Abstract
Differences in political culture have been observed at the cross-national and subnational levels, and political culture corresponds with a wide array of important social and political phenomena. However, possible psychological correlates of political culture are less clear. Building on research in personality psychology and cross-cultural psychology, this study contemplates whether aggregate personality measures compiled in the American states correspond with patterns in political culture. Using measures of personality traits provided by more than 600,000 survey respondents, parallels with state-level measures of citizen ideology, political culture, and civic culture are examined. Possible mechanisms linking personality and political culture are discussed.
Virtually all political processes unfold at multiple interrelated levels. As an example, to understand why an individual voted for the challenger rather than the incumbent may require that we account not only for the person’s political orientations, sophistication, and psychological dispositions but also for aspects of the individual’s social networks and contexts, for characteristics of the candidates, and for elements of the national political arena. After doing all of this, the real purchase may come only once we link our understanding of the actions of the individual voter to political phenomena that occur at the collective level—the winning and losing of elections, the postelection composition of legislatures, and patterns in public policy.
Attention to micro-level processes is needed to understand what prompts individuals to form attitudes and to engage in actions. However, the value of such research may be limited if we lose sight of links between micro-level processes and collective-level outcomes. Eulau (1968, 209) made this point well, noting that “although concrete political action is invariably the behavior of individual human actors, the politically significant units of action are groups, associations, organizations, communities, states, and other collectivities.”
In recent years, students of personality and cross-cultural psychology have considered the link between individuals and aggregates with focus on the Big Five personality trait structure. This research addresses whether collective-level variation in personality corresponds with state-, regional-, and national-level phenomena. Building on this work, the present study puts the focus on politics, and does so via attention to the fifty American states. We explore connections between measures of the Big Five 1 trait structure and data on state political culture. In doing so, we have three goals. First, following Eulau’s recommendation, we seek to demonstrate that micro-level psychological dispositions can be linked to tangible collective-level variables. Second, we hope to shed new light on the nature and significance of state political culture, and especially on connections between political culture and measures emanating from citizen-level data. Third, and perhaps most critically, we hope to spark questions regarding why personality and culture may be linked and what such connections may imply with respect to state-level differences in politics and governance.
The remainder of the article includes three primary sections. First, we outline the rationale for exploration of a correspondence between collective personality and state political culture, and we specify our expectations regarding possible links between personality and culture. The next section discusses the key empirical indicators we employ, and previews our modeling strategy. Finally, the article’s third main section presents a series of tests exploring possible links between collective personality and state-level measures of citizen ideology, and civic and political culture.
Politics and Psychology at the Collective Level
The core assumption of the current research is that aggregate personality traits capture something meaningful regarding differences between the residents of various locales. Embedded within this assumption are numerous questions. First, if, as scholars have contended, the Big Five trait structure exists across cultures and the trait dimensions are highly heritable, why would meaningful geographic variation exist? Second, in terms of measurement, are there obstacles that complicate the representation of collective-level personality? Third, if such variation exists, what are its correlates? Fourth, and perhaps most critically, what process or processes give rise to any such correlations? Is it possible to suggest, test, and identify causal relationships?
Some scholars have pursued these questions for decades, but it is only in recent years that empirical inquiry focused on broad representations of personality has taken root on a large scale. 2 Allik and Realo (2009) noted that studies integrating the concepts of personality and culture were rare in the 1980s, whereas more than one hundred such studies appeared per year in the few years leading up to their review. One factor sparking this wave of research is the development of the Big Five framework. First, since emerging in the late-1980s, this typology has established a dominant position in the field. Scholars describe it as marking a paradigm shift (John, Naumann, and Soto 2008), and they note that the volume of research on this approach dwarfs that concerning prominent earlier personality typologies (e.g., Gosling, Rentfrow, and Swann 2003). Second, scholarly curiosity has been piqued by the “universal” character of the Big Five structure (e.g., McCrae and Costa 1997), with numerous studies establishing that the Big Five trait dimensions can be observed in nations around the world. Third, the efforts of Robert McCrae, a key figure in development of the Five-Factor model, have been especially consequential. McCrae et al. (2005a, 2005b) launched the “Personality Profiles of Cultures” initiative, which led to the acquisition and study of cross-national data on personality.
We build on these efforts in two key manners. First, whereas most work on collective personality has employed cross-national designs, we explore variation within one nation by examining personality and its correlates in the American states. Second, rather than considering general elements of culture, we focus on political culture. We first revisit the questions outlined above: why collective personality should be expected to vary across geographic units, whether any obstacles hamper the study of collective personality, what likely correlates of collective personality should be considered, and what processes underlie any correspondence between collective-level personality and other geographically varying phenomena.
Geographic Variation in Personality
The first task researchers have addressed in the new wave of inquiry on personality and culture has been the demonstration that personality varies in sensible, systematic ways, across geographic units. The particular matters that require discussion are why geographic variation might exist, and what patterns have been observed. Given the wealth of scholarly attention to these issues, the present discussion is necessarily brief.
The Big Five dimensions are highly heritable, with error-corrected heritability estimates in excess of 0.50, and as high as 0.79 (e.g., Riemann, Angleitner, and Strelau 1997). If variation in personality within a population is reflective of biological influence, is there reason to expect variation in collective personality when regions are compared? There are at least three reasons such variation might be seen. 3 First, enculturation within geographic units may occur. For instance, if enculturation in one region had the effect of increasing every resident’s extraversion by 10 percent, and a corresponding process in another region reduced extraversion by 10 percent, two phenomena would be observed: within each region, variation in extraversion would reflect the influence of biology, yet between regions, a difference in collective-level values would be seen. Second, personality-related alleles prevalent within one culture may vary relative to those in another due to differences in the ancestries of the populations. That is, biological differences may be clustered geographically, and persist across generations. Third, selective migration may occur, with personality likes attracting likes, or with people with similar personality structures attracted to features of a particular geographic area. What is important about each of these accounts is that they allow for biological influence on personality and the occurrence of collective-level differences when personality means are compared across geographic locales.
