Political Information, Political Involvement, and Reliance on Ideology in Political Evaluation

Polit Behav (2013) 35:89 112 DOI 10.1007/s11109-011-9184-7 ORIGINAL PAPER Political Information, Political Involvement, and Reliance on Ideology in Political Evaluation Christopher M. Federico Corrie V. Hunt Published online: 30 October 2011 Ó Springer Science+Business Media, LLC 2011 Abstract Many studies have focused on the relationship between political information and the use of ideology. Here, we argue that two evaluative motivations general investment of the self in politics and extremity of partisanship serve as moderators of this relationship. Specifically, we use data from two recent national surveys to test whether the possession of information is more strongly associated with a tendency to approach politics in an ideological fashion among individuals high in both types of evaluative motivation. Results supported this hypothesis, revealing that information was more strongly associated with ideological constraint and with a tendency to give polarized evaluations of conservatives and liberals among those who highly invest the self in politics and those with more extreme partisanship. As such, this study suggests that information and involvement interact to shape the use of ideology. Keywords Ideology Political expertise Political involvement Despite the centrality of philosophical concepts like liberalism and conservatism to mainstream political discourse, modern public opinion research has generally concluded that most citizens are unable to effectively use these concepts when Electronic supplementary material The online version of this article (doi:10.1007/s11109-011-9184-7) contains supplementary material, which is available to authorized users. C. M. Federico (&) Departments of Psychology and Political Science, University of Minnesota, Twin Cities, 75 East River Road, Minneapolis, MN 55455, USA e-mail: federico@umn.edu C. V. Hunt Hart Research Associates, 1724 Connecticut Avenue, NW, Washington, DC 20009, USA

90 Polit Behav (2013) 35:89 112 making political judgments (Converse 1964). Nevertheless, those who possess a high level of political sophistication in the form of factual political knowledge, interest, and involvement in political affairs, and so on are more likely to structure their attitudes and perceptions in ideological terms (Delli et al. 1996; Judd and Krosnick 1989; Zaller 1992). But how do the various components of sophistication contribute to the use of ideology? In this article, we argue that the use of ideology should be understood in terms of an interaction between two factors that are often lumped together under the general heading of sophistication: political information and the motivation to use that information in an evaluative fashion. Specifically, we suggest that increased involvement in the form of general interest or investment of the self in politics and extremity of partisanship should motivate people to use political information to approach politics in an ideological fashion, leading to a stronger relationship between information and ideological coherence in one s political judgments among those highly invested in politics and those with extreme partisan orientations. Information and the Use of Ideology The topic of how ideological the general public is has captivated researchers ever since Campbell et al. (1960) found that most American citizens failed to conceptualize politics in ideological or even near-ideological terms. Subsequent research has established that the attitudes of a large percentage of the population also do not show evidence of ideological constraint, i.e., consistency between where they identify on the left right spectrum and their attitudes toward specific issues (vertical-constraint) and ideological consistency among attitudes toward various political issues (horizontal constraint; Bennett 2006; Converse 1964, 2000; Judd and Krosnick 1989; Kinder and Sears 1985; Zaller 1992). However, individuals who are high in political sophistication, engagement, or awareness catch-all terms encompassing the possession of political information, interest in politics, and even general cognitive ability seem to represent the exception (Converse 1964; Zaller 1992). In this vein, a wealth of research has demonstrated that various indices of sophistication are associated with an increased tendency to think about politics in ideological terms (Campbell et al. 1960; Judd and Krosnick 1989; Luskin 1990), greater attitude constraint and stability (Abramowitz 2010; Abramowitz and Saunders 2008; Bennett 1989, 2006; Hamill et al. 1985; Jacoby 1989, 1991, 1995; Kinder and Sears 1985; Stimson 1975; Zaller 1992), a greater tendency to evaluate liberal and conservative objects in an opposed fashion (Federico 2007), and greater ideological extremity (Federico 2004; Sidanius and Lau 1989). Finally, by contributing to belief-system consistency in these ways, sophistication also plays a key role in the emergence of political polarization at a more general level. In this vein, sophistication is associated with a stronger linkage between ideological self-placement and partisanship (Abramowitz 2010; Abramowitz and Saunders 1998, 2008) and greater ideological polarization in issue attitudes as a function of partisanship (Abramowitz 2010).

