Abstract
Conventional wisdom suggests that partisanship is the “unmoved mover” in the minds of American voters. Revisionist theories hold that party updating is conditional on individual/contextual factors. By delimiting the scope conditions of the Michigan model, revisionist models do not fundamentally challenge the classic view. This paper proffers an unconditional model of party revision. We theorize that beliefs about government activism—operational ideology—are widely available and heuristically efficacious, and easily map onto party labels. Hence, operational ideology should drive party updating. Using data from seven panel studies covering 1990–2012, we demonstrate that (1) party shapes operational ideology, (2) operational ideology shapes party, (3) the ideology-to-party effects are as large as the party-to-ideology effects, and (4) neither sophistication nor education or elite polarization condition these relationships. These results push the revisionist model of party farther than it has gone before and suggest that operational ideology is a core predisposition in mass belief systems.
Party identification represents a bedrock predisposition in the minds of citizens. Classic and contemporary works establish that partisan attachments develop early in the life cycle, grow stronger as the years roll by, hold steady over time, and wield pervasive influence over political judgment (Bartels 2002; Campbell et al. 1960). Partisanship biases views of presidential performance and political candidates. It distorts perceptions of seemingly uncontestable facts. Perhaps most troubling, blind loyalty to a party colors views about public policy. If the merits of a proposed government action matter less than the party backing it, we cannot put much stock in citizens’ policy “preferences” as valid inputs into the democratic process.
But perhaps this goes too far. Research that exploits the power of panel data to uncover evidence of attitude change suggests that party ID responds to policy views under limited conditions. In a pioneering study, Carsey and Layman (2006) show that people who know where the parties stand on single issues they care about bring their partisan identities in line with their policy positions. Yet even among this limited subset of voters, the effects of partisanship on issues surpass those of issues on party. Dancey and Goren (2010) discover that during periods of intense elite debate, citizens update their party ID to better reflect their preferences on emotionally charged issues such as welfare and gay rights, but again, party-based updating of issues outpaces issue-based party revision. Abramowitz and Saunders (1998) and Highton and Kam (2011) show that issues shape party more than the reverse, but because each study relies on a single panel survey that ended in the mid-1990s, it remains unclear whether these results speak to the endogeneity of partisanship as a general rule or to the endogeneity of partisanship during Bill Clinton’s volatile first term. In sum, although policy views seem to motivate partisan revision for some individuals in some situations, the party-to-policy link remains the leading causal pathway. Therefore, revisionist studies do not alter the paradigmatic view that party dominates policy for most people most of the time. Instead, these works suggest that party revision is conditional on individual and contextual variables (cf. Groenendyk 2013).
In this paper, we develop a novel theory that posits that operational ideology functions as a core political predisposition that drives party updating for most citizens in the contemporary United States. The theoretical crux of our argument holds that all citizens possess genuine and durable views about whether the federal government should ensure some measure of security against the vicissitudes of market life, that operational beliefs about government occupy a central node in belief systems alongside party identification, and that these beliefs have the power to shape a wide range of political evaluations (Ellis and Stimson 2012; Goren 2013). We theorize that operational beliefs about government are uniquely positioned to shape—as well as be shaped by—partisan affinities. Breaking with the standard revisionist theories, we propose that the propensity to ground partisanship in beliefs about government activism is not conditional on sophistication, education, or elite polarization. Instead, most citizens regularly bring their partisan attachments into line with their operational beliefs about government.
Using data from seven American National Election Studies (ANES) and General Social Survey (GSS) panels covering the years 1990–2012, we furnish strong support for our theoretical conjectures. Three key findings emerge from our efforts. First, operational ideology and party identification constrain one another over time. Second, the influence of operational ideology on party ID usually matches and sometimes eclipses that of party on operational ideology. Third, political knowledge, education, nor elite polarization moderates these relationships. Our results suggest that a greater appreciation of the theoretical significance of operational ideology is warranted, along with heightened sensitivity to the endogeneity of partisanship.
Concepts
Classic and contemporary accounts define party identification as an enduring psychological attachment to a political party that shapes political perception and judgment (Campbell et al. 1960). Party identification functions as a strong predisposition that guides information processing and decision making across multiple contexts. Democrats and Republicans perceive, interpret, evaluate, and integrate new information in a manner that yields pro-party conclusions. For these reasons, the causal arrow runs primarily from party ID to political attitudes and beliefs (Bartels 2002).
This does not mean that party lies beyond the influence of all political beliefs. In revisionist frameworks, party ID behaves like a “running tally” whereby citizens adjust their partisan attachments to ensure correspondence with their stands on issues (Franklin and Jackson 1983) or retrospective judgments about the economy and foreign affairs (Fiorina 1981). Recent scholarship affirms this model. In an exemplar of this work, Carsey and Layman (2006) demonstrate that respondents who care passionately about a single issue and know where the parties stand on it bring their party ID in line with preferences on this issue. However, even when evidence of partisan endogeneity arises, the effects of party on policy outweigh the reciprocal effects of policy on party. And for people who know the party position on issues they do not care about, party dominates issue opinions. To their credit, Carsey and Layman (2006, 474) concede the centrality of partisanship without equivocation: “we view our findings as further confirmation that party identification is a ‘moving force in politics,’ that tends to be moved itself only in special circumstances” (emphasis added). Without denying the importance of this qualification, it is clear that this study provides some of the best evidence to date that many people revise their partisan identities when they have reasons for doing so.
