Perceptions of Taxing and Spending: A Survey Experiment
abstract. This Note presents the results of an original survey experiment on whether the public prefers “tax expenditures” to “direct outlays”—that is, whether members of the public are more likely to support government spending that takes the form of a tax credit rather than a check or cash. Using a survey that spans a wide variety of policy areas—and with important variations in wording and information—we show that the public strongly prefers tax expenditures even when the economic substance of the proposed policies is identical. We also show that the public views tax expenditures as less costly than equivalent direct outlays. These results support a longstanding but largely unstudied hypothesis that tax expenditures hide the costs of government spending, and have implications for why tax expenditures have continued to grow in size and complexity.
author. Conor Clarke: Yale Law School, J.D. expected 2015; London School of Economics, M.Res. 2011; University of Birmingham, M.Sc. 2010; Amherst College, B.A. 2008. Edward Fox: Yale Law School, J.D. expected 2015; University of Michigan, Ph.D. Candidate in Economics; Columbia University, B.A. 2007. We gratefully acknowledge the support of the Oscar M. Ruebhausen Fund at Yale Law School, which made this research possible. We are also indebted to Chris Milione for his careful and conscientious editing, as well as to Devon Porter, Meng Jia Yang, Rachel Bayefsky, and the many patient editors of the Yale Law Journal. For helpful comments and conversations we thank Anne Alstott, Ian Ayres, Fred Goldberg, Jacob Goldin, Jacob Hacker, Henry Hansmann, Zach Liscow, Yair Listokin, Ben Lockwood, Daniel Markovits, Kyle Rozema, Tom Tyler, Edward Zelinsky, and session participants at the National Tax Association’s annual conference. We contributed equally to this paper, and any errors are our own.
A stubborn question in tax law and policy is why some spending programs are organized through the tax code rather than as direct outlays. Both methods are common. For example, Social Security payments are issued directly into the recipients’ accounts.1 In contrast, the Earned Income Tax Credit (EITC), as its name suggests, takes the form of a credit against the recipient’s income tax: eligible beneficiaries simply owe less or are entitled to a refund when they file their taxes.2 Social Security payments are made by the Social Security Administration.3 The EITC, like all federal tax expenditures, is handled centrally at the IRS.4
For some tax experts, the widespread use of “tax expenditures”—policy spending through the tax code that departs from taxing “accepted concepts of net income”5—is concerning. The great tax scholar Stanley Surrey argued famously that one of several problems with tax expenditures is that they are a disguised form of spending, spending that is poorly managed by Congress and almost completely overlooked by the American people. Most tax expenditures, wrote Surrey, “seem almost to live a life of their own, undisturbed and unexamined,” and with “[n]o agency [that] really studies or controls them.”6 Surrey concluded unhappily that this “is no way to run a tax system.”7
In the decades since Surrey’s writing on the subject, questions about the role and value of tax expenditures have become only more relevant. While the Tax Reform Act of 1986 eliminated many tax expenditures in the Internal Revenue Code, total tax expenditures have since grown and are now more than eight percent of GDP—$1.4 trillion in 2014.8 In the aggregate, individual income tax expenditures are now larger than either defense spending or Medicare spending.9 Yet despite the fact that academic studies of tax salience and behavioral taxation have become increasingly popular, basic questions about the public perception of tax expenditures remain largely unanswered. In particular, many of Surrey’s original concerns have avoided rigorous testing: is the true cost of a tax expenditure really hidden or diminished by virtue of being part of the tax code? Is it true that the public sees spending through the tax code as different?
These questions are urgent for an additional reason. In some ways, Surrey got his wish. Tax expenditures are subject to more oversight than ever. “Tax expenditure budgeting,” an annual process by which the federal and state governments account for their spending through the tax code, has become the American norm. Federal law requires the United States Treasury to produce an annual tax expenditure budget,10 and most states have adopted similar processes.11 These procedures would no doubt please Surrey, but the continued growth of tax expenditures would not. Indeed, that growth presents a paradox: spending through the tax code has continued to rise faster than government spending has as a whole, despite repeated efforts to publicize and rein in the costs of tax expenditures.
Our Note helps to explain this apparent paradox. We offer evidence on how the American public thinks about tax expenditures as opposed to spending programs organized outside the tax code. We obtain this evidence through a survey experiment that tests how support for (and perceptions of) public spending policies vary not based on the substance or expense of a policy, but simply on whether a policy is described as a tax expenditure or direct outlay. We use Google Consumer Surveys (GCS) to ask panels (which aim to be demographically representative of the United States Internet-using population) for their views on a variety of policy options, including hypothetical subsidies for the housing market and the disabled.12 We keep the substance and total cost of the policies functionally identical. We then ask one group for its views on a policy that is described as a direct expenditure, and the other for its views on a policy that is described as a tax break.
By asking similar panels for their views on such questions, we are able to study the way in which a policy’s framing affects public support and public perceptions, and we are able to isolate this framing effect in an empirically rigorous manner. In particular, this method allows us to test the hypothesis that citizens are more likely to support “hidden” spending that occurs through the tax code, rather than “direct” spending that occurs through another policy mechanism, such as payment via cash or check. And this method allows us to test whether tax expenditures simply appear less expensive than direct outlays.
Our results are strongly consistent with both hypotheses. Americans are more likely to support policies when they are described as tax expenditures, and they are more likely to view tax expenditures as cheaper than direct outlays. In our baseline comparisons, respondents were ten percentage points more likely to support our hypothetical, economically equivalent policies when we framed them as tax breaks rather than as direct outlays. These results held true across a variety of policy areas, and they held true when we varied the amount of information that we offered about how tax expenditures work. Respondents were also more likely to say that a program added “a lot” to the deficit if it was described as a direct outlay instead of a tax expenditure, even though the programs were listed with the same explicit cost.