Polit Behav (2013) 35:89 112 91 But which sophistication-related variables matter the most in this regard? Many lines of work point toward information about key political figures, processes, and institutions as a particularly crucial variable (Delli Carpini and Keeter 1996). More generally, a variety of research suggests that political information strengthens the impact of various predispositions, group memberships, and background characteristics on policy judgments, voting, and even judgments about reality (Achen and Bartels 2006; Althaus 1998, 2001; Bartels 1996; Delli Carpini and Keeter 1996; Gilens 2001; see also Abramowitz 2010). This also appears to be the case with respect to the use of ideology. In this vein, studies suggest that factual political information is usually the strongest predictor of constraint and other variables indicative of strong ideological understanding (e.g., the speed with which ideologically-relevant judgments are made; see Fiske et al. 1990). Thus, information has moved to the forefront in studies of the relationship between sophistication and the use of ideology in political science (Federico and Schneider 2007; Zaller 1992). The Role of Evaluative Motivation Despite this evidence, we argue that it is unwise to think of the relationship between sophistication and the use of ideology solely in terms of information. As noted previously, research has connected sophistication-related variables other than information to outcomes like constraint. For example, Jennings (1992) noted large differences in ideological constraint between elite and mass samples that differed extensively in political involvement; he also found significant but smaller differences in constraint among segments of the mass public differing in involvement. Other studies have found similar effects for involvement-related measures like interest in politics (e.g., Baldassarri and Gelman 2008; Campbell et al. 1960; Jacoby 1995; Judd et al. 1981; Luskin 1990). Moreover, information is often lumped together with other indices of sophistication such as interest in politics to form composite measures of sophistication (e.g., Abramowitz 2010; Luskin 1990; McClosky and Zaller 1984; Stimson 1975). Finally, information is itself correlated with these other aspects of sophistication and often used as a generalized proxy for them, raising the possibility that the effects of information may be conflated with the effects of other dimensions of political sophistication (Zaller 1992). So, how do we unpack these various aspects of sophistication and their respective consequences for the use of ideology in mass publics? Ultimately, we argue that the sophistication needs to be understood in terms of two things: information and the motivation to use that information in specific ways. Like other perspectives, our approach attributes particular significance to the role of information: To use ideology, citizens need to possess well-developed knowledge structures containing information about key political ideas. However, we contend that even if people possess the information needed to understand the world ideologically, they also need to be motivated to use that information to approach politics in a highly ideological way. Specifically, the relationship between information and reliance on ideology should depend on motivational variables that strengthen people s tendency

92 Polit Behav (2013) 35:89 112 to evaluate people and things as good or bad. Because it provides an overarching rubric for the evaluation of many different objects (e.g., Hagner and Pierce 1982; Jacoby 1991), information about the content of different ideological positions should be of greater importance to people motivated to make evaluative judgments about the world. Sophistication-related variables other than the central construct of information may provide this motive to evaluate. Perhaps the most important of these is personal involvement or interest in politics. By personal involvement, we refer to the extent to which a person believes that an entity or object here, the domain of politics is important and relevant to the self (Boninger et al 1995). Researchers have identified interest as the motivational dimension of political sophistication i.e., as a chronically accessible goal that encourages individuals to devote cognitive resources to political judgment as opposed to something else (Delli Carpini and Keeter 1996; Luskin 1990) and which increases their desire to attend to, think more seriously about, and establish more confidence in judgments about politics (see Eagly and Chaiken 1993; Petty and Cacioppo 1979; see also Lavine et al. 2000). Personal involvement in the form of investment of the self in politics may thus provide a domain-specific motive to evaluate objects. Other sophistication-related variables may also provide citizens with a strong motive to evaluate in the political context. In particular, extremity of partisanship which covaries notably with information and interest (e.g., Abramowitz 2010; Sidanius and Lau 1989) may also do so. Specifically, due to a relatively strong alignment of the self with a salient political grouping and its fate, extreme partisans are particularly motivated to think about politics and to engage in political evaluation (often with a bias toward the formation of opinions based on their own party s ideology; Campbell et al. 1960; Cohen 2003; Goren et al. 2009; Green et al. 2002; Greene 1999; see also Boninger et al. 1995; Eagly and Chaiken 1993). Putting all of this together, we argue that these two sophistication-related sources of evaluative motivation general investment of the self in politics and extremity of partisanship should be associated with a stronger relationship between political information and the use of ideology. That is, the extent to which people approach politics in an ideological fashion should depend both on the possession of adequate information structures and various motivations to use those structures for evaluative purposes. People may possess a lot of political information, but if they are not interested or engaged, then they should not be motivated to rely on ideology. 1 Put another way, in our framework, information provides citizens with the potential to 1 It may be hard to imagine a person who knows about politics but who does not feel personally invested in politics. Nevertheless, we argue that many citizens acquire substantial information simply by being well-educated, interacting with politically informed individuals, or paying attention to the news. Despite not exerting any interested effort on their part, people can learn about politics by fulfilling general requirements in high school and college. Delli Carpini and Keeter (1996) did indeed find that people who were more educated also tended to know more about politics, controlling for other factors (including involvement). Receiving a quality education does not necessarily mean that people will also become engaged or interested in politics. People may remain uninterested in politics for a variety of reasons, including simply finding it boring or being turned off by the partisan bickering. Without any sort of motivation to think about political issues in an ideological fashion, we argue that political information will not translate as strongly into the use of ideology in political judgment.

