An Ideology by Any Other Name

Abstract

The terms ‘liberal’ and ‘conservative’ are prominent features of political discourse in the United States, and many citizens choose to identify with one of these ideological labels. Yet, many citizens do not fit comfortably in either of these categories, and comparatively little is known about the breadth and importance of other ideological labels in the mass public. We pose a novel but simple survey question to a large nonprobability sample of survey respondents (n = 4,655) to measure self-identification with up to 14 different ideological labels, and trace the associations of this self-labeling with individual differences and political attitudes. We find that identification with alternative ideological labels is widespread in our sample and contains important information about political attitudes that common survey questions on ideology do not capture. In a preregistered conjoint experiment (n = 2,433), we show that attachment to alternative labels predicts vote choice in both primary and general election contexts. Our novel approach to measuring multidimensional ideological attachments provides a foundation for expanded scholarship on ideology beyond the confines of the liberal-conservative framework.

The literature on mass ideology in the United States revolves around issues of conceptualization and measurement (Feldman 2013). The dimensionality of mass ideology concerns both. While many argue that multiple dimensions are necessary to accurately represent mass ideology (Feldman and Johnston 2014; Treier and Hillygus 2009), others suggest that a single dimension, ranging from liberal to conservative, is sufficient for practical purposes, and captures much of the interesting variation in political behavior (Fowler et al. 2023; Jost 2006; Marble and Tyler 2022). Yet most of this work focuses on the covariance structure of political beliefs and attitudes; for example, the extent to which policy attitudes are structured by one or more latent dimensions. In this paper, we build on and extend a much smaller literature on the nature of ideological identification (e.g., Conover and Feldman 1981; Malka and Lelkes 2010; Sniderman et al. 1991). Some research suggests that liberal-conservative self-placement is only weakly held as an identity (Kinder and Kalmoe 2017; Levendusky 2009; but see Devine 2015) and may largely capture feelings about cultural conservatism (Ellis and Stimson 2012) or partisan identity rather than issue preferences (Barber and Pope 2019; Mason 2018). Other work suggests this common variable may obscure important differences between social and economic liberalism-conservatism (Klar 2014). Still other research argues that identities beyond liberal and conservative play a meaningful role in structuring American political behavior, such as “libertarian” (Iyer et al. 2012).

We extend this literature to consider identification with a significantly larger set of ostensibly ideological labels. Consistent with the limited work on this topic, our purpose is largely descriptive and exploratory. To this end, we consider several questions. First, how prevalent is ideological identification beyond liberal and conservative? Second, does alternative identification reflect psychological commitment to a political label or group? Third, how well do alternative forms of identification describe specific policy attitudes? Fourth, do alternative ideological signals affect vote choice?

We adopt a simple, novel approach to measuring ideological identification. We ask a large nonprobability sample of US adults (n = 4,655) to self-identify with up to 14 different ideological labels, alongside measures of political attitudes and demographic characteristics. We map identification to issue preferences and political values, and assess the extent to which alternative identification describes meaningful deviation from the standard form of self-placement. We argue that these labels convey unique information about mass opinion by allowing respondents to express a wider range of attachments without an assumed unidimensional structure. In a preregistered conjoint experiment (n = 2,433), we then explore the impact of candidate self-labeling in primaries and general elections, and evaluate whether correspondence between voter and candidate self-identification increases the probability of preferring that candidate. We find significant and substantively meaningful effects, comparable in size to candidate position-taking on salient issues, that suggest the importance of alternative labels to political decision making.

Importantly, our aim is explicitly exploratory. We do not examine self-identification with every possible political label, nor do our nonprobability samples allow for generalizable population inferences. Nonetheless, the labels we measure include a diverse array of possible identities and we would not expect our primary conclusions to be altered significantly by the addition of other labels. We intend this paper to serve not as a definitive statement about the distribution of ideology in the American public, but rather as initial evidence of substantial identification with alternative ideological labels and a foundation for future research.

Existing Research on Ideological Identification

Most past research on ideological identification in the U.S. concerns the meaning and importance of self-placement on a 7-point scale ranging from “extremely liberal” to “moderate” to “extremely conservative” (e.g., Boutyline and Vaisey 2017; Conover and Feldman 1981; Jost 2006; Malka and Lelkes 2010; Philpot 2017; Sniderman et al. 1991). The two most recent, comprehensive investigations of this variable are Kinder and Kalmoe (2017) and Ellis and Stimson (2012).

Kinder and Kalmoe (2017, 47) “establish three elementary, but consequential facts”: many Americans (over one in four) decline to place themselves on the liberal-conservative scale; those who are willing to self-locate choose “moderate, or middle of the road” at very high rates (again, about one in four); and “conservative” is substantially more popular than “liberal” among those who choose a non-moderate category. They argue that self-placement does not represent a strong psychological commitment for most Americans, in contrast to partisanship (see also Levendusky 2009). This weak attachment is reflected in lower stability over time, weaker associations with vote choice, and more responsiveness to significant political events. They ultimately conclude that liberal-conservative identification is not a prime mover of American mass politics. Rather, “liberal” and “conservative” are labels that, when asked explicitly, some citizens recognize as reflecting their broader political orientation.

