Asymmetry in Ideological Accountability - Political Communication Lab

Asymmetry in Ideological Accountability - Political Communication Lab

All in the Eye of the Beholder: Asymmetry in Ideological Accountability Shanto Iyengar∗ Gaurav Sood† March 11, 2017 Abstract Political elites have ...

744KB Sizes 0 Downloads 14 Views

All in the Eye of the Beholder: Asymmetry in Ideological Accountability Shanto Iyengar∗

Gaurav Sood†

March 11, 2017

Abstract Political elites have increasingly moved to the ideological extremes. But the public at large is still mostly moderate. Thus, rank and file partisans today are ideologically distant not only from the leaders of the opposing party, but also from the leaders of their own party. Still, partisans continue to feel warmly toward the party they identify with, and (as we demonstrate) its leaders. What explains this continued warmth? We find that both, distorted perceptions, and asymmetric discounting of known policy differences are implicated. Partisans perceive co-partisan leaders to be more moderate than what their voting records suggest. Furthermore, even when provided with information about policy positions, partisans support more extreme spatially distal co-partisan candidates no less than relatively more moderate more spatially proximate candidates of their party.



Shanto Iyengar is Chandler Chair in Communication and Professor of Political Science at Stanford University. He can be reached at: [email protected] † Gaurav can be reached at [email protected]

It is well-established that American political leaders are ideologically polarized (Aldrich 1995; Stonecash, Brewer and Mariani 2003; McCarty, Poole and Rosenthal 2006). It is also wellcorroborated that polarization harms the body politic; polarization causes “gridlock” in legislatures (Jones 2001; Binder 2005; McCarty, Poole and Rosenthal 2006), impedes the work of judicial and administrative bodies (Binder 2005; McCarty, Poole and Rosenthal 2006), and reduces trust in government (Hetherington 2005). Worryingly, the polarization of American legislatures continues apace. The worry turns to puzzlement when we juxtapose three observations. First, over the last four decades, elected representatives of both major parties have become more ideologically extreme (Aldrich 1995; Stonecash, Brewer and Mariani 2003; McCarty, Poole and Rosenthal 2006). Second, there has been far less centrifugal movement among the rank and file; the median voter remains moderate on many of the issues (DiMaggio, Evans and Bryson 1996; Evans 2003; Fiorina, Abrams and Pope 2005; 2008; Hill and Tausanovitch 2014). Third, supporters of the two parties still express strong affection for their party (Iyengar, Sood and Lelkes 2012), and its leaders (as we show later), while expressing hostility towards the opposing party and its leaders.1 In this paper we offer an explanation for the paradox of unabated partisan support despite increased ideological divergence within parties between followers and leaders. Either due to biases in information provided by trusted sources, e.g. communiqués from co-partisan elites, invention of partisan congenial facts, or biases in processing of information, partisans’ perceptions of parties’ and party leaders’ policy positions are strikingly distorted. For instance, as we show later, despite the increasingly stark differences between elite and mass preferences, the distance between partisans’ ideological self-placements and where they place their own party has remained stable over the past four decades. 1 While it is possible that ideological sorting of rank-and-file partisans (see for e.g. Levendusky 2009) explains some of the rise in partisan animus, and continued warmth toward own party, as Mason (2014) shows, once you control for partisan identification, difference between sorted and unsorted partisans on ‘thermometer bias’, ‘anger at outgroup candidate’, and ‘activism’ is statistically indistinguishable (see Figure 5, Mason 2014).

2

There is more to the explanation than biases in information exposure and wishful thinking. Partisans appear to use the information they learn in a manner inconsistent with the dominant spatial model (more spatially proximal candidates get more support); when provided information about candidates’ policy positions, partisans approve of co-partisan candidates with more extreme (and more distal) positions no less than, and occasionally more than co-partisan candidates with more moderate (and more proximate) positions. We begin by examining the relationship between representatives’ actual ideology (estimated via roll-call votes or campaign contributions) and their popularity among partisans. To assess the relationship, we use a very large collection of state level polls that ask about approval of elected officials, and combine it with estimates of officials’ ideology. We find that the most ideologically extreme representatives draw nearly as much support from co-partisans as the more moderate representatives. On the other hand, notably fewer partisans support more extreme out-party representatives than their more moderate out-party counterparts. Next, we investigate whether asymmetric changes in approval are founded in asymmetries in the spatial calculus used by voters to judge candidates. Jointly scaling respondents’ and congressional officials’ positions using data from the 2006 Cooperative Congressional Election Study (CCES), we find this asymmetry is partly a consequence of the different spatial calculus that partisans use to judge co-partisan and opposing party representatives. We find that independents are most sensitive to spatial distance, and that partisans are far more sensitive to ideological position of elected officials of the opposing party than their own party. To explain this differential discounting of actual ideological distance, we first consider the role of perceptions by examining perceptual bias with data from the 2006 CCES, comparing perceptions of Senators’ votes with their actual voting record. We find that co-partisan Senators are perceived as more moderate than their actual record, with bias greater in perceptions of co-partisan Senators’ voting records than out-party Senators’ voting records. Lastly, we explore how partisans’ judge politicians when information about policy positions is provided to them. 3

Even when given information about candidate policy positions, partisans approve of co-partisan candidates with more extreme positions no less than —and sometimes more than —co-partisan candidates with more moderate positions. All in all, these results paint a sobering picture of voters’ ability to hold elected officials of their own party accountable for their ideological positions.

Partisan Affect, Perceptual Asymmetries, and Motivated Processing That most people attach themselves to political parties for policy-based reasons is an appealing, even obvious, theory. But empirical evidence for the theory has always been scant. One of the earliest challenges to the theory came from Campbell et al. (1960), who discovered that only a few people reasoned about politics in ideological terms, and theorized that partisan identification is primarily affective.2 This more “primal” view of partisanship has since been validated by considerable work on the underpinnings of partisan identity (Green, Palmquist and Schickler 2004; Greenstein 1965; Jennings, Stoker and Bowers 2009). The claim that the bond between voters and parties is primarily affective is buttressed by extensive evidence showing that partisans know little about the policy positions of party elites (Campbell et al. 1960; Carpini 1996; Luskin 2002; Bawn et al. 2012).3 Evidence of mass ignorance led researchers to investigate how citizens fend in the absence of information. The ensuing studies suggested that people use what little information they have (or what information is made available to them) to infer information they do not have (Brady and Sniderman 1985; Conover and Feldman 1989; Feldman and Conover 1983; Popkin 1991; Lau and Redlawsk 2001). For instance, on learning that a candidate is a woman, some voters infer that the candidate supports 2

Though see also Berelson, Lazarsfeld and McPhee (pg. 310-311, 1954), who summarize some of their findings thus: political preferences have ‘their origin in ethnic, sectional, class, and family traditions,’ and are ‘characterized more by faith than by conviction and by wishful expectation rather than careful prediction of consequences.’ 3 Ignorance about parties’ positions need not mean that partisan affiliation is unprincipled. See Achen (2002) for a theoretical sketch of one such scenario.