Polit Behav (2013) 35:89 112 93 be ideological, whereas investment of the self in politics and partisan extremity provide people with the desire to use the information that they already possess to approach politics and political judgments in an ideological way. Prior Work on Information and Evaluative Motivation Unfortunately, existing work has done little to disentangle information from the sophistication-related motivations to use it in an evaluative fashion. As noted previously, information generally correlates strongly with other dimensions of sophistication that have similar main effects on the use of ideology, including specifically motivational ones; often, the informational and motivational indices are combined into a single measure as well. Moreover, while a number of studies have included measures of both information and domain-specific evaluative motivation (e.g., political interest) in models predicting reliance on ideology in political judgment (e.g., Jacoby 1995; Luskin 1990), none of these studies have examined how information and evaluative motivation interact or more specifically, how motivation may moderate the effect of information. 2 Nevertheless, some evidence that we are on the right track with this interactive model comes from studies that examine sources of evaluative motivation that fall outside the conceptual domain of political sophistication and outside the domain of politics itself. For example, Federico and his colleagues (Federico 2004, 2007; Federico and Schneider 2007) have used the need to evaluate scale (Bizer et al. 2004; Jarvis and Petty 1996) to tap into a non-political source of evaluative motivation. The need to evaluate is an individual-difference construct measuring the degree to which a person is generally motivated to evaluate experiences, objects, and ideas as good or bad (Jarvis and Petty 1996; see also Bizer et al. 2004; Federico 2004). These studies find that experts high in the need to evaluate show higher levels of ideological constraint (Federico and Schneider 2007) that the need to evaluate strengthens the relationship between information and the tendency to evaluate opposing ideological groups, political candidates, and parties in an affectively opposed fashion (Federico 2007). They also suggest that the relationship between political information and how extremely liberal or conservative people see themselves is stronger among those with a high need to evaluate (Federico 2004) and that experts are more likely to conceptualize partisan differences in ideological terms when they are also high in the need to evaluate (Federico and Schneider 2007). 2 Luskin (1990) does include measures related to information and motivation as predictors of ideological sophistication in a multiplicative form model, which makes the effect of any one predictor conditional on all others. Nevertheless, this general multiplicative form does not allow one to identify specific moderating effects of a particular variable on the relationship between two other particular variables. Moreover, Luskin s (1990) measure of ideological sophistication fails to include a key element of the use of ideology (i.e., constraint), while including measures which can be plausibly regarded as measures of information as well as reliance on ideology.

94 Polit Behav (2013) 35:89 112 Overview In this study, we explore the possibility that two things commonly lumped together under the general heading of political sophistication political information and the motivation to use that information in an evaluative fashion interact to shape the use of ideology. Specifically, we examine the hypothesis that investment of the self in politics and partisan extremity should moderate the relationship between political information and reliance on ideology in political judgment. We test this hypothesis using data from two recent surveys of American adults: the 2004 American National Election Study (ANES) and the 2008 Information, Motivation, and Ideology Study (IMIS). We operationalize the second of our two evaluative motivations partisan extremity in the same way across the two datasets, whereas we use slightly different measures of investment of the self in politics in each study in an effort to provide convergent support for our hypothesis. While we measure investment of the self in politics using standard measures of interest in politics in the 2004 ANES, we use a more direct measure of the extent to which respondents political beliefs were central to their sense of self in the 2008 IMIS. Moreover, across the surveys, we use multiple measures of the use of ideology, including measures of ideological constraint, and an additional measure of ideological bipolarity, or the extent to which the respondent correctly evaluated conservatives and liberals as opposites. Data and Methods Datasets As noted above, our analyzes relied on the 2004 ANES and a national survey of our own construction, the 2008 IMIS. These datasets were chosen for their recency and their inclusion of all the measures we needed to test our hypotheses. 3 The 2004 ANES interviewed N = 1,212 respondents before the 2004 elections using a nationally representative sample; all interviews were carried out on a face-to-face basis. The pre-election interviews produced a response rate (RR1) of 66.1%. Following the election, 1,066 of the pre-election respondents were re-interviewed, producing a RR1 of 88%. Since some of our key variables were taken from the postelection survey, only those N = 1,066 individuals who completed both waves were used in the analyzes. The 2008 IMIS interviewed N = 1,511 respondents during the fall of 2008. The survey was conducted by Knowledge Networks, Inc. (KN) using their web-enabled panel. In order to reach a nationally representative sample, KN chooses potential panel respondents through a scientific probability sample initially contacted via random-digit dialing telephone interviews. Adults successfully contacted in this way are invited to participate in the KN web panel. If they agree, panel members are provided with a WebTV interface and free internet access in 3 Since data from the full set of factual-knowledge items for the 2008 ANES have not been coded or released as of this writing (http://www.electionstudies.org/studypages/2008prepost/2008prepost.htm), we rely on the less-recent 2004 ANES.

Polit Behav (2013) 35:89 112 95 return for completing a weekly survey (for representativeness evidence, see Chang and Krosnick 2002; Huggins and Eyerman 2001). The IMIS survey used a probability sample of all panel members 18 years of age or older. Among panel members randomly selected for the IMIS, 65.7% completed the survey. However, considering the rate at which households were recruited for the web panel (20%) and the rate at which at least one individual in each household completed an overall profile survey (54.5%), the final cumulative RR1 for the IMIS was 7.2%. 4 As described below, various ANES and IMIS items were combined into scales in order to operationalize our key variables; details are provided in the Appendix. Measures from the 2004 ANES Dependent Variables In the 2004 ANES, we used two different indices of the use of ideology as our dependent variables. Our first dependent variable was a measure of composite ideological constraint focused on the extent to which respondents displayed ideologically consistent issue positions. This is perhaps the most commonly used index of the use of ideology (Converse 2000; Federico and Schneider 2007). As noted above, ideological consistency is thought to consist of two components: vertical-constraint, or ideological agreement between one s general left right selfplacement and issue attitudes; and horizontal constraint, or ideological agreement between attitudes toward different issues. The two are related in the sense that horizontal constraint can be thought of as a consequence of the higher-order vertical linkage between issue positions and a central ideological self-placement; if an individual judges a greater number of issues in terms of their overall ideological identity, then attitudes toward those issues should be ideologically consistent as well (Converse 1964; Judd and Krosnick 1989; see also Stimson 2004). Thus, reliance on ideology in an individual s political judgments should be more evident to the extent that attitude structures display both of these features. In order to generate a summary measure of constraint that tapped both of these dimensions and overcame the measurement error associated with any single index of ideological constraint, we constructed a composite based on three separate indices. These indices were based on ten issue items, the post-election 7-point ANES ideology scale, and thermometer ratings of liberals and conservatives. For our measures, both the 7-point ideology item and the ten issue items were recoded to run from 0 to 1 with higher scores indicating greater conservatism. Two indices measured vertical-constraint. The first vertical-constraint measure was based on the average absolute distance between each of the ten recoded issue items and the midpoint of 0.5 on a composite ideology measure consisting of (1) responses to the recoded 7-point ideology scale and (2) the respondent s thermometer rating of conservatives minus their rating of liberals (recoded to run from 0 to 1). Since all individual items were recoded to run in the same ideological direction and have the 4 The cumulative response rate is computed by multiplying these three component rates together (i.e., 20 9 54.5 9 65.7%; see Callegaro and DiSogra 2008).