Similarly, Highton and Kam (2011) apply structural equation modeling techniques to data from the parent and youth samples of the 1973–1982 and 1982–1997 waves of the Political Socialization Panel Study to adduce support for both classic and revisionist models of party ID. They show that the classic model describes the party–issue relationship from 1973 to 1982 for the parent and youth samples, whereas the revisionist model applies in the 1982–1997 youth panel (parents were not reinterviewed in 1997). The latter result obtains presumably because the rise of elite polarization that began in the 1970s provided citizens with clear-cut policy cues that facilitated issue-based party updating. The authors conclude that the party–issue relationship is context dependent. These results are intriguing, but a word of caution is in order. As Highton and Kam (2011, 208, n. 21) note, as the youth sample was drawn from the population of high school seniors in 1965, it is unrepresentative of those who did not finish high school. So far as education proxies for responsiveness to elite signals, the results obtained in the youth panel may not generalize to the broader electorate. Once again, issue-based party updating may be confined to a politically engaged subset of the population.
Space limitations preclude a full discussion of every study that addresses party–issue dynamics. At this juncture, we note that most of this work finds that (1) the effects of party on issues are stronger than the effects of issues on party, (2) the magnitude of these effects increases in response to some individual or contextual moderators, and (3) even under these conditions, the results lend stronger support to the classic model of partisanship than to the revisionist model (Bartels 2002; Dancey and Goren 2010; Goren 2005; Levendusky 2009; but see Abramowitz and Saunders 1998). As such, revisionist models of party ID are conditional. These models posit that people adjust their partisan affinities under special circumstances. Under normal circumstances, party dominates policy.
Against this view, we argue that operational ideology drives party ID as a general rule in the contemporary era. Said otherwise, the influence of operational ideology on party ID is universal rather than conditional. Before elaborating, we need to define operational ideology. Over the years, scholars have proffered several distinct conceptualizations of ideology, including ideological sophistication, ideological constraint, and policy ideology (Jessee 2012). We stress the narrowly defined dimension of operational ideology, by which we mean the degree to which one believes that the federal government should pursue a more or less active role in the economic and market life of the country (Ellis and Stimson 2012). Operational ideology represents a general belief rather than a discrete issue preference because it revolves around claims about whether the federal government should look after the material well-being of the American people rather than on behalf of a specific group (e.g., the working poor) or through a particular policy (e.g., the minimum wage). Operational liberals favor a strong government mobilized to ensure economic opportunity and security for the people, whereas operational conservatives prefer a smaller government that knows it limits. Said another way, operational liberals want more government intervention to help people cope with the market economy, whereas operational conservatives seek less government intervention. 1
We propose that operational ideology functions as a core predisposition for most citizens. If this is true, people must hold genuine and durable beliefs about government activism that guide political judgment and behavior. There are compelling reasons to believe that citizens satisfy these conditions. To begin, questions about government activism have divided political elites and the national parties since Franklin Roosevelt and Congressional Democrats built the welfare state in the 1930s to combat the destruction wrought by the Great Depression. Thereafter, recurring conflict over government activism has animated domestic politics like no other cluster of issues. Without gainsaying the significance of other controversies that have periodically exploded onto the national agenda, few can approach, let alone match, the reach of the welfare state as an enduring political cleavage (Ellis and Stimson 2012). Moreover, the ideological distance between the national parties over the federal government’s role has increased sharply since the 1970s (McCarty, Poole, and Rosenthal 2008), thereby further elevating the salience of this dimension in the public mind.
The visibility of a given policy cleavage carries no assurance that people will comprehend the stakes involved. This begs the question of whether most people understand the idea behind the cleavage. For government activism, we posit that they do. At heart, operational ideology poses a simple, stark choice: should the federal government become more or less involved in the economic welfare domain. There is no confusion over what it means to say government does too much or not enough. Citizens are not required to master a wealth of intricate details about rival policy proposals or the best technocratic means to achieve a desired policy end. In short, we think that most people hold real beliefs about operational government power that shape political judgment.
Having defined operational ideology, we now explain how it differs from other policy orientations. To some, it may seem that operational ideology is synonymous with liberal–conservative identification. This is not the case. Research demonstrates that liberal–conservative identifications (i.e., symbolic ideology) are evaluations of religious, social, and political symbols associated with these labels (Conover and Feldman 1981; Ellis and Stimson 2012). Some people equate these labels with “gays” or “big business” or “Bible thumpers.” For others, the terms evoke images such as “wasteful taxes,” “permissive,” “traditional,” and so on. That is, liberal–conservative identities are symbolic predispositions largely devoid of policy content. By contrast, operational ideology is all about policy. Indeed, it is hard to imagine a more fundamental policy question than whether the federal government should be more or less involved in public life. Last, operational ideology differs from issue preferences. The latter can be defined as positional views about a specific controversy (e.g., gun control). Given that an issue references a single policy, whereas operational ideology emphasizes government activism across multiple issues, the conceptual distinction should be clear.
Theoretical Framework
How might party identification and operational ideology influence one another in a dynamic sequence? We begin with the “partisan influence” hypothesis. Leading theories of partisan reasoning maintain that political source cues activate latent partisan predispositions that guide information processing and decision making in a straightforward manner (Goren, Federico, and Kittilson 2009; Zaller 1992). Party sources signal how partisans should evaluate the political world and what they should believe about public affairs. If the message source and message recipient share a party label, the latter usually follows the party line without assessing the merits of the proposal. But if the source and target identities conflict, the recipient will likely reject the communication without much thought. This line of reasoning suggests that in an information environment flush with partisan cues, individual-level partisanship will dynamically constrain the positions people take on operational liberalism/conservatism (Dancey and Goren 2010). Democratic identifiers should adopt the pro-government position advocated by party leaders, whereas Republican followers should toe the Grand Old Party (GOP) leadership line for smaller government (Hypothesis 1).