4

social welfare programs (McDermott 1997; Lawless 2004). Perhaps a more widespread heuristic is the use of affect to derive politicians’ positions on issues. People (dis)like some groups instinctively, and use that “information” to infer policy positions of group members (Brady and Sniderman 1985; Achen and Bartels 2006).4 The “likability heuristic” —those you like appear ideologically closer to you than those you dislike —is especially relevant in the current era of strong partisan affect (Iyengar, Sood and Lelkes 2012; Iyengar and Westwood 2013).5 The implications of the heuristic are straightforward —uninformed partisans will believe that representatives of their party are ideologically close and representatives of the opposing party, distant. To this point we have discussed how partisans cope in the absence of information. But affect-based reasoning can intercede, even when information is made available. For instance, a large proportion of Republicans believe that President Obama is a Muslim (Hollander 2010) although reliable evidence to the contrary is widely available. The tendency to reject uncongenial information —where congeniality is defined on the basis of group affiliation —is well-established (Lord, Ross and Lepper 1979; Kunda 1987). Seminal work by Lodge and his collaborators has documented the pervasive use of confirmatory bias and motivated skepticism (toward disconfirmatory messages) in information processing; both heuristics allow partisans to protect their sense of identity from short-term threats (Taber and Lodge 2006; Kim, Taber and Lodge 2010; Lodge and Taber 2013).6 Aside from biases in exposure and processing stemming from partisan affect, institutionlevel processes also likely contribute to distorted perceptions of candidates’ policy positions. During election campaigns, many ideologically extreme politicians present themselves as centrists 4 See also Bartels (2002) and Sood, Iyengar and Dropp (2012) who show that partisans learn (or infer or invent) in a motivated manner over the course of election campaigns. 5 While we anticipate increased use of the heuristic during periods of strong affective polarization, we make no claims about its reliability in getting voters to correct positions (see Luskin 2002). 6 See also Bartels (2014); Druckman and Bolsen (2011); Druckman, Peterson and Slothuus (2013); Harrison (2012); Petersen et al. (2013).

5

and their opponents as extremists (Henderson 2013). Famously, George W. Bush, the most ideologically extreme of recent presidents7 ran as a “uniter, not a divider.” This combined with the fact that ads are most effective at persuading voters who share the partisanship of the sponsoring candidate (Ansolabehere and Iyengar 1995) suggests that campaigns likely contribute to distorted perceptions of candidates’ positions. Changes in the market for broadcast news, most notably, the success of Fox News and the emergence of MSNBC as the cable network on the left, mean that there are news outlets that deliver biased rhetoric. And while the number of partisans who limit their exposure to partisan channels remains small, a fair number are exposed to some partisan information regularly via their online social networks (Facebook) (Messing 2014). Due to all these reasons, we expect ideological differences to matter less to partisans than to independents. We further expect partisans to be more sensitive to the actual ideological distance of representatives of the opposing party than representatives of their own party.

Asymmetry in the Ideology-Affect Relationship We begin by showing that the relationship between a representative’s ideology and approval depends on the match between representative and respondent. While approval among co-partisans is nearly unrelated to representatives’ ideology, with more extreme representatives enjoying nearly as much support among co-partisans as their more moderate counterparts, more extreme representatives have much lower approval ratings among supporters of the opposing party than more moderate representatives. In the next section, we show that these differential patterns of approval among in- and out-partisans are in part due to differential accounting of actual ideological proximity by partisans. Data and Measures The data come from a series of monthly state-level polls conducted by SurveyUSA be7

http://voteview.com/blog/?p=735

6

tween 2005 and 2011. SurveyUSA uses Random Digit Dialing and automated interviewing for its polls. All the polls we use have a sample size of at least 600 respondents. In all, we have 1,873 polls, covering all fifty states, with approximately 1.08 million respondents rating over a 100 US Senators. Polls asked respondents about their opinion about officials from their own states. The approval question was typically worded as follows —“Do you approve or disapprove of the job Senator [NAME] is doing?”8 We merged these approval data with DW-Nominate estimates of ideology (McCarty, Poole and Rosenthal 2006). We also present analyses with Campaign Finance based estimates of ideology (CF-Scores)(Bonica 2013). Analytical Strategy To assess the relationship between ideology and support among different partisan groups, we modeled the support that an elected representative enjoys among different partisan groups within a state as a function of representatives’ ideology, office held, and the match between representatives’ and respondent’s party affiliation. To this model, we made three amendments. First, given that a Democrat from Massachusetts is not the same as a Democrat from Texas (a Democrat from Texas is likely to be more conservative than one from Massachusetts and thus expected to like conservative politicians more and liberal ones less), we added state fixed effects to make within state comparisons.9 Second, since we have data on the same representative at multiple time points, we added random effects for representatives. And third, to account for polls that contributed more than one observation to our dataset, we clustered standard errors by polls. Letting i index polls, j index representatives, and s index states, our model takes the form — 8

A very small set of polls —fifteen to be precise —offered more expansive response options, going from strongly approve to strongly disapprove. We dichotomized these ratings, coding neutral and over as ‘approve’ and somewhat or strongly disapprove as disapprove. Removing this polls or dichotomizing using an alternate scheme: counting only ‘approve’ and ‘stronly approve’ as approval makes little difference to our results. 9 Given respondents from Texas were asked to assess only Texas senators, not Massachusetts senators, the stateto-state variance is potentially already covered by the endogeneity of election winning. However, we opt for a conservative specification as the main specification given the size of the data, efficiency is not a concern but bias is. We try out other specifications in section A3 of Appendix A, including introducing state level estimates of ideology of Republicans and Democrats.

7

yij ∼ N (Xj β + αi + γj + δs ; σ2 ) αi ∼ N (0, σα2 ) γj ∼ N (0, σγ2 ) In all, we modeled approval, y, as a function of ideology (X); α captures idiosyncratic variation across senators, γ the dependence between observations, and δ the state fixed effects. To ease interpretation, we rescale all our measures to lie between 0 and 1. For approval, going from 0 to 1 means going from 0% approval to 100% approval. For DW-Nominate and CF-Score, the 0 to 1 range means going from the most liberal representative (DW-Nominate = -.512; CFScore= -1.124) to the most conservative representative (DW-Nominate = .900; CF-Score= .977) in our sample. Results In several national polls conducted between December 2008 and August 2010, Sarah Palin’s support never once slipped below 69% among Republicans. Typically, it was higher than 75%. As we will see below, such robust support for ideologically extreme party leaders is typical. The most conservative Republican Senators (in roughly the top decile of ideology in their own party, with DW-Nominate scores of over .8) enjoyed an average approval rating of over 65% among Republicans. On the other side of the aisle, the most liberal Democratic Senators (with DW-Nominate scores of less than -.4), enjoyed approval from more than 68% of Democrats on average. We start by plotting the locally smoothed relationship between ideological extremity10 and approval separately for in-party and out-party representatives. Figure 1 shows that approval of co-partisan Senators is only weakly related to their ideology. However, the relationship between approval and ideology is much stronger when the Senators are of the opposing party. 10

We coded ideological extremity as follows: We took the absolute value of DW-Nominate scores, folded the scores by party, and rescaled them to lie between 0 (most moderate) and 1 (most extreme).