Several studies have reported systematic variation in personality, cross-nationally and among the American states. Two efforts have been especially prominent. First, McCrae and his colleagues (e.g., McCrae and Allik 2002; McCrae et al. 2005a, 2005b) have reported Big Five data obtained from two cross-national studies, each involving respondents from at least thirty-six nations. This research identifies consistent patterns in personality means across cultures, and shows this variation to be related to numerous national-level indicators. Second, Rentfrow and his colleagues (Rentfrow 2010; Rentfrow et al. 2009; Rentfrow, Gosling, and Potter 2008) examine variation in Big Five data across the American states. Rentfrow and his coauthors have sought to pinpoint patterns in state-level personality, and to link this variation to differences in other state-level indicators.
Potential Pitfalls in Measuring Collective-Level Personality
Comparisons of survey data across geographic units routinely face a series of perils. This is particularly true in cross-national inquiry where researchers must grapple with language differences, cross-cultural differences in response styles, and possible effects on levels of measurement error due to cross-national differences in reading comprehension. Cross-national research on personality also may be compounded by reference-group effects; if people in a geographic area rate their personalities via comparisons between themselves and others around them, values aggregated at the geographic level would have little meaning because the metric would shift from one context to the next (e.g., Heine et al. 2002). 4 These concerns have sparked extensive discussion among psychologists (e.g., Church 2001; Church et al. 2011; McCrae et al. 2007). All scholars see these matters as serious, although disagreement persists as to whether these issues pose insurmountable obstacles to meaningful cross-cultural comparisons.
Although these concerns should never be taken lightly, they are much less problematic for comparisons at the subnational level. Here, we examine state-level personality data. Because respondents share the same national culture, and most share the same language, the obstacles seen in cross-national analyses are dramatically less troublesome. We draw on the data collected and examined previously by Rentfrow and his colleagues. Those authors have reported thorough analysis concerning the properties of the state-level personality measures, and have generated convincing evidence that the personality scales are reliable and that they support valid inferences regarding collective personality.
Linking Personality and Political Culture
We examine three aspects of the political character of American states: citizen ideology, political culture, and civic culture. As a prelude, we assess the possible links between personality and three other correlates of state-level political phenomena: minority diversity, white ethnic diversity, and an indicator of state socioeconomic status (Hero 2007; Hero and Tolbert 1996). Although most past research on the correlates of collective-level personality has not directly examined political constructs such as these, prior work does provide guidance regarding likely patterns of results. Likewise, past individual-level research on the political correlates of personality offers a foundation for corresponding collective-level assessments.
We will examine citizen ideology at the state level as a validity check. In individual-level analyses, claims of a possible link between personality and ideology enjoy a long history (e.g., Eysenck 1954; McClosky 1958). More recently, numerous studies have sought to identify links between the Big Five trait dimensions and ideology. Although all five traits have produced significant effects in at least some studies, the most consistent relationships are between openness to experience and ideological liberalism, and between conscientiousness and conservatism (e.g., Gerber et al. 2010; Jost 2006; Mondak 2010; Mondak and Halperin 2008; van Hiel, Kossowska, and Mervielde 2000). We are unaware of corresponding efforts to link aggregate personality and ideology. However, Rentfrow (2010; see also Rentfrow et al. 2009) has examined the relationship between state-level personality and voting patterns in the 1996, 2000, 2004, and 2008 U.S. presidential elections. Results reveal significant correlations between openness and the Democratic share of the vote, and between conscientiousness and the Republican vote—findings that prompted Rentfrow (2010, 552) to observe that “a macro-level psychological perspective may be useful for understanding the American political divide.” Our expectation is that familiar individual-level patterns will be observed at the collective level, and especially for links between state-level measures of openness and liberalism, and conscientiousness and conservatism.
The centerpiece of our analysis is Daniel Elazar’s (1984) three-part typology of political culture. 5 In Elazar’s framework, each state was classified as being characterized by one of three cultural types—moralistic, individualistic, or traditional. Elazar described moralistic state cultures as viewing government as a positive force for the promotion of the public good. Among citizens, service to the collective good is especially valued. In individualistic cultures, government is seen in more practical and less idealistic terms. The private sphere trumps the public, and government should seek to foster, not replace, effective private action. Finally, traditionalistic cultures emphasize a hierarchical social order. Government is seen as valuable, but primarily within the confines of existing social structures and class divisions, not as a means to dismantle them.
Elazar’s typology has sparked inquiry, prompted questions, and been met with a degree of scholarly skepticism. Elazar developed his typology nearly fifty years ago. Hence, one issue is whether recent measures of collective personality should be related to Elazar’s classification. Culture and personality are presumed to move sluggishly, yet a decades-long gap still may be sufficient to obscure relationships that may have been observed with contemporaneous measures. Conversely, if cultural and demographic dynamics worked to solidify earlier patterns, the passage of time may lead to a strengthening of any link between culture and personality. A second issue is how the typology should be operationalized, and how any operationalization might relate to personality. Elazar framed his classification in multidimensional terms, yet he and many subsequent researchers also described a unidimensional continuum on which states are arrayed from moralistic to individualistic to traditional. Finally, the collective personality measures draw on data aggregated from mass surveys, yet past research (e.g., Nardulli 1990) has struggled to find evidence consistent with Elazar’s framework in analyses of survey data. Thus, even if political culture corresponds with facets of government structure and output, questions remain regarding whether the mass public’s collective psychological orientation plays any role.