We now take up the “operational ideology influence” hypothesis. Prior work indicates that some general principles (e.g., equal opportunity) exercise little influence over party bonds (Goren 2005). Why should operational ideology be any different? We posit that operational ideology is a central heuristic that facilitates efficient and effective political judgment across a range of decisions. When constructing issue preferences citizens need not work through the pros and cons of whatever policy proposals capture their attention. The less taxing way to proceed is by choosing the alternative more consistent with one’s broader views about government activism. Likewise, when it comes to electoral choice citizens need not evaluate a multitude of rival proposals advanced by candidates to decide who would be a better fit on policy grounds. Voters can quickly render quality decisions by comparing their positions to each candidate on the operational liberalism/conservatism dimension. In short, we posit that operational ideology is better situated than other policy orientations to shape partisan identities. Indeed, given the centrality of questions about the size and reach of government in American political history, American political culture, and contemporary elite debate, we suspect that most if not virtually all citizens have developed genuine beliefs about the role of government.
If our reasoning is on the mark, citizens across the sophistication spectrum should hold genuine beliefs about government activism and use these actively. Accumulated research supports this view. For example, Feldman and Zaller (1992) show that the sophisticated and unsophisticated rely equally on such beliefs to constrain their positions on political issues. Goren (2013) applies measurement modeling techniques to data from multiple ANES surveys covering a twenty-year period to demonstrate that unsophisticated, moderately sophisticated, and highly sophisticated individuals hold structured beliefs about government activism. He further shows that these beliefs are very stable over time and drive candidate preferences in multiple presidential elections for citizens at different levels of sophistication. In conjunction, these and other works that examine macro opinion (Enns and Kellstedt 2008) indicate that beliefs about operational governmental power are meaningful, stable, and consequential for all citizens.
Given that party ID is not immutable, operational ideology is central to political judgment, and an information environment in which both are readily and repeatedly linked to one another, we predict that people who discern value in government efforts to ameliorate economic uncertainty and market externalities should come to identify more strongly with the Democratic Party, whereas those who disdain big government should find the GOP team more appealing (Hypothesis 2). After testing these “main effects” hypotheses, we turn to the question of whether two individual and situational variables (i.e., political sophistication and elite polarization) moderate the dynamic relationship between operational ideology and party ID.
Data, Measures, and Estimation
We use data from the 1990–1992, 1992–1994, and 1994–1996 ANES panels and the 2006–2008, two 2008–2010, and 2010–2012 GSS panels to explore attitude stability and change in cross-lagged models of the party–policy relationship. 2 We measure party ID with the standard 7-point self-categorization scale, which ranges from 0 (strong Democrat) to 6 (strong Republican). Turning to operational liberalism, we follow Ellis and Stimson (2012) by using items that ask respondents whether federal spending should be increased, decreased, or kept about the same on a variety of government programs, such as welfare, food stamps, aid to the poor, the homeless, aid to blacks, public education, health care, the environment, Social Security, and so on (see Online Appendix A at http://prq.sagepub.com/supplemental/ for question wording). The only exception to the three-category response items is the ANES government services question, where respondents are given seven response options. We use a linear transformation that scales the index to run from 0 (strong operational liberal) to 6 (strong operational conservative). 3 Because we use multiple indicators to tap beliefs about government activism (five to eight items in a given survey), these scales are also reliable. Cronbach’s alpha reliability coefficient varies from .75 to .81 across the seven panels.
How might we estimate the dynamic relationship between party ID and operational ideology? Green and Palmquist (1990) show that the failure to account for random measurement error in the predictors of party updating can lead to erroneous inferences because the parameter estimates for error-laden predictors are biased and inconsistent. Building on this insight, many works deploy structural equation modeling (SEM) techniques to evaluate dynamic relationships between latent variables uncontaminated by random measurement error (Carsey and Layman 2006; Goren 2005; Highton and Kam 2011). However, other researchers have sharply criticized the SEM approach because the technique relies on hidden and untestable assumptions about the distribution of latent variables and the structure of the error terms (Achen 1983; Levendusky 2009; Luskin 1987). In addition, some critics argue that to the extent unexplained variation in survey responses results from faulty respondent who lack crystallized attitudes rather than from faulty questions, SEM techniques “overcorrect” the estimates, and convey a misleading impression about the stability and structure of public opinion (Converse 1980; Zaller 2012). Ansolabehere, Rodden, and Snyder (2008, 216) summarize scholarly skepticism toward SEM techniques as follows: “Confronted with complex structural models with many layers and parameters, skeptical readers see an unintelligible black box and are left with the impression that the findings have been manufactured by technique.”
In our judgment, both sides in the debate raise legitimate points. Until we have a firmer grasp on how to apportion random measurement error in survey responses between respondents and questions, it seems risky to base one’s inferences on the results of a single estimator. In light of this, we estimate cross-lagged equations using ordinary least squares (OLS) and errors-in-variables (EIV) regression. The EIV estimator uses reliability calculations to adjust the parameter estimates to reflect the relationship between “true” scores on the latent variables. We use alpha reliability estimates as inputs into the EIV estimator for operational ideology. For party ID we estimate Wiley-Wiley models on the three-wave panels to obtain error variance estimates. With the error variance estimates in hand, we then calculated the reliability of party ID. Because we lack a third panel wave in the 1990–1992 ANES, we use the party reliability estimate from the 1992 wave of the 1992–1994–1996 ANES panel as an input into the 1990–1992 EIV model.