8

Figure 1: Relationship between ideology and approval by whether or not respondent and representative’s party match

.8 ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ●

.7 ● ● ●● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

.6

● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ●

.5 ● ● ● ● ● ● ● ● ●

● ● ● ●

●● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ●

● ● ● ●

● ● ●● ● ● ● ● ● ● ●●

● ● ● ●

● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ●

● ● ● ●

● ● ● ● ● ●● ●

● ● ● ●

● ● ● ●

● ● ● ●● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ●

● ● ● ● ●● ● ●● ● ● ●● ● ●● ●● ● ●● ● ● ●● ● ●● ● ●●● ●● ● ● ●● ● ● ● ● ●● ● ●● ●● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ●

● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

● ● ●

.1

.2

● ●● ● ●● ● ● ●

.3

.4

.5

Out−Party ● ● ● ●

● ● ● ● ● ●● ●● ●● ● ● ●● ● ● ● ● ●

.6

Ideological Extremity

9

● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ●● ● ●● ● ●● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●

● ● ●● ●

.2

● ● ● ●● ● ● ● ● ● ● ●

● ● ● ● ● ●

● ● ● ● ● ●● ● ● ● ●●● ●● ● ● ●● ●● ● ● ●● ● ● ●● ●● ● ● ● ●●

● ● ● ● In−Party

● ● ● ●

● ● ● ● ● ● ● ● ● ●

● ● ● ●● ● ● ● ● ● ● ● ●

● ● ● ●

.3

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ● ●

●● ● ●

● ● ● ●

● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

.4 ● ● ● ●

● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ●

● ●● ● ●● ● ●● ●● ●● ● ● ● ● ● ●●

● ● ● ●



● ● ●

● ● ● ● ●

● ● ● ● ● ● ●

● ● ● ● ● ● ●

● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ●

● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ●● ● ● ● ●● ● ● ● ●● ●●●

● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Approval

● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ●● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●

.7

.8

.9

For ease of exposition, we split the analyses by party of the elected official and respondent.11 Beginning with Republicans’ evaluations of Republican Senators, more extreme Republican Senators elicit somewhat lower approval ratings (b = −.18, p = .14; see Table 1). In particular, moving from the most liberal Republican Senator in our data (-.05 on DW-Nominate) to the most conservative (.90 on DW-Nominate), changes approval among Republicans by only about 13%. Democrats’ evaluations of the same Republican Senators, however, respond strongly to ideology (b = −.65, p < .001), a more than three-fold increase in responsiveness. Table 1: Approval of Senator as a function of ideology (DW-Nominate). Estimated separately for Republican and Democratic Senators, further split by party of the respondent, with fixed effects for states. Fixed effects for states are not reported.

Republican Senators Republicans Democrats Intercept .87∗∗∗ .91∗∗∗ DW-Nominate −.18 −.65∗∗

Democratic Senators Republicans Democrats .16∗∗∗ .62∗∗∗ 1.09∗∗∗ .37

N Polls 920 920 953 R2 .59 .69 .69 ∗∗∗ ∗∗ ∗ + Note — p < .001, p < .01, p < .05, p < .1

953 .46

As expected, Republicans evaluate more liberal Democratic Senators more harshly (b = 1.09, p < .001).12 But when Democrats rate the same Senators, they are far less responsive to ideology (b = .37, p > .10); ratings of the most conservative Senator (-.05 on DW-Nominate) are merely 12% higher than those of the most liberal Senator (-.512 on DW-Nominate). Given these sizable differences in how partisans respond to in and out-party Senators, the hypothesis that coefficients are equal is easily rejected; the chances of the two coefficients being equal are literally less than one in a million.13 (The results look substantively the same when we substitute 11

These analyses drop states that have only one representative of a party, and where the ideology of the representative doesn’t change across different Congresses. Section A2 in Appendix A presents estimates from a model that pools Republicans and Democrats, using a more complete set of data. 12 The positive coefficient stems from the fact that bigger ideology scores imply more conservative Democrats. And these more conservative Senators elicit more support from Republicans. 13 Since we expect intercepts to vary only by “in” and “out” party, and not depend on whether the “in” or “out” party is Democratic or Republican, we also analyzed the data after splitting by “in” and “out” party. These results

10

CF-Scores for DW-Nominate estimates of ideology. These results are presented in Table A1 of Appendix A.14 ) It is well known that evaluations of incumbents are increasingly polarized (see Jacobson 2006).15 Scholars typically attribute the increased polarization in approval ratings to shifts in elite ideology (see Jacobson 2006; Fiorina, Abrams and Pope 2008). Our results suggest that this account is at best only half accurate. There is a penalty for ideological extremity, but it is generally limited to officials of the opposing party.

Asymmetric Discounting of Actual Ideological Differences A modest relationship between approval and ideological extremity for co-partisan officials does not necessarily mean that partisans’ account for ideological distance differently depending on the party of the representative. The weakness in the relationship can, for instance, be a consequence of lower approval among partisan moderates being offset by greater approval among partisan extremists. To assess the extent to which partisans’ responses to ideological proximity vary by party of the elected official, we estimated the relationship between approval and ideological distance between the representative and the voter. If partisans are evenhanded in their evaluations of party leaders, the relationship between ideological distance and approval should not depend on the party of the representative. If, however, our theory holds, actual ideological distance to leaders of own party should matter less to approval than distance to leaders of the opposing party. We used data from the 2006 CCES. (Appendix B provides details about sampling.) The survey included questions on several key roll-call votes by Senators including votes on (1) the ban on are presented in Table A2 of Appendix A. 14 Redoing analyses using item response theory based measures of ideology by Clinton, Jackman and Rivers (2004) also produce results that are substantively the same. These results are available from the authors upon request. 15 For recent data on polarization in evaluation of presidential performance, see http://www.gallup.com/poll/152222/obama-ratings-historically-polarized.aspx, and Pew: http://www.pewresearch.org/2009/04/02/partisan-gap-in-obama-job-approval-widest-in-modern-era/

11

partial-birth abortion, (2) federal funding of stem-cell research, (3) extending lower tax rates for capital gains, (4) ratifying the Central America Free Trade Agreement (CAFTA), (5) immigration reform, (6) withdrawing troops from Iraq, (7) raising the minimum wage. (See also Ansolabehere and Jones (2010) who use a similar list of votes.) In addition to assessing respondents’ beliefs about the votes of their Senators, the survey asked respondents how they would have voted on each of the issues. We recoded the votes attributed to Senators and the respondent’s votes so that in each case 1 indicated support for the conservative position and 0 indicated the opposite. We jointly scaled the votes by Senators and respondents using a two parameter latent trait model (Lord, Novick and Birnbaum 1968). We modeled the probability of respondent i supporting policy j as a function of xi , respondent’s unobserved ideological position, γj , the policy “discrimination parameter” tracking how strongly respondents’ ideological positions are related to their probabilities of supporting the policy, and αj , the policy “difficulty parameter” or the general level of support enjoyed by a policy —

P (yij = 1) = Θ(γi xi − αi ) Θ is the cumulative distribution function of the standard normal. Scaling the votes using the 2PN latent trait model yields estimates of ideology that are strongly correlated with alternate measures of ideology. For instance, Senators’ positions recovered from these votes correlate with their DW-Nominate scores at .92. Similarly, scaling a survey respondent’s positions on the votes yields a measure that correlates with the 7-point party identification question at .70. Respondent partisan self-identification was assessed using the conventional party identification question. We pooled independents that leaned toward a party with self-identified partisans. Approval of Senators was measured on a four point scale going from “strongly disapprove” (coded as 0) to “strongly approve” (coded as 1). Our measure of spatial ideological distance between Senators and respondents is simply the absolute distance between the Senator’s and re-