Ellis and Stimson (2012) note that the popularity of the conservative label is surprising because the American “public mood”—the diffuse ideological orientation of the public in left-right space—is consistently left-of-center over time (see also Claassen et al. 2015). They demonstrate that conservative self-identifiers express a wide range of political beliefs, and a large percentage are “conflicted” or “cross-pressured” in that they combine conservative identification with left-wing policy preferences (see also Feldman and Johnston 2014). Ultimately, they argue that the label “conservative” reflects a broader, less explicitly political orientation in the American public. When people identify as conservative, they are primarily indicating their orientation toward culturally conservative values, such as “conventional behavior and appearance, playing by the established rules, and fitting into established social patterns” (131). The popularity of the conservative label thus reflects a general tendency toward social conformity rather than strong support for politically conservative policies.

These findings are qualified somewhat by Feldman and Johnston (2014), who find significant heterogeneity in how people implicitly understand the liberal-conservative dimension. They find at least three groups, with the first mapping only social issues to these labels, another group mapping only economic issues, and a third mapping both social and economic issues to liberal and conservative. This suggests that these labels may be too broad to capture the myriad ways people view themselves politically. Weber and Federico (2013) provide additional evidence for this, demonstrating substantial diversity in the moral principles endorsed by distinct groups of self-identified conservatives.

A smaller literature probes alternative measures of ideological self-identification. Klar (2014) explores distinct varieties of self-identification with liberalism and conservatism in the American public. In addition to the traditional 7-point scale for overall political views, Klar asks respondents to place themselves on the same scale, but with respect to their views on social issues and, separately, with respect to their views on economic issues (resulting in three distinct 7-point scales). While the correlations among these items are large, they leave substantial room for idiosyncrasy, with a correlation of 0.60 between social and economic self-placement. This again suggests that many combinations of social and economic self-labeling exist in the American population, which the labels “liberal” and “conservative” are too broad to capture.

Very little work has gone beyond these common labels. In a 2014 study, the Pew Research Center finds that about 11% of Americans both know what the term “libertarian” means and agree that the label describes them well.¹ Moreover, Iyer et al. (2012) find that libertarian self-identifiers show substantial differences in values and psychological traits compared to both liberal and conservative self-identifiers. This reiterates that there are potentially important subgroups of the public for which liberal and conservative are insufficient.

Several scholars have also studied identification with the environmental movement (Dunlap and McCright 2008; McCright & Dunlap 2015; Pichardo Almanzar et al. 1998). McCright and Dunlap (2015) find that about 5 to 10% of Americans identify as an “active participant” in the environmental movement. Since 1989, Gallup has polled Americans on whether they identify as environmentalist and, if so, whether that identification is strong.² Such identification has declined over time, yet remains high; in 2021, approximately four in ten Americans identified as an environmentalist and one in five identified strongly with the label. Moreover, Gallup’s surveys suggest that environmentalist self-identification is associated with significant differences in attitudes concerning the environment and environmental policy.

This brief review of the literature suggests that a substantial proportion of Americans may identify with ideological labels beyond liberal and conservative. Nevertheless, there has been little attempt to explore alternative forms of ideological identification in a more comprehensive way, and a number of important questions remain.

First, investigations of alternative forms of identification to date have narrowly focused on a particular label; we do not know what percentage of Americans identify with other labels, whatever those labels might be. Moreover, we cannot assume that (say) libertarians and environmentalists identified in prior studies are discrete groups, because those who identify with one alternative may identify with others as well. Given the substantial proportion of Americans who do not self-identify as liberal or conservative, the possibility that many people would self-identify with other labels is compelling, but remains largely untested.

Second, we do not know whether identification with alternative labels reflects a meaningful psychological commitment, or if such identification is merely a way of describing or signaling specific political opinions or cultural values. Building on Kinder and Kalmoe (2017), we consider whether alternative ideological self-identification reflects what might be considered a proper social identity (Devine 2015), similar to some forms of partisanship (Greene 2004; Huddy et al. 2015).

Finally, we do not know whether alternative forms of ideological identification are predictive of political attitudes or behavior. We consider whether these labels consistently describe patterns of public opinion, above and beyond what can be learned by knowing someone’s self-placement on the traditional 7-point scale. Further, ideological labels associated with political actors and parties provide important information regarding their general policy commitments, in part because such actors have electoral incentives to maintain a consistent political brand (Hinich & Munger 1994). Knowing that a candidate is associated with a particular ideological label can thus serve as a powerful heuristic when casting a vote, even if a voter is not strongly attached to the label in the psychological sense. To the extent that alternative self-identification reflects political beliefs and social identities that differ from typical left-right conceptions, candidates may benefit from demonstrating affinity with these alternatives to attract idiosyncratic or cross-pressured voters, who tend to be more persuadable (Treier & Hillygus 2009) and can be consequential in close elections.

