Values, partisanship, and ideology

The effort to systematically evaluate individuals’ values originates in social psychology (Rokeach Reference Rokeach1973). Shalom Schwartz (Reference Schwartz1992, 4) posits that ‘values (1) are concepts or beliefs, (2) pertain to desirable end states or behaviours, (3) transcend specific situations, (4) guide selection or evaluation of behaviour and events, and (4) are ordered by relative importance’. Values have genetic origins (Hatemi et al. Reference Hatemi, Medland, Klemmensen, Oskarsson, Littvay, Dawes, Verhulst, McDermott, Nørgaard, Klofstad, Christensen, Johannesson, Magnusson, Eaves and Martin2014; Schermer et al. Reference Schermer, Vernon, Maio and Jang2011), are developed early on in life (Vecchione et al. Reference Vecchione, Schwartz, Alessandri, Döring, Castellani and Caprara2016b; Vecchione et al. Reference Vecchione, Schwartz, Davidov, Cieciuch, Alessandri and Marsicano2020), and are held stably throughout the course of one’s lifetime (Searing et al. Reference Searing, Jacoby and Tyner2019). Values also correlate with individuals’ political and personal behaviours (Enke et al. Reference Enke, Rodríguez-Padilla and Zimmermann2023; Schwartz et al. Reference Schwartz, Cieciuch, Vecchione, Torres, Dirilen-Gumus and Butenko2017). This article focuses on political values, which are beliefs about the ways in which society should be ordered and the priorities that society should have. Some commonly studied political values are moral traditionalism, egalitarianism, individualism, and patriotism (Feldman Reference Feldman1988; Goren Reference Goren2005; Jacoby Reference Jacoby2014).Footnote ¹

Values constitute one type of core predisposition – that is, a set of durable inclinations or evaluations – which could influence issue stances (Alvarez and Brehm Reference Alvarez and Brehm2002). This article tests the role of values relative to two other drivers of issue stances: ideology and partisanship. The concept of partisanship – operationalized here as an individual’s identity with a political party – is clearly differentiated from values and ideology. Ideology and values, however, are conceptually similar and require careful differentiation.

The term ‘ideology’ is used inconsistently in political science. Gerring (Reference Gerring1997) notes that the common threads across definitions are (1) constraint across idea elements, (2) contrast relative to other ideologies, and (3) stability through time (980). As political scientists use the term, (4) the subject matter of ideology is politics rather than beliefs about the world more broadly, distinguishing ideology from worldviews or belief systems (982). Most public opinion scholarship in American politics makes two additional assumptions: (5) ideology operates in a hierarchical fashion, ‘in which more specific attitudes interact with attitudes towards the more general class of objects in which the specific object is seen to belong’, (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960, 190) and (6) ideology operates as a one-dimensional scale on a liberal-conservative spectrum.Footnote ² In using the term ideology in this article, I refer to the concept as delineated by these six components.

Like this left-right concept of ideology, values offer constraint across idea elements, stand in contrast to other values, are stable over time, and operate hierarchically.Footnote ³ However, there are a few key differences between values and ideology. First, values are developed with primary application to an individual’s everyday surroundings (Vecchione et al. Reference Vecchione, Döring, Alessandri, Marsicano and Bardi2016a). For example, traditionalism may manifest in church attendance. By comparison, however, left-right ideology is defined in terms of policy content (for example, preferences over the size of the government), with potential secondary effects on one’s personal life. While political values and ideology are both ‘political’ in the broad sense of the term, the left-right spectrum – unlike values – is intrinsically defined in terms of policy. Second, while liberalism and conservatism are logical opposites on a spectrum, values do not inherently go together or oppose each other – one can favour traditionalism and egalitarianism, but the left-right ideological concept does not permit one to be both a liberal and a conservative. Instead, each value operates in its own dimension, and individuals differ in the extent to which they prioritize a value (Jacoby Reference Jacoby2006).Footnote ⁴ Since values are widely held, politicians rarely attack them (unlike how they might criticize left- or right-wingers); instead, politicians try to explain how their actions are consistent with widely held values. Third, values cut across typical understandings of left-right ideology. The same freedom-centric rhetoric of ‘my body, my choice’ can be used to justify abortion rights by liberals or opposition to vaccine mandates by conservatives. Values constitute substantive visions for the ways that society should be ordered that do not necessarily align with ideology.

