Political Choices in Hard Times: Voting in the 2010 U.S. House Elections

Abstract

This paper assesses the explanatory power of rival models of voting behavior in the 2010 elections for the U.S. House of Representatives. Multivariate analyses of data gathered in a 2008–10 national panel survey indicate that a combination of national-level valence and positional issues had strong effects on the choices voters made. Campaigning in 2008 during the biggest financial crisis since the Great Depression, Barack Obama had boldly reiterated the mantra “Change You Can Believe In” to propel his successful race for the presidency. There were serious political consequences when the expectations he raised went unfulfilled. High unemployment and anaemic growth continued to beset the economy in 2010, and the President's landmark health-care legislation and policy proposals in areas such as climate change and immigration were debated in a context of widespread disappointment with his performance. This context enhanced voters' susceptibility to Republican claims that the President's policies initiatives were ill-advised. The result was a politically toxic mix of valence and positional issues which corroded Obama's image and worked strongly against Democratic candidates.

Notes

In a political context a shellacking means a decisive defeat.

The idea that mid-term congressional elections may be viewed as referendums on presidential performance has a lengthy history and has stimulated lively controversy. See, e.g., Tufte (1975); Kernell (1977); Abramowitz (1985); Campbell (1997); Bafumi et al. (2010).

The classic American Voter analyses are updated in Lewis-Beck et al. (2008).

See Clarke et al. (2009a). A substantial portion of the literature on congressional elections focuses on factors affecting outcomes of these contests, rather than on forces affecting individual-level voting behavior. For reviews, see, e.g., Campbell (1997); Jacobson (2004).

The 2010 Political Support in America Study was funded by the National Science Foundation (U.S.), with additional financial support provided to Scotto by the Economic and Social Research Council (U.K.), and to Clarke and Stewart by the University of Texas at Dallas. The authors thank these organizations for their generous assistance. Fieldwork for the 2010 PSA national pre- and post-election Internet panel survey (N = 3,800) was conducted by YouGov/Polimetrix under the direction of Elizabeth Christie. All 2010 PSA respondents were initially 2008 Cooperative Campaign Analysis Project (CCAP). Vote shares in the 2010 survey are Democrat = 47.1%, Republican = 50.5%, Other = 2.4%. These figures differ by only 2.3%, 1.1% and 1.2%, respectively, from the actual national vote totals (average absolute difference = 1.5%). See Vavreck and Rivers (2008) for information on YouGov/Polimetrix respondent selection procedures. On the utility of internet surveys for studying political attitudes and behavior in the U.S., see Ansolabehere and Schaffner (2011). Survey questionnaires are available at http://www.tandf.co.uk/journals/fbep. The survey data may be downloaded from the PSA website: http://politicalsupportamerica.org.

The 2008 Obama “like-dislike” data are from the authors' question module (N = 1,000) in the post-election wave of the 2008 Cooperative Campaign Analysis Project (CCAP).

The AIC statistic imposes a penalty for the richness of model parameterization. AIC values are calculated as -2*model log-likelihood + 2k where k is the number of estimated parameters. See, e.g., Burnham and Anderson (2002).

The possibility that the composite vote model estimates suffers from simultaneity bias involving congressional voting and the Obama image variable was investigated. A Hausman test (Davidson & MacKinnon, 1993; Hausman, 1978) using the composite vote model and the results of the Obama image model analysis discussed below was performed. The test result (t = -0.500 p = 0.614) indicates that the composite vote model estimates are not confounded by such a bias. Further confidence in the results is gained by noting that all of the component variables in the Obama image factor-score variable were measured in the pre-election wave of the 2010 survey and none of these questions were asked in proximity to questions asking respondents about their vote in the forthcoming congressional election.

Probabilities were calculated using the CLARIFY program. See Tomz et al. (1999).

All of the 2010 PSA respondents were originally interviewed in the multi-wave 2008 Cooperative Campaign Analysis Project (CCAP). The question measuring attitudes toward Obama in the November 2008 CCAP post-election survey is: “Here is a list of politicians (list includes Obama, McCain, Biden, Palin). How favorable is your impression of each person, or haven't heard enough to say?” Response categories are “very favorable” = 5, “somewhat favorable” = 4, “neutral/haven't heard enough” = 3, “somewhat unfavorable” = 2, “very unfavorable” = 1. Including a measure of feelings about Obama in 2008 raises the possibility that model misspecification might create a simultaneity bias. Given the elaborate theoretically driven, specification of the Obama image model, we believe this is doubtful. Also, the nature of the bias created by the presence of a lagged endogenous variable works to increase the size of the coefficient of that variable while diminished the size the coefficients associated with other predictors in the model. That, in turn, means that estimates of the significance of those predictors will be conservative rather than exaggerated. See, e.g., Ostrom (1990).

Quote from remarks by President Obama in Chicago, November 26, 2008 as he named an Economic Recovery Advisory Board.