Traditional conceptualizations of ideology’s political functions—according to which individuals should derive their political preferences from internally elaborated “belief systems” (see Converse 1964) and engage in spatial comparisons to make electoral decisions (Downs 1957)—assume a sophisticated understanding of ideological abstractions that is, for the most part, not characteristic of the American public (Campbell et al. 1960; Converse 1964; Lewis-Beck et al. 2008; but see Achen 1975; Nie et al. 1976). Kinder (1983, p. 416) summarizes: “the political thinking of much of the public cannot be adequately described as ideological in the sense of deductive reasoning from an overarching set of integrated principles about politics and the social world.”

Nonetheless, ideology remains empirically relevant, even essential, to understanding political behavior. Ideology is one of the most consistent and powerful predictors of myriad political attitudes and behaviors, including vote choice (Jost 2006; Levitin and Miller 1979; Luttbeg and Gant 1985), candidate evaluations (Zaller 1992), policy preferences (Feldman 2003; Jacoby 1991), and party identification (Abramowitz and Saunders 1998; Levendusky 2009).

For over a half-century, scholars faced with this apparent disconnect have struggled to characterize ideology’s impact on political attitudes and behavior. One of the most significant efforts to advance this debate comes from a line of research that reconceptualizes ideological identification as a symbolic attachment to ideological groups not dependent upon, or even necessarily related to, an individual policy orientation. Primary evidence of “symbolic ideology” is found in studies of the “symbolic-operational paradox,” which examine Americans’ longstanding and overwhelming preference for conservative self-identification in spite of an aggregate preference for liberal policies (Ellis and Stimson 2012; Free and Cantril 1967). A similar pattern holds at the individual level; symbolic ideology (ideological self-identification) and operational ideology (policy preferences) are modestly correlated, “suggesting that they are, in fact, different concepts and should be analyzed separately” (Popp and Rudolph 2011, Footnote 5).

Symbolic ideology scholars primarily attribute ideological identification to the influence of social groups and messages. One of the best predictors of ideological self-identification is an individual’s evaluation of ideological groups, and those evaluations largely are based upon attitudes toward the social groups and symbols with which an ideological label is most closely associated, e.g. businessmen and the military for conservatives and feminists and ethnic minorities for liberals (Conover and Feldman 1981). In fact, studies indicate that conservative self-identification is primarily a reaction against liberalism and its associated social groups while hostility toward conservative groups has become an increasingly powerful predictor of liberal self-identification in recent years (Zschirnt 2011). Social influences figure prominently in scholars’ explanations of the symbolic-operational paradox, as well. Ellis and Stimson (2012) find that Americans’ overwhelming preference for conservative self-identification is due in large part to extra-political cultural connotations and patterns of elite discourse that favor the conservative label and disfavor the liberal label. Symbolic attachments are also shown to significantly influence expressed policy preferences, but their effects are contingent upon social cues from ideological group members; absent in-group cues, the relationship between ideological self-identification and policy preference is much weaker or not statistically significant (Malka and Lelkes 2010; Popp and Rudolph 2011).

Symbolic ideology research helps to clarify the apparent contradiction between evidence showing that the American public is lacking in ideological sophistication yet influenced in its political attitudes and behaviors by ideological identification. Contrary to the traditional conceptualization of ideology as, in its meaningful form, an internal application of abstract principles (see Campbell et al. 1960; Converse 1964), symbolic ideology embraces the notion that ideology derives much of its explanatory power from the influence of social factors giving meaning to ideological labels and direction to ideological group members. In short, it is an interaction of internal categorization and external cues that makes ideology relevant to the political behavior of the mass public. Moreover, symbolic ideology does not require a high level of political sophistication; attitudes toward, and applications of, ideological labels primarily are based upon symbolic evaluations, not elaborated policy orientations.Footnote 1

The achievements of symbolic ideology research recommend further theoretical and empirical development. A promising opportunity for such development, for reasons that I elaborate upon below, is found in Social Identity Theory (SIT). SIT is one of the most influential theories in the social sciences and in recent years it has assumed greater prominence in the political science literature, motivating studies of subjects as diverse as judicial behavior (Baum 2006), political intolerance (Gibson and Gouws 2000), political participation (Fowler and Kam 2007), national identity and patriotism (Althaus and Coe 2011; Huddy and Khatib 2007; Theiss-Morse 2009), and party identification (Green et al. 2005; Greene 1999; Iyengar et al. 2012; Weisberg and Hasecke 1999). By emphasizing the attitudinal and behavioral significance of internalized social group membership and external social group cues, SIT maps clearly onto the symbolic ideology literature’s reconceptualization of ideological identification. At the same time, the SIT literature includes many theoretical and empirical innovations relevant to understanding symbolic ideology, and ideology more generally, but never applied toward that end. In fact, no study in the symbolic ideology literature or elsewhere has directly measured and analyzed ideological identification as a social identity.

The purpose of this analysis is to establish a measure of ideological social identity (ISI), or psychological attachment to an ideological in-group, and use it to answer fundamental questions including: how common is ISI within the American public? Who is most, or least, likely to have a strong ISI? Does ISI vary across environmental contexts, and if so how? Is ISI relevant to understanding political behavior, particularly in relation to the influence of ideological identification? The value of such an analysis is twofold: first, evidence of meaningful and consequential ISI will further validate symbolic ideology research while also extending it theoretically and empirically; second, establishing the reality of ISI in the mass public is a necessary first step toward more fully exploiting the theoretical and empirical insights of the SIT literature in future studies of ideological identification, as detailed below.

