Public Attitudes Toward Social Spending in the United States: The Differences Between Direct Spending and Tax Expenditures

Abstract

This paper uses a survey experiment to examine differences in public attitudes toward ‘direct’ and ‘indirect’ government spending. Federal social welfare spending in the USA has two components: the federal government spends money to directly provide social benefits to citizens, and also indirectly subsidizes the private provision of social benefits through tax expenditures. Though benefits provided through tax expenditures are considered spending for budgetary purposes, they differ from direct spending in several ways: in the mechanisms through which benefits are delivered to citizens, in how they distribute wealth across the income spectrum, and in the visibility of their policy consequences to the mass public. We develop and test a model explaining how these differences will affect public attitudes toward spending conducted through direct and indirect means. We find that support for otherwise identical social programs is generally higher when such programs are portrayed as being delivered through tax expenditures than when they are portrayed as being delivered by direct spending. In addition, support for tax expenditure programs which redistribute wealth upward drops when citizens are provided information about the redistributive effects. Both of these results are conditioned by partisanship, with the opinions of Republicans more sensitive to the mechanism through which benefits are delivered, and the opinions of Democrats more sensitive to information about their redistributive effects.

This paper examines public attitudes towards government spending in the USA. There is, of course, a wealth of excellent scholarship regarding public support for spending programs (e.g., Cook and Barrett 1992; Gilens 1999; Jacoby 1994, 2000; Rudolph and Evans 2005; Soroka and Wlezien 2010; Stimson 2004; Wlezien 1995). This line of research, however, focuses almost exclusively on direct social spending—the funding of social benefits through budgetary appropriations—ignoring public opinion towards federal involvement in private social welfare programs (but see Haselswerdt and Bartels 2011; Mettler 2011). This focus on direct spending alone limits our understanding of public opinion towards government spending, given that the federal government finances social welfare benefits not only through direct public appropriations, but also by subsidizing the private provision of social benefits through the use of targeted tax expenditures.

The distinction between public and private benefits in the United States’ “divided” (Hacker 2002) social system is important for several reasons. First, there are partisan differences in how policymakers finance social programs, as Democrats prefer to finance public social programs through direct appropriations, while Republicans prefer to finance private programs through tax expenditures (Faricy 2011). Second, there are policy differences between direct social spending and social tax expenditures: direct spending and tax expenditures are produced through different policy processes, assist different populations, and redistribute federal money to different economic and social groups (e.g., Hacker 2002; Howard 1997). Finally, there may be differences in the public visibility of direct spending versus tax expenditures: some have argued that tax expenditure programs (and the redistributive effects of such programs) are “hidden” from public view in a way that makes public preferences on these programs less well-informed than preferences on direct spending (Mettler 2008, 2011). In ignoring the private social system and the tax subsidies that finance it, public opinion scholars miss the opportunity to understand attitudes toward a substantial portion of the American welfare state, and are unable to gauge differences in how the public perceives programs financed through public or private means.

This paper addresses these issues experimentally, focusing on differences in how citizens view government spending done through direct appropriations and spending done through tax expenditures. Our goal is to see how support for the exact same social programs changes when the delivery mechanism of those programs (direct spending or tax expenditures) changes, and the information that citizens are provided about the redistributive effects of those programs changes.

We derive three expectations. First, we expect that citizens will be less likely to support upward-redistributing social programs when provided information about such programs’ redistributive effects. Second, we expect that programs which are portrayed as being delivered through tax expenditures will be more popular than identical programs portrayed as delivered through direct spending. Third, we expect that Republicans and Democrats will react in different ways to both the redistributive- and delivery-mechanism frames: changes in the “redistributive” frame should be particularly important to the attitudes of Democrats, while changes in the “delivery” frame should be most important to the attitudes of Republicans. The results demonstrate that citizens’ support for social spending programs is in part driven by both how programs are delivered and whether citizens are thinking about the programs’ redistributive effects. More importantly, the results shed light on the ‘real world’ ways in which policymakers frame social spending policies, and the ways in which citizens evaluate such policies.

Direct Social Spending and Social Tax Expenditures

The USA has a divided social welfare state in that the federal government uses both direct and indirect means of spending to finance the provision of social benefits. While direct spending is easily recognizable as money spent by the federal government on programs and agencies such as Social Security, Medicare, the Department of Education, and Temporary Aid to Needy Families—the concept of social tax expenditures requires a bit more explanation.

Social tax expenditures refer to efforts by the federal government, through the use of targeted tax breaks, to incentivize the private provision and consumption of social benefits.¹ These activities provide tax incentives for individuals to purchase social benefits from private companies—by, for example, contributing to 401(k) plans, purchasing employer-based health insurance policies, or taking out and paying interest on a home mortgage—in lieu of having the government provide those benefits directly. The government, in other words, can choose to finance social benefits either through direct spending (by writing a check to citizens or otherwise directly covering the cost of those benefits), or through tax expenditures (by allowing some citizens who use government-approved private social services to write smaller checks to the federal government in the form of income taxes).