Past research on the correlates of collective personality has not considered political culture. However, aspects of this work are suggestive of possible connections. First, cross-national research has noted relationships between Hofstede’s (1980) measures of national culture and Big Five data (e.g., Hofstede and McCrae 2004). The fact that a cultural typology could be linked to data on collective personality is encouraging. Further encouragement comes from Rentfrow’s (2010; Rentfrow, Gosling, and Potter 2008) analyses. Rentfrow reports a positive correlation between agreeableness and Putnam’s (2000) state-level social capital index, and corresponding negative correlations between social capital and conscientiousness and neuroticism. Rentfrow, Gosling, and Potter (2008) found similar correlations with state-level measures of social engagement.
Drawing on these findings and individual-level research on the Big Five, 6 our strongest expectations regarding Elazar concern conscientiousness and agreeableness. Conscientiousness includes a tendency toward deference to authority, an emphasis on personal responsibility, and a link with ideological conservatism. Thus, states with high levels of conscientiousness should exhibit traditional and perhaps individualistic political cultures, and should be relatively unlikely to embrace the moralistic culture’s sense of big government as a positive force. Based on the findings of Rentfrow and his colleagues, similar results for neuroticism also can be expected. Agreeableness, a trait related to ideological liberalism and to concern with the welfare of others, should be positively related to moralistic culture. However, given the role of family and social connections in traditional culture, agreeableness also may be observed in traditionalistic states. It is in states with individualistic cultures, ones that emphasize isolated action rather than collective endeavors, where high levels of agreeableness are unlikely.
In addition to Elazar’s typology, we will also consider links between personality and state-level civic culture. The foundation for research on civic culture is Almond and Verba’s (1963) classic cross-national study. Putnam (1993) extended work on civic culture to the subnational level in groundbreaking research in Italy. Rice and Sumberg (1997) followed by developing indicators of civic culture in the American states. Our expectations are relatively straightforward. Rice and Sumberg note that their measure is correlated with a unidimensional Elazar scale, and thus, the conscientiousness, neuroticism, and agreeableness effects noted above again should be expected. These expectations are bolstered because Rentfrow, Gosling, and Potter (2008) identified correlations between state-level personality and several indicators similar to those Rice and Sumberg used to measure civic culture.
The Question of Process
Scholars in cross-cultural psychology and personality psychology have begun offering accounts regarding the processes that may connect collective-level personality to other geographically clustered phenomena. 7 Although intriguing perspectives have been advanced, our view is that a definitive identification of process most likely will not emerge for quite some time. Any puzzle involving relationships among deeply rooted aggregate variables is inherently difficult to solve. A first step is identification of correlations, as doing so can help spark insights as to the possible causal processes at work. Recognizing this, our objective in the present study is not to pinpoint causal processes, but rather to tackle the prefatory matters of determining whether collective personality and state political culture are related at all, and how any such relationships are affected by accounting for other known correlates of culture.
Although we will offer no definitive claims regarding causal processes, we see heuristic value in the model developed by Rentfrow, Gosling, and Potter (2008). Their model highlights five types of causal mechanisms. The first involves the direct impact of personality on behavior. Where particular psychological dispositions are common, behaviors influenced by those dispositions also should be prevalent. For instance, if agreeableness is abundant in an area, then civility in social relations also should be common. Second, Rentfrow et al. suggest that patterns of behavior may manifest themselves in aggregate social indicators. In the present example, a pattern of civility may eventually produce a relatively low crime rate. 8 Third, social influence may affect behavior. For instance, individuals who do not score high in agreeableness may nonetheless be induced to behave politely (or at least not criminally) toward others due to their exposure to local norms. Fourth, institutions may affect behavior. Where agreeableness is high, there may be a strong investment in schools, charitable organizations, public recreational facilities, and so on, all of which may work to steer individuals away from criminal activity. Finally, social norms may affect collective levels of traits. If traits are malleable at the individual level, exposure to social norms may produce changes in people’s actual psychological dispositions. But even if traits do not change at the individual level, a region’s collective personality may be altered if prevailing norms prompt social migration. If the crime rate in Region A is the most disconcerting to individuals high in agreeableness, those individuals may seek to move from Region A to Region B. As a result, the mean level of agreeableness in Region A will decline, and the corresponding level in Region B will rise.
The mechanisms suggested by Rentfrow, Gosling, and Potter (2008) are not mutually exclusive. Thus, precise allocation of causal influence among these paths will be difficult. The first task remains identification of correlations. Inquiry regarding possible connections between personality and political culture potentially will bring advantages irrespective of success in identification of causal paths. Most importantly, this work may establish that personality and culture possess greater theoretical and substantive reach than has previously been recognized.
Data and Method
The measures used in this study draw on sources reported in past research, along with data from publicly available indicators. Five measures require discussion: state-level personality, three control variables, state-level citizen ideology, the Elazar political culture framework, and state civic culture and its subcomponents. Empirical analyses will seek to determine whether personality is correlated with various indicators of state politics and culture. To account for collinearity among the Big Five measures, data will be examined in a series of ordinary least squares (OLS) regression models, complemented at one point with models estimated via multinomial logistic regression.
The Big Five
Measures of the Big Five are drawn from a voluntary, self-administered Internet survey that included a forty-four-item Big Five Inventory. As with any opt-in Internet-based survey, the representativeness of the sample is a valid concern. That said, the properties of the state-level Big Five data, which are described in detail elsewhere (Rentfrow, Gosling, and Potter 2008), are such that reasonable confidence is warranted with respect to analyses conducted between states. The personality data come from more than 600,000 respondents who completed the survey between December 1999 and January 2005. The data set includes 619,397 respondents; because we omit Washington, D.C., our measures draw on data from 617,242 respondents. The number of respondents ranges from 1,536 in Wyoming to 71,873 in California. Rentfrow, Gosling, and Potter (2008) showed that these response levels are in close proportion to the actual populations of the states (using 2000 U.S. Census data, r = .98).