12

spondent’s position on the vote-based ideology scale. We rescale it to lie between 0 (no difference between positions) to 1 (all positions are different). We assess asymmetry in the relationship between ideological distance and approval by comparing the relationship across independents, partisans assessing out-party Senators, and partisans assessing in-party Senators. We start by plotting a locally smoothed estimate of the relationship between ideological distance and affect (see first panel of Figure 2). As is clear from the plot, a linear relationship approximates the relationship well. We thus next move to regressing Senatorial approval on the absolute distance between the policy positions of Senators and respondents. Since each respondent assessed two Senators, we cluster standard errors by respondent. Our expectation is that vis-à-vis independents, the relationship between ideological distance and approval will be weaker among partisans. In line with the theory and results above, we further expect the relationship to be still more attenuated when respondents are evaluating Senators of their own party. Simply because the Senator belongs to the same party as the respondent, he or she gets a large bump in their approval. Conditional on actual ideological disagreement, Senators’ approval is on average 13 percent higher among co-partisans than among independents, and 23 percent higher among co-partisans than among opposing partisans (see Table 2; see also the second panel of Figure 2.) (See Table C1 in Appendix C for results from a model that includes a broad set of covariates. The results are nearly identical to the results presented in Table 2.) Table 2: Approval of candidate as a function of actual ideological distance estimated separately among independents, partisans evaluating opposing party Senators, and partisans evaluating co-partisan Senators.

In-Party Out-Party Independents Intercept .80∗∗∗ .57∗∗∗ .67∗∗∗ Absolute distance −.34∗∗∗ −.48∗∗∗ −.54∗∗∗ N 15118 13054 3255 2 R .05 .16 .18 ∗∗∗ ∗∗ ∗ + Note — p < .001, p < .01, p < .05, p < .1

13

1.0

Figure 2: Relationship between actual ideological distance and approval among independents, partisans evaluating opposing party Senators, and partisans evaluating co-partisan Senators Lowess

L nea App ox ma on

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ●● ●●●●● ●●●● ●● ● ●●●● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ●● ●●●●● ●●●●● ● ●●●● ●●

0.8

In−Party

Independents ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ●●● ● ●● ● ●

Approval

0.6

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ●● ● ● ●●● ●● ● ●

A

O

0.4

Out−Party

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●

● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●

0.0

0.2

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

0.0

0.2

0.4

0.6

0.8

1.0

Actual Ideological Distance

A

D

Actual ideological disagreement expectedly increases disapproval among all subgroups However there are striking differences in the strength of the relationship The relationship is strongest among independents (b = −.54, p < .001) somewhat weaker when the respondent belongs to a different party than the Senator (b = −.48, p < .001 Diff

nd

Ou pa y

= .06 p < .05)

and appreciably weaker when partisanship of the respondent is the same as the Senator’s (b = −.34, p < .001 Diff

nd

n pa y

= .20 p < .01) 16

To get a sense of the effect of these differences we simulated how approval ratings would have looked like had partisans used the same spatial calculus as independents —mean approval of co-partisan representatives would have been nearly 20 points lower and mean approval of opposing-party representatives approximately 6 points higher on average Perhaps still more pertinent to our finding of high approval ratings of co-partisan extremists is the fact that partisans are less likely to account for ideological differences with their own party’s elites if those elites are more ideologically extreme than them (to the right of Republicans or to the left of Democrats) (b = −.27, p < .001) than when they are more moderate than 16

We rep cated th s ana ys s us ng a s mp e average across votes Resu ts are substant ve y s m ar wh ch suggests rough y equa we ght ng across ssues

14

them (to the left of Republicans, or to the right of Democrats) (b = −.37, p < .001; bDiff. = .10, pDiff. < .01). In all, partisans react less strongly to ideological differences with their representatives than independents, with partisans being particularly averse to penalizing ideologically distant representatives of their party. Additionally, partisans react less strongly to ideological differences with elites from their own party who are more extreme than them, than to differences with copartisan elites who are more moderate than them.

Explaining the Partisan Asymmetry To this point, we have focused on documenting the asymmetry in the relationship between ideology and partisan affect. But what explains this asymmetry? We have two conjectures —bias in perceptions of ideology of the representatives and the parties, and differences in how partisans use information to judge approval. Both biases are best understood as caused by partisan affect, or mechanisms related to the sense of partisan identity. We first provide evidence of biases in perceptions, and then document differential weighting of information.

Bias in Perceptions of Senators’ Positions Moving from parties to placements of elected representatives, we again turn to CCES 2006 and the questions on Senators’ voting behavior. A majority of the partisan respondents were either uninformed or misinformed about the position of their Senators on each of the votes. We jointly scaled respondents’ positions, respondents’ beliefs about Senators’ positions using a two parameter latent trait model, assuming the data were missing at random. To motivate our discussion of perceptual errors, we assessed the relationship between approval and perceived ideological distance. Expectedly, partisans appear to react far more strongly to perceived ideological differences than to actual ideological differences. (See Table D1 in Ap15

pendix D.) Moreover, the relationship is much more symmetric. While undoubtedly, some of the strength of relationship between approval and perceived differences is a consequence of approval causing perceived distance, endogeneity concerns are likely less severe given proximity measures were built indirectly from positions on separate lists of items, and not via semantic selfand candidate-placement scales that follow each other. Either way, the structure of the boost in coefficients between actual and perceived ideological distance – far greater for in-partisans than out-partisans – strongly hints at the asymmetry we note above. Different kinds of errors can weaken correlations between actual ideological distance and approval. The first possibility is that perceptions of co-partisan Senators’ positions carry more random error than perceptions of opposing party Senators’ positions. We find little evidence for it (see section D2, Appendix D), though we find some differences in how the error is distributed, with error more strongly correlated with ideology of in-partisan representatives than out-partisan representatives (see section D2, Appendix D). The key conjecture about errors in perceptions, however, is about directional error. Consistent with the bias in perceptions of positions of parties that we find in the previous section, we find that on average Democrats thought Republican Senators were more right-wing than Republicans, and Republicans thought Democratic Senators were more liberal than Democrats (see Figure 3). More consequentially, the bias in ratings of co-partisan Senators was considerably higher than ratings of opposing party partisans. Democrats understated the extremity of Democratic Senators by a considerable margin (Mean Diff.perceived - true = .42). Meanwhile, the bias in Democrats’ perceptions of the Republican Senators was considerably smaller(Mean Diff.perceived - true = -.09). Conversely, Republicans understated the extremity of Republican Senators (Mean Diff.perceived - true = -.27), and the extent of error in their perceptions of Democratic Senators was somewhat lower (Mean Diff.perceived - true = .24). Thus, the magnitude of the average directional error in perceptions of co-partisan Senators’ positions exceeded the error in perceptions of opposing-party Senators’ 16

Figure 3: Perceived ideological location of Democratic and Republican Senators by Party of Respondent Democratic Senators