By applying different estimators to multiple data sets covering the past three decades of political experience, we can assess the degree of empirical support accruing to the conventional and revisionist theories. Should either hypothesis receive consistent support across the OLS and EIV results, we can have greater confidence in the results than if support is confined to a single estimator. 4
Statistical Analysis
If the “partisan influence” hypothesis holds, we should observe significant coefficients for lagged party ID on current operational ideology, holding lagged ideology and demographic controls constant. 5 If the “operational ideology influence” hypothesis is not incorrect, we should uncover evidence that lagged operational ideology shapes current party ID, ceteris paribus. Recall that both party ID and operational ideology lie on a 0–6 scale, and are keyed so that higher scores denote stronger Republican identification and operational conservatism. Therefore, we expect positive coefficients in the regression models.
The first set of results comes from the ANES panel covering the period 1990–1992. Table 1 contains the OLS and EIV cross-lagged regression estimates. As called for by the partisan influence hypothesis, in the OLS model, lagged party systematically affects current beliefs about government, ceteris paribus (
Cross-Lagged Models for the 1990–1992 ANES Panel, OLS and EIV Estimates.
Source. 1990–1992 ANES panel.
The standard errors are in parentheses and have been adjusted to account for the complex sample design. Party ID and operational ideology lie on a 0–6 scale, and are coded so higher scores reflect more Republican and operationally conservative responses, respectively. The cross-lagged coefficients have been highlighted to facilitate comparisons across equations in a given pair of panel models. Control variables have been omitted for clarity (full model estimates appear in Online Appendix B1). The first difference score reports the percentage difference in the predicted score for a given dependent variable between respondents scoring at the 95th and 5th percentile on shaded independent variable. For instance, in the first OLS equation, respondents at the 95th percentile on lagged party ID (6 = strong Republicans) are predicted to score 6.7% more conservative on operational ideology92 than respondents at the 5th percentile (i.e., 0 = strong Democrats). The 95th percentile for operational ideology is a score of 4.17 (lean conservative) and the 5th percentile is a score of 0.33 (strong liberal). ANES = American National Election Study; OLS = ordinary least squares; EIV = errors-in-variables.
p < .05 (one-tailed).
The EIV estimates in the last two columns affirm the OLS results. As indicated in column 5, once measurement error has been accounted for party ID90 predicts change in the expected direction on operational ideology92 (p < .001,) with lagged ideology and the controls held constant. Statistically, a 1-unit move in the Republican direction in 1990 translates into a .04 increase on operational conservatism92, all else constant. Likewise, lagged operational ideology predicts current partisanship with a 1-unit increase in operational conservatism leading to a .19 shift in the GOP direction (p < .001).
Note that the magnitude of the operational ideology-to-party coefficient is nearly three times as large as the coefficient for party on ideology in the OLS model (.20 > .07) and almost five times as large in the EIV model (.19 > .04). These results suggest that the effect of operational ideology on party is much larger than the reciprocal effect of party on ideology. However, these comparisons obscure the fact that there is more variation in the actual distribution of party ID compared with that of operational ideology (see the standard deviations for each scale as reported in Online Appendix D). A naïve comparison of the raw coefficients obscures this fact.
To supplement the comparisons of the unstandardized coefficients, we report the first difference, in percentage terms, between the dependent variable score for respondents scoring at the 95th and 5th percentile on the lagged independent variable. In the OLS models, we see that moving from the 5th to 95th percentile on party ID (i.e., from strong Democrat to strong Republican) results in a 6.7 percent increase in operational conservatism, whereas a similar shift on operational ideology (i.e., from strong liberal at 0.33 to conservative leaner at 4.17) results in a 12.7 percent movement in the Republican direction. The first differences from the EIV models are 4.1 and 12.5 percent, respectively. In short, the first difference results suggest that the effect of operational ideology on party ID exceeds that of party ID on operational ideology, at least in the 1990–1992 data.
The 1992–1994–1996 estimates appear in Table 2. In the first OLS model in column 2, party92 has a significant effect on government activism94 (p < .01). Likewise, as indicated in column 3, operational ideology92 predicts party ID94 (p < .001). Similar to 1990–1992, the effect of lagged operational ideology on party is much greater than the corresponding effect of lagged party on ideology (.19 > .05). The EIV results revise the story in a way that damages the partisan influence hypothesis. In the operational ideology94 model, party92 fails to exert a significant effect on operational ideology (
Cross-Lagged Models for the 1992–1994–1996 ANES Panel, OLS and EIV Estimates.
Source. 1992–1994–1996 ANES panel.
