jost_j_t_kay_a_c_thorisdottir_n_social_and_psychological_bas

Table 13.1 Multiple regression analyses of the state-level personality, sociodemographic, and political models for the Democratic Presidential candidates in the 1996, 2000, and 2004 elections. (Continued )

Note. Cell entries are nonstandardized partial regression coeffi cients (and their standard errors). State-level personality estimates for 1996 and 2000 were derived from Sample 1; state-level personality estimates for 2004 were derived from Sample 2.

a Sample 1, b Sample 2;

+p < .1, * p < .05, ** p < .01, *** p < .001

Democratic candidates (R = .730, .772, and .755, for Clinton, Gore, and Kerry, respectively). As can be seen in the middle portion of Table 13.1, the percentage of blue-collar workers accounted for more variance than any of the other sociodemographic predictors and was negatively related to votes for each of the candidates. Two variables that accounted for slightly smaller proportions of the variance were percentage of residents with college degrees and percentage of African Americans, which were both positively related to votes in each election.

Political model. The political model included a political behavior change variable, which was the difference between the percentage of votes for the Democratic candidate in the two previous elections (i.e., 1988 and 1992 to predict 1996 votes, 1992 and 1996 to predict 2000 votes, and 1996 and 2000 to predict 2004 votes). Analyses of the political model revealed that it shared substantial proportions of variance with votes for the three candidates (R =

.298, .244, and .570, Clinton, Gore, and Kerry, respectively). The bottom portion of Table 13.1 shows that the changes in statewide Democratic votes in the two elections prior to those under investigation were significant positive predictors of the percentages of same-party votes in the 1996 and 2004 elections and a marginally significant positive predictor in the 2000 election.

Comparison of the models. Each of the three models accounted for significant proportions of variance in the percentage of votes for the Democratic candidates in all three elections. To determine whether state-level personality accounted for unique variance in voting patterns after adjusting for the effects of sociodemographic and political variables, we conducted a series of stepwise regressions comparing changes in the R between the state-level personality model and each of the other two models.

We first examined the combined and unique effects of the state-level personality and sociodemographic models on the percentages of votes for each Democratic candidate. When combined, the two models shared 82%, 84%, and 85% of the variance in votes for Clinton, Gore, and Kerry, respectively. Comparisons of the changes in the Rs showed that, even when the sociodemographic model was included, the state-level personality model still explained significant proportions of variance in the percentage of votes for Clinton (∆ R = .09, F (5, 39) = 3.479, p < .01), Gore (∆ R = .07, F (5, 39) = 3.053, p < .05), and Kerry (∆ R = .10, F (5, 39) = 4.340, p < .01).

We next examined the unique and combined effects of the state-level personality and political models. When both models were combined, they accounted for 64%, 69%, and 84% of the variance in votes for Clinton, Gore, and Kerry, respectively. Analyses of the proportions of variance accounted for by the state-level personality model when the political variable was controlled revealed that it accounted for significant proportions of unique variance in the percentage of votes cast for Clinton (∆ R = .34, F (5, 44) = 4.697, p < .01), Gore (∆ R = .45, F (5, 44) = 7.028, p < .001), and Kerry (∆ R =

.27, F (5, 44) = 11.089, p < .01).

Republican Candidates. State-level personality model. Analyses of the percentages of Republican votes cast in the three elections revealed that the statelevel personality model shared significant proportions of variance with votes cast for Dole (in 1996) and Bush (in 2000 and 2004) (R = .614, .672, and .728, respectively). As shown in the top portion of Table 13.2, the patterns of relations between each of the Big Five personality dimensions and the percentage of Republican votes cast in each election were virtually identical across the three elections. The partial regression coefficients revealed that openness again accounted for the most variance, but this time it was negatively related to votes cast for each candidate. Conscientiousness was a significant and

positive predictor of votes for each of the Republican candidates. Extraversion was a significant negative predictor of the percentages of votes cast for Dole and Bush. Neuroticism was a significant (negative) predictor of the percentage of votes cast for Bush in 2004 but not in 2000. Thus, the findings from our analysis of Republican candidates were precisely opposite to those obtained in the analysis of Democratic candidates.

Table 13.2 Multiple regression analyses of the state-level personality, sociodemographic, and political models for the Republican Presidential candidates in the 1996, 2000, and 2004 elections.

