Public Opinion and Fiscal Politics

Public Opinion and Fiscal Politics Abstract There has been a lot of controversy over federal spending over the past few years, as Tea Party rallies have garnered significant media attention and governors of many states have rejected federal funds to expand Medicaid. Yet, despite large literatures on distributive politics and public opinion and representation, we still know very little about the role of public opinion in fiscal p olitics. Our paper uses multilevel regression and poststratification (MRP) to estimate state-level opinion on a few general categories of spending. We then analyze the relationship between state-level opinion and the Senate roll-call vote on the stimulus bill, as well as the distribution of stimulus money to states. In sum, we find a p ositive correlation between constituent support for the stimulus and Senators’ votes on the measure, but no relationship between opinion and appropriations. Looking beyond the stimulus to net spending more broadly (i.e. total federal money received by each state divided by federal taxes paid), we find a p ositive c orrelation with several different measures of opposition to federal spending. These results suggest that the role of public opinion in fiscal p olitics i s c omplex, a nd d eserves further study. 1 Introduction There has been quite an uproar over federal spending over the past five years. Critiques have come both at the mass level, through acerbic posters at Tea Party rallies, and also at the elite level, with many state governors rejecting federal money to expand Medicaid, for example. This wave of opposition to spending might not seem strange in the wake of a financial meltdown. Yet, it appears to belie one of the central tenets of the distributive politics literature in political science—that people almost always want goodies, and elected officials almost always want to procure them. This dissonance raises an important question: what is the role of public opinion in fiscal politics? We know surprisingly little about this, given the existence of robust (though separate) literatures on distributive politics and public opinion. Distributive politics scholars tend not to consider public preferences, focusing instead on institutional factors affecting appropriations. And those studying public opinion and representation tend to focus on roll-call votes, rarely expanding their analyses to other dependent variables like appropriations. Methodological challenges may have discouraged work in this area. Below the highest level of aggregation, comparing national public opinion on spending to the size of the federal budget over time, measurement becomes tricky. Many bills affect spending, but are not directly about spending, per se. And identifying a particular lawmaker’s role in the labyrinthine federal budget process is much more difficult that observing her position on, say, gun control. How, then, do we know if a member of Congress is acting to increase or decrease spending? Moreover, measures of subnational preferences have been notoriously elusive. Most polls are designed to be representative at the national level, making it difficult for scholars to get accurate measures at the state or Congressional district levels. Our paper takes on these challenges of studying money and representation. On the opinion side, we use multilevel regression and poststratification (MRP) to estimate state-level preferences on spending using national polls. On the outcome side, we take advantage of the 2009 American Recovery and Reinvestment Act, an unusually direct expression of lawmakers’ positions on federal spending. Our analysis of the Act goes beyond the industry standard, as we study not only the roll-call vote on the stimulus package, but also the distribution of stimulus funds. We then step back and analyze the relationship between opinion and appropriations more broadly, given recent controversies surrounding so-called “red state socialism.” When the House of Representatives delayed voting on a Hurricane Sandy relief bill in January 2013, New Jersey Governor Chris Christie lambasted co-partisan Speaker John Boehner, arguing that hard-hit states New Jersey and New York were “donor states” to the federal coffers and deserved assistance in their own time of need. This theme resurfaced more recently as he and Senator Rand Paul engaged in a public battle over federal spending in their respective states. Dubbing Christie the “King of Bacon,” the Kentucky legislator claimed the New Jersey governor was part of a fiscally reckless group “unwilling to cut the spending, and they’re ‘Gimme, gimme, gimme— give me all my Sandy money now”’ (Blake 2013; Gold 2013). Christie fired back against Paul’s claim that he was “bankrupting the government,” noting that Kentucky receives $1.51 from the federal government for every dollar it contributes in taxes, while New Jersey receives only 61 cents. This argument revived a popular infographic on so-called “red state socialism,” which had appeared on the front page of internet aggregator sites Digg and Reddit leading up to the 2008 presidential election (see F IGURE A.1). It shows that net “takers” of federal money (i.e., those receiving more federal funds than they pay in federal taxes) tend to be red states, where voters presumably prefer smaller government and less spending. Of the 32 net beneficiary states in fiscal year 2005, 27 voted for George W. Bush in the 2004 presidential election and, more recently, 24 voted for Mitt Romney. Conversely, net “givers” tend to support Democrats, who presumably promote the kinds of redistributive policies that appear not to benefit these states. Fourteen of the 18 net contributor states in fiscal year 2005 voted for John Kerry in 2004 and 17 (all but Texas) voted for Barack Obama in 2012. Now, there is nothing pernicious or even necessarily surprising about the existence of “giver” and “taker” states. Reallocation of wealth between people and regions is common practice in liberal democracies. Indeed, the notion that government should mitigate human suffering through redistribution played a key role in the New Deal, which established the modern American state. It would be curious, however, if support for federal spending was really systematically lower in states receiving relatively large amounts of federal money. This phenomenon requires more rigorous academic analysis before we can accept it. What appears to be a puzzle could easily be an illusion, as the “red state socialism” infographic and related narratives have a critical shortcoming. Republican vote share is a very coarse proxy for attitudes about spending. Just because someone votes Republican does not mean he is fiscally conservative. Many factors go into partisanship and vote choice. Over the past few decades, social issues have become just as important to the Republican brand as fiscal issues. Also, Republican vote share may be correlated with outlays for reasons unrelated to opinion on spending. For example, retirement benefits, which comprise a large portion of federal spending, accrue to individuals over a particular age; and we know that in recent presidential elections, older people tended to vote Republican (Abramson, Aldrich and Rohde 2009). But, there is not necessarily a relationship between fiscal conservatism and federal payments here. To see if there is a true negative correlation between support for spending and outlays to states, we need to analyze state-level opinion specifically on spending. Our paper proceeds as follows. We begin by detailing some theoretical and empirical challenges in the study of opinion and fiscal politics, and how we handle them. We then review the process by which we estimated state-level opinion on spending, and provide a summary of our estimates. We then analyze the relationship between statelevel opinion and the Senate roll-call vote on the stimulus, as well as the distribution of stimulus money to states. In sum, we find a positive correlation between constituent support for the stimulus and Senators’ votes on the measure, but no relationship between opinion and appropriations. Looking beyond the stimulus to net spending more broadly (i.e. total federal money received by each state divided by federal taxes paid), we find a positive correlation with several different measures of opposition to federal spending. In other words, drilling down below the “red state socialism” infographic, we find that there is in fact a federal spending paradox, even once we substitute more direct measures of attitudes on spending for Republican vote share and analyze different types of spending and question wordings. While a more precise analysis reveals nuances, it does not erase the odd correlation between opposition to spending and net outlays. Overall, we find that the role of public opinion in fiscal politics is complex and deserves further attention. 