Several additional properties of these data are addressed by Rentfrow, Gosling, and Potter (2008), two of which we note here. First, the data fare well in representing variation between the states on race and ethnicity, and reasonably well for social class, albeit with a modest underrepresentation of the lowest and highest class strata. Respondents are relatively young as a whole, with a median age of 24, although diverse on age (SD = 9.8 years), particularly when compared with student samples. 9 Second, Rentfrow, Gosling, and Potter report that the Big Five measures are highly reliable at the individual and state levels.
For our analyses, the state-level Big Five measures, which are state means, have been recoded to range from 0 (lowest observed value) to 1 (highest observed value). Descriptive statistics for the scales are as follows: openness, M = 0.69, SD = 0.20; conscientiousness, M = 0.56, SD = 0.19; extraversion, M = 0.39, SD = 0.19; agreeableness, M = 0.74, SD = 0.16; and neuroticism, M = 0.52, SD = 0.21. 10
Rentfrow, Gosling, and Potter (2008) reported relationships between the state-level Big Five measures and social indicators concerning matters such as crime rates, religiosity, and mortality. Significant relationships were found in twenty of forty tests, bringing confidence that the personality scales capture something meaningful at the state level. What is not yet certain, though, is whether collective-level personality data correspond with indicators of states’ political cultures.
Covariates
In all analyses reported below, we will contemplate possible relationships involving the personality measures while accounting for the effects of other known correlates of state-level political phenomena. Our focus will center on Hero and Tolbert’s (1996) measures of minority diversity and white ethnic diversity, and one of the measures of state-level socioeconomic status employed by Hero (2007), high school graduation rates. The minority diversity and white ethnic diversity data are drawn directly from Hero and Tolbert; the former has a mean of 0.23 (SD = 0.15) and the latter has a mean of 0.09 (SD = 0.06). High school graduation rates are taken from 2000 Census data indicating the percentage of adults age twenty-five and older who are high school graduates (M = 86.46, SD = 3.65).
In models addressing ideology and political culture, two versions of each model will be estimated: one in which the factor in question is regressed only on the Big Five measures, and one that includes the personality variables and the covariates. Provided that personality is found to be related to ideology and culture, this strategy offers a first step toward highlighting possible causal paths. First, this approach will enable us to gauge the extent to which any effects of personality operate over and above previously identified influences. Second, if personality and the covariates overlap themselves, it would signal possible mediation.
State Ideology
The Berry et al. (1998) measure of state-level citizen ideology builds on the logic that legislators in a state collectively represent the ideological profile of the citizenry. The measure is constructed using data on the ideology of the major-party congressional candidates in each district, weighted by vote share, and then aggregated to the state level. Berry and his colleagues present an array of evidence regarding the measure’s reliability and validity. The database has been updated to 2008. For 2005, the year we use, state ideology values range from 29.06 (most conservative) in Oklahoma to 89.85 in Vermont (M = 53.39, SD = 13.90).
Although the Berry et al. measure has been subjected to multiple validation tests, it represents citizen ideology indirectly. The only input from a state’s citizens comes via the votes they cast. To tap ideology more directly, a second measure of state ideology is drawn from the 2006 Cooperative Congressional Election Study (CCES; Ansolabehere 2006). Respondents were asked to gauge their own ideological leanings on 5-point and 101-point scales. Data on the two measures are highly correlated at the individual level (r = .95). To combine data from the items, we recoded the 101-point scale so that it ranged in value from 1 to 5, 11 aggregated each measure by state, and then summed the two aggregate measures. 12 Lower values indicate ideological conservatism. State means range from 5.07 in Alabama to 6.14 in Massachusetts (M = 5.56, SD = 0.26). Unlike the Berry et al. scale, this measure taps citizen ideology directly, but it also suffers from an inherent limitation in that state samples are relatively small in some instances, with Ns less than one hundred in four states. Fortunately, the Berry et al. and CCES scales are highly correlated at the state level (r = .75), a finding that supports the validity of both measures.
We will examine the correspondence between state ideology and state-level personality primarily as a validity check. Given that numerous studies have found individual-level relationships between openness and liberalism, and between conscientiousness and conservatism, the search for corresponding collective-level relationships offers a means to establish whether variation in the personality measures maps to a core state-level political indicator.
Elazar’s Political Culture Typology
A key challenge with Elazar’s typology is how to operationalize it for use in empirical analyses. Two issues complicate this task. First, Elazar categorized not just states but also regions within states. As a result, state culture can be quantified fractionally. For instance, two states may be primarily moralistic, but one may be somewhat more so than the other, and it may be worthwhile to capture that distinction. Second, questions persist as to whether the Elazar framework is best represented unidimensionally or multidimensionally. Elazar was inconsistent on this point, and two states are classified as having cultures that mix the two endpoints of the unidimensional scale, traditional and the moral. It follows that nuance may be lost if state political culture is forced to fit an exclusively unidimensional scale.
Our approach is to use two different scales, one that is unidimensional and one that is multidimensional. The first is Sharkansky’s (1969) recoding of the Elazar data. Sharkansky drew on Elazar’s discussion of regional variation within states to construct a precise numerical estimate of each state’s position. Use of the Sharkansky coding allows us to account for relatively fine distinctions among states—provided, of course, that a unidimensional representation of state political culture is appropriate. Our second approach is to use Elazar’s original three-category classification, and to estimate models via multinomial logistic regression. This alternate glosses over nuanced distinctions within each category but has the virtue of relaxing the assumption that the three cultural types are arrayed unidimensionally.
