Social Policy and Public Opinion: How the Ideological ... · PDF fileSocial Policy and Public Opinion: How the Ideological Direction of Spending Influences Public Mood Christopher

Social Policy and Public Opinion: How theIdeological Direction of Spending InfluencesPublic Mood

Christopher Ellis Bucknell University

Christopher Faricy Washington State University

This article develops a model of public responsiveness to social policy in the United States, focusing in particular onthe public’s ability to distinguish between direct and indirect government spending as means of financing socialbenefits. We argue that public opinion should be responsive to changes in both direct (appropriations) and indirect(tax expenditures encouraging the private provision of social goals) spending. Further, the public should respond tochanges in direct and indirect spending in distinct ways consistent with the divergent resource and interpretiveeffects of the two types of spending. We find that while public opinion is not responsive to the total amount offederal social spending, it is attentive to changes in direct and indirect spending, considered as separate concepts.The results show that the electorate treats changes in the relative allocation of government spending as representingimportant shifts in the ideological direction of public policy.

The ability of the mass public to perceive andrespond to the actions of policy makers is acornerstone of representative democracy in

the United States.1 Because of this, a substantial bodyof research works to understand how public opinionresponds to policymaking activity. In particular,empirical work has found that public preferencesfor government spending change in response to thelevel of appropriations for public social programs(e.g., Soroka and Wlezien 2010; Wlezien 2004).Policy scholars, however, understand the Americansystem of social policy more broadly, as a ‘‘divided’’system in which the government finances both publicprograms—through direct spending via budgetaryappropriations—and private benefits—through indi-rect spending otherwise known as ‘‘tax expenditures’’(Faricy 2011; Hacker 2002).

These two different policy tools, although bothtreated as ‘‘spending’’ for federal budgetary purposes,represent divergent roles of government in the econ-omy. The American public has recently observed andparticipated in important debates over the dividedsocial system, on such issues as President Bush’s plan

for the ‘‘partial privatization’’ of Social Security (whichwould replace some direct government spending onincome security with government-based incentives forprivate retirement savings) and Democrats call for a‘‘public option’’ to compete with private health careinsurance plans (with would increase direct govern-ment spending onpublic health, perhaps at the expenseof some indirect spending which incentivizes theprivate provision of health care). These more recentdiscussions highlight long-standing debates over theproper role of the federal government in facilitating theprovision of generally popular social goals such asincome security, education, and public health (Hacker2004; Howard 2007).

The exclusion of indirect spending in empiricalwork on public responsiveness to government activity isimportant for two central reasons. First, indirect spend-ing represents the government’s substantial role insubsidizing private-sector social benefits—accountingfor close to $600 billion in federal spending in 2009alone. This is particularly important given that indirectspending plays a far different role in the economy thanthat of direct spending. Indirect spending does more to

The Journal of Politics, Page 1 of 16, 2011 doi:10.1017/S0022381611000806

! Southern Political Science Association, 2011 ISSN 0022-3816

1An online appendix with supplementary material for this article is available at www.journals.cambridge.org/jop. Data to reproduce thenumerical results in this article will be made available at http://dvn.iq.harvard.edu/dvn/dv/CGFaricy on the date of publication.

1

privilege market- over government-based conceptionsof power, tends to fund programs that benefit primarilywealthier citizens, and tends to redistribute wealthupward rather than downward. Second, the theoreticalfocus on direct appropriations implicitly assumes thatpublic reaction to government spending is drivenwholly by the amount that government spends inparticular social domains, rather than accounting forhow that money is spent, or to whom the benefits ofgovernment spending accrue.

In this article, we work to develop a more generaltheory of public opinion and social policy in the UnitedStates, encompassingmass responsiveness to bothdirectand indirect social spending. Considering social policyas deviating paths provides for a broader theoreticalunderstanding of the relationship between public pref-erences and social policy, allowing us to more closelyintegrate scholarshipwhich examines dynamicopinion-policy linkages in American politics—in particular, the‘‘thermostatic’’’ model of policy responsiveness offeredby Wlezien (1995)—with ‘‘social spending’’ policy asunderstood by scholars of American public policy. Weargue that the public has the ability to recognize andrespond to changes in both indirect and direct socialspending. In addition, we argue that citizens do notand should not respond broadly to ‘‘governmentspending,’’ defined as the sum total of governmentexpenditures on social programs. Instead, the diver-gent redistributive and ideological effects of directversus indirect social policy cause the public torespond to the two conceptions of social spending inopposing ways.

Using novel measures of social spending, we findthat increases in indirect social expenditures aretreated by the public as conservative public policyand move public preferences in a liberal direction.Direct social appropriations, by contrast, are viewedas liberal policy and tend to make public preferencesmore conservative. The public does not respond tosocial spending as an aggregated concept; rather, ittreats the two conceptions of social spending asdifferent types of policy tools and adjusts its prefer-ences accordingly. Responsiveness to spending policyis strongest, but not limited to, the most educatedportion of the electorate.

The results show that public opinion dynamicsare not driven by the total amount of federal govern-ment spending, but rather by the relative allocation ofsocial welfare expenditures across direct and indirectmeans. More generally, the extension of the thermo-static logic to indirect social spending also providesan initial step (see also Enns and Kelly 2010) towardmore closely linking the rich tradition of work on

opinion-policy relationships as conceived by scholarsof public opinion, with a fuller understanding of‘‘social policy’’ as conceived by policy scholars, in away that could enrich work in both fields (Hackerand Pierson 2009; Mettler and Soss 2004).

The Two Faces of Social Spendingin the United States

Most empirical research exploring the relationshipbetween public opinion and social policy typicallyfocuses strictly on direct governmental appropria-tions for social programs. But policy scholars (e.g.,Hacker 2002; Howard 1997, 2007) have demonstra-ted that there are two social systems in the UnitedStates: one public and the other private. Social policy,commonly defined, is any government effort toprovide economic security to citizens through pro-tection against income loss and the guarantee of aminimum standard of living.2 In this study, weconceive of social policy in this broader sense, as achoice between these two systems: direct socialspending on public social programs versus indirectsocial expenditures that subsidize the private market.

Direct spending is easily recognizable: it is simplyfederal spending on social programs that is allocatedthrough the budgetary appropriations process. Indi-rect spending, or ‘‘tax expenditures,’’ is a measure ofthe revenue loss to the U.S. Treasury from tax breaksused to incentivize the private provision of socialprograms. A central purpose of the tax expenditureconcept is to allow a side-by-side comparison betweenprograms funded through the tax code and pro-grams financed through the appropriations process(Surrey 1974).

Tax expenditures are considered a component ofgovernment spending because this policy instrumenttargets money to specific populations or activities andhas the same effect as direct spending on beneficiaries,and on the budget (Howard 1997; Surrey 1974). Forexample, a new tax expenditure program for employer-based pensions that cost $100 billion would have asimilar effect on the budget deficit as introducing an

2In recent decades, there has been substantial growth in thediversity of policy tools used to finance social policy in thisbroader sense, using both public and private means. There arenumerous and varied government tools used to promote policyobjectives: appropriations, tax expenditures, grants, regulations,loan guarantees, government corporations, and loans (Hacker2002; Kettl 1997; Salamon 2002).

2 christopher ellis and christopher faricy

expansion of Social Security that also cost $100billion.3 Tax expenditures are a familiar concept instudies of social policy in both American public policyand comparative politics (e.g., Adema and Ladaique2005; Hacker 2002). The majority of tax expendituresfor social policy go towards financing employer-basedsocial benefits such as private health insurance andpensions.

For our purposes, direct social spending can bemeasured using a familiar and oft-employed method,the appropriations data contained in the PolicyAgendas Project.4 To create a measure of direct socialappropriations, we combine total spending from thefollowing budgetary categories: Health; Medicare;Income Security; Education, Training, Employment,and Social Services; Social Security; and VeteransBenefits and Services. To account for inflation andpopulation growth, the measure we use is per-capitaspending in constant (2006) dollars.