2 Theoretical and Empirical Challenges in the Study of Opinion and Outlays Since politics boil down to “who gets what, when, how,” it should come as no surprise that the dynamics of distribution have long interested political scientists (Lasswell 1936). Thus far, scholars have focused primarily on institutional factors. Numerous studies have shown that an area’s “volume” of representation (i.e., its number of legislators per capita) is a robust predictor of allocations thereto. When a state gains a representative in the United States House due to the constitutionally-mandated de- cennial reapportionment, more federal outlays follow (Elis, Malhotra and Meredith 2009). Smaller states, which have more Senators per capita, also benefit disproportionately from federal spending in the United States (Atlas et al. 1995; Lee 1998; Lee and Oppenheimer 1999; Lee 2000).1 Similar patterns appear at the subnational level as well. Before the Supreme Court’s 1960s equal apportionment decisions, malapportionment in U.S. state legislatures affected the distribution of state funds to localities (Ansolabehere, Gerber and Snyder 2002). In short, more representation means more money. Findings regarding other factors, like party, are more mixed. Some scholars argue that representation by a majority party member may affect a state or Congressional district’s receipts from the federal government (Carsey and Rundquist 1999; Alvarez and Saving 1997; Balla et al. 2002). Others argue this influence extends only to certain types of outlays. For example, Berry, Burden and Howell (2008) find that districts represented by majority party members received slightly more federal funding from low-variation entitlement programs, but enjoyed no advantage in new program spending (see also Lee 2003; Levitt and Snyder 1995). Examining both the substance and volume of spending, Bickers and Stein (2000) show that while the content of outlays shifted away from entitlement programs to contingent liability programs when Republicans assumed control of Congress after the 1994 elections, overall levels of outlays to districts changed little. While these studies have enriched our understanding of institutional influences on distributive politics, the role of public opinion in this arena remains mysterious. This is surprising, since many have studied the public’s influence in other parts of the policy process. In their seminal work on policy responsiveness, Page and Shapiro (1983) show that large shifts in opinion (5 points or more) correlate with policy change, 1 This is also true in other countries with bicameral legislatures and malapportioned upper chambers (e.g. Horiuchi and Saito 2003). though opinion and policy were in conflict in a significant minority of cases. Erikson, Mackuen and Stimson (2002) reach a comparable conclusion using an aggregate policy measure, which they find follows public “mood,” and Lax and Phillips (2009a, 2012) discover a similar relationship between opinion and policy at the state level. Moving from policy to roll-call votes, Krimmel, Lax and Phillips (2013) observe a positive relationship between policy-specific constituent opinion and lawmakers’ votes on gay rights at the national level, despite significant amounts of incongruence between constituent preferences and roll-call votes amongst Republican members of Congress. Given these findings, we might expect to see a positive correlation between opinion and fiscal outcomes as well—Members of Congress representing constituents who are more supportive of spending might be more likely to vote for a significant spending measure like the stimulus, and be especially motivated to secure appropriations for their states or districts. Yet, the distributive politics literature has mostly overlooked the role of public preferences. A few studies have examined the relationship between public opinion and the federal budget. Hartley and Russett (1992) show that opinion influences defense spending. Looking more broadly, Wlezien (2004) finds that responsiveness varies across different types of spending. On some issues, like crime and foreign aid, Congress does not appear to consider public preferences in setting the budget. On others, like welfare, the budget does reflect changes in public support. Still, these studies do not tap into the geographic nature of fiscal politics. Below the highest level of aggregation, looking at national opinion and the federal budget, we know little about the relationship between opinion and appropriations. Why have distributive politics scholars overlooked public opinion, and scholars of representation ignored fiscal outcomes? Perhaps the relationship between money and opinion seems too obvious to be interesting. If people always want goodies and rep- resentatives always want to please their constituents, then lawmakers should have unbounded incentives to increase cash flow to their states or districts. Recent controversies in fiscal politics suggest a more nuanced story, however. While the distributive politics literature tends to assume more money is always better, on public opinion side, we know Americans have complicated views on spending. The gap between these two literatures remains unbridged, and our understanding of fiscal politics incomplete. As described earlier, measurement challenges may have impeded work in this area as well. It is important to overcome these hurdles, given the unfortunate pattern of recent budget crises paralyzing the government. The parties’ inability to negotiate a timely solution to the 2011 debt ceiling crisis led to a historic downgrade of the nation’s credit. Their more recent failures to reach budget deals have triggered sequestration, caused the first government shutdown in almost two decades, and further undermined already low public confidence in Congress. As there is no clear end in view, we need to better understand the sources of conflict on fiscal issues. Analyzing the role of public opinion in fiscal politics is a key step in this direction. The relationship between state-level opinion on spending and federal outlays could also have important implications for democratic quality. If states whose populations oppose spending do indeed get more money from the federal government, as the “red state socialism” narrative suggests, we may be facing some type of system performance problem. Perhaps those most exposed to federal spending are most likely to observe its pathologies (e.g., inefficiency, poor program quality, etc.). It is also possible that a disproportionately large percentage of people in “taker” states, many of which have long histories of racial segregation, oppose spending because they believe it benefits people not “like them.” These possibilities raise different but serious concerns about the relationship between government and its citizens, as well as citizens’ relationship to each other. Both affect the quality of a liberal democracy. 