The Divergent Policy Effects of Direct Social Spending and Social Tax Expenditures

While tax expenditures, like direct spending, are utilized to deliver popular social benefits to citizens, the two spending types diverge in how they deliver benefits, and to whom those benefits tend to accrue. The majority of tax expenditures have regressive effects on the national income distribution. Since tax deductions are typically claimed against a progressive income tax structure, benefits accrue disproportionately to those in higher marginal-rate tax brackets. Social welfare benefits directed through tax deductions, exclusions, and non-refundable tax credits thus reduce the progressivity of the tax structure.² For example, recent reports from the Joint Committee on Taxation (JCT) find that households in the top 20 % of the income distribution receive 80 % of the income benefits from tax deductions related to employer-based retirement plans (JCT 2011). In contrast, public social programs largely aid lower-income citizens and have progressive effects on income distribution (Kelly 2009).

In addition, the demographic profile of those who receive social services through direct spending and through tax expenditures are generally different. White collar, full-time employees in large companies are much more likely to be offered, enrolled in, and utilize company health and pension benefits (and the federal tax subsidies that finance them) than are other workers (Faricy 2011). According to studies from the Employee Benefits Research Institute (2009), the modal recipient of an employer-based retirement plan was a white male with a college education, earning over $50,000, and working in a large firm. This ‘government welfare’ recipient stands in stark contrast to the types of citizens who typically receive direct federal social benefits—who tend to be comparably poorer and older (Howard 2007). It also stands in contrast to the type of person that the public often perceives as receiving government assistance—poor, out of work, and a member of a racial minority group (e.g., Iyengar 1994; Gilens 1999; Kellstedt 2000).

The two types of social spending also represent two distinct views of the proper role of the government in the economy. Social tax expenditures are used to predominantly subsidize private business and tend to privilege market (as opposed to directly government-based) solutions to social problems, while direct spending advocates for a direct government role in providing social benefits as a means to address societal inequalities (Hacker 2002). Consistent with what might be expected, we see partisan differences with respect to the choice of spending tool used to finance social benefits: when Republicans control the elected branches of the federal government, the percentage of all social spending devoted to social tax expenditures increases. When Democrats are in control, the opposite is true (Faricy 2011).

Public Attitudes Toward Government Spending

What affects citizens’ attitudes toward government spending on social programs? A substantial amount of literature has been devoted to answering this question (e.g., Gilens 1999; Jacoby 1994, 2000; Stimson 2004; Wlezien 2004). This research shows broad support for federal involvement in the provision of social goods and services. The idea that the government should play a role in providing housing for the poor, pensions for the elderly, education for all children, and a variety of other social benefits is popular with the public, even among Republicans and Conservatives (e.g., Stimson 2004).³ Citizens tend to prefer a government that does more to equalize wealth over one that does less (Page and Jacobs 2009).

At the same time, though, the public tends to dislike ‘the government’ in a symbolic sense, and (despite favoring increased spending on nearly all major government social welfare programs), expresses abstract preferences for a smaller, leaner, government (Ellis and Stimson 2012). This ‘ambivalence’ (Cantril and Cantril 1999) toward the federal government’s role in providing social benefits has been the subject of much research in American public opinion (e.g., Free and Cantril 1969; Schiffer 2000; Stimson 2004; Ellis and Stimson 2012).

Most of this literature, however, focuses exclusively on government spending for public programs such as Social Security or welfare. To date, there has been little empirical research on public preferences towards social tax expenditures. While both direct spending and tax expenditures are government-based programs designed to provide social benefits to citizens, the two types of spending vary in both the role that they accord to the government and the types of benefits that they provide.

The public policy literature provides reason to think that citizens might view direct social spending and social tax expenditure programs differently. In particular, many policy scholars are concerned about the relative lack of visibility and transparency of tax expenditure programs, and the resultant effects on how the public views such programs (Mettler 2008). Citizens are clearly not deeply informed about the particulars of any sort of social policy, whether funded through direct spending or tax expenditures (Gilens 2001). But there is reason to think that tax expenditures—and the policy processes that create them—are particularly shielded from public view. Hacker (2002), for example, refers to the policy process that creates tax expenditures as “subterranean,” since tax expenditures are created outside the formal budget process and the administration of benefits is activated through a complex mix of public and private actors.

The resulting low visibility of tax expenditures makes it difficult for citizens to understand what this sort of social policy does and what effects it has. Social tax expenditures, in other words, may be ‘government actions’ that are not recognized as such by the public. Mettler (2010), for example, shows that most citizens who receive social benefits through the tax code are unaware that they are even receiving government social benefits. This work finds that a majority of respondents who claimed to have used a tax expenditure program (such as the mortgage deduction or child care credit) also claimed to have never benefited from a federal social program of any sort. Recipients of direct spending programs, by contrast, were more likely to acknowledge that the government had helped them in a time of need.