Data for this analysis come from an original survey experiment conducted among a nationally representative participant sample. Included in the survey experiment are direct measures of ISI, as well as measures of relevant demographic and political characteristics, vote choice measures, and experimental manipulations designed to capture variation in ISI levels across different electoral contexts. I use these data to test a series of hypotheses addressing the prevalence, causes, and behavioral significance of ISI. The results of this analysis indicate that ISI is an empirically distinct component of ideological identification that is prevalent in the mass public and subject to variation, with respect to strength and underlying causes, across ideological groups and electoral contexts. Also, it is behaviorally consequential; ISI conditions the effect of ideological self-placement on vote choice in actual and hypothetical elections.

Theory

Henri Tajfel and colleagues developed SIT in order to better understand the nature and causes of inter-group relations. Through a series of experimental studies, they demonstrated that mere categorization within a social group, once internalized, induces high levels of in-group identification, inter-group differentiation, and in-group bias—even if the group is an arbitrary one and participants have no basis for perceiving common in-group interests (Billig and Tajfel 1973; Tajfel et al. 1971; see also Allen and Wilder 1975; Brewer and Silver 1978; Doise and Sinclair 1973). Experimental participants, in effect, were adopting a social identity, defined by Tajfel (1978, p. 63) as “that part of an individual’s self-concept which derives from his knowledge of his membership in a social group (or groups) together with the value and emotional significance attached to that membership.”

Foundational to SIT are two assumptions: first, individuals define themselves in large part by the groups to which they belong; second, individuals strive for positive self-evaluation and enhanced self-esteem (Tajfel et al. 1971; see also Brewer 1991; Turner et al. 1987). In order to positively evaluate oneself, then, an individual must also positively evaluate the in-group that in part defines the self. Thus, social identity will motivate an individual toward attitudes and behaviors that promote the in-group, perhaps but not necessarily at the expense of the out-group (Brewer 2002), and maintain in-group membership through compliance with in-group norms. In essence, social identity represents a process of depersonalization, “whereby people come to perceive themselves more as the interchangeable exemplars of a social category than as unique personalities” (Roccas and Brewer 2002, p. 50).

SIT has significant implications for understanding ideological identification. Typically, about half of all survey respondents identify themselves as liberals or conservatives, and two-thirds identify as moderates, liberals, or conservatives. To the extent that those identities are internalized, as suggested by respondents’ willingness to self-categorize, SIT research indicates that ideological group members will be psychologically motivated to positively evaluate the in-group, maintain in-group norms, and advance in-group interests.Footnote 2 Opportunities for ISI to influence political attitudes and behaviors are numerous and profound.

Take, for example, a party primary or general election in which a candidate is perceived to be at odds with an intra-party ideological faction: members of the opposing ideological in-group may be less willing to support that candidate—through voting, financial contributions, volunteering, or persuasion of other voters—for fear of weakening the ideological in-group, in the event that the out-group candidate succeeds, and in turn suffering negative personal consequences. One could imagine this explanation applying to conservative and Tea Party voters in the Republican Party primaries and presidential elections of 2008, 2010, and 2012. Or consider a policy debate in which party leaders promote a policy that conflicts with the views of a dominant intra-party ideological faction, such as Bill Clinton’s support for welfare reform in the 1990s or George W. Bush’s support for immigration reform in the 2000s. To the extent that prominent ideological in-group members, such as radio talk show hosts and interest group leaders, voice opposition, individuals might be motivated by ISI to oppose party leaders in order to increase the ideological in-group’s chances of successful opposition or at least avoid violating in-group norms and destabilizing their self-concept.

The symbolic ideology literature provides some basis for expecting such effects. Among other examples, Popp and Rudolph (2011) and Malka and Lelkes (2010) show that ideological group cues significantly influence individual policy judgments. Analyzing ideological identification within the framework of SIT would add theoretical weight to such findings while also opening up new opportunities for theoretical and empirical extension. For instance, the effects of such cues might be stronger among individuals with a high level of ISI, since in-group membership is a more central part of their self-concept, or conversely among more moderate members of an ideological group who are amenable to compromise on policy grounds but sensitive to the risk of violating in-group norms and destabilizing their self-concept. Social identity could also help to explain how individuals respond when ideological in-group cues conflict with partisan in-group cues; individuals with a stronger ISI, relative to partisan social identity, may be more likely to accept the ideological in-group cue, and vice versa.

Also, it is important to note that the symbolic ideology literature typically uses ideological self-placement to measure its central concept. The self-placement scale, however, does not directly measure the in-group attachments with which symbolic ideology scholars are concerned; while it does reveal the ideological group within which an individual self-categorizes, its most explicit function is to measure ideological extremity, e.g. “slightly liberal” versus “liberal” versus “very liberal”. To isolate the depth and specific effects of in-group attachment requires a more sensitive measure explicitly designed toward that end. The SIT literature provides such empirical measures that can be easily adapted to capture ideological in-group attachment, as detailed below. Moreover, interacting the self-placement and ISI measures, as I do in this analysis, enables scholars to gauge the behavioral impact of ideological in-group attachments across levels of self-reported ideological extremity.