3 Data and Methodology To create a measure of net outlays to states, we need data on both total taxes paid by state and total outlays to states. Our data cover fiscal years 1998 to 2008. Data on state tax burdens come from IRS Databook Table 5, Internal Revenue Gross Collections by State. Total internal revenue collections, or total taxes paid, include corporate income tax, individual income tax, individual employment tax, estate tax, gift tax, and excise tax. The data on federal expenditures to states come from the Census Bureau’s Consolidated Federal Funds Report (CFFR). Federal expenditures fall into one of four categories: grants-in-aid to state and local governments, procurement contracts, direct payments, and salaries and wages. Grants-in-aid to state and local governments include both block and categorical grants, awarded by competitive processes, by formulas, or both. Depending on the nature of the grant, the funds may go directly to state, local, or tribal governments or to non-profit organizations. The original grant recipient may pass funds to subcontractors, like lower levels of government, organizations, or individuals. In FY 2010, the federal government awarded $683.4 billion in grants, representing 20.9% of total expenditures. Procurement contracts include obligations to private domestic contractors from the procurement activities of all federal agencies. This excludes procurements from the legislative and judicial branches and foreign procurements. In FY 2010, these contracts were worth $516.7 billion, 15.8% of all federal expenditures.2 Direct payments are divided into two categories: retirement and disability payments to individuals and all other direct payments, the vast majority of which are also payments to individuals. Retirement and disability payments include all types 2 In general, grants-in-aid and procurements represent obligations and may not equal actual expenditures. of social security payments, government employee retirement benefits, and certain veterans’ benefits. Other direct payments include Medicare benefits, unemployment compensation, housing assistance, and student financial assistance. In FY 2010, direct payments composed over half of all federal expenditures to states, with retirement and disability payments totaling $914.8 billion (27.9%) and other payments totaling $818.6 billion (25.0%). Salaries and wages includes the salaries and wages of all federal employees and also other personal compensation like housing allowances. In FY 2010, this totaled $342.9 billion, or 10.5% of federal expenditures to states.3 To gather public opinion data, we searched the Roper Center’s iPOLL Databank for survey questions containing any variation of the word “spend” (e.g., spend, spending, spends, etc.). For this analysis, we used general questions on spending from surveys that included state identifiers and used national adult samples of at least 1,000 people, or which could be pooled with similar questions from other surveys to yield samples of this minimum size.4 TABLE A.1 lists all of the survey questions we employed. Combining responses to similar questions yielded six categories: Stimulus, Economy vs. Deficit, Jobs vs. Deficit, Pork, Infrastructure, and Jobs in District. We did not pool questions before and after 2008 because this was the year of the economic collapse and Barack Obama’s election, which seem to have energized opposition to spending. This decision only affected questions in the “pork” category. Below is a list of typical questions for each category. • Stimulus. “Do you think the federal government should spend more money to try to boost the economy in a way that creates jobs, or do you think that whether or not jobs are created should be left to the private sector?” 3 The CFFR also contains data on contingent liabilities (i.e., loans and insurance). These are excluded from our analysis. 4 Lax and Phillips (2013) show that MRP creates accurate state-level estimates with national polls of at least 1,000 respondents. • Economy vs. Deficit. “Which of these do you think is more important right now— increasing federal spending to try to improve the economy, even if it sharply increases the federal budget deficit, or avoiding a big increase in the federal budget deficit, even if it means not increasing federal spending to try to improve the economy?” • Jobs vs. Deficit. “Which comes closer to your own view? The federal government should spend money to create jobs, even if it means increasing the budget deficit. The federal government should not spend money to create jobs and should instead focus on reducing the budget deficit.” • Pork. “Members of Congress sometimes add provisions to legislation that include government spending projects for their own home states and districts, sometimes known as ’earmarks’ or ’pork.’ Do you think this practice is generally acceptable or not acceptable?” • Infrastructure. “As you may know, the government is considering several other proposals to address economic problems facing the nation. Do you think it is the right thing or the wrong thing for the government to spend billions of dollars to substantially increase spending on roads, bridges, and other public works projects?” • Jobs in District. “Do you want your representative in Congress to fight for more government spending in your congressional district, in order to create jobs, or do you want your representative to fight to cut government spending, even if it means fewer jobs in your district?” All of these surveys were conducted using samples that are representative at the national, but not state level. To calculate opinion by state, we employ a two-stage technique called multilevel regression and poststratification (MRP), developed by Gelman and Little (1997) and refined by Park, Gelman and Bafumi (2004) and Lax and Phillips (2009b, 2013). First, we estimate a multilevel model of individual opposition to spending for each of our seven question categories as a function of respondent demographic characteristics and state characteristics. From this model, we can predict how likely a particular type of respondent living in a particular state would be to favor, say, reducing the deficit over spending money to create jobs. By partially pooling respondents across states and explicitly modeling the hierarchical structure of the data, the multilevel model allows us to use all of the information about each demographic variable in the data to learn about the association between that variable and the survey response.5 Specifically, we model the probability of opposing spending as a function of demographic categories: gender (g=male, female), race (r=black, hispanic, white/other), age (a=18-29, 30-44, 45-64, 65+), and education (e=less than high school, high school degree, some college, college degree, and post-graduate degree), plus state (50 states and the District of Columbia, indexed by s). This allows us to estimate survey responses for 6,120 different demographic-geographic types (indexed by c). The model is gender age race education state P r(yc = oppose) = logit−1 (β + αg[c] + αr[c] + αa[c] + αe[c] + αs[c] ). (1) Each demographic characteristic (αggender , αrrace , αaage , αeeducation ) is a modeled (or ran5 The amount of pooling depends on features of the data themselves. Using a standard regression model on grouped data, a researcher has only two options: either pool groups completely, ignoring any group level variation, or do not pool them at all and estimate separate regressions by group, ignoring any similarities between individuals across groups. A multilevel model is a compromise between these two extremes, a weighted average that is a function of across group variation, within group variation, and the number of observations in a group. Group level parameters are pulled more toward their overall mean (leading to more “shrinkage”) when across group variation is low, within group variation is high, or within group sample size is small (Gelman and Hill 2007; Lax and Phillips 2013). dom) effect for respondents of that type. For example, 2 αrrace ∼ N (0, σrace ), for r = black, hispanic, white/other (2) For state modeled effects, we allow the mean to be a function of a measure of state-level public opinion liberalism (ideology): 2 ), for s = Alabama, ... , Wyoming αsstate ∼ N (β ideo ideologys , σstate (3) One can think of state modeled effects as a corrective for what a purely demographic model would predict about opinion. They account for the possibility that people’s preferences are influenced not only by their personal demographic characteristics, but also their surroundings. An average white woman aged 18-29 with a high school diploma living in Connecticut may have different preferences over spending than a woman with the same personal demographic characteristics living in Louisiana, for example. State modeled effects can be a function of any number of state characteristics (e.g., average income, presidential vote, percent Evangelical, region, etc.). Lax and Phillips (2013) show that using one state-level predictor is enough and that the measure of ideology we use, a demographically-purged state predictor of opinion liberalism (DPSP) that they created, performs best.6 DPSP is a vector of state random effects estimated from a demographic model of survey responses over a wide variety of issues—essentially, what is left