State Civic Culture
Unlike Elazar’s framework, consensus exists that civic culture is unidimensional. Putnam’s (1993) work in Italy offers a clear case for inclusion of four elements—civic engagement, equality and inequality, trust and tolerance, and group membership. Rice and Sumberg (1997) developed civic culture estimates for the American states. Values on their measure range from −1.53 in Mississippi to 1.52 in Vermont. Rice and Sumberg demonstrate a relationship between their measure and Elazar’s (e.g., states categorized as moralistic tend to have high values on the civic culture scale), and between civic culture and government performance. To examine links with personality, we initially use Rice and Sumberg’s scale, followed by tests employing updated versions of some of the components of that scale.
Results
Findings are reported in four stages. Initially, we explore possible links between personality and our three covariates. Next, we consider whether state-level personality and ideology are correlated. Analyses then turn to Elazar’s (1984) political culture typology and Rice and Sumberg’s (1997) representation of state-level civic culture.
Covariates
In all analyses to follow, we explore the influence of personality while accounting for the possible effects of minority diversity, white ethnic diversity, and state high school graduation rates (Hero 2007; Hero and Tolbert 1996). As a preface to those analyses, we first consider whether personality and these covariates are related. We do so by regressing each of the three control measures on the state-level Big Five variables. Results are reported in Table 1.
Personality and Correlates of Political Culture.
Cell entries for each column are ordinary least squares regression coefficients, with standard errors in parentheses. The column heading lists the dependent variable for each model.
p < .10. *p < .05. **p < .01. ***p < .001.
Coefficient estimates reveal substantial overlap between personality and the other variables, and especially between personality and white ethnic diversity. As to the Big Five measures, conscientiousness and neuroticism yield significant effects in all three models, whereas the other three trait dimensions are only related to white ethnic diversity. 13
These results bring likely implications for the analyses to follow. First, given the minimal overlap between openness, extraversion, and agreeableness and our covariates, relationships between these traits and the phenomena considered below most likely would be novel effects that operate over and above the influences of the covariates. Second, the opposite pattern is more likely for conscientiousness and neuroticism. Given that these traits are related to all three covariates, effects of these variables in models of state-level ideology and political culture likely would, to some extent, overlap with those of the two diversity measures and educational attainment. If this proves to be the case, such findings might suggest avenues for future research regarding possible causal paths. For instance, perhaps minority and ethnic diversity lead to variation in collective personality, which, in turn, shapes ideology and culture.
Ideology
In individual-level research, identification of a link between openness and ideological liberalism is common, as is a parallel relationship between conscientiousness and conservatism. Modest relationships also often are detected between neuroticism and conservatism, and between agreeableness and liberalism. Although individual-level findings need not be matched in aggregate-level analyses, these past results provide clear guidance for the present inquiry.
Table 2 reports results for models in which the 2005 version of the Berry et al. (1998) state ideology measure and state ideology as represented on the 2006 CCES are regressed on the measures of state personality, then on both the personality measures and the three controls. To reiterate, our modeling strategy employs regression techniques not because we view these specifications in cause-and-effect terms but rather as a means to account for collinearity among the Big Five and our other predictors.
State-Level Personality and Ideology.
Cell entries are ordinary least squares regression coefficients, with standard errors in parentheses. The dependent variable in the first two models is the 2005 version of the Berry et al. (1998) measure of citizen ideology in the American states. Observed values range from 25.24 (most conservative) to 91.85 (most liberal). The dependent variable in the third and fourth models is ideology as measured on the 2006 CCES; data from 5-point and 101-point ideology measures from more than 32,000 respondents are aggregated by state. Observed values range from 5.07 (most conservative) to 6.14 (most liberal). Personality data are state means derived from Internet samples of more than 600,000 respondents (see Rentfrow, Gosling, and Potter 2008), recoded to range from 0 (lowest observed value on each trait dimension) to 1 (highest observed value on each trait dimension). CCES = Cooperative Congressional Election Study.
p < .10. *p < .05. **p < .01. ***p < .001.
Starting with the models that include only personality, we see that two of the Big Five measures, openness and conscientiousness, produce significant coefficients in both initial models in Table 2. Both effects for extraversion (p < .09) approach significance. Because the personality variables have been recoded to a 0 (lowest observed value) to 1 (highest observed value) scale, their substantive effects are easily interpreted. For openness, the 34.46 coefficient in the first model means that an increase in openness from the lowest to the highest observed value corresponds with a 34-point rise in state liberalism. The ideology measure has an observed range of only 61 points. In the second model, there is a 0.87-point rise in liberalism across openness, whereas the observed range on this ideology measure is 1.07 points. In light of past individual-level findings of strong relationships between two traits, openness and conscientiousness, and ideology, it is reassuring that the largest collective-level effects emerge for these same traits. It is all the more reassuring that the pattern of results is virtually identical using two different measures of state ideology.
These findings help validate the collective-level personality measures. Were openness and conscientiousness unrelated to state-level ideology, concern would exist as to whether the measures capture anything relevant to political phenomena. Absent such evidence, if null results were found in tests involving state political culture, it would be difficult to gauge whether they speak to an actual lack of substantive relationships or potential limitations in the measures of collective personality. With significant links between state-level personality and ideology having been established, we can be confident that the personality measures support valid inferences.
As expected, including all eight predictors at once has virtually no effect on results for openness, with that personality trait dimension continuing to maintain significant relationships to ideology. In contrast, the effect for conscientiousness is muted once the controls are added. This pattern, which will be repeated below, highlights the potential value in further inquiry on the interrelationship between these constructs.
Elazar’s State Political Culture Typology
Our approach entails using two procedures for estimation of possible links between the Big Five and political culture. In one, we employ Sharkansky’s (1969) unidimensional measure. Sharkansky’s coding builds on Elazar’s, adding nuance via attention to regional differences within states. The original 1 to 9 scale is modified to permit fractional differences, to two places past the decimal. In actuality, the forty-eight states (the data omit Alaska and Hawaii) receive thirty-one different values between 1 and 9. Our second procedure acknowledges the multidimensionality of Elazar’s original framework. We code each state in terms of its dominant cultural component (0 = traditional, 1 = moral, 2 = individual), and then estimate a multinomial logit model.