The symbolic ideology literature itself attests to the relevance of SIT. Ellis and Stimson (2012) title Sect. 6.3 of Ideology in America “Another Explanation, Part 2: ‘Conservatism’ As Social Identity.” However, the authors do not situate their analysis within the SIT literature and they operationalize social identity using child-rearing values that are most commonly associated with authoritarianism (see Hetherington and Weiler 2009). Malka and Lelkes (2010, p. 160), on the other hand, draw heavily upon the SIT literature in their description of ideology’s effects, explaining at one point that “A self-categorization as ‘conservative’… will constitute a social identity when one’s self-perception as conservative is experienced as a point of similarity with other ingroup members and as a point of collective difference with outgroup members.” However, Malka and Lelkes make no direct use of social identity measures, instead operationalizing ideological identification as self-placement. In short, the concept of ISI is not new to the symbolic ideology literature but it is quite underdeveloped. I aim to develop the concept through direct measurement and analysis of ISI, and in doing so to provide a valuable theoretical and empirical extension of the symbolic ideology literature amenable to advancement through future research.

Hypotheses

The first objective of this analysis is to establish a valid measure of ISI. The traditional measure of ideological identification, used throughout the symbolic ideology literature and the political science literature more broadly, is the self-placement scale that typically asks respondents to place themselves on a seven-point ideological continuum ranging from very liberal to very conservative. The self-placement scale is explicitly designed to measure ideological extremity, which is qualitatively different from feelings of psychological attachment to an ideological in-group. Indeed, there is no basis in the SIT literature for expecting ISI to be exclusive to extreme ideologues; self-categorization alone is sufficient to produce social identity and its attendant effects (Tajfel et al. 1971). Of course, it does stand to reason that ideological extremists will tend to feel more passionately about politics and thus find political in-groups more relevant to self-definition. Given these considerations, I expect self-placement and ISI measures to be empirically related but also clearly distinguishable; specifically, I hypothesize a statistically significant but modest correlation between the two scales.

The next objective of this analysis is to estimate the prevalence of ISI in the mass public. I hypothesize that a substantial proportion of ideological identifiers will score at least moderately on the ISI scale and that their mean ISI score will exceed the scale’s median value, given the SIT literature’s finding that self-categorization within a social group is sufficient to produce feelings of social identity.

The SIT literature also indicates that the strength of a social identity may vary in response to environmental stimuli and individual characteristics; in particular, an increase in the contextual salience of a relevant identity group is positively associated with levels of social identity and inter-group bias (Gaertner et al. 1993; Mullen et al. 1992; Turner et al. 1987). A number of hypotheses follow from this empirical finding. First, ISI levels should be higher when individuals are primed to think about electoral competition, and particularly party primaries in which ideological differences often constitute the major cleavage between candidates. Second, ISI levels should be highest among self-identified conservatives since elite discourse is characterized by a disproportionate and generally positive use of the conservative label (Ellis and Stimson 2012). Third, ISI levels should be higher among individuals who use a great deal of new media sources such as cable news, talk radio, and internet blogs, since those sources tend to be more ideological in nature than traditional news sources such as newspapers, network television news, and local television news. In essence, I assume that each of these factors increases ideology’s salience to the individual in question, and in turn strengthens ISI.Footnote 3

Finally, I hypothesize an interaction effect whereby ISI conditions the impact of ideological self-placement on vote choice; specifically, for individuals with an ISI that is at least moderate in strength, I expect the effects of self-placement to be statistically significant and to strengthen as ISI strengthens. The more conventional alternative, of course, would be to test the direct effects of ISI on vote choice. However, doing so would entail treating self-placement merely as a statistical control. To the contrary, the effects of these two components of ideological identification are more likely to be complementary than independent; a very conservative individual with a strong ISI has greater motivation to support a conservative candidate than a slightly conservative individual with a strong social identity or a very conservative individual with a weak social identity. In short, ISI should reinforce the effects of ideological self-placement, and vice versa. This approach is also consistent with my goal of extending, not fundamentally challenging, symbolic ideology research. Rather than model ISI as an alternative to self-placement, and argue that the former should supplant the latter, I interact the two measures in hopes of a more constructive result: clarifying the conditions under which self-placement—the standard measure of symbolic ideology—is empirically relevant to vote choice, and when its effects are strongest and weakest.

Data and Empirical Methods

Data for this analysis come from an original online survey experiment that was administered by Knowledge Networks (KN)Footnote 4 and funded with financial support from Time-Sharing Experiments for the Social Sciences.Footnote 5 KN selected 1,666 potential participants from its ongoing Knowledge Panel for e-mail solicitation in April 2010, 66 % of whom completed the survey experiment for a total sample of 1,089 participants.Footnote 6

The survey experiment included three experimental manipulations following a 3 (election condition) × 2 (social identity question order)Footnote 7  × 2 (candidate presentation order),Footnote 8 between-groups factorial design. Participants were randomly assigned to each condition. The first manipulation was designed to test variation in ISI levels across electoral contexts, and to measure vote choice in a controlled information environment. For this manipulation, treatment group participants read an experimental stimulus, provided in this article’s appendix, describing a hypothetical general election or party primary for a US Senate seat.Footnote 9 The general election (party primary) stimulus characterized the two candidates in terms of their partisan affiliations (ideological reputations), major campaign themes, and the implications of each candidate’s victory for the partisan (ideological) direction of the country (party). Immediately after reading the stimulus, treatment group participants were asked to report the candidate for whom they would vote in the election. Control group participants read no election stimulus and so they were not presented with a hypothetical vote choice question.