over after accounting for the relationship between demographics and opinion liberalism. Unlike other possible predictors of state-level opinion, like presidential vote, DPSP is not itself correlated with demographics, and so it explains more of the residual variation in opinion that demographics cannot predict, 6 Additional state-level predictors do not help, and sometimes hurt the accuracy of the estimates. making it a better corrective.7 Estimating MODEL 1 allows us to determine the predicted level of opposition to each type of spending for each of the 6,120 demographic-geographic types created by our model (e.g., the probability that a black man from New York, aged 30-44 with a college degree, will oppose stimulus spending). In the second stage, poststratification, we use these predictions to estimate the level of opposition to each type of spending at the state level. We do so by weighting the predicted opposition for each demographicgeographic type (θc ) by the frequency of that type in each state (Nc ), according to Census data. Thus, the MRP estimate of state-level opposition to spending is simply: oppositionMRP state, s = X c∈s Nc θc / X Nc (4) c∈s Several studies have demonstrated MRP’s accuracy in estimating state and even congressional district-level opinion (Park, Gelman and Bafumi 2004; Lax and Phillips 2009b, 2013; Warshaw and Rodden 2012). 4 Descriptive Results: Opinion on Spending What does public opinion on spending look like? We begin by presenting some basic national-level results to demonstrate overall levels of opposition to the stimulus and a few other types of general spending. We then proceed to discuss state-level variation. 4.1 National Level F IGURE 1 displays the percentage of survey respondents giving the fiscally conservative answer to each type of question.8 Overall, people are quite divided over government spending. When asked about general stimulus spending, a large minority (46%) op7 Lax and Phillips (2013) constructed DPSP using opinion data from 39 surveys sets with 200,000 observations. DPSP values are included in the MRP package for R. One possible limitation of DPSP is that, in its current incarnation, it may be useful only for recent survey data. 8 All percentages are survey weighted averages of respondents who enter into the MRP model, those for whom we have complete demographic information. posed it. This is the most general category in our analysis, as the questions do not refer to any particular program, pose any sort of tradeoff, or use words with negative connotations like “pork.” These, and almost all of the other surveys we use, were conducted after the 2008 financial crisis began. Even having experienced the worst economic collapse since the Great Depression, Americans are conflicted over the government’s role in boosting the economy. Opposition is as high or higher when questions pose a tradeoff between spending and the deficit. A majority of respondents (53%) oppose general stimulus spending if it would increase the national deficit, and 46% oppose spending to create jobs if it would increase the national deficit. National−Level Fiscal Conservatism Stimulus Economy vs. Deficit Jobs vs. Deficit Pork (pre−2008) Pork (post−2008) Infrastructure Jobs in District 0 10 20 30 40 50 60 70 Figure 1: National-Level Results. This graph summarizes the national-level averages for the seven question categories in our analysis, accounting for survey weights. When asked about infrastructure spending, the classic example of a pork project in the distributive politics literature, opposition shrinks by more than half (to 22%), compared to general stimulus spending. This relatively low opposition to public works spending is consistent with the distributive literature’s assumption that constituents like pork. However, when asked explicitly about pork and earmarks, people are far less supportive.9 Pork and earmarks were unpopular before 2008, and became even 9 Half of the respondents were asked about earmarks, and half were asked about earmarks or pork, more so afterwards (opposition grew from 63% to 73%). The economic crisis could have made people more fiscally conservative, or this could be a partisan response to Barack Obama’s election, which spurred the growth of the Tea Party (Skocpol and Williamson 2012). Interestingly, despite this widespread opposition, people do not think their own representatives are abstaining—only about 20% of people think their representatives do not sponsor measures containing pork.10 Unlike “public works,” the terms “pork” and “earmarks” are associated with waste. This could explain why support for this type of spending is weaker when these terms are used. It is also possible that people are weighing the pork they might get against the pork that might go to other districts, and dislike the latter more than they desire the former. It is not necessarily surprising that people would oppose spending that could benefit other districts. Even when the survey specifies that spending would benefit people living in the respondent’s own district, however, the level of opposition remains striking. When asked “Do you want your representative in Congress to fight for more government spending in your congressional district, in order to create jobs, or do you want your representative to fight to cut government spending, even if it means fewer jobs in your district?” a large minority (39%) say they would like their representative to fight to cut spending. This is a strong finding, considering the question poses no tradeoff with the deficit and does not mention unpopular terms like “pork” or “earmarks.” Moreover, the question asks specifically about spending for jobs, which people support more than general stimulus spending. Together, these national level results suggest that people’s preferences about spending depend on the type of spending, even within these very general categories. Overall, and there were no meaningful differences between the two groups. People seem to see these two things as synonymous today. 10 See table A.1, category “MC Pork” for the question wording. however, support for spending is not very strong. These results—particularly the high level of opposition to spending in one’s own district—challenge the popular assumption in the distributive politics literature that goodies are always good. 4.2 State Level While national-level results provide a useful summary of opposition to spending, they obscure significant variation across states that has, until now, remained unobserved. This has been unfortunate since distributive politics are, by nature, geographicallybased political phenomena. There is a wealth of data on state-level outlays, taxes, and political and demographic characteristics. But, due to the absence of state-level polling, it has not been possible to analyze these data in conjunction with public opinion. F IGURE 2 plots our estimates of state-level opposition to general stimulus spending. As with the national-level results, all state-level results report the percentage of people offering the fiscally conservative answer (here, that would be opposition to stimulus spending). States labeled in red supported John McCain in the 2008 presidential election, while states labeled in blue supported Barack Obama. The dotted vertical line marks the national average response, as in F IGURE 1. The range in opposition is large across states, from a low of 40% in New York to a high of 59% in Utah.11 The fact that this range includes the 50% mark means majorities in different states want different things. In 15 states, a majority opposes stimulus spending, while in 35 states a majority favors it.12 Support for spending is further complicated by its correlation with state population. Because more populous states tend to favor spending more than less populous states, the national average level of opposition to spending is skewed down11 If anything, MRP’s partial pooling understates cross-state variation, especially if the true range is large and if the sample size is small. Given data limitations, one can minimize shrinkage by using a state level predictor when creating the estimates, as we do (Lax and Phillips 2013). 