These results make several contributions to the existing literature. First, we apply a better methodology to a wider range of contexts than did past studies to help answer significant outstanding questions in the political science, economics, and tax-law literatures. Second, we test the robustness of the idea that “spending through the tax code” produces a framing effect by varying the amount of information we provide to our survey respondents; this question speaks to the issue of why citizens are so inclined to favor tax expenditures. Finally, we connect our results to key debates in the economics and political science literatures, and we discuss the implications for economic welfare, modern tax law, and democratic decision making about public spending.
The rest of this Note is divided into five Parts.13 Part I positions our contribution in the relevant literature on tax expenditures and the behavioral-economics approach to taxation. Part II describes our methodology in more detail. Part III offers a fuller description of our results. Part IV discusses limitations and implications. Part V concludes.
In the tax-law literature, the general distinction between spending inside and outside the tax code is typically associated with the work of Stanley Surrey, who reportedly coined the term “tax expenditure.”14 Surrey is well-known for emphasizing that “[t]he federal income tax system consists really of two parts,” one which “comprises the structural provisions necessary to implement the income tax,” and another that “comprises a system of tax expenditures under which Governmental financial assistance programs are carried out through special tax provisions rather than through direct Government expenditures.”15
Surrey had many criticisms of the tax-expenditure system: he thought it confused Congress,16 muddled the administration of social programs,17 and made the tax code more complicated.18 But a particularly notable theme of Surrey’s work is that tax expenditures are “hidden.”19 Despite the fact that tax expenditures are now identified and budgeted like other expenditures—a longtime goal of Surrey’s20—the hidden nature of tax expenditures is a theme that still runs through contemporary literature on taxation and public policy. In his book The Hidden Welfare State, for example, the political scientist Christopher Howard writes that “tax expenditures with social welfare objectives are largely invisible to citizens, policy makers, and academics who study U.S. social policy.”21
The premise that tax expenditures are or would be treated differently from direct outlays is, in some sense, counterintuitive from the perspective of public finance. Most scholars of public finance would consider tax expenditures to be “conceptually equivalent” to direct outlays.22 Indeed, tax expenditures can always be described in a manner that makes them seem identical to direct spending—one in which (as Howard puts it) “taxpayers write a check to the government for their full tax liability, and the government issues them a check to cover those activities exempted from taxation.”23 As consumers of government benefits, taxpayers should value a dollar of cash just as much as a dollar of tax relief. As taxpayers who fund government programs and vote for elected officials, they should view a dollar of government spending as equivalent to a dollar of forgone tax revenue.24 Considered at this level of abstraction, public support for a spending program should not depend on whether that spending goes through the tax code.
But do real-life taxpayers actually treat a dollar of direct spending as equivalent to a dollar of tax expenditure? In our opinion, the tax-expenditure literature generally answers this question in the negative—but it has received surprisingly little empirical attention. One of the few tax-law papers to study this subject directly is a 2005 article by Edward Zelinsky, which used a student survey to assess how different types of financial support for firefighters affected how respondents perceived their “volunteer” status.25 Zelinsky’s subject matter was drawn from a real policy dilemma: increasingly stringent requirements for training firefighters make it difficult for communities to recruit volunteers, but many communities are nonetheless reluctant to pay firefighters directly. As a result, some communities apparently offer tax breaks to their volunteer firefighters, such as property tax reductions.26 The idea is that the tax breaks offer a financial inducement to become a volunteer firefighter—but not an inducement that is so explicit as to threaten the volunteer status of the position.
To see how the different forms of compensation affected public perceptions of these “volunteers,” Zelinsky distributed questionnaires to several groups of law students at the Benjamin N. Cardozo School of Law.27 Half were asked whether a direct payment affected the volunteer standing of firefighters, while the other half were asked whether a tax exemption did the same thing.28 Zelinsky finds strong evidence that respondents are more likely to view recipients of a tax break (rather than a direct payment) as volunteers in good standing.29
An older attempt to study such questions empirically is a book chapter by Steven Sheffrin.30 Sheffrin looks primarily at how the public conceives of a fair sharing of the tax burden, but he also briefly considers the question of how public views diverge from views commonly held by professional economists.31 To see if the public shares economists’ view that tax expenditures and direct outlays are equivalent, Sheffrin asked 150 students in an economics class about their opinions of an investment subsidy plan for firms.32 He described the plan in one instance as a $1 million tax break and in the second as a $1 million payment.33 In the baseline scenario, the students had a similar view of the favorability of the tax break and direct subsidy programs.34
However, when asked a follow-up question in which tax credits reduced the companies’ tax liability to zero, the students had a substantially more favorable view of the direct subsidy program.35 In other words, they preferred the program under which the firms “paid” some taxes, even if this payment was exactly offset by a check from the government. Sheffrin attributes this result to respondents’ belief that “[e]ntities should pay taxes” and the fact that they were not looking solely at the company’s net position.36
While the tax literature has recognized the importance of tax expenditures, therefore, relatively little empirical work has been done on whether the public actually thinks of them as different from direct outlays.
