Economic Voting and Welfare Programs: Evidence from the U.S. States

Transcription

1 Economic Voting and Welfare Programs: Evidence from the U.S. States Matthew M. Singer Department of Political Science University of Connecticut 341 Mansfield Road, U-1024 Storrs, CT Phone: (860) Fax:(860) Forthcoming in European Journal of Political Reesrach Abstract: While scholars have hypothesized that a strong welfare state should reduce voter incentives to base their vote on economic outcomes, evidence for this proposition remains mixed. I test whether differences in welfare protections across American states affect the relationship between economic performance and support for the president s party in 430 state legislative elections over Analyzing the results of over 42,000 contests in which an incumbent was running for reelection, I find that while unemployment insurance programs do not affect the importance of economic performance, the electoral fortunes of presidential copartisans are less strongly tied to the national economy in states with generous anti-poverty programs. Thus by reducing vulnerability to poverty, economic safety-nets lower the salience of the economy and provide electoral cover for politicians during economic slowdowns. Forthcoming in European Journal of Political Research Acknowledgements: The author thanks William Berry and Lyle Scruggs for providing access to their data and also thanks John Aldrich, Herbert Kitschelt, Karen Remmer, Lyle Scruggs, Virginia Gray, and the reviewers for their comments on previous drafts.

2 Economic Voting and Welfare Programs: Evidence from the U.S. States Economic fluctuations have a larger effect on outcomes in some elections than in others, varying in importance across political units and over time (see Anderson 2007). Prominent explanations for this variation identify conditions that allow voters to assign political credit or blame for economic outcomes. 1 Much less attention, however, has been devoted to the possibility that the economy s salience may also vary across economic and social contexts. There can be little debate but that citizens generally desire that incumbents manage the economy effectively. Yet citizens also care about government performance with respect to crime, social policy, foreign affairs, and other issues (Fiorina 1981, Echegaray 2005) and face cognitive limits on the number of issues that they can consider at a given time (Payne 1976, Redlawsk 2004). Thus as other issues compete with the economy for voters attention, the electoral salience of economic management is likely to vary across individuals and time periods as some voters evaluate the incumbent on non-economic grounds. In the extant studies on the economy s salience, the relative accessibility of economic performance compared to other potential bases of performance voting is hypothesized to be a function of three factors (Singer 2011). The economy s salience may be higher when recessions or volatility focus voter attention on the incumbent s ability to specifically manage the crisis (e.g. Bloom and Price 1975). Voter engagement with economic issues is likely to drop, in contrast, when voters attention is diverted by foreign policy or governance problems (e.g. Abramson et al 2007). Finally, individuals tend to pay the most attention to issues that personally affect them (e.g. Lavine et al 2000; Krosnick 1988). Thus voters who are personally vulnerable to economic 1 This literature follows Lau and Sears (1981) at the individual level and Powell and Whitten (1993) at the aggregate level. Numerous examples are reviewed in Anderson (2007). 1

3 dislocations (e.g. due to poverty or insecure employment) may tend to emphasize incumbents abilities to manage the economy while voters who feel economically secure may have fewer incentives to base their vote on economics (e.g. Hellwig 2001, Dorussen and Taylor 2002). However, most economic-voting studies assume that the economy is always an important issue to voters (Wlezien 2005, 556). Usually that assumption is implicit, with no discussion of factors that might determine the economy s relative importance. This assumption is especially transparent in studies that pool observations from different segments of society, years, or countries without considering their comparability. 2 The empirical support for the hypothesis that the economy s impact varies is also mixed, with a recent study concluding that economic voting is widespread, affecting all classes and conditions of voters almost equally (van der Brug et al 2007, 136). Whether the economy s salience varies thus remains an open question. 3 In this paper I focus on an implication of the hypothesis that increases in economic vulnerability cause voters to give greater emphasis to the incumbent s economic record: protections provided by the welfare state should reduce the economy s salience. Pacek and Radcliff (1995) argue that social programs reduce economic uncertainty during downturns by providing benefits (income, insurance, etc.) and smoothing consumption. In countries with generous welfare systems, citizens short-term prospects are less dependent upon changes in the business cycle. Voters protected by a strong welfare state should thus have fewer incentives to base their vote on economic performance because they know they will have access to some minimal guaranteed resources even if their personal economic situation should change during a recession. In regions where the state does not provide those guarantees, in contrast, will place comparatively greater value on the incumbent s economic competence. 2 Even when they consider that there may be differences in how groups hold politicians responsible for outcomes. 3 Though see Singer (2011) 2