Results are reported in Table 3. Estimates in the first two columns are from OLS models using the Sharkansky data. In the first model, two Big Five variables, openness and extraversion, produce marginally significant effects and two more, conscientiousness and neuroticism, yield strong effects. Increased values on openness and extraversion correspond with a move toward a moralistic state culture, and a move away from a traditional culture. The opposite pattern emerges for conscientiousness and neuroticism. Together, these coefficients signal a pronounced link between state political culture and collective personality. For instance, the coefficient for neuroticism is nearly as large as the range of the Sharkansky measure, and the coefficient for conscientiousness exceeds that range. The model’s R2 value is an impressive .49.
Personality and the Elazar Political Culture Typology.
Standard errors are in parentheses. Coefficients for the first two models are OLS regression estimates where the dependent variable is Sharkansky’s (1969) linear recoding of the Elazar typology. Values on the dependent variable range from 1 (most moralistic) to 9 (most traditional), with fractional values to two places past the decimal. Coefficients for the remaining models are multinomial logit estimates from models using Elazar’s original typology, with all states coded as being primarily traditional, moral, or individualistic, with the latter serving as the reference category. “SC” entries indicate whether a variable makes a statistically significant (p < .05) contribution to the full model. Big Five data are state means derived from Internet samples of more than 600,000 respondents (see Rentfrow, Gosling, and Potter 2008), recoded to range from 0 (lowest observed value on each trait dimension) to 1 (highest observed value on each trait dimension). OLS = ordinary least squares.
p < .10. *p < .05. **p < .01. ***p < .001.
Two aspects of these correlations should be noted. First, skeptics have raised reasonable questions regarding whether Elazar’s typology captures anything tangible. The multiplicity of significant coefficients in Table 3’s initial model establishes that Elazar’s framework does, in fact, correspond with measurable attributes of states’ citizenries. Second, the magnitude of these correlations is noteworthy given that these results connect something that is nominally apolitical, collective personality, with a model expressly designed to represent state political culture.
The second column in Table 3 adds our covariates to the mix. The effects for openness and extraversion are virtually unchanged when all independent variables are included. In contrast, the coefficient for neuroticism drops by a third in magnitude, and the coefficient for conscientiousness drops by half. Collectively, these models demonstrate not only that variation in the Big Five traits across states corresponds with variation in the unidimensional variant of the Elazar political culture measure but also that two of these effects exhibit substantial overlap with relationships between our control variables and political culture.
The final four columns in Table 3 present data from multinomial logit models. An individualistic culture—the center value when the data are arrayed unidimensionally—serves as the reference category. Thus, coefficients in the first column for each model signal effects for the traditional versus individual distinction and coefficients in the second column are for moral versus individual. In a multinomial logit model, a χ2 statistic can be computed for each variable to gauge whether the variable adds to model performance as a whole; we use the superscript SC to identify significant variable χ2 statistics (p < .05). Note that the second model omits three variables—openness, extraversion, and minority diversity. When we attempted to estimate the full eight-variable model, it failed to converge. That outcome is unsurprising given that there were sixteen coefficient estimates in a model with only forty-eight observations. Openness and extraversion yielded insignificant effects in the first multinomial logit model, and minority diversity did the same when we regressed the dependent variable on only the three controls; thus, we opted to drop these predictors as a means to craft a reduced model with stable estimates.
Three of the Big Five variables exhibit strong relationships with political culture in these specifications. Those for conscientiousness and neuroticism are consistent with those seen in the OLS model. Compared with an individualistic culture, states with high means on these traits are more likely to have a traditional culture, and less likely to have a moralistic one. The consistency for these findings across different models is reassuring. However, there also are discrepancies between the multinomial logit and OLS results. First, the marginally significant relationships seen earlier for openness and extraversion are not repeated. Second, agreeableness, which produced a null effect in the OLS model, yields a substantial effect in the multinomial model. A look at the logit coefficients reveals why. Values on agreeableness correspond with cultural type, but not in a manner consistent with a unidimensional scale that moves from moralistic to individualistic to traditional. Instead, high levels of agreeableness correspond with moralistic cultures, ones in which the government is perceived as a positive force for the promotion of the public good. Furthermore, consistent with expectations, low levels of agreeableness correspond not with traditional cultures but instead with individualistic ones.
The substantive links between political culture and conscientiousness, agreeableness, and neuroticism are quite large. With other variables held at their means, results from the first multinomial logit model indicate that a state with the lowest observed level of conscientiousness has a 0.99 estimated likelihood of having a moral culture, whereas a state with the highest value has a 0.96 likelihood of having a traditional culture. Similar values are obtained for neuroticism: a 0.99 likelihood of a moral culture at low levels of neuroticism versus a 0.74 likelihood of a traditional culture (and a 0.26 likelihood of an individualistic culture) where neuroticism is high. Low agreeableness corresponds with a 0.98 likelihood of an individualistic culture versus a 0.91 likelihood of a moralistic culture when agreeableness reaches its highest value.
The addition of the three control variables brings interesting nuance to the account. In Table 3’s final multinomial logit model, all five of the retained variables contribute to model performance. However, the variables differ in the nature of their contributions. All three personality effects are statistically significant for the moral versus individualistic distinction, but insignificant for the contrast between traditional and individualistic. The opposite pattern is seen with graduation rates and white ethnic diversity, with both of these variables reaching statistical significance only for the traditional versus individualistic contrast.