While this article focuses on the left-right notion of ideology standard to political science, another prominent theory in psychology defines ideology as a type of motivated social cognition (MSC) (Jost et al. Reference Jost, Glaser, Kruglanski and Sulloway2003). On this view, the nucleus of (conservative) ideology that unites epistemic, existential, and ideological motives for conservatism is ‘psychological attempts to manage uncertainty and fear. These, in turn, are inherently related to the two core aspects of conservative thought … resistance to change and the endorsement of inequality’ (Jost et al. Reference Jost, Glaser, Kruglanski and Sulloway2003, 351). One reading of this theory suggests that values and ideology are interchangeable – that is, perhaps traditionalism (as a measure of resistance to change) and egalitarianism (as a measure of non-acceptance of inequality) themselves constitute ideology (Malka and Lelkes Reference Malka and Lelkes2010). While this seems to be a plausible reading at first glance, researchers in the MSC tradition do distinguish between values and ideology in their empirical specifications; in particular, they typically operationalize ideology primarily through symbolic self-identification (Jost Reference Jost2006; Jost, Nosek, and Gosling Reference Jost, Nosek and Gosling2008). MSC researchers have specified a theoretical model instead in which psychological beliefs drive values, which in turn drive ideology; in doing so, they have found that egalitarianism only predicts left-right ideology in certain national contexts (Thorisdottir et al. Reference Thorisdottir, Jost, Liviatan and Shrout2007). This sort of specification and the ensuing findings suggest that those operating in the MSC tradition treat values as distinct from ideology.Footnote ⁵ I show that these concepts are empirically separable in Appendix Section I.

Beyond ideology, voters employ additional predispositions to make sense of politics. Group identities and affective judgments about other groups are important tools for voters in interpreting politics (Converse Reference Converse and Apter1964; Elder and O’Brian Reference Elder and O’Brian2022; Sniderman et al. Reference Sniderman, Brody and Tetlock1991; Tesler Reference Tesler2012). Another type of predisposition is a premise or a belief about the way the world is rather than how it should be (Barker and Marietta Reference Barker and Marietta2022). I discuss the relationship between other predispositions and issue stances in the penultimate section.

Do values drive issue stances?

This article tests whether values drive individuals’ issue stances. Under this theory, values operate in a hierarchical fashion, exerting ‘downward’ effects on issue stances (Goren Reference Goren2013). Existing work has also posited a hierarchical role for ideology on issue stances (for example, Peffley and Hurwitz Reference Peffley and Hurwitz1985). There are two reasons why values can serve in this hierarchical role. The first is that values are centrally available heuristics for voters seeking cognitive shortcuts (Alvarez and Brehm Reference Alvarez and Brehm2002; Sniderman et al. Reference Sniderman, Brody and Tetlock1991; Goren Reference Goren2013). It requires more cognitive effort for a voter to study policy particulars in order to develop an opinion than it does to take an accessible, closely held general belief and quickly apply it to a particular case. A voter may come to favour the Affordable Care Act through egalitarianism (since it enables patients with pre-existing conditions to receive insurance coverage) or oppose it by reference to freedom (since it forces voters to purchase insurance). Second, values serve as evaluative standards for good versus bad policies. Even if a voter carefully researches a policy question, she will generate an opinion by applying previously held dispositions to the particulars at hand. Values offer one such standard.

Many studies show correlations between values and issue stances even after controlling for partisanship, ideology, and voter demographics (Berinsky Reference Berinsky2002; Brewer Reference Brewer2003; Feldman Reference Feldman1988; Feldman and Zaller Reference Feldman and Zaller1992; Feldman and Steenbergen Reference Feldman and Steenbergen2001; Jacoby Reference Jacoby2006). However, cross-sectional studies can generate observational equivalence between causal and non-causal accounts in a number of ways (Lenz Reference Lenz2012). The biggest risk in interpreting correlations between values and issue attitudes as causal is that there may be other omitted variables which drive issue stances, which, when accounted for, would attenuate the observed correlations. The benefit of panel studies is that we have multiple observations of the same individuals, which mitigates concerns regarding time-invariant confounders such as the individual’s upbringing and other unmeasured individual-level characteristics that are likely to be stable over time (for example, personality traits). Furthermore, since causality requires that cause precedes effect, it is difficult to extract causal relationships from cross-sectional data. The panel studies used in this article let us explore how values and issue stances evolve over time to ‘stress test’ the role of values on political attitudes.Footnote ⁶