12 These counts exclude Washington, DC. Stimulus 20 30 40 50 60 UT SD NE ND OK WV WY TN AR KY AL ID IA IN MT KS MS MO MN WI AK NC OH OR SC CO ME NV AZ MI VA TX LA PA GA NH FL HI WA DE IL VT NM NJ CT CA RI MD MA NY 20 30 40 50 60 Figure 2: Opposition to Spending by State. This graph plots state-level opposition to general stimulus spending. John McCain won the states labeled in red in the 2008 election, while Barack Obama won the states labeled in blue. The dotted vertical line indicates the national average. ward. Only 14 states have below average opposition to stimulus spending, leaving 36 states with above average opposition. Thus, national level data conceal potential sources of conflict over fiscal politics. There is a clear correlation between support for the Republican presidential candidate and opposition to stimulus spending, though the states are not perfectly sorted by party. For example, solidly red states Georgia and Louisiana display below average to middling opposition to spending (here and in F IGURE A.2). This underscores the point that Republican vote share is a rough substitute for opinion. Because national parties bundle issues in ways that pose dilemmas for individuals with cross-cutting preferences, a Republican vote does not necessarily suggest fiscal conservatism. In analyzing the spending paradox, it is therefore important to use actual measures of state-level opinion rather than partisan proxies. F IGURE 3 illustrates regional patterns. For general stimulus spending, the bluest areas (indicating the lowest opposition) are in the northeast and on the west coast. We see the strongest opposition (states with yellow to orange shading) in the southern and Plains regions. Nowhere do we see very dark blue or red shading, which would indicate extreme support or opposition. Stimulus 0.7 0.6 0.5 0.4 0.3 Figure 3: Map of Opposition to Spending. Shading indicates the degree of opposition to general stimulus spending using the same data as in F IGURE 2. F IGURE A.2 plots state level estimates of opposition to spending for survey questions that juxtapose support for spending with support for decreasing the deficit. The patterns here are similar to those for stimulus spending in general. When posed as a tradeoff with the deficit, opposition to spending to create jobs and spending to stimulate the economy vary by approximately 20 points across states. Also, as with general stimulus spending, some traditionally conservative states appear closer to traditionally liberal states than one might expect. For example, in the case of “Jobs vs. Deficit”, Georgia, Louisiana, Mississippi, and South Carolina display low to middling opposition to spending. This seemingly odd pattern becomes less surprising when we consider that Georgia, Mississippi, and South Carolina ranked 39th, 43rd, and 46th out of all states in unemployment, with rates 0.6, 0.9, and 1.6 percentage points above the national average (an already high 9.6%) in 2010, the year these polls were conducted (Bureau of Labor Statistics 2010). Once again, opinion on spending does not translate straightforwardly into support for a particular party in presidential elections. F IGURE A.3 displays our state-level estimates for questions about pork barrel spending. Compared with more general spending questions, the cross-state variation here is very small, particularly for opposition to pork before 2008. In every state, large majorities support pork-barrel type (i.e., infrastructural) projects, unless they are called “pork” or “earmarks” explicitly, in which case large majorities oppose them. Senator Paul might be surprised to discover that only 65% of his constituents oppose pork barrel spending, one percentage point more than Governor Christie’s constituents in New Jersey. More broadly, there does not appear to be much disagreement across states on pork. F IGURE A.4 plots opposition to spending for jobs in one’s own district. Most striking here is the large range of opposition to this type of presumably universally attractive spending. If ever we should see overwhelming support, it should be for outlays of this nature—distributive spending for you and the people who live close to you that creates local jobs and is not tainted by the terms “pork” or “earmarks.” Interestingly, the states are not as cleanly sorted by presidential vote as they are for other categories of spending. In fact, they are in a very different order here. We will return to this point in the discussion. F IGURE A.5 illustrates differences between the regional pattern for this question and the others. 5 State-Level Opinion and the American Recovery and Reinvestment Act In February 2009, President Obama signed the The American Recovery and Reinvestment Act of 2009 (ARRA or the “Stimulus”) into law. The broad purpose of the ARRA was to provide relief from the Great Recession, the economic downturn that began in December 2007, through direct spending in the economy and expansion of programs to aid individuals. As of December 2013, funds allocated under the ARRA total $840 billion, including $290.7 billion in tax benefits, $264.4 billion in entitlements, and $261.2 billion in contracts, grants, and loans.13 5.1 Roll Call The ARRA passed the House 244 to 188 and the Senate 61 to 37.14 In thinking about issues of representation, scholars typically want to know whether and how public opinion predicts votes on a bill. In this case, we can compare the Senate vote on the ARRA with state-level opinion estimates on the stimulus.15 Our opinion measure is the level of opposition in a state to using government money to create jobs. The level of opposition to stimulus spending in the state does indeed predict how 13 See http://www.recovery.gov/arra/Transparency/fundingoverview/Pages/fundingbreakdown.aspx Rep. Ginny Brown-Waite (R-FL5) and Sen. Judd Gregg (R-NH) abstained. Sen. Al Franken (D-MN) had not yet taken office due to a disputed election result. 15 We are working on congressional district level estimates as well. 14 a Senator voted on the ARRA. F IGURE 4 shows the bivariate logit curve for the relationship between opposition to stimulus spending and vote against the ARRA. This relationship is positive and statistically significant (with logit coefficient 0.23 and standard error 0.06, clustered by state).16 The dotted lines in the figure mark fifty percent opposition to stimulus spending in the state and a fifty percent chance of a Senator voting against the ARRA. The logit curve nearly passes through the point where these dotted lines cross, meaning that when opinion on stimulus spending is equally divided in a state, that state’s Senator was, on average, just as likely to oppose the ARRA as support it. 2009 Stimulus Senate Vote Against ARRA 1.0 FLGA LA TXAZ NV SC OH NC AK MO MSKSINIA IDAL KYTNWYOK NE SD UT 0.8 0.6 0.4 0.2 0.0 NJNMVT NY MD MA CA RI CT DE ILWA FL HINH PA LA VA MI ME CO NV OR OH NC AK MN MO WI MT INIA 40 45 AR WVND NE SD 50 55 60 Opposition to Stimulus Figure 4: Opposition to Stimulus and Senate Vote. Bivariate logit curve. The positive relationship between state opposition to stimulus spending and vote 16 This is robust to the exclusion of Utah. against the ARRA in the Senate persists even after controlling for indicators of economic need in the state. Controlling for change in unemployment from 2008 to January 2009, a measure of short term need in a state, and (log) state median income, a measure of persistent need, we still find a positive and statistically significant relationship between state level opposition to stimulus spending and vote against the ARRA (with logit coefficient 0.22 and standard error 0.06, clustered by state). Interestingly, change in state level unemployment does not predict the vote on the ARRA, and higher median income makes a Senator less likely to oppose the ARRA. Of course, the positive relationship between opinion and voting does not tell the whole story. There were still incongruent votes for and against the ARRA. That is, when a state majority opposed stimulus spending, a Senator sometimes voted for the ARRA nonetheless, and vice versa. Sixty-seven of the ninety-eight votes cast in the Senate were congruent with state majorities.17 Incongruence did not cancel out, as most of the incongruent votes were against the ARRA.18 One might reasonably conclude that this provides some evidence in favor of a positive connection between pubic opinion and roll call voting on an issue of spending. However, if we broaden our inquiry from the ARRA roll call vote to actual outcomes (i.e. distribution of stimulus money to states), public opinion seems to play little role at all. 5.2 Spending The ARRA included a provision for the creation of Recovery.gov, a website allowing users to track the distribution of entitlements and benefits to agencies and the distri17 This is roughly the same congruence rate Krimmel, Lax and Phillips (2013) find for Congressional roll-call votes on gay rights issues. Studying congruence between state-level opinion and state policy across a wide range of issues, Lax and Phillips (2012) find a lower level of congruence, on average (48%). None of those policies were specifically about spending, however. 