0.8

Republican Senators

0.6

0.0





−0.6

0.2

−0.4

0.4

−0.2



0.0





● ●

Democrats

Republicans

Democrats

17

Republicans

positions. These differences in the structure of error in perceptions of in- and out-party candidates have consequences for correlations. The correlation between DW-Nominate scores and perceived ideology of in-party Senators is about .69, about 7% lower than the same correlation for opposing party Senators (r = .75; pDiff. < .05). Lastly, we assess consequences of perceptual errors. We expect the greatest errors in perceptions of co-partisan Senators’ positions to be among respondents furthest away from their own party (so the greater the error, the lower the approval). Conversely, we expect the greatest errors in perception of opposing party Senators’ positions to be among respondents closest to the opposing party (the greater the error, the higher the approval). To assess this claim, we regressed approval on absolute distance between perceived and actual positions. Expectedly, the errors are positively correlated with approval for opposing partisans, and negatively correlated with approval for in-party elites (see Table 3). Further, and in line with expectations, there was no relation between errors made by independents and approval. However, as the low r-squares suggest, errors do not explain much of the variance in approval, though taken jointly, they go some ways toward explaining the asymmetry. Table 3: Approval of candidate as a function of absolute distance between perceived policy positions and actual policy positions estimated separately among independents, partisans evaluating opposing party Senators, and partisans evaluating co-partisan Senators.

In-Party Out-Party Independents Intercept .78∗∗∗ .20∗∗∗ .44∗∗∗ Error −.30∗∗∗ .43∗∗∗ .04 N 15118 13054 3255 R2 .03 .04 .00 ∗∗∗ ∗∗ ∗ Note — p < .001, p < .01, p < .05, + p < .1

In all, we find —1) greater bias in perceptions of in-party candidates than opposing party candidates, 2) greater bias (and greater error) in perceptions of co-partisan candidates who are more extreme than the respondent, than co-partisan candidates more moderate than the respon18

dent, and 3) that these errors matter modestly, partly explaining the asymmetry we observe in the relationship between actual ideology and approval.

Impact of Provision of Information on Approval As we have acknowledged, the relationship between perceptual errors and approval is likely endogenous —no doubt, errors in perceptions affect approval, but approval also likely causes perceptions. So how would representatives fare if their policy positions were made known to partisans? To understand how partisans account for known differences, we report results from an experiment in which we attributed policy positions to hypothetical candidates at random. The experiment allows us to estimate the effects of policy information on candidate evaluations. When voters are made aware of candidates’ positions, to what extent do they factor policy distance into their evaluations of the candidates? As we discuss below, we find that partisans approve of more distant (and more ideologically extreme) co-partisan candidates no less than more proximate (and more moderate) co-partisan candidates. Research Design We conducted the survey experiment in early 2013 on a sample drawn from the online optin panel maintained by YouGov. (See Appendix B for details about sampling.) The experiment was limited to self-identified or leaning partisans (n = 954). The experiment manipulated positions of hypothetical candidates on a social welfare issue (government spending), and a moral issue (abortion). We first asked respondents to place themselves on a scale that ranged from 1 (most conservative) to 7 (most liberal) and then presented them with the position of a hypothetical Democrat (or Republican) candidate on the same scale. To increase ecological validity of our experiment, we only randomized over partisan consistent positions. So the Democratic candidate could only be at 5 or 7 and the Republican candidate could only be at 1 or 3 on both the issues. After presenting the respondents with candidate positions, we asked respondents to evaluate the candidate on the basis of their policy position. Respondents rated the candidates on a five point 19

scale that ranged from strongly approve to strongly disapprove; we recoded these ratings to lie between 0 and 1 with higher values indicating greater approval. (For exact wording of each of the questions and a screenshot of one of the treatment conditions, see Appendix E.) Results On the 7-point government services scale, on average, Democrats placed themselves at 4.9 while Republicans placed themselves at 2.2. The corresponding figures on abortion were 5.5 and 3.1. The means for either policy did not differ significantly across conditions. Thus, on both issues, the average Democrat was considerably closer to the Democratic representative with the more moderate position (5) than to the Democratic representative with the more extreme position (7), and the average Republican was much closer to the Republican representative with the more moderate position (3) than to the Republican representative with the more extreme position (1). In such situations, the commonly used symmetric spatial utility model predicts greater support for the representative with the more moderate position. However, in 2 of the 4 cases, the representative with the more extreme position enjoyed greater support from co-partisans than the representative with the more moderate position (see Table 4). Democrats’ mean approval rating of the Democratic candidate who held the most liberal position on abortion was .74, comfortably higher than the average rating of .60 received by the Democratic candidate with the more moderate position (p < .01). Similarly, Republicans’ ratings of Republican candidate with the most conservative position on government services were considerably higher than for the candidate with the more moderate position (Mean Diff. = .10, p < .05). In the other two cases, the difference in support for extreme and moderate representatives was statistically indistinguishable from zero. Thus, even when given information, partisans are disinclined to penalize more distal and more extreme co-partisans. On the flip side, partisans consistently responded to the position of opposing partisan candidates in expected ways: opposing party candidates with more extreme positions drew far less support than more more moderate candidates. In all four cases, the out-party representative 20

Table 4: Approval of candidate by candidate position on the issues.+

Dem. Candidate 7 5 Diff.

Rep. Candidate 3 1 Diff.

Gov. Services Democrat .61 .65 −.04 .33 .19 .14∗ N 129 128 121 128 ∗ Republican .10 .27 −.17 .70 .80 −.10∗ N 97 111 106 134 Abortion Democrat .74 .60 .14∗ .25 .16 −.09∗ N 142 125 127 112 ∗ Republican .23 .33 .10 .60 .62 −.02 N 126 108 109 105 Note ∗ p < .05 + Independents were not interviewed in the study. with the more extreme position received less support. The average decline in support across the four cases was roughly .12 (p < .05). Finally, we present some suggestive analysis that sheds additional light on why increases in ideological extremity do not erode support among co-partisans. We think it likely that partisans who hold more extreme views often also care more about ideology. The striking finding that partisans sometimes support the co-partisan candidate with the more extreme, and on average, more distant position may thus be a consequence of partisans with more extreme positions being more responsive to ideological distance than partisans with more moderate positions. To assess this possibility, one can regress approval on self-placement, distance to the candidate, and the interaction between self-placement and distance. However, given candidates were only allowed to take certain positions, distance to the candidate and self-placement can be perfectly correlated. For instance, Democrats most distant to the most liberal Democratic candidate also have the most conservative position. Given this limitation of the design, we opted for something simpler. We compared how approval ratings of the two Democratic candidates by Democrats with the most liberal position (7) differed from ratings by more moderate Democrats (who placed themselves 21

at ‘5’) (i.e., comparing a distance of 0 and 2 points away from the candidate). We replicated the analysis for Republicans, comparing ratings of the two Republican candidates by Republicans with the most conservative position (1) and Republicans who placed themselves at 3 (again 0 and 2 points away from the candidate). In each of the four cases, more extreme partisans approved of the co-partisan candidate with the same position as theirs considerably more than the more moderate partisans, with average ratings of around .9 versus .7 (p < .01). Moreover, the drop in approval in response to increasing policy distance by two points was always considerably sharper among the more extreme partisans (an average decline of .3) than among more moderate partisans (an average decline of .1) (Mean Diff. = .2, p < .01). In all, the evidence suggests that extremist partisans’ greater sensitivity to ideological distance partly explains why more moderate partisan representatives, closer to the party mean, generally fare no better than considerably more extreme representatives.