These sorts of methodological concerns are especially justified for the question at hand because the cross-sectional relationships suggesting that values drive partisanship and ideology have not fully stood up to panel-based designs. Panel studies produce mixed findings on whether values influence partisanship (Evans and Neundorf Reference Evans and Neundorf2020; Lupton and McKee Reference Lupton and McKee2020) or partisanship influences values (Goren Reference Goren2005), and such studies also suggest that ideology influences moral foundations to a greater degree than vice-versa (Hatemi et al. Reference Hatemi, Crabtree and Smith2019). The use of panel studies has also caused scholars to revisit previously held assumptions regarding the causal primacy of other predispositions, such as identity characteristics (Egan Reference Egan2020) and racial beliefs (Engelhardt Reference Engelhardt2021; Enns and Jardina Reference Enns and Jardina2021).

A handful of studies have used panel surveys to find that values have a positive impact on future issue stances, indicating that values do exert the causal role this article theorizes, but they suffer from important limitations (Goren Reference Goren2013; McCann Reference McCann1997; Peffley and Hurwitz Reference Peffley and Hurwitz1993). While their use of panels obviates concerns regarding time-invariant omitted variables, they do not eliminate concerns about time-varying omitted variables that may influence issue attitudes. Two of the most commonly posited forces of the hierarchical constraint of issue stances are partisanship and left-right ideology, but only Peffley and Hurwitz (Reference Peffley and Hurwitz1993) control for these variables and compare the relative effects of different predispositions. This makes it hard to conclude that values are driving issue stances as opposed to partisanship or ideology. These studies also rely on a single panel each, which limits the findings’ external validity.Footnote ⁷ The question of generalizability is especially salient in twenty-first-century American politics since ideology and partisanship have become increasingly correlated with issue stances (Bafumi and Shapiro Reference Bafumi and Shapiro2009). Thus, it may be that values are no longer a necessary heuristic for the development of issue stances.

I generate three hypotheses to test the effects of values. In each of these hypotheses, I compare the effects of values to those of partisanship and ideology. If the effects of values are no longer statistically significant after accounting for these variables, then partisanship and ideology might instead be the drivers of issue attitudes. I also compare the coefficients on each to see whether values play a larger role than partisanship or ideology.

I begin with the basic across-voter relationships between values and issue stances. While existing work says that we ought to expect a respondent who is more traditionalist in her values to hold more value-congruent policy views, we do not know whether her attitudes instead stem from unmeasured characteristics such as her upbringing. By introducing respondent fixed effects, we hold constant many factors that might influence the individual’s attitudes. I posit that there is a within-individual correlation between values and issue stances (H1) – that is, if we compare two different snapshots for the same individual, the snapshot in which she espouses more egalitarian values will also be the one in which she favours policies that enhance societal opportunity relative to the other snapshot.

H1 does not account for the temporal dynamics of the panel, though. To more thoroughly evaluate the causal role of values, I test whether values pull issue stances towards congruence with the value over time (H2). Put simply, if someone is more traditionalist at time 1, H2 posits that her attitudes at time 2 will be more aligned with traditionalism after accounting for her attitudes at time 1. If H2 is true, it would be indicative of a higher-order belief – namely, values – ‘pulling’ specific attitudes towards the position consistent with the value. This is the type of effect we would expect to see if values operate hierarchically on attitudes. Figure 1 shows two examples of the hypothesized relationship.

Figure 1. Two examples of effects under H2, where ${p_2}$ (policy attitudes at time 2) are a function of ${p_1}$ (policy attitudes at time 1) and ${v_1}$ (values at time 1). Red arrows reflect non-zero regression to the centre point (of 0) from ${p_1}$ , and blue arrows reflect a positive effect of ${v_1}$ .

The first two hypotheses rely on changes in values and policy stances in order to achieve identification. While these hypotheses constitute evidence towards a causal role for values, they cannot detect causal effects among those whose values and attitudes are unchanging during the panel, even if their issue stances do, in fact, derive from their values. My third hypothesis addresses this limitation by testing whether values shape attitudes in new policy domains on which voters are less likely to have had stable and well-considered policy views at the time of the first survey wave (H3).

Data and measurement

I employ three panel studies from the American National Election Studies (ANES): one with observations in 1992, 1994, and 1996; one with observations in 2000, 2002, and 2004; and one with observations in 2016 and 2020. I also use three panel studies from the General Social Survey (GSS): the first with observations in 2006, 2008, and 2010; the second with observations in 2010, 2012, and 2014; and the third with observations in 2016, 2018, and 2020 (the third consists of two two-wave panels, wherein half of the 2020 respondents were originally surveyed in 2016 and the other half were originally surveyed in 2018). In analyses with data pooled across survey waves, I use inverse weights such that data from each survey wave is given equal weight.