Overall, our evidence suggests that measuring a wider range of ideological attachments can aid scholars in mapping mass policy attitudes and understanding political behavior, especially in domains where the standard scale is less informative. Our investigation includes two studies with distinct data collection efforts. We first discuss the design and results of an observational study before turning to a conjoint experiment on vote choice.³

Study 1: Observational Data

Sample

Our data for Study 1 is a large nonprobability sample of US adults collected in three waves via Lucid, with each respondent participating in only one wave. The waves were fielded in August 2018 (n = 4,148), October/November 2019 (n = 1,265), and July 2020 (n = 1,803). We pool these respondents into a single sample of 7,216 respondents. After removing respondents who failed an attention check (n = 1,142), who broke off before our ideology measure (n = 1,323), one uniform non-responder (on all ideology, personality, partisanship, and issue measures; n = 1), and those likely providing insincere⁴ responses on our main ideology measure (n = 95), our final pooled sample consists of 4,655 respondents.⁵ This nonprobability sample differs from the US adult population on several known demographic dimensions (see descriptive statistics in Appendix C.2) and likely differs on unobserved dimensions as well. Our data therefore cannot provide generalizable population inferences about the U.S. public. Nevertheless, our sample is quite diverse and provides a useful first exploration of broad-based alternative ideological attachments. Our aim with Study 1 is not to define mass ideology with representative precision, but rather to establish notable patterns among a wide range of ideological labels that can serve as a basis for further research.

Variables

Self-Description in Terms of Ideological Labels

All respondents were asked: “Which of the following describes your political beliefs and attitudes? Please check all that apply.” We provided a randomly-ordered list of 14 ideological labels: liberal, progressive, green, conservative, traditional, libertarian, fascist, socialist, communist, communitarian, nationalist, populist, cosmopolitan, and environmentalist. Respondents could select as many of the labels as they wished, including zero. Those who selected zero labels were asked whether there is “some other word or phrase you would use to describe your political attitudes and beliefs?” This list is not intended to be comprehensive, but to cover a diverse array of labels that appear with non-negligible frequency in contemporary political discourse. It inevitably reflects the judgment of the authors and we expect future work to expand or refine our list. Importantly, however, very few respondents selected zero labels (74), and even fewer (5) volunteered another alternative label, which suggests we captured most of the salient and important alternatives to liberal and conservative.

Strength of Ideological Identification

Respondents who selected multiple labels were asked which of their chosen labels best describes their political beliefs and attitudes. For each respondent’s best-describing label (or the only one chosen, if applicable), we asked several questions to determine psychological attachment to that label, adapted from Huddy et al. (2015; see also Devine 2015), e.g. how important is being a [label] to the respondent, or how often does the respondent use “we” instead of “they” when talking about [label]s. We scale each response to vary between 0 (weakest attachment) and 1 (strongest attachment), and take a simple average to generate a single measure of identity strength for each respondent.

Political Engagement

We measure political engagement as the average of three variables: attention to news about national politics (5-point scale), weekly consumption of political news (0 to 7 days), and political knowledge (number of correct answers to five objective knowledge items). We recode each variable to vary between 0 and 1, and average the three for a single overall measure of political engagement.

Political Orientation and Political Values

We asked about political partisanship using the standard branched question format to generate a 7-point scale of partisan identity (PID), from strong Democrat to strong Republican. We also asked respondents to place themselves on the standard 7-point liberal-conservative ideological scale used in most American public opinion surveys. We measured moral traditionalism with four standard 7-point Likert items from the American National Election Studies (ANES), which we average into a single measure. We measured preferences for limited government with the three standard binary items from the ANES, which we average into a single measure.

Policy Attitudes

We asked respondents for their opinions on ten policy issues: military spending, government involvement in health insurance markets, marriage equality, affirmative action, abortion access, the minimum wage, free trade, taxes on the wealthy, social security privatization, and immigration. Exact question wording is provided in Appendix C.3.

Results

Self-Labeling Beyond Liberal and Conservative

Do people self-identify with ideological labels beyond liberal and conservative? Fig. 1 provides the proportion of respondents who self-identified with each of the 14 ideological labels. Respondents could select multiple labels to describe themselves; indeed, many did so, and the median respondent chose two.⁶ Very few chose none (1.6%). In comparison, 33.1% of our respondents answered “moderate” for liberal-conservative self-placement on the 7-point scale (or did not answer the question), suggesting that the standard measure may miss substantial ideological information that these alternative labels can capture.

Despite the wide range of available labels, both conservative and liberal are among the most popular labels, although traditional is narrowly more popular in our sample. Consistent with the extant literature (e.g., Ellis and Stimson 2012; Kinder and Kalmoe 2017) we find that traditional (37.6%) and conservative (37.6%) garner substantially more identifiers than liberal (29.8%) and progressive (20.9%). 23.8% of our sample self-identifies as environmentalist, similar to recent Gallup estimates of strong environmentalists (22%).⁷ Several other labels also garner substantively important levels of self-identification in our sample, including green, socialist, libertarian, and nationalist, each of which are each chosen by at least 7% of respondents. Perhaps unsurprisingly, communist (1.6%) and fascist (0.9%) attracted the fewest self-identifiers, while the cosmopolitan, populist, and communitarian labels were each selected by less than 5% of the sample.