The implication of this is that tax expenditure programs, though they provide government-directed benefits to citizens, are not viewed as ‘government programs’ in the same way as programs financed through direct appropriations. As a result, the experimental work of Haselswerdt and Bartels (2011) suggests that providing citizens information about the explicit governmental role in facilitating social goals such as homeownership (through the mortgage interest tax deduction) leads citizens to consider these issues in a different light. In addition, some recent work suggests that the ‘submerged’ nature of tax expenditure policies leads citizens to be largely unaware of the redistributive effects of tax expenditures (e.g., Hacker and Pierson 2010), and that prompting citizens to think about particular social policies through the lens of redistribution—as opposed to the goal that it is trying to facilitate or the social norms that it is helping to promote—leads citizens to consider the policies differently as well (Mettler 2011).

Understanding Attitudes Toward Spending: An Experimental Approach

This discussion has several implications for understanding why citizens support or oppose particular social spending programs. We know that citizens generally support greater spending on social programs in part because they see value in the ends—better education, retirement security, and the like—that spending on social programs promises (Jacoby 2000). But the impulse to support direct social spending conflicts with the public’s long-standing skepticism toward “government intervention” in the market economy (Sears and Citrin 1981; Ellis and Stimson 2012). Tax expenditures, by contrast, comport well with public demands for a smaller direct governmental role in managing the economy. But most tax expenditures, in practice, lead to regressive redistributive effects that citizens tend to dislike.

Taken together, this implies that the design and delivery mechanism of particular public policies—and the degree to which the effects of those policies are visible—should be important determinants of whether citizens respond positively to them (Mettler 2010). More specifically, it suggests that the way a social program is presented and framed to the public will have a substantial impact on citizens’ support for it. In what follows, we work to understand how the framing of social spending programs affects public support for them. Our approach is experimental: we wish to see whether differences in how policy programs are framed have an impact on public support or opposition, and whether the effect of framing will vary across people with different political values. We first develop hypotheses regarding how the framing of social spending programs will affect support for such programs. We then develop and discuss the results of a survey experiment designed to test these hypotheses.

First, we expect that the redistributive effects of government benefits—and citizens’ awareness of those redistributive effects—matters to public support. Citizens generally tend to dislike government programs that redistribute wealth upward, particularly to social groups that they do not view as deserving of government aid. By contrast, citizens are generally tolerant of (if not entirely supportive of) programs that redistribute wealth downward (e.g., Cantril and Cantril 1999; Page and Jacobs 2009). Citizens are not often aware, however, of how and to whom government programs redistribute wealth, and this general lack of awareness is particularly true for programs that rely on tax expenditures (Mettler 2010). Taken together, this suggests that citizens will be less likely to support certain kinds of spending programs if they were made aware of the regressive redistributive effects of such programs. We thus expect that:

Hypothesis 1

  For social programs which redistribute wealth upward, providing information about such programs’ redistributive effects will decrease public support for them.

We do not expect, by contrast, that providing information to respondents about a downward-redistributing program should negatively affect support for the program. The notion that government should be redistributing wealth downward and providing services to those who need them is, at least on balance, popular with the American public (Ellis and Stimson 2012).⁴

Next, we expect that the mechanism of delivery of benefits matters to how citizens view social programs. Benefits distributed through tax expenditures are not often perceived as being government-provided—even by recipients of such benefits. This implies that citizens do not necessarily view tax expenditures as an explicit government-based effort to intervene in social outcomes (nor do they view beneficiaries of tax expenditure programs as recipients of “government assistance”), but rather view tax expenditures as ways through which citizens “help themselves” and save on their taxes as a result.⁵ Because of this, a tax expenditure program is less likely to be viewed through the prism of “government” and “government spending”—with all of the negative connotations that those phrases contain—than is direct spending.

This implies that, all else equal, a program that is portrayed as providing tax credits from the government to individuals for engaging in particular behaviors will be received more positively by the public than one portrayed as providing direct payments from the government to the same individuals for the same behaviors. We thus expect that:

Hypothesis 2

  Social programs portrayed as “tax expenditures” will be supported more strongly than social programs portrayed as “direct spending.”

Third, we expect Democrats and Republicans to react differently to both the delivery-mechanism and redistributive effects frames. Differences in how Democrats and Republicans view these programs manifest themselves clearly at the elite level: Democrats tend to prioritize direct spending, and Republicans tax expenditures, when deciding how to best provide social benefits (Faricy 2011). In addition, inequality on average increases at a faster rate during Republican presidencies than Democratic ones, suggesting a greater tolerance for inequality among elite Republicans than among elite Democrats (Hacker and Pierson 2006; Bartels 2008).

At the mass level, there are also many reasons to think that Democrats and Republicans would view social spending—and how social spending is framed—differently. Mass party identifiers differ in how they view the proper role and scope of government in remediating social inequality. Though Republicans are on balance tolerant of government efforts to redistribute wealth downward (Page and Jacobs 2009), they are also more strongly opposed to governmental intervention in the economy, and more strongly prize individualism as a core political value, than Democrats (Feldman and Zaller 1992; Franklin and Jackson 1993; Ura and Ellis 2012). While Democrats are not exactly “pro-government,” they are typically more supportive of government efforts to help poor citizens and groups, and less likely to react negatively to the idea of activist government. They are also more likely than Republicans to espouse the value of egalitarianism, and to resist policies that benefit wealthier citizens over poorer ones (Goren 2001).