Discussion Elected representatives are far more ideologically extreme than their supporters (Bafumi and Herron 2010). This state of affairs has been termed a ‘disconnect’ (Fiorina and Abrams 2012). Our evidence suggests that the term is a misnomer. On average, partisans believe that their party is as close to their own ideological position as it was before the start of the current divergence between the parties. Partisans also feel warmly towards their own parties (Iyengar, Sood and Lelkes 2012) and approval of co-partisan leaders bears little relationship with their ideological extremity. Thus, the chances of entry of a more moderate third party look remote. In fact, some of our evidence suggests that if there is an opening for new entrants, it is further to the extremes, as indicated by the recent rise of Tea Party movement. Our findings also potentially help explain the conundrum of polarized ratings of ideologically extreme elites by less-than-extreme voters. It appears that low approval ratings of opposing 22

party elites are to a large degree indeed a consequence of the “positions they take” (Jacobson 2006), but high approval ratings of the same elites among co-partisans are likely a result of motivated reasoning (or elite miscommunication). It is likely that this basic asymmetry also helps explain the over-time pattern in partisan affect —near constant warm feelings toward the in-party and steadily cooler feelings toward the opposing party. In discovering this asymmetry in partisan perceptual bias, our study also makes a contribution to our understanding of the nature of partisan bias. To date, partisan bias has been assumed to be symmetric, i.e. equal bias in perceptions of policy positions of co-partisan and opposing party leaders. Our study documents that bias is greater in perceptions of the positions of the in-party, and its leaders. This finding converges with some other findings on perceptual bias. For instance, over the course of the campaign, partisans’ perceptions of their proximity to the in party candidate shows little change, while the perceived distance between their position and that of the out-party candidate increases sharply (Sood, Iyengar and Dropp 2012). In many other ways, the evidence we present here is consistent with the more standard account of partisanship. For instance, we find that partisans are less responsive to ideological distance from partisan representatives than independents. This finding is consistent with evidence from other scholars who find that independents are more likely to vote for the spatially proximate candidate than partisans (Jessee 2012; Shor and Rogowski 2012). Other studies that have manipulated positions of hypothetical candidates have also found evidence of partisan bias in voting decisions (Sniderman and Stiglitz 2012; Rogowski 2014). Sniderman and Stiglitz (2012), however, offer a different explanation for the bias, arguing that voters take both the candidate’s and party’s position into account when evaluating a candidate. This may well be true though the data underlying the claim merely suggest that those who know the parties’ policy reputations choose between candidates as if they were. We think it more plausible that attenuation of the weight informed partisans put on ideological disagreements with candidates is more due to “unprincipled” reasons. 23

At any rate, even if informed partisans account for policy reputations, the heuristic may still be normatively unappealing. For it is not clear to us whether partisans should average across party reputations and candidate positions when judging candidates. In a legislative system lacking mechanisms to ensure party discipline, with a tradition of sincere voting (so much so that the most frequently used ideology estimates of representatives are based on how they vote in Congress), and no tradition of coalition politics, it is far from clear why people should vote for anyone other than the spatially proximate candidate. There are reasons to prefer the more distant candidate of one’s own party than the more spatially proximate candidate only if one expects some dilution in ideology due to coalition politics. If representatives vote their ideological positions, voters should choose the representative closest to them in the ideological space. All of this is not to say that the average voter has a utility function that is spatially symmetric. Indeed our data are not consistent with that hypothesis. Instead, many of our results are consistent with a directional model of voting (see Rabinowitz and Macdonald 1989). In line with Rabinowitz and Macdonald (1989), we find that candidates have a wide berth in choosing their position on issues, and “that candidates can compete successfully by taking extreme stands on issues” (Rabinowitz and Macdonald 1989). Neither is it to say that all voters have the same utility function. Parallel to the finding among legislators by Carroll et al. (2013), data from the experiment strongly suggests that more extreme partisans are more sensitive to ideological differences. There are psychological theories of persuasion that bear on the partisan asymmetry documented in this paper. In particular, our findings bear a striking resemblance to the classic “assimilation-contrast” model of attitude change developed by Hovland and his colleagues in the 1950s (Hovland, Harvey and Sherif 1957; Sherif and Hovland 1961). When voters have strong attachments to political parties, the model predicts a “boomerang” effect whereby the position attributed to a disliked party perceived as ideologically distant is pushed even further away from the receiver’s position (the contrast effect). For instance, a Democrat who encounters a Republican campaign ad on government spending enlarges the discrepancy between herself and the 24

Republican on the issue. On the other hand, when the initial discrepancy between the party and receiver is relatively small, and the party is held in high regard —the case of intra-party communication —the receiver shrinks any difference and assimilates the party’s position to her own. (For evidence of assimilation and contrast effects in voter perceptions, see Merrill, Grofman and Adams (2001).) In the current era of ill will across the party divide, this model provides a parsimonious account of the polarizing effects of campaign communication. In closing, we note that our findings have pessimistic implications for the prospect of representation based on ideological proximity between candidates and voters. Party elites enjoy considerable leeway to stake out positions at odds with the preferences of their supporters. And, given the extent to which partisans dislike each other (Iyengar, Sood and Lelkes 2012; Iyengar and Westwood 2013), candidates have weak incentives to take positions that appeal to the supporters of the other party. Finally, partisan asymmetry in candidate evaluation is also likely to have an impact on mass polarization. Democrats witnessing Republican voters’ adulation of Sarah Palin in 2008 may have concluded that ordinary Republicans were just as right wing as their standard bearer. Thus, the failure of voters to penalize —and the propensity of activists to reward —ideologically extreme candidates from their own party gives partisans on the other side good reason to (mistakenly) infer that opposing partisans are just as extreme as their leaders.

25

References Achen, Christopher H. 2002. “Parental socialization and rational party identification.” Political Behavior 24(2):151–170. Achen, Christopher H and Larry M Bartels. 2006. It feels like weŠre thinking: The rationalizing voter and electoral democracy. In Annual Meeting of the American Political Science Association, Philadelphia. Aldrich, John H. 1995. Why parties? The origin and transformation of party politics in America. Vol. 15 Cambridge Univ Press. Ansolabehere, Stephen and Brian F Schaffner. 2011. Re-examining the validity of different survey modes for measuring public opinion in the US: Findings from a 2010 multi-mode comparison. In AAPOR Annual Conference, Phoenix AZ. pp. 12–15. Ansolabehere, Stephen and Philip Edward Jones. 2010. “Constituents’ Responses to Congressional Roll-Call Voting.” American Journal of Political Science 54(3):583–597. Ansolabehere, Stephen and Shanto Iyengar. 1995. Going negative. Simon and Schuster. Bafumi, Joseph and Michael C Herron. 2010.