4 Three extant studies test the hypothesis that welfare programs reduce the economy s salience, but reach mixed conclusions. Pacek and Radcliff divide West European and North American countries into two groups depending upon the percentage of GDP spent on socialwelfare protection in 1975; those that spent more than 20% of GDP on welfare are considered to have institutionalized welfare states. 4 They find that the impact of changes in per capita income on governing party s support is significantly smaller in institutionalized welfare states. However, they do not control for the institutional environment, specifically the coalition governments that Powell and Whitten (1993) and others show reduce economic voting and tend to correlate with strong welfare states. Subsequent studies that use alternative specifications of welfare efforts and control for institutions find no evidence that welfare protections reduce the economy s impact (van der Brug et al 2007, Palmer and Whitten 2002). Prior research on welfare programs and economic voting have all been conducted in Europe, however, and the strong correlation between welfare state protections and the frequency of coalition governments in that region may explain why it has proven difficult to separate the effect of safety-nets from political institutions. Thus in this note I examine the relationship between welfare protections and economic voting in U.S. state legislative elections over Subnational elections in the United States provide an ideal forum for testing the hypothesis that welfare protections reduce the economy s electoral impact because there is substantial variance in welfare state protections across states and within them over time. These variations exist both in the overall levels of welfare effort and in the type of benefits that states provide. At the same time, there is limited variation in the institutions that may affect economic voting. 4 Specifically, among the OECD countries Denmark, Italy, Germany, Norway, and Sweden are considered to have institutionalized welfare states. 3

5 Our data support the hypothesis that voters protected by the welfare state pay less attention to economic fluctuations when voting: there is a weaker relationship between fluctuations in the national economy and reelection rates for members of the governing party in states where welfare protections are strong. This implies that differences in citizen vulnerabilities to economic downturns affect the salience of economic management, providing additional evidence that the economy s salience varies in meaningful ways. An unexpected finding of our data, however, is that these effects are limited to long term anti-poverty programs; states with generous short-term unemployment insurance programs have the same levels of economic voting as do those with smaller replacement rates. Economic Voting in U.S. States The majority of the work on economic outcomes and subnational elections in the United States has focused on governors. Governors from the president s party are more likely to win reelection when the national economy is strong, and governors also benefit from strong state economies and low fiscal deficits (e.g. Niemi et al 1995, Atkeson and Partin 1995, Hansen 1999, Partin 1995). The strength of the relationship between economic outcomes and support for the governor, however, is contingent upon the latter s ability to control instruments of policy via formal powers and support in the legislature (Rudolph 2003, Lowry, Alt, and Farree 1999). The election fortunes of state legislative candidates differ from those of governors. 5 While governors gain from a strong state economy, state legislative candidates do not after controlling for support for the state governor. Instead, a state legislative candidate's electoral support is determined by shifts in the national economy (Chubb 1988). The electoral fortunes of candidates from the President's party vary directly with the status of the economy, whereas 5 Though many studies of state legislative races do not consider the economy (e.g. Carey et al 2000, Hogan 2005) 4

6 candidates from the opposition party gain, or at least are not punished, during periods of economic turmoil (Berry et al 2000, Jacobson 1990). In fact, Berry et al (2000) find that the economy only affects the support for the president s party but neither hurts nor harms incumbents from the national opposition party. The strength of national-level variables in predicting support for the president s party in state legislative outcomes is somewhat surprising because no state legislature has any real control over the national economy. It is also surprising that voters focus on the party that governs nationally even if that party is in opposition at the state level. The implication is that state legislative races inasmuch reflect economic outcomes, they are partially a referendum on the national president and his economic management. The lack of control that state legislators have over national level economic outcomes makes these races an interesting laboratory for exploring economic voting. Variations across the states in the economy s effect, if documented, are unlikely to reflect differences in state legislative responsibility for the economy or any of the other variables that get the most attention in the comparative economic voting literature. No state legislature has much control over national economic outcomes. Differences in economic voting across states thus instead likely reflect differences in the economy s salience. Specifically, I expect that the economy will have a smaller impact in states where voters have access to large welfare programs. Welfare Policies in the U.S. States In this analysis, I focus on the effect of Aid to Families with Dependent Children (AFDC)/Temporary Assistance for Needy Families (TANF) and Unemployment Insurance. These programs provide basic income guarantees against the personal economic dislocations of unemployment and poverty. The majority of the funds are provided by the federal government 5