Taken together, these differences in values suggest that Republicans and Democrats will react differently to how social spending programs are framed. For Republicans, we expect that whether a program is portrayed as “direct spending” or “tax expenditures” will have a substantial impact on program support. Even if they are the same in cost and intent, tax expenditures—given that they are perceived as a decrease in the amount of one’s own tax burden—are easier to view as a reward for one’s own individual initiative and effort (as opposed to being viewed as ‘government assistance’) than direct spending—which is realized as a direct payment from the government to an individual. An awareness of the regressive redistributive effects of a program, by contrast, should be less important to Republicans.

For Democrats, we expect the reverse to be true. The portrayal of a program as being financed through direct spending or tax expenditures means should matter less for Democrats, given greater levels of abstract support for activist solutions to social problems. The awareness of the redistributive effects of a program, by contrast, should matter more to Democrats than it does to Republicans. While citizens of both parties are generally unaware of the redistributive effects of most tax expenditure programs, making citizens aware of such effects is more likely to conflict with egalitarian values among Democrats, and thus more likely to dampen support for these programs. More formally, we expect that:

Hypothesis 3

  For social programs which redistribute wealth upward, providing information about a program’s redistributive effects will decrease support for the program more strongly among Democrats than among Republicans.

Hypothesis 4

  The framing of a program as “direct spending” as opposed to “tax expenditures” will decrease support for that program more strongly among Republicans than among Democrats.

Experimental Design

To address these hypotheses, we develop an experiment which allows us to view the effects of two different kinds of frames—a “spending type” frame (i.e., whether a program is portrayed as direct spending or tax expenditures) and a “redistributive type” frame (i.e., whether respondents are given some information about the redistributive impact of a program)—on public support for three types of social programs. Though we greatly simplify these programs when describing them to respondents, all three programs mirror existing federal programs in cost, intent, and actual redistributive effects. The programs are:

• (a) Mortgage interest tax deduction. This program allows individuals who itemize their tax returns to deduct their mortgage interest payments from their federal taxes. The mortgage interest deduction is one of the largest tax expenditure programs in the USA, and has been the subject of considerable scholarly research because of both its size and its upward redistribution effects (Hacker 2002; Howard 1999; Mettler 2010, 2011).

• (b) Retirement savings tax deduction. This program allows individuals to contribute to employer-provided retirement plans on a pre-tax basis. This program is also widely considered to redistribute wealth upward: it is available only to those who are employed in benefits-eligible positions, and gives the most benefits to those who are able to save the most money for retirement. In addition, by incentivizing investment in private retirement plans—rather than using the same amount of money to increase government-provided retirement benefits—this program privileges private- rather than public-sector solutions for saving for retirement (Kelly 2009; Ellis and Faricy 2011).

• (c) Food stamps. This program, a component of the federal welfare system that is administered by states, provides financial assistance for families to purchase groceries and other necessities. This program is currently constituted as a direct spending program, one which provides direct payments rather than tax credits to recipients. As a means-tested program, it redistributes wealth downward rather than upward.

It is important to note the differences between the first two types of programs and the third one. First, the mortgage interest and retirement security programs are currently delivered through the tax code, while the food stamp program is currently delivered through direct spending. In addition, the programs differ in their relative familiarity to citizens. Despite being two of the largest federal tax expenditure programs, the mortgage interest and retirement spending programs are often cited as examples of the ‘submerged’ welfare state (Mettler 2011), programs for which even beneficiaries are often unaware of their functioning or existence. The food stamp program, by contrast, one for which most citizens have at least a passing (if limited) understanding of its intent, target beneficiaries, and distributive effects (e.g., Gilens 1999, 2001). In fact, the food stamp program is one of the small set of programs—those which explicitly redistribute wealth to the poor—that the public most closely connects with the broader concept of “government spending” (e.g., Cook and Barrett 1992; Jacoby 2000).

For each of these programs, we develop a short description of them that summarizes their intent and cost. We then vary the description of each in two ways. First, we vary the delivery mechanism of the program—i.e., whether the program’s benefits will be delivered through tax credits (tax expenditures) or government payments (direct spending). Second, we vary the information about redistribution that respondents are provided about the programs, either saying nothing about redistributive effects or describing briefly the population to which the benefits of the program will mostly accrue. Each program can thus be presented in four ways: (1) as a tax expenditure, with no redistributive effects, (2) as tax expenditures, with redistributive effects, (3) as direct spending with no redistributive effects, or (4) as direct spending with redistributive effects. For example, the four questions for the mortgage interest program are:

• Tax Expenditure Some have endorsed a program that would provide individuals and families assistance in making the monthly payments on their homes. Under this program, homeowners would be allowed to deduct their mortgage interest payments from their federal taxes, helping to offset the amount of money that they pay in interest on their mortgage. The total cost of this program is expected to be about $100 billion per year.