Outside of legal scholarship on taxation, two recent political science papers have used survey evidence to study the public’s perception of spending programs. Christopher Faricy and Christopher Ellis asked university students about their opinions of three social spending programs: the mortgage interest deduction, the deduction for retirement savings, and food stamps.37 They presented the programs to some students as tax expenditures and to others as the equivalent direct outlays,38 and they found weak evidence that respondents preferred identical programs couched as tax expenditures.39 For each program, the tax expenditure equivalent was more popular than the direct outlay, but this difference was small and only statistically significant for one of the three programs.40 The authors also found that the effect is bigger for Republican students than for Democratic students.41
Jake Haselswerdt and Brandon Bartels take a similar approach in an unpublished working paper.42 Unlike the other papers mentioned above and below, Haselswerdt and Bartels do not use a student sample. Instead, like us, they use a survey that attempts to reach a representative sample of the U.S. population.43 They asked about three programs: the mortgage interest deduction, job training, and paid parental leave.44 Among other things, they describe the programs as either a tax expenditure or an equivalent direct outlay.45 They find that each program is significantly more popular when described as a tax expenditure.46
The studies above follow a method that is widely employed in behavioral research: asking two demographically similar groups of respondents a question in which the substance is identical but the framing is different.47 This research agenda seeks to isolate what is now called the framing effect.48 Amos Tversky and Daniel Kahneman popularized this approach in a classic series of behavioral experiments,49 finding, for example, that identical life-saving policies are more popular when the outcomes are framed in terms of lives saved rather than lives lost.50 As applied to tax expenditures, Zelinsky derives a similar result, finding that “policies unacceptable when framed as direct expenditures become supportable when labeled as tax subsidies, even though the economic substance of the policies is the same.”51
Political scientists have also considered the relationship between public spending and framing. They have studied, among other things, how different political parties frame their approaches to spending;52 how political framing differs from political persuasion;53 and how framing interacts with political competition and the formation of citizen preferences.54
In continuing to study how tax expenditures are viewed in comparison to direct outlays, we also join a growing literature on behavioral-economic approaches to tax policy.55 This field is increasingly interested in how general principles of behavioral economics can be applied to tax policy, and in developing original experiments that might inform tax-law design.56 Many of these studies find that individuals react to taxes in ways not predicted by standard economic theory.57
Our approach complements and builds on existing work in several ways.
We believe our sample is a substantial improvement over existing work. First, we reach a diverse, non-expert sample that is close to representative of the electorate. All of the other papers that compare tax expenditures and direct outlays, except Haselswerdt and Bartels’s working paper,58 use student samples. Such samples can be problematic. For example, Sheffrin’s study is designed to illustrate that the public’s views diverge from the views of conventional economics, but the students he surveys were in an economics class.59
That is a specific example of a general phenomenon. Student samples often differ from the general population in systematic ways that matter. One survey paper found that student responses “differed substantially” from those of the population at large in 48% of social science studies where they could be compared.60 For our purposes, students are likely to differ substantially from the average population on at least two important dimensions: their level of education and their experience paying taxes.
Highly educated people, particularly those trained in quantitative fields, may be less susceptible to framing.61 Zelinsky hypothesizes that this explains differences in the framing effect among his students,62 and this may be one explanation for why Sheffrin does not initially find a framing effect.
Moreover, students are less experienced with tax expenditures than the population as a whole, and this might make them poor proxies. For example, like Faricy and Ellis, and Haselswerdt and Bartels, we ask about the mortgage interest deduction. There are some differences in wording, but all three studies asked about support for the deduction in light of its $100 billion annual cost. Our study and that of Haselswerdt and Bartels—studies aimed at the population at large—found that support was roughly 55%.63 However, when shown the costs, only 21% of Faricy and Ellis’s students supported the deduction.64 Faricy and Ellis’s students were likewise not affected by the frame, whereas both our paper and that of Haselswerdt and Bartels found a difference in support of 25 to 35 percentage points.65 As a result, we believe that our results are a more reliable measure of the framing effect than those derived from student samples.
In addition to studying a more representative sample, our approach differs from previous work in several ways. First, we explore the framing effect in a range of contexts, including support for housing programs, aid to the disabled, and the question of whether people prefer to receive cash or equivalent tax credits and whether they perceive tax expenditures to be less expensive. Taken together, this variation across policy areas helps show that the framing effect is wide-ranging and sheds light on the source of that effect.66
Second, we show that the public’s preference for tax expenditures persists even when we describe the mechanics of a tax expenditure in some detail. This helps ensure that the source of the framing effect is not confusion over how the tax programs work or who gets the benefit. In their surveys, Haselswerdt and Bartels, as well as Faricy and Ellis, do not clarify the mechanics of their programs in the same way. For example, while Faricy and Ellis express concern that citizens do not understand tax expenditures, they do not illustrate the insight empirically.67 Instead, the questions in their study and in that of Haselswerdt and Bartels are somewhat unclear. For example, Haselswerdt and Bartels ask respondents whether they support making those who take a job retraining class “eligible for a tax break, that is, a reduction in the income tax they owe to help cover the cost of the class,” and similarly describe the direct expenditure as simply “a cash payment to help cover the cost of the class.”68 Respondents may simply assume that the payments will be larger or smaller depending on which vehicle is used for payments. Even in other questions in which the researchers specify the total cost, the programs are not necessarily equivalent in terms of how much each person receives, particularly since often the tax expenditures are phrased as deductions, but the direct outlays appear more similar to credits.69
Third, we avoid using existing programs—except the mortgage interest deduction—to measure the framing effect. All of the programs in the Faricy and Ellis study are currently implemented in the United States,70 as are two-thirds of those in the Haselswerdt and Bartels study.71
Asking about these “status quo” programs is potentially problematic for a number of reasons. Rational respondents might prefer not to change only the mechanism by which an existing program is delivered, since switching the mechanism is presumably costly and the substance of the program will remain the same. In addition, respondents might simply be confused as to why an existing program is being framed as a hypothetical choice. They may pick the status quo when a choice is too complex; they may pick it in protest.72 Or respondents may simply be affected by the well-known status quo bias, which has been extensively studied in behavioral economics.73 The notable point is that, in each of these scenarios, something other than a question’s framing—as either a tax expenditure or direct outlay—may partially influence the response.
Such status quo issues are well understood to be a problem in survey design.74 Indeed, the difference between new and status quo options can be enormous, even when the real payoff is the same. One well-known study of individual biases toward risk, for example, found that only 27% of survey respondents were willing to pay $700 for a safety measure that guaranteed a 0.5% reduction in the risk of an injury in a given year; that number jumped to 60% when the safety feature in question was described as an industry standard.75 In at least one influential study of stated-preference survey design, these problems were considered worrying enough that respondents who always selected a status quo were categorized along with respondents who selected “I don’t know.”76 Haselswerdt and Bartels are aware of this issue, and in fact one of their goals is to measure whether the status quo “communicat[es] to the public how different problems should be viewed and solved.”77 But we are not sure that they can separate this effect from the other status quo issues discussed above.