18 Nineteen Senators from states where majorities were in favor of stimulus spending voted against the ARRA. Only twelve Senators from states where majorities were opposed to stimulus spending voted for the ARRA. bution of contracts, grants, and loans to their recipients. Because contracts, grants, and loans are tagged by recipient zip code, it is possible to track the geographic distribution of these particular funds.19 In order to facilitate comparisons across states, we created a relative per capita spending index for the contracts, grants, and loans allocated to each state under the ARRA. The spending index SI for state s is the ratio of the per capita allocation of contracts, grants, and loans to that state (CGLs divided by state population) to the per capita allocation of all contracts, grants, and loans to all states.  SIs =  CGLs pops P  s CGLs uspop (5) The denominator represents what the per capita allocation would be if ARRA contracts, grants, and loans were distributed evenly across the US.20 Thus, if a state’s SIs = 1, then that state received the same amount of CGL it would have had the per capita distribution been the same across all states. A state with SIs > 1 received more than it would have under even allocation, and a state with SIs < 1 received less. For example, a state with SIs = 1.2 received twenty percent more CGL under the ARRA than it would have had the ARRA distributed per capita CGL equally across states. The states with the median SIs for contracts, grants, and loans are Michigan (1.02) and New York (1.03). The state that received the least per capita is Florida (SIs = 0.69), and that state that received the most is Alaska (SIs = 3.8).21 It would not be unreasonable to expect a positive correlation between support for 19 It is not possible to track the geographic distribution of tax benefits because the ARRA did not require the Treasury Department to report this information (per communication with Recovery.gov webmaster). It is possible to track entitlements by compiling information from each agency report, and we are in the process of doing this. 20 Population measures are from the 2010 Census. 21 Alaska is frequently an outlier in receipt of per capita outlays. However, each of the findings discussed in this section are robust to Alaska’s exclusion. the stimulus and the distribution of contracts, grants, and loans. High support could reflect high levels of demand for stimulus money, after all, which could lead to more applications for contracts, grants and loans. However, this is not what we observe. 2009 Stimulus Relative Per Capita Stimulus Funding AK 3.5 3.0 2.5 2.0 VT MT 1.5 1.0 SD ND NM WA AZ WY CO ID TN NV DEHI ME SC WV MS IL NYCA MIOR MOKS KY VA AR OK NC AL MN OH IA NH LA PA NE GA CT TX WI IN NJ FL MD MA RI 40 45 50 55 UT 60 Opposition to Stimulus Figure 5: Opposition to Stimulus and ARRA Outlays. States colored by Senate vote: red for two votes against, blue for two votes in favor, purple for split vote. F IGURE 5 shows the relationship between state level opposition to stimulus and state relative per capita spending for contracts, grants, and loans from the ARRA. States colored red in the figure are states in which both Senators voted against the ARRA, states colored blue are those in which both Senators voted for the ARRA, and states colored purple are those in which Senators split their votes. We find no clear relationship between state-level opposition to stimulus spending and SIs . This is still true after controlling for change in state unemployment from 2008 to January 2009 and (log) median income.22 22 Neither change in state unemployment from 2008 to January 2009 and (log) median income are These results encourage us to think more critically about what support for spending means. We might expect it to signify demand for spending in a state or district, but preliminary evidence from the ARRA suggests this is not the case. It is possible that a disproportionately large number of contracts, grants, and proposals from states supporting the stimulus were rejected. However, it seems more likely that the meaning of public opinion on spending simply needs more scrutiny. Of course, it is possible that preferences on spending do not have real substantive meaning. Most people probably do not have true opinions on technical or procedural matters; indeed, Wawro and Schickler (2010) show that short-term partisan thinking affects opinion on the filibuster today. Spending could be like the filibuster, in that support might be contingent on (or at least strongly affected by) content. Or, people— Keynsians, for example—could have true opinions on spending. It is also possible that spending gets conflated with other things in people’s minds; that is, spending might become a proxy for other issues. While we cannot settle this matter fully in one paper, we can look to see if what we find with respect to the ARRA is true more broadly. What relationship, if any, is there between opinion on different types of general spending and total outlays to states? Is there really a federal spending paradox, as the “red state socialism” infographic suggests? We employ our state-level opinion estimates to answer this question. Our data also offer a glimpse into the possibility that opinion on spending is not simply about money, but serves as a proxy (or quasi-proxy) for racial resentment. 6 Broader Relationship Between Opinion and Outlays While the graphs in the section presenting descriptive results are suggestive, we need a more systematic analysis to determine if there is truly a positive correlation between associated with relative per capita CGL in the state. This is consistent with what ? find at the county level. opposition to spending and net outlays. We begin by confirming the relationship between net spending and Republican vote share, and then replace Republican vote share with more precise measures of opinion. To summarize, whether we look at support for the party of fiscal conservatism or actual opposition to spending, we find a positive relationship to net outlays. 6.1 Outlays and Republican Vote Share The analysis underpinning the charge of “red state socialism” in the popular infographic was based on the ratio of total federal outlays to states from the CFFR in fiscal year 2005, total tax burden by states in fiscal year 2005 determined by the Tax Foundation using a proprietary method, and vote for George W. Bush in 2004.23 Our reconstruction of the tax side of the ratio using the data described above yields substantively similar results. We find that 30 states were net “takers” of federal money in fiscal year 2005 and, of those, 22 went for Bush in 2004. Of the 20 states that were net “givers” in 2005, 11 voted for Kerry.24 This positive association between net “takers” and Republican presidential voting is not confined to the 2004 presidential election. F IGURE 6 displays year-by-year graphs of the ratio of total federal outlays to total taxes paid by state and the two-party vote for the Republican presidential candidate in the current or previous election. The black line indicates the linear fit. In each year, the relationship between the two is positive. We can also pool this data and account for common time shocks with year fixed effects. We find that a ten percentage point increase in the two party vote for the Re23 The Tax Foundation adjusts the annual tax burden in states to be deficit neutral. That is, the amount of the annual deficit is allocated to states based on their share of the actual yearly tax burden. The Tax Foundation also accounts for “income shifting and economic incidence” (email correspondence with William McBride, Chief Economist). 