State-level personality and presidential vote share in the 2008 and 2012 presidential elections

Abstract

Recent research has examined how variation across the states in the “Big Five” personality trait taxonomy helps explain the proportion of votes the presidential candidates receive in the states, concluding that state personality traits had a direct effect on presidential vote share in the 1996, 2000, and 2004 presidential elections. The current study has three goals: First, to examine the influence of personality traits on Barack Obama's vote share in the 2008 and 2012 elections; second to test whether the influence of personality traits on vote share holds under stricter controls for political factors and white racial prejudice; and, third, to test for potential meditating effects of state-level political characteristics and white racial prejudice in linking state-level personality traits with Obama's vote share. The findings indicate that two state personality traits – conscientiousness and openness – had indirect effects on Obama's 2008 and 2012 vote share through their influence on state ideology, partisanship, and white racial prejudice.

Keywords:

• Personality
• Presidential vote
• Subcultures
• Citizen liberalism
• Policy mood
• White racial prejudice

Notes

1 Importantly, regional subcultures and regional personality traits are distinct concepts. On one hand, a subculture is a “way of life and system of shared values that legitimate a preferred set of social relationships…[and] provide individuals with their basic social identities (the identity function), norms for socially acceptable behavior (the boundary maintenance function), and standards for judging social institutions (the legitimating function)” (Lieske, 2010, p. 540). On the other hand, “personality” refers to “a multifaceted, enduring, internal psychological structure” (Mondak et al., 2010, p. 86), and is primarily a function of genes (Hatemi, Medland, & Eaves, 2009).

2 Other research establishes the influence of individual-level personality traits and candidate support in presidential elections. For recent examples, see Jost, West, and Gosling (2009) and Osborne and Sibley (2012).

3 See http://www.census.gov/compendia/statab/.

4 See http://rcfording.wordpress.com/state-ideology-data/ for the Berry et al. measure of citizen liberalism and http://academic.udayton.edu/SPPQ-TPR/index.htm for the Carsey and Harden measures of state policy mood and partisanship.

5 OLS regression is used to predict Obama vote share in individual years. For the pooled analysis, two models are estimated. The first uses generalized least squares (GLS) regression and the second uses OLS regression with panel corrected standard errors (PCSE). Both models allow for heteroskedasticity and first order autocorrelation, and are appropriate when estimating models that include both cross-sectional and panel data (Beck & Katz, 1995). Appendix A includes the bivariate correlations between all of our variables. As evident from this table, some of the independent variables are highly correlated. However, variance inflation factors scores show that the models do not suffer from dangerously high levels of collinearity. Variance inflation favor scores are all below eight, which is considered the cutoff between benign and harmful collinearity. Only two variables have variance inflation factor scores above 4.5 – policy mood (VIF = 6.60 in 2008 and 2012) and white racial prejudice (VIF = 5.8 in 2008 and 5.5 in 2012). These findings leave us confident that, while some variables have high correlations, our analysis does not suffer from multicollinearity. Finally, tests reveal the presence of heteroskedasticity in the regressions. Consequently, robust standard errors are estimated, which relax the assumption that errors are independent and identically distributed.