The standard errors are in parentheses and have been adjusted to account for the complex sample design. Party ID and operational ideology lie on a 0–6 scale, and are coded so that higher scores reflect more Republican and operationally conservative responses, respectively. The cross-lagged coefficients have been highlighted to facilitate comparisons across equations in a given pair of panel models. Control variables have been omitted for clarity (full model estimates appear in Online Appendix B2). The first difference score reports the percentage difference in the predicted score for a given dependent variable between respondents scoring at the 95th and 5th percentile on shaded independent variable. For instance, in the first OLS equation, respondents at the 95th percentile on lagged party ID (6 = strong Republicans) are predicted to score 4.5% more conservative on operational ideology94 than respondents at the 5th percentile (i.e., 0 = strong Democrats). The 95th percentile for operational ideology in 1992 is a score of 4.75 (conservative) and the 5th percentile is a score of 0.60 (liberal). The 95th percentile for party identification in 1994 is a score of 6 (strong Republican), and the 5th percentile is a score of 0 (strong Democrat). The 95th percentile for operational ideology in 1994 is a score of 5 (conservative), and the 5th percentile is a score of 0.75 (liberal). ANES = American National Election Study; OLS = ordinary least squares; EIV = errors in variables.
p < .05 (one-tailed).
Substantively, the effects for the 1992–1994 models mimic the 1990–1992 effects. Again, we can compare the predicted difference in the dependent variable for respondents at the 95th and 5th percentile on the key lagged independent variable. For the OLS models, we see that moving from strong Democrat (5th percentile score = 0) to strong Republican (95th percentile score = 6) on party ID leads to 4.5 percent movement in the conservative direction on operational ideology over time. Conversely, operational conservatives (95th percentile score = 4.75) score 12.9 percent more Republican on party ID than operational liberals (5th percentile score = 0.60). For the 1992–1994 EIV models, the corresponding first differences equal −1.1 and 13.1 percent.
Turning to the 1994–1996 OLS estimates, we find that party and operational ideology simultaneously influence one another (p < .001 and p < .05, respectively). Here, the effect of ideology on party lags slightly that of party on ideology (.07 < .09). The first difference estimates show that strong Republicans (i.e., 95th percentile on party ID94) score 9.1 percent higher on operational conservatism96 than strong Democrats (i.e., 5th percentile on party94). Conversely, movement from liberal to conservative on operational ideology94 (i.e., 5th percentile score of 0.75 to the 95th percentile score of 5) produces a 5.1 percent increase in the strength of GOP partisanship96. In contrast to the 1994–1996 OLS estimates, neither cross-lagged variable’s effect can be reliably distinguished from 0 in the EIV models.
Table 3 presents the results for the 2006–2008–2010 GSS. Once again, we discover that both lagged party and operational ideology exert a reciprocal influence on each other. The OLS results (columns 2, 3, 6, and 7) show significant effects at p < .05, or better, in the expected direction, for all four models. Note further that the coefficient for operational ideology06 on party08 is more than five times as large as the reverse effect of party06 and ideology08 (.21 > .04), whereas the effect of operational ideology08 on party10 is almost twice as large as the effect of party08 on ideology10 (.15 > .08). But again, because a simple comparison of the unstandardized coefficients does not tell the whole story, we turn to the first difference calculations. Movement from the 5th percentile score (strong Democrat at 0) to the 95th percentile score (strong Republican at 6) on party ID shifts opinion on operational ideology by 4 percent in 2006–2008 and 8 percent in 2008–2010. Comparable movement on lagged operational ideology (i.e., from strong liberal at 0 to moderate at 3.43) generates increases of 12.2 and 8.5 percent in GOP affinities across the panel waves.
Cross-Lagged Models for the 2006–2008–2010 GSS Panel, OLS and EIV Estimates.
Source. 2006–2008–2010 GSS panel.
The standard errors are in parentheses and have been adjusted to account for the complex sample design. Party ID and operational ideology lie on a 0–6 scale, and are coded so that higher scores reflect more Republican and operationally conservative responses, respectively. The cross-lagged coefficients have been highlighted to facilitate comparisons across equations in a given pair of panel models. Control variables have been omitted for clarity (full model estimates appear in Online Appendix B2). The first difference score reports the percentage difference in the predicted score for a given dependent variable between respondents scoring at the 95th and 5th percentile on shaded independent variable. For instance, in the first OLS equation, respondents at the 95th percentile on lagged party ID (6 = strong Republicans) are predicted to score 4.1% more conservative on operational ideology08 than respondents at the 5th percentile (i.e., 0 = strong Democrats). The 95th percentile for operational ideology in 2006 is a score of 3.43 (moderate) and the 5th percentile is a score of 0 (strong liberal). The 95th percentile for party identification in 2008 is a score of 6 (strong Republican), and the 5th percentile is a score of 0 (strong Democrat). The 95th percentile for operational ideology in 2008 is a score of 3.43 (moderate), and the 5th percentile is a score of 0 (strong liberal). GSS = General Social Survey; OLS = ordinary least squares; EIV = errors in variables.
p < .05 (one-tailed).
The EIV results are less clear-cut. In the 2006–2008 models (columns 4 and 5), lagged party identification exerts no systematic influence over contemporary operational ideology (
The final analyses come from the 2008–2010–2012 GSS panel. The estimates appear in Table 4. In the first set of OLS equations, we find that party ID in 2008 shapes beliefs about the federal government in 2010 (
Cross-Lagged Models for the 2008–2010–2012 GSS Panel, OLS and EIV Estimates.
Source. 2008–2010–2012 GSS panel.