• Direct Spending Some have endorsed a program that would provide individuals and families assistance in making the monthly payments on their homes. Under this program, homeowners would receive annual payments from the federal government to help to offset the amount of money they pay in interest on their mortgage. The total cost of this program is expected to be about $100 billion per year.

• Tax Expenditure, Redistributive Some have endorsed a program that would provide individuals and families assistance in making the monthly payments on their homes. Under this program, homeowners would be allowed to deduct their mortgage interest payments from their federal taxes, helping to offset the amount of money that they pay in interest on their mortgage. The tax deductions will be provided only to those who own homes, with the largest deductions typically going to those who own the most expensive homes (and thus pay the most in mortgage interest). The total cost of this program is expected to be about $100 billion per year.

• Direct Spending, Redistributive Some have endorsed a program that would provide individuals and families assistance in making the monthly payments on their homes. Under this program, homeowners would receive annual payments from the federal government to help to offset the amount of money they pay in interest on their mortgage. These payments will be provided only to those who own homes, with the largest payments typically going to those who own the most expensive homes (and thus pay the most in mortgage interest). The total cost of this program is expected to be about $100 billion per year.

Wording for the other two questions is available in the Appendix. Each question ends by asking respondents whether they “strongly support,” “somewhat support,” “neither support nor oppose,” “somewhat oppose,” or “strongly oppose” each program.

Respondents for the survey experiment were recruited from undergraduate classes at a large university in the Pacific Northwest, and were offered extra course credit for their participation in the experiment. Upon agreeing to complete the survey, respondents were directed to an online questionnaire. After completing a series of basic demographic questions, respondents were randomly assigned (separately for each program) to one of the four question types for each social spending program, and asked for their views on (in order) mortgage interest, retirement savings, and food stamps. After answering the experimental questions on the four social spending programs, respondents then completed a short battery of other attitudinal and demographic questions.

The survey yielded 373 total respondents (158 females, 188 males). As a convenience sample employing student respondents, the sample is clearly not ideal in many respects. But the sample was politically diverse (32 % of respondents identified as Republican, 36 % as Independent, and 32 % as Democratic), and held political and social values that reflect those of the USA at large. We can certainly imagine ways in which students might understand and process information about these social programs differently than other citizens (and might even hold different aggregate distributions on them).⁶ But we see little theoretical reason to think that the use of our sample will systematically affect the direction or significance of the framing effects across different types of questions, the main threat to drawing inferences from randomized experiments conducted on non-representative samples (Druckman and Kam 2011).⁷

Top-Line Results

As one would expect given the political diversity of the sample, attitudes toward these three programs were quite mixed. Averaging across all experimental conditions, the food stamp program was the most popular of the programs, with 54 % of respondents either ‘strongly supporting’ or ‘supporting’ it, and 30 % ‘opposing’ or ‘strongly opposing it.’ The retirement security program saw 47 % of respondents supporting it and 26 % opposing it. The mortgage interest deduction was least popular, garnering 27 % support and 49 % opposition.

Before evaluating our core hypotheses, we first note the factors that predict attitudes toward each of these spending programs in general, ignoring the question format to which respondents were exposed. To do so, we run basic regression models predicting respondents’ level of support for each of these programs (with the responses scaled such that ‘strongly support’ is coded as 5, and ‘strongly oppose’ is coded as 1) as a function of a straightforward battery of demographic and political covariates, as well as a question asking how deserving the target beneficiaries of the programs are of government assistance.

Table 1 presents the results of these models. While these results are not central to our hypotheses, it is useful to note that attitudes on each of these three programs are fairly well predicted by intuitive factors: most important among these factors was a perception of ‘deservingness’: believing that a program’s particular target group was deserving of government aid also led to greater levels of support for social spending across all three programs. In general, Democrats and egalitarians are also more supportive of these programs than are Republicans and individualists. These results suggest that attitudes toward these issues are predictable in ways that one would expect given past research on the determinants of social spending preferences (Jacoby 2000; Rudolph and Evans 2005). This helps to reassure us that our student sample is viewing these programs like the adult population would, and provides some confidence that attitudes toward these issues are not so fleeting that they will be dictated entirely by question framing.⁸

Table 1 OLS models predicting attitudes toward social spending programs

Experimental Results

With these results in hand, we turn to evaluations of our first two hypotheses. Figure 1 provides a basic look at the results from the experiment, showing mean levels of support, for each of the three spending programs and each of the four framing conditions. For mortgage interest and retirement security, these tables show us that framing matters: basic ANOVA analyses provide evidence that the mean levels of support are not equal across framing conditions (p < 0.05 for both programs). For the food stamp program, by contrast (again, the only program that is in reality constituted as a direct spending program, and the only one with clear downward redistributive effects), the ANOVA results show no evidence that question framing matters. We see differences in levels of support for the program across the four framing conditions that are consistent with our hypotheses. Both the mortgage interest and retirement security programs are the least popular when the programs are framed as providing direct payments to recipients and when respondents are given some information about to whom the benefits are most likely to accrue.