Fig. 1

Data from Study 1

What about those who call themselves “moderate” on the standard 7-point self-placement scale (or do not provide a response)? As we show in Appendix Figure A. 1, more than half of these respondents (50.8%) identify as traditional, with substantial self-identification as environmentalist (25.0%), progressive (18.0%), or green (12.1%). Some also self-identified as conservative (23.3%) or liberal (13.4%), with 64.7% of those respondents also selecting at least one other label. This evidence suggests that the standard 7-point scale may confuse ideological identification with overall left-right orientation in a spatial sense. Some who feel that liberal or conservative is self-descriptive nonetheless place themselves as moderate on a bipolar scale. The high rate of self-selection of other labels further indicates that the standard measure misses meaningful variation within the “moderate” category.

Because all respondents could select multiple labels, in Appendix A.2 we consider the probability that a respondent identified with each label conditional on having also identified with another. These probabilities provide a sense of the dependency or nesting of labels. For example, the probability that a liberal identifier also chooses progressive is 0.36, whereas the probability that a progressive identifier also chooses liberal is 0.52. This suggests that progressive is, to some extent, a subcategory of liberal. Figure 2 provides a network representation of this matrix.⁸ The size of each node represents the proportion of respondents who identify with that label. Directed edges represent the probability, expressed by edge width, that a respondent who identifies with the label at the origin of the edge also identifies with the label at the head of the edge. For example, respondents who identify as nationalist are very likely to also identify as traditional, but the reverse is not true. Edges representing probabilities less than 0.30 are excluded. Overall, the plot suggests that, for many people, alternative labels enable them to differentiate their ideological identity within a broader category. That is, people who identify with an alternative label (e.g., socialist) are quite likely to identify with one of the more popular labels as well (e.g., liberal). In this sense, many alternatives are at least partially nested within the broader categories, rather than opposing identities.

Fig. 2

Data from Study 1. Directed edges represent the probability, expressed as edge width, that a respondent who identifies with the label at the origin of the edge also identifies with the label at the head of the edge. Node size represents the proportion of total respondents who identify with that label. Edges representing probabilities below 0.30 are excluded

Respondents who selected two or more labels were asked to choose which label is most important to them.⁹ Panel A of Fig. 3 shows the overall proportion of respondents that chose each label as the most important. Panel B of Fig. 3 instead shows the proportion of respondents who selected each label as the most important, conditional on self-identifying with that label (possibly among several). In other words, while only 29.6% of our sample stated that conservative was their most important label, 77.6% of all conservative self-identifiers did so.

Fig. 3

Data from Study 1

Figure 3 shows several noteworthy patterns. Consistent with the prevailing literature, conservative is a more popular label in our sample than liberal: a higher proportion of conservative identifiers than liberal identifiers also named that respective label as the most important. Panel A shows that conservative (29.6%) and liberal (21.1%) labels are most commonly chosen as the most important overall, followed closely by the traditional label (19.6%). Far fewer chose another label as most important: 9.6% chose progressive, 6.7% environmentalist, 3.6% libertarian, and—despite much discussion of populist policies in contemporary elite debates—just 0.5% of respondents chose populist as their most important label. These patterns are reinforced in Panel B: conservatives (77.6%) and liberals (69.7%) overwhelmingly chose that label as the most important, whereas traditionalists (51.2%), progressives (45.3%), and libertarians (45.2%) did so only about half the time. Identifiers with the remaining labels were overwhelmingly likely to choose something else as the most important.

How strong are respondents’ psychological attachments to their most important label? Fig. 4 shows the distribution of respondents’ average responses on the identity strength scale, ranging from 0 (weakest attachment) to 1 (strongest attachment). Each facet of Fig. 4 shows a histogram of strength of identity among respondents who chose that particular label as their most important. Among these identifiers, the mean identity strength is marked by the dashed line and reported in the upper left of each facet. The shade of each histogram bin reflects the number of respondents from the overall sample that belong to that bin.

For all ideological labels, the primary mass of each histogram leans rightward, showing that the typical respondent who chose that label as the most important expressed more agreement than disagreement with the identity strength items.¹⁰ But two other patterns bear comment. First, the distribution of attachment strength for all labels is wide, including for liberal and conservative; many respondents express quite strong attachment, but some also expressed weak attachment to the label, with most somewhere in the middle. Second, identifiers with the most popular label (traditional) are more heavily concentrated in the middle of the scale than other major ideological labels (i.e., conservative, liberal), indicating a relatively tepid attachment to the term. ¹¹

Fig. 4

Data from Study 1. The dashed line indicates the per-label mean strength of identity

We next examine how label identification relates to partisanship. Figure 5 depicts the proportion of respondents occupying each position on the standard 7-point partisan identification scale that identifies with each ideological label (again excluding “nones”). Liberal identifiers comprise a large majority of strong Democrats (65.8%), but only 39.8 and 42.8% of weak Democrats and Democrat-leaning Independents (respectively), and much smaller numbers of “true” Independents and Republicans. Most Republicans self-identified with the conservative label, from 56.6% among leaners to 77.1% among strong Republicans. But the conservative label was also popular among Independents and weak Democrats, with 25.4 and 23.4% (respectively) identifying as conservative. Indeed, weak Democrats in the sample are more likely than both strong Democrats and Democrat-leaning Independents to identify as conservative or traditional—indicating a cross-pressured policy conflict with the modern Democratic party.