I focus on two values which have been studied extensively in the values literature: moral traditionalism and egalitarianism. Traditionalism refers to the value of sustaining historically received traditions in order to maintain a virtuous society. The family, community, and organized religion are core institutional pillars of traditionalism (Goren Reference Goren2013). Egalitarianism refers to the importance of ensuring equality of opportunity and giving everyone a fair chance to succeed in society (Feldman Reference Feldman1999). I test these values for two reasons. The first is that we have large-scale panel surveys testing these two values, enabling thorough tests of the effects of values on issue stances. The second is that these values are not influenced by lagged partisanship or ideology in the ways that some other values are in experimental and panel studies (Goren Reference Goren2005; Goren et al. Reference Goren, Federico and Kittilson2009; Lupton and McKee Reference Lupton and McKee2020).Footnote ⁸ This suggests that these values are stable and enduring concepts.

I use four questions in the ANES for traditionalism and five for egalitarianism, along with six from the GSS for traditionalism (the GSS does not track egalitarianism). The ANES and GSS ask different types of questions about traditionalism – while both tap the same concept, the ANES questions exclusively use broader language like ‘traditional family ties’ and ‘new lifestyles’. On the other hand, the GSS has a mix of broader questions (for example, the permissibility of different types of personal morality) and more specific questions (for example, whether sex before marriage is wrong). As such, I separate out the analyses between the ANES and GSS. By using two different operationalizations of traditionalism, I ensure that the findings do not derive from a specific set of values questions.

I explore the effects of values on issue stances that are conceptually linked to the value. This is because values are ‘domain-specific organizing principles’ (Zaller Reference Zaller1992, 26), which means that they have particular applications when dealing with attitudes tied to that domain. Thus, as the extant literature assumes, the relevant question is whether values drive beliefs over policies that are substantively related to those values.

Traditionalist values relating to the morality of certain behaviours and changing lifestyles bear most clearly on policies that instantiate those beliefs into law (for example, legal restrictions on unorthodox lifestyles). In this vein, past work has found relationships between traditionalism and opposition to LGBT rights (Brewer Reference Brewer2003). I construct a policy index that contains questions relating to LGBT issues, abortion issues, and certain school issues. Following the terminology of Goren and Chapp (Reference Goren and Chapp2017), I label this the ‘cultural orthodoxy’ policy index. For example, this linkage suggests that those who favour maintaining traditional families and oppose new lifestyles will oppose same-sex marriage or legalized abortion. Especially insofar as traditional society reflected a Christian consensus, those who hold such values may also favour policies such as school prayer or vouchers for private schools (which are often used toward religious schools).

I also construct an opportunity policy index, which contains items relating to policies that improve access to basic resources and attempt to generate social mobility. This linkage is consistent with past work showing relationships between egalitarian values and attitudes on education spending (Feldman Reference Feldman1988). I use questions relating to the basic social safety net, educational opportunity, and non-discrimination in employment. Those who believe that we need to do more as a society to give everyone an equal chance to succeed might be more likely to favour government spending on financial aid or non-discrimination laws, for example. Notably, this index does not contain items on policies such as transportation spending or climate spending – as such, it is not merely a measure of ‘fiscal liberalism’.

While the value questions reflect voters’ beliefs about the priorities that society should have, one’s values do not necessitate landing on a given side of a policy debate. One could favour the maintenance of traditional family ties while supporting non-discrimination laws against LGBT individuals; likewise, one could favour equality of opportunity while opposing increased spending on the poor.

The question wordings for each item are in Appendix Section A, along with scale reliabilities and stabilities. In order to provide easily interpretable coefficients, I create linear additive scales of values and domain-specific policy views that range from −0.5 to 0.5. I scale the standard partisanship and ideology questions into the same unit domain. This approach enables straightforward interpretation of a shift in values, ideology, or partisanship on policy views. However, it implicitly assumes equal weight of all the items in a given index. In Appendix Section C, I run robustness checks that load the items onto factors rather than using additive scales (using feeling thermometers to construct three-item scales for partisanship and ideology) and find similar results. In each of this model’s articles, I also test the effects of values on those of ideology and partisanship. While the unit scale invites clear interpretation, the variables may have different variances, and so in the coefficient comparison tests, I test significance using standardized models.