To quantify indirect social spending, we rely on anew dataset compiled from estimates of tax expendi-tures from the Joint Committee on Taxation (here-after JCT). The JCT estimates tax expenditures interms of revenues lost to the U.S. Treasury for eachspecial tax provision included in the U.S. tax code.A provision has traditionally been listed as a taxexpenditure if it departs from the normal income taxstructure and it results in more than a de minimisrevenue loss ($50 million). Under the JCT method-ology, the normal tax structure for an individualincludes the following: one personal exemption foreach taxpayer and one for each dependent, thestandard deduction, the existing tax schedule, anddeductions for investment and employee businessexpenses. Most tax benefits to individual taxpayerscan be classified as exceptions to this normal incometax law. Each tax estimate is a function of subtractingtwo predicted streams of revenues: predicted reve-nues under the current law from predicted revenueunder new and expanded tax provisions.5

The tax expenditure data is organized by the JCTin the same budget categories as appropriationsspending. Following Faricy (2011), we create meas-ures of indirect social expenditures by combiningexpenditure estimates from the same budgetarycategories as used for direct spending. Again, weadjust this measure for both population size andinflation. Figure 1 graphs the total amount of directand indirect social spending for the period 1974–2006.6 Together, these two measures of spending sumto over $2 trillion in 2006. On average, indirect spend-ing represents about 20% of all social spending—although, as we will see, fluctuations around thisaverage are systematic and important. Even afteradjusting for inflation, the amount of both directand indirect social spending has increased substantiallyover the past 30 years.

The Divergent Consequences ofIndirect and Direct spending

Both indirect and direct spending provide ways forthe federal government to finance the provision ofgenerally popular social goals—public health, educa-tion, income security, and social welfare.7 But thechoice of financing social programs through direct orindirect spending has considerable implications forthe distribution of income, the targets of socialbenefits, and the relationship between governmentand market power. In addition, indirect and directspending policy are crafted through different means.More specifically, we identify at least fourcritical differences between direct and indirect socialspending.

First, direct and indirect spending vary intheir income redistribution effects. Appropriations

3Since 1974, the nonpartisan Congressional Joint Committee onTaxation (hereafter the JCT) has annually estimated tax expen-ditures in terms of revenues lost to the U.S. Treasury for eachspecial tax provision included in the U.S. tax code. The Congres-sional Budget and Impoundment Act of 1974 (CBA) officiallycodified tax expenditures and requires that these figures arereported to both Congress and the White House on an annualbasis. The JCT and the CBO both report tax expenditure data.We use the more common measurement from the JCT since ithas a more consistent methodology over time.

4See www.policyagendas.org and Baumgartner and Jones (2009).

5According to the JCT, these estimates have been excellentpredictors of actual changes in government tax receipts ascalculated by I.R.S. returns.

6Although tax expenditure data is nominally available going backto 1967, we restrict the analysis to the 1974-and-beyond periodbecause 1974 was the first year in which the collection and use oftax expenditure data was officially routinized by the JointCommittee on Taxation (as a function of the CongressionalBudget and Impoundment Act of 1974). The Joint Committee onTaxation experimented with slightly different calculations from1967–72, using varying assumptions for determining the standardbaseline income, but settled on a baseline equation that has beenused since 1974 report.

7It is well known, for example, that social programs of the sortfinanced through direct and indirect spending are quite popularin the electorate. When faced with the basic question of whethergovernment should spend more or spend less on education,social security, and public health for example (the three largestcategories of indirect social spending), substantial majorities saygovernment should spend and do more (e.g., Page and Jacobs2009; Stimson 1999).

social policy and public opinion 3

spending on social programs progressively redistrib-utes income to poorer citizens through the directfederal provision of services and programs designed tobenefit mainly older populations. Kelly (2008), forexample, shows that 53% of total Social Securitybenefits go to citizens in the lowest income quartile,and only 23% goes to the top two income quartilescombined. Total public social spending reduces thegini coefficient in the United States by over 12% (Kelly2008). The majority of tax expenditures, by contrast,have regressive effects on income redistribution (Sur-rey 1974). Since the income tax code is progressive bystructure, tax expenditures formulated as deductionsor exclusions from that structure generally reducethe progressivity of the tax system. Tax expenditures,in other words, regressively redistribute income byreducing tax rates more for higher marginal ratetaxpayers than for lower marginal rate taxpayers.8

Second, the public and private social systemsserve different populations. The beneficiaries offederal programs financed by direct social spendingare typically the elderly, the disabled, the unem-ployed, and the poor. Further, such programs aremore likely to serve racial and ethnic minorities aswell as single female-headed households. The benefi-ciaries of indirect spending, by contrast, tend to bewealthier, White, and more likely to be employed in

professional occupations than the public at large(Faricy 2011).9

Third, direct and indirect spending differ in thepower they accord to the private sector. While directspending explicitly calls for a direct federal role inproviding benefits, indirect spending allocates substan-tially more power to private markets in distributingsocial benefits (Howard 2007).10 Rather than directlyappropriating funds to provide health insurance, forexample, indirect spending uses tax breaks to incen-tivize private businesses and citizens to provide andpurchase health insurance in a market-based system.11

FIGURE 1 Indirect and Direct Social Spending, 1973–2006

8Taxpayers that are at or below poverty level and those that donot itemize their personal deductions receive none of thisgovernment aid. These and other tax breaks go mainly to thosewho itemize their income taxes, disproportionately wealthierhomeowners (Stanley and McDaniel 1985). For example, thechild care credit accrues more average tax savings to families asthey move up in income level. In 1999, a family making over$200, 000 was given an average credit of $485 while a familyearning between $40, 000 and 50, 000 was allocated only $387 forthe same child care credit (Toder, Wasow, and Ettlinger 2002).

9Workers enrolled in employment-based health insurance pro-grams, who are the recipients of indirect spending, for example,are more likely to be White, economically secure, and workingfull time as a professional in a large firm as compared to theaverage citizen (Purcell 2008). In 2008, according to the Em-ployee Benefits Research Institute (EBRI), roughly 70% of Whitesreceived health insurance through their employer. Conversely,49% of Blacks had coverage and only 41% of Hispanics hadprivate coverage. In the same year, more than 66% of workers inmanagerial and professional occupations had personal health careinsurance compared with 34% of workers in the service sector. Inaddition full-time, full-year workers (66%) and workers in firmswith 1,000 or more employees (64%), were substantiallymore likelyto receive employer-provided health insurance than part-time work-ers (34%) or workers in firms with fewer than 10 employees (26%).

10There are notable exceptions such as Medicare Part D fundingprivate companies and tax expenditure programs that makeSocial Security tax free but the majority of direct expendituresfinance public programs while the vast majority of indirectspending is allocated to the private sector.

11This difference in conceptions of government versus marketpower fits with the political history of social spending in theUnited States: Republicans have typically moved to increase theproportion of the federal social budget funded through indirectmeans, while liberal Democrats have sought to expand the sizeand proportion of the federal social budget funded by directappropriations (see Faricy 2011; Ventry 2002).


Fourth, indirect anddirect spending are also thoughtto differ in the openness of the respective policymakingprocesses themselves. The use of indirect spending tofinance social programshas been referred to as ‘‘hidden,’’‘‘a shadow state,’’ and ‘‘subterranean’’(Gottschalk 2000;Hacker 2004; Howard 1997). The limited number ofcongressional committees involved in the policy proc-ess of indirect spending along with the lack of annualreview provide institutional barriers for the masselectorate to learn, understand, and react to spendingchanges. Indirect spending provisions are created orexpanded in revenue or tax reform bills and notthrough the annual appropriations process. The SenateFinance and House Ways and Means committees haveexclusive jurisdiction over indirect spending bills andserve as both the approving and ‘‘appropriating’’committees. Direct spending measures, by contrast,must pass through standing committees with jurisdic-tion over a specialized policy area as well as the Budgetand Appropriations committees that fix spending levelsfor each budget category.

