Why do members of Congress sometimes cast roll-call votes that are incongruent with the opinions of their constituents? A strong electoral connection between lawmakers and those they represent forms the foundation of democratic representation and creates a powerful incentive for lawmakers to follow district opinion. When lawmakers behave in ways that are out of step with their district, citizens can, and do, vote them out of office (Canes-Wrone, Brady, and Cogan Reference Canes-Wrone, Brady and Cogan2002; Mayhew Reference Mayhew1974). Despite this electoral incentive, members of Congress frequently cast votes contrary to the preferences of their constituents, even on the most high-profile issues. More puzzling is the fact that Republicans tend to vote against district opinion more frequently than Democrats even though the logic of the electoral connection should apply equally to both parties (Clinton Reference Clinton2006). Nevertheless, Republican lawmakers are not systematically voted out of office at higher rates than Democrats.
What explains this? Drawing on research from political psychology, we argue that past scholarship on roll-call voting and representation has incorrectly conceptualized constituency preferences by dismissing symbolic attachment to ideological labels as a source of real attitudes used to evaluate legislative position taking. Indeed, political psychology research has shown that operational policy preferences and symbolic ideology are distinct dimensions of ideology that are not necessarily strongly correlated with one another (Converse Reference Converse1964; Ellis and Stimson Reference Ellis and Stimson2012). Mason (Reference Mason2018a) calls these two dimensions “issue-based ideology” and “identity-based ideology.” We argue that both dimensions of ideology—issue/policy and identity—influence public evaluations of elected representatives’ roll-call votes. Moreover, because people often hold policy preferences that do not match their symbolic, identity-based attitudes, a lawmaker’s roll-call decision can create internal conflict in how constituents evaluate their performance. A single roll-call vote can either satisfy both dimensions of a constituent’s preferences, neither dimension, or only one.
When these two preferences diverge, we argue that rank-and-file Democrats reliably prioritize policy preferences over symbolic attachments, but rank-and-file Republicans tend to reconcile the conflict in favor of their symbolic attachments to their ideological identity. These differences in the behavior of Democrats and Republicans, we argue, are a function of the “source cues” that citizens receive from political elites—cues that help structure the opinion of rank-and-file partisans. Due to the ideological composition of each party’s coalition, the elite cues differ systematically between Democrats and Republicans. On the one hand, Republican identifiers overwhelmingly also identify themselves as conservatives—i.e., they embrace a conservative identity—but they hold diverse operational policy preferences. On the other hand, Democrats are a diverse coalition of ideological identities, but all largely endorse liberal policy positions.
As a result, Republican and Democratic lawmakers face different incentives when it comes to the way they frame political issues and engage in roll-call behavior. Republican lawmakers are more likely than Democrats to cast roll-call votes that are incongruent with district opinion on high-profile policy issues because their constituents often value symbolic loyalty to “conservatism” more than they care about the content of the public policy being advanced, while the opposite is true for Democratic lawmakers. As a result, lawmakers of both parties are following their electoral incentives, but they serve districts that demand different patterns of representation. Our argument ultimately stands in contrast to recent work depicting Republicans as more motivated by ideology and Democrats by identity (see Grossmann and Hopkins Reference Grossmann and Hopkins2016).
To test this theory of roll-call representation, we rely on two types of evidence. First, we use individual data from the Cooperative Congressional Election Studies (CCES) from 2008 to 2014 to show that partisans from each party react differently to roll-call behavior when evaluating their representative’s performance. We show that Democrats are more likely than Republicans to approve of representatives who cast votes in line with their specific policy preferences. Conversely, Republicans are more likely than Democrats to approve of representatives who vote in line with their ideological identity, even if they sometimes vote against their preferred policy outcome. In other words, Republican identifiers reward support for in-group loyalty to the conservative team but Democratic identifiers reward support for their individual policy positions.
Second, we use the Cooperative Congressional Election Study (CCES) data from 2008-2014 to create district-level aggregate measures of symbolic and operational opinion to estimate models that predict roll-call voting. Consistent with our theory, we also find that both constituents’ policy opinions and symbolic attitudes predict roll-call behavior. Among districts represented by Republicans, though, operational opinion has a weaker association with roll-call decisions than it does among districts represented by Democrats, while the reverse is true for symbolic, identity-based attitudes. The district’s symbolic attitudes predict roll-call voting for Republicans in Congress more than they do for Democrats. We argue that these findings have important implications for how we conceptualize representational congruence, our understanding of the internal dynamics of the parties in American legislatures, and the way scholars can integrate recent advances in political psychology and policy representation.
The Policy Congruence Tradition of Representation
An established line of research going back to Miller and Stokes (Reference Miller and Stokes1963) has explored the relationship between constituency opinion and roll-call voting. From the beginning, this work has produced divergent findings, with some authors finding relatively high levels of responsiveness (Erikson Reference Erikson1978) and some finding very little (Bafumi and Herron Reference Bafumi and Herron2010; Jacobs and Shapiro Reference Jacobs and Shapiro2000). However, most work has found that the relationship is contingent on a host of different things, including lawmaker, district, issue, and contextual factors (Achen Reference Achen1978; Bailey and Brady Reference Bailey and Brady1998; Butler and Nickerson Reference Butler and Nickerson2011; Gerber and Lewis Reference Gerber and Lewis2004; Griffin Reference Griffin2006; Hurley and Hill Reference Hurley and Hill2003; Jackson and King Reference Jackson and King1989; Miller and Stokes Reference Miller and Stokes1963). These findings have led researchers to conclude that lawmakers sometimes engage in instructed delegate representation, but sometimes they follow different models (Hill, Jordan, and Hurley Reference Hill, Jordan and Hurley2015).