Do values correlate with policy attitudes for a given voter?

Existing work shows that values correlate with domain-specific policy attitudes, but these models may suffer from omitted variable bias. I first replicate the cross-sectional models by estimating regressions of the form ${Y_{it}} = {\alpha _t} + {\beta _1}{V_{it}} + {\beta _2}{I_{it}} + {\beta _3}{P_{it}} + \xi {X_{it}} + {\varepsilon _{it}}$ where ${Y_{it}}$ indexes policy attitudes, ${V_{it}}$ is values, ${I_{it}}$ is symbolic ideology, ${P_{it}}$ is party identification, X is a matrix of demographic covariates (race, gender, income, and education), and ${\alpha _t}$ are wave fixed effects (since the data is pooled across multiple years). To discern the degree of omitted variable bias, I replicate these models but include a term ${\gamma _i}$ as individual fixed effects. These models attempt to identify effects by leveraging within-individual variation over time (Allison Reference Allison2009). I cluster standard errors at the individual level. I use data from all the survey waves described above except for the 2002 and 2004 ANES waves since those do not ask values questions.

In Table 1, the odd-numbered columns show across-voter models and the even-numbered ones show within-voter models. Beginning with the former set, the odd-numbered columns in Table 1 show substantively large correlations between values and issue stances. A maximally traditionalist respondent in the ANES has policy views 0.44 units more culturally orthodox than one who is minimally traditionalist (holding constant her partisanship, ideology, and basic demographics). Put in the context of one of the underlying policy items, a 0.44 unit shift is nearly equal to the difference between opposing any recognition of same-sex couples and permitting civil unions (0.5 units). The equivalent coefficients are 0.31 for egalitarianism in the ANES and 0.52 for traditionalism in the GSS. The coefficients on values are always larger than those of ideology and partisanship, and the gaps are always statistically significant except for the values-ideology gap in column 3.

Table 1. Cross-sectional relationships between values and policy views

Note: Data weighted to give each survey wave equal weight. All models include survey wave fixed effects and demographic controls. Standard errors clustered at the respondent level. Ideology and partisanship variables are coded such that higher values are more conservative/Republican for cultural orthodoxy policy models and more liberal/Democratic for opportunity policy models. * indicates p < 0.05 (two-tailed tests).

The coefficient sizes attenuate when we introduce respondent fixed effects in the even-numbered columns. These models suggest that the across-voter estimates are contaminated by a considerable degree of omitted variable bias. Despite this, the even-numbered columns of Table 1 show that if a given survey respondent is a minimal traditionalist in one ANES wave and a maximal traditionalist in another, her policy views will be 0.08 more units more culturally orthodox in that latter wave (column 2). These effect sizes are 0.12 for egalitarianism and 0.09 for GSS traditionalism (columns 4 and 6). While such within-individual swings are unlikely, we can benchmark the effect sizes by comparing them to those of ideology and partisanship. Although we see that when respondents identify as more Republican or more conservative, they also espouse more culturally orthodox policy views (and the same for Democrats/liberals on pro-opportunity policies), these coefficients are only significant in column 4. In each within-individual test, the values coefficient is larger than that of partisanship or ideology, and in three of the six comparisons, the difference is statistically significant.

These findings give us initial reason to believe that past literature on the relationships between values and issue stances does not arise merely due to omitted variable bias. However, these models treat the timing of the waves as essentially interchangeable; to more rigorously examine claims of causality, I look at how attitudes shift within individuals over time.

Do values drive changes in policy views?

H2 asks whether values induce changes in policy attitudes between one period and another. In this test, each row in the dataset corresponds to an individual, and so the model is now ${Y_{i,t+1}} = {\alpha _t} + {\beta _1}{V_{it}} + {\beta _2}{I_{it}} + {\beta _3}{P_{it}} + {Y_{it}} + {\varepsilon _{it}}$ where ${Y_{i,t+1}}$ reflects policy views measured two or four years later.Footnote ⁹ We should expect that the strongest predictor of current policy views is past policy views in any given domain. The main question here, though, is whether values influence policy views after accounting for attitude stability. Are these attitudes being drawn towards voters’ values, as Figure 1 posits? And if this is the case, does this force appear above and beyond the extent to which voters’ attitudes are pulled towards their partisan and ideological identities?