Several alternative labels show strong partisan leans. The progressive and socialist labels were selected primarily by Democrats, but hardly any Independents or Republicans in the sample, while nationalist identifiers were almost exclusively Republican. These high rates of alternative identification among leaners suggest perceived policy conflicts with the nearer party that do not derive necessarily from moderate or centrist opinions. For example, leaning Democrats are more likely than strong or weak Democrats to identify as progressive or environmentalist. Similarly, leaning Republicans are more likely to identify as populist and nationalist than strong or weak Republicans. Scholars have long noted that leaning partisans often behave like other partisan identifiers (Keith et al. 1986; Klar and Krupnikov 2016). Our results agree, in that leaners identify with similar ideological labels. But our results also add nuance: the leaners identify at higher rates with some of the less mainstream labels and at lower rates with the mainstream ones. This may suggest that at least some leaning partisans are more extreme on salient policy dimensions, rather than more moderate (as implied by their location on the 7-point PID scale).

Fig. 5

Data from Study 1

Other alternative labels are cross-cutting, drawing identifiers from across the partisan spectrum. Traditional and libertarian both lean slightly Republican, yet garner many identifiers across the partisan divide. The environmentalist and green labels also draw many adherents from both Democrats and Republicans, attracting only slightly more Democrats.

Policy Preferences

The 7-point ideology measure has been a staple of political science and public opinion scholarship for decades. Can knowledge of Americans’ other ideological attachments help us understand their political preferences and values? We argue that the range and combinations of ideological attachments provide meaningful information not captured by the standard scale.

To test this possibility, we asked respondents for their preferences on ten policy issues (military spending, government health insurance, marriage equality, affirmative action, abortion access, the minimum wage, free trade, taxes on the wealthy, social security privatization, and immigration), plus a scale measuring moral traditionalism and another measuring preference for limited government. We use subsets of these items to estimate respondent positions on two latent dimensions of left-right space: an economic conservatism dimension described by limited government values and attitudes on taxes, minimum wage, social security, and health insurance; and a social conservatism dimension described by moral traditionalism and attitudes on marriage equality, abortion access, affirmative action, and immigration.¹² We then estimate a hierarchical linear model for each dimension to map the positions of “typical” self-identifiers for each label within the space. Each model nests respondents within 502 groups of unique observed patterns of identification, such as “progressive green” or “conservative traditional nationalist,” including one group with no identification. The model includes a fixed effect for each of the 14 ideological labels (the typical position for all identifiers with that label across all groups) and a random intercept for each observed group (group-level deviation away from the sum of its component fixed effects).¹³

Beyond estimating the mean position for specific labels, the hierarchical models allow us to evaluate whether the labels are essentially additive—that is, whether a “progressive libertarian” has views equivalent to the position of a “progressive” plus the position of a “libertarian,” or whether particular combinations of self-identification reveal new information not captured by the component parts. We do so by examining each model’s conditional intraclass correlation coefficient (ICC), which describes the proportion of variance in latent ideology attributable to the 502 observed groups, above and beyond the variance explained by the fixed effects from each label. The conditional ICC can vary between 0 and 1, with values close to 0 indicating almost no variance is attributable to the particular combinations beyond their additive effects.

Figure 6 plots the fixed effects of each ideological label on both economic and social dimensions—that is, the marginal effect of identification on economic ideology and (separately) on social ideology. The error bars indicate 95% confidence intervals for the marginal effect on the respective latent dimension. Liberal, progressive, environmentalist, and socialist identification strongly predict left-leaning positions on both dimensions in our sample, while the nationalist and conservative labels indicate right-leaning positions on both dimensions. Notably, several alternatives exhibit a strong lean on one dimension but not the other. The traditional label predicts strong social conservatism but only weakly predicts conservative economic attitudes. Conversely, the libertarian label strongly predicts economic conservatism, but does not predict social conservatism. Several other labels, grouped in the center of the plot, have no significant marginal effect on policy preferences ceteris paribus, including green, populist, communitarian, cosmopolitan, and communist.¹⁴

Fig. 6

Data from Study 1. Figure shows fixed effect coefficients estimated via a hierarchical model that also estimates a random intercept for each of 502 unique label identification patterns. Bars indicate 95% confidence intervals

We find that the effect of each label is largely additive: the conditional ICC is 0.028 for the economic conservatism model and 0.027 for the social conservatism model, indicating that specific combinations of ideological attachments do not provide much additional information beyond the sum of their components. Yet the additive properties of the labels neatly convey both heterogeneity and extremity of policy positions in our sample. Figure 7 plots the predicted group-level positions (linear fixed effects plus random intercept; positions shown in grey) on the economic and social dimensions for the 57 unique identification groups with at least 10 members in our sample (circle size indicates group size). The hollow circles show the random intercepts for 7-point self-placement positions (estimated in separate models as the only predictors).