Because the process of policymaking is essentiallya ‘‘closed’’ process, there is concern among somepolicy scholars that citizens will be unable to under-stand policy changes with respect to indirect spend-ing or responsiveness will be restricted only to a verynarrow subset of politically engaged citizens with thecapacity to follow the tax expenditure policymakingprocess and understand its consequences. If thepublic cannot respond to indirect policymakingactivity in a particular realm, it has no basis to judgethe actions of policy makers and no way to hold themaccountable for their actions—thus potentially open-ing the door for a greater role for special interests andpolitical elites in shaping policy (Hacker 2002;Mettler 2008).

This discussion suggests that funding social pro-grams via direct or indirect means are distinct policychoices with divergent political and social consequen-ces. The raw amount of both direct and indirectspending has increased over time. But if direct andindirect spending truly represent divergent policy toolswith divergent consequences, then what is of conse-quence when understanding public opinion change isnot necessarily the total amount of social spending, butrather the balance between direct and indirectspending.

In this light, Figure 2 displays a ratio of indirectto direct social spending. This measure shows nolinear increase; rather, the proportion of socialspending that is indirect varies quite a bit, growingto over 25% of direct spending during two timeperiods during 1982–1987 and 2002–07, and settling

as low as 17% in the early 1970’s. This ratio measureperhaps best represents the dynamics of the choicefaced by policy makers in considering how to dealwith social spending.

A Model of Public Responsiveness toSocial Policy

It is clear that the American public lacks the capacity orinterest to be deeply informed about the particulars ofpublic policy in many if not most contexts. But there isnevertheless evidence that the public, at least in theaggregate and in some policy domains, is attentive andresponsive to the policymaking activity of the federalgovernment. The dominant theoretical frameworkused to explain mass responsiveness to public policy,particularly mass responsiveness to social policy anddirect social spending, is the ‘‘thermostatic’’ modeloffered by (Wlezien 1995; Soroka and Wlezien 2010).

In essence, this theory provides a model ofnegative feedback to policy makers, as the publicadjusts its relative preferences for government activityin a direction opposite the dominant ideologicaldirection of policymaking. As governmental activitymoves public policy in a particular direction, at leastsome citizens will change their preferences, goingfrom wanting ‘‘more’’ policy change in that directionto wanting no policy change at all, or (if the govern-ment overshoots their preferred level of policy)wanting ‘‘less.’’ In the aggregate, this means that publicdemands for policy to move further in that directionweaken, and public preferences change, to demandless policy change in that direction. When publicpolicy moves in a liberal direction, for example,citizens notice these changes and respond by de-manding comparably more conservative policy (andvice versa). The end result is the provision of negativefeedback from the aggregate public to policy makers.

The thermostatic model is fairly pervasive, shownto be applicable across many types of policy domains(Wlezien 2000), levels of government (Johnson,Brace, and Arceneaux 2005; Cashore and Howlett2007), and Western democracies (Jennings 2009).This type of responsiveness is critically important forrepresentative democracy, as it provides strong in-centives for elected officials to consider the wishes ofcitizens when crafting policy (Erikson, Mackuen, andStimson 2002). The model may at first glance suggesta reasonably sophisticated response on the part of theelectorate to changes in the policy environment. Butit does not require high levels of sophistication or


public attentiveness to policymaking processes orpolicy details. At its core, it requires citizens to usefairly simplified pieces of information to update verybroad political judgments. The level of politicalattentiveness necessary for citizens to be able toprovide thermostatic feedback in relevant policydomains is generally quite low: a general sense ofthe direction and magnitude of policy change. Itsimply requires that citizens, in other words, be atleast broadly cognizant of the direction which policyis moving, and be able to react accordingly.12 As such,we see evidence of this sort of responsiveness not onlyamong political sophisticates, but also among thosewith lower levels of education or political engagement(Enns and Kellstedt 2008; Ura and Ellis 2008).

This model is most commonly applied to thedynamic relationships between public opinion anddirect spending: as government spending increases(or decreases) relative to public demand in a particulararea, for example, citizen demands for comparablymore (or less) spending in that domain decrease. Thefocus on direct spending alone, however, obscures asignificant portion of federal spending, particularlywhen it comes to social policy and the provision ofsocial benefits. This view not only misses a significantportion of federal spending, but may also have theeffect of overstating (or at least misstating) the public’sability to respond sensibly to changes in federal policy,particularly given the ostensibly ‘‘closed’’ nature of theindirect spending policymaking process. A fuller viewof thermostatic public opinion, in other words, needs

to take into account both types of social spending andconsider not only the total amount that the federalgovernment spends on social programs, but also theideological direction of that spending.

Two Types of Social Spending andthe Thermostatic Model

Building on the thermostatic model, and drawingfrom Pierson’s (1993) feedback theory of publicresponsiveness, we develop a more general model ofthe dynamic relationship between public preferencesand social policy in the United States. We argue threepoints. First, the same kinds of factors that permitpublic responsiveness to direct spending will permitresponsiveness to indirect spending. Second, differ-ences in the redistributive effects of direct and indirectsocial policy should lead the public to respond tochanges in the two types of spending in opposite ways.Third, and related, because of the ideologically distinctnature of direct and indirect spending, the dynamics ofpublic responsiveness to spending should not beviewed as responsiveness to the total amount of federalsocial spending (direct plus indirect expenditures),but is instead best conceived as responsiveness to thebalance of direct and indirect spending, considered asseparate and distinct concepts.

The first of these points is the most straightfor-ward, and the most important: we expect that thethermostatic logic generalizes well to indirect, as wellas direct, social policy. The thermostatic model, whilewidely applicable, is not universal, as there are severalpolicy areas on which scholars find minimal evidence

FIGURE 2 Indirect and Direct Social Spending, 1973–2006

12It does not, in addition, assume that citizens have an exact‘‘ideal point’’ of policy in mind; rather, it simply assumes thatsome people want policy to move in one ideological directionmore strongly than do others (see Wlezien 1995, 982).


of thermostatic public feedback (Wlezien and Soroka2010). In particular, past research suggests twogeneral conditions that must be satisfied for theexpectation of thermostatic public responsiveness tobe met. The level of aggregation at which a particularpolicy domain is understood must be sufficientlyhigh, high enough to accord with the levels at whichthe public processes political information. The publicmay not be able to react sensibly to proposals tochange funding formulas for public health benefits,or to technical details that govern the administrationof health benefits. But it is much more likely tounderstand whether the federal government’s directrole in providing health benefits is getting larger orsmaller, and react accordingly (Wlezien 1995). Inaddition, the policy domain under considerationmust be of sufficiently high salience that the publiccan obtain at least limited information about thelikely intent or consequences of changes in policy.

This information can come in several forms. Itcan come most obviously through a detailed under-standing of the policymaking process itself. But it canalso come through more ‘‘gut’’ level understandingsof or experiences with the consequences of policy,through a rudimentary understanding of changes ofpolicy as reported in the press, or by taking cues fromother citizens who are more informed or interested inthe particular policy area than they. As Soroka andWlezien state:

‘‘all that is required is that some meaningful pro-portion of citizens have a basic preference for policychange in one direction or another and that theyadjust this preference over time in reaction to whatpolicymakers do, based on the experience thosecitizens receive from the media, political groups, orfriends and family, as well as daily experience withgovernment services, and with society more gener-ally.’’ (2010, 42)

The ability of the public to provide feedback to policymakers simply demands, in other words, that theactual or likely consequences of a particular policyissue are important and broad enough to be signifi-cant to large numbers of people, and that the effectsof a particular policy are apparent enough—eitherthrough media reports, or partisan cues, or throughcitizen experiences with the policy itself—for citizensto understand the consequences of particular changesin public policy (see also Erikson, MacKuen, andStimson 2002).