96 Polit Behav (2013) 35:89 112 same 0 1 scale, this served as a measure of the average proximity between an individual s general ideological self-placement and the ideological tenor of their issue positions. The second vertical-constraint measure was based on the proportion of the ten issue items where the individual responded on the same side of the issue as their overall position on the 7-point ideology scale. This served as a less strict indicator of vertical-constraint by focusing merely on the extent to which respondents adopted issue positions on the same side of the left right divide as their general ideological self-placement. Finally, the third index measured horizontal constraint by taking the standard deviation of each individual s responses to the ten recoded issue items and then reversing the score by multiplying by -1 (Barton and Parsons 1977). This provides an indication of the lack of variability across a respondent s issue positions, once the latter were placed on the same interval and ordered in the same ideological direction; to the extent that a respondent places herself in a similar left right position across issues, scores on this measure should be high. 5 All three indices were recoded to run from 0 to 1 and then averaged to form a scale (a = 0.67; M = 0.51, SD = 0.14). Our second dependent variable was a measure of ideological bipolarity. This index indicates the extent to which respondents evaluate conservatives and liberals in an opposed or bipolar fashion, e.g., by evaluating liberals negatively if they evaluate conservatives positively and vice versa. Higher levels of bipolarity indicate a greater understanding that liberalism and conservatism are conceptual opposites in political discourse and that an attraction to one should thus imply a corresponding distance from the other. In doing so, it also provides a measure of the degree to which individuals have polarized impressions of key ideological actors a pattern of perceptions that has been closely linked to ideological constraint and reliance on ideology in general in research on the ideological basis of contemporary political divisions (e.g., Abramowitz 2010). It was computed using the formula C L, where C is the conservative feeling thermometer and L is the liberal thermometer. Once computed, scores were then recoded to run from 0 to 1. Higher scores indicate a stronger tendency to give conservatives and liberals opposed evaluations (M = 0.23, SD = 0.24). 6 5 This measure can be constructed so as to assess consistency with respect to predefined liberal or conservative positions or consistency with respect to centrist positions defined by the sample mean on each item. The former measure is constructed in the fashion described above; the latter is constructed by standardizing all issue-attitude scores before computing standard deviations (Barton and Parsons 1977). Since we are interested in consistency with respect to the options offered by the items themselves which reflect the conservative and liberal options defined by elites we opt for the former operationalization. However, in both datasets, our results were identical when horizontal constraint was constructed using standardized issue scores. 6 This measure is similar to one used by Federico (2007), with two differences: (1) the present measure reverses the Federico measure so that higher scores indicate more rather than less bipolarity; and (2) the Federico measure includes an additional correction for the extent to which the respondent gave highly positive evaluations of both conservatives and liberals by subtracting the respondent s average evaluation of the two from the absolute difference used here. Our results were identical when the Federico index was substituted for our measure.

Polit Behav (2013) 35:89 112 97 Independent Variables Three key independent variables were considered. Political information was measured using seven factual-knowledge items (Delli Carpini and Keeter 1996; Fiske et al. 1990). Responses were coded on a correct/incorrect (0/1) basis and averaged to form a scale (a = 0.82; M = 0.46, SD = 0.32). In addition to information, two sophistication-related evaluative motivations were considered: investment of the self in politics and extremity of partisanship. As noted previously, investment of the self in politics was operationalized in the 2004 ANES in terms of general interest in politics. Interest was measured using six items tapping attention to public affairs and campaigns, political discussion, and political involvement. Responses to these items were coded on a no/yes (0/1) basis and averaged to form a scale (a = 0.77; M = 0.64, SD = 0.27). A measure of extremity of partisanship was constructed by folding the 7-point ANES pre-election party identification item at its pure independent midpoint. Like our other predictors, this measure was recoded to run from 0 to 1 (M = 0.61; SD = 0.33). Higher scores on these scales indicate higher levels of information, interest, and partisan extremity, respectively. Several control variables typically related to ideological consistency were also considered (Federico and Schneider 2007). Need for cognition a general tendency to engage in and enjoy effortful cognitive endeavors (Cacioppo et al. 1996) was assessed using two items developed by Bizer et al. (2004) for use in the ANES. Responses to these items were averaged; higher scores indicated a higher need for cognition (a = 0.63; M = 0.58, SD = 0.35). Need to evaluate was assessed using Bizer et al. s (2004) two-item version of the measure. Responses were averaged to form a scale, with higher scores indicating a higher need to evaluate (a = 0.64; M = 0.58, SD = 0.21). Earlier work suggests that completion of a college degree is critical in the development of complex attitude structures (Sniderman et al. 1991), so a dummy variable indicating whether respondents had completed a college degree (1 = yes) or not (0 = no) was also included. Finally, four demographics were also included in the analyzes: age (in years), income (in thousands of dollars per year), race (0 = nonwhite, 1 = white), and gender (0 = female, 1 = male). Measures from the 2008 IMIS Dependent Measure In the 2008 IMIS, we only had the measures needed for a single measure of composite ideological constraint. It was constructed from two indices: (1) one similar to the second vertical-constraint measure used in the 2004 ANES and (2) one similar to the horizontal constraint measure used in the 2004 ANES. These indices were constructed from eight issue items and a 7-point ideology scale identical to the one used in the 2004 ANES. Both were recoded to run from 0 to 1 and then averaged to form a scale (a = 0.62, M = 0.37, SD = 0.14). 7 7 While it was possible to construct all three constraint indices used in the 2004 ANES in the 2008 IMIS as well, we excluded the vertical-constraint index based on the average absolute distance between 7-point

