Individual Unemployment, Layoffs, and Voting in U.S. Presidential Elections

This paper uses new methods and unique new data to both demonstrate that individual-level unemployment affects individual voting decisions and that county-level layoffs had a substantial effect on the outcome of the 2008 presidential election. First, panel data from a series of elections is used to show that changes in non-subjective measures of economic wellbeing strongly predict an individual’s vote decision. Second, results based on a comprehensive dataset of all layoff announcements from January 2008 through January 2009 shows that local layoffs had a large effect both on county-level election returns and the national outcome, suggesting that previous research findings of a lesser role for local conditions may reflect measurement issues. The timing of layoff announcements is used to provide evidence that the relationship between local economic conditions and local voting behavior is a causal one. The paper concludes by considering evidence from county-level employment and income changes which suggests that, at the local level, voters particularly respond to employment concerns. * Contact information: [email protected]; 1 LMU Drive, University Hall 4229, Los Angeles, CA 90045 The large literature on economic voting suggests that the national economic downturn was a more likely cause of Barack Obama’s success in the 2008 election than individual-level or local economic concerns. While some research has found evidence that individual-level economic concerns affect voting decisions in U.S. presidential elections (e.g., Markus 1988), even those papers argue that national, or sociotropic, concerns have a substantially larger effect on election outcomes. Some scholars remain unconvinced that personal economic conditions even have a significant effect on the individual vote decision. Moreover, given the literature’s focus on individualand national-level economic concerns, previous work has devoted little attention at all to local economic conditions, with the existing literature concluding that local conditions are much less important than national ones in determining voting decisions (Eisenberg and Ketcham 2004). 3 However, previous research has not adequately addressed the issue of measurement error in estimates of local economic conditions or emphasized the importance of local-level changes in unemployment, which may be more visible to voters than changes in local-level income. In this paper, I use new methods with existing data and analyze a unique new dataset to provide evidence that personal economic circumstances indeed do affect voting decisions and that local economic conditions can have a substantial and significiant impact on national election outcomes. 1 Seminal papers include Kramer (1971), Fair (1978), Kinder and Kiewiet (1979), Kramer (1983), and Markus (1988). In the 1990s, over 200 papers were published on economic voting (Lewis-Beck and Paldam 2000). See Lewis-Beck and Stegmaier (2000) for a review. 2 For example, Ansolabehere et al. (2006) write that “The literature on economic voting notes that voters’ subjective evaluations of the overall state of the economy are correlated with vote choice, whereas personal economic experiences are not.” 3 While previous research has had difficulty in establishing a clear link between local economic conditions and election outcomes, local events have been shown to influence elections in other domains. For example, Grose and Oppenheimer (2007) show that local-level Iraq casualties drive 2006 congressional vote results. 4 Ansolabehere, Rodden, and Snyder (2008) demonstrate that measurement error has led to longstanding misconceptions about the stability of issue preferences. Correcting for measurement error by averaging across multiple measures of issue preferences leads to much stronger predictive power of issue preferences for presidential vote choice. I am currently attempting to use multiple measures of the local economy to determine if correcting for measurement error in this way changes the predictive power of local economic conditions (Healy 2008). While the finding that local economic conditions can be a critical driver of voter behavior is new to the academic literature, campaign strategists and news reports have long operated on the assumption that local economic conditions affect elections. Many examples from the recent presidential campaign support this idea. John McCain pulled his campaign out of Michigan after his poll numbers dipped following bad economic news for the locally-based automobile manufacturers (Nagourney and Zeleny, October 4, 2008). Indiana, one of the states hardest-hit by the economic downturn, experienced a swing of twenty percentage points in the Democrats’ favor from the previous election, compared to a roughly ten percentage point swing for the Democrats in the national vote. These examples are consistent with the idea that local economic conditions significantly affect voting decisions and election outcomes. The results presented in this paper show that this sort of anecdotal evidence is actually not in conflict with the large literature suggesting the primacy of national economic conditions in influencing voting. While voters may use national economic growth to evaluate the incumbent’s performance, individual-level and local economic conditions appear to enter the decision-making calculus more through concerns about employment. To the extent that previous research has generally focused on voters’ subjective evaluations of their current economic state (the better off/worse off measure), it may have focused only indirectly on the measure of individual wellbeing that primarily determines the vote choice. Previous work considering the impact of local economic conditions on voting behavior has generally looked at income growth, generally finding a zero effect. This zero effect is not surprising when the large amount of measurement error in local income is considered. The analysis in this paper suggests that these earlier results are only half wrong, however. Local conditions do significantly affect voting when properly measured, but the effect operates primarily through employment as opposed to income. Before estimating the effect of local unemployment, I use panel data from five elections to estimate the effect that