The standard errors are in parentheses and have been adjusted to account for the complex sample design. Party ID and operational ideology lie on a 0–6 scale, and are coded so that higher scores reflect more Republican and operationally conservative responses, respectively. The cross-lagged coefficients have been highlighted to facilitate comparisons across equations in a given pair of panel models. Control variables have been omitted for clarity (full model estimates appear in Online Appendix B2). The first difference score reports the percentage difference in the predicted score for a given dependent variable between respondents scoring at the 95th and 5th percentile on shaded independent variable. For instance, in the first OLS equation, respondents at the 95th percentile on lagged party ID (6 = strong Republicans) are predicted to score 4.1% more conservative on operational ideology10 than respondents at the 5th percentile (i.e., 0 = strong Democrats). The 95th percentile for operational ideology in 2008 is a score of 3.5 (moderate), and the 5th percentile is a score of 0 (strong liberal). The 95th percentile for party identification in 2010 is a score of 6 (strong Republican), and the 5th percentile is a score of 0 (strong Democrat). The 95th percentile for operational ideology in 2010 is a score of 4.1 (lean conservative), and the 5th percentile is a score of 0.4 (strong liberal). GSS = General Social Survey; OLS = ordinary least squares; EIV = errors-in-variables.
p < .05 (one-tailed).
Turning to the 2008–2010 EIV results in columns 4 and 5, we uncover support for the partisan influence hypothesis only (
To summarize the estimates from Tables 1 to 4, the statistical and substantive results affirm our hypotheses. First, consistent with the conventional model of the partisan perceptual screen, we found that lagged party ID predicts current operational ideology in all seven OLS models and three of seven EIV models. Second, as called for by our rival model of issue-based party updating, lagged operational ideology shapes current partisan affinities in six of seven OLS models and four of seven EIV models. Said otherwise, the evidence supports both the classic and revisionist models of party ID regardless of how we deal with random measurement error. 8
Intriguingly, the OLS and EIV models show that the operational ideology-to-party effect often exceeds—sometimes by a wide margin—the party-to-ideology estimate. We can see this in two ways. First, the mean regression coefficient for party IDt − 1 on operational ideologyt equals .06 in the OLS models and .02 in the EIV models. The average regression coefficient for ideologyt − 1 on partyt equals .14 in the OLS models and .09 in the EIV models. 9 Second, the mean effect size of party IDt − 1 on operational ideologyt, as captured by the difference between respondents at the 95th and 5th percentile on lagged party, equals 6.3 percent in the OLS models and 2.4 percent in the EIV models. The mean effect size of lagged ideology on current party equals 8.6 percent in the OLS models and 5.9 percent in the EIV models. We do not want to overemphasize these differences. Instead, we simply point out that the conventional wisdom, that party identification resists influence from policy orientations, needs to acknowledge the considerable influence that beliefs about the size and scope of the federal government wield over party ID for most people most of the time.
Do Moderators Matter?
Given that operational ideology systematically drives party change over time, we have suggestive evidence that it functions like a core predisposition in the minds of many or perhaps even most people. However, we cannot be sure this is the case until we examine whether the influence of operational ideology is conditional on political sophistication, which we define as the degree to which someone is cognitively engaged with public affairs (Zaller 1992). The “sophistication interaction” hypothesis and the related “education interaction” hypothesis (see Sniderman, Brody, and Tetlock 1991) propose that the unsophisticated experience has great difficulty linking ideological predispositions to other political orientations, and hence, these perspectives imply that the relationships we observed above might be stronger at higher levels of cognitive ability.
We test this as follows. First, per Bartels (1996) and Ansolabehere, Rodden, and Snyder (2008), we tap political sophistication in the ANES panels using the interviewer’s rating of the respondent’s general level of information about public affairs. For the GSS data, we classify respondents holding a bachelor’s degree as politically sophisticated. Second, we split the ANES sample into low/high information groups and the GSS samples into noncollege/college graduates. Third, we estimated cross-lagged regression models using both OLS and EIV (see Online Appendix E).
If the sophistication interaction hypothesis holds, then the party-to-ideology and ideology-to-party effects should be statistically significant in the high groups, statistically insignificant in the low sophistication groups, and demonstrably larger in magnitude in the high groups. As revealed by the estimates in Online Appendix E, the data fail to confirm these hypotheses. In the OLS equations that predict operational ideology, party ID is significant in six of seven models for the less sophisticated and significant in all seven models for the high groups. 10 Moreover, the effect of party on operational ideology is somewhat larger on average in the high group (.08 > .05), although few of the differences are significant. When party ID serves as the dependent variable in the OLS models, ideology has a significant effect in five of seven equations for the low group versus four equations for the high group. Here, the coefficient is roughly comparable across groups (.14 > .12). In the EIV models, we find that the effect of party on operational ideology is stronger statistically and substantively in the low sophistication group compared with the high group (i.e., more significant effects and a slightly larger average coefficient), whereas the effect of ideology on party is statistically substantively comparable across the groups (note that the difference in the average magnitude of the coefficient is slight: .11 > .08).
Overall, this pattern of results suggests that sophistication may not systematically condition the ideology–party relationship. However, it may be that the coefficients bounce around erratically due to our crude sophistication splits and/or the imprecision resulting from using relatively small samples. This is not to say that our tests are without value. Indeed, if consistent differences emerged across groups, we would have strong evidence that sophistication matters. But as we did not find such evidence, the proposition that individual differences matter remains tentative pending the results of more powerful tests.