Table 13.2 Multiple regression analyses of the state-level personality, sociodemographic, and political models for the Republican Presidential candidates in the 1996, 2000, and 2004 elections. (Continued )

Note. Cell entries are nonstandardized partial regression coeffi cients (and their standard errors). State-level personality estimates for 1996 and 2000 were derived from Sample 1; state-level personality estimates for 2004 were derived from Sample 2.

a Sample 1, b Sample 2;

+p < .1, * p < .05, ** p < .01, *** p < .001

Sociodemographic model. The sociodemographic model significantly predicted the percentage of votes for the Republican candidates in all three elections (R = .634, .776, and .754, respectively). As shown in the middle portion of Table 13.2, the percentage of blue-collar workers accounted for significant proportions of variance, and was positively related to the percentage of votes for the Republican candidates in each election. Percentage of residents with college degrees was the only other variable that accounted for substantial proportions of variance, and was negatively related to votes for Republican candidates.

Political model. The bottom portion of Table 13.2 shows that the political model accounted for significant proportions of variance in votes for Bush in 2000 and 2004 (R = .622 and .631, 2000 and 2004, respectively), but not Dole (R = .211). In both 2000 and 2004, change in same-party votes was positively related to votes for Bush, but negatively related to votes for Dole in 1996.

Comparison of the models. Comparisons of the state-level personality and sociodemographic models indicated that, when both models were combined,

they shared 77% of the variance with votes for Dole and 85% of the variance in votes cast for Bush in 2000 and also in 2004. Comparisons of changes in the Rs showed that even after adjusting for sociodemographic variables, the state-level personality model still explained a significant proportion of variance in the percentages of votes for Dole (∆ R = .14, F (5, 39) = 3.855, p < .01) and Bush in both 2000 (∆ R = .07, F (5, 39) = 3.232, p < .05) and 2004 (∆ R = .10, F (5, 39) = 4.597, p < .01).

When the state-level personality and political models were combined, they accounted for 66% of the variance in votes for Dole, 84% of votes for Bush in 2000, and 86% of Bush votes in 2004. Furthermore, analyses of the unique contribution of the state-level personality model indicated that it remained a significant predictor of votes for Dole in 1996 (∆ R = .45, F (5, 44) = 6.170, p < .001) and Bush in 2000 (∆ R = .221, F (5, 44) = 9.900, p < .001) and 2004 (∆ R = .226, F (5, 44) = 11.086, p < .001). Overall, the portrait that emerged from our analyses of Republican voting patterns almost perfectly complemented the picture derived from the Democratic data.

Third-Party Candidates. Our analyses of third-party candidates focused only on the candidate who received the highest percentage of third-party votes in each of the three elections. Unfortunately, because there was not always a Reform, Green, or Independent party candidate in the two consecutive elections preceding those under investigation, we were not able to compute political behavior variables for any of the third-party candidates. Therefore, we examined the independent, shared, and unique effects of the state-level personality and sociodemographic models on the percentages of votes cast for Perot in 1996 (Reform Party), Nader in 2000 (Green Party), and Nader in 2004 (Independent). Analyses of Nader votes in 2004 were based only on the 35 states in which he was on the ballot.

State-level personality model. Analyses of the percentages of third-party votes cast in the three elections revealed that the state-level personality model shared significant proportions of variance with votes cast for Nader in 2000 and 2004 (R = .604 and .624, respectively) but not for Perot in 1996 (R =

.325). As shown in the top portion of Table 13.3, state-level openness was a marginally significant negative predictor of the percentage of votes for Perot, but not for Nader in either 2000 or 2004. Conscientiousness and neuroticism were negative predictors of the percentages of votes cast for Nader in both elections, but they had no effect on support for Perot.

Sociodemographic model. The results displayed in the middle portion of Table 13.3 show that the sociodemographic model shared significant proportions of variance with the percentages of votes cast for Perot in 1996 and Nader in 2000 and 2004 (R = .751, .755, and .740, respectively). The partial

Table 13.3 Multiple regression analyses of the state-level personality, sociodemographic, and political models for the third-party Presidential candidates in the 1996, 2000, and 2004 elections.