7 but states define both eligibility criteria and benefit generosity (See Howard 1999 for a review). In 2006, the American states and Federal government spent approximately $25 billion on TANF and $32 billion on unemployment insurance benefits. 6 The welfare program that has received the most attention from scholars is AFDC/TANF (see Plotnick and Winters 1985, Brown 1995, Volden 2002, and Barrilleaux et al 2002 for three prominent examples). AFDC was primarily designed as the baseline social protection against poverty (Hanson 1983). The specific focus of the program, however, is families with children-- recipients must have dependents living at home. 7 States decide how much to contribute to the program, what the monthly limit on benefits should be, and what criteria should be used for families to qualify. There is substantial variation across states in the generosity of this antipoverty program. In the period of time I study here, monthly benefits for a family with three children and no income range from $125 to over $1200 (all figures in 1995 dollars). Real benefits decreased in almost every state during the period but these protections declined faster in some states than others (Berry et al 2003). Ohio, for example, provided the most generous protections in 1960 but had only the 27 th most generous welfare protections in Vermont, on the other hand, increased its real benefits over the same period and improved from having the 33 rd most generous protection regime to the third most generous in Generous AFDC benefits tend to come in states that have small minority populations, liberal legislatures, organized poor sectors of society, and those that have neighboring states that are also generous, though there are debates about these findings (see Rodgers 2005 for a review). 6 Calculated from and 7 In 1995, the program was changed to limit the amount of time that families could receive benefits and renamed TANF. 6

8 Unemployment insurance is designed to smooth short-term dislocations by replacing a percentage of the worker s previous income for a fixed period of time. Like AFDC, most of the administration of this program is at the discretion of the states. The federal government sets a minimal payroll tax but states can set a higher tax rate at their discretion. While there is little variance in the amount of time that individuals are allowed to receive benefits (nearly all states have a 26 week limit), there is substantial variation in the size of the benefits that states allow. Most states replace 30-40% of workers wages when they become unemployed but replacement rates vary from a low of 24% to a high of nearly 68% (Scruggs 2007). Again there are substantial differences across states and within them over time; Washington raised its ranking from 45 th to 7 th over the period while New Jersey fell from 9 th to 43 rd over the same period. Unemployment benefits are strongly shaped by state partisanship, with heavily Democratic states giving higher benefits, and also increases with higher levels of unemployment and the state s income level while decreasing with the presence of large minority populations (Scruggs 2007, Howard 1999). Together, these two programs create two forms of safety-nets that citizens can rely on during an economic downturn. Unemployment insurance provides short-term relief from financial vagarities while AFDC provides shelter from absolute poverty if the dislocation proves to be longer term. Both programs have been shown to lower personal bankruptcy rates (Fisher 2005), with the two programs largely acting as substitutes for each other. There is also some evidence that states with generous unemployment insurance programs spend less on poverty 7

9 relief programs, which implies that the states see these programs as complements to each other. 8 However, there is substantial variation across states in the extent of the programs: some states are generous for both sets of programs (e.g. Oregon, Wisconsin, and Minnesota), some are relatively non-generous for both programs (e.g. Tennessee, Alabama, and Missouri), and some states are generous with one program and not the other (e.g. Pennsylvania is only generous with unemployment insurance while California and Vermont are only generous with AFDC). Thus Howard (1999) describes the United States as having 50 different welfare states, though given the variation within many states over time the number of different welfare regimes available for study in the U.S. context is even larger than that. At the same time, the partisan and institutional variance across states is less varied than in the European context; while states differ in their partisan coloring most states have at most two effective parties and the coalition governments which have been shown to strongly affect economic voting patterns in Europe thus do not exist. 9 I take advantage of this tremendous variation in our main independent variable while holding constant other confounding variables to test whether states with generous welfare programs reduce the effect of national economic outcomes on state legislative races. Data To test if welfare expenditures intervene between the economy and the vote as expected, I model whether an incumbent state legislator is reelected given the political and economic environment and his previous levels of electoral support. The basic model specification is drawn 8 Specifically, I estimate an error correction model of AFDC benefits over time using the data described in the next section and find that there is a long-run negative relationship between unemployment insurance generosity and AFDC benefits. Thus generous unemployment insurance is associated with smaller expenditures on AFDC. 9 In the web appendix, I control for divided government at the state and national level because this may make it more difficult for voters to assign responsibility to a specific actor (Powell and Whitten 1993). 8

10 from Berry et al (2000)'s study of state legislative elections. 10 As the measure of national economic conditions, I use the percentage change per capita GDP in the year of the election. 11 Increases in the growth rate should increase support for the incumbent s copartisans. Our measure of welfare expenditures combines data from the benefit levels provided by the two main safety-nets that vary across states. I measure AFDC generosity using the maximum monthly benefits available to a family with three children and no income. The data are compiled by Berry et al (2003) and are in real 1995 dollars. 12 I operationalize unemployment insurance generosity using Scruggs (2007) measure of maximum replacement rates, 13 which averages the replacement rate for the two observations per calendar year. I combine these in a single indicator by standardizing the two measures of generosity and taking their average. 14 In the analysis I also 10 I mine grateful to Berry and his coauthors and to Scruggs for making the data available-all errors remain mine. 11 Using changes in real disposable income as the independent variable yield similar results. 12 The data presented below do not adjust for costs of living. I have also run the analysis using the ratio of AFDC limits to the average cost of living in a state, and the substantive results do not change from those reported below. The two measures are very highly correlated, meaning that the most generous states are generous even when you control for average income in those states. 13 The replacement rate is taken as the ratio of the maximum weekly benefit to the average weekly wage in any given state. Estimates of benefit generosity are based on the maximum benefits provided in the Department of Labor s Significant Provisions of State Unemployment Insurance Laws. The weekly maxima are for single individuals, i.e., they exclude supplements for dependents. Wage data is based on UI tax data from the Bureau of Labor Statistics Quarterly Census of Employment and Wages, and, prior to 1976, wage estimates based on state per capita income. (Scruggs 2007, 8). 14 To facilitate the interpretation of these interaction terms, I rescale the welfare measure so that the least generous state takes the value of 0. 9