Last, the external political environment might affect the ideology–party relationship (Highton and Kam 2011). Given that elite polarization accelerated throughout the period covered by our data, we can see whether the cross-lagged effects increased in magnitude from the Bush I presidency to the Obama presidency. Figure 1 plots the cross-lagged OLS (top panel) and EIV (bottom panel) coefficients from Tables 1 to 4 over time. There is no discernable tendency for either effect to grow stronger over the years. For the ideology-to-party effect, the coefficient fluctuates unpredictably. For the party-to-ideology pathway, the effect holds steady. It may be that the effects of operational ideology on party and vice versa peaked prior to 1990, in which case, ceiling effects may be in play. Although plausible, our data cannot speak to this possibility. However, our data do imply that since 1990, elite polarization has not altered the operational ideology–party relationship.
Plot of OLS and EIV cross-lagged coefficients over time.
Summary and Conclusions
Scholars have long maintained that party ID functions as the fountainhead in mass belief systems. A small band of researchers has argued that policy orientations shape party ties under limited conditions, but even when evidence of partisan endogeneity emerges, party typically overwhelms policy. We have identified a critical exception to this rule. Drawing on work on operational ideology (Ellis and Stimson 2012), we have argued that beliefs about government activism in the economic welfare domain function as durable predispositions that facilitate fast and frugal political judgment in the minds of ordinary people (also see Goren 2013). From this framework, we derived two hypotheses and found support for each one. First, consistent with the canonical view of party, people bring their views of government into closer alignment with their partisan leanings. Second, as called for by our framework, citizens habitually update party ties to reflect their operational views of government. Finally, exploratory analyses suggested that neither individual nor contextual moderators identified in prior literature condition the policy–party links we discovered.
Our work makes several contributions to the study of public opinion. First, this paper advances our understanding of the nature of ideological thinking in the American public. If one takes ideology to mean a well-developed and fully integrated belief system, there is no doubt that Americans are “innocent of ideology.” But as operational ideology centers on the simpler idea of more versus less government in national life, a policy cleavage that has animated American political discourse for decades, the proposition that citizens hold meaningful beliefs about government and use these to guide political judgments seems uncontroversial. The empirics strongly support this view and, in conjunction with related work on the power of operational ideology to inform issue positions and candidate choice (Goren 2013), make a compelling case that operational ideology functions as a core policy predisposition—perhaps the core policy predisposition—in the minds of voters. 11
Second, we have adduced some of the strongest evidence to date that party ID is deeply informed by programmatic policy concerns. Whereas some studies show that the influence of policy on party is confined to those who care about a given issue and/or know where the parties stand (Carsey and Layman 2006; Sniderman and Stiglitz 2012), we have demonstrated that sophisticated and unsophisticated—and educated and less educated—citizens alike ground partisan identities in beliefs about the government to roughly comparable degrees. And although some studies reveal that the influence of policy on party waxes and wanes over time (Dancey and Goren 2010; Highton and Kam 2011), we have shown that operational postures toward government activism powerfully motivate party updating year after year.
Beyond this, our evidence reveals that the relationship between partisanship and operational ideology runs both ways. In contrast to work on partisan sorting (Levendusky 2009), we show that partisanship and operational ideology work in tandem, with voters adjusting both views to home in on their underlying political positions. Although Levendusky presumes a primarily partisan-driven process of sorting, we put forth a nuanced view of partisanship and operational ideology. Put simply, we push the revisionist conceptualization of party ID farther than it has gone before.
We would be remiss if we failed to underscore the limits of what we have done and can claim. The first caution is that although the findings provide a robust affirmation of our revisionist theory of party updating, it must be remembered that the lion’s share of evidence backs the claims advanced by the American Voter conceptualization of party ID. We recognize of course that accumulated research shows that party ID dominates other core principles such as equal opportunity and moral tolerance (Goren 2005; Goren, Federico, and Kittilson 2009), which raises the question about why our results differ. Given the long history and unrivaled salience of debates about the size and scope of government in elite discourse and American political culture, we believe that operational ideology may be unique (or nearly so) in its ability to detach people from their partisan moorings.
Second, we have suggested that the ability to link welfare state views and party ID does not seem to depend on political sophistication or education—at least in the data sets we examined. This finding breaks sharply with a long established line of research attesting to the power of sophistication to enhance reliance on general policy predispositions (Jessee 2012; Sniderman, Brody, and Tetlock 1991; Zaller 1992, 2012; but see Goren 2013). Of course, other individual-level moderators, such as ambivalence or civic motivation, may condition the policy-to-party links (Groenendyk 2013; Lavine, Johnston, and Steenbergen 2012). We look forward to future research on this score.
Third, our study focuses on a single dimension of ideology. There is no doubt that questions about government activism in the economic and social welfare domain are highly salient at the system and individual level. That said, ideology is a multidimensional concept (Feldman and Johnston 2014), and we have not addressed the dynamic relationship between other salient dimensions, such as symbolic ideology (i.e., liberal–conservative identification) and party ID, on one hand, and social ideology (e.g., positions on moral and cultural issues such as abortion, gay rights, gun control, and so on) and party, on the other hand. We look forward to future research that takes up these relationships.
Finally, the data we examined cover a period of intense elite polarization. As such, we cannot say whether the propensity of individuals to ground their party loyalties in their operational beliefs about government held in the pre-polarization era arose in response to it or reinforced elite polarization. 12 In any case, as polarization appears here to stay for the foreseeable future, we see little reason to expect things to change moving forward. To the degree that the national parties continue to divide along these lines—a path they have followed since the 1930s—citizens will update their party ties to ensure consonance with their underlying beliefs about the role of government in American society. In light of the alleged ideological innocence and political incompetence that characterizes many influential accounts of the American voter, our results suggest a less severe and more nuanced appraisal is in order.