Results in Table 3 reveal that, irrespective of how the Elazar political culture concept is operationalized, cultural types correspond with collective personality. Further examination of the link between personality and culture requires that we shift gears, and consider an alternate representation of culture—Rice and Sumberg’s (1997) measure of civic culture.
Rice and Sumberg’s Civic Culture Scale
Civic culture maps relatively closely to the Elazar description of moralistic culture, and thus personality correlations similar to those observed in Table 3 are expected. Confirmation of the presence of such relationships would add further evidence that collective personality corresponds with state political culture. This extension focuses on an alternate representation of political culture, adding breadth to the effort to specify links between culture and personality.
Table 4 presents OLS regression results for models that include civic culture and the state-level Big Five variables. In the first model, statistically significant effects emerge for conscientiousness and neuroticism, the same two personality variables that consistently produced significant effects in Table 3’s Elazar models. Current results reveal significant negative relationships between these trait dimensions and state-level civic culture. However, when we add control variables in the third model, the effects for conscientiousness and neuroticism wash out. Hence, unlike with the Elazar framework, the relationship between personality and the Rice–Sumberg civic culture measure appears to fully overlap with the effects of established correlates.
Personality and the Rice–Sumberg Measure of State Civic Culture.
Cell entries are ordinary least squares regression coefficients, with standard errors in parentheses. The dependent variable is Rice and Sumberg’s (1997) measure of civic culture in the American states. Observed values range from −1.53 to 1.52. Big Five data are state means derived from Internet samples of more than 600,000 respondents (see Rentfrow, Gosling, and Potter 2008), recoded to range from 0 (lowest observed value on each trait dimension) to 1 (highest observed value on each trait dimension).
p < .10. *p < .05. **p < .01. ***p < .001.
As a follow-up, we consider whether subcomponents of civic culture exhibit different patterns of relationships with the Big Five measures. Once controls are introduced, personality is unrelated to the full Rice–Sumberg scale, but these null results may mask more specific effects, ones involving individual components of civic culture. The Rice–Sumberg scale integrates data on four phenomena: civic engagement; political equality; solidarity, trust, and tolerance; and social structures of cooperation. Rice and Sumberg (1997) used multiple objective indicators of three of these four, the exception being social structures, which is measured only with data on nonprofit organizations per capita. We partly replicate their approach by using circa 2005 data on nonprofit organizations (per 10,000 state residents), along with two objective indicators for each of the other three components: Internet access and library book circulation (civic engagement), income inequality and the percentage of state legislators who are women (political equality), and violent crime and lawyers per 10,000 residents (solidarity, trust, and tolerance). For most of these, earlier versions of these measures were employed by Rice and Sumberg. One exception is that we replaced the earlier measure of newspaper circulation with data on Internet access. 14
Results for models focused on the components of civic culture are presented in Table 5. To conserve space, we include only the full eight-variable models. All of the Big Five variables except agreeableness produce significant effects, indicating that personality is correlated with the subcomponents that contribute to state civic culture. Across the seven tests, openness yields four effects that are at least marginally significant (p < .10). However, these effects conflict in terms of their overall relationship to civic culture. The positive effects for Internet access and women in office correspond with higher levels of civic culture, but links between openness and income inequality and an abundance of lawyers move in the opposite direction. Thus, openness corresponds with facets of civic culture, but, as revealed in Table 4, is unrelated to civic culture at the summary level because the subcomponent-level effects offset one another.
Personality and State-Level Indicators of Civic Culture.
Cell entries for each column are ordinary least squares regression coefficients, with standard errors in parentheses. Internet use is the percentage of state residents with Internet access at home, 2009. Library book circulation is the total library circulation per capita, 2005. Income inequality is the ratio of the average income of the top 5% of families to the average income of the bottom 20%, 2006. Women in state legislatures is the percentage of all state legislators who are women, 2005. Violent crime is the number of violent crimes per 100,000 people, 2005. Lawyers per 10,000 is the number of lawyers in the state per 10,000 residents, 2007. Nonprofit organizations is the number of registered nonprofit organizations per 10,000 people, 2008.
p < .10. *p < .05. **p < .01. ***p < .001.
In Table 4, conscientiousness and neuroticism were shown to be negatively related with state civic culture. Results in Table 5 add nuance to those basic findings. High levels of conscientiousness tend to be found in states with low levels of Internet access and high rates of violent crime. Neuroticism is also associated with low levels of Internet access, and also library book circulation, along with the election of few women to state legislatures. Thus, unlike with openness, the cumulative force of the component associations for conscientiousness and neuroticism results in strong overall associations with civic culture.
As to the control variables, the greatest number of significant effects is found for the measure of high school graduation rates, with significant coefficients found for variables related to all four components of civic culture. The effects of minority diversity and white ethnic diversity are more narrow. The latter is related only to Internet access and lawyers per capita. Minority diversity is very strongly related to both measures of the solidarity, trust, and tolerance component of civic culture. Specifically, greater minority diversity is associated with higher rates of violent crime and more lawyers per capita.
Revisiting the Question of Process
Rather than offering definitive answers to questions of causal process, our purpose has been to highlight empirical patterns that may prompt renewed attention to political culture and its correlates. Findings reported thus far raise complex questions, questions for which concrete answers are not yet available. We end with a brief discussion that we hope will be illustrative. Specifically, we speculate on the individual-level and social dynamics that may have given rise to some of the findings reported above. We do this first with respect to our core results regarding correlations between personality and political culture. We then expand on this discussion by briefly contemplating the roles of our three control variables.
We focus on one example, the relationship between personality and violent crime seen in Table 5. There, a strong positive link was observed between state-level conscientiousness and crime. At question is what may have caused this relationship. Addressing this question, even on a speculative level, helps to highlight the complexity of the forces in operation.