“Leapfrog representation and extremism: A

study of American voters and their members in Congress.” American Political Science Review 104(03):519–542. Bartels, Larry M. 2002. “Beyond the running tally: Partisan bias in political perceptions.” Political Behavior 24(2):117–150. Bartels, Larry M. 2014. “Remembering to Forget: A Note on the Duration of Campaign Advertising Effects.” Political Communication 31(4):532–544.

26

Bawn, Kathleen, Martin Cohen, David Karol, Seth Masket, Hans Noel and John Zaller. 2012. “A Theory of Political Parties: Groups, Policy Demands and Nominations in American Politics.” Perspectives on Politics 10(3):571–597. Berelson, Bernard, Paul Felix Lazarsfeld and William N McPhee. 1954. Voting: A study of opinion formation in a presidential campaign. Binder, Sarah A. 2005. “Elections, Parties, and Governance.” The legislative branch pp. 148–70. Bonica, Adam. 2013. “Ideology and interests in the political marketplace.” American Journal of Political Science 57:294Ű–311. Brady, Henry E and Paul M Sniderman. 1985. “Attitude attribution: A group basis for political reasoning.” The American Political Science Review pp. 1061–1078. Campbell, Angus, Philip E Converse, Warren E Miller and Donald E Stokes. 1960. The American Voter. New York: Wiley. Carpini, Michael X Delli. 1996. What Americans know about politics and why it matters. Yale University Press. Carroll, Royce, Jeffrey B Lewis, James Lo, Keith T Poole and Howard Rosenthal. 2013. “The structure of utility in spatial models of voting.” American Journal of Political Science 57(4):1008– 1028. Clinton, Joshua, Simon Jackman and Douglas Rivers. 2004. “The statistical analysis of roll call data.” American Political Science Review 98(02):355–370. Conover, Pamela Johnston and Stanley Feldman. 1989. “Candidate perception in an ambiguous world: Campaigns, cues, and inference processes.” American Journal of Political Science pp. 912– 940. 27

DiMaggio, Paul, John Evans and Bethany Bryson. 1996. “Have American’s social attitudes become more polarized?” American journal of Sociology pp. 690–755. Druckman, James N, Erik Peterson and Rune Slothuus. 2013. “How elite partisan polarization affects public opinion formation.” American Political Science Review 107(01):57–79. Druckman, James N and Toby Bolsen. 2011. “Framing, motivated reasoning, and opinions about emergent technologies.” Journal of Communication 61(4):659–688. Evans, John H. 2003. “Have Americans’ Attitudes Become More Polarized?ŮAn Update*.” Social Science Quarterly 84(1):71–90. Feldman, Stanley and Pamela Johnston Conover. 1983. “Candidates, issues and voters: The role of inference in political perception.” The Journal of Politics 45(04):810–839. Fiorina, Morris P, Samuel A Abrams and Jeremy C Pope. 2008. “Polarization in the American public: Misconceptions and misreadings.” The Journal of Politics 70(02):556–560. Fiorina, Morris P and Samuel J Abrams. 2012. Disconnect: the breakdown of representation in American politics. Vol. 11 University of Oklahoma Press. Fiorina, Morris P, Samuel J Abrams and Jeremy C Pope. 2005. Culture war?: The Myth of a Polarized America. Pearson Longman New York. Green, Donald, Bradley Palmquist and Eric Schickler. 2004. Partisan hearts and minds: Political parties and the social identities of voters. Yale University Press. Greenstein, Fred. 1965. Children and Politics. Transaction Publishers. Harrison, Brian F. 2012. “Bully Partisan or Partisan Bully?”. Annual Meeting of the American Political Science Association. Henderson, John. 2013. “Issue Distancing in Congressional Elections.” Working Paper. 28

Hetherington, Marc J. 2005. Why trust matters: Declining political trust and the demise of American liberalism. Princeton University Press. Hill, Seth and Chris Tausanovitch. 2014. “Non-Polarization in the American Public.” Working Paper. Hollander, Barry A. 2010. “Persistence in the perception of Barack Obama as a Muslim in the 2008 presidential campaign.” Journal of Media and Religion 9(2):55–66. Hovland, Carl I, OJ Harvey and Muzafer Sherif. 1957. “Assimilation and contrast effects in reactions to communication and attitude change.” The Journal of Abnormal and Social Psychology 55(2):244. Iyengar, Shanto, Gaurav Sood and Yphtach Lelkes. 2012. “Affect, Not Ideology A Social Identity Perspective on Polarization.” Public opinion quarterly 76(3):405–431. Iyengar, Shanto and Sean Westwood. 2013. Fear and Loathing Across Party Lines: New Evidence on Group Polarization. In Annual Meeting of the American Political Science Association. Jacobson, Gary C. 2006. “The polls: polarized opinion in the states: partisan differences in approval ratings of governors, senators, and George W. Bush.” Presidential Studies Quarterly 36(4):732–757. Jennings, M Kent, Laura Stoker and Jake Bowers. 2009. “Politics across generations: Family transmission reexamined.” Journal of Politics 71(3):782–799. Jessee, Stephen A. 2012. Ideology and Spatial Voting in American Elections. Cambridge University Press. Jones, David R. 2001. “Party polarization and legislative gridlock.” Political Research Quarterly 54(1):125–141. 29

Kim, Sung-youn, Charles S Taber and Milton Lodge. 2010. “A computational model of the citizen as motivated reasoner: Modeling the dynamics of the 2000 presidential election.” Political Behavior 32(1):1–28. Kunda, Ziva. 1987. “Motivated inference: Self-serving generation and evaluation of causal theories.” Journal of Personality and Social Psychology 53(4):636. Lau, Richard R and David P Redlawsk. 2001. “Advantages and disadvantages of cognitive heuristics in political decision making.” American Journal of Political Science pp. 951–971. Lawless, Jennifer L. 2004. “Women, war, and winning elections: Gender stereotyping in the postSeptember 11th era.” Political Research Quarterly 57(3):479–490. Levendusky, Matthew. 2009. The partisan sort: How liberals became Democrats and conservatives became Republicans. University of Chicago Press. Lodge, Milton and Charles S Taber. 2013. The rationalizing voter. Cambridge University Press. Lord, Charles G, Lee Ross and Mark R Lepper. 1979. “Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence.” Journal of Personality and Social Psychology 37(11):2098. Lord, Frederic M, Melvin R Novick and Allan Birnbaum. 1968. Statistical theories of mental test scores. Addison-Wesley. Luskin, Robert C. 2002. “From denial to extenuation (and finally beyond): Political sophistication and citizen performance.” Thinking about political psychology pp. 281–301. Mason, Lilliana. 2014. “ŞI Disrespectfully AgreeŤ: The Differential Effects of Partisan Sorting on Social and Issue Polarization.” American Journal of Political Science .