Footnotes
Acknowledgements
We thank Bill Jacoby for his advice on an earlier version of this paper. We are responsible for any remaining errors in the manuscript.
Authors’ Note
Dr. Chen has moved to the University of Calgary. All data are publicly available from the American National Election Studies (ANES) and General Social Survey (GSS) websites. ANES data can be found at http://www.electionstudies.org/studypages/download/datacenter_all_NoData.php. GSS data can be found at http://gss.norc.org/Get-The-Data. Replication files are available online at 
.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
1.
Our definition of operational ideology follows directly from Stimson’s work on both macro- and micropolicy mood (Ellis and Stimson 2012; ). As noted by Ellis and Stimson, operational ideology and symbolic ideology are distinct concepts in American politics. Accordingly, the correlation between the two constructs in our data ranges from .31 in the 1990 American National Election Studies (ANES) to .44 in the 1996 ANES, with an average correlation of .36.
2.
The General Social Survey (GSS) ran separate panels from 2006–2008–2010 and from 2008–2010–2012. Hence, we have two separate panels that cover the period 2008–2010. Although the panels overlap in this period, the participants represent separate samples and are not linked.
3.
We include respondents who answered more than half the available operational ideology questions in a given panel. For example, eight operational ideology questions appear on the 2006–2008–2010 GSS, and we calculate operational ideology scores for everyone who answered five or more questions. Missing data rates are minimal across all panels. The percentage of individuals excluded due to item nonresponse ranges from 0.64 percent in the 1996 ANES to 4.93 percent in the 2008 GSS. Moreover, the rates vary little across levels of sophistication (cf. ).
4.
As a robustness check, we applied the structural equation modeling (SEM) techniques used by Goren (2005) and Carsey and Layman (2006) to model the dynamic relationship between “latent” party ID and “latent” operational ideology in the 1992–1994–1996, 2006–2008–2010, and 2008–2010–2012 panels. We could not use the 1990–1992 ANES because we need three waves of data to identify the single indicator measurement model for party ID. As indicated in Online Appendix C () and Note 8, the same pattern of results emerges across the ordinary least squares (OLS), errors in variables (EIV), and SEM models.
5.
We control for sex (1 = female, 0 = male), married (1 = married, 0 = other), black (1 = black, 0 = other), college graduate (1 = graduate, 0 = other), and Southern resident (1 = Southerner, 0 = non-Southerner). We expect sex and black to be negatively related to Grand Old Party (GOP) identification and operational conservatism, whereas married, college graduate, and Southern resident should be positively related to GOP ID and operational conservatism. To preserve space, the tables report the effects on lagged party and ideology on their contemporary values. We relegate the full set of estimates to Online Appendix B.
6.
It is possible that operational ideology, rather than functioning as a core predisposition, proxies for group affect or core political values. To account for this, we tested whether our results hold when we control for feelings toward the poor, feelings toward African Americans, and the value of equal opportunity. Our results do not change with the addition of these controls. A description of the results and the regression estimates appear in Online Appendix F.
7.
We note that ideologically driven movement in party ID may more readily reflect changes in the strength of identification rather than changes in the direction of partisanship. As an additional check, we reestimated our models using the 3-point self-categorization scale for party identification as the dependent variable and calculated predicted probabilities of partisan categorization from these models. The estimates reported in Online Appendix G show that movement in lagged operational ideology produces statistically and substantively significant effects on partisan self-categorization across all the panels.
8.
We remind readers that our inferences do not change if we use the SEM approach adopted in prior research by Goren (2005) and . As indicated in Online Appendix C, lagged party predicts operational ideology in three of seven models (p < .05), whereas lagged ideology predicts current party in four of seven models.
9.
The average magnitude of the SEM coefficient for the ideology-to-party effect is larger than the mean party-to-ideology effect (.14 > .03; see Online Appendix C).
10.
Given the small sizes of several high sophistication samples, we use p < .10 (one-tailed test) for rejecting the null. This ensures that the tests are not stacked against the sophistication interaction hypothesis.
11.
We remind readers that average effect size of operational ideologyt − 1 on party IDt, as captured by the difference between respondents at the 95th and 5th percentile on ideology, equals 8.6 percent in the OLS and 5.9 percent in the EIV models. Although these may strike some as modest effects, several points should be kept in mind when evaluating their magnitude. First, the ideology variable is measured two years prior to party ID. Second, we control for lagged partisanship, which makes ours a tough test of the ideological influence hypothesis. Third, party ID is a very stable political predisposition. Fourth, we show in Online Appendix G that lagged operational ideology produces meaningful changes in the 3-point party self-categorization variable. Fifth, the effects hold for everyone. In light of these considerations, we believe a fair conclusion is that operational ideology induces substantively meaningful shifts in party ID.
12.
Unfortunately, we do not have comparable measures that enable us to go back to the 1956–1958–1960 or 1972–1974–1976 ANES panels to explore this possibility. The differences between the measures we have and the measures that are available on the earlier ANES panels are sometimes stark (i.e., there are differences in branching and labeling, treatment of “don’t know” and “no opinions,” and differences in item wording), which makes us reluctant to undertake these kinds of temporal comparisons in the absence on nearly identical measures. , however, do show reciprocal associations between party identification and social welfare attitudes in both the 1956–1958–1960 and 1972–1974–1976 ANES panels. Although these attitudes are not substitutable for our measure of operational ideology, they do imply that these types of reciprocal relationships can exist outside of our current, intensely polarized political environment.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