Interpreting the diversity of findings is complicated by the fact that scholars in this area use different measures of constituent opinion and disagree over how to estimate its relationship to roll-call voting. Measures of statistical covariation between constituency opinion and roll-call patterns are the most common measure of representation, even as some have questioned the validity of this approach on both statistical and theoretical grounds (Achen Reference Achen1977; Hill, Jordan, and Hurley Reference Hill, Jordan and Hurley2015; Matsusaka Reference Matsusaka2001). There is also disagreement over the proper way to conceptualize district opinion. Some conceive of it as the mean or median position on a one-dimensional ideological scale that summarizes many issue positions (Bafumi and Herron Reference Bafumi and Herron2010; Gerber and Lewis Reference Gerber and Lewis2004; Tausanovitch and Warshaw Reference Tausanovitch and Warshaw2013), while others argue that measuring opinion on specific issues separately is a more valid characterization of mass attitudes (Ahler and Broockman Reference Ahler and Broockman2018).
However, what prior studies have in common is that they define constituent opinion as average preferences about policy, with the debate boiling down to whether it is more appropriate to aggregate issue positions or measure them separately. This debate is important for clarifying the nature of policy preferences among groups of voters, but it does not address the fact that voters’ operational policy preferences may not be the only attitudes they use to evaluate their member of Congress’s (MC) roll-call votes. Symbolic ideology—i.e., identity-based ideology—can also influence how people want their representative to behave.
Indeed, one of the most consistent findings in public opinion research is that people lack coherent belief systems (Converse Reference Converse1964). Nevertheless, most people still describe themselves using ideological labels (Conover and Feldman Reference Conover and Feldman1981). Scholars conceptualize the adoption of these labels as a form of social identity, whereby people form symbolic attachments to ideological groups without consciously endorsing an entire policy program (Conover and Feldman Reference Conover and Feldman1981; Devine Reference Devine2015; Ellis and Stimson Reference Ellis and Stimson2012; Kinder and Kalmoe Reference Kinder and Kalmoe2017; Malka and Lelkes Reference Malka and Lelkes2010). These ideological attachments are usually inherited from other areas of people’s lives such as their family members, religious affiliations, peers, or elsewhere, and they serve important social and cognitive functions (Devine Reference Devine2015; Jost, Federico, and Napier Reference Jost, Federico and Napier2009). In addition to helping people navigate the complexities of the political world, these attachments help them cope with uncertainty, find meaning in their lives, and develop a sense of belonging in their communities.
As a result, ideological labels are not simply an error-laden shorthand for policy preferences; they represent a distinct dimension of political attitudes (Mason Reference Mason2018a). Group identification is a form of social identity tied to self-esteem that leads one to perceive common group interests and to have more favorable attitudes towards members of their perceived in-group (Tajfel Reference Tajfel1978). Ideological labels in American politics constitute a group identity that leads people to favor legislation simply because they learn that members of their group support it (Devine Reference Devine2015; Malka and Lelkes Reference Malka and Lelkes2010), even if they may not personally agree with the policy. As such, ideological group identification can have an independent effect on support for policies and evaluations of representatives.
The implication of this line of research is that ideology is a multi-dimensional construct composed of an operational dimension that reflects citizens’ policy preferences and a symbolic dimension grounded by in-group identity (Carmines and D’Amico Reference Carmines and D’Amico2015; Ellis and Stimson Reference Ellis and Stimson2012; Free and Cantril Reference Free and Cantril1967; Jost, Federico, and Napier Reference Jost, Federico and Napier2009; Popp and Rudolph Reference Popp and Rudolph2011). Even though both dimensions are often politically salient, they are only marginally correlated, which means people routinely hold conflicting attitudes across the two dimensions, producing an “operational-symbolic disconnect” (Ellis and Stimson Reference Ellis and Stimson2012). The disconnect results from the fact that while a majority of the American public embraces a conservative social identity, only about one-third consistently endorse conservative policy positions. By contrast, a large majority of the American public endorses liberal policy positions but only about one-third identifies as liberal (Ellis and Stimson Reference Ellis and Stimson2012).
The most prominent feature of this disconnect, however, is its asymmetry—that is, symbolic conservatives are far more likely to express liberal policy preferences than symbolic liberals are to endorse conservative policies. In fact, according to Ellis and Stimson (Reference Ellis and Stimson2012) whereas there is little conflict among liberals between their symbolic identification and their operational preferences, two-thirds of symbolic conservatives experience conflict with their operational policy preferences on economic issues, cultural issues, or both.Footnote 1
While research on policy representation has long focused on the effects of public opinion about policy, political psychology suggests that other attitudes derived from a person’s ideological identity might affect public reactions to lawmaker behavior as well. These often incompatible policy and symbolic attitudes not only create ambivalence among citizens when they come into conflict, but these cross-pressured constituents can also create incentives for lawmakers to vote against their district’s stated policy preferences. As we will explain in more detail, these incentives are more prevalent among Republican members of the House than Democratic members.
A Theory of Symbolic and Operational Representation
How Voters Evaluate Roll Calls
Partisan identification is the most important concept in American politics (Bafumi and Shapiro Reference Bafumi and Shapiro2009; Bartels Reference Bartels2002; Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Johnston Reference Johnston2006). It represents a deeply rooted socio-psychological attachment to one of the two major parties, and it serves as most people’s primary political identity. It structures the way people make sense of their political reality, and it is the primary psychological construction that tethers citizens to the American political system. In other words, “partisanship pulls together conceptually nearly every aspect of electoral politics” (Achen and Bartels Reference Achen and Bartels2017, 268).