How well do the two conceptions of socialspending fit within this framework of policy feed-back? The case for responsiveness to direct spendingis relatively clear and has been made in detail

elsewhere (Wlezien 1995). While few possess deeplynuanced information about the federal budget, citi-zens regularly encounter information about federalbudgetary activity through the media, political opin-ion leaders, and the rhetoric of party elites. Further,citizens have regular experiences with federal agenciesand government benefits and, at least at the margins,are able to understand whether those services arebecoming more or less expansive. Citizens can see orfeel the consequences of policy, even if they are notattentive to the policymaking process itself, and thesetypes of ‘‘gut’’ understandings are generally sufficientfor citizens to understand the direction in whichfederal spending is moving, the basic conditionrequired for thermostatic public responsiveness.13

The case for public responsiveness to indirect spend-ing is more complex. Unlike with direct spending—where citizens have at least a vague knowledge of wherethe appropriations budget comes from—citizens havesubstantially less knowledge regarding what indirectsocial spending is and how such policy is made (Mettler2010). Nevertheless, we argue that citizens are able torespond to changes in indirect spending by using manyof the same mechanisms that are thought to underlieresponsiveness to direct spending.

Pierson (1993) theorizes that there are two typesof conditions that precipitate public feedback topolicy makers: resource effects, which determinehow policies shape the distribution of incentivesand outcomes, and interpretive effects, which influ-ence how policy conveys information about thepolitical environment to citizens. The choice betweendirect and indirect spending is at least in part apartisan concern: the two major parties have distinctpreferences for indirect and direct social spendingratios, with Republicans favoring more indirect socialspending and Democrats favoring direct expendituresfor public programs (Faricy 2011). This partisanconflict helps citizens to place changes in social policyaction into a broader interpretive lens, aligning nicelywith established partisan and other heuristics (Lupiaand McCubbins 1998). In addition, the two typesof social expenditures allocate benefits to differentpopulations and redistribute income in opposingdirections, facilitating different kinds of ‘‘resourceeffects’’ that affect how citizens relate to government.In the language of Pierson, the ‘‘outputs’’ ofsocial spending choice shape economic and politicalconditions in a way that facilitates the ‘input’’ of

13See Wlezien and Soroka (2010, chap.2) for an extendeddiscussion of this point.


citizen responsiveness to government action (Pierson1993, 595).14

Further, and as with direct spending, large segmentsof the citizenry have ‘‘street-level’’ experiences with taxexpenditures of various sorts and thus can develop ageneral understanding of whether these expenditures aregetting larger or smaller. In 2008 for example, over 158million citizens received health insurance through theiremployer, and more than 100 million were enrolled inemployer-sponsored pension plans. In addition, indirectspending constitutes a substantial portion of the federalbudget: the U.S. government, in 2008, allocated morethan $600 billion to private-sector social programs in theform of tax subsidies. Taken together, these sorts ofactions should permit the kind of rudimentary under-standing of how policy is changing and roughly howmuch it is changing in a way that allows the public torespond in a basic thermostatic way.15

Second, we expect that the two kinds of socialpolicy tools have distributive effects that shouldproduce distinct types of public response. An increasein direct spending targets social benefits to vulnerablepopulations such as the elderly, the working poor,and the unemployed. Public social programs producean income effect by redistributing cash and financialbenefits downward. They also increase the role of thefederal government in directly providing benefits andsocial services. Given these expected outcomes—andthe fact that citizens use the expected outcomes of apolicy change as a way to gauge the ideologicaldirection of that change (Soroka and Wlezien

2010)—direct social spending should be perceivedas ‘‘liberal’’ policy, given the combined effects oftargeting vulnerable populations, progressive redis-tribution, and direct government administration ofbenefits (see also Feldman 1984; Jacoby 1994).

As we have discussed above, indirect spendingtends to accrue social benefits to wealthier and moreeconomically and professionally secure citizens anddoes so by enhancing the position of private andmarket-based actors in the provision of benefits.These ideas are considered, in the main, to beright-of-center policy goals. In addition, the vastmajority of tax expenditure programs designed asdeductions and exemptions accrue more incomebenefits to wealthier citizens (mainly homeowners),due to the progressive structure of the income tax.

The resource effects of the two types of socialspending are reinforcing, then, in that they servepolitically distinct sectors of the population, whichallows the mass public to more readily discern theideological direction of the policy effects. We thusexpect that indirect social expenditures, because theirintent is to allocate public resources to businesses andprivate organizations while redistributing wealth up-ward, should be perceived as conservative policy(Lupia et al. 2007). Following the thermostatic logic,citizens should thus respond to increases in directsocial spending by increasing their demand forconservative public policy solutions and respond toincreases in indirect spending by increasing theirdemand for liberal policy solutions.

Third, because of the differences in the ideologicalintent and substantive consequences of direct andindirect spending, we expect that thermostatic respon-siveness to government spending will not be depend-ent on the amount of government spending of socialprograms, but rather on the relative balance of directand indirect spending in financing those programs.Much prior research on this topic has conceived of theopinion-policy link with respect to spending as aquestion of ‘‘more’’ or ‘‘less’’—if spending in a par-ticular policy domain increased, the public, all elseequal, would call for less spending (and vice versa).But if we consider government spending in thisbroader sense, the critical driver of public responsive-ness is not whether government is spending ‘‘more’’ or‘‘less’’ in a particular domain, but rather the way inwhich government is allocating that spending. We thusexpect that the public will not react to changes in thetotal amount of governmental spending when indirectand direct spending are aggregated. What we shouldsee instead is differential responsiveness to the twotypes of spending, considered separately.

14After an election that produces new majorities for the Repub-lican Party, for example, a resulting policy action might be toincrease indirect social spending. The majority party, often usingthe bully pulpit of the presidency, will communicate through allavailable sources the benefits of recently passed legislation, usingtypical partisan language of promoting economic freedom. Themedia will cover the general rise in tax breaks, and due to thetradition of ‘balanced’ presentation, will communicate the policyas both needed tax relief desired by Republicans and theDemocratic criticism of tax breaks for the rich. Over time, theout party (and allied interests) has an incentive to portraythe current political environment as being too conservative orbenefitting only the rich, which will over time be represented inchanges to mass public opinion. Groups of voters supporting theout party are more receptive to learning about the current policyenvironment and mobilizing against it, and the party out ofpower has incentive to inform and educate their supporters tocreate energy and action for the next election.

15Citizens respond to policy change in part because of reactionsto the policy change itself, but also in part because they either usepolicy change to anticipate changes in social outcomes, orbecause they react to the near-term policy outcomes producedby a policy themselves. Citizens should respond to indirectspending, in other words, for many of the same reasons thatguide response to direct spending according to the basic thermo-static logic (Soroka and Wleizen 2010).


Measuring Public Opinion

Our initial measure of public opinion is Stimson’s(1999, 2004) Public Policy Mood. Culled from theresponses to thousands of survey questions fromdozens of survey houses, Mood is one of the mostcommonly used empirical measures of aggregatepublic opinion, particularly in work dealing withrelationships between public opinion and publicpolicy. This measure is particularly useful for ourpurposes for several reasons. First, Mood can besubstantively interpreted as a measure of preferencestoward the proper size of the federal government andits role in distributing benefits, regulating the econ-omy, and providing social services (see Stimson 1999,71). This dimension is generally considered to be a‘‘scope of government’’ dimension, capturing publicsentiment on the long-standing ‘‘liberal-conservative’’divide over the appropriate balance of governmentaland market power (see also Kelly 2008).16 Muchvariance in public opinion on these sorts of issuestends to load on a single dimension, with aggregatepreferences for a wide range of issues moving togetherover time. Since we have argued that tax expendituresand direct appropriations reflect different conceptionsof the proper role of government, this measure isparticularly well-suited for the task at hand. Mood isalso a highly aggregated measure, encompassing pref-erences from a wide range of different issues. Becausewe argue that public responsiveness to the two types offederal spending should be primarily global, a measurewhich focuses on this kind of broad preference isespecially helpful.

Mood can also be understood as a relativemeasure, which deals broadly with public demandsfor ‘‘more’’ or ‘‘less’’ direct government power (seeErikson, MacKuen, and Stimson 2002). Rather thanmeasuring the public’s absolute preferences (i.e., ‘‘howliberal a government do you want? ‘‘), Mood insteadtaps the public’s relative sense over whether govern-ment should move in a ‘‘liberal’’ or ‘‘conservative’’direction. In this sense, it captures the difference

between the public’s ideal level of public policy andcurrent levels of public policy (see Soroka and Wlezien2010, 25). As we have seen, this relativistic perspectiveis important to the thermostatic logic.