One possibility is that personality exerts a direct causal force. Where more people possess a given personality profile, concordant behavior should be expected. Unfortunately, this account breaks down for conscientiousness and crime. Conscientious individuals are hardworking, punctual, and responsible, yet states with high levels of collective conscientiousness also have high crime rates. Unless these states are populated by punctual, hardworking criminals, a direct causal effect is unlikely. A second possibility is that crime prompts dynamics that alter state-level patterns in personality. Suppose, for example, that a state has experienced a long period in which violent crime was prevalent. High crime might cue calls for greater personal responsibility and for efforts such as the formation of neighborhood watches. In short, the occurrence of crime may highlight the importance of conscientiousness to a state’s noncriminals.
These first two scenarios focus on violent crime in itself rather than contemplate its link to civic culture. Rice and Sumberg (1997) used crime rates as an indicator of a relative dearth of civic culture. Hence, a third possibility is that there is a complex dynamic at work, one linking civic culture, personality, and crime. For instance, where conscientiousness is high, preferences will be against social liberalism and a large, active government. As a result, public spending on education and youth social and recreation programs may be low, which, in turn, may nudge more individuals toward crime. That crime may, in turn, trigger even less spending on education and social programs (what money there is would now be devoted more toward law enforcement), and may prompt patterns in migration that further entrench the state’s personality profile.
Matters become even more complicated once we factor in the effects of our three control variables. This is demonstrated by adding minority diversity to our discussion of personality and crime. At the state level, conscientiousness, minority diversity, and violent crime are positively correlated with one another. If we set aside the possibility of reciprocal relationships, we might surmise that one of these three variables is the base cause, a second plays a mediating role, and the third is the effect. But which variable holds which position in this causal chain? With only a few moments’ thought, one easily can devise plausible scenarios in which, in turn, each variable is assigned any of the three causal positions. Furthermore, we know from Table 5 that any mediation is incomplete in that conscientiousness and minority diversity both retain significant correlations with violent crime when the effects are estimated simultaneously.
Our objective in highlighting these interrelationships is twofold. First, we believe the relationships are complex, and thus, researchers should tread cautiously when speculating about causal paths. But second, efforts to sort out causal patterns are vital. Identification of covariance is a key first step toward improving our understanding of political culture, but it is nonetheless merely a first step. Careful development of theory, construction of creative research designs, and the acquisition of multifaceted longitudinal data all likely will be needed if we ultimately are to devise comprehensive accounts of political culture and its antecedents and consequences.
Conclusion
Historically, most of the contributions of research on psychological processes have come in improved understanding of the individual actor. However, as noted at the outset of the present study, it is also important that research on the individual contemplates the collective. This is particularly vital for inquiry concerning political processes. After all, it is when individuals come together that causes are advanced, elections are won, and policies are formed.
In recent years, scholars in the areas of cross-cultural and personality psychology have taken up the challenge of connecting individual-level processes to collective-level phenomena. In a short span of time, this research has made tremendous progress in identifying patterns of geographic variation, cross-national and subnational, in collective personality, and in linking aggregate personality with a wide array of variables. This work has called attention to the existence and likely significance of geographic variation in personality, and it has generated insightful and intriguing questions regarding the processes that are in operation. The present study has built on these efforts in two manners. First, whereas most recent research on collective personality has examined cross-national patterns, we have considered subnational variation in personality via focus on collective-level measures of the Big Five trait dimensions in the fifty American states. Second, this study’s main objective has been to determine whether variation in collective personality corresponds with indicators of state politics and political culture.
The tests conducted here reveal strong relationships between the Big Five and a host of politically relevant variables. These include state-level ideology and political culture, and also civic culture and its component indicators. These results are striking in at least two manners. First, personality stands as a micro-level property, yet present results establish its connection to a variety of broad subnational factors. Second, collective personality levels are seemingly apolitical, at least on their surface, but variation in personality across the American states corresponds quite strongly with states’ core political characteristics.
For students of cross-cultural psychology and personality, these results contribute further evidence of meaningful subnational geographic patterns in personality, along with the first clear demonstration that collective personality corresponds with representations of political culture. For researchers in political psychology, this study offers stark evidence of the promise inherent in moving beyond exclusively individual-level inquiry when studying personality.
Current results establish what has been an elusive connection, a link between cultural typologies and attributes of the states’ citizenries. We have emphasized that the relationship between personality and political culture raises intriguing questions of process. To the extent that personality influences behavior, political culture may in part be a product of citizens’ nominally apolitical psychological dispositions. Conversely, if enculturation and selective migration occur, then political culture and its corresponding processes may produce effects well beyond the political realm. Even modest enculturation and migration may be sufficient over time to entrench differences between the states not only in politics but also in collective personality. For instance, if a few aficionados of moralistic culture find themselves out of place in Alabama, and trade places with traditionalists who are seeking to leave Vermont, the result may be a solidification of each state’s political culture, along with a strengthening of its collective personality.
At this point, it seems most likely that the correspondence between personality and political culture occurs as the consequence of multiple interrelated processes, quite possibly involving the influence of personality on behavior and the reciprocal influence of culture and its attendant norms on the distribution and expression of personality. And, as we have seen, the plot thickens when we contemplate the roles of other covariates, factors such as minority and ethnic diversity. For now, what we can state definitively is that collective personality and key facets of political culture are related. The existence and especially the impressive strength of those relationships signal the importance of efforts to determine what mechanisms drive those connections, and with what likely implications.
Footnotes
Acknowledgements
We wish to thank Peter Rentfrow and Sam Gosling for sharing their data on personality in the American states. We also thank this journal’s anonymous reviewers for their instructive comments. Damarys Canache acknowledges the assistance of the College of Liberal Arts and Sciences at the University of Illinois, which provided her with release time via a "Study in a Second Discipline" award to support this research.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.