Fourth, and perhaps most important, we consider the implications of these findings in ways that differ greatly from previous work. We explore the likely underlying causes of the framing effect, as well as the relationship between framing effects and welfare economics. We then highlight the implications of citizens’ preference for tax expenditures for the way in which the Internal Revenue Code is written. We use our results to suggest a new and counterintuitive explanation for why tax expenditures have grown: because the public is actually paying more attention to government budgets. We also suggest that the public’s fondness for tax expenditures should be added to the traditional list of factors that drive the increasing complexity of the tax code and is perhaps one of the best explanations for why ordinary taxpayers find the tax system so maddeningly complex. Finally, we make a new connection between tax expenditures and other areas of the law in which increasing transparency has potentially serious drawbacks.
We developed a survey instrument using Google Consumer Surveys (GCS), a popular and relatively inexpensive online survey tool designed for use by both companies and researchers.78 GCS is a relatively new service—the product was launched in March 201279—but it has already been used to produce peer-reviewed papers in a variety of fields, including political science,80 psychology,81 and business.82
We discuss in detail below the extent to which our panel is likely to be representative of the U.S. population. We conclude that there are some reasons to believe that our panel is not fully representative—but, in the end, it is likely to be close, and it offers substantial advantages over the classroom surveys utilized in existing publications.
Unlike several other online survey tools—which hire a representative panel of respondents83 or otherwise manage a marketplace for survey questions84—GCS presents survey questions to general Internet users in the form of a “wall” preventing access to premium Internet content. Just as a visitor to a website might be required to watch an advertisement or pay a fee before accessing premium content (typically known as a “paywall”), GCS lets Internet users answer a survey question.85 This “surveywall” is intended to be relatively brief and painless. Google’s theory is that, “[b]y reducing the burden [of responding to a survey] to just one or two clicks, we increase the response rate of the survey.”86 According to Google, this produces an average response rate of 16.75%.87 Google argues that this response rate compares favorably to other commonly used Internet or traditional phone survey tools.88
Unlike many other survey tools, Google does not ask respondents to report their age, gender, location, income, or other demographic information. Instead, Google reports that “Consumer Surveys infers approximate demographic and location information using the respondent’s IP address and DoubleClick cookie,” which Google uses to “ensure each survey receives a representative sample and to enable survey researchers to see how sub-populations answered questions.”89 Not asking for this information improves response rates and allows questions to be asked free of any survey “priming.”90 We discuss the accuracy of Google’s methods below and include further details in a short Appendix.
The representativeness and reliability of the GCS survey population have been tested and discussed favorably in two studies—one from the Pew Foundation91 and one from Google itself.92 Google’s study compared GCS surveys to “gold standard” national telephone surveys—one private and one conducted by the Centers for Disease Control—by using questions identical to those in the “gold standard” surveys.93 Google also hired two well-respected Internet survey firms to ask the same questions.94 The search giant concluded that its own survey tool outperforms other Internet survey providers on several benchmarks.95
The Pew Research Center, meanwhile, performed “a series of tests covering a wide range of topics and question types to compare results from Pew Research telephone surveys to those obtained using the Google Consumer Surveys method.”96 GCS performed relatively well overall. Across forty-eight questions, the median difference between GCS and the Pew surveys was 3%.97 Of particular interest for our survey, Pew concluded that “the Google Consumer Surveys sample appears to conform closely to the demographic composition of the overall internet population.”98 In terms of political views, GCS respondents were “broadly similar [to the U.S. population], though some larger differences were observed.”99 Nor was there a consistently conservative or liberal bent to these differences.100 In fact, Nate Silver concluded that GCS was the second most accurate 2012 presidential poll, beating out CNN, Quinnipiac, Gallup, and YouGov, among others.101
Nonetheless, there are good reasons to believe that GCS panels are not perfectly representative of the entire U.S. population. First, Google surveys only the U.S. Internet-using population, and 15% of the U.S. population does not use the Internet.102 These individuals are disproportionately older and less educated, and this likely biases any survey of Internet users.103 Similarly GCS’s model does not guarantee that each panel is a random sample of all Internet users. Nevertheless, the research discussed above suggests it is quite likely that GCS is close to representative—and certainly a large improvement over the classroom panels used in prior research. It is also likely to be much more representative than Mechanical Turk, where survey respondents are paid for each survey that they complete, raising a number of problems—including self-selection and potential misrepresentation.104 In spite of these problems, Mechanical Turk studies have found generally receptive audiences.105
GCS does have some important drawbacks. First, we can ask only short questions. Google imposes a 175-character limit on questions, which forced us to think hard about how we worded our questions, and made it a challenge to ask several questions about technical tax policy.106 Second, because GCS questions pop up instantaneously, respondents see the question before committing to answer—an issue that affects most Internet surveys but is arguably more problematic in our case. Third, the fact that individual respondents will see only one question makes it impossible to study an individual respondent’s answers across questions.
On the other hand, GCS also offers some practical benefits. Consumer Surveys are inexpensive—which allowed us to gather many thousands of fairly representative observations at low cost—and have a relatively high response rate. In our surveys, an average of 18% to 24% of Internet users who saw each question responded. In addition, the fact that Google infers demographic data means that we did not need to rely on respondents’ self-reporting to obtain a representative panel. We also did not have to rely on respondents’ self-reporting about sensitive matters like age and income. And, unlike Mechanical Turk, GCS respondents do not answer questions for money.
The central goal of our survey was to ask two demographically equivalent groups of respondents whether they supported economically equivalent policy proposals—one described in the form of a tax expenditure, the other in the form of a direct outlay. Our central hypothesis, consistent with the notion that spending through the tax code disguises the true cost, was that respondents would be more likely to support policies that take the form of tax expenditures.