Finally, and as an empirical matter, Stimson’smeasure has been used in a wide variety of analyseswhich attempt to explore governmental responsivenessto public opinion. Although the exact nature of policyresponsiveness varies, changes in public opinion asmeasured by Mood have been found to matter to thepolicymaking actions of all three branches of govern-ment, and have the potential to affect the degree towhich lawmakers pay attention to the preferencesof business, interest groups, and nonelected policyactors (e.g., Binder 1999; Coleman 1999; Mishler andSheehan 1993; Smith 2000). The measure thus hassome ‘‘face validity’’ as a way to explore public opinionchange on a dimension of preference known to be rel-evant to the actions of policy makers. The Mood indexis a simple 0–100 scale, with higher values indicatinggreater public ‘‘liberalism’’ (i.e., greater demands fora stronger direct federal role in economic and socialwelfare matters).

Public Opinion and Social Spending

The central tests of our theory are presented inTable 1, which relates public opinion to direct andindirect federal spending on social programs asoperationalized in Figures 1 and 2. The most basicrepresentation of the thermostatic model relates thepublic’s relative preferences (here, preferences formore or less liberalism in policy) to policy itself(here, levels of direct and indirect spending). Themodel can be written as:

DYt ! a" b1 It#1$ % " b2D$t#1% " b3W " e $1%

Where Y is the public’s relative preference for ‘‘more’’or ‘‘less’’ policy at time t, I and D are lagged levels ofindirect and direct spending, and W represents exog-enous drivers of public opinion not explicitly capturedin either I or D. At the outset, we will focus on directand indirect spending alone, but we will later considerthe role of common indicators of W. Consistent withthe most basic representation of the thermostaticmodel, then, we model changes in public preferencesas a function of lagged levels of spending in each of thetwo domains. Our expectations are that increases indirect appropriations will decrease public demand fora direct federal role in social welfare matters (and thusdecrease public ‘‘liberalism’’ as measured by Mood),

16In addition, this index does not tend to encapsulate preferenceson issues like abortion, gay marriage, or other ‘‘cultural’’ typesof concerns. Issues such as this tend to not share muchcommon variance with scope-of-government concerns, and thesetypes of issues remain conceptually distinct from scope-of-government concerns at the individual level (Carsey and Layman2010). Mood is thus best thought of as a measure of scope-of-government public opinion, not ‘‘public opinion’’ more broadlydefined.


while increases in tax expenditures will have theopposite effect (and thus increase public ‘‘liberalism’’).

The first column of Table 1 models opinion as afunction of lagged levels of total social spending,combining direct and indirect expenditures into asummary measure. The results are as expected: theeffect of policy on opinion is substantively very smalland statistically indistinguishable from zero (t 51.06, p 5 30). We have argued that there is littlereason to expect the public to respond to total levelsof social spending, since the total spending measure isactually a conflation of two separate concepts. Andwhen government spending is considered simply as‘‘total spending,’’ the conclusion, indeed, is thatpublic opinion does not respond to spending policy.

The results are different when we considerindirect and direct spending as separate concepts.The second column models public responsiveness tosocial welfare spending as we have conceived of itabove, including separate measures of direct andindirect spending. The results here are supportive ofour expectations: the public is responsive to changesin both indirect and direct social spending, but reactsto the two types of spending in opposite ways. A$1,000 increase in per-capita tax expenditures nowtranslates into a 4.5 percentage point liberal move inpublic opinion. The effect is relatively modest (the$1,000 increase needed to move public opinion4.5 points represents roughly one-half of the rangeof per-capita expenditures over this time period), butgiven that the observed range of Mood over this timeperiod is only 12 percentage points, even fairlymodest changes in public opinion can have signifi-cant consequences for the political system (seeErikson, MacKuen, and Stimson 2002).

The effects of direct appropriations are also statisti-cally significant, but as expected, run in the oppositedirection: increases in per-capita direct social welfare

spending tend to make the public more conservative.Here, a $1, 000 increase in direct spending moves thepublic roughly 1 percentage point in a conservativedirection. Direct expenditures are a larger portion of thetotal social welfare budget, so the smaller coefficientobviously does not necessarily imply a smaller substan-tive effect: indeed, there is no statistically significantdifferencebetween the absolute valueof the standardizedcoefficients for direct (B5 -.76) and indirect (B5 .99)spending. In general, then, we find that public prefer-ences are responsive to both direct and indirect socialspending. But as would be expected given their redis-tributive effects and implications for government power,it reacts to direct and indirect spending in oppositeways.Further, we see that the magnitudes of these reactionsare similar, consistent with the idea that the samefactors that allow the public to respond systematicallyto changes in direct social appropriations also allow thepublic to respond systematically to indirect ones.Despite clear differences in how indirect and directspending policy is made (and perhaps differences inthe openness of those processes to public view), thepublic takes into account available information aboutboth direct and indrect spending when updating itsrelative preferences for the role of government.

Perhaps more importantly, the public is broadlyattentive to the ratio of indirect and direct socialspending, a measure which perhaps best captures thereal trade-offs that policy makers face in decidingwhether to pursue social goals through direct orindirect means.17 The third column of Table 1 models

TABLE 1 Effects of Social Spending on Public Opinion

Dependent variable: changes in public liberalismt

Total Spending per capita(t-1) 0.21 (0.20)Indirect Spending per capita(t–1) 4.52* (1.81)Direct Spending per capita(t–1) -1.09* (0.57)Social Spending Ratio(t–1) 28.74* (9.56) 23.28* (11.95)Public Policy (important laws)(t–1) -0.06 + (0.04)Changes in unemployment(t) -0.28 (0.34)Inflation(t–1) 0.18 (0.14)Constant -1.02 (1.04) -0.14 (1.04) -5.65* (1.91) -5.23* (2.47)R2 0.03 0.19 0.23 0.30N 33 33 33 33

Note: * p, .05, + p, .10, standard errors in parentheses.

17The ratio measure provides a different means of understandingthe broad balance between government and market power in theprovision of social benefits and can be conceived of as a summarymeasure of the ideological direction of social spending policy.Changes in the ratio measure represent the degree to which theideological direction of social spending is moving, on balance, tothe left or to the right.


public opinion as a function of the proportion of totalfederal social welfare spending that is indirect. Theresult is what would be expected from the first twocolumns: increases in tax expenditures as a proportionof all social spending tends to move the public in aliberal direction. Simply stated, the results in this tablesuggests that public preferences are attentive to bothhow much the federal government is spending onsocial welfare programs and the balance of how thatmoney is being allocated. These effects suggest a publicthat is attentive to the redistributive consequences ofvarious types of public policy actions and one that isable to perceive both ‘‘direct’’ and ‘‘indirect’’ spendingin remarkably similar ways.

For the sake of clarity, our core model focusessolely on the effects of direct and indirect spending. It isimportant to note, however, that the results hold evenafter including a variety of other controls thought inpast research to matter to public opinion. Most directly,we find that our results regarding public responsivenessto the ideological direction of spending hold aftertaking into account other nonspending shifts in thesize and scope of government. As such, it is critical tounderstanding changes in the ideological direction ofpolicy to which citizens respond. But spending is onlyone (albeit a central) thing that the federal governmentdoes. Erikson, MacKuen, and Stimson’s (2002) study ofdynamic representation, for example, codes the annualideological content of Mayhew’s (1991) ‘‘importantlaws’’ measure to derive a summary measure of thenumber and ideological direction, of laws passed in anygiven year (see Mayhew (1991) for a more detaileddescription of the concept of important laws, Erikson,MacKuen and Stimson (2002) for an application to thisLaws measure to dynamics of Mood, and Kelly (2008)for an updated measure).18 The Laws measure (seeColumn 4 of Table 1) performs in a way consistent withpast research (e.g., Erikson, MacKuen, and Stimson2002; Kelly and Enns 2010): the public responds

thermostatically to the broad ideological direction ofpublic policy, with the enactment of ‘‘liberal’’ lawsmoving the public in a conservative direction and viceversa. But the inclusion of this different measure ofpolicymaking does not do much to diminish the effectsof the tax expenditure proportion measure.19

Similarly, core models of Mood have suggestedthat the economic conditions of unemployment andinflation can affect Mood in some contexts. Unem-ployment and inflation are the two critical economicoutcome variables, for example, considered byErikson, MacKuen, and Stimson’s classic study of policyresponsiveness (see also Enns and Kellstedt 2008;Kellstedt, Peterson, and Ramirez 2010). The inclusionof these economic variables also does not materiallychange the effects of spending. While this is clearly notan exhaustive list of factors that could conceivablyaffect Mood, they are important, theoretically relevantones prominent in past research on the subject.20 Wecan be confident, at least, that the effects of socialspending are robust to well-utilized, plausible controls.