Fig. 1

Mean levels of spending support, by framing condition

Table 2 provides a direct evaluation of our first two hypotheses. This table reports results comparing mean levels of support across the ‘spending type’ and ‘redistributive effects’ conditions for each of the three programs. For both the mortgage interest and retirement programs, inclusion of some information about redistributive effects significantly dampens support for the programs, reducing mean levels of support for the programs by roughly three-tenths of a point on the five-point scale. In the case of the retirement security program, in fact, this change in spending frame causes the majority opinion to change: when respondents are not provided with information about redistributive effects, 56 % of respondents say that they support this program. Only 42 % of respondents provided with information about the redistributive effects of retirement security, however, support this program. The mortgage program is on balance unpopular regardless of frame, but support drops from 33 to 21 % when respondents are provided with information about redistributive effects.⁹

Table 2 Mean levels of support for spending programs across framing conditions

These results are consistent with our hypotheses, showing that providing respondents information about the wealth distribution effects of major tax expenditures causes reduces support for those programs. The inclusion of the redistributive frame, by contrast, has no significant or substantively meaningful effects for the food stamp program. This result makes sense, given the fact that the food stamp program, unlike the other two programs, tends to redistribute wealth downward.

We see similar, but somewhat weaker, results when it comes to the effects of the ‘spending type’ frame. For the retirement security program, framing a program as providing direct payments rather than tax credits significantly diminishes support for the program. Again, this shifts the retirement security program from one which garners majority support to one which garners plurality opposition (54 % of respondents support the retirement security program when it is framed as a ‘tax credit,’ while only 40 % support it when it is framed as a ‘government payment’). The results for the mortgage interest program are in the expected direction (with ‘government payments’ being less popular than ‘tax deductions’), but the mean levels of support across the two frames are not significantly different from one another.

Again, the results for these two programs contrast with those for the food stamp program, where the mean levels of support across the spending conditions are essentially identical. This runs counter to our hypotheses, since regardless of a program’s redistributive effects, we expected that framing a program in terms of a tax credit would be more palatable than framing it as a direct cash payment. Respondents’ greater familiarity with the food stamp program—and the notion that the mention of ‘government spending’ causes citizens to think primarily of ‘redistributive social welfare spending’ such as money spent on food stamps (Jacoby 1994)—suggests that perhaps attitudes on this program are simply more resistant to framing than they are for the other programs.

In addition to these results, it is useful to note the implications of some of the differences across the four different types of framing conditions considered separately. Most importantly, we see that (again, at least for retirement security and mortgage interest) evidence that framing a program as providing benefits through ‘tax credits’ rather than ‘government payments’ influences not only levels of support for the program, but also the degree to which citizens will tolerate upward redistributive effects in the programs. The difference in levels of support for the ‘direct-redistributive’ frame and ‘tax expenditure-redistributive’ frame are particularly strong and statistically powerful for the retirement security program (and somewhat weaker for the mortgage interest program). But both models suggest that citizens will be more willing to support a tax expenditure program that redistributes wealth upward than a direct payment program that does the same thing. Again, for the food stamp program, there is little evidence that framing affects program support.

Taken together, these results provide general support for our first two hypotheses. For the mortgage interest and retirement programs, providing respondents some information about the programs’ redistributive effects significantly dampened support for them. And at least with respect to retirement security, framing the program as a tax expenditure rather than as a direct payment decreased support as well. But again, these results do not hold for the food stamp program, where support remained similar across each of the four framing conditions.

Partisanship and Issue Framing

Our final two hypotheses suggest that partisanship should play a role in how respondents react to changes in the framing of social spending programs. In particular, we expect that Republicans will be particularly sensitive to changes in the spending frame, while Democrats will be more sensitive to changes in the redistributive effects frame. To test these hypotheses, Fig. 2 presents results comparing mean levels of support across the redistributive and spending frames for Democratic and Republican identifiers.

Fig. 2

Effects of spending and redistributive frames, by party identification

The results are supportive of our hypotheses.¹⁰ For Republicans, the inclusion of the redistributive effects frame does not significantly lessen support for any of the three programs (in the case of mortgage interest, stating that the benefits accrue primarily to owners of expensive homes modestly, though not significantly, increases support for the program). Republicans, it appears, decide whether to support or oppose these programs in a way that does not take into account the degree to which they redistribute wealth. Democrats, by contrast, are significantly less supportive of both the mortgage interest and retirement security programs when they are told that benefits accrue primarily to owners of expensive homes and those who save the most for retirement. These results are consistent with our expectations: given Democrats’ greater concern with income inequality, it stands to reason that Democrats would view these programs differently when redistributive effects are explicitly primed.¹¹

Again, though, we see no results for the food stamps program: though the redistributive effects frame reminds respondents that this program helps the poorest citizens, Republicans are not less likely to support the program as a result. This makes some sense, given that Republicans are thought to be at least tolerant of downward redistribution (Page and Jacobs 2009).