We also tested several secondary questions by varying the details of our questions. One secondary question was whether the hypothesis above works because respondents view tax expenditures as “cheaper.” We tested this by asking respondents how they perceived the costs of equivalent tax expenditures and direct outlays, and whether they would prefer to receive a tax credit or a check. Another secondary question was whether taxpayers’ preference for tax expenditures would hold true across a range of policy types; we tested this question by asking about hypothetical policies that support the housing market and hypothetical subsidies for the disabled. A third secondary question was whether respondents’ views vary depending on whether the policy in question is a new, hypothetical policy, or an existing and salient policy; we probed this distinction by asking about the home mortgage interest deduction.
A fourth secondary question was whether respondents’ views change depending on how much information we provide about the policy proposals in question. Because tax policy is relatively technical, in our view it is important to try to distinguish between the framing effect and simple ambiguities in (or misunderstandings of) how the policies in question operate. For example, it might be apparent to those steeped in tax law or public finance that a dollar of “refundable tax credit” is the conceptual equivalent of a dollar in cash—but perhaps not to the average citizen. To get a sense of how this affects our results, we varied our descriptions of the tax expenditure. In some questions, we spelled out in detail how refundable tax credits operate; in others we did not. We also tested to see whether using the term “tax expenditure” itself affects the results.
Finally, we asked several questions designed to rule out common alternative explanations that might indicate a bias toward spending through the tax code, and to see whether policy preferences diverged from individual consumption preferences.
In our first wave, we asked the following nine questions107:
Q1. Would you support the government offering annual $1000 cash payments to each family, to help cover rent?
Q2. Would you support reducing each family’s taxes by $1000 to help cover rent? If a family owes less than $1000, they get the rest in cash.
Q3. Would you support the government offering a $1000 refundable tax credit for each family, to help cover rent?
Q4. Do you support the government letting homeowners deduct their mortgage interest payments?
Q5. Would you support the government replacing existing tax aid for homeowners by matching 25% of their mortgage interest with cash?
Q6. Tax aid for homeowners costs $100 billion a year. At the same cost, would you instead support matching 25% of their mortgage interest with cash?
Q7. Tax aid for homeowners costs $100 billion a year. Do you support the government continuing to let them deduct their mortgage interest payments?
Q8. Would you support an annual $1000 government cash payment to each disabled person?
Q9. Would you support a tax credit reducing each disabled person’s taxes by $1000? If a person owes less than $1000, he or she gets the rest in cash.
We released these questions in November 2013, spread over a weekend and three weekdays.
The demographics of respondents and responses we received did not vary by day of the week, indicating that the pool of potential respondents was similar during weekdays and weekends.108 We received a little over 1,000 responses to each question. However, as can be seen in Table 1 below, not all of these responses were usable since some respondents chose to opt out and others lacked full demographic data.109 There is some evidence that more respondents opted out of the more complex questions.110 But we remain confident that this skew is relatively minor. Even if, for the sake of argument, each additional person who opted out of the more complex questions reduces the significance of our results, our results still show that tax expenditures are substantially preferred and the results are still statistically significant.111
In February 2014 we conducted a relatively small second survey designed to ascertain whether applying the label “tax expenditure” to spending through the tax code made any difference. In April 2014 we asked a larger sample about their preferences for personally receiving direct-payment subsidies or tax credits. Finally, in September 2014, we asked a large sample about how they perceived the costs of direct outlays and tax expenditures.
Our results are summarized in the following tables.
survey of preferences for using direct subsidies or tax credits
survey of preferences for receiving direct subsidies relative to tax credits
survey of respondents’ perception of policy costs
comparisons of survey answers
Our results strongly confirm the central hypothesis that individuals prefer spending through the tax code to direct expenditures. In our baseline comparisons, respondents were about 10 percentage points more likely to support policy proposals when they were described as tax expenditures rather than as direct outlays. This held true across policy types: respondents were 10.5 percentage points more likely to support a subsidy for the disabled when it was phrased as a tax expenditure and 9.8 percentage points more likely to support a subsidy for the rental market when described in similar terms. Respondents also very strongly preferred the existing mortgage-interest deduction to an alternative policy in which the government directly matched a portion of homeowner mortgage payments. All of these results are statistically significant at the 1% level.
We had hypothesized that more information about how tax expenditures work would push some respondents to think about the underlying similarities between spending through the tax code and direct spending. For example, noting that “[i]f a family owes less than $1000, they get the rest in cash” probably underscores these similarities more than explaining that each family gets “a $1000 refundable tax credit.” But surprisingly and notably, respondents were only slightly more likely to support a tax expenditure when we offered less information about it, and the difference was not statistically significant. The additional information on the mechanics of the tax expenditure did not appear to alert respondents to the functional equivalence of tax expenditures and direct outlays, at least not to a statistically significant extent.
Similarly, in our later survey, when we asked people whether they preferred a $1000 check or an equivalent tax credit, more information did not reduce the impact of the frame: when given more information, people were a bit more likely to conclude that the two options were equally good, but that difference was not statistically significant. On balance, we found these results surprising and believe that they have important implications for tax policy (discussed below). We do not find that using the term “tax expenditure” has any effect relative to simply describing the tax reduction.
We also found that respondents have a strong personal preference for receiving a check (rather than a tax reduction) from the government: that is, while individuals would prefer to see tax expenditures enacted as policy, they would prefer to receive direct outlays. This would be consistent with the hypothesis that tax expenditures seem cheaper: the other side of the coin is that they might seem less valuable.
Finally, and also consistent with the hypothesis that tax expenditures seem cheaper, we found that respondents were less likely to think that tax expenditures contributed substantially to the deficit. Specifically, 37% of respondents said that a $6 billion dollar direct spending program added “a lot” to the deficit, while only 32% of respondents said equivalent spending through the tax code would add as much.
The results presented in the previous Part support the idea that individuals prefer spending through the tax code to direct expenditures, and these results are consistent with the framing effect. The results also suggest that the framing effect holds true in a sample of survey respondents that are representative of the national Internet-using population, across a range of policy types, and when more information on tax expenditures is provided. Our evidence also suggests that the framing effect is at least partly driven by a perception that tax expenditures are cheaper from the point of view of the fisc. Likewise, individual respondents are less likely to view a tax break as equally valuable as a direct subsidy of the same size.