Even though partisan identities are socialized through family and social networks early in life, their ideological content is largely elite driven. Following Converse’s (Reference Converse1964) notion of sociological constraint, Noel (Reference Noel2012, 158) explains that “the content of ideology is the result of some small set of idea organizers who define what it means to be liberal or conservative,” and “voters merely respond to this organization” as ideas trickle down from the elite to the mass levels. Among these “idea organizers” are interest groups, opinion leaders, and activists who form coalitions around policy programs, and they push these programs as part of an extended party organization that selects candidates during nominating contests (Bawn et al. Reference Bawn, Cohen, Karol, Masket, Noel and Zaller2012). Throughout the last half of the twentieth century into the early twenty-first century, the policy content of modern liberalism is largely derived from the active state interventions of the New Deal, the Great Society, and the Civil Rights eras. By contrast, the modern conservative movement developed its policy program in reaction to the growth of the modern welfare state. The three pillars that define modern conservatism are a reverence for small government, a hawkish foreign policy built around a strong military, and traditional cultural values.
Politicians adopt party platforms to mobilize these ideological groups into viable electoral coalitions. The starkest example of this process occurred in the mid-twentieth century, when the political realignment of the South motivated liberal elites to realign themselves into the Democratic Party and conservative elites started to identify as Republicans more consistently (Levendusky Reference Levendusky2009). The rank and file gradually adopted the parties’ preferred positions and symbols, decided to switch parties, or were aged-out and replaced by younger, more sorted cohorts. As a result, today the term “liberal” is virtually synonymous with “Democrat,” while the term “conservative” is indelibly linked to the Republicans.
While rank-and-file Democrats and Republicans should both value representatives who behave in line with both their symbolic attitudes and policy preferences, each party will ultimately favor one over the other for three reasons. First, the asymmetry that exists in the operational-symbolic disconnect in public opinion poses a different problem for leaders of each party. For Republicans, because conservative identifiers have diverse policy opinions, opinion leaders and strategists alike have strong incentives to appeal to conservative identity and to minimize the salience of policy differences. These identity-based ideological appeals, focused on symbolism, in-group loyalty, and out-group threat (Mason Reference Mason2018a), are the key for maintaining a viable electoral coalition that can support a conservative government in power. By contrast, because appeals to liberal identity cannot rally a viable electoral coalition, Democratic strategists and opinion leaders have strong incentives to downplay symbolic appeals to liberalism and emphasize the salience of liberal policy goals, particularly goals that appeal to a diverse coalition (Grossmann and Hopkins Reference Grossmann and Hopkins2016).
Second, ideological identity is more salient for conservatives than liberals. One’s conservative ideological identity more commonly overlaps with their partisan, racial, and religious identities than for those who espouse a liberal identity, giving conservatives a simpler identity structure, or a “mega identity” (Mason Reference Mason2018b), which increases the salience of identity conflict and reduces tolerance (Brewer and Pierce Reference Brewer and Pierce2005). Liberal identifiers are overwhelmingly Democrats, but most Democrats are not liberals and liberal identity is not as closely associated with racial, religious, and other identities. In other words, Republican identifiers are a far more homogeneous group than Democratic identifiers. This research is bolstered by findings that show that conservative identity is stronger than liberal identity, the content of which is “primarily a reaction against liberalism and its associated social groups” (Devine Reference Devine2015, 510; Zschirnt Reference Zschirnt2011). Similarly, the increased emphasis on identity-based ideology also makes conservative identifiers more sensitive to elite social cues that signal what newly politicized policies are deemed consistent with the prescribed meaning of conservatism (Barber and Pope Reference Barber and Pope2019).
Third, other differences between liberal and conservative identifiers should lead Republicans to favor identity over policy. For example, according to moral foundations theory—which identifies patterns that recent research suggests might be socialized rather than innate (Smith et al. Reference Smith, Alford, Hbbing, Martin and Hatemi2017)—conservative identifiers are more likely than liberals to value binding moral foundations like loyalty, conformity, and purity in politics, as well as other aspects of life (Graham, Haidt, and Nosek Reference Graham, Haidt and Nosek2009). They also tend to be more sensitive to the violation of group norms and values (Haidt Reference Haidt2012). These tendencies should lead many conservatives to favor lawmakers who side with their tribe and retain ideologically pure voting records, even when they disagree with the specific policy. Conversely, liberals, for whom loyalty and purity are less salient, tend to embrace individualizing moral foundations and should place less value on ideological conformity for its own sake from their representatives. Instead, these liberal identifiers, who overwhelmingly identify at Democrats, ought to prefer congruence with their opinions on specific issues. Perhaps another way to phrase it is that conservatives are more willing than liberals to put the group’s wishes ahead of their own.
As a result of these three factors, rank-and-file Democrats are more likely to both receive and respond to operational policy messages from their party, while Republicans are more likely to both receive and respond to symbolic appeals to conservative identity. Moreover, as party messages are disseminated downward and absorbed by rank-and-file partisans, the positive feelings that partisans have toward their party ultimately extends to the messages as well, defining what liberalism and conservatism mean in practice (Lenz Reference Lenz2013; Levendusky Reference Levendusky2009; Malka and Lelkes Reference Malka and Lelkes2010; Mondak Reference Mondak1993). As a result, Democratic voters should care more about policy appeals and have greater attachment to their party’s policy positions than Republican voters, who will be more attached to conservative symbols and identity. These theoretical expectations are supported by past research showing that in fact Democratic lawmakers make more policy based appeals to voters, while Republican lawmakers tend to make symbolic appeals (Vavreck Reference Vavreck2001).