98 Polit Behav (2013) 35:89 112 Independent Variables There were three of these in the 2008 IMIS. Political information was measured using eight factual-knowledge items similar to those used in the 2004 ANES. As before, responses were coded on a correct/incorrect (0/1) basis and averaged to form a scale (a = 0.65, M = 0.71, SD = 0.24). As in the ANES, the sophisticationrelated evaluative motivations considered in our analyzes were investment of the self in politics and extremity of partisanship. Investment of the self in politics was operationalized in the 2008 IMIS more directly, in line with the conceptual definition of personal involvement (Thomsen et al. 1995). Specifically, a measure of centrality of politics to the self was created using two Likert-type items adapted from the importance of-identity subscale of Luhtanen and Crocker s (1992) collective self-esteem scale (e.g., My political attitudes and beliefs are an important reflection of who I am ). Responses were recoded from 0 to 1 and averaged to form a scale (a = 0.84, M = 0.59, SD = 0.26). In turn, extremity of partisanship was operationalized by folding a 7-point party identification item identical to the one used in the 2004 ANES at its pure independent midpoint. Like our other predictors, this measure was recoded to run from 0 to 1 (M = 0.64; SD = 0.35). Higher scores on these scales indicate higher levels of information, centrality of politics, and partisan extremity, respectively. Finally, we considered a set of controls similar to those employed in the 2004 ANES. Need to evaluate was again assessed using Bizer et al. s (2004) two-item version of the measure, which was coded in the same way it was in the 2004 ANES (a = 0.66, M = 0.58, SD = 0.19). As an additional sophistication-related measure, a dummy variable indicating whether respondents had completed a college degree (1 = yes) or not (0 = no) was again included. Finally, four demographics were also included in the analyzes: age (in years), income (in thousands of dollars per year), race (0 = nonwhite, 1 = white), and gender (0 = female, 1 = male). Results Interest in Politics, Partisan Extremity, and Ideological Constraint in the 2004 ANES We begin by examining our hypothesis with respect to ideological constraint in the 2004 ANES. To test our hypothesis, we estimated a series of nested ordinary leastsquares regression models. In these models, composite ideological constraint was regressed on the controls, information, interest in politics, partisan extremity, and Footnote 7 continued ideological self-placement and individual issue positions. The latter index was excluded because it considerably lowered the reliability of the composite constraint measure. However, even when this lessreliable three-item measure was substituted for the two-item measure used in the primary IMIS analyzes, all of the critical interactions remained significant. Moreover, when the third omitted item was used on its own as a dependent variable, all of the critical interactions were significant. These analyzes are available on request.

Polit Behav (2013) 35:89 112 99 Table 1 Interest in Politics, partisan extremity, and the relationship between information and ideological constraint Composite ideological constraint Model 1 Model 2 Model 3 Predictor B SE b SE b SE Age -.001*** (.0002) -.001*** (.0002) -.001*** (.0002) Income.001 (.001).001 (.001).001 (.001) Race (1 = white).01 (.01).01 (.01).01 (.01) Gender (1 = male).01 (.01).01 (.01).01 (.01) College degree (1 = BA or more).04*** (.01).03** (.01).03*** (.01) Political information.11*** (.02).10*** (.02).11*** (.02) Need for cognition -.03* (.01) -.03* (.01) -.03* (.01) Need to evaluate.01 (.02).001 (.02).003 (.02) Interest in politics.08*** (.02).07*** (.02).08*** (.02) Partisan extremity.03** (.01).04** (.01).02 (.01) Information 9 interest in politics.21*** (.08) Information 9 partisan extremity.16*** (.05) Constant.50*** (.02).50*** (.02).50*** (.02) F (degrees of freedom) 19.45 (10, 917)*** 19.22 (11, 916)*** 19.31 (11, 916)*** Adjusted R 2.186.198.198 p \ 0.10; * p \ 0.05; ** p \ 0.01; *** p \ 0.001 Note. Entries are unstandardized OLS regression coefficients and HC3 robust standard errors (N = 928 for all models) the Information 9 Interest in Politics and the Information 9 Partisan Extremity interactions over multiple steps. To avoid unnecessary multicollinearity, each interaction was examined in a separate step. 8 In order to avoid heteroskedasticity problems, HC3 robust standard errors were used in all models (Long and Ervin 2000). Except for age and income, all independent variables were recoded on a 0 1 basis; information, interest in politics, need for cognition, and the need to evaluate were also centered prior to the computation of interaction terms. Table 1 summarizes the results for ideological constraint in the 2004 ANES. Model 1 examined only the effects of the first-order terms and controls. As the coefficients in the first column indicate, respondents with college degrees (p \ 0.001) and respondents with greater information (p \ 0.001), greater interest in politics (p \ 0.001), and more extreme partisan leanings (p \ 0.01) showed greater ideological constraint; older respondents showed a slight tendency toward 8 In order to highlight each, we report separate models for each interaction in the main text. However, in order to be sure that interest and partisan extremity both independently moderated the impact of information, we ran an additional model in which both the Information 9 Interest in Politics and the Information 9 Partisan Extremity interactions were included. The estimates from this model revealed a significant Information 9 Interest in Politics interaction (b =.16, p \.01) and a significant Information 9 Partisan Extremity interaction (b =.13, p \.01), with an adjusted R 2 of 0.204 for the full model. Complete results are available upon request from the authors.