Table 2 shows the results.Footnote ¹⁰ The first two columns show that a one-unit increase in traditionalism in the ANES models leads to 0.18 and 0.12 unit increases in culturally orthodox policy views two and four years later, respectively. The third and fourth columns indicate that a maximal egalitarian’s policy views become 0.16 units more pro-opportunity both two years later and four years later relative to a minimal egalitarian’s. Finally, the GSS models show that maximal traditionalists shifted their policy stances 0.23 units in the culturally orthodox direction relative to minimal traditionalists both two and four years later. This effect size is equivalent to a one-response option difference on the same-sex marriage question in the GSS cultural orthodoxy scale (for example, going from somewhat opposed to same-sex marriage to being neutral on the question).

Table 2. Relationship between values and changes in policy views

Note: ANES data pooled from three two-year panels (92–94, 94–96, 00–02) and three four-year panels (92–96, 00–04, 16–20). GSS data pooled from five two-year panels (06–08, 08–10, 10–12, 12–14, 18–20) and three four-year panels (06–10, 10–14, 16–20). Data was weighted to give each survey wave equal weight. All models include survey wave fixed effects. Standard errors clustered at the respondent level. Ideology and partisanship variables are coded such that higher values are more conservative/Republican for cultural orthodoxy policy models and more liberal/Democratic for opportunity policy models. * indicates p < 0.05 (two-tailed tests).

In each of the models, symbolic ideology generates shifts in policy stances; for example, more conservative individuals’ views shift towards cultural orthodoxy. However, the coefficient on partisanship is only significant in three of the six specifications. In other words, once we account for values and ideology, we do not see Republicans and Democrats shifting their viewpoints in the culturally orthodox or pro-opportunity directions. Table 2 also shows that in every case, the magnitude of the values coefficient is larger than that of partisanship and ideology. The difference between values and ideology is not significant in any of the ANES models but is significant in both of the GSS traditionalism models. The gap between values and partisanship, though, is significant in every model except for the two-year egalitarianism panel (column 3).Footnote ¹¹ The difference between values and partisanship is especially notable insofar as it runs counter to the theory that partisanship is the primary driver of policy views (for example, Achen and Bartels Reference Achen and Bartels2016).

I verify these results by testing each of the two-wave panels separately in Appendix Section D. The results in that section show that the effects of values relative to those of ideology and partisanship remain consistent over time. This mitigates concerns that values mattered in the 1990s but have since been supplanted by partisanship or ideology. Even in the 2016–2020 ANES panel, the coefficient on values is larger than that of both partisanship and ideology in three of the four comparisons at the $\alpha =0.05$ level.

One limitation of testing differences in policy stances between time 1 and time 2 is that values and issue stances could simultaneously shift. As a refinement of H2, then, I use the three-wave panels to test whether changes in values affect subsequent changes in policy views – that is, whether shifts in values between times 1 and 2 drive changes in the respondent’s policy views in time 3. Lenz (Reference Lenz2012) employs a version of this test to causally identify the effects of persuasion on political evaluations. This test is overly strong insofar as a values shift could generate an ensuing attitude shift before the second wave is captured. To take an example with the 2010-2012-2014 panel, if values shift between 2010 and 2011 and then attitudes shift between 2011 and 2012, this test would fail to uncover an effect on 2014 attitudes because the effects had already been completed by the 2012 wave. This test also identifies effects only among those whose values change. Nonetheless, I find evidence in Appendix Section E that shifts in values drive subsequent shifts in issue attitudes. Furthermore, these shifts are larger than the effects stemming from equivalent shifts in ideology and partisanship (although the gaps are significant in only one of the four tests).

Across a variety of tests and robustness specifications, it appears that values drive the development of issue stances, and these effects are at least as large, if not larger than those of ideology and partisanship.

Do voters use values to develop issue attitudes?

Welfare reform

H3 tests the role of values in shaping attitudes in emergent policy domains. Welfare policy in the 1990s presents a useful case study because we see a steep rise in the salience of the issue combined with a rapid change in terms of the policy debate. Only 7 per cent of Americans in 1992 considered welfare one of the country’s most important problems, but 27 per cent did so in 1996 (Soss and Schram Reference Soss and Schram2007), due in large part to a sharp rise in elite discussion of welfare between 1990 and 1995 (Schneider and Jacoby Reference Schneider and Jacoby2005). The policy terrain also shifted drastically in this period. While welfare reform was not a new political issue, Bill Clinton’s promise to ‘end welfare as we know it’ in 1992 shifted the debate from preserving the status quo program (Aid to Families with Dependent Children) to devising new policy solutions, culminating in the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (Weaver Reference Weaver2000).