We are left with an understanding of constituent ideological preferences consisting of two equally real—and often incongruent—attitudes: operational preferences and symbolic attachments. Voters should genuinely value both policy congruence and fidelity to the ideological in-group from their representatives. However, because the rank and file of each party have been socialized to value different behaviors from their representatives, they should place more weight on different dimensions of ideology when they evaluate their representatives’ roll-call votes.
This logic leads to three hypotheses:
H1: Symbolic ideological and policy congruence are both associated with higher individual evaluations of lawmaker performance holding the other constant.
H2: Rank-and-file Republican will reward (punish) their representatives more when they cast votes that are congruent (incongruent) with their ideological identity than rank-and-file Democrats.
H3: Rank-and-file Democrats will reward (punish) their representatives more when they cast votes that are congruent (incongruent) with their policy opinion than rank-and-file Republicans.
Roll-Call Representation
Those social forces shaping the way citizens evaluate roll-call votes also structure the incentives facing electorally minded members of Congress. When MCs cast a roll call, they want to avoid backlash and minimize the electoral penalty for their position. Roll-call decisions become complicated when key supporters disagree. The symbolic-operational disconnect embedded in public opinion (Ellis and Stimson Reference Ellis and Stimson2012) implies that legislators can be cross-pressured by their districts when the majority simultaneously holds attitudes that could lead it to either approve or disapprove of a vote for or against a policy. When, as often happens, voters are caught between their issue preferences and their ideological identity, lawmakers are often uncertain about which dimension voters will use to evaluate their roll calls. This uncertainty means that the district’s predominant symbolic ideology and operational opinion both affect the representative’s roll-call decisions. Previous research has tended to treat one as a proxy for the other, but we argue that they are conceptually different, and lawmakers respond to both. This understanding yields a number of testable expectations. First, it suggests that lawmakers should be motivated by both their district’s operational and symbolic preferences.
H4: District symbolic and operational conservatism each have a positive effect on the likelihood that a member of Congress casts a conservative roll call holding the other factor constant.
Second, even though both dimensions have an independent effect on lawmaker behavior, congressional parties can help minimize the electoral risk to their party members and help their potential challengers win in the next election cycle by cultivating an ideological reputation that will lead to a favorable party brand capable of garnering electoral majorities (Cox and McCubbins Reference Cox and McCubbins2005; Lee Reference Lee2016). These brands are achieved through coordinated messaging strategies in which party leaders are incentivized to bolster the party’s ideological reputation using different dimensions of ideology.
On the one hand, Republican members of Congress understand that they can win a majority by appealing to symbolic attachment to conservative identity, but not necessarily a shared policy program, and the way to maintain its support is to demonstrate loyalty to conservative symbols. They also understand that conservative identity is far more popular than liberal identity. When casting roll-call votes on policy questions lawmakers are taking operational positions, but they have control over how they present those decisions to constituents. Republican lawmakers want to be able to credibly claim that they have “conservative voting records,” that they “stand with” conservative activist groups, and they want opinion leaders in their partisan coalition to praise them. To accomplish this, they take operationally conservative positions that are often unpopular, but rely on communications strategy to represent these votes as rooted in conservative symbols and opposition to liberal out-groups.
On the other hand, Democrats represent districts that are symbolically diverse, including many self-identified liberals, moderates, and even conservatives. As such, Democratic lawmakers, who campaign more on concrete policy positions, must be far more attentive to the policy opinions of the majority in their district. More than Republicans, Democrats should be aware of which issues are, and are not, part of their electoral appeal and will vote accordingly, sometimes even taking the conservative position on a roll-call vote when they perceive it to be more popular in their district. Because Democrats campaign on policy, by emphasizing policy benefits to key constituencies instead of making broad symbolic appeals, Democratic members of Congress are less constrained by the need to be consistently in step with the liberal position.
H5: Republican-held districts should be symbolically conservative but operationally diverse while Democratic districts should be symbolically diverse but operationally liberal.
H6: The effect of a district’s operational preferences on roll-call voting will be stronger among Democratic lawmakers than among Republican lawmakers.
H7: The effect of a district’s symbolic ideology on roll-call voting will be stronger among Republican lawmakers than among Democratic lawmakers.
Data and Methods
To test our expectations, we conduct two studies. The first is an analysis of representative approval using public opinion and roll-call data from the 2008–2014 Cooperative Congressional Election Study (CCES). This allows us to see if each type of congruence has a different effect depending on a person’s partisanship. The second study uses measures of district-level operational and symbolic ideology, aggregated from the CCES, and high-profile roll-call votes about which the CCES directly measures public opinion to see if representatives of each party follow different dimensions of ideology.
The CCES is a two-wave internet panel study run by Stephen Ansolabehere and Brian Schaffner, and it was founded with the intent to specifically study district-level public opinion, public perceptions of members of Congress, and congressional representation. The pre-election wave takes place at the beginning of October, while the post-election wave occurs shortly after the November elections. Because it is an online opt-in survey, the CCES employs a sample matching methodology for creating a representative sample from non-random pools of respondents. The matching procedure attempts to approximate the random-digit dialing used by Pew Research Center. Unlike most survey studies, which have modest sample sizes (1,000-2,000 observations), the CCES trades on its unusually large sample sizes, which allows for relatively large samples from each congressional district. Its other major advantage is that it asks its large sample to weigh in on several individual roll -all questions, making it possible to compare individual-/district-level opinion with the votes of representatives.