To measure ISI in this survey, I adapted Mael and Tetrick’s (1992) Identification with a Psychological Group (IDPG) scale. The IDPG is an empirically robust and demonstrably valid measure of social identity (see Brewer and Silver 2000) utilized often in the social science literatures and in many previous studies of partisan social identity (see Green et al. 2005; Greene 2000; Weisberg and Hasecke 1999). The full IDPG scale consists of ten statements with which participants state their level of agreement on a seven-point scale ranging from “Strongly Disagree” to “Strongly Agree.” Space and cost constraints necessitated including a three-item IDPG subset in the survey experiment,Footnote 10 consisting of the following statements: “When someone praises (this group), it feels to me like a personal compliment”; “(This group’s) successes are my successes”; “If a prominent (member of this group) got caught in a scandal, I would not feel embarrassed at all” (reversed). I inserted partisan labels (Republicans/Democrats/Independents) in the parentheses when measuring partisan social identity and ideological labels (political conservatives/political liberals/political moderates) when measuring ISI.Footnote 11

The social identity scales used in this analysis consist of the “compliment” and “success” items only. I exclude the “scandal” item from analysis due to its poor reliability.Footnote 12 The two-item scale proves to be a reliable measure of social identity for all partisan and ideological groups, with Cronbach’s Alpha estimates of 0.78 for conservatives, moderates, and Republicans; 0.77 for Democrats; 0.75 for liberals; and 0.67 for Independents.Footnote 13 To create ISI and PSI scores, I coded each participant’s response to the relevant “success” and “praise” items as ranging from one (weakest social identity) to seven (strongest social identity) and then calculated participants’ average scores on the two scales.

I test the hypothesized causes of variation in ISI levels using a linear regression model. The key independent variables in this model represent assignment to the general election treatment group, coded one if treated and zero otherwise; assignment to the party primary treatment group, coded one if treated and zero otherwiseFootnote 14; and self-reported weekly exposure to six different types of news sources, including network television news, local television news, cable television news, newspapers, radio news, and blogs, each coded from zero (never) to four (three times a week or more). Additional independent variables include: a folded ideological self-placement scale, coded from zero (moderate) to three (very liberal/conservative); a folded party identification scale, coded from zero (independent) to three (strong Democrat/Republican); age, coded continuously; education, coded from one (eighth grade education) to eleven (professional or doctoral degree); religious attendance, coded from zero (never) to five (more than once a week); annual household income, coded from zero (less than $5,000) to eighteen ($175,000 or more); gender, coded one for female and zero for male; race, coded one for white and zero for non-white.

Next, I test the hypothesized interaction effect of ISI and ideological self-placement on vote choice, using two logistic regression models. The first model predicts vote choice in the hypothetical Senate general election, for participants in that treatment group only. The second model predicts vote choice in the 2008 presidential election for all participants, using information previously reported in the Knowledge Panel profile data. Votes for a Democratic candidate are coded one and votes for a Republican candidate are coded zero.

The key independent variables in these models are ideological self-placement, ISI, and the interaction of both scales. The self-placement scale is coded to range from one (very liberal) to seven (very conservative). The ISI scale is coded to range from one (weakest social identity) to seven (strongest social identity). The interaction term multiplies a participant’s score on both of these scales. Additional independent variables include: party identification, ranging from one (Strong Democrat) to seven (Strong Republican); a partisan social identity scale, constructed in the same manner as the ISI scale; age, gender, race, education, religious attendance, and income variables coded as described above. Finally, to control for potential order effects, I include a variable in the hypothetical Senate election model representing the order in which participants read about the candidates, coded one if the Republican candidate preceded the Democratic candidate and zero if the Democratic candidate preceded the Republican candidate.

Results

As hypothesized, the correlation between ISI and self-placement is statistically significant but modest in magnitude, at 0.283. Disaggregating by ideological groups reveals a somewhat higher correlation for conservatives (0.322) than for liberals (0.232), but no substantive differences from the overall finding. The correlation between scales is too low for them to be measuring the same concept, unless perhaps a high level of measurement error is disguising their redundancy. If, however, one or both measures were contaminated by a great deal of error, one would expect to find a less consistent and predictable relationship than is evident: comparing across the ideological self-placement scale, the mean ISI score is highest for very liberal/conservative participants (4.82), followed by liberal/conservative participants (4.60), followed by slightly liberal/conservative participants (3.88). The same pattern holds when analyzing liberals and conservatives, separately. In other words, ideological extremity on one measure tracks ideological extremity on the other measure—an indication of construct validity, I would argue. The relationship between the two scales is perfectly consistent with the argument that they tap distinct dimensions of the same underlying concept (ideological identification): ideological self-placement and ISI vary systematically, and in a predictable pattern, while correlating only modestly.

Next, I use the data from my nationally representative participant sample to estimate ISI levels in the mass public. The kernel density plot in Fig. 1 shows that a sizable proportion of ideological identifiers, in general, have what could be reasonably characterized as an ISI. In the aggregate, 41.5 % of participants score above the ISI scale’s neutral point (“neither agree nor disagree”) and nearly 29 % at least slightly agree with the ISI statements, on average. More concretely, a t test reveals that the mean ISI score for all participants (M = 4.21, SD = 1.20) is significantly higher than the scale’s neutral point of four, t(1,079) = 5.68, p = 0.000. Thus, on average, participants exhibit at least a moderate degree of ISI.Footnote 15

Fig. 1
figure 1

Kernel density plot of ideological social identity scores, aggregated and by ideological group. Ideological social identity scores represent a participant’s average level of agreement, on a 1 (strongly disagree) to 7 (strongly agree) scale, with the following items: “When someone praises [political liberals/conservatives/moderates], it feels to me like a personal compliment;” “[Political liberals’/conservatives’/moderates’] successes are my successes.” Participants completed these measures for their ideological group only