On the elite side, we see a sharp increase in public discourse not just regarding welfare generally but with respect to policy specifics, as Figure 2 shows. Coverage of ‘welfare’ in 1995 is 1.9 times that of coverage in 1990, but usage of policy-specific phrases ranges from 7.4 to 28 times higher in 1995 relative to 1990. Public opinion on particular policy components is initially volatile but stabilizes by the mid-1990s. For example, support for family caps increased by almost thirty percentage points between early 1992 and late 1993 before stabilizing in 1995 (Weaver Reference Weaver2000, 179).

Figure 2. Newspaper frequency of different welfare-related terms by year from newspapers.com. Each term count is divided by the frequency of ‘Monday’ to index relative to aggregate newspaper volume (Beach and Hanlon Reference Beach and Hanlon2022), and the ensuing ratio is indexed to the 1990 ratio for that term.

Existing work shows cross-sectional correlations between values and welfare attitudes (Berinsky Reference Berinsky2002; Feldman and Zaller Reference Feldman and Zaller1992; Feldman and Steenbergen Reference Feldman and Steenbergen2001), but the 1992-1994-1996 ANES panel enables us to see how lagged values in 1992 and 1994 drive 1996 policy attitudes. Voters had an opportunity to learn about the policy programme from 1992 to 1996, and one way that they could do so was by applying broadly applicable values (for example, equality of opportunity) to particular policy debates (for example, not restricting eligibility for welfare programs). To test this, I construct an index from the ANES welfare policy questions (see Appendix Section A). Higher levels indicate support for increases in spending and more lenient rules for eligibility (that is, opposition to the reforms).Footnote ¹² I compare the effects of egalitarian values to those of partisanship and ideology on these attitudes. Schneider and Jacoby (Reference Schneider and Jacoby2005) use the same ANES panel to show that ideology and partisanship drive shifts in welfare policy attitudes – the question here is whether they continue to do so once voters’ values are taken into account.

Table 3 shows the results of predicting welfare attitudes from prior predispositions and attitudes. The empirical specification is similar to that of Table 2, except it employs welfare policy questions rather than opportunity policy ones. Column 1 indicates that maximal egalitarians in 1992 developed attitudes that were 0.16 units more opposed to welfare reform by 1996 relative to minimal egalitarians. While symbolic ideology in 1992 predicts subsequent opposition to welfare reform, partisanship is not significant after accounting for ideology and values, suggesting that voters were not following partisan cues. The muted effects of partisanship perhaps stem from elite Democrats’ ambiguous positions on welfare. The second column shows that 1994 egalitarianism, ideology, and partisanship all drive 1996 views on welfare policy. We see similar coefficients on the predispositions across models but a marked increase in the effects of lagged attitudes. This suggests that voters were developing views on the topic between 1992 and 1994 – perhaps due to the rising salience of the issue as a component of the Republican’s Contract with America in 1994. In each model, values play a major role in the development of 1996 welfare attitudes, but we cannot say that the effect is greater than that of ideology or partisanship at the $\alpha =0.05$ level.

Table 3. Relationship between values and future views on welfare reform

Note: Ideology and partisanship variables are coded such that higher values are more liberal/Democratic. * indicates p $ \lt $ 0.05 (two-tailed tests).

Of course, other values such as hard work and individualism surely influence welfare attitudes along with predispositions such as racial group attitudes and external factors such as racialized media coverage (Gilens Reference Gilens1996). I account for one of these factors in Appendix Table F.5 by showing that the results hold after controlling for racial group attitudes. At a minimum, it appears that egalitarian values offered voters one tool with which they could develop attitudes towards welfare policy in the 1990s.

Transgender policies

Although debates around transgender issues have existed for decades, they became increasingly salient in the 2010s. President Obama signed Executive Order 13672 in 2014, which prevented discrimination on the basis of gender identity in federal hiring and government contracting. In 2016, his administration directed public schools to permit transgender students to use bathrooms corresponding to their gender identities and reversed the ban on transgender people openly serving in the military. In 2017, the Trump administration reinstated the ban, in 2021, the Biden administration revoked Trump’s ban renewals, and in 2025, the Trump administration once again reinstated the ban. These questions have also arisen in state politics, as North Carolina famously passed a law in 2016 requiring individuals to use the bathroom corresponding with the sex listed on their birth certificate. The law produced outcry from activists and corporations, and the state partially repealed the law. In recent years, policy fights have arisen over gender reassignment surgeries for minors and participation by transgender individuals on women’s sports teams.

