Ideological Social Identity: Psychological Attachment to Ideological In-Groups as a Political Phenomenon and a Behavioral Influence

The symbolic ideology approach, while influential, is hardly universal among scholars; critics argue that ideological identifications are primarily based upon substantive policy views rather than symbolic evaluations (see, for example, Abramowitz and Saunders 2006). The present analysis is not designed to demonstrate the superiority of either approach, but to refine scholars’ understanding of symbolic ideology with the view that it, as well as operational ideology, has significant theoretical and empirical value.

This explanation is not contingent upon a belief in the symbolic ideology view, contested by operational ideology advocates, that ideological identification is primarily based upon evaluations of social groups associated with ideological labels (Conover and Feldman 1981; Zschirnt 2011; for a critical analysis, see Abramowitz and Saunders 2006), although that is a view with which I generally agree. Self-categorization, or internalization of a group identity, is sufficient to trigger the social identity processes described above; whether an individual arrives at the point of ideological identification based on symbolic evaluations or policy preferences should not be determinative, only the fact of self-categorization. Social identity and its attendant processes are evident among a wide range of groups, after all, from those based on arbitrary assignment (see, for example, Tajfel et al. 1971) to those seemingly based on social and political convictions, such as feminism (Huddy 1997). Once an individual develops an ideological social identity, however, the ideological in-group becomes a source of psychological and affective attachment motivating political behavior based on group considerations rather, or at least more so, than independent policy judgments. In this sense, the ideological in-group takes on a symbolic function regardless of whether it was adopted for symbolic or substantive reasons.

Iyengar et al. (2012) adopt a similar approach in their study of partisan affective polarization, explaining: “the more salient the affiliation, the more biased the individual’s beliefs about in-group and out-group members. Salience itself can depend on either dispositional factors… or characteristics of the information environment…” (pp. 407–408). Their analysis indicates that campaign environments have the effect of strengthening partisan identities and confirming partisan stereotypes; in particular, exposure to negative campaign advertisements is a statistically significant and positive predictor of affective polarization. Thus, I find in this research direct support for emphasizing the role of salience in social identity processes, and indirect support for the hypothesized effects of electoral competition and exposure to more ideologically-biased media sources on ideological social identity.

Knowledge networks (KN) uses random-digit dial and address-based sampling methods to recruit a representative group of Americans into its Knowledge Panel. KN also provides Internet access and hardware, including computers, when needed in order to facilitate representative sampling.

TESS uses funds from a National Science Foundation grant (SES-0818839) to provide financial support for scholars to conduct online survey experiments among nationally representative participant samples. For more information, see www.​tessexperiments.​org.

Knowledge networks included a post-stratification weight in the deliverable data to correct for demographic unrepresentativeness. I use this weight in my analysis when appropriate.

I included this manipulation to test whether social identity levels varied depending on the order in which participants completed ISI versus PSI items. One-way ANOVAs show no such effect. I exclude this manipulation from further analysis.

This manipulation randomly varied the order in which participants read about party candidates in the general election condition or ideological candidates in the party primary conditions. There is no reason to expect that candidate order influences reported social identity levels. It could, however, affect vote choice and so I include it in the experimental election vote choice model.

Participants in the latter condition were assigned to party primaries based on reported party affiliation in the KN profile data: Republican (Democratic) identifiers read about a Republican (Democratic) primary and Independents were randomly assigned to either party’s primary.

Green et al. (2005) use a three-item subset of the IDPG scale, and Weisberg and Hasecke (1999) four items, to study partisan social identity. Both scales prove to be valid and reliable.

Participants completed one set of PSI measures and one set of ISI measures, corresponding to their partisan and ideological affiliations as previously reported in the KN profile data.

Cronbach’s Alpha estimates range between 0.02 and 0.48 when pairing the scandal item with either of the other two items, well below acceptable reliability standards.

As an additional check on the performance of the two-item scale, I used data from a related study of 2008 national party convention delegates that included five social identity measures: the three already described plus two additional measures adapted from the IDPG. The compliment-success scale performs exceptionally well in this analysis. As a measure of liberal, conservative, and Democratic—but not Republican—social identity, this scale performs far better than any other two-item scale combination, and better than expanded scales including all five items or four after excluding the scandal item.

Control group participants constitute the baseline category in this analysis, since they are coded zero for both of the electoral condition dummy variables.

Ideological social identity, it should be noted, proves not to be a proxy for partisan social identity. Although the two variables are highly correlated, at 0.619, this correlation is almost identical to that of the traditional ideological and partisan self-placement measures, at 0.603. Moreover, a majority of participants have distinct scores on the ISI and PSI scales, with more than a quarter (27.4 %) exhibiting a stronger ideological than partisan social identity. In fact, participants identifying as conservatives and Republicans score significantly higher in terms of ideological social identity (M = 4.62, SD = 1.20) than they do in terms of partisan social identity (M = 4.40, SD = 1.15), t(350) = 4.19, p = 0.000, while moderate Independents score significantly higher on PSI (M = 4.10, SD = 1.04) than on ISI (M = 3.60, SD = 1.01), t(48) = −3.54, p = 0.001. Also, liberal Democrats have a higher PSI (M = 4.46, SD = 1.22) than ISI, but this difference reaches only marginal levels of statistical significance (M = 4.33, SD = 1.22) t(211) = −1.83, p = 0.070. Thus, while ideological and partisan social identities are empirically related, they do represent distinct constructs and warrant distinct analysis here as well as within the political science literature at large.

While ISI levels are lowest among moderates, some of these respondents do exhibit a strong ideological social identity: 26.8 % of moderates score above the scale’s midpoint, and 15.8 % at least slightly agree with the ISI statements. Given the prevalence of ideological moderates in the mass public, including 35.5 % of the survey experiment sample, moderate social identity could be politically significant. For instance, moderate social identity may help to empirically explain responses to partisan and ideological polarization such as third-party voting and ticket-splitting. Future research would be useful in exploring such effects.

Additional ANOVA tests confirm that ISI does not vary significantly across education levels, when measuring education across its full range or as a median-split dichotomous variable.

Folded ideological self-placement is excluded from the moderate social identity model, since moderate identifiers inhabit a single position on the scale and this precludes variation.

Likelihood ratio tests provide an additional indicator of ideological social identity’s empirical contribution. These tests confirm that the addition of the ISI and interaction variables significantly improves model fit, both in the hypothetical Senate vote choice model and in the 2008 presidential vote choice model, at the 0.05 confidence level.

To provide some initial insight into this relationship, in separate analyses of ANES data I created an approximate measure of ISI by calculating the difference between liberal or conservative respondents’ feeling thermometer ratings of their ideological in-group and out-group, such that a higher relative rating of the in-group indicates a higher level of ISI. It is worth noting that Iyengar et al. (2012) use a parallel strategy to measure partisan affective polarization (see Fig. 3; Table 3), a concept framed explicitly in terms of Social Identity Theory. Logistic regression models predicting vote choice in the 1984–2008 presidential elections yield results remarkably similar to those presented in this analysis and consistent across elections; even when controlling on five operational ideology measures (government spending and services; health care; abortion; defense spending; government aid to African-Americans), as well as relevant political and demographic variables, the marginal effect of ideological self-placement on presidential vote choice is statistically significant but only for respondents with at least a moderately strong (approximated) ISI, and its effects become stronger as ISI become stronger. One or more of the five operational ideology measures is also statistically significant in each vote choice model, attesting to the at least somewhat independent effects of these operational and symbolic ideology measures.