Education and Responsiveness toSocial Spending

Evidence of aggregate public responsiveness to bothtypes of social spending broadens our view ofthermostatic responsiveness to social policy, showingthat the public is responsive not to the total amountof social spending, but rather to the ideologicaldirection of that spending, and the relative balanceof spending across direct and indirect means. In thissection, we wish to extend that logic further. As we

18Other representations of opinion-policy relationships havetaken different approaches: Erikson, MacKuen, and Stimson(2002) and Soroka and Wlezien (2009), for example, use thetraditional OLS lagged dependent variable approach to modelingopinion-policy relationships (modeling public opinion as afunction of its own lagged values and lagged values of policy).Jennings (2009) conceives of thermostatic opinion-policy dy-namics in the United Kingdom as an error-correction processand uses error-correction modeling techniques. We present thereduced form model used in Table 1, modeling changes inopinion as a function of lagged levels of spending, because itmost closely parallels the original thermostatic model of Wlezien(1995). But using these alternative specifications produce sub-stantive results, and statistical significance levels of variables ofinterest, that are similar to the model that we present here (seethe online appendix for results of these models.)

19Mayhew’s Lawsmeasure essentially captures broad trends in theideological direction of public policy activity. As such, it includesboth matters that impact federal direct and indirect spending(such as laws that impact funding for Social Security, education,or means-tested benefits)and matters far divorced from spending(such as changes to trade policy, environmental protectionpolicy, or changes in gun or voting-rights laws). The Lawsmeasure is thus conflated at least to some extent with thespending policy measures. We have also run the analyses with arevised version of Mayhew’s measure with laws that directlyimpact social spending purged from the analysis. Doing soincreases the impact of Laws on Mood, but does not change thestatistical significance or substantive importance of the socialspending ratio.

20In addition, we have estimated models controlling for otherfactors thought to move Mood—in particular, consumer senti-ment (Durr 1993) and income inequality (Kelly and Enns 2010).These factors (see online appendix) generally behave in a wayconsistent with past research, but do not reduce the substantiveimpact or statistical significance of the spending variables.


have discussed, a growing body of research suggeststhat because of the relatively low informationdemands placed on citizens for thermostatic respon-siveness to occur, we see evidence of dynamicresponsiveness to policy not only among a politicallyengaged and connected elite, but at least to someextent, citizens with lower levels of attentiveness aswell. This is true when it comes to direct spending, asthermostatic responsiveness in important social pol-icy domains occurs among the less educated and lesspolitically sophisticated (Soroka and Wleizen 2010).Here, we see whether these implications of thethermostatic model also extend to indirect spen-ding—whether, in other words, public responsivenessto indirect social policy runs as ‘‘deep’’ into theelectorate as responsiveness to direct social policy.

Data from the General Social Survey (GSS) allowsus to explore the effects of political engagement onpublic responsiveness to social spending. More di-rectly, we can see whether the aggregate results inTable 1 hold for citizens in different subsets of thepopulation. Mood is an aggregate indicator of publicpreferences comprised from a number of differentquestions that are not asked consistently over time. Itis thus not possible to directly disaggregate Moodinto subgroup ‘‘Moods’’ for any particular subset ofthe population. But the General Social survey con-tains a battery of questions, broadly relating togovernment action in a variety of social areas, whichcan be used to proxy Mood.21 Although the measureis much simpler and more limited in scope, theaggregates of it correlate with the broader Moodindex at 0.85 and capture the same shifts and changesin aggregate opinion as Mood. Since the GSS makesbasic demographic data for all respondents, we canmeasure ‘‘Moods’’ at a level below the public as awhole by aggregating these individual-level scoresover relevant subgroups of the population.

The GSS does not consistently ask questions thatmeasure political sophistication or engagement di-rectly. Instead, we proxy it using the GSS measuresfor education. Although education and engagement areclearly not the same thing, they are strongly correlatedwith one another, and education does represent aprimary way through which citizens learn to receiveand process information about the political context(Abrajano 2005). We thus divide the population intothree groups based on levels of formal education: thosewithout a high school education, those with a highschool education but no further degree, and those witha college degree or more. We then can create separatesubgroup ‘‘Moods’’ for each of these groups.

In Table 2, we replicate the core model fromTable 1—opinion as a function of the social spendingratio—for each of the educational groups.22 The firstcolumn of Table 2 shows results mirroring thepenultimate columns in Table 1 using as the depend-ent variable the aggregate Mood proxy instead ofStimson’s Mood index. This set of results shows astrong correspondence between the results using Mooditself, and the results using the GSS proxy for it:whatever the limitations of the GSS proxy measure, itrecovers results nearly identical to those shown inTable 1, providing some degree of confidence that theresults we see in the rest of Table 1 would be obtainedif we were able to disaggregate Mood itself.23

The final three sets of results illustrate the effects ofspending on opinion for each of the three subgroupsconsidered separately. These results illustrate twopoints. First, the greatest levels of responsiveness tothe social spending ratio, appears to be in the mosteducated subgroup—although because of the compa-rably higher standard errors for the ‘‘more than highschool’’ model, we cannot say that this group is more

21In particular, the General Social Survey asks respondents fortheir preferences on 10 major policy domains (see the onlineappendix for questions and coding of the GSS data). Preferenceson these 10 domains can be coded for ideological content andsummed together into a single score. While the particularquestions that comprise this proxy measure are by no means aperfect match for the concept of interest to us here, what is mostimportant is that when aggregated across domains, they tend toreflect citizens’ underlying preferences on the ‘‘scope of govern-ment’’ dimension as defined by Mood. See Ellis, Ura, and Ashley-Robinson (2006), Kellstedt, Petersen, and Ramirez (2010), andEnns and Wlezien (2011) for discussions regarding the measure-ment and properties of GSS proxies for Mood.

22Because the equations for each of the three subgroups are beingestimated using the same time frame and the same set ofindependent variables, we estimate the models using SeeminglyUnrelated Regression(Zellner 1962), which allows for the errorsof the equations (for example, idiosyncratic factors left out of themodel that may affect Mood for all three groups) to be correlatedwith one another.

23The same patterns of results hold for this and all remainingcolumns in Table 2 if we use model opinion as a function ofseparate measures of direct and indirect spending per capita,rather than the social expenditure ratio. This is particularlyimportant in light of past research suggesting that the indirectsocial policymaking process is more comparably hidden frompublic view than that of direct spending. It is certainly possiblethat our aggregate findings mask the fact that responsiveness toindirect spending in particular, is restricted to a small, highlyengaged, and politically elite group of citizens.


responsive than the group with only a high schooleducation (X2 test of difference 5 1.94, p 5 .16).24

Second, responsiveness to spending is not re-stricted to only a tiny subset of the population, butneither is it universal. The coefficients for both the‘‘more than high school’’ and ‘‘high school’’ groupare both significant, in the expected direction, andsubstantively important. The high school group,for example, responds to aggregate spending inways that might look like the aggregate public. Theleast educated group, by contrast, appears to beunresponsive to changes in spending. The coefficientfor the spending ratio variable is not significantlydifferent from zero for this group, and this group issignificantly less responsive than the citizens with ahigh school diploma (X2 5 4.56) or a college degree(X229.53).