11 use each one separately to test if both have the same political effects. 15 Testing the hypothesis that welfare expenditures have an intervening effect on economic voting requires the use of a multiplicative interaction term. I interact the welfare generosity limit with the measure of economic conditions. If welfare expenditures cushion the economic blow for voters as expected, the interaction term will push the effect of the economy towards 0 as welfare benefits increase. The economic and welfare variables are the focus of our analysis, but I include a larger set of controls previous studies suggest are important determinants of incumbent political support. Because the impact of these variables is tangential to our analysis, I only list them here and refer the readers to Appendix 1 for more information on their coding. At the individual incumbent level, I control for the incumbent's previous electoral margin and whether or not the race was even contested previously to capture individual-specific factors that should shape their willingness to draw a high quality challenger. I also control for whether the incumbent is running after state-level redistricting controlled by his own party or by the opposition, with the expectation that redistricting efforts will help members of the party that controlled the redistricting. I also control for the incumbent s partisanship to ensure that there are no long-term differences in how voters evaluate Republicans and Democrats. 16 At the state level, I control for 15 Because I believe that the economy s salience should be affected by the overall size of the welfare state and not the origin of its funding, I do not control for the portion of welfare benefits that come from state sources and not the federal government. Adding this control does not affect the substantive results. 16 In alternative specifications not reported here, I have tested for whether welfare benefits affect Republicans and Democrats differently and find no significant interactive effects between partisanship and welfare benefits. Nor do I find that welfare policies differentially affect the relationship between the economy and reelection chances across the two parties when their party controls the national executive. 10

12 the degree of legislative professionalism and staff that allows candidates to overcome challengers or a weak economy by casework (King 1991, Berry et al 2000), the potential coattails effect of presidential, gubernatorial, and senatorial elections that coincide with the state legislative elections, the electoral system in which the election is contested (SMD, MMD with posts, and pure MMD), and if the legislator only served a two-year term. 17 Finally, I control for the candidate s partisanship to control for any differences in partisan advantages over this period. 18 I examine this model of incumbent reelection in a dataset of state legislature results compiled by Berry et al (2000). Their original dataset contains the results of 42,820 contested partisan elections for the state legislature where an incumbent sought reelection from When merged with Berry et al's (2003) data on AFDC limits in the 48 contiguous states from , I are left with 42,096 specific incumbents in 430 legislative elections over 20 years for which I also know the state's welfare protection limits. Scrugg s dataset on unemployment insurance covers this entire period. Because our model of incumbent reelection combines variables from three separate units 17 In addition to the controls listed here, Berry et al (2000) argue that the impact of economic conditions on state legislative outcomes is conditional upon legislative professionalism. Specifically, they argue that more professional legislatures provide the staff resources and expertise needed for legislators to actively cultivate a personal vote and insulate themselves from external political and economic forces (see also Hibbing 1999, Moncrief and Thompson 1992, Rosenthal 1998, Squire 1992). However, in the results reported in the web appendix, this interaction term is neither in the expected negative direction nor significantly different from 0 at conventional levels. This may reflect our use of a hierarchical model- hierarchical models tend to have larger standard errors than models which use other techniques to correct for clustering by states or years. 18 In additional specifications not reported here, I have also tested whether the economy s effect is different in years when there is no presidential election and find that while its effect may be slightly smaller, it is not significantly so and that the effect of welfare policies is the same in presidential election years as in other years. 11