Full size image

Figure 1 also plots separate ISI scores for liberals, conservatives, and moderates. The data provide initial support for the hypothesis that conservatives have stronger ideological social identities, on average, than liberals and moderates. A majority of conservatives score above the ISI scale’s neutral point and 40 % of conservatives at least slightly agree with the ISI statements, on average, in both cases exceeding liberal and moderate performance. According to one-way ANOVAs, conservatives (M = 4.44, SD = 1.25) do, in fact, score significantly higher than moderates on the ISI scale (M = 3.91, SD = 1.07), F (1, 826) = 42.67, p = 0.000. However, the difference between conservative and liberal (M = 4.25, SD = 1.21) scores falls just short of conventional significance levels, F(1, 698) = 3.57, p = 0.059.Footnote 16 Taken together, this evidence indicates that conservatives typically have somewhat stronger ISI than liberals but the difference between them is quantitative rather than qualitative; ISI is common to both liberals and conservatives, with the latter’s advantage being marginally significant.Footnote 17

Next I test two other hypothesized causes of variation in ISI levels through linear regression analysis: the salience of electoral competition and exposure to “new media” news sources. Also included in these models are a series of demographic and political variables relevant to ideological identification and thus capable of influencing ISI levels. The value of this analysis, in essence, is to identify who is most likely to have a strong (or weak) ISI, and when ISI is strongest. To provide general conclusions about the causes of variation in ISI, first I test this model among all participants. Then, to detect potential differences between ideological groups, I separately test the same model among conservative, liberal, or moderate participants only.

Table 1 presents the results of the linear regression models, beginning with the full participant sample in Column 1. The results in Column 1 generally confirm the hypothesized impact of electoral competition and new media exposure on ISI levels. First, in comparison to the excluded category of control group participants, who read no experimental stimulus priming thoughts of electoral competition, participants in the general election and party primary treatment groups exhibit statistically significant increases in ISI levels. This finding is consistent with research in the SIT literature indicating that social identity becomes stronger in response to environmental stimuli that make the relevant identity more salient. However, contrary to my expectations, the magnitude of that increase is essentially equal for participants exposed to a general election stimulus, which refers only to party competition, and participants exposed to a party primary stimulus, which explicitly references competition between ideological groups. Apparently, electoral competition in general strengthens ISI, regardless of whether ideological groups are made implicitly or explicitly salient.

Table 1 Predictors of the strength of ideological social identity
Full size table

New media exposure also predicts increased ISI; the cable television news and internet blog variables are statistically significant and positively signed in Column 1, while talk radio and each of the traditional news variables are not significant. This finding is most likely attributable to the ideological nature of many cable news programs and blogs. Of course, the same can be said about talk radio, which is not a significant predictor of ISI. Perhaps this discrepancy is due to the diversity of talk radio, which includes not only ideological news programs but also traditional news programming such as National Public Radio and local news, as well as numerous sports programs and stations.

Other statistically significant variables in the full sample model include gender, age, and the folded party identification and ideological self-placement measures. Males and older people exhibit higher levels of ISI. Also, not surprisingly given the preceding analysis, more partisan and ideological participants exhibit higher levels of ISI. Ideological self-placement’s effects provide additional insight into the empirical relationship between that variable and ISI; whereas preceding evidence of the relationship is merely correlational, the results from Table 1 demonstrate that it is robust to a substantial battery of relevant covariates. Yet, at the same time, the relationship is hardly determinative, given the size of self-placement’s coefficient (0.260) and the independent effects of other significant variables in the model. These results further indicate that ideological self-placement and ISI are empirically related yet distinct constructs.

The variable most notable for not reaching statistical significance is education. Political behavior models often use education as a proxy for political sophistication (see, for example, Zaller 1992), and it is a common finding in political science research that political sophisticates understand and utilize ideological concepts more capably than non-sophisticates (Jacoby 1991; Knight 1985). Education’s failure to reach statistical significance in this analysis indicates that ISI is not generally limited to political sophisticates.Footnote 18

The second, third, and fourth columns of Table 1 present results from the same empirical model, tested respectively among conservative, liberal, and moderate identifiers. Comparing across models provides the opportunity to identify variation in the factors predicting ISI for different ideological groups. Indeed, the hypothesized effects of electoral competition and new media exposure vary across ideological groups. In terms of the former, assignment to the general election and party primary conditions is a statistically significant predictor of increased conservative and moderate social identity. However, neither variable is significant in the liberal social identity model. The explanation for this discrepant result is unclear, and worthy of future analysis given its departure from the empirical findings of the SIT literature. The effect among conservatives and moderates, however, is consistent with expectations and also worthy of further analysis. In particular, the finding that party primaries strengthen ISI among conservatives and moderates has important implications for scholars’ understanding of partisan polarization. To the extent that increased ISI among these groups leads to greater inter-group bias in behavior and attitudes, as the SIT literature would predict, this finding might help to explain polarized attitudes and voting preferences within parties, particularly among conservatives and moderates of the Republican Party, and their downstream effects on general elections and governance. Specifically, as ideological divisions within the party become more salient, ISI and its attendant effects become more pronounced among ideologues as well as moderates. Since ideologues typically dominate primary electorates, however, the end result is likely to be decreased support for moderate candidates and policies and increased support for more ideological candidates and policies, thus leading to greater polarization within and later between the parties. Given recent debates among scholars over the extent and causes of mass polarization (see, for example, Abramowitz 2010; Fiorina and Abrams 2008), this relationship is well worth exploring in future research.