24 On average, our simpler ratio is $0.16 higher than the Tax Foundation’s (median difference was $0.08). This affected the giver/taker status of 10 states. Four states that were givers under the Tax Foundation’s calculus became takers in our analysis: Florida, New Hampshire, Oregon, and Washington. Six states that were takers according to the Tax Foundation became givers in our analysis: Arkansas, Georgia, North Carolina, Nebraska, Ohio, and Oklahoma. Spending Paradox: 1998-2008 1998 1999 WVNM AK MS ND LAAL MT HI ME KY WY SC VA SD AZ OK IA UT VT ID FL NC TN MO IN AR PA RI OR KS MD WA NH NV GA TX NE OH MA MIWI CO NYCTCA IL NJ MN DE MT MS ND AK LA HI AL SD ME KY WY VA SC AZ VT IAFL OK ID UT TN NC KS OR PA MD MO IN RI AR NH CA WA NV TX GA NE WI CO MA OH MI NY CT IL NJ MN DE 2002 2001 NMWV WVNM MT MS AK HI ND ME AL SD VA KY LA AZ SC WY VT IAFL OK ID TN NC UT RI MD OR AR KS WA NH PA IN MO CA NV TX CO MA NE WIOH GA MI NY CT IL NJ MN DE Spending / Taxes 2000 2003 2004 NM WV MS NDAK MT AL SC HI ME LA VA KY SD WY UT IA AZ VT OK TN NC INKS ID MO OR RI PAFL ARGA WA NE MD NV NH OH CA MIWI TX CO MA NYCTIL NJ MN DE 2005 NM NMWV WV NM MS NDAK MT HI AL SD SC KY VA VT UT WY LA OK MD IA AZ ID TN NC OR MO INKS NE PAFL WA RI NH ARGA CA NV MI TX WI MA OH CO NYCTIL NJ DE MN ME 2006 NM HI ME MS MT HI ND AL ME SC VA KY SD AZ LA WY MD VT UT OK ID FL OR TN IA KS MO PA IN NC WA NH TX CA RI AR MA GA CO OH NYCT MIWINV NE IL NJ DE MN 2007 AK AL MT SDND KY MD AK MS MT ND AL SD ME VA SC KY AZ LA MD WYUT VT HI IAFL TN OKID INKS OR NH MO PA WA NC RI CA NV MI AR GATX WI MA CO NY OH NE CT ILNJ MN DE WVMS AK ND MT AL SD SC VA KY WY LA AZ VT MD ID UT MOTNIN IAFL OR KS PA NH WA NC OK WI MA RI CAMI ARGATX NE NV CO OH NY CT ILNJ MN DE ME HI 2008 MS NM WV WV LA VA AZ WY IA MO INKS ID UT FL TN SC OR PA NH MI OK NC OH WA RI CA WI MA CO NVARGATX NE NY CT ILNJ MN DE VT NM WV AK WV NM MS AK KYAK AL SC AL ME MT HI ME VA MTSD VA SC HI ND SDND VT KY AZ VT MD AZ MD LA LA ID WY MO TN MI IA IN IAFL MO IN OKID WYUT KS FL OR OR OK NH PA PA MI UT OH TN NV NH NCGATX GA KS CAWA WA MA WI WI NC RI CA AR OH MA NV CO TX NE RI IL NY ILCT CO NEAR NY CT NJ NJ MN MN DE DE Two-party Republican Vote for Last President Figure 6: Republican Presidential Voting and Ratio of Total Outlays and Total Taxes. States that voted for the Republican in the current or previous presidential election are displayed in red and those that voted for the Democrat are in blue. The black line is a linear fit. Total outlays include all federal outlays to states and total taxes include all internal revenue collections. publican presidential candidate is associated with an increase of 20 cents in spending per dollar in taxes paid (i.e., roughly the difference between California and Kansas in 2008).25 The positive relationship holds whether we look at total outlays or each specific category thereof (e.g., direct payments to individuals, grants-in-aid to states and localities, etc.).26 25 If we use mean total ratio over the decade and Republican vote share from 2008, we get a similar result (18 cents in spending per dollar paid in taxes). If rerun the original model with Republican vote share in next election instead of previous, we also get a similar result (20 cents). All of these relationships are statistically significant at the .05 level or better. This and subsequent findings are robust to choosing either the next or the last presidential election for non-election years. 26 In examining different categories, the dependent variable is the ratio of outlays in that particular category to total taxes (since tax burdens are not categorized in the same manner). Since Republicans are theoretically the party of small government and low spending, this tension between spending outcomes and partisanship is interesting. But, the Republican brand encompasses many issues. The correlation we find does not necessarily demonstrate a mismatch is between net federal spending and opinion thereon. We could use the same results to argue that hawkishness or opposition to abortion are associated with more federal outlays. Or, we could claim states favoring lower taxes get more money per tax dollar. There is no reason to choose one interpretation over another. What we really want to know is whether or not there is a tension between spending outcomes and opinion on spending specifically. To this, we now turn. 6.2 Outlays and Opinion on Spending F IGURE 7 plots opposition to stimulus spending on the x-axis against the decade mean ratio of outlays to taxes on the y-axis. The black line indicates the linear relationship, which is positive and statistically significant at the 0.05 level. In other words, when we use actual opinion instead of Republican vote share, the paradox persists. A one standard deviation increase in Republican vote share leads to approximately the same change in net spending as a one standard deviation increase in opposition to general stimulus spending. This finding is not restricted to stimulus spending. While, as discussed earlier, there is much more opposition to some types of spending than others, all types demonstrate a positive relationship to net outlays. F IGURE A.6 plots state-level opinion for different types of federal spending on the x-axis against the decade mean ratio of outlays to taxes. The solid black line shows the linear relationship between these two variables, which is always positive and statistically significant at the 0.05 level. These relationships are substantively significant as well. TABLE 1 summarizes our findings. The first column lists the change in outlays, in cents per tax dollar, associated with a 10 point increase in opposition to spending. The second column lists the change Stimulus NM WV MS 2.5 Spending/Taxes AK MT 2.0 ND AL HI 1.5 VT ME VASC LA AZ SD KY WY IA ID FL TN OK MOKSIN OR PA NC WANH RI GA AR MI CA NV WI TX MA NE COOH NY CT IL NJ MN DE MD 1.0 0.5 40 45 50 55 UT 60 Opposition to Spending Figure 7: Net Spending and Opinion on Stimulus. This figure plots opposition to general stimulus pending against the ratio of federal outlays to federal taxes for each state. The solid black line represents the linear relationship between these two variables, which is positive and statistically significant. associated with a one standard deviation increase in opposition to spending, and the third column shows the magnitude of one standard deviation. A 10 point rise in opposition to stimulus spending is associated with an increase of $0.37 in federal spending per dollar of taxes. For a state with an average tax burden, this would have meant an additional $20 billion in outlays from the federal government in 2008. To compare across different types of spending, we can look at the second column of TABLE 1. It is not necessarily meaningful to compare the effect of a 10 point increase in opposition to pork before 2008 to the same size increase in opposition to stimulus spending, since their standard deviations are so different. In general, change in net outlays associated with a one standard deviation change in opposition to spending are quite consistent across different types of opposition. A one standard deviation increase in opposition to spending to stimulate the economy is associated with an increase of $0.17 in outlays per dollar paid in taxes. The change is slightly larger for Economy v. Deficit ($0.18) and Infrastructure ($0.19), and slightly smaller for Pork before 2008 Category Effect: 10 points Effect: 1 SD 1 SD Stimulus Economy v. Deficit Jobs v. Deficit Pork (pre-2008) Pork (post-2008) Infrastructure Jobs in District Republican Vote Share $0.37 $0.42 $0.28 $2.33 $0.61 $0.69 $0.16 $0.20 $0.17 $0.18 $0.14 $0.15 $0.11 $0.19 $0.06 $0.16 4.6 4.3 5 0.66 1.7 2.8 3.6 9 Table 1: Opposition to Spending and Outlays. This table summarizes the statistically and substantively significant relationship between opinion on spending and net outlays. The first column lists the change in outlays, in cents per tax dollar, associated with a 10 point increase in opposition to spending. The second column lists the change associated with a one standard deviation increase in opposition to spending, and the third column shows the magnitude of one standard deviation. ($0.15) and Jobs v. Deficit ($0.14). While still statistically significant, the change is smaller for Pork after 2008 ($0.11), and even more modest for Jobs in District ($0.06). In sum, we find that there is in fact a spending paradox at the state level. States in which populations express greater opposition to federal spending tend to get more money, per dollar contributed in taxes, from the federal government. By creating the first estimates of state-level public opinion on federal spending, we are able to show that this paradox is not simply an illusion stemming from the use of a rough proxy for opinion in Republican vote share. Looking at opinion specifically, we still find a tension between attitudes and outlays. Estimating state-level opinion on spending also allows us to gauge the magnitude of the relationship. It is interesting to note that this kind of relationship does not exist at the individual level. When we estimate an individual model of opposition to spending, we find a positive relationship between household income and opposition to spending. Thus, the state-level paradox we find is not simply an aggregation of individual-level tensions. 