13 (individual factors, state factors, and national factors), I use a three tiered hierarchical model whose design corrects for the clustering of errors within states or the national economy that would potentially result in artificially deflated standard errors and overly optimistic test statistics (Steenbergen and Jones 2002). There is substantial variation in reelection probabilities not only across individual legislators but also between states and over time (the three tiers modeled below). 19 Our dependent variable (whether or not the incumbent is reelected) is a dichotomous variable and so I model it using a logit equation. Equations 1-3 below outline the individuallevel (first tier) equation of whether incumbent i seeking reelection in state j at time t will be reelected as a function of the specific individual-level predictors outlined above. Prob (Incumbent Reelected ijt = 1 ) π = ϕ ijt log[ ϕ /(1 ϕ )] = Y ijt ijt * ijt Y * ijt = π π 0jt 3 jt ( Opposition Controlled Redistricting) + π 5it + π ( Contested Previously) + π 1jt + π ( Republican) 2jt ( Margin of Victory) + 4 jt ( Own Party Controled Redistricting) [1] The intercept of equation 1 (designed as π 0 ) is the average probability that an incumbent from state j at time t will be reelected, with deviations from that mean explained by the individual specific characteristics captured by the controls. The potential effect of state-level variables is to change the average reelection rate in a state, holding the characteristics of individual legislators constant. Hence the state-level (second tier) model, equation 2 below, models the average conditional reelection rate for state j at time t (estimated in equation 3) as a 19 The largest variance component is for the first tier; candidate-level differences account for 46% of the total variation in reelection rates. State characteristics explain 32% of the variance in reelection probabilities while national-level factors explain the remaining 22%. 12

14 function of the state-level predictors outlined above. π 0jt = β + β ( Welfare) + β ( Governor Election Year) + β ( Governor Vote) 0t 4t 7t 1t 2t + β ( Senate Election Year) + β ( Senate Vote) + β ( President Vote) + β ( MMD with Posts) + β ( SMD) + β ( 2 Year Term) + β 8t 5t 9t 6t 3t 10t ( Professionalism) + δ 0 [2] I incorporate national conditions into our explanation of incumbent reelection in a similar fashion by using them to model differences in reelection probabilities that remain after controlling for differences in individual traits and the institutions that exist in a state, differences driven by changes in national-level (third tier) factors that change over time. Specifically, the third-tier equation (equation 3 below) models the intercept from the state-level model as a function of national economic conditions, whether there is a presidential election occurring, and whether there is divided government at the national level or not. β 0 = γ 0 + γ 1(Economy) + γ 2 (Presidential Election Year) + γ 3 (Divided National Government) + μ 0 [3] These three equations, estimated simultaneously, make up the basic structure of our hierarchical model of incumbent reelection and allow us to model the importance of national economic conditions for members of state legislatures. Our main hypotheses, however, regard the conditional impact of the economy depending upon whether or not the incumbent is a presidential copartisan or in opposition and the level of welfare state protections in state j at time t. Because these hypotheses combine the effect of variables across levels of analysis, I estimate an additional equation to generate the interaction term of interest by modeling the effect of state welfare levels on reelection as being a function of the national economy. The resulting equation generates an estimate for the interactive effect of the economy and welfare levels (γ 10 ) that can then be used to test for welfare benefits conditioning role on economic voting. β = ( + μ [4] 1 γ 10 + γ 11 Economy) 1 Our expectation is that a strong national economy will propel members of the president s 13

15 party toward reelection, i.e. that γ 1 will be positive for presidential copartisans. If welfare benefits reduce the economy s effect, then γ 11 will be negatively signed. Results Table 1 (below) estimates the hierarchical logit model of incumbent reelection using the variables described above. Because I expect the economy s effect to differ between members of the president s party and members of the national opposition, I estimate the models of reelection for the two groups separately. (Table 1 about here) Looking first at the controls, the incumbent s previous electoral experience proved to be a strong predictor of his future vulnerability. Incumbents who faced major party challengers in the previous election were more likely to be defeated in the current one but increases in the previous margin of victory are associated with increased probabilities of winning reelection. The statelevel estimates also support Berry et al (2000) s proposition that increases in the amount of professional staff and resources available to incumbent state-legislators provides an incumbency advantage that increases the probability of being retained in office. These effects are true for all incumbents, independent of their relationship to the president. Taken together, these two results reaffirm the importance of incumbency advantages at the state level. Finally, the type of electoral system used in the state does has a significant effect on reelection rates, though their effect is largely limited to members of the opposition party. The results in Table 1 present some interesting null findings as well. I find that most redistricting does not have a significant effect on incumbent reelection, which may be evidence that redistricting activities in this period neither protected incumbents, independent of their party 14