The effects of new media exposure also vary across ideological groups: cable news exposure is a significant and positive predictor of conservative and moderate social identity only, while blog exposure is a significant and positive predictor of liberal social identity only. A possible explanation for the discrepancy between liberals and conservatives is that the former rely more upon blogs, such as the Daily Kos and Firedoglake, for ideological content while the latter rely more upon cable news programs, such as Fox News’ Hannity and The O’Reilly Factor, for ideological content. Again, the explanation for this discrepancy is a worthy topic for further research. Also, it is worth noting that the only traditional news source to reach statistical significance in these models, local television news, is negatively associated with moderate social identity. The explanation for this finding is not immediately clear; perhaps the relative lack of ideological salience in local news coverage depresses social identity for moderates who use it as their primary news source.

Among the other variables included in these models, the folded ideological self-placement scale is a statistically significant and positive predictor of conservative and liberal social identity,Footnote 19 while the folded party identification scale is a statistically significant and positive predictor of liberal and moderate social identity but only a marginal and positive predictor of conservative social identity. The only other variable to reach conventional levels of statistical significance in any of these models is education, which positively predicts liberal social identity. For reasons detailed above, this is a particularly interesting finding. All other tests indicate that a high level of education, and by proxy political sophistication, does not influence the strength of an ISI. However, for liberals it is relevant; more educated, or sophisticated, liberals tend to feel a greater sense of psychological attachment to their ideological in-group. Absent further research, I can only speculate as to why this is the case. A plausible explanation is that the one-sided and negative nature of social messages concerning the liberal label, including elite ideological discourse and extra-political connotations, discourages strong and positive attachment to that label except or less so among those individuals who possess sufficient political sophistication to evaluate the liberal in-group based upon an advanced understanding of shared policy preferences and values. Once more, further research would be useful in developing a more concrete explanation for this finding.

Having analyzed the validity, prevalence, and causes for variation in ISI, the final objective of this analysis is to evaluate its impact on political behavior. To do so, in accordance with the hypothesis described above, I estimate two logistic regression models testing the interaction effect of ISI and ideological self-placement on vote choice in a hypothetical US Senate election, described to participants in the experimental general election condition, and in the 2008 presidential election. The results of the hypothetical vote choice model are presented in the first column of Table 2, and the results of the presidential vote choice model are presented in the second column.

Table 2 Predictors of vote choice in hypothetical and actual elections
Full size table

Several variables in Table 2 reach conventional levels of statistical significance, including party identification and education, in both models; age, in the experimental vote choice model; and gender and race, in the presidential vote choice model. Following conventional methodological precedent, I do not substantively interpret the coefficients and significance levels of the lower-order interaction variables or the interaction term since it comprises two continuous variables (see Brambor et al. 2006; Braumoeller 2004). Instead, I use graphs of the interaction effect to evaluate it.Footnote 20

Figures 2 and 3 plot the marginal effect of ideological self-placement on vote choice across levels of ISI. The solid line in each figure represents the marginal effect of self-placement on vote choice in that election, as a participant’s ISI increases, and the dashed lines represent the upper and lower 95 % confidence intervals for the marginal effect. Ideological self-placement’s effect on vote choice is statistically significant at any point on the ISI scale that the 95 % confidence intervals do not include a value of zero.

Fig. 2
figure 2

Interaction effect of ideological self-placement and ideological social identity on vote choice in a hypothetical US Senate election. This figure is based upon logistic regression model results presented in the first column of Table 2, predicting vote choice among participants assigned to an experimental condition describing a hypothetical US Senate election between Democratic and Republican candidates. The dependent variable is coded one for Democratic vote choice, zero for Republican vote choice, and missing otherwise. To evaluate a hypothesized interaction effect, Fig. 2 plots the marginal effect of ideological self-placement on vote choice across levels of ideological social identity (ISI). The solid line in this figure represents the marginal effect of self-placement on vote choice in the hypothetical Senate election, as a participant’s ISI increases, and the dashed lines represent the upper and lower 95 % confidence intervals for the marginal effect. Ideological self-placement’s effect on vote choice is statistically significant at any point on the ideological social identity scale that the 95 % confidence intervals do not include a value of zero. ISI scores represent a participant’s average level of agreement, on a 1 (strongly disagree) to 7 (strongly agree) scale, with the following items: “When someone praises [political liberals/conservatives/moderates], it feels to me like a personal compliment;” “[Political liberals’/conservatives’/moderates’] successes are my successes.” Participants completed these measures for their ideological group only

Full size image
Fig. 3
figure 3

Interaction effect of ideological self-placement and ideological social identity on vote choice in the 2008 US Presidential election. This figure is based upon logistic regression model results presented in the second column of Table 2, predicting vote choice among all participants who reported voting for Barack Obama or John McCain in the 2008 presidential election. The dependent variable is coded one for Democratic vote choice, zero for Republican vote choice, and missing otherwise. To evaluate a hypothesized interaction effect, Fig. 3 plots the marginal effect of ideological self-placement on vote choice across levels of ideological social identity (ISI). The solid line in this figure represents the marginal effect of self-placement on vote choice in the 2008 presidential election, as a participant’s ISI increases, and the dashed lines represent the upper and lower 95 % confidence intervals for the marginal effect. Ideological self-placement’s effect on vote choice is statistically significant at any point on the ideological social identity scale that the 95 % confidence intervals do not include a value of zero. ISI scores represent a participant’s average level of agreement, on a 1 (strongly disagree) to 7 (strongly agree) scale, with the following items: “When someone praises [political liberals/conservatives/moderates], it feels to me like a personal compliment;” “[Political liberals’/conservatives’/moderates’] successes are my successes.” Participants completed these measures for their ideological group only

Full size image

Figure 2 shows that the marginal effect of ideological self-placement on vote choice is statistically significant only at or above the midpoint of the ISI scale (four), in the hypothetical Senate election. Thus, for participants with a weak psychological attachment to their ideological in-group, self-placement on the ideological scale does not significantly influence vote choice in this election. However, for participants with a moderate to strong ISI, a majority of the survey experiment sample, self-placement does significantly influence vote choice. These results provide initial support for the hypothesized interaction effect; ideological self-placement influences vote choice, but only insofar as an individual also feels at least moderate psychological attachment to the ideological in-group.