7 Deservedness and Opinion on Federal Spending Why is opposition to federal spending so strangely high in states presumably benefitting the most from it? There could be, and probably are, multiple reasons for this paradox. One possibility is that opinion on spending isn’t just about money, per se. Opinions on issues outside the strict scope of fiscal politics might affect preferences about spending. Language about “deservedness” has entered public discourse on federal spending. Skocpol and Williamson (2012) show perceptions about recipients’ deservedness have motivated Tea Party activity. Tea Partiers almost universally disapprove of welfare programs, which they think benefit younger and lower-income people, whom they view as lazy moochers. In fact, they find “this moral social geography, rather than any abstract commitment to free-market principles, underlies Tea Party fervor to slash or eliminate categories of public benefits seen as going to unworthy people who are ‘free-loading’ on the public sector” (Skocpol and Williamson 2012, 66). This view is not limited to Tea Party supporters. Many people today view spending as something that benefits “others.” Suzanne Mettler has shown that 94% of people who deny having used a government program actually have (Mettler 2011a,b). In response to her finding that high percentages of people receiving Social Security, unemployment and Medicare benefits claimed not to have used a government program, New York Times columnist Paul Krugman notes, “Presumably, then, voters imagine that pledges to slash government spending means cutting programs for the idle poor, not things they themselves count on” (Krugman 2012). Our state-level estimates of opinion on spending offer a modest view into this type of thinking. Specifically, we can examine the difference between opposition to spending generally and opposition to spending in one’s own district. The more people think spending generally means spending on people “not like them,” the larger the difference we should observe between their opinions on spending generally and their opinions on spending in their own area. If people think spending generally benefits everyone generally, then we should not see a significant difference. For this analysis, we used the two most comparable questions in our data pool: Jobs in District and Jobs v. Deficit. The 5.5 percentage point gap between these questions at the national level is not especially surprising (see F IGURE 1). Things get more interesting when we disaggregate the gap by state. We calculated the difference between state-level opposition to stimulus spending to create jobs and state-level opposition to spending in one’s own district. F IGURE 8 illustrates these differences, which range from a low of roughly 2 percentage points in California to 12 in Utah.27 The party sort on this index is not perfect, but it is close. In general, there is a larger gap between opposition to “spending for me and my neighbors” and “spending for everyone” in red states than blue states. Plotting this gap against the ratio of outlays to taxes, F IGURE 9 shows this relationship does not rely on using Republican vote share as a proxy for opinion on spending. It is not simply that people who live in net taker states dislike spending; compared to those in net giver states, they are more likely to disproportionately dislike spending outside their own district. In other words, they draw a bigger distinction between spending generally and spending for themselves and those who live close to them. Why might this be? It seems unlikely that pure personal selfishness would be more prevalent in some states than others. It seems rather more likely that more people have negative perceptions about the deservingness of “others” or, more broadly, “everyone”, in states with particularly intense histories of racial tension. These include not only southern states with Jim Crow legacies, but also southwestern mountain states experiencing major demographic changes due to immigration. Many of these states receive more money from the federal government than they pay in federal taxes. 27 These results reflect all of the data we have on these two questions. We ran two alternative analyses, first looking only at white opinion, and then using only data from the Kaiser/Harvard/Washington Post poll in which both questions were asked (which had an unusually large sample of over 2,000 respondents). Both variations had some effect on individual states, but did not change the overall pattern between red and blue states. General Spending − Spending in District 0 3 6 9 12 15 UT TX AR MS NM NE OK IN SD WY MO AZ SC ID LA ND AL KS WV PA TN MN KY FL MT AK NV VA GA CO DE NC IA WI WA NH MD ME HI NJ IL VT MI OR RI CT OH MA NY CA 0 3 6 9 12 15 Figure 8: Difference between “Jobs v. Deficit” and “Jobs in District” by State. This graph plots the difference between state-level opposition to spending to create jobs, given a deficit tradeoff, and spending to create jobs in one’s own district. John McCain won the states labeled in red in the 2008 election, while Barack Obama won the states labeled in blue. The dotted vertical line indicates the national average. Gilens (1999) shows that perceptions about deservedness are strongly related to race. People do not oppose welfare because they think it is too expensive or inefficient, he finds, but because they think many recipients do not really need or deserve it. And by recipients, they largely mean African-Americans. People grossly overestimate the percentage of the poor who are black, erroneously associating means-tested welfare programs with African-Americans, even though most recipients are in fact white. Additional negative stereotyping compounds this problem. Gilens shows a strong correlation between the belief that blacks are lazy and opposition to welfare spending. In Opposition to Spending − Opposition to Spending in District NM WV MS 2.5 AK MT Spending/Taxes 2.0 HI ME VA ND AL SD SC LA KY 1.5 AZ WY VT MD 1.0 0.5 IA FL TN OR NC NH WA GA MI CA RI WI NV MAOH CO NY CT IL NJ PA ID OK KS MO IN UT AR TX NE MN DE 2 4 6 8 10 12 Difference Figure 9: Net Spending and District-Centered Views on Spending. This figure plots the difference between “Jobs v. Deficit” and “Jobs in District” against the ratio of federal outlays to federal taxes for each state. The solid black line represents the linear relationship between these two variables, which is positive and statistically significant. contrast, when people view other racial groups as lazy, it does not appear to affect their opinions about welfare spending. Our findings suggest that the effect of negative racial attitudes may extend beyond welfare to spending more broadly. This, however, is a very preliminary hypothesis. What is clear is that state-level opinions on spending issues do not translate straightforwardly into indicators of demand for spending in states, nor into receipt of federal outlays to states. 8 Conclusion and Next Steps Now that we know this discord between outlays and attitudes exists, how might we explain it? We can begin addressing this question by asking whether or not the tension is historically unique. If it turns out to be a contemporary phenomenon, as we suspect it is, then we will need to explain not only why we see it now, but also why we did not see it before. What changed over time to produce this curious relationship between opinion and outlays? While this adds an extra dimension to the puzzle, it can also offer clues to its origins. By tracing the spending paradox over time, we can see not only when but how it emerged.