Note. Cell entries are nonstandardized partial regression coeffi cients (and their standard errors). State-level personality estimates for 1996 and 2000 were derived from Sample 1; state-level personality estimates for 2004 were derived from Sample 2.

a Sample 1; b Sample 2; § N = 35

+p < .1, * p < .05, ** p < .01, *** p < .001

regression coefficients reveal that the percentage of African Americans was negatively related to the percentage of votes cast for Perot and Nader in 2000 and 2004. In the 2000 and 2004 elections, the percentage of white-collar workers was a significant positive predictor of votes cast for Nader.

Comparison of the models. Analyses of the shared effects of the state-level personality and sociodemographic models on third-party votes revealed that together they accounted for 81%, 82%, and 85% of the variance in votes for Perot in 1996, Nader in 2000, and Nader in 2004, respectively. Comparisons of changes in the Rs showed that when the sociodemographic model was included in the regression equation, the state-level personality model accounted for a marginally significant proportion of unique variance in the percentage of votes for Perot (∆ R = .058, F (5, 39) = 2.070, p < .1) and Nader in 2000 (∆ R = .060, F (5, 39) = 2.192, p < .1), and a significant proportion of the variance in votes for Nader in 2004 (∆ R = .110, F (5, 23) = 2.827, p < .05).

General Discussion

Previous research indicates that differences in historical settlement patterns, cultural diversity, and local economic circumstances contribute to regional voting patterns (Agnew, 1987a, 1987b; Bensel, 1984; Elazer, 1994; Heppen, 2003; Hero, 1998). The present work set out to examine an additional potential influence on voting patterns. Using multiple samples, elections, and sources of data, our results converged to indicate that statewide differences in basic personality traits account for variation in Presidential election voting patterns.

The results were largely consistent with previous research on the links between personality and political orientation at the individual level (Caprara et al., 1999; Carney et al., 2008; Gosling et al., 2003; Jost et al., 2003a, 2003b; McCrae, 1996; Ozer & Benet-Martinez, 2006) and shed new light on the psychological characteristics of individuals inhabiting different geographical regions and their relations to voting patterns. Consistent with our predictions, state-level openness and conscientiousness scores differentially predicted the percentages of votes cast for Democratic and Republican candidates in the three presidential elections between 1996 and 2004. More specifically, states with higher average levels of openness and lower levels of conscientiousness were more likely to vote for Democratic candidates and less likely to vote for Republican candidates. On the basis of these findings, people living in “blue states” would be characterized as more curious, creative, imaginative, intellectual, and tolerant of differences than people living in “red states.” At the same time, residents of “red states” will tend to

be more traditional, reliable, organized, efficient, and self-disciplined than residents of “blue states.”

Although no explicit predictions about the relation between extraversion and political orientation were made, state-level extraversion scores were positively related to preferences for Democratic candidates and negatively related to preferences for Republican candidates. Thus, inhabitants of “blue states” could be characterized as more talkative, enthusiastic, energetic, and sociable and less inhibited, quiet, and reserved than people living in “red states.” Our extraversion findings are inconsistent with those of Caprara and colleagues (1999, 2003), who found that energy was associated with centerright (rather than center-left) political preferences in Italy. Future research is needed to better understand the causes of cross-national differences in psychological predictors of political orientation (see Thorisdottir, Jost, Liviatan, & Shrout, 2007).

With regard to independent and third-party candidates, at least two general patterns seemed plausible. First, the same personality characteristics (e.g., high openness or low agreeableness) could have predicted statewide support for minority party candidates in general, to the extent that all thirdparty candidates are likely to depart from the political mainstream. Alternatively, the personality characteristics that predicted support for Republican candidates could have also predicted support for relatively conservative third-party candidates (like Perot), and the characteristics that predicted support for Democratic candidates could have predicted support for liberal third-party candidates (like Nader). We found more evidence in support of the latter possibility. Specifically, individuals in states in which Perot received stronger support in 1996 were somewhat lower in openness (like Republican states), whereas individuals in states where Nader received more support were lower in conscientiousness (like Democratic states).

The results from our analyses comparing the relative contributions of personality, sociodemographic, and political variables provide further evidence for the robust effects of state-level personality on political preferences. Out of a total of 15 model comparison analyses, the state-level personality model accounted for significant proportions of unique variance in 13 (87%) of them. These findings strongly suggest that personality traits common in a region affect how individuals in the region vote, independent of the sociodemographic, economic, and political characteristics of the region. Although the other two models accounted for substantially more unique variance, the present findings nevertheless suggest that state-level personality can inform our understanding of regional voting patterns over and above the information provided by more traditional variables. It is worth emphasizing

that the patterns of findings were replicated in two separate personality samples and three presidential elections.