When it comes to the spending frame, we also see results consistent with our hypotheses. Both Republicans and Democrats support the mortgage interest and retirement programs less when they are framed as direct payments rather than tax credits. These results make clear sense for Republicans, given traditional Republican resistance to direct government spending programs. They also fit broadly with what we know about Democrats’ political values and attitudes toward government, which are typically more conflicted than those of Republicans. While Democrats are more receptive than are Republicans to direct government intervention in the economy, the mainstream Democratic Party still prefers, all else equal, a smaller and less intrusive government to a more expansive one (Feldman and Zaller 1992). So it is not surprising that Democrats would prefer private over public means of achieving the same policy objectives.

But importantly, we see that the effect of the spending frame on Republican preferences is significantly greater than the effect of framing on Democratic preferences for both the retirement security and food stamp programs. For the mortgage interest program, the mean difference in Republican support across framing conditions (0.43) is greater than the mean difference in Democratic support (0.28). For the retirement program, the mean difference in Republican support (0.41) is nearly double the mean difference in Democratic support (0.22). Again, though, the effects of the spending frame are restricted to these two programs. Though Republicans do have modestly greater levels of support for the food stamp program when it is framed as a tax credit rather than a direct payment, neither Republican nor Democratic support for the program is significantly affected by the mode of program delivery.

Conclusions

Our examination of the effects of policy framing on preferences for social spending reveals several things. First, we found that the mechanism through which social benefits are delivered matters to how much citizens support programs designed to provide such benefits: support for social spending programs is generally higher when programs are presented as tax expenditures than when the exact same programs are presented as direct spending. Second, we found that public support decreases for social programs which distribute wealth upward when citizens are made aware of such programs’ redistributive effects. Finally, we found that Democrats and Republicans reacted to both the delivery mechanism and the mention of redistributive effects differently, with Republican support more sensitive to the mechanism through which programs were financed, and Democratic support more sensitive to the mention of redistributive effects. In all cases, though, we found framing effects only for the mortgage interest and retirement security programs, and not for the food stamp program.

Though we cannot generalize to the universe of social programs from these examples, these results provide some guidance on when and how framing is likely to matter to public support for such programs. The fact that the ‘redistributive effects’ frame was more important to opinions for our two upward-redistributing programs (retirement security and mortgage interest) than our downward-redistributing program (food stamps) is not surprising: citizens generally assume that government social spending distributes wealth downward, so explaining that a particular program actually redistributes wealth upward is more likely on balance to lead them to reconsider their views on such programs.

Evidence that the delivery-mechanism frame matters to the retirement and mortgage interest programs also suggests that citizens, all else equal, will more strongly support social programs financed through the tax code than otherwise identical programs financed through direct government payments. Counter to our hypotheses, though, we saw no effect of spending type frame for the food stamp issue: respondents did not seem to care about how the program was financed. These results illustrate the limits of framing, and of our hypotheses. They also suggest that framing is likely to matter more to the support of programs on which citizens have less familiarity. The retirement security and mortgage interest programs are ones that some argue are particularly unfamiliar to the public (e.g., Howard 1997; Mettler 2011), while a program such as the food stamp program is more easily recognized as an archetypical example of ‘government spending’ (Schneider and Jacoby 2005). It would be useful for future research to explore the impact of framing on other issues, perhaps singling out direct spending programs (such as those to protect the environment, or to fund education or scientific research, for example) that are not as closely associated with the general concept of ‘government spending’ as is food stamps.

Even with these limitations, the results provide several insights on how citizens evaluate social spending programs outside of the lab. For example, we find that the effects of the delivery-mechanism frame on support for social programs was conditioned by partisanship. The fact that the effects of the delivery-mechanism frame were stronger for Republicans than Democrats could serve as an explanation for why Republicans, generally opposed to the idea of ‘big government,’ also on balance report preferences for higher levels of government spending on specific social programs (Ellis and Stimson 2012). At least some Republicans might view tax expenditure policy as a way to have its policy cake and eat it, too: providing desirable social benefits while still adhering to the values of individual initiative and support for the private sector.

Our finding regarding the differential importance of the redistributive frame to Democrats and Republicans adds nuance to extant findings on how information about income redistribution impacts attitudes towards social spending. Our results, which show that highlighting information about the regressive redistributive effects of tax expenditure programs dampens Democratic support for these programs more than it does Republican support, makes sense given the parties’ general stances on the proper role of government in redistributing wealth. It also suggests that while a strategy to provide citizens more information about the regressive effects of major tax expenditure programs may decrease overall support for these programs, it would do so in a way that affects Democrats more than Republicans.

In addition, and though we are clearly limited by our survey-experimental approach, we believe that the results here are of value in explaining how ‘real-world’ policy elites frame proposals for the financing of social benefits. In general, citizens generally support both the abstract idea of ‘limited government’ and a strong government role in financing many specific aspects of a strong social welfare state. In this light, Howard (2007) argues that policymakers, particularly conservative ones, turn to tax expenditures as a way to thread the political needle of funding popular social goals while seemingly advancing the virtues of individual initiative and smaller government. The fact that tax expenditures seemingly fit both roles—small government, and a strong system of social welfare—suggests that such programs have a ‘framing advantage’ that makes it easier for policymakers to persuade the public to support private rather than public solutions for social problems. Tax expenditure programs, in other words, make it possible for policymakers to claim that a program both finances popular social goals while also largely avoiding the specter of ‘government involvement.’ Direct social spending, by contrast, fulfills public demands for more social services, but conflicts with public impulses to limit government.