Study 1: Public Evaluations of Lawmaker Roll-Call Behavior
For the first part of our analysis, we investigate whether members of Congress are rewarded for the asymmetrical roll-call strategy that we expect. Our theory of operational and symbolic representation is driven by the idea that the emphasis Democratic lawmakers place on operational opinion and Republicans on symbolic ideology is driven by the preferences of Democratic and Republican voters respectively. That is to say, they are responding to the demands of the constituencies they represent via the electoral connection. To see if this is in fact true, we want to see how Republicans and Democrats assess the performance of their individual representatives in the House when those members cast voters that are incongruent with both their operational policy opinion and their symbolic ideology.
Data and Measures
Representative approval.Footnote 2
To investigate the differential effects of operational and symbolic opinion on roll-call behavior we use the CCES from four years: 2008, 2010, 2012, and 2014. The dependent variable in our analysis is representative approval. The CCES routinely asks respondents whether they approve or disapprove of their representative’s performance in office. The question produces a four-point measure where respondents can either strongly or somewhat approve/disapprove. The measure ranges from 0-3, with 0=strongly disapprove and 3=strongly approve. The average response is μ=1.57 and the standard deviation is σ=0.993, which translates into the average respondent somewhere between somewhat disapproving and somewhat approving of their representative.
Operational and symbolic congruence
We created two measures of ideological congruence using the roll-call votes for which the CCES asks respondents their opinion. These roll-call votes are among the most consequential and high-profile each session. The first measure of congruence captures the proportion of votes each member of Congress cast that are the same as the roll-call preferences of the respondent. For example, if the respondent favored the Affordable Care Act but their MC voted against it, then that vote was incongruent with that respondent’s policy preferences. As a result, it is coded as 0. However, if the respondent opposes the Affordable Care Act, then it would be coded as 1. We repeat this process for all the roll-call votes in the dataset and average them to create a proportion of policy congruence between each respondent and their member of Congress (μ=0.549 σ=0.301). Zero signifies complete incongruence and 1 denotes complete congruence.
The second measure is symbolic congruence, which captures the proportion of the votes a respondent’s representative takes that are congruent with the respondent’s ideological identity. These are votes that the respondent may not necessarily personally agree with as a matter of policy, but they are votes in the policy direction associated with the group to which they identify. We classify votes as conservative if they oppose or reduce federal intervention in the economy, increase federal activity on national security, promote traditional morality, or support free trade, with the opposite positions being liberal. We then classify conservative roll calls as symbolically congruent with respondents who identify as conservative and liberal roll calls as symbolically congruent with those who identify as liberal. For example, take a respondent who identifies as a conservative but supports the Affordable Care Act. If their member of Congress votes against the ACA, then that vote is coded as being symbolically congruent with the conservative respondent, even though it was incongruent with their stated policy preferences (due to the respondent’s support for the ACA). As with policy congruence, the measure ranges from 0–1 (μ=0.506 σ=0.354). As expected, the correlation between symbolic and operational congruence is low, just 0.237.
Controls
Since our theory is conditional on partisanship, we are also including a standard seven-point partisan identification measure, ranging from strong Democrat (0) to strong Republican (6). In addition to our key independent variables of interest, we also controlled for a standard set of demographic controls, including age, gender, education, and race. We controlled for these demographics because they are often associated with attitudes towards politicians, symbolic attachments, and issue positions. As such, we wanted to account for the most well-established demographic confounders. A control for household income was also included because past research suggests that wealthier individuals are more likely to prefer and to receive operational policy representation (Harden Reference Harden2015).
Results
The CCES represents a nested data structure with respondents clustered within congressional districts. As such, we estimated a linear multilevel random-intercept model. Coefficients in linear multi-level models are interpreted just like OLS coefficients, but they also let us estimate a separate intercept for each congressional district. This approach allows us to model the clustered nature of the data—namely, spatial autocorrelation. Because we are interested in how Democrats and Republicans value policy and symbolic congruence differently, we introduced interaction terms between our congruence measures and the respondent’s partisan identification. We also include dummies for each congressional term to deal with any temporal variation in representative approval across survey years.Footnote 3
Table 1 presents our results predicting the approval of individual members of Congress. Interestingly, age, being nonwhite, and being female are all positively related to representative approval. Women are more likely to approve of their congressperson than men, the old register higher levels of approval than the young, and people of color have higher evaluations than white Americans. Levels of political interest are also positively associated with representative approval. However, neither education nor income had a statistically significant relationship with approval. This suggests that once one controls for political interest, socio-economic status has no effect on approval, which is interesting given that high SES Americans are better represented than everyone else (Bartels Reference Bartels2008; Miler Reference Miler2010).
Table 1 Predictors of Representative Approval
Note: Linear Multilevel model. **denotes p<.05. Std errors are in parentheses.
Moving to the direct tests of our hypotheses, both policy and symbolic congruence between the respondent and their congressperson predict higher representative approval, but those effects are conditional on the respondent’s partisanship, meaning the relative influence of each are different for Democrats and Republicans. The interaction between partisanship and policy congruence (β=-0.075, SE=0.004, p<0.001) shows that it is more strongly associated with approval among Democrats than Republicans. Figure 1a shows the conditional slopes for the effect of policy congruence on lawmaker approval across the range of the standard seven-point measure of partisanship. As figure 1a reveals, the differences between Democratic identifiers and Republican identifiers become more extreme as they embrace their partisan identity more strongly. For example, the marginal effect of policy congruence on representative approval for strong Democrats (1.098) is nearly twice the size of the marginal effect for strong Republicans (0.647).