16 positions, not had a consistent partisan logic 20, or it could be evidence that these effects take longer than a single election to manifest themselves. Moreover, there is very little evidence of coattail effects, though support for a party s gubernatorial candidate seems to have some effect on other subnational races (Hogan 2005). Interestingly, I find more null relationships than Berry et al (2000) do with the same data; the important difference is that our multi-level model even further corrects for clustering within contexts that would otherwise reduce the standard errors. Of the variables that are significantly associated with incumbent reelection, the incumbent s previous margin of victory has the largest impact-a swing of one standard deviation below the mean value for the sample to one standard deviation above the mean increases the probability of being reelected from 0.74 to 0.98 when all other values are at their mean. A similar change in a candidate s party s gubernatorial candidate increases the probability of being reelected from 0.84 to Finally, a similar increase in legislative professionalism by two standard deviations minimally changes the probability of being reelected from 0.91 to Our primary interest, however, is on the effect of the economy. On average, incumbents from the president s party run strong when the national economy is strong and are weaker when then national economy is weak. However, economic growth does not have a significant effect on reelection chances for incumbents who belong to the national opposition party or who are independent. This is also what Berry et al (2000) had previously found looking at the same data Consistent with Abramowitz et al s (2006) findings with respect to the U.S. House of Representatives. 21 This may consistent with the argument by Hibbing and Alford (1981) and Stein (1990) that incumbents who belong to the president's party are at least responsible for the state of economy, but nonincumbents (e.g. challengers and open-seat candidates) of the same party are not. Thus they do not benefit from a strong economy nor do they pay the full price for a weak one. 15

17 However, the effect of economic performance is not constant across states. Consistent with the main theoretical argument, strong welfare protections reduce the impact of economic factors. Given that the economy does not affect incumbents from the national opposition party, it is not surprising that welfare states cannot make the economy s effect even further insignificant. Yet for members of the president s party, the negative interactive relationship between growth and welfare benefits implies that the marginal impact of economic growth becomes smaller in states where the state provides larger guarantees to the poor and unemployed. Thus economic growth rates are significantly associated with reelection in states with very small welfare states but become less important in regions with generous government programs. Interestingly, in other model specifications not reported here I do not find welfare benefits to have a direct effect on reelection probabilities. Large welfare states do not enhance incumbent reelection or undermine them unconditionally. Nor do I find in other analyses that welfare protections help Democratic incumbents differently than Republican ones; there is no interactive relationship between a member s partisanship, the size of the welfare state, and their chance of being reelected. Welfare polices only electoral affect in this sample is to shape the marginal effect of swings in the national economy. The results in Table 1 thus support the hypothesis that welfare benefits affect the economy s salience. However, our combined measure of welfare generosity leaves open the question of whether both anti-poverty and unemployment programs reduce economic voting. In Table 2, I estimate the models using only one of the two measures as our primary indicator. The results show that AFDC benefits significantly reduce the importance of economic performance. As a result, in states where real AFDC benefits are about $775 a month (which includes 25% of the elections in this sample), the effect of changes in the GDP growth rate is so small that it is 16

18 insignificantly different from 0. (Table 2 about here) In contrast, unemployment benefits do not have a significant effect on economic voting; while their effect is in the expected negative direction it is both substantively small and statistically indistinguishable from 0. Thus while unemployment insurance reduce short-term insecurity and compensates from a portion of income loss, these programs do not seem to be reducing the reactions of citizens to economic swings. This difference in how welfare programs affects voters in unanticipated by the extant literature, which does not differentiate between types of programs when considering their potential electoral effects. This difference merits further analysis in other contexts, but the difference may be explained by two possible factors. First, the scope of short-term unemployment insurance programs may be less visible to voters than the overall size of anti-poverty programs; AFDC generally has more members and spends more than does unemployment insurance, especially in non-recession years. 22 Second, the two programs have different emphases; AFDC provides a floor against absolute poverty in the medium term while unemployment insurance is only designed to be a short-term coping strategy. AFDC benefits are also available to individuals who are under employed or otherwise affected by a recession while unemployment insurance only helps the newly unemployed. In any case, the data in Table 2 imply that it is investments in a strong safety-net against poverty that reduce the salience of economic fluctuations for voters. To further confirm the robustness of these results, I have rerun the models in Table 2 with controls for other aspects of the state s economic conditions (e.g. the average wage and income 22 E.g. and for data on enrollment. 17

19 level in a state) and institutional climate (divided government at the national and state level, legislative professionalism) that may also affect levels of economic voting. 23 I do not find any evidence that any of these factors condition economic voting. Moreover, these additional controls do not change the estimated effect of AFDC benefit levels on state-level economic voting. Thus variations in economic voting at the state level do not reflect differences in attributed responsibility for economic outcomes nor can I easily ascribe these effects to other institutional or economic differences across states. Taken together, the results in Table 1 and Table 2 imply that anti-poverty welfare programs significantly shape the role that economic factors play in elections involving an incumbent who belongs to the president s party. One way of considering the effect of antipoverty protections is to estimate reelection probabilities for an average legislator who belongs to the president s party during a severe recession. During the 1982 recession, for example, the predicted probability that a Republican state legislator was reelected is 0.77 in states where AFDC benefits for a family with 3 children equaled $ If the welfare benefits were about $500 (approximately the median value for the sample), the reelection probability of the same average legislator rise to 0.81 despite the recession and would equal about 0.84 if AFDC benefits equaled $775 (the 75 th percentile). Thus a strong welfare state reduces the electoral costs of a large downturn by cushioning the citizenry from its effect. More broadly, inclusive anti-poverty programs reduce the marginal impact of economic swings. The predicted difference between reelection rates when economic growth is at its maximum (the 1984 recovery) and its minimum (the 1982 recession) is a decrease of about See the web appendix 24 That they remain even that high speaks again to the power of incumbency. 18