Figure 3 provides additional support for the hypothesized interaction effect, this time using data from an actual election. The marginal effect of ideological self-placement on vote choice in the 2008 presidential election becomes statistically significant starting at a score of 2.5 on the ISI scale. Thus, self-placement has no discernible effect on vote choice for participants with very weak psychological attachment to an ideological in-group, while it does have a significant effect for participants with moderately weak to very strong ISI.

For a more precise estimation of the interaction effect, in Table 3 I present the predicted probabilities of voting for the Democratic candidate in the 2008 presidential election and the experimental general election. For both models, I predict the probability of voting for a given candidate at each category of ideological self-placement, and at the neutral and maximum values of ISI. When calculating predicted probabilities, all variables are held at their mean or modal values except for ideological self-placement, ISI, and the interaction term.

Table 3 Predicted probabilities of Democratic vote choice, by ideological self-placement and strength of ideological social identity
Full size table

In both elections, the predicted probability of voting for a Democratic candidate changes in the expected direction when moving from the midpoint to the maximum of the ISI scale and holding self-placement constant. The predicted probability of voting for the Democratic candidate increases as ISI becomes stronger for very liberal, liberal, and slightly liberal participants, excepting only slightly liberal participants in the 2008 presidential election. Likewise, the predicted probability of voting for the Democratic candidate decreases as ISI becomes stronger for very conservative, conservative, and slightly conservative participants, in both elections.

The pattern of change in predicted probabilities varies substantially between the two vote choice models, with conservatives showing greater change across ISI values in the presidential election and liberals showing slightly greater change across ISI values in the hypothetical Senate election. Also, the change in predicted probabilities for liberals is far greater in the experimental general election than in the presidential election, while it differs minimally for conservatives between the two elections. Barack Obama’s advantage in the 2008 election might help to explain these discrepancies; liberals at all levels of ideological extremity and ISI were especially likely to vote for the Democratic candidate under unusually favorable electoral conditions, while conservatives were more susceptible to variation given unfavorable electoral conditions and a Republican nominee oft-maligned for ideological unorthodoxy. In any case, the essential conclusion from both models is the same: ideological self-placement’s effect on vote choice is not statistically significant when ISI is weak, and it becomes stronger as ISI also becomes stronger. As expected, these two distinct but related components of ideological identification reinforce one another, motivating more ideologically consistent behavior as extremity increases.

To explain this relationship more clearly, I would argue that ISI represents an additional motivation not to defect from an ideological in-group and, in this case, support the out-group’s candidate. An individual who is, for example, very conservative is unlikely to vote for an out-group candidate on the basis of symbolic preference alone. However, when tempted the individual is less likely to actually defect if he or she feels a strong sense of psychological attachment to the ideological in-group. To do so would not only go against in-group interests that reflect upon self-evaluation, but it would be a violation of group norms that jeopardizes a valued in-group membership and thus risks the significant psychological discomfort of destabilizing the self-concept. The psychological incentive to engage in ideologically consistent behavior therefore weighs more heavily on individuals with a strong, versus weak or nonexistent, ISI.

Such an explanation of ideology’s impact on political behavior has not been previously established in the symbolic ideology literature, primarily due to its reliance upon ideological self-placement as a measure of the central concept. Indeed, the symbolic ideology literature, to date, has not adequately addressed the conceptual and empirical shortcomings of measuring ideological in-group attachment through ideological self-placement. The fundamental contribution of this article is to develop a credible, direct measure of ideological in-group attachment and through its empirical application clarify the conditional nature of ideological self-identification’s impact on vote choice, with potential expansion to other political behavior and attitudes.

Discussion

The results of this analysis underscore ISI’s theoretical and empirical potential to expand scholars’ understanding of ideological identification, in general, and symbolic ideology, in particular. The ISI scale is derived from a valid and reliable measure of social identity used prominently within the SIT literature, and evidence indicates that it is empirically distinct from the ideological self-placement scale that is standard within symbolic ideology research. According to data obtained from a nationally representative survey experiment sample, ISI is common within the American public; in fact, the mean ISI score among ideological identifiers significantly exceeds the scale’s neutral point, and nearly one-third of participants at least slightly agree with statements tapping psychological attachment to an ideological in-group.

This analysis also identifies who is most likely to hold a strong ISI, and when those identities are strengthened by environmental stimuli. Comparing across ideological groups, ISI is strongest among conservatives and weakest among moderates. Feelings of psychological attachment to an ideological in-group are not, however, exclusive to conservatives; liberals and conservatives, on average, score significantly higher than the neutral point on the ISI scale. ISI also is not exclusive to sophisticates; education does not significantly predict the strength of ISI, in general, although it is significantly and positively related to liberal social identity. I also find evidence that ISI is responsive to environmental stimuli that increase the salience of ideological concepts. In general, participants who read experimental stimuli describing a hypothetical general election or party primary scored significantly higher on the ISI scale, and about equally so between conditions, although this effect is not found among liberal identifiers. The effect of party primaries on ISI, for moderates and conservatives, potentially has significant implications for scholars’ understanding of partisan polarization, as it may help to explain moderates’ struggles to win party nominations and subsequent office; to the extent that primaries stimulate increased ISI, ideological majorities are more likely to view moderates within their party as an out-group and discriminate accordingly. Participants who frequently use “new media” news sources, such as cable television news and internet blogs, score significantly higher on the ISI scale as well; exposure to cable news is associated with increased conservative social identity, and exposure to blogs is associated with increased liberal social identity.