The substantive implication of these models isthat public responsiveness to the policy context doesnot necessarily filter down to all subsets of thepopulation. There are segments of the population—in

ourmodels, those with less than a high school diploma,but more broadly citizens with exceptionally low levelsof political engagement—who cannot or donot updatepreferences in response to policy. But nor is respon-siveness to spending restricted to a very narrow slice ofthe electorate, as citizens with even a high schooleducation respond systematically to both direct andindirect spending. In addition, we see whatever roleeducation plays in shaping responsiveness to policyapplies equally to both direct and indirect spending.25

This provides suggestive, though not definitive,evidence that the same kinds of factors which permit(or not) responsiveness to direct spending alsopermit responsiveness to indirect spending.

Conclusion

This article demonstrates that the mass public followsand properly interprets changes to the modality ofsocial spending as a shift in the ideological directionof public policy. Social programs funded throughdirect and indirect means—education, income secur-ity, public health—are generally popular (Ellis andStimson 2009). The changes in public opinion inresponse to spending policy that we have observedhere are not reflective of the public’s desire for‘‘more’’ or ‘‘less’’ spending: the coefficient for ‘‘totalspending’’—aggregating across indirect and directrealms—in Table 1, for example, was substantively

TABLE 2 Effects of Social Spending on Public Opinion, by Education Level

Dependent variable: changes in public liberalismt

GSS Mood ProxyLess than

High School High SchoolMore ThanHigh School

Social Spending Ratio(t–1) 25.70* (8.67) 7.27 (11.55) 27.67* (6.30) 42.57* (14.90)Constant -5.07* (1.75) -1.30 (2.31) 0.40 (0.67) 0.01 (1.57)R2 0.23 0.03 0.37 0.20

Notes: Table entries are OLS coefficients (column 1) and Seemingly Unrelated Regression Coefficients (columns 2-8).Standard errors in parentheses. N533 for all models. * p, .05

24Average educational levels have clearly increased over this timeperiod, so the proportion of citizens who fall into each of thesecategories has changed over time. We choose to measure usingabsolute level of education both because of its simplicity andbecause it is consistent with past work using similar kinds of data(see, e.g., Enns and Kellstedt 2008; Soroka and Wleizen 2010).While raw educational attainment is the cleanest way to oper-ationalize education, there are concerns with using such ameasure—in particular, that average levels of educational attain-ment have been increasing over the past 30 years and thateducation’s role in shaping political attentiveness is in part as asocietal ‘‘sorting’’ function (see Nie, Junn, and Stehlik-Barry1996). We have also estimated the models in Table 2 (resultsavailable in online appendix) dividing the population into twogroups, based on whether their average educational attainmentwas above or at or below the average education of the mediancitizen in their birth year. These results, if anything, are moresupportive of the idea that reaction to direct and indirectspending are not restricted to the better educated: in thisspecification, we see significant responsiveness among high-and low-education respondents, and the magnitudes of theresponses are not significantly different across groups.

25In addition, there is no evidence that the ability to respond toindirect spending specifically, as opposed to the spending ratiomore generally, increases with education level: models whichestimate the effects of direct and indirect spending separately(available upon request) show that the size of the indirectspending coefficient relative to the direct spending coefficientdoes not grow larger the higher one progresses up the incomescale.


small and statistically indistinguishable from zero.Rather, we argue that the proper way to conceive ofopinion change in response to spending is best under-stood as changing public perceptions regarding therole of the government in the economy as it goes aboutfinancing the provision of social benefits—morespecifically, about the choice between indirect anddirect spending.

Policy makers tend to view direct and indirectspending as alternative tools toward achieving popularsocial goals, and the differences between views ofthese tools tend to be ideologically driven (Faricy2011). The idea of retirement-age income security,for example, is exceedingly popular, and directgovernment programs to ensure such security (mostobviously, Social Security) have proven to be toowidely supported for frontal assaults from politicalconservatives. Consequently, conservative elites havesupported a long-term strategy of building a private(but government-incentivized) alternative to thepublic social system. After the Republicans weredefeated in their attempt to reduce Social Securityin 1982, the Heritage Foundation began promotingthe idea of using 401ks and IRAs to construct analternative to Social Security (Rich 2004). Teles(2007) argues that the Republican Party, with assis-tance from the Heritage Foundation, designed adisentrenchment plan for Social Security built pri-marily not on attacking the system itself, but throughefforts to increase public familiarity with and relianceon private alternatives.

More generally, this example illustrates thatpolicy decisions which alter the balance of directand indirect spending alter the balance betweenpublic and private power by changing the allocationof government money from direct financing of publicsocial programs to the subsidization of private ones(and vice versa). At a minimum, our results suggestthat the public, at least in broad terms, understandsthe ideological differences between direct and indirectgovernment spending in a way consistent with howthese differences are understood by policy makersand has the capacity to react accordingly.

In addition, our results have implications forunderstanding the relationship between the publicand the ‘‘hidden’’ (Howard 1999) social welfare statein a way that helps to merge insights from twodistinct lines of scholarship. It is generally arguedthat most citizens do not have an even moderatelydetailed understanding of the ‘‘tax expenditure’’concept (Mettler 2010) and that the process ofindirect social policymaking is less routinized, andmore concealed from public view, than many other

policymaking processes (e.g., Howard 2007).Nothing in this article casts doubt on either of theseideas. But the extension of the thermostatic logic toindirect spending shows that public responsiveness toindirect social policy may still be systematic andsensible.

Despite clear limitations in both public under-standing of indirect spending policy and the processthrough which such policy is made, the publictypically reacts to increases in indirect spending—andin the ratio of indirect to direct spending—bymoving its preferences in a liberal direction. Thisfinding comports well with what we know fromresearch into public opinion and direct spending:the public’s lack of knowledge regarding the budget-ary process, or even the rough size of the directappropriations budget in any number of categories(Gilens 2005), does not preclude public opinion fromresponding systematically to changes in the size ofthat budget in those same categories (Soroka andWlezien 2010). More generally, we believe that thesefindings help to integrate what we know about policyfeedback from scholars of American public policywith what we know about mass preference changefrom public opinion scholars and helps to provide afirst step upon which future work can build inproductively linking these two lines of research(Mettler and Soss 2004).

It is important to be clear about the limitations ofour argument. We are not arguing that publicopinion is optimally responsive to all types of publicpolicy, nor are we suggesting that the indirect policy-making process does not, in theory, invite thedisproportionate influence of special interests. Wesimply argue that citizens have the capacity todevelop a general sense about the direction in whichpolicy is moving and can react accordingly. This is, tobe sure, a highly simplified understanding. But eventhese broad kinds of public responsiveness tend tofacilitate policy-maker attentiveness to public opin-ion (Soroka and Wlezien 2010).

The next step, of course, is to examine whetherpublic opinion does, in fact, affect the balance ofdirect and indirect social spending. We would expectto see changes in public preferences result in changesto social spending policy. But given this framework,changes in public sentiment should not affect abso-lute levels of government spending, but rather theallocation of that spending across direct and indirectmeans. Given the deep popularity of social programsand benefits, we would not expect that electedofficials will interpret an increase in mass conserva-tism to be a directive for less government spending,


but instead, as a directive to use spending tostimulate and subsidize the private market.

Acknowledgment

An earlier version of this article was presented at the2010 meetings of the Midwest Political ScienceAssociation. The authors wish to thank Frank Baum-gartner, Bill Jacoby, Phil Paolino, James Stimson, andthree anonymous reviewers for helpful commentsand suggestions on this manuscript.

References

Abrajano, Marisa A. 2005. ‘‘Who Evaluates a PresidentialCandidate by Using Non-Policy Campaign Messages?’’Political Research Quarterly 58: 55–67.

Adema, Willem, and Maxime Ladaique. 2005. ‘‘Net SocialExpenditure, 2005 Edition: More Comprehensive Measuresof Social Support.’’ OECD Social, Employment and MigrationWorking Papers.