Figure 1 Effect of operational and symbolic representation on approval of representative performance
Turning to the substantive effect of operational congruence on Democratic and Republican approval, when strong Democrats are faced with a representative who votes in ways that are 100% incongruent with their policy preferences, holding all else equal, then their predicted approval is a mere 0.903 on a scale ranging from 0 to 3. This means the average lawmaker evaluation among strong Democrats is somewhere between strongly disapprove and somewhat disapprove. However, when that same representative’s votes are 100% congruent, the predicted level of approval for strong Democrats is 2, or somewhat approve. That is more than a one standard deviation increase in representative approval. On the other hand, Republicans put far less weight on their Congressperson voting in line with their policy opinions. Strong Republicans evaluate their representative’s performance more highly (1.317) when their representative’s votes are 100% incongruent with their own policy opinions compared to strong Democrats in the same situation. Moreover, even though strong Republicans evaluate their representatives about the same as strong Democrats when their voting record matches their policy preferences 100% of the time (1.96), the min-to-max change is only an increase of about two-thirds of a standard deviation in representative approval.
In other words, while it is true that both Democrats and Republicans reward their representatives when their votes are congruent with their policy preferences, Democrats reward them a lot more and are far less approving of representatives whose votes are not congruent with their own policy preferences. This finding is consistent with our argument about differences in policy and symbolic representation between the parties and shows that Democratic lawmakers have a greater incentive to follow their district’s policy opinion. Simply put, Democratic representatives are punished more by Democratic constituents for not following their policy preferences than Republican representatives are punished by Republican constituents for doing the same thing. Even though survey data makes causal identification difficult, this finding is consistent with past experimental evidence showing that Democrats are more responsive to persuasive appeals that include policy-based ideological cues than Republicans, which was noted as an anomaly by Bullock (Reference Bullock2011).
These partisan differences are even more stark when looking at the effect of symbolic congruence. Indeed, the conditional effect of symbolic congruence—that is, when lawmakers vote in ways that are congruent with a respondent’s ideological identity—is even stronger than policy congruence (β=0.147, SE=0.004, p<0.001). As our theory predicts, the effect of symbolic congruence is far and away stronger for Republicans than Democrats. The essential pattern we observed with operational congruence is reflected here, except in the reverse. Figure 1b presents this result visually. While the marginal effect of symbolic congruence for strong Republicans is 1.496, the marginal effect for strong Democrats is less than half that size (0.613).
To put this into clearer perspective, the predicted min-to-max change from complete symbolic incongruence to complete congruence for strong Democrats is only about a two-thirds standard deviation increase in representative approval, much smaller than the more than one and one-half standard deviation difference in approval between strong Republicans whose representative is completely incongruent symbolically as opposed to completely congruent.
We find strong support for the idea that Republican and Democratic lawmakers face very different incentive structures for how to represent the preferences of their constituents. While Democrats have a stronger incentive to follow their constituents’ policy preferences, Republican representatives are incentivized to follow voters’ symbolic preferences, even if it means betraying their district’s policy preferences to do so. After all, if Republican representatives were completely faithful a person’s policy preferences, they could expect a performance evaluation of about 1.9 out of 3, all else being held equal. However, if they were completely faithful to that same garden-variety constituent’s symbolic preferences, even if they occasionally voted against that person’s policy preferences, they would be rewarded with an evaluation of roughly 2.3 out of 3.
Even though this evidence is observational, it is consistent with previous experimental evidence in political psychology showing that Republicans are more likely than Democrats to accept policies when their partisan or ideological in-group supports them (Bullock Reference Bullock2011; Barber and Pope Reference Barber and Pope2019). It is also consistent with similar experimental evidence showing that Republicans are more likely than Democrats to punish their representatives when they take positions that are at odds with the rest of the party (Arceneaux Reference Arceneaux2008). The differing effects across the parties were either viewed as anomalous or idiosyncratic by these scholars, but they fit our argument about asymmetries in partisan behavior.
Study 2: Explaining Roll-Call Voting
Data and Measures
To investigate the differential effects of operational and symbolic opinion on roll-call behavior we again use the Cooperative Congressional Election Study (CCES) common content and roll-call data from 2008–2014. The key independent variables are average policy conservatism on the issue and average symbolic conservatism at the district level.Footnote 4 Policy conservatism is the proportion of constituents supporting the conservative position on the roll call. Symbolic conservatism is average self-placement on a five-point ideological scale, the standard measure of ideological identity (Conover and Feldman Reference Conover and Feldman1981; Devine Reference Devine2015), at the district level rescaled so that the minimum value is zero and the maximum is one. Higher values indicate more conservative identifiers in the district. Roll-call votes are coded such that the conservative position has a value of one and the liberal position a value of zero. Refer to online appendix A for a list of roll-call votes and survey measures used in this analysis, distributions of sample sizes by congressional district, and a number of robustness checks.
These two measures allow us to capture the unique effect of each dimension of a district’s attitudes on roll-call decisions. This measurement strategy means that each lawmaker appears in the data set multiple times in each congressional term because the unit of analysis is the roll-call decision with a different policy opinion measure for each vote. To ensure that our results are not an artifact of inflated statistical power we use clustered standard errors (by district) in the pooled models and estimate separate models for each congressional term. We also estimate mixed effects logistic regression with random slopes and intercepts for each congressional district and dummy variables for each congressional term. Relevant coefficients from both of these models are presented in the online appendix.