Finally, ISI has a meaningful impact on vote choice and in a way that has significant implications for the symbolic ideology literature. Whereas symbolic ideology studies routinely test the direct effects of ideological self-placement on political attitudes and behavior, an interaction of self-placement with the ISI scale shows evidence of a conditioning effect. Specifically, self-placement’s effects on vote choice are limited to participants who feel at least a moderate degree of psychological attachment to their ideological in-group, and the effects of self-placement increase as ISI increases. The value of this empirical finding is not to challenge the utility of the ideological self-placement scale or to recommend its replacement as a measure of symbolic ideology, but to identify the conditions under which self-placement’s effects are limited or strengthened by other aspects of ideological identification. Also, from a methodological standpoint, this analysis underscores the potential for, and value of, developing more direct measures of ideological in-group attachment than the self-placement scale that is standard in symbolic ideology research. These contributions have significant potential to improve scholars’ understanding and measurement of symbolic ideology and its behavioral significance. To further understand the nature and significance of ISI, future analyses should also include measures of operational ideology, such as issue preferences, in their predictive models. Data limitations preclude such an analysis in this article. However, if it were carried out, I would expect to find that the effects of ISI are robust to the inclusion of operational ideology measures; the theoretical and empirical underpinnings of this article provide no basis for expecting otherwise.Footnote 21

Most importantly, ISI represents a valuable opportunity for the theoretical extension of symbolic ideology research. SIT and symbolic ideology share many commonalities, in particular their emphasis on the interaction between internal categorization and external group cues. Indeed, a number of symbolic ideology scholars have invoked the language and theoretical assumptions of SIT to explain the nature of ideological identification. To this point, though, the symbolic ideology literature’s incorporation of SIT has been underdeveloped and implicit more so than explicit. By directly measuring ideological identification as a social identity, as well as analyzing its causes and behavioral effects, this article represents a first step toward more fully exploiting the theoretical and empirical potential of SIT in order to improve scholars’ understanding of ideological identification, in general, and symbolic ideology, in particular.

To further this argument, consider the implications of Self-Categorization Theory (Turner 1985; Turner et al. 1987). SCT is an adaptation of SIT that distinguishes itself, in part, by the assumption that individuals subjectively evaluate in-groups and out-groups in comparison to the perceived prototype of a superordinate category to which both groups belong. Thus, inter-group discrimination occurs when individuals perceive a relevant out-group as deviating from the norms and values of the superordinate group also encompassing the individual’s in-group. If the in-group and out-group are not perceived as belonging to a common superordinate group, or if the out-group is perceived to be more relatively prototypical of the superordinate group than the individual’s in-group, social discrimination is unlikely to occur.

This theory has significant implications for understanding, among other things, intra-party ideological divisions and their effects on collective action within a party. After all, party primaries often play out as contests between candidates representing opposing ideological factions, both of which struggle not just for the nomination but to define the party’s image and expressed principles. Once the nomination is decided, party leaders and the nominee in particular try desperately to heal factional wounds and unite the party to achieve victory in the coming general election, as Mitt Romney did in the 2012 presidential election and John McCain did in the 2008 presidential election. But, as Romney and McCain both found out, ideological opponents often resist—if not through outright defection than by withholding the fruits of enthusiastic support.

Self-Categorization Theory is potentially relevant to understanding this process. For instance, SCT would suggest that the most effective way for party leaders to reduce social discrimination against members of the victorious intra-party ideological faction—such as withholding means of support or actively promoting defection—is to redefine the losing faction’s superordinate category, from one that excludes the nominee to one that includes the nominee. The candidate could do so by engaging in divisive partisan rhetoric that raises the salience of inter-party differences, or by selecting a vice presidential nominee from the losing ideological faction so that its supporters will perceive the party as mutually inclusive. Also, social identity could be helpful in understanding whether, when, and to what degree members of competing intra-party ideological factions agree to reconcile past differences and unite for collective action. I would suspect that ISI negatively predicts reconciliation for members of the losing ideological faction, who fear strengthening the out-group through electoral success, and positively predicts reconciliation for members of the winning ideological faction, who hope to strengthen the in-group through electoral success.

Another opportunity for theoretical advancement comes from Mummendey and Wenzel’s (1999) In-Group Projection Model. The IPM posits, in part, that individuals project the characteristics of their subgroup onto the superordinate in-group, and therefore view fellow subgroup members as prototypical of the superordinate group and members of opposing subgroups as deviant. The IPM could be helpful in understanding, among other things, evaluations of candidate ideology. It is not uncommon for individuals to inaccurately project their ideological identification onto a favored candidate (see, for example, Carmines and Berkman 1994), or for popular candidates to be perceived by members of different ideological groups as a member of their in-group. If the IPM is relevant to understanding such phenomena, it could be that mischaracterizations of a candidate’s ideology are less the product of ignorance or a lack of ideological sophistication than a psychological motivation to project an ideological subgroup identity onto the leader of a partisan superordinate group.

These are but two examples of how a more explicit and thorough integration of SIT stands to enrich scholars’ understanding of ideological identification, and particularly symbolic ideology. More direct evidence comes from the results of this analysis, which indicate the credibility and relevance of a social identity approach to ideology.