Baumgartner, Frank R., and Bryan D. Jones. 2009. Agendas andInstability in American Politics. Chicago: University of ChicagoPress.

Binder, Sara 1999. ‘‘The Dynamics of Legislative Gridlock, 1947--1996.’’ American Political Science Review 93: 519–33.

Carsey, Thomas M., and Geoffrey C. Layman. 2010. ‘‘PartyIdentification, Party Polarization, and Conflict Extension inthe U.S. Electorate.’’ In Understanding Public Opinion, ed.Barbara Norrander and Clyde Wilcox. Washington, DC: CQPress.

Cashore, Benjamin, and Michael Howlett. 2007. ‘‘PunctuatingWhich Equilibrium? Understanding Thermostatic Policy Dy-namics in Pacific Northwest Forestry.’’ American Journal ofPolitical Science 51: 532–51.

Coleman, John J. 1999. ‘‘Unified Government, Divided Govern-ment, and Party Responsiveness.’’ American Political ScienceReview 93: 821–35.

Durr, Robert H. 1993. ‘‘What Moves Policy Sentiment?’’ Amer-ican Political Science Review 87: 158–70.

Ellis, Christopher, and James A. Stimson. 2009. ‘‘SymbolicIdeology in the American Electorate.’’ Electoral Studies 28:388–402.

Enns, Peter, and Paul Kellstedt. 2008. ‘‘Policy Mood and PoliticalSophistication: Why Everybody Moves Mood.’’ British Journalof Political Science 38: 433–54.

Enns, Peter, and Christopher Wlezien. 2011. Who Gets Repre-sented? New York: Russell Sage Press.

Erikson, Robert S., Michael B. MacKuen, and James A. Stimson.2002. The Macro Polity. Cambridge: Cambridge UniversityPress.

Faricy, Christopher. 2011. ‘‘The Politics of Social Policy inAmerica: The Causes and Effects of Indirect versus DirectSocial Spending.’’ Journal of Politics 73: 74–83.

Feldman, Stanley. 1984. ‘‘Economic Self-Interest and the Vote:Evidence and Meaning.’’ Political Behavior 6: 229–51.

Gilens, Martin. 2005. ‘‘Political Ignorance and Collective PolicyPreferences.’’ American Political Science Review 95: 379–96.

Gottschalk, Marie. 2000. The Shadow Welfare State: Labor, Busi-ness, and the Politics of Health-Care in the United States.Ithaca, NY: Cornell University Press.

Hacker, Jacob S. 2002. The Divided Welfare State: The Battle overPublic and Private Social Benefits in the United States. NewYork: Cambridge University Press.

Hacker, Jacob. 2004. ‘‘Privatizing Risk without Privatizing theWelfare State: The Hidden Politics of Social Policy Retrench-ment in the United States.’’ American Political Science Review:243–60.

Hacker, Jacob, and Paul Pierson. 2009. ‘‘The Case for Policy-Focused Political Analysis.’’ Prepared the annual meeting ofthe American Political Science Association.

Howard, Christopher. 1997. The Hidden Welfare State: TaxExpenditures and Social Policy in the United States. Princeton,NJ: Princeton University Press.

Howard, Christopher. 2007. The Welfare State Nobody Knows.Princeton, NJ: Princeton University Press.

Jacoby, William G. 1994. ‘‘Public Attitudes toward GovernmentSpending.’’ American Journal of Political Science 38: 336–61.

Jennings, Will. 2009. ‘‘The Public Thermostat, Political Respon-siveness and Error-Correction: Border Control and Asylum inBritain, 1994--2007.’’ British Journal of Political Science 39:847–70.

Johnson, Michael, Paul Brace, and Kevin Arceneaux. 2005.‘‘Public Opinion and Dynamic Representation in the Amer-ican States: The Case of Environmental Attitudes.’’ SocialScience Quarterly 86: 87–108.

Kelly, Nathan. 2008. The Politics of Income Inequality in theUnited States. Cambridge: Cambridge University Press.

Kelly, Nathan, and Peter Enns. 2010. ‘‘Inequality and theDynamics of Public Opinion: The Self-Reinforcing LinkBetween Economic Inequality and Mass Preferences.’’ Amer-ican Journal of Political Science 54: 855–70.

Kellstedt, Paul, David Petersen and Mark Ramirez. 2010. ‘‘TheMacro Politics of a Gender Gap.’’ Public Opinion Quarterly 74:477–98.

Lupia, Arthur, and Mat McCubbins. 1998. The DemocraticDilemma: Can Citizens Learn What they Need to Know?Cambridge: Cambridge University Press.

Mayhew, Richard. 2005. Divided we Govern. New Haven, CT:Yale University Press.

Mettler, Suzanne and Joe Soss. 2004. ‘‘The Consequencesof Public Policy for Democratic Citizenship: BridgingPolicy Studies and Mass Politics.’’ Perspectives on Politics 2:55–73

Mettler, Suzanne. 2008. ‘‘The Transformed Welfare State and theRedistribution of Political Voice.’’ In The Transformation ofAmerican Politics, ed. Paul Pierson and Theda Skocpol.Princeton, NJ: Princeton University Press.

Mettler, Suzanne. 2010. ‘‘Reconstituting the Submerged State:The Challenges of Social Policy Reform in the Obama Era.’’Perspectives on Politics 8: 803–824.

Mishler, William and Reginald S. Sheehan. 1996. ‘‘Public Opin-ion, the Attitudinal Model, and Supreme Court Decision-Making.’’ Journal of Politics 58: 169–200.

Page, Benjamin I., and Lawrence R. Jacobs. 2009. Class War?Chicago: University of Chicago Press.

Pierson, Paul. 1993. ‘‘When Effect Becomes Cause: Policy Feed-back and Political Change.’’ World Politics 45: 595–628.


Purcell, Patrick. 2008. ‘‘Pension Sponsorship and Participation:Summary of Recent Trends.’’ Federal Publications: 543.

Rich, Andrew. 2004. Think Tanks, Public Policy, and the Politics ofExpertise. New York: Cambridge University Press.

Smith, Mark. 2000. American Business and Political Power.Chicago: University of Chicago Press.

Stimson, James A. 1999. Public Opinion in America. Boulder, CO:Westview Press.

Soroka, Stuart, and Chris Wlezien. 2010. Degrees of Democracy.Cambridge: Cambridge University Press.

Surrey, Stanley S., 1974. Pathways to Tax Reform: The Concept ofTax Expenditures. Cambridge, MA: Harvard University Press.

Surrey, Stanley S., and Paul R. McDaniel. 1985. Tax Expenditures.Cambridge, MA: Harvard University Press.

Teles, Steven M. 2007. ‘‘Conservative Mobilization against En-trenched Liberalism.’’ In The Transformation of AmericanPolitics, ed. Paul Pierson and Theda Skocpol. Princeton, NJ:Princeton University Press.

Toder, Eric, Bernard Wasow, and Michael P. Ettlinger. 2002. BadBreaks All Around. New York: The Century Foundation Press.

Ura, Joseph, and Christopher Ellis. Income, Preferences, and theDynamics of Policy Responsiveness. PS: Political Science andPolitics 41: 785–94.

Ventry, Dennis, and Joseph J. Thorndike. 2002. Tax Justice: TheOngoing Debate. Washington, DC: Urban Institute Press.

Witte, John F. 1985. The Politics and Development of the FederalIncome Tax. Madison: University of Wisconsin Press.

Wlezien, Christopher 1995. The Public as Thermostat: Dynamicsof Preferences for Spending. American Journal of PoliticalScience 39: 981–1000.

Christopher Ellis is an Assistant Professor atBucknell University, Lewisburg, PA 17837.

Christopher Faricy is an Assistant Professor atWashington State University, Pullman, WA, 99164.


Documents

Social Policy and Public Opinion: How the Ideological ... · PDF fileSocial Policy and Public Opinion: How the Ideological Direction of Spending Influences Public Mood Christopher