Limitations of the Present Research

As in most studies that rely on self-report methodologies, it is possible that responses to the BFI were affected by social desirability concerns. That is, participants might have attempted to answer the personality items in such a way as to present themselves in a favorable light (Borkenau & Amelang, 1985; Paulhus, 1991); the problem with this is that the findings could reflect social desirability rather than personality per se. Given that responses were completely anonymous and that participants’ primary incentive for completing the questionnaire was to receive feedback on their personalities, it is unlikely that self-presentational concerns drove their responses. Nonetheless, to rule out this possibility empirically, we adopted the well-known strategy of computing for each participant a social desirability index from the mean of all personality items after recoding the items in a socially desirable direction (Paulhus, 1991). We then recalculated state-level personality estimates after removing the top 20% of scorers on this social desirability index and re-ran the multiple regression analyses to predict voting patterns. The effects were virtually identical to those obtained when all respondents were used, suggesting that socially desirable responding did not drive the effects of state-level personality on voting patterns.

Another potential limitation of the present work is that the samples were gathered using a relatively novel Internet methodology, raising the question of whether our findings would generalize to the population as a whole. The samples were not representative because they were based on self-selected groups of participants who chose to complete the personality measure online. However, recent work comparing self-selected Web-based samples to traditional convenience samples of non–self-selected participants suggests that this problem is not as serious as one might expect. For one thing, the Big Five scale reliabilities in the present work were very similar to those obtained in research using non–self-selected participants (Gosling et al., 2003; John & Srivastava, 1999), suggesting that our participants were not responding less carefully or systematically than participants from other samples. In addition, research demonstrates that self-selected online respondents tend to provide clearer and more complete responses than do respondents who are not similarly self-selected (Gosling et al., 2004; Walsh, Kiesler, Sproull, & Hesse, 1992). Research also shows that participants are less likely to engage in socially desirable responding when completing Web-based surveys than paper-and-pencil questionnaires or telephone interviews (Richman, Kiesler, Weisband, & Drasgow, 1999). Moreover, we found that our samples were

more culturally representative of the population than are most convenience samples. At the same time, we recognize that future research on the effects of state-level personality on voting patterns would benefit from work using even more representative sampling methods.

CONCLUSION

The present work demonstrates that personality variables can be used to illuminate regional voting patterns. We found that the psychological characteristics of people inhabiting certain geographical regions independently contribute to their political preferences, even after adjusting for standard sociodemographic and political indicators. Importantly, the effects of statelevel personality on statewide voting patterns generalized across two different personality samples and three presidential elections. Our findings provide empirical support for previous observations that individuals “cluster in places where people share their cultural aesthetic and, as it turns out, political values” (Brooks, 2004, p. A27). That is, “blue states” are disproportionately high in openness, and “red states” are disproportionately high in conscientiousness. By taking into account the aggregate effects of individual-level personality estimates, our work sheds light on the complex interplay between regional and psychological characteristics in shaping electoral outcomes.

From our perspective, the main advantage of research on the effects of state-level personality is that it provides information about the psychological processes underlying regional differences in political preferences. Whereas most macro-level research generally draws inferences about the psychological characteristics of individuals on the basis of aggregate-level data (e.g., Elazer, 1994; Florida, 2002), state-level personality research allows for a direct test of these inferences. For example, Elazer’s description of “traditionalistic” political culture suggests that people living in the agrarian South should be especially concerned with respecting authority and maintaining social order and tend to vote for conservative rather than liberal candidates. This description is supported by our empirical findings, which show that residents of “red states” are relatively conventional, concerned with order, and less tolerant of new and diverse ideas.

Among the many variables that distinguish regions from one another is the availability of certain forms of human capital and economic growth. Whereas metropolitan regions generally have strong economies with large proportions of creative human capital (i.e., individuals who encourage and support innovation), rural regions tend to have comparatively less economic growth and higher rates of social capital in terms of family-, community-, and civic-oriented participation (Florida, 2002; Putnam, 2000). Both of these