In this Part, we discuss our results more fully. In Part IV.A, we discuss some limitations and hypotheses that we hope will provide a basis for future research on why taxpayers prefer spending through the tax code. In Part IV.B, we discuss the implications of our results for tax law and policy.
Our results are limited in a few respects. Some of these are general problems of single-question Internet surveys: for example, our respondents did not spend hours thinking about these questions. The average response time was about twenty seconds.112 While this response time in some ways limits inferences about respondents’ “true preferences,” as discussed below, it also perhaps captures how some voters actually perceive and evaluate these questions in the political marketplace. In the context of political advertisements and platforms, voters do not necessarily spend long periods of time puzzling over the details.113
Nonetheless, a few remaining issues strike us as especially interesting and relevant for further discussion and research. In this section, we focus on two possible explanations for why respondents seem to have a stubborn preference for tax expenditures. First, we discuss the well-known framing effect and why it might exist here. Second, we discuss possible reasons for why a “rational citizen” or “rational voter” might prefer to channel spending programs through the tax code. While we are ultimately skeptical about these latter explanations, we think they are important to discuss alongside our results.
As described above,114 there is a large literature in psychology, economics, and political science attempting to clarify how the framing of a decision affects responses. If one takes the view that a dollar spent inside the tax code is functionally equivalent to a dollar spent outside of it, then our results support the existence of a framing effect in this context.
Our results suggest that taxpayers prefer tax expenditures in part because they perceive them to be less expensive for the public fisc. Our respondents felt that direct spending programs increased the deficit more substantially than equivalent tax expenditures, even though the explicit cost was the same. The other side of this coin is that our respondents were more likely to prefer receiving a dollar of cash over a dollar of tax relief because they perceive a dollar of cash to be more valuable, even when we subtly emphasized that they are the same (for example, in Question 13).
This fits our intuitions—and the scholarly literature—about how citizens view the tax code. Providing a dollar of tax relief might be viewed as letting a person keep something she already possesses; some citizens might not even realize that they are in fact receiving a benefit from a government policy.115 This taps into an intuitive and common—though, in many ways problematic—assumption about the nature of taxation and government, in which one’s “pre-tax income” represents a natural state of justice that precedes government intervention. We think this view is somewhat misguided: after all, one’s pre-tax income depends crucially on a system of public order that could not exist without government intervention (and, hence, taxation). But the view of pre-tax income as naturally just is commonly held.116
This theory has a subtle connection to the first studies of the framing effect, in which respondents preferred a triage policy that emphasized the lives saved rather than lives lost, even if the totals were the same.117 Likewise, we find it plausible that respondents prefer policies that let citizens keep their own hard-earned money to those that give citizens benefits from an amorphous government larder.
We think that this view likely explains much of the framing effect we witness here. We also think that teasing apart these explanations more directly would be a valuable direction for future research.
An assumption of our paper is that tax expenditures and direct outlays of equivalent size are, in fact, equivalent. But we do not (and cannot) eliminate every possible reason why a “rational” voter might prefer tax expenditures to direct outlays.
That said, we do attempt to rule out some of these alternative explanations. For example, it could be that voters view tax expenditures as more politically stable or permanent. Zelinsky, for example, raises this prospect when he notes that a tax expenditure, “if embedded in a permanent tax code, may be more secure politically than a cash payment, which must be appropriated annually.”118 While we think this is plausible, we are unconvinced that political stability explains a large proportion of the apparent framing effect. The results from our later waves of survey questions—which included a question emphasizing that both the tax expenditure and the direct outlay are “one-time” only—still display a robust framing effect.
Alternatively, it could be the case that voters and policymakers view spending through the tax system as less amenable to “regulatory capture” than a program administered by a specialist agency that interacts repeatedly with a specific part of the economy.119 While we have not tested this hypothesis, we think it unlikely that aversion to regulatory capture explains a large proportion of the apparent framing effect. Details of tax and spending administration are not particularly salient to the public,120 and we think it improbable that most taxpayers respond on the basis of a difference in program administration—much less a difference in the likelihood of capture based on the administrative scheme.
Finally, spending through the tax code might be preferred based on how voters evaluate a tradeoff between specialization and coordination. If voters think that certain spending activities are most efficiently clustered together in the tax system, then they will prefer tax expenditures over direct outlays.121 However, given the simplicity of the programs that we proposed in our experiment, we find it unlikely that this last explanation plays a role in explaining our responses.122
We should note one element of our results that we think can be explained by rational behavior: the questions about mortgage interest showed the widest gap between the proposed direct spending program and existing spending routed through the Code, and there are many plausible explanations for these results. We felt it important to ask these questions because they deal with one of the best-known tax expenditures. Because the mortgage deduction is an existing program, the public’s preference for keeping this policy might simply reflect a quite rational preference for the status quo—as opposed to switching to a new and very similar system and incurring related costs.123 That said, the status quo bias cannot explain the entire preference for spending through the Code, since the framing effect is seen in responses to questions that propose hypothetical programs unrelated to the status quo.
Our findings have several implications for tax law and the debate over tax expenditures, and we discuss these implications here.
One set of implications concerns public welfare. Indeed, the framing effect raises a question that appears frequently in behavioral economics: how can we evaluate the welfare consequences of seemingly irrational public preferences?
If the arguments in Part IV.A.2 are correct, then taxpayers are not rational in the manner predicted by classical economics: they should not prefer a dollar of tax spending to a dollar of direct spending. Public support and public welfare should be the same in both cases—but we show that this does not hold true.