Results
We begin our analysis by looking at some descriptive statistics. We expect that lower rates of congruence among Republicans stem from the fact that the two parties represent different kinds of districts. Figure 2 demonstrates this by presenting kernel density plots of district policy and symbolic conservatism in the roll-call data. The top panel shows that while both parties represent districts that span the full range of policy opinion, the Democratic districts are concentrated slightly to the left of center and Republican districts appear evenly split between liberal and conservative opinion on the roll calls in the study, consistent with H5.
Figure 2 Distribution of district policy opinion and symbolic identity by lawmaker party
Contrast this distribution with the second panel showing that while Democratic districts span the full range of symbolic ideology with most of them near the middle of the scale, almost all Republican-held districts are symbolically conservative. This clustering to the right means that casting a conservative roll call nearly always puts Republicans in step with the symbolic attitudes of their districts, but only puts them in congruence with its policy opinion half of the time. For Democrats, by contrast, a reliably liberal vote may or may not match the symbolic attitudes of most of their constituents, so they need to be more sensitive to district policy opinion on salient issues.
Figure 2 also lends credibility to the premise that Republicans are more likely than Democrats to represent districts where prevailing policy opinion is not congruent with symbolic and partisan attachments. In other words, it suggests that the symbolic-operational disconnect that researchers have found at the individual level is reflected at the aggregate congressional district level as well. This disconnect suggests that Democratic and Republican lawmakers face very different incentives given the composition of their respective electorates across each ideological dimension—i.e., incentives that encourage Democrats to be more congruent with their districts’ prevailing policy preferences in order to maintain their electoral coalition and Republicans to be more symbolically congruent.
Figure 3 shows the proportions of each party’s roll calls in our data that are congruent with the prevailing policy opinion and symbolic ideology in their districts. The parties are nearly mirror images of one another, with Democrats voting consistent with policy opinion 71% of the time and with symbolic ideology 56% of the time. Republicans, on the other hand, only vote congruently with policy opinion 57% of the time but are in step with their district’s predominant symbolic ideological identity 75% of the time. This finding shows some tentative support for H6 and H7.
Figure 3 Operational and symbolic congruence on roll call voting by party
However, the descriptive analysis does not tell us whether symbolic and policy conservatism are associated with roll-call decisions holding the other constant, nor whether the parties differ in their responses to these factors. We now turn to modeling roll-call decisions to estimate the relationship between each kind of opinion on legislative voting. We model conservative roll-call voting using logistic regression. To test the first part of our argument—that policy and symbolic opinion have independent effects on roll-call decisions—table 2 presents three models, one including district policy opinion and symbolic ideology and controlling for the lawmaker’s party, and separate models for each party including our two key independent variables. The models include a control for the percentage of the district that graduated from college, because liberal Democrats tend to be more educated than conservative Republicans and education can be associated with greater operational liberalism as well as a preference for policy representation.
Table 2 Logit models of conservative roll calls
Notes: ** denotes p < 0.05
Values are logit coefficients with standard errors clustered by district in parentheses.
The models show that policy opinion and symbolic attitudes are both associated with roll-call positions even when each factor is held constant, and that the effect holds when controlling for the lawmaker’s party and when estimated separately for each party. Model 1 estimates very similar coefficients for symbolic and operational conservatism, both of which are statistically significant and in the expected direction, supporting H4. Operational and symbolic conservatism both predict conservative roll-call voting, even controlling for the other factor and the lawmaker’s party.
Models 2 and 3 directly test our expectations that Democrats are more responsive to issue opinion and Republicans to symbolic attitudes. We find support for these expectations as well. A Democrat representing a district that is 25% in support of the conservative position has a 7.8% chance of casting a conservative roll call, but if their district is 75% in favor of the conservative position the odds increase to a 36.6% chance, a 28.8-point increase. A Republican representing a district that is 25% in favor of the conservative position has a 67% chance of casting a conservative roll call, but those chances increase to 84.1% if the district is 75% in favor, only a 17.1-point increase. The p-value of the difference between the coefficients on operational opinion is less than 0.0001.
Models 2 and 3 also show that Republicans are nearly three times as responsive to the district’s symbolic ideology. Democrats representing a district with a symbolic conservatism score of 0.38 out of 1 (the minimum observed for Republican districts and thus allowing a direct comparison) cast conservative roll calls 13% of the time, compared to 24% for those representing districts with a symbolic conservatism score of 0.75, a difference of eleven points. Republicans representing districts with a symbolic conservatism score of 0.38 cast conservative roll calls 50.6% of the time, but when their district score is 0.75, they cast conservative roll calls 82.3% of the time, a difference if 31.7 points. The p-value of the difference in coefficients is 0.0002. These findings support H6 and H7. Predicted probabilities of a congruent voteFootnote 5 for each lawmaker show that Republicans are expected to side with district operational opinion 55.6% of the time while Democrats are expected to vote with district operational opinion 70.2% of the time, very close to the observed rates of congruent voting.