Fig. 7

Data from Study 1. Figure shows predicted group positions estimated via two hierarchical models with label fixed effects and group random intercepts. Grey circles indicate a label pattern group; hollow circles indicate a liberal-conservative self-placement group position

Figure 7 demonstrates two points. First, the label combinations describe substantial heterogeneity away from the unidimensional arrangement expressed by the 7-point scale. Consider our sample’s 19 “conservative green traditional environmentalists,” located on the upper-left periphery of the “slightly conservative” (point 5) on the 7-point scale. While their conservative traditionalism ably expresses their socially conservative views, their more left-leaning pro-environmental beliefs are also captured, moving them to a firmly moderate position on economic matters relative to their “conservative traditional” peers, situated between “conservative” (point 6) and “extremely conservative” (point 7). Similarly, all libertarian-affiliated groups (denoted by “I”) are positioned below the main diagonal, expressing their substantial economic, but less social, conservatism.

Second, Fig. 7 demonstrates that the label identification patterns capture substantial extremity in ideology beyond the “extremely liberal” (point 1) and “extremely conservative” (point 7) endpoints of the standard scale. That is, identifying with several labels strongly aligned on the left or on the right predicts a consistency and extremity of attitudes on both dimensions that is well beyond what the truncated 7-point scale can measure. This difference is especially pronounced on the left in our sample, where the extremity of a large percentage of left-wing respondents is substantially underestimated by the standard scale.

In Fig. 8, we show that these useful properties extend to those in our sample who identify as “moderate” on the standard 7-point scale (or offer item non-response; total n = 1,543). For example, when estimating group positions in this subsample, “libertarian” moderates look very similar to “slightly conservative” full-sample respondents on the economic dimension, but look more similar to “slightly liberal” full-sample respondents on the social dimension. Similarly, “conservative traditional” moderates and “progressive” moderates are located near the “slightly conservative” and “slightly liberal” points on the social dimension (respectively), but have somewhat more moderate views on the economic dimension. In contrast, “progressive environmentalist” moderates exhibit views similar to the average “slightly liberal” respondent on both dimensions. Even among “moderates” on the standard 7-point scale, our expanded measure indicates that some still identify with liberal or conservative labels, which Fig. 8 suggests are meaningful self-categorizations and are strongly predictive of positions on the two latent dimensions.

Fig. 8

Data from Study 1. Figure shows predicted group positions among “moderate” respondents on the standard self-placement scale, estimated via two hierarchical models with label fixed effects and group random intercepts. Grey circles indicate a label pattern group among the “moderate” (or item non-response) subsample; hollow circles indicate a liberal-conservative self-placement group among the full sample

Narrowing to individual policy issues, we see similar patterns. Figure 9 shows the fixed effects for self-identifying with each ideological label, estimated via hierarchical models (for each issue) that also estimate a random intercept for the 502 observed groups in the full sample with unique identification patterns. Once again, the marginal effects of labels are additive: the conditional ICC value for several of these models approaches 0, and the highest (affirmative action) is an underwhelming 0.035, meaning that specific identification patterns explain little variance in policy attitudes beyond the sum of their component effects.¹⁵

Fig. 9

Data from Study 1. Figure shows fixed effect coefficients are estimated via a hierarchical model that also estimates a random intercept for each of 502 unique label identification patterns. Bars indicate 95% confidence intervals. Positive values indicate a more conservative policy attitude

As with the latent dimensions of economic and social ideology, the marginal effects of the labels on individual policy attitudes convey both heterogeneous cross-pressures for some groups and homogenous consistency pressures for others. For example, the traditional label is reliably associated with conservativism on social issues, but not economic issues. The libertarian label, by contrast, predicts a conservative position on nearly all issues in the economic domain, but not for most social policies, and predicts a liberal attitude on marriage equality. The environmentalist label is associated with liberal positions on most economic matters and much in the social domain, but not on abortion or affirmative action, two areas well removed from environmental regulation. Yet the additive marginal effects also usefully predict extremity of opinion, in addition to cross-pressures: a “liberal progressive environmentalist” in our sample is likely to have very liberal views on many issues, whereas a “traditional conservative nationalist” is likely to have very conservative views.

Study 2: Conjoint Experiment

We fielded a preregistered conjoint experiment in July 2023 to evaluate the importance of alternative ideological attachments for voting in both primary and general election settings. In particular, we assess whether a candidate taking on a specific ideological label (other than “liberal” and “conservative”) affects support for that candidate on average (Hypothesis 1), and whether a match between the candidate’s and voter’s self-identified labels increases the probability of choosing that candidate (Hypothesis 2).¹⁶

To test these hypotheses, we recruited a nonprobability sample of 2,733 US adults via the Lucid platform. We removed 300 participants that failed preregistered data quality checks, providing an analysis sample of n = 2,433.¹⁷ Each participant made seven consecutive choices between random pairs of candidate profiles, generating n = 34,062 total candidate observations. We randomly assigned participants to make choices between candidates for the US House of Representatives in either a general election context (between a Democrat and a Republican) or in a primary election context (between two Democrats or between two Republicans, conditional on the respondent’s partisanship¹⁸).