During this period, transgender rights became a core tenet of the Democratic Party’s platform. In the 2012 platform, a single sentence mentions gender identity as one of many dimensions along which the party opposes discrimination. The issue becomes more prominent in the 2016 platform, with two subsections relating to LGBT rights and references to particular issues (for example, transgender individuals being denied service at restaurants). In the 2020 platform, not only is there a section on LGBTQ+ rights, but policy questions relating to transgender people and gender identity appear in the sections on the military, immigration, education, employment, housing discrimination, criminal justice, homelessness, and hate crimes, among others. In short, transgender policy goes from a single sentence in the 2012 platform to infusing almost every policy domain in the 2020 platform.

Figure 3 shows that newspaper coverage of these issues rose exponentially from 2010 to 2016 before tapering off. While discussions of the terms transgender and bathroom in tandem have subsided relative to their 2016 peak, most of these terms in 2022 are near or past their 2016–2017 highs. The 2016–2017 coverage peak reflects the media shining attention on an issue that had previously received little coverage, but the tapering off in media coverage should not lead one to conclude that the issue also peaked in salience in the electorate at that time. Rather, Google Search Trends indicates that the most commonly searched phrase with ‘transgender’ between 2010 and 2022 is ‘transgender’ + ‘meaning’, indicating that people are attempting to learn what the meaning of transgenderism is. Figure 4, showing search trends for those terms over time, reveals steadily increasing traffic for the search, with significant spikes in 2021 and 2022 relative to past years. This suggests that the public is increasingly learning about these issues.

Figure 3. Newspaper frequency of different transgender-related terms by year from newspapers.com relative to 2010 (see Figure 2 for methodology).

Figure 4. Google Search trends by month for ‘transgender’ + ‘meaning’.

It appears that transgender issues were a quickly emerging policy domain during the 2010s and one that Americans were still learning about. This offers a prime opportunity for traditional values (that is, opposing new lifestyles and changes in traditional gender roles) and egalitarian values (that is, treating all people equally before the law) to drive attitudes on these policies. I test whether these values, along with ideology, partisanship, and transgender policy attitudes, all measured in 2016, influence transgender policy attitudes in 2020. The ANES asks two questions with respect to transgender policy. The first asks whether transgender people should be allowed to use the bathroom of their choice (2016 and 2020), and the second asks whether they should be allowed to serve in the military (2020 only).

Table 4 shows the results. Traditional values are the largest driver of transgender policy attitudes in 2020 out of all four predispositions, even after accounting for lagged views. The coefficient is larger than that of egalitarian values, ideology, and partisanship in each model at the $\alpha =0.05$ level. This suggests that if we compare two otherwise identical voters in terms of egalitarianism, ideology, partisanship, and attitudes towards transgender people in bathrooms in 2016, a maximal traditionalist in 2016 would end up 1.5 response options more anti-transgender on the bathroom question in 2020 (for example, feels very strongly that transgender people should use the bathroom of their birth gender) compared to a minimal traditionalist (for example, landing between feels ‘a little strong’ and ‘moderately strong’ on the question). Similarly, the maximal traditionalist would, on average, respond 2 points more opposed to transgender individuals serving in the military relative to the minimal traditionalist (for example, oppose a moderate amount versus neither favour nor oppose). These effect sizes are quite meaningful given that we have already controlled for the respondents’ attitudes towards transgender bathroom policies in 2016. While egalitarianism also influences attitudes on transgender policy, it does so at rates that are comparable to ideology and partisanship.

Table 4. Relationship between values and future views on transgender policy

Note: Ideology and partisanship variables are coded such that higher values are more conservative/Republican. The egalitarianism variable is reverse-coded such that higher values indicate lower degrees of egalitarianism. * indicates p < 0.05 (two-tailed tests).

One possibility is that welfare reform and transgender policy issues constitute the most likely causes for the effects of values since they are ‘easy’ policies for politicians to connect to values such as traditionalism, individualism, and egalitarianism. However, politicians can make these connections even on technically complex and ‘hard’ legislation, as they did on the Affordable Care Act with egalitarianism and freedom. Lee and Culpepper (Reference Lee and Culpeppern.d.) show that egalitarian values are one of the strongest correlates of voters’ policy attitudes on the more complex topic of financial regulation. In sum, it appears that values are a key tool for voters to develop opinions on new issues, making them a central component of voters’ belief systems.