Might public welfare remain the same, even if public preferences are susceptible to the framing effect? Some argue that this might be the case. Weisbach and Nussim, for example, suggest that outcome equivalence implies welfare equivalence.124 In other words, the welfare effect—that is, the effect on utility—of a $1000 check should be the same as the welfare effect of a $1000 tax credit, even if the public says it prefers one over the other.125 But our results suggest that welfare equivalence does not necessarily hold true: people may react differently if they receive the same $1000 in a different manner (having their taxes reduced as opposed to paying the higher tax and receiving a $1000 check).
These results showing the framing effect are nonetheless hard to interpret. Indeed, some scholars have concluded that this kind of “equivalency framing,” in which two identical options receive different levels of support depending on how they are described, renders preferences “uninterpretable.”126 Nonetheless, in studying the potential divergence between public welfare and public preferences, we contribute to the growing interest in interpreting the welfare implications of behavioral economics.127
Perhaps more centrally, our results also have implications for a longstanding debate in tax law and policy: why have tax expenditures continued to increase, despite repeated efforts to expose and restrain their cost? Since Surrey’s original writings on the subject, there has been a concerted effort at both the federal and state levels to make tax expenditures a more public part of the budgeting process—in other words, to make sure that policymakers are forced to publicly account for their spending through the tax code.128 According to some recent critics, however, these efforts have failed to slow the growth of tax expenditures.129
There have been several prominent efforts to address this apparent paradox.130 One line of analysis emphasizes that, but for the budgeting requirements, the growth rate of tax expenditures would have been “even more robust.”131 Another argues that efforts to publicize tax expenditures have so far been meek and ineffective. Edward Kleinbard, for example, has argued that the current federal tax expenditure budgeting scheme “is expressly designed to avoid leaving any visible imprint on the budget, and the programs so favored have not been forced to compete with other spending programs for scarce Government resources in other committees or among the members as a whole.”132 A third line of analysis, favored by Edward Zelinsky, argues that tax expenditure budgeting “legitimates tax expenditures and encourages a scramble [among interest groups] for parity in the form of comparable tax benefits.”133
Our results have implications for the debate over why tax expenditures have continued to grow. In particular, we show that there is likely to be more demand for spending through the tax code even if more transparent “budgeting” or “disclosure” takes place. A major constraint on a legislator’s willingness to enact new spending measures is the way in which new policies will be viewed by constituents—and, naturally, politicians running for elected office have incentives to propose policies and frame policies in a manner that appeals to the median voter. Our results suggest that constituents may view tax expenditures more positively than general outlays, and this in turn suggests that the scales might always be tilted in favor of spending measures that flow through the tax code.134
Spending through the tax code is what Chong and Druckman refer to as a “strong frame”: a frame that “emerge[s] from public discussion as the best rationale for contending positions on the issue.”135 Since the “typical political strategy is to connect a proposal to a positive idea or value that is widely available in the population,”136 we believe that the basic and stable public bias in favor of spending through the tax code helps to explain the enormous rise of tax expenditures.
A powerful tradition of legal thought suggests that transparency is normally good (or, in Justice Brandeis’s memorable phrase, that sunlight is the best disinfectant137). Likewise, the emphasis of the conversation on tax expenditures has been about making them more transparent through a process of tax expenditure budgeting.138 The thought is simple: “[I]f policymakers, the media, and the general public lack information about tax expenditures, they cannot fully participate in decisions about how to allocate state resources.”139
The question of when and how to give the public information about tax expenditures is very much alive. Tax transparency is a perennial subject of the policy debate,140 and the tax expenditure budgets are becoming increasingly common. The federal government is required to produce an annual tax expenditure report, as are most states.141 While tax expenditures still typically receive less scrutiny than direct spending,142 the trend continues to favor more disclosure about tax expenditures. In some states, like Connecticut143 and Minnesota,144 the reporting requirements and publications are extensive. Furthermore, as of 2011, only four states did not require some form of tax expenditure report, falling from nine in 2009.145
This trend toward tax expenditure budgeting and related transparency is usually viewed as beneficial. As the Joint Committee on Taxation recently put it, “[t]ax expenditure analysis can help both policymakers and the public to understand the actual size of government, the uses to which government resources are put, and the tax and economic policy consequences that follow from the implicit or explicit choices made in fashioning legislation.”146 Yet despite the enthusiasm for tax expenditure disclosure, the growth of tax spending has eclipsed the growth of government spending as a whole.147
Our results suggest that recent efforts to expose the true costs of tax expenditures can be ineffective and perhaps even counterproductive. In broad strokes, existing efforts to disclose tax expenditures might have two effects. First, they might make the public more aware of the general equivalence between taxing and spending programs, and perhaps as intended, educate the public on the true costs of tax expenditures. Second, however, efforts to disclose tax expenditures might make the public think more about how the budget is organized and thereby normalize the broad range of policies that are organized through the tax code. This normalization in turn may make it seem perfectly appropriate to use tax expenditures to implement a new social benefit scheme in the future, whereas, without the additional information, tax expenditures seemed unusual and only appropriate in specific areas.148
If such normalization occurs, we should expect it to increase both the relative spending on tax expenditures and total spending. Relative spending increases because legislatures, responding to public preferences, substitute spending through the tax code for direct spending. Absolute spending increases because the public thinks spending through the tax code is spending on the cheap.149 Therefore, if the public consistently prefers spending through the tax code—even if the public has more information about the equivalence between that spending and direct outlays—providing more information about the size and prevalence of tax expenditures might actually make tax expenditures more popular.
Our results do suggest that the framing effect is relatively robust to additional information—that is, in our survey the framing effect persisted even when we provided additional information. People’s opinions did not change when they were given clear information about the mechanics of the tax expenditure. Even when respondents were given explicit information about the cost of the program—exactly the type of information that they would see in a tax expenditure budget—they continued to think that tax expenditures should be classified as less expensive.150
More information might help. People who spend a lot of time thinking about the equivalence of tax expenditures and direct outlays—like tax professors and other well-informed tax mavens—are probably less likely to care about the frame. But what our results do show is that the type of information typically disclosed in tax expenditure budgets does not have much effect on the public’s preference for tax expenditures. This isn’t necessarily a bad thing; it might simply mean that the public is getting more of what it wants and that people are happier as a result. But the public’s stubborn preference for tax breaks does suggest that if the modern-day Surreys want to press their campaign against tax expenditures, they might want to consider a different approach.
