Partisan Differences in Nonpartisan Activity: The Case of Charitable Giving

Giving USA has been producing annual reports on Americans’ giving patterns for over 60 years and is regularly referred to by respected news outlets such as the Chronicle of Philanthropy (Sandoval 2016). Since 2000, Giving USA’s reports are based on research conducted by the Lilly Family School of Philanthropy at Indiana University.

In 2014, the Giving USA foundation estimated total donations at $358 billion, with $115 billion (32%) going to religious charities (Velasco 2015).

As Brooks writes, “People who favor government income redistribution are significantly less likely to behave charitably than those who do not” (Brooks 2006, p. 55). This view is also sometimes reflected in elite rhetoric. For example, Paul Ryan, the 2012 Republican vice-presidential candidate and now Speaker of the House, stated a Mitt Romney administration would fight poverty by supporting private charities (Achenbach 2012). Similarly, Wall Street Journal reporter John D. McKinnon defended Romney’s low tax burden relative to his personal income, arguing that “Republicans favor a world in which people pay fewer taxes and give more to charity, believing that private spending is more effective than that of the federal government” (McKinnon 2012).

There is also a well-known relationship between giving and actual, as opposed to perceived, income (Bekkers 2015; Eckel and Grossman 2003, 2004; Karlan and List 2006; Bekkers and Wiepking 2011; James and Sharpe 2007; Wiepking 2007; Giving USA 2009; List 2011; Reich and Wimer 2012).

While the American National Election Study has asked about giving in the past, it does not ask about amounts donated, but only whether any donations were made; the ANES also does not distinguish between types of recipient organizations. In an analysis of different methods for asking about donations, Wilhelm (2007) concludes that survey questions asking about dollar amounts, as well as types of organizations, yield more accurate responses.

The SCCBS also included a nationally representative sample of about 3000 respondents. Our results are insensitive to which sample we use, but we rely on the larger sample as it includes more donors.

The SCCBS ideology question reads: “Thinking politically and socially, how would you describe your own general outlook—as being very conservative, moderately conservative, middle-of-the-road, moderately liberal, or very liberal?” By explicitly asking respondents to consider their views on social issues when answering the question, the SCCBS may have measured ideology with error. Consistent with this claim, in the Online Appendix we show that self-identified conservatives in the SCCBS have lower incomes than self-identified liberals, which is the opposite of what we find in our other two surveys.

To ensure that the SCCBS donation measures are comparable with our other two data sources, we convert these intervals to raw amounts by taking the midpoint of the two endpoints for each range, converting the scale to $50, $300, $750, $3000, and $5000.

SSI recruits participants through various online communities, social networks, and website ads. SSI makes efforts to recruit hard-to-reach groups, such as ethnic minorities and seniors. These potential participants are then screened and invited into the panel. When deploying a particular survey, SSI randomly selects panel participants for survey invitations. We did not employ quotas, but instead asked SSI to recruit a target population that matched the (18 and over) census population on education, gender, age, geography, and income. The resulting sample is not a probability sample but is a diverse national sample. Several studies using SSI samples have been published recently in political science (Berinsky et al. 2014; Kam 2012; Malhotra and Margalit 2010; Malhotra et al. 2013).

The first wave of the survey was conducted between October 17 and October 31, 2012. The second wave of the survey was conducted between November 13 and November 27, 2012. 75% of the post-election surveys were completed by November 19, 90% by November 21.

One concern with our measures of giving is that they rely on self-reports, rather than directly observed measures (Dawes et al. 2011; Bolsen et al. 2014). While we lack a direct measure, we note we find similar results when we use state-level measures of giving based on the Panel Study of Income Dynamics, a high-quality survey of American families’ economic resources and activity (Kim and Stafford 2000; Becketti et al. 1988; Duncan and Hill 1985). We present estimates of the relationship between state partisanship and state charitable contributions in the Online Appendix.

We code “leaners” as identifiers with one group or the other, such that only respondents at the midpoint category are coded as moderates/independents.