100 Polit Behav (2013) 35:89 112 less constraint (p \ 0.01), as did respondents who were lower in the need for cognition (p \.05). In turn, Model 2 added the Information 9 Interest in Politics interaction to Model 1. The estimates for this model revealed a significant effect for the theoretically critical interaction (b = 0.21, p \ 0.001). To unpack the interaction, we computed conditional effects for the relationship between information and constraint at interest levels one standard deviation above (high interest) and below (low interest) the mean of the interest variable (Aiken and West 1991). These estimates indicated that the relationship between information and constraint was significant and positive among those high in interest (b = 0.15, p \ 0.001) but small and only marginal among those low in interest (b = 0.04, p \ 0.10). Since all variables have been recoded to run from 0 to 1, these estimates indicate that moving from the lowest to the highest level of information would produce a 15% increase in constraint among those high in interest, but only a 4% increase among those low in interest. Finally, Model 3 added the Information 9 Partisan Extremity interaction to the set of predictors in Model 1. The estimates for this model revealed a significant effect for the theoretically critical interaction (b = 0.16, p \ 0.001). To unpack the interaction, we computed conditional effects for the relationship between information and constraint at levels one standard deviation above (high extremity) and below (low extremity) the mean of partisan extremity. While information predicted greater constraint at both extremity levels, the conditional-effect estimates indicated that the positive relationship between information and constraint was stronger among those high in partisan extremity (b = 0.16, p \ 0.001) than it was among those low in partisan extremity (b = 0.11, p \ 0.001). In other words, moving from the lowest to the highest level of information would produce a 16% increase in constraint among those high in extremity, but a somewhat lower 11% increase among those low in extremity. Thus, our 2004 ANES results are supportive vis-a-vis ideological constraint. 9 Interest in Politics, Partisan Extremity, and Ideological Bipolarity in the 2004 ANES In the 2004 ANES, we also examined our key hypothesis with respect to ideological bipolarity in evaluations of liberals and conservatives. Since those with a strongly ideological approach to politics should have opposed views of these two political categories, this measure provided us with a second index of reliance on ideology. It also provided us with a look at the extent to which individuals had polarized views of the ideological order (e.g., Abramowitz 2010). In order to do this, we estimated Models 1 3 from Table 1, substituting our index of ideological bipolarity for the composite measure of ideological constraint as the dependent variable. Model estimation was otherwise identical. 10 Table 2 reports these results. Model 1 again 9 The Information 9 Centrality of Politics and the Information 9 Partisan Extremity interactions in Models 2 and 3 from Table 1 remain significant when the Information 9 Need to evaluate interaction (see Federico and Schneider 2007) is added to the models (both ps \ 0.01). 10 To be sure that interest and partisan extremity both independently moderated the impact of information on bipolarity, we again estimated a model in which both the Information 9 Interest in