This suggests that, at the margins, those in favor of privatizing social services through tax expenditure policy should have an easier time mobilizing public support for their plans, particularly if these programs are pitted against similar programs financed through direct means. It also suggests that those who wish to protect existing tax expenditure programs, even highly regressive ones, are advantaged by the fact that such programs will be viewed in a more positive light than they would be if the same benefits were delivered through direct payments. The fact that even progressive Presidents have added to or resisted challenging entrenched tax expenditure programs speaks to the high level of support that such programs have.

Highlighting the upward-redistributing effects of many real or proposed tax expenditure programs, however, may in part negate the ‘framing advantage’ of tax expenditure programs. Ellis and Stimson (2012), for example, document the failure, driven by public backlash, of George Bush to partially privatize Social Security, a failure that was driven in large part by opponents’ efforts to highlight the redistributive consequences of such a plan. Generally high levels of abstract public support for “letting citizens invest their own payroll tax contributions in private retirement accounts,” in other words, waned when citizens became aware of the effects that this would have on the long-standing social safety net for poor- and middle-class seniors.

The tepid response to Paul Ryan’s plan to partially privatize Medicare illustrates a similar issue. Opponents’ efforts to highlight who would be hurt by a Medicare voucher plan—President Obama has attacked this plan as one that would assist wealthier senior citizens at the expense of the elderly poor—have dampened public support for it. By highlighting the redistributive effects of such a plan, Obama has worked to diminish what might be abstract support for a Medicare plan financed through the private marketplace. In any case, the results suggest that policymakers behave in a way consistent with what we show here: those in favor of expanding tax expenditure programs highlight the fact that such programs reduce direct government influence, while opponents seek to highlight such programs’ often regressive nature.¹²

Appendix: Question Wording for Retirement and Food Stamp Questions

Retirement

Tax Expenditure Some have endorsed a program that would assist individuals in saving for retirement while still paying day-to-day bills. Under this plan, individuals who invest money in a private retirement plan would be able to deduct the amount that they save for retirement from their taxable income. This tax deduction would reduce the amount of taxes that workers must pay to the federal government. The total cost of this program is expected to be roughly $115 billion per year.

Direct Spending Some have endorsed a program that would assist individuals in saving for retirement while still paying day-to-day bills. Under this plan, individuals who earn invest money in a private retirement plan would be eligible for government payments that would help to offset the amount of money that they save for retirement. The total cost of this program is expected to be roughly $115 billion per year.

Tax Expenditure, Redistributive Some have endorsed a program that would assist individuals in saving for retirement while still paying day-to-day bills. Under this plan, individuals who invest money in a private retirement plan would be able to deduct the amount that they save for retirement from their taxable income. This tax deduction would reduce the amount of taxes that workers must pay to the federal government. These tax deductions would go to those who save for retirement, with the largest payments going to those who save the most money for retirement. The total cost of this program is expected to be roughly $115 billion per year.

Direct Spending, Redistributive Some have endorsed a program that would assist individuals in saving for retirement while still paying day-to-day bills. Under this plan, individuals who earn invest money in a private retirement plan would be eligible for government payments that would help to offset the amount of money that they save for retirement. These government payments would go to those who save for retirement, with the largest payments going to those who save the most money for retirement. The total cost of this program is expected to be roughly $115 billion per year.

Food Stamps

Tax Expenditure Some have endorsed a program that would provide low-income citizens with assistance in paying for groceries and other necessities. Under this program, citizens would be eligible to deduct the amount of money that they pay for groceries from their taxable income. These tax credits would reduce the amount of federal income tax that these citizens pay each year. The total cost of this program is expected to be $65 billion per year.

Direct Spending Next, some have endorsed a program that would provide low-income citizens with assistance in paying for groceries and other necessities. Under this program, certain citizens would be eligible for cash grants from the federal government that can be used to purchase groceries. The total cost of this program is expected to be $65 billion per year.

Tax Expenditure, Redistributive Next, some have endorsed a program that would provide low-income citizens with assistance in paying for groceries and other necessities. Under this program, citizens would be eligible to deduct the amount of money that they pay for groceries from their taxable income. These tax credits would reduce the amount of federal income tax that these citizens pay each year. Since the program is designed to help those who struggle to pay for basic necessities, the largest tax credits would go to those with the lowest incomes. The total cost of this program is expected to be $65 billion per year.

Direct Spending, Redistributive Next, some have endorsed a program that would provide low-income citizens with assistance in paying for groceries and other necessities. Under this program, certain citizens would be eligible for cash grants from the federal government that can be used to purchase groceries. Since the program is designed to help those who struggle to pay for basic necessities, the largest tax credits would go to those with very low or no incomes. The total cost of this program is expected to be $65 billion per year.