We specify the effect of income to be linear, but including indicators for each income category gives similar results.

We use state when it is available (SCCBS and SSI) and region when it is not (GSS).

We recode such that +1 equals a preference for less spending, 0 equals a preference for current levels, and −1 equals a preference for more spending.

Each question is originally on a seven-point scale. We first code responses above the midpoint as +1, below the midpoint as −1, and at the midpoint as 0. We then sum responses to the three questions.

Results are robust to specifications where we trim the 99th percentile of the giving variables (to account for outliers) and when using Tobit regressions (to account for the large number of zero donations). More details on our control variables, including how Democrats and Republicans differ across these dimensions, are available in the Online Appendix.

We compute this quantity as follows: first, we use the coefficient estimates to compute predicted values of log giving among a full sample of liberals, with one predicted value for each respondent (i.e., we use the “observed-value” approach for estimating substantive effects, as advocated by Hanmer and Kalkan (2013)). We then take the average predicted value, exponentiate it, and subtract one (given that we added one when we took the log). Call this \(\hat{Y}_0\). We then repeat this procedure, predicting values of log giving for a full sample of conservatives, taking the average predicted value, exponentiating and subtracting one. Call this \(\hat{Y}_1\). We then take the difference between \(\hat{Y}_1\) and \(\hat{Y}_0\). See Boas et al. (2014) for another application of this procedure.

We find the same general results when, using the Panel Study of Income Dynamics, we aggregate individual-level giving to the state level and use presidential vote share as a state-level measure of partisanship. These results are available in the Online Appendix.

One alternative explanation for this finding is that a person’s religious tradition—for example, being an evangelical Protestant, mainline Protestant, or Catholic—may explain these differences, rather than religious attendance. Using Steensland et al.'s (2000) denominational coding scheme, we explore this possibility and present the results in the Online Appendix. While church attendance is positively associated with total levels of giving along with giving to religious causes, the relationship between church attendance and giving is essentially the same for members of different religious faiths. Additionally, we may be concerned about religious homogamy. Andreoni et al. (2003) note that men and women have different tastes when it comes to charitable giving, requiring married couples to compromise and bargain when making donation decisions that occur at the household level. Consequently, if Republicans and Democrats differ in their propensity to marry within their religious traditions, this could account for some of the partisan variation in donation decisions. However, we find no evidence that partisans differ in their rates of marrying within their religious faith.

We find substantively and statistically similar results when we replicate the findings excluding religious non-identifiers, who should be especially unlikely to donate to religious organizations and congregations. These results are available in the Online Appendix.

The sample size in the GSS drops considerably because not all respondents were asked about charitable giving and the items used to construct operational conservatism. Unfortunately, the SCCBS lacks questions we could use to construct a measure of operational conservatism.

While the difference in giving between operationally liberal and operationally conservative Democrats borders on statistical significance, such a negative relationship actually contradicts the claim that policy conservatism and giving are positively linked. Additionally, there are very few operationally conservative Democrats. Below we report the results of a test that accounts for this lack of overlap.

We present similar results using a more flexible model specification in the Online Appendix where we follow Ellis and Stimson in classifying respondents as “consistent conservatives”, “consistent liberals”, and “conflicted ideologues”. This alternative specification, which accounts for the fact that there are few operationally conservative Democrats, produces the same substantive results.

Based on their analysis of baby names, Oliver et al. (2016) conclude Republicans are more motivated to signal economic as opposed to cultural capital.

We likely find changes in economic perceptions among Republicans and not Democrats because Obama’s re-election was expected among Democrats, but not Republicans. Because the lack of an effect on giving could be a result of a lack of statistical power, in the Online Appendix we test whether there is an interaction between “surprise”—the respondent’s pre-election prediction of who would win—and party. Surprised respondents should theoretically exhibit larger changes in economic behavior (see Quek and Sances 2015). While we find substantively large interactions between Republican identity and predicted Obama vote share for economic perceptions and vacation spending, we find no evidence of an interaction for giving.