Polit Behav (2013) 35:89 112 101 Table 2 Interest in politics, partisan extremity, and the relationship between information and ideological bipolarity Ideological bipolarity Model 1 Model 2 Model 3 Predictor B SE b SE b SE Age -.001* (.0004) -.001 (.0004) -.001 (.0004) Income.002 (.001).001 (.001).002 (.001) Race (1 = white).04** (.02).04** (.02).04* (.02) Gender (1 = male) -.001 (.01) -.01 (.01) -.003 (.01) College degree (1 = BA or more).01 (.02).004 (.02).01 (.02) Political information.16*** (.03).13*** (.03).15*** (.03) Need for cognition -.01* (.02) -.01 (.02) -.01 (.02) Need to evaluate.13*** (.04).11** (.04).12** (.04) Interest in politics.13*** (.03).12*** (.03).14*** (.03) Partisan extremity.15*** (.02).15*** (.02).12*** (.02) Information 9 interest in politics.46*** (.10) Information 9 partisan extremity.34*** (.08) Constant.18*** (.03).17*** (.03).17*** (.03) F (degrees of freedom) 21.19 (10, 868)*** 21.27 (11, 867)*** 21.71 (11, 867)*** Adjusted R 2.196.214.213 p \ 0.10; * p \ 0.05; ** p \ 0.01; *** p \ 0.001 Note. Entries are unstandardized OLS regression coefficients and HC3 robust standard errors (N = 879 for all models) included only the first-order terms and controls. As the estimates in the first column indicate, whites (p \ 0.01) and respondents with greater information (p \ 0.001), greater interest in politics (p \ 0.001), and more extreme partisan leanings (p \ 0.01) all showed a stronger tendency to give opposed or bipolar evaluations of liberals and conservatives. Older respondents and those who were lower in the need for cognition showed less ideological bipolarity (both ps \ 0.05). Model 2 again added the Information 9 Interest in Politics interaction to the Model 1 equation. These estimates revealed a significant effect for the interaction (b = 0.46, p \ 0.001). To unpack the interaction, we computed conditional effects for the relationship between information and the tendency to give liberals and conservatives opposed evaluations at points one standard deviation above (high interest) and below (low interest) the mean of the interest variable. These estimates indicated that the relationship between information and bipolarity was significant and positive among those high in interest (b = 0.25, p \ 0.001) but small and non-significant among those low in interest (b = 0.001, p [ 0.50). Put another way, these effects Footnote 10 continued Politics and the Information 9 Partisan Extremity interactions were included. The estimates from this model revealed a significant Information 9 Interest in Politics interaction (b = 0.37, p \ 0.001) and a significant Information 9 Partisan Extremity interaction (b = 0.27, p \ 0.001), with an adjusted R 2 of 0.223 for the full model. Complete results are available upon request from the authors.

102 Polit Behav (2013) 35:89 112 indicate that moving from the lowest to the highest information level would produce a 25% increase in constraint among those high in interest, but only a 0.1% increase among those low in interest. Finally, Model 3 added the Information 9 Partisan Extremity interaction to the set of predictors in Model 1. The estimates for this model revealed a significant interaction (b = 0.34, p \ 0.001). To follow-up on this, we calculated conditional effects for the relationship between information and bipolarity at levels one standard deviation above (high extremity) and below (low extremity) the mean of partisan extremity. While information was associated with greater bipolarity at both extremity levels, the conditional-effect estimates indicated that the positive relationship between information and bipolarity was stronger among those high in partisan extremity (b = 0.27, p \ 0.001) than it was among those low in partisan extremity (b = 0.15, p \ 0.001). 11 These estimates indicate that moving from the lowest to the highest level of information would produce a 27% increase in constraint among those high in extremity, compared to a 15% increase among those low in extremity. As such, these analyzes provide further support for our hypotheses about the moderating effects of interest and partisan extremity, and they extend our findings to an outcome more directly related to the broader phenomenon of ideological polarization. 12 Centrality of Politics to the Self and Partisan Extremity in the 2008 IMIS In an effort to replicate and extend the supportive ANES results, we turned to our second dataset the 2008 IMIS. This time, we operationalized investment of the self in politics in a more face-valid way: instead of relying on standard measures of political interest, we directly asked respondents several questions about how important their political beliefs were to their sense of who they were. Extremity of partisanship was again operationalized in terms of one s distance from pure 11 As with ideological constraint, the Information 9 Interest in Politics and the Information 9 Partisan Extremity interactions in Models 2 and 3 from Table 2 remain significant when the Information 9 Need to Evaluate interaction is added to the models (both ps \.001). 12 Since information was correlated with interest (r = 0.52, p \ 0.001) and partisan extremity (r = 0.19, p \ 0.001), we ran additional models that included the two quadratic terms for the variables involved in the interaction included in any given model to be sure that any significant interaction between them was not masking a significant curvilinear effect of either constituent variable (Ganzach 1997). In the Information 9 Interest in Politics models (parallel to Model 2), the key interaction remained significant with respect to constraint (b = 0.25, p \ 0.01) and bipolarity (b = 0.49, p \ 0.001). Similarly, in the Information 9 Partisan Extremity models (parallel to Model 3), the key interaction remained significant with respect to constraint (b = 0.21, p \ 0.001) and bipolarity (b = 0.29, p \ 0.001). We also looked at whether the portions of variance in interest and extremity that are independent of information moderate the relationship between information and each dependent variable. We first created a net interest variable by regressing interest on information and saving the residuals. These residuals were recoded to run from 0 to 1 and then used as the net interest variable. The same was done to create a net extremity variable. When each dependent variable was regressed on information, net interest, the Information 9 Net Interest interaction, and the controls, information and net interest significantly interacted to predict constraint (b = 0.27, p \ 0.001) and bipolarity (b = 0.45, p \ 0.001). Similarly, when each dependent variable was regressed on information, net extremity, the Information 9 Net Extremity interaction, and the controls, information and net extremity significantly interacted to predict constraint (b = 0.17, p \ 0.001) and bipolarity (b = 0.33, p \ 0.001).