20 in a state with where the AFDC benefit is $100 for a family with 3 children (see figure 1). If, however, that state had enacted a welfare program that placed it in the 75 th percentile (e.g. when AFDC benefits equal $775) then the same fall in economic growth rates would only decrease the predicted reelection probabilities by about 0.07, a reduction in the economy s impact by about 63%. The implication is that citizens in states with encompassing welfare protections have fewer incentives to punish the incumbent party for a recession or to reward it for a strong economy. (Figure 1 about here) If I focus on elections held at less extreme conditions, the effect of welfare on economicbased voting is similar even if the overall effect of the economy is smaller. The difference in reelection rates in a state that held elections in 1981 (when per capita GDP grew by 1.52%) and in 1981 (when the economy shrank by 1.19%) is a decrease of 0.07 in a state with AFDC benefits at $100. However, the same difference in economic performance only decreases incumbent reelection by 0.25 in a state with AFDC benefits at the $775 level. Again, a change in welfare policies of this magnitude reduces the economy s relative magnitude by about 64%. Discussion Although the puzzle of why the economy s effect varies across electoral contexts has been extensively researched, the evidence for the hypothesis that voters protected by welfare safety-nets have fewer incentives to focus on short-term economic performance in evaluating potential choices remains mixed.. American state elections provide a fresh set of observations with which to revisit this hypothesis while simultaneously holding the democratic institutions and levels of party-system fragmentation constant. Thus the American state data provide a most-likely case for observing welfare programs reducing the economy s electoral impact even though this hypothesis has never been tested in the American context. 19

21 The data analyzed here are consistent with the hypothesis that anti-poverty programs reduce the linkage between economic outcomes and electoral ones. During economic downturns, members of the president s party generally become more likely to suffer electoral setbacks but the electoral retribution tends to be smaller in states that have strong anti-poverty programs. Thus at least some welfare programs provide protection for both vulnerable citizens and vulnerable incumbents. Interestingly, the programs which reduce attention to the economy are those which provide a minimum safety net and which are long-term in nature, while programs whose benefits vary across recipients depending upon their previous wage and employment situation and which are short-term do not seem to alleviate voter anxiety during an economic downturn. Different economic effects lead to different political effects. For students of economic voting who have largely focused on institutional factors, these findings will hopefully serve as a reminder that non-institutional variables also affect voters calculations. Duch and Stevenson (2008) and Hellwig and Samuels (2007) have recently provided evidence that economic factors like the degree of volatility in the economy and exposure to globalization affect the inference that citizens can draw about the incumbent s competence from economic performance. However, the welfare state has an alternative mode of affecting economic voting: namely, by changing the salience of economic considerations. Voters can potentially evaluate incumbents on a variety of issues, especially as party competition shifts towards post-materialist issues in some polities or reorients itself toward questions of governance and security in others. If government policies can reduce the connection between short-term economic fluctuations and citizen welfare, this opens space for voters to focus on other concerns. More generally, if economic voting scholars accept that welfare states can reduce the economy s salience by minimizing exposure to economic slowdowns, it may also be worth 20

22 considering the relevance off other factors that reduce citizen vulnerability during a downturn (secure employment, access to savings or social networks, etc).. Some preliminary hypotheses on this topic have been provided by Hellwig (2001) and Dorrussen and Taylor (2002), mainly focusing on the economic sector in which people are employed, though van der Brug et al (2007) dispute the latter set of findings. Further work that directly attempts to measure the economy s perceived salience and which relies on more direct measures of citizen vulnerability (e.g. poverty, previous access to safety nets, perceived and real insecurity in their present employment) may further address this hypothesis. At the very least, however, the data from U.S. states remind us that the salience of the economy may vary systematically across political contexts. Only as voters become systematically vulnerable to economic dislocations does it s the economy, stupid really become true. 21

23 Appendix 1: Variables and Coding Unless otherwise noted, the data source for these variables is Barry (2000). Winnext: Did the incumbent win (1) or lose (0) Opposition Candidate: Is the incumbent a member of the president's party (0) or in opposition (1) Contested Previously: Did the incumbent face a major party challenger in the previous election (1) or not (0) Previous Margin: Continuous measure of margin of victory, time t-1 Own Party Redistricting: Did the incumbent's party control any redistricting in the state that occurred subsequent to the previous election (1) or not (0) Opposition Party Redistricting: Did the other party control any redistricting in the state that occurred subsequent to the previous election (1) or not (0) 2 Year Term: Do state legislators serve two year terms (1) or not (0) SMD Election: Are state legislators elected from SMD's (1) or not (0) MMD with Posts Election: Are state legislators elected from MMD's with posts (1) or not (0) Presidential Vote : Percentage of the vote received by the presidential candidate for the incumbent's party in his state if the current state legislative race is concurrent with a presidential election; 0 otherwise. Gubernatorial Vote: Percentage of the vote received by the gubernatorial candidate for the incumbent's party in his state if the current state legislative race is concurrent with a gubernatorial election; 0 otherwise. Senatorial Vote: Percentage of the vote received by the senatorial candidate for the incumbent's party in his state if the current state legislative race is concurrent with a senate election; 0 22