Fig. 10

An example screenshot of a decision task in the general election condition of Study 2

For each selection task, each candidate profile provided a limited set of information: a policy position on 5 of 27 possible issues (adapted from Ahler and Broockman 2018), an ideological label (or no label), and the candidate’s political party. For each policy position, Democratic (Republican) candidates expressed a left-leaning (right-leaning) stance with probability 0.8 and a right-leaning (left-leaning) stance with probability 0.2. The ideological label for each candidate profile was similarly randomized: Democratic (Republican) candidates could be labeled as “Liberal,” “Progressive,” “Socialist,” or “Environmentalist” with probability 0.16 (0.04) each; as “Conservative,” “Nationalist,” “Libertarian,” or “Traditional” with probability 0.04 (0.16) each; or left unlabeled with probability 0.2 (0.2). We thus assigned each candidate’s policy and ideology signals using weights conditional on the candidate’s partisanship to improve the plausibility of the candidates, while still allowing for variation that captures useful information about individual preferences vis-à-vis abnormal candidates. For each choice, the respondent was asked “Which candidate do you prefer?” with four options: “Strongly” or “Slightly” prefer Candidate 1, or “Slightly” or “Strongly” prefer Candidate 2. Figure 10 shows an example candidate selection task in the general election condition. As preregistered, we binarize the candidate support variable to indicate whether the respondent “Slightly” or “Strongly” supported a given candidate (1), or did not (0).

As with many conjoint experiments, our design includes deliberate deviations from the typical setting in which voters perform the decision task under study. In particular, the profiles we present elevate the salience and accessibility of specific policy positions far above what would usually be true for voters completing a ballot. We also do not include any demographic, occupational, or personality information about the candidates that is often known to voters. Because ideology is particularly informative as a proxy for unknown policy positions (Hinich and Munger 1994), our design thus provides a hard test of our hypotheses by making substantial policy information both known and salient in the decision task. Statistically significant average marginal component effects (AMCEs; Hainmueller et al. 2014) from the ideology signals would therefore offer strong evidence that alternative ideological attachments constitute meaningful considerations in choosing among candidates, even under these conditions.

Results

We first consider whether expressing alternative ideological perspectives helps or hurts candidates on average—that is, across all “voters” (respondents) in a given election. We consider Democratic primary candidates (n = 9,352), Republican primary candidates (n = 7,728), Democratic general election candidates (n = 8,491), and Republican general election candidates (n = 8,491) separately.¹⁹ We do so by regressing (binary) candidate support on each of the eight possible ideology labels (with unlabeled candidates as the reference category), and include each of the 27 possible policy issues (-1 for left-leaning stance, 0 for no stance, and 1 for right-leaning stance) as covariates. We use an ordinary least squares estimator and cluster the standard errors at the respondent level.

Figure 11 shows the estimated AMCE of each ideological signal (AMCEs for the issue positions are not shown; regression results are provided in Appendix B.1) for each type of candidate. While the effects of expressing a conservative ideology in a primary (bottom panels) are not unexpected—a strong penalty for Democrats (p < 0.001) and a strong benefit for Republicans (p < 0.001)—we also find that several alternative ideological labels meaningfully affect candidate selection, providing support for H1. Democratic primary candidates (bottom-left panel) who signal a nationalist ideology suffer a penalty on average (p = 0.019). Similarly, Republican primary candidates (bottom-right panel) suffer a penalty for signaling a socialist ideology (p = 0.021), and benefit from signaling a traditionalist ideology (p = 0.011). The effect of ideological signals is less impactful for general election candidates; partisanship is, after all, an enormously powerful signal (Levendusky 2009; Rahn 1993). Even so, we see suggestive evidence that alternative ideological signals can matter in general elections, such as the positive average effect of embracing traditionalism as a Democrat (p = 0.058).

Fig. 11

Data from Study 2. Figure shows estimated AMCE of each candidate type signaling one of eight ideologies. See Appendix B.1 for full results

The AMCE estimates shown in Fig. 11 indicate the average effect of providing a given ideological signal on voters’ support, not just among those who identify with a specific label. While these results have implications for candidate strategy, they do not necessarily indicate the importance of ideological labels to individual voters. Indeed, we expect that those who self-identify with the same ideological label as a candidate should be particularly likely to support that candidate (H2).²⁰

Fig. 12

Data from Study 2. Figure shows estimated AMCE of a match between candidate-expressed and voter self-identified ideology. See Appendix B.2 for full results

In Fig. 12, we show that H2 is broadly supported: the AMCE of an ideological “match” between the voter and the candidate—that is, the candidate signals an ideology that the voter also identifies with—is both statistically significant and substantively meaningful for candidates of either party, in both primary and general elections, and for matches on the alternative labels as much as for matches on the mainstream liberal and conservative labels. The estimated effect of an alternative label match ranges from 0.046 (p = 0.008) for Republican general election candidates to 0.076 (p < 0.001) for Democratic general election candidates, with similar effect sizes for primary candidates of both parties (model results are reported in Appendix B. 2). Of particular note is that matches have an effect even in the general election context where partisanship is available as a cue, at a magnitude of approximately half to two-thirds the effect size of a voter moving one step along the 7-point partisanship scale. Further, we find meaningful negative effects for candidates providing cross-cutting ideological signals that voters do not identify with, as shown by the significant negative AMCE estimates for nationalist Democratic primary candidates and socialist Republican primary candidates. In sum, our evidence makes clear that liberal and conservative are not the only ideological signals that matter to voters—embracing other ideological labels can also help or hurt political candidates.