Polit Behav (2013) 35:89 112 103 Table 3 Centrality of politics, partisan extremity, and the relationship between information and ideological constraint Composite ideological constraint Model 1 Model 2 Model 3 Predictor b SE b SE b SE Age -.001** (.0002) -.001*** (.0002) -.001*** (.0002) Income -.0003 (.001) -.0003 (.001) -.0003 (.001) Race (1 = white) -.03** (.01) -.02** (.01) -.03** (.01) Gender (1 = male) -.01* (.01) -.01* (.01) -.02 (.01) College degree (1 = BA or more).03*** (.01).02** (.01).03*** (.01) Political information.05*** (.02).07*** (.02).07*** (.02) Need to evaluate -.02 (.02) -.02 (.02) -.01 (.02) Centrality of politics.09*** (.02).09*** (.02).09*** (.02) Partisan extremity.05*** (.01).05*** (.01).05*** (.01) Information 9 centrality of politics.27*** (.05) Information 9 partisan extremity.22*** (.04) Constant.44*** (.02).44*** (.02).44*** (.02) F (degrees of freedom) 15.36 (9, 1477)*** 16.78 (10, 1476)*** 18.52 (10, 1476)*** Adjusted R 2.091.105.110 p \ 0.10; * p \.05; ** p \ 0.01; *** p \ 0.001 Note. Entries are unstandardized OLS regression coefficients and HC3 robust standard errors (N = 1487 for all models) independence on a standard ANES-style partisanship measure. In order to analyze the 2008 IMIS data, we estimated a set of regressions similar to those used in the ANES. In these models, composite ideological constraint was regressed on the controls, information, centrality of politics to the self, partisan extremity, and the Information 9 Centrality of Politics and the Information 9 Partisan Extremity interactions over multiple steps. Again, each interaction was examined in a separate step. 13 As before, HC3 robust standard errors were used in all models. Except for age and income, all independent variables were recoded on a 0 1 basis; information, centrality of politics, and the need to evaluate were also centered prior to the computation of interaction terms. Table 3 summarizes the results for the 2008 IMIS. The estimates for Model 1 which included only the first-order terms and controls indicate that respondents with college degrees, who had greater information, who saw politics as more central to the self, and who had more extreme partisan identifications all showed greater ideological constraint (all ps \ 0.001); older respondents (p \ 0.01), men 13 To ensure that centrality and partisan extremity both independently moderated the impact of information on bipolarity, we ran a model in which both the Information 9 Centrality of Politics and the Information 9 Partisan Extremity interactions were included. The estimates from this model revealed a significant Information 9 Centrality Politics interaction (b = 0.17, p \ 0.01) and a significant Information 9 Partisan Extremity interaction (b = 0.17, p \ 0.001), with an adjusted R 2 of 0.115 for the full model. Complete results are available upon request from the authors.

104 Polit Behav (2013) 35:89 112 (p \ 0.05), and whites (p \ 0.001) showed less constraint. Model 2 added the Information 9 Centrality of Politics interaction to the set of predictors in Model 1. As expected, the interaction was highly significant (b = 0.27, p \ 0.001). To probe the interaction, we computed conditional effects for the relationship between information and constraint at centrality levels one standard deviation above (high centrality) and below (low centrality) the mean of centrality of politics to the self. These estimates indicated that the positive relationship between information and constraint was notably stronger among those who saw politics as highly central to the self (b = 0.14, p \ 0.001) than it was among those did not see politics as central to the self, where the slope did not reach significance (b =-0.003, p [ 0.80). In terms of predicted changes, these estimates indicate that moving from the lowest to the highest level of information would produce a 14% increase in constraint among those high in centrality, while actually resulting in.3% decrease in constraint among those low in centrality. 14 Finally, Model 3 added the Information 9 Partisan Extremity interaction to the predictors included in Model 1. The estimates indicated a significant effect for the hypothesized interaction (b = 0.22, p \ 0.001). Again, we computed conditional effects for the relationship between information and constraint at levels one standard deviation above and below the mean of partisan extremity. These estimates indicated that the relationship between information and constraint was strong, significant, and positive among those high in partisan extremity (b = 0.14, p \ 0.001), but non-significant among those low in partisan extremity (b =-0.01, p [ 0.50). In other words, moving from the lowest to the highest level of information would produce a 14% increase in constraint among those high in partisan extremity, while actually resulting in a 1% decrease in the same among those low in extremity. In sum, the 2008 IMIS replicates and extends our ANES results. We found that centrality of politics to the self moderated the relationship between information and ideological constraint just like interest in politics did in the 2004 ANES; we also found that partisan extremity was again a significant moderator of the relationship between information and constraint. 15 14 Since information was significantly correlated with centrality of politics (r = 0.37, p \ 0.001) and partisan extremity (r = 0.19, p \ 0.001), we also ran additional models that included the two quadratic terms for the variables involved in the interaction included in any given model to be sure that any significant interaction was not conflated with significant curvilinear effects. In the Information 9 Centrality Politics model (parallel to Model 2), the key interaction remained significant (b = 0.18, p \ 0.01). Similarly, in the Information 9 Partisan Extremity model (parallel to Model 3), the key interaction remained significant (b = 0.18, p \ 0.001). As before, we also looked at whether the portions of variance in centrality and extremity that are independent of information moderate the relationship between information and ideological constraint. To do this, we created net centrality and net extremity variables using the same regression procedures employed in the 2004 ANES. When constraint was regressed on information, net centrality, net extremity, the Information 9 Net Centrality interaction, and the controls, the interaction was significant (b = 0.26, p \ 0.001). Similarly, when constraint was regressed on information, net centrality, net extremity, the Information 9 Net Extremity interaction, and the controls, the interaction was significant (b = 0.22, p \ 0.001). Thus, alternative analyzes again replicate the results of our main analyzes. 15 The key Information 9 Centrality of Politics and the Information 9 Partisan Extremity interactions in Models 2 and 3 from Table 3 remain significant when the Information 9 Need to Evaluate interaction is added to the models (both ps \ 0.01).