24 otherwise. Legislative Professionalism: The legislative operating budget per member in real dollars AFDC Benefits: Monthly benefits limit for a family with three children and no income in 1995 dollars. Recentered so that the lowest benefit level received a score of 0. From Berry et al (2003). Unemployment Insurance: the ratio of the maximum weekly benefit amount to the average wage in the state at time t, calculated as a percentage replacement rate. Taken from Scruggs (2007). Gubernatorial Election Year: Was the state legislative race concurrent with a gubernatorial election (1) or not (0) Senatorial Election Year: Was the state legislative race concurrent with a senate election (1) or not (0) State Divided Government: Was the governor for a different party than the majority party in the state legislature (1) or not (0) National Divided Government: Was the president for a different party than the majority party in the national legislature (1) or not (0) Income: real per capita income in state in 1995 dollars. From Berry et al (2003) Average Wages: Average retail wages in the state, in 1995 dollars. From Berry et al (2003). Presidential Election Year: Was the state legislative race concurrent with a presidential election (1) or not (0) National Economic Conditions: Per capita GDP growth, in a percent. From World Development Indicators. Republican: Incumbent belonged to the Republican Party (1) or not. 23

25 Works Cited Abramowitz, Alan I., Brad Alexander, and Matthew Gunning Incumbency, Redistricting, and the Decline of Competition in U.S. House Elections. Journal of Politics 68 (1): Abramson, P. R., Aldrich, J. H., Rickershauser, J. & Rohde, D. W. (2007). Fear in the voting booth: The 2004 presidential election. Political Behavior, 29 (June), Anderson, Christopher J The End of Economic Voting? Contingency Dilemmas and the Limits of Democratic Accountability. Annual Review of Political Science 10: Atkeson, Lonna Rae and Randall W. Partin Economic and Referendum Voting: A Comparison of Gubernatorial and Senatorial Elections. The American Political Science Review 89 (Mar): Barrilleaux, Charles, Thomas Holbrook, and Laura Langer Electoral Competition, Legislative balance, and American State Welfare policy. American Journal of Political Science 46 (April): Berry, William D, Michael B. Berkman, and Stuart Schneiderman Legislative professionalism and incumbent reelection: The development of institutional boundaries. American Political Science Review 94 (December): Berry, William D, Richard C. Fording, Russell L. Hanson Reassessing the race to the bottom in state welfare policy. The Journal of Politics 65 (May): Bloom, H. S., Price, H. D. (1975). Voter Response to Short-Run Economic Conditions: The Asymmetric Effect of Prosperity and Recession. American Political Science Review, 69 (December), Brown, Robert D Party-cleavages and welfare effort in the American states. The American Political Science Review 89 (March):

26 Carey, John M., Richard G. Niemi and Lynda W. Powell Incumbency and the Probability of Reelection in State Legislative Elections. Journal of Politics 62 (3): Chubb, John E Institutions, the Economy and the Dynamics of State Elections. American Political Science Review 82 (March): Dorussen, Han and Michaell Taylor Economic Voting. New York: Routledge. Echegaray, Fabian Economic Crises and Electoral Responses in Latin America. New York: University Press of America. Fisher, Jonathan D The effect of unemployment benefits, welfare benefits, and other income on personal bankruptcy. Contemporary Economic Policy 23 (4): Fiorina, Morris Retrospective Voting in American National Elections. New Haven: Yale University Press. Hansen, Susan B Life Is Not Fair": Governors' Job Performance Ratings and State Economies. Political Research Quarterly 52 (1): Hellwig, Timothy and David Samuels Voting in Open Economies: The Electoral Consequences of Globalization. Comparative Political Studies 40 (3): Hellwig, Timothy Interdependence, Government Constraints, and Economic Voting. The Journal of Politics 63 (4): Hibbing, John R Legislative Careers: Why and How I Should Study Them. Legislative Studies Quarterly 24 (May): Hibbing, John R. and John R. Alford The Electoral Impact of Economic Conditions: Who is Held Responsible? American Journal of Political Science 25 (Aug): Hogan, Robert E Gubernatorial Coattail Effects in State Legislative Elections. Political Research Quarterly 58 (4):