Concluding Remarks

We began with the aims of exploring (1) the breadth of identification with alternative ideological labels, (2) the degree of psychological commitment to alternative labels, (3) the relationship of alternative labels to policy attitudes, and (4) the extent to which alternative ideological labels are meaningful for vote choice. We consider our findings with respect to each in turn.

At the most basic level, we find that identification with alternative ideological labels is quite broad in our sample. Only 25.8% endorsed the liberal or conservative label exclusively, eschewing all other options. The traditional, environmentalist, and progressive labels were each endorsed by at least one in five respondents, while the green, socialist, libertarian, and nationalist labels were each endorsed by at least one in twenty. Several alternatives, such as traditional and environmentalist, dominated identification among self-placed “moderates” on the 7-point scale, and both of these labels were endorsed widely by both Democrats and Republicans. The list of ideological labels that our respondents could choose was not exhaustive—there are surely other political labels with which a significant portion of the American public identifies—but our results offer confidence that ideological identification beyond liberal and conservative is not narrow in scope.

Psychological commitment to alternative labels appears modest in our sample, but no weaker than for liberal and conservative. Liberal and conservative labels were most frequently chosen as our respondents’ most important self-descriptors, but those who chose an alternative label as their most important displayed comparable levels of psychological attachment as liberal and conservative identifiers.

We find that the range of alternative ideological labels that we capture allow us to map two important phenomena. The first is substantial variation away from the unidimensional left-right continuum that the 7-point scale approximates, such as the cross-pressures faced by a “traditional environmentalist” with slightly liberal views on most economic issues but more centrist views on social policy. The second phenomenon is a better estimation of ideological extremity and consistency. On the left, our typical respondent who identifies solely as liberal shows moderate liberalism on both economic and social dimensions, landing roughly halfway between “slightly liberal” and “liberal” on the 7-point scale. However, knowing that a liberal respondent also identifies as progressive allows us to estimate their position near the endpoint of the 7-point scale (“extremely liberal”), while knowing that the liberal progressive respondent also identifies as environmentalist or socialist allows us to estimate their position well past the endpoint of the 7-point scale. And as we show in Appendix A. 5, we find that some labels are more predictive of policy extremity on specific policy issues than others. For example, the nationalist label is predictive of right-leaning attitudes on military spending, but not on affirmative action. In other words, each label describes (on average) certain sets of ideas, not all of which fall consistently on the left or the right. As ideological attachments to several left-leaning or several right-leaning labels stack up, we can better predict ideological extremity and consistency across issues.

To be clear, the standard 7-point scale remains a powerful predictor of policy attitudes, and some alternative labels appear to be at least partially understood as subcategories of liberal or conservative. Yet the range of ideological affinities we capture provides valuable information above and beyond the standard scale.

In Study 2, we demonstrate that these alternative labels can matter for voting behavior in both primary and general election contexts. Our evidence indicates that a range of ideological signals from candidates—not just “liberal” or “conservative”—affect candidate selection, even when a high degree of salient policy information about the candidates is readily available. Voters who self-identify with the same label as a given candidate are especially likely to vote for that candidate—and in some cases, voters who do not share a cross-cutting ideological self-identification with a candidate are significantly less likely to support that candidate.

Notably, we measure a broad range of ideological attachments using a single, simple, and quickly administered survey question. While our measure is slightly more time-intensive than the 7-point scale (median time of 18 versus 6 s), it is substantially simpler than the typical method of measuring multidimensional ideology via multi-item scales (e.g., Ansolabehere et al. 2008) and still escapes the pitfalls of assuming a unidimensional, bipolar structure to mass attitudes. Scholars may find our measure particularly useful when assuming a unidimensional structure is especially risky, such as when examining the political behavior of self-identified “moderates” and others not well described by the standard scale, or when examining public opinion in specific policy domains where key alternative attachments are likely to be meaningful.

To be sure, our nonprobability samples offer only limited insights about the exact degree of alternative ideological attachments among the U.S. public. And, undoubtedly, there are still other ideological attachments that we do not measure here but likely have meaningful bearing on mass political attitudes. Yet our evidence offers confidence that the mass public’s alternative ideological attachments are substantial in scope and have important implications for political behavior—which makes them well worth measuring, in both probability and nonprobability contexts (Jerit and Barabas 2023). Our analysis can thus provide a helpful foundation for continued scholarship that expands our understanding of ideology beyond the narrow liberal-conservative frame.