, and a negative effect on charitable behavior. 5
The story with regard to the effect of public assistance on volunteer hours suggests that welfare receipt (current or past accumulation) make no difference in volunteering. Although the estimates’ coefficients are negative, none of them is statistically significant. Our prior analysis of the 2003 COPPS data showed no effect of current receipt, but an increase in hours associated with an increase in public assistance income (Guo & Peck, 2009). The 2005 COPPS data do not show the same results.
Conclusion
This work examines how public assistance use influences charitable activity. In contrast to some prior work (e.g., Brooks, 2002, 2004), we find that welfare receipt has no suppressing effects on charitable giving and volunteerism, except potentially in cases of long-term prior public assistance experience, but the identification of that effect is questionable. To clarify, neither current public assistance receipt nor the amount of public assistance income has any effect on money and time donations; yet prior duration of public assistance has a negative effect on money donations regardless of whether the household currently receives assistance, though it has no significant effect on volunteer hours. These findings arise while we control for a full array of other relevant variables that influence charitable activity outcomes. The former finding challenges prior research that suggests a negative relationship between public assistance receipt and charitable giving (e.g., Brooks, 2002). 6 The latter finding about the negative effect of prior public assistance use is partially consistent with the notion of a “culture of welfare,” which suggests that long-term welfare dependency may suppress many pro-social attitudes (Brooks, 2002). This observation also seems to echo the proposition that “welfare benefits discourage political involvement by cultivating personal traits of dependence” (Soss, 1999, p. 363). However, the effect of prior public assistance use is likely confounded by other factors such as household vulnerability and long-term depletion of resources which might negatively affect charitable behavior. Moreover, the fact that people do not volunteer less, even when on long-term government assistance, strengthens the case that households that receive government assistance are not less generous (at least in terms of time). Further research is needed in this area.
The data we examine to explore the relationship between public assistance and charitable activity comes from a period of economic expansion. Prior to our current recession, the most recent recessionary period occurred in 2001. It is possible that the increase in volunteer hours associated with public assistance that we observed in our prior work was a function of the 2003 COPPS, measuring 2002 charitable activity, coming on the tail end of that recession. Perhaps people act more charitably during hard economic times. More specifically, when greater segments of society experience economic hardship, those people getting help may volunteer in response. If this is the case, then examining future waves of the COPPS into the current, much more severe recession, may allow testing this hypothesis.
Future studies should also test the mechanisms explaining how welfare receipt affects giving and volunteering. One mechanism discussed earlier is that welfare receipt increases financial dependence and passivity. From this perspective, we would expect welfare receipt to reduce amounts donated and hours volunteered. Another possible mechanism relates to the opportunity cost of volunteering. From this perspective, we would expect the number of volunteer hours to increase as one moves from a full-time job to welfare receipt, but less so when moving from a part-time job to welfare receipt. Along this line, it would also be worthwhile to consider disaggregating the dependent variables into participation and amount variables: The decision to donate time might be considered as distinct from the amount of time donated; and the decision to donate money might be considered distinct from the amount of money donated. This inquiry could reveal the extent to which welfare receipt at all affects the likelihood of volunteering and amount or time donated. These results would inform greater understanding of the possible mechanisms: For example, one would expect opportunity costs to influence volunteer hours more than the decision to volunteer; and one would expect passivity to affect other pro-social behaviors and social activities as well, in addition to giving and volunteering.
From a methodological perspective, this work has value in that it compares two approaches to dealing with selection bias and observes variation in results: discretely defined comparison groups and propensity score analysis. We observe that, while a discrete comparison group approach has demonstrated its value in reducing selection bias, our two specific analyses reduce the amount of bias in impact estimates perhaps more than is optimal. That is, our two more appropriate comparisons—between public assistance recipients and poor household or prior recipients—generate smaller point estimates than the analysis based on the propensity score-matched sample, which generates results that are the least biased of our estimates.
In sum, this work contributes to the small but important body of research that considers whether and how public assistance receipt might affect one’s charitable activity. We conclude that current public assistance recipients behave no differently than their non-recipient counterparts when it comes to donating time and money to charity, but that the cumulative effect of past public assistance use appears to suppress charitable donations. Because of inconsistency in research results regarding the effect of public assistance on volunteering, we propose that future research explore this topic more explicitly, over a longer period of time.
Footnotes
Acknowledgements
We are grateful to the agencies that fund the Panel Study of Income Dynamics (PSID), Atlantic Philanthropies for funding the collection of data in the first four waves of the Center on Philanthropy Panel Study (COPPS), and the Bill and Melinda Gates Foundation for funding the 2007 and 2009 data collection of the Center Panel as well as the dissemination of the 2005 data. We appreciate the research assistance of Andrea Mayo at Arizona State University and Daniel Boada and Will Huguenin at Abt Associates Inc. We also acknowledge participants in the Arizona State University School of Public Affairs Research Colloquium, in the ASU Center for Population Dynamics Colloquium, participants at our Fall 2011 panel at the Research Conference of the American Evaluation Association, and in Abt Associates’ Journal Authors Support Group for their useful input.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.