entering unemployment has on the individual voting decision. By looking at individual-level changes in economic conditions, the analysis provides clearer evidence than can be gained from cross-sections, where individual-level changes cannot be observed. The results also do not rely upon subjective measures of wellbeing. The results are of a similar size as those that Markus (1988) found using subjective evaluations to measure a voter’s economic wellbeing, where he employed an instrumental variables strategy to find unbiased estimates. As a result, the evidence presented here corroborates his estimates of the effect of pocketbook motivations on the voting decision. Since relatively few voters transition into unemployment for any given election, the fact that transitions to unemployment affect the individual vote choice does not lead to the conclusion that personal or local economic conditions and concerns have affected election outcomes. To demonstrate that link is a difficult empirical problem. Local economic conditions suffer from greater measurement error than national ones, so that a regression of vote totals on local and national economic variables is likely to lead to a downwards bias in the coefficient on local conditions. Moreover, changes in unemployment rates are notoriously difficult to interpret, since some symptoms of economic troubles such as individuals giving up in the search for work, can actually lead to lower unemployment rates. Autor and Duggan (2003) find, for example, that important reductions in labor force employment due to increases in the disability rolls were entirely missed by the reported unemployment rate. Here, I obtain estimates of the effect of local unemployment shocks for the 2008 presidential election by considering all layoff announcements from January 2008 to January 5 Looking at changes in variables such as income growth and unemployment will exacerbate the measurement error problem, as well. 2009. The timing of each announcement is precisely dated so that it is possible to compare the effect of those changes in local economic conditions that occur soon before the election to those that occur soon after. Through this comparison, it is possible to identify the causal effect that changes in the local economic environment have on elections. The results suggest that changes in the economic environment in a voter’s county have a large effect on voters’ decisions. For the 2008 election, at least, it appears that local economic conditions may have had as large an effect on the outcome as the national economy is generally presumed to have. The regression estimates suggest that for every vote the incumbent loses among laid off workers, it may lose an additional five due to deteriorating local economic conditions. In Section 2, I introduce a simple model of voting behavior, a strategy for estimating that model, and the data used for estimation. Section 3 contains the results for the individual-level analysis. In Section 4, I present the results obtained by considering the relationship between layoff announcements and county-level election returns. Section 5 concludes. Model, Empirical Strategy, and Data I first consider the model that will be used to estimate the effect that individual transitions to unemployment have on the individual vote choice, as well as the strategy and data to be utilized to estimate the model. Then, I describe an aggregation of that model to the county level that I will employ to estimate the effect that layoff announcements have on vote choice. Individual-level analysis To analyze the effect of individual-level unemployment on voting behavior in American presidential elections, I utilize data from the five available panel waves of the American National Election Studies (ANES). The available panels go from 1956-60, 1972-76, 1990-92, 1992-96, and 2000-04. For each of these panels, the dependent variable under consideration is the respondent’s change in vote choice from the previous election. The independent variable under consideration is the respondent’s change in employment status. To estimate the vote choice, I consider a simple model of voting behavior. Define Voteit to be the vote choice by voter i in the election at time t, to be 0 if the voter selected a candidate other than the incumbent party, 1⁄2 if they did not vote, and 1 if the voter selected the incumbent party. The results are unaffected if we restrict the analysis to vote choices of one of the two major parties. Define an individual fixed effect to represent all the time-invariant characteristics that affect the voter’s preferences for the incumbent party, and a time fixed effect i α t γ to account for factors that determine an incumbent’s national support, such as per-capita income growth. Define Economyit to be a measure of personal economic conditions for the voter. Then the equation for the vote is modeled as: (1) 1 it i t it it it Vote Economy Controls u α γ β φ = + + + + In the above equation, Controlsit is a vector of control variables, such as an interaction between a voter’s race and a dummy for the Democrats being the incumbent party, that have been used in previous research. To obtain unbiased estimates of the effect of the economy on voting requires both a method for dealing with omitted variables that may be contained in and an appropriate measure of personal economic conditions. Using panel data makes it possible to control for an individual’s previous vote in the regression, which effectively accounts for any fixed individual i α characteristics that we cannot observe but that influence the voter’s preferences for the incumbent party. Previous work on economic voting (e.g., Markus 1988) has generally relied upon survey respondents’ answer to the question about the state of their current financial situation. I employ a different measure of a respondent’s personal economic situation that is not subjective and thus less susceptible to bias. I define to be 1 if a respondent was employed two years before the election and unemployed at the time of the election and 0 if the respondent was employed two years and not unemployed at the time of the election or was not unemployed at either time. Results are substantively the same if the control group is restricted to those respondents who were employed at the time of the previous election. it Economy