These results support the main assertions about how the two dimensions of ideology relate to roll-call voting in Congress. The same symbolic-operational disconnect that often exists at the individual level also describes congressional districts, leading Democrats to better represent majority policy opinion in their districts than Republicans when they both toe the party line. Symbolic and policy ideology are also both related to roll-call behavior, but their effects differ by party, with policy opinion better predicting Democrats’ votes and symbolic ideology better predicting Republicans’, consistent with our theory. Differences in the roll-call behavior of Democratic and Republican lawmakers make complete sense given that those differences are in line with our first analysis showing that rank-and-file Democrats and Republicans reward their lawmakers differently. Instead of sacrificing district representation for party loyalty, as some have claimed, members of Congress from both parties are achieving both at similar rates. They are just representing different dimensions of their constituents’ ideological preferences.
Discussion
Elected members of Congress have strong electoral incentives to follow the ideological preferences of their constituents, but ideology is a multi-dimensional concept that includes both a policy dimension that captures people’s opinions on issues and a symbolic dimension that reflects people’s ideological identity as either a liberal or a conservative. Past research on representation has focused on policy preferences or assumed that ideological identity is an aggregate measure of policy opinion, overlooking the fact that these two dimensions represent different constructs and are weakly correlated. In this paper, we have shown that policy preferences and symbolic ideology both explain public evaluations of roll-call votes, but with partisan differences stemming from the nature of each party’s coalition. Republicans are more likely to reward loyalty to the symbols of conservative identity, while Democrats prefer operational congruence on matters of policy. The result is that lawmakers of each party provide different kinds of representation. While this finding is consistent with past research (Jackson and King Reference Jackson and King1989), the implication here is that Republicans are not necessarily abdicating their responsibility as faithful stewards of their constituent’s preferences; rather, they are providing the kind of representation that their constituency wants.
These findings have at least two other implications. First, they build on an emerging line of research suggesting that the two political parties are not mirror-images of each other, and in fact they are quite different in their goals and organization (Bartels Reference Bartels2018; Ellis and Stimson Reference Ellis and Stimson2012; Grossmann and Hopkins Reference Grossmann and Hopkins2016). For example, in their influential book, Grossmann and Hopkins (Reference Grossmann and Hopkins2016) argue that the fundamental difference between the two parties is that Democrats form their attachments to their party based on group identity, while Republicans report their party allegiance to be a function of ideology. The authors ultimately come to this conclusion by making the same mistake as many others: they conflate symbolic ideology with an endorsement of a specific policy program. This conceptualization leads them to conclude that because Republican identifiers also identify as conservative much more consistently than Democrats identify as liberal, they are ultimately more “ideological.” Conversely, they argue that by virtue of Democrats being symbolically diverse and invoking group sympathies to explain their partisanship, they seek group benefits instead of an ideological vision, in spite of the fact that Democrats are more consistently liberal on policy than Republicans are consistently conservative. In so doing, they tacitly deny the existence of a meaningful ideological dimension defined by operational policy preferences.
However, the findings here suggest a different interpretation of similar evidence. Our conception of ideology suggests that it is rank-and-file Republicans who expect lawmakers to prioritize group identity and loyalty to the symbols of the social group, while Democrats place more emphasis on policy representation—the dimension that most people think of when they think about ideology. In other words, each party pursues an ideological agenda, but Democrats and Republicans think about ideology differently. Nevertheless, while our account of representation differs from the version implied by Grossmann and Hopkins (Reference Grossmann and Hopkins2016), we suspect future research can better reconcile these accounts, especially as one looks at the linkages connecting political elites to the mass public through different messaging strategies.
Second, we contribute to how we ought to think about different models of representation. While a number of researchers have grappled with the many dimensions of this question (Hill, Jordan, and Hurley Reference Hill, Jordan and Hurley2015; Stimson, Erikson, McKuen Reference Stimson, Erikson and MacKuen1995; Miller and Stokes Reference Miller and Stokes1963), this literature rarely addresses partisan differences in representation and never fully accounts for citizens’ symbolic preferences. In one of the few exceptions, Barker and Carmen (Reference Barker and Carman2012) argue that Democrats and Republicans engage in systematically different styles of representation. They claim that Democrats prefer instructed delegate representation and Republicans favor trustee representation. While the evidence here is merely suggestive, we offer a slightly different interpretation. We suggest that the difference is not over delegate versus trustee representation; rather, it is about policy versus symbolic representation.
While the policy representation that we argue Democratic lawmakers engage in is consistent with the instructed delegate model, the implication of our theory of symbolic representation for Republicans is different from an unconstrained trustee model. Trustee representation implies that citizens trust their representatives to exercise independent judgement in office. Even though rank-and-file Republicans might be somewhat forgiving when their representatives deviate from their operational policy preferences, our findings suggest that Republicans voters do still want their representatives to take the symbolically conservative position on an issue and are punished when they fail to do so. After all, if Republicans preferred trustee representation they would not mind when their representatives cast a liberal vote with which they disagree. Our findings suggest that is not the case. Instead, they are more likely to tolerate votes out of step with their policy preferences only when those votes are in-step with their ideological identity. In other words, rather than exercise their independent judgement, Republican identifiers still demand in-group loyalty in their roll-call behavior. Of course, scholars differ as to how loyal to previous ideological commitments a trustee should be.
We hope the findings presented here will stimulate further investigation into the role of symbolic identity in legislative representation and in the way the public uses both identity and policy preferences to evaluate elected officials. Future work should isolate causal relationships and exploit cross-state and panel or time-series data to examine the conditions under which symbolic identities rival policy preferences when people evaluate politicians. It should also further explore the implications of symbolic preferences for other models of policy representation, especially when looking at linkages among elites, activists, and the mass public.
Supplementary Materials
To view supplementary material for this article, please visit https://doi.org/10.1017/S1537592720003746