In some ways, the notion that tax expenditure budgets might normalize tax expenditures should not be surprising: it connects to a growing literature that is skeptical of whether mandatory disclosure will always help.151 Disclosing CEO salaries, for example—a step intended to shame companies away from excessive compensation packages and curb agency problems—might actually increase the overall level of CEO pay, since it makes price competition for CEOs all the more explicit, and CEOs likely want to be paid more than average.152 Likewise, informing employees of their coworkers’ salaries might reduce job satisfaction, since no one wants to get paid less than the median employee.153 The transparency of medical prices could, in some instances, increase medical costs, since no one wants to pay for a cheaper than average surgeon.154 Likewise, publicly documenting the popularity of tax expenditures might have the effect of making tax expenditures more popular.
The public’s stubborn preference for tax expenditures may also help explain another much-derided feature of tax policy that has frustrated many scholars: tax complexity. A public bias in favor of spending through the tax code is an overlooked explanation for why the tax code has become so complex. Typical explanations for tax complexity focus on interest group pressure, tax fairness, and tax fraud reduction.155 But a robust public predisposition for spending through the tax code—combined with an electoral system in which politicians are motivated to pursue policies and framing that appeal to the median voter—also adds to the confusion of the tax system.
When ordinary taxpayers confront their taxes, the most complex items they deal with are generally tax expenditures. For salaried or wage employees, calculating gross income is relatively easy: enter your W-2 and, if you have any investments, take the income figures from the 1099 provided by your broker or financial institution. On the other hand, calculating deductions and credits is much trickier. A few relevant deductions are expenses related to earning taxable income.156 But the rest are tax expenditures: the mortgage interest deduction, deduction for state and local taxes, mortgage insurance deduction, the EITC, the adoption credit, the child tax credit, the “astonishingly complex credits designed to offset the cost of college tuition,”157 the charitable donation deduction, retirement accounts—the list goes on. The IRS estimates that the average individual spends eight hours on their taxes each year, between record keeping and actually filing.158 We think it is likely that the vast majority of this time is spent dealing with eligibility for tax expenditures.159 Taxpayers’ preference for spending through the tax code is thus part of what drives 84% of Americans to call the federal tax system complex.160
Tax expenditures are an important part of government spending. As noted above, the CBO estimates that federal tax expenditures will be $1.4 trillion in the 2014 fiscal year.161 If total spending is higher because (all else being equal) citizens have a persistent and systematic bias in favor of spending through the tax code, then billions of dollars are at stake. For example, if tax expenditures are a mere 2% higher because of greater public support for spending through the Code, then this $28 billion would be as much money as the President’s proposed energy budget (including clean energy initiatives—$28.4 billion)162 or the entire budget of the DOJ (including federal prisons and the FBI—$27 billion).163
Are citizens competent to make these important decisions of tax policy and public spending? A relatively uncontroversial feature of democratic theory is that the government should be responsive to citizen preferences. But this suggests an analogous conclusion, discussed at some length in the political science literature: in order for government to be responsive to citizen preferences, citizens must be capable of forming preferences.164
What does it mean for citizens to be capable of forming competent preferences? Some political scientists argue that competent preferences “should not be based on arbitrary aspects of how an issue or problem is described”165—a criterion that resembles the well-known “independence of irrelevant alternatives” condition in the social choice literature.166 But when framing effects cause the public to respond differently to equivalent proposals, the democratic competence of the citizenry may be thrown into doubt.167
This problem is different from one that is often raised in analyses of the public’s relationship to the tax system. Often, the emphasis is on public ignorance of the mechanics of tax administration.168 But the framing effect creates a different worry. It suggests that even if the public were much more fully aware of how the tax system works, public preferences would still be susceptible to “arbitrary” changes in wording and frames.
Frames are, of course, an inevitable feature of life. But they are an underappreciated part of the ways in which tax law and policy have been shaped over the last several decades. Much of academic and policy focus has been on publicizing the details and cost of our tax policy choices. But our analysis here suggests this emphasis is incomplete. However worthy those efforts may be, they run up against the stable and enduring public bias in favor of pursuing policies through the tax code.
As discussed in the body of the Note, Google Consumer Survey (GCS) is a “survey wall” that pops up when users want to access premium content, much like a paywall. GCS is used by “130 publishers in the U.S.” including “[t]hree of the top 10 newspapers, seven of the top 15 . . . sites like the New York Daily News, Christian Science Monitor, and . . . the LA Times.”169 The network also includes small news sites like Lima, Ohio News and the Texas Tribune as well as Pandora and YouTube, and various other arts and entertainment sites.170
Visitors to these sites cannot opt into the surveys. Instead, they are solicited using a model that is designed to ensure a representative sample. The algorithm over-samples groups that are currently underrepresented in the sample. That is, if, say, women aged 18-24 from the South were underrepresented in the sample relative to the portion of the underlying Internet-using population they make up (as measured by the Census’ Current Population Survey), then that group would receive proportionally more survey requests.171 This is known as stratified sampling, but GCS’s ability to receive responses in real time makes it different from traditional stratified methods where the stratification (that is, the issue of which groups are over sampled) is set before the survey begins.
The screenshot below, displaying Question 10, shows how one of our questions would show up. The order in which the answers are displayed is randomized (that is, the “Yes” option will show up after “No” as often as before “No”).
Taking Question 10 as an example, the image below shows the distribution of how long participants took to respond to the question.
On average, participants took thirty-six seconds. Excluding participants who took more than two minutes to respond, the average response time was twenty seconds. There does not appear to be any binding upper time limit.
One respondent took fifty-four minutes to respond (though we doubt that he was lost in thought about tax expenditures the whole time).