30

McCarty, Nolan M, Keith T Poole and Howard Rosenthal. 2006. Polarized America: The dance of ideology and unequal riches. mit Press Cambridge. McDermott, Monika L. 1997. “Voting cues in low-information elections: Candidate gender as a social information variable in contemporary United States elections.” American Journal of Political Science pp. 270–283. Merrill, Samuel, Bernard Grofman and James Adams. 2001. “Assimilation and contrast effects in voter projections of party locations: Evidence from Norway, France, and the USA.” European Journal of Political Research 40(2):199–221. Messing, Solomon. 2014. Selective and inadvertent exposure in an age of social media: How social and mobile web technologies are reshaping the news media landscape PhD thesis Stanford University. Petersen, Michael Bang, Martin Skov, Søren Serritzlew and Thomas Ramsøy. 2013. “Motivated Reasoning and Political Parties: Evidence for Increased Processing in the Face of Party Cues.” Political Behavior pp. 1–24. Popkin, Samuel L. 1991. The reasoning voter: Communication and persuasion in presidential campaigns. University of Chicago Press. Rabinowitz, George and Stuart Elaine Macdonald. 1989. “A directional theory of issue voting.” The American Political Science Review pp. 93–121. Rivers, Douglas. 2007. Sampling for web surveys. In Joint Statistical Meetings. Rivers, Douglas and Delia Bailey. 2009. Inference from matched samples in the 2008 US national elections. In Proceedings of the Joint Statistical Meetings. pp. 627–639. Rogowski, Jon. 2014. “Voter Decision-Making with Polarized Choices.” Working Paper. 31

Sherif, Muzafer and Carl I Hovland. 1961. “Social judgment: Assimilation and contrast effects in communication and attitude change.”. Shor, Boris and Jon C Rogowski. 2012. “Congressional Voting by Spatial Reasoning, 2000-2010.” Working Paper. Sniderman, Paul M and Edward H Stiglitz. 2012. The reputational premium: A theory of party identification and policy reasoning. Princeton University Press. Sood, Gaurav, Shanto Iyengar and Kyle Dropp. 2012. Coming to Dislike Your Opponents: The Polarizing Impact of Political Campaigns. In Annual Meeting of the American Political Science Association. Stonecash, Jeffrey M, Mark D Brewer and Mack D Mariani. 2003. Diverging parties: Social change, realignment, and party polarization. Westview Pr. Taber, Charles S and Milton Lodge. 2006. “Motivated skepticism in the evaluation of political beliefs.” American Journal of Political Science 50(3):755–769. Tausanovitch, Chris and Christopher Warshaw. 2013. “Measuring Constituent Policy Preferences in Congress, State Legislatures and Cities.” The Journal of Politics 75(2):330–342.

32

Appendix A: Relationship Between Ideology of Representatives and Approval A1: Impact of Ideology on Senatorial Approval Using CF-Scores We re-estimated the relationship between affect and ideology using CF-Scores. Expectedly, substituting CF-Scores for DW-Nominate made little difference to our substantive results. Republicans’ ratings of Republican Senators hardly respond to Senator’s ideology (b = −.00; see Table A1). Meanwhile Democrats ratings of the same Senators are much more responsive to their ideology (b = −.39). The reverse pattern holds for Democratic Senators. Their ratings by Democrats move much more modestly to changes in their ideology (b = .12) than their ratings by Republicans (b = .42). Table A1: Approval of Senator as a function of ideology (CF-Scores). Estimated separately for Republican and Democratic Senators, further split by party of the respondent, with fixed effects for states. Fixed effects for states are not reported.

Predictors Intercept CF-Score

Republican Senators Republicans Democrats .75∗∗∗ .73∗∗∗ −.00 −.39∗∗

Democratic Senators Republicans Democrats .25∗∗∗ .66∗∗∗ .42∗∗∗ .12

N Polls 920 920 953 2 R .60 .67 .71 ∗∗∗ ∗∗ ∗ + Note — p < .001, p < .01, p < .05, p < .1

953 .46

A2: Relationship between Ideology and Affect by Whether or Not Representative and Respondent Ideology Match To estimate the impact of ideology on approval of Senators that belong to the respondent’s party, and Senators that belong to the main opposing party, we split ideology data by party and rescaled it to lie between 0 (most moderate Senator in our data for the party) to 1 (most extreme Senator in 33

our data for the party). The asymmetry is clear. Respondents’ approval of Senators of their own party is mostly linearly unassociated with extremity (b = −.02, see Table A2 but also Table A3). The reaction to extremity of Senator of the opposing party is considerably sharper (b = −.27, p < .001). Aside from differences in slopes is a sharp difference in intercepts, best interpreted as credit given to Senators for simply belonging to respondents’ party. Table A2: Approval of Senators as a function of ideological extremity (DW-Nominate). Estimated separately where respondent and Senator party match, and where they don’t, with fixed effects for states. Fixed effects for states are not reported.

Intercept Extremity (DW-Nominate)

Own Party Senators .76∗∗∗

Opposing Party Senators .59∗∗∗

−.02

−.27∗∗∗

N Polls 1457 1457 2 P suedoR .44 .58 Note —∗∗∗ p < .001, ∗∗ p < .01, ∗ p < .05, + p < .1

A3: Alternate Specifications We model the support an elected representative enjoys among different partisan groups within a state as a function of representatives’ ideology, her position, and whether or not representatives’ party matches that of the partisan group. Since our data come from state level polls, we must account for inter-state differences in kinds of partisans that can affect relationship between ideology and approval. For instance, if there is positive correlation between “state ideology” and the (state level) representative’s ideology, as is likely, estimate of the relationship between ideological extremity and approval is liable to be biased towards zero. In light of these concerns, the most conservative model —our main model in the paper —allows for only within state comparisons. However, one can also model cross-state differences. This modeling strategy should produce unbiased estimates of differences in how partisan groups react as long as ideology of partisan 34

groups moves in tandem across states. To that end, one can estimate a model that controls for state level ideology. A still more conservative model —in that it requires even fewer assumptions controls for ideology of Democrats and Republicans separately. We estimate this model using measures of state party ideology from (Tausanovitch and Warshaw 2013). Results from this model are presented below. Aside from this model, we also estimated other other models in which we control for year fixed effects, gender of the elected representative, and years in position. Our results are robust to these modifications as well. Results from these latter specifications are available from authors upon request. Table A3: Approval of Senators as a function of ideology(DW-Nominate) with random effects of states.

Republican Senators Republicans Democrats Intercept .31∗ .91∗∗∗ DW-Nominate −.03 −.53∗∗∗ Rep. Ideology .38∗ Dem. Ideology .17

Democratic Senators Republicans Democrats .64∗∗∗ .73∗∗∗ 1.03∗∗∗ .09 ∗ −.34 .03

N Polls 916 916 953 2 R .77 .80 .83 Note —∗∗∗ p < .001, ∗∗ p < .01, ∗ p < .05, + p < .1

35

953 .76

Appendix B: Details About YouGov Sampling Methodology The YouGov sampling methodology is as follows. They first draw a random sample from a large high quality RDD sample (designated as the target sample), and then recruit a sample from their online panel that matches the target sample on designated attributes. The end result is a roughly “representative” sample (Rivers 2007; Rivers and Bailey 2009; Ansolabehere and Schaffner 2011). For our survey experiment, YouGov used the 2007 American Community Study (ACS) as the sampling frame to which it added data on party identification, ideology, and political interest from the 2008 Pew Religious Life Survey using matching. The target sample was selected by stratifying on age, race, gender, education using simple random sampling within strata. Next, YouGov used nearest neighbor matching to find respondents from the online panel that most closely resembled the characteristics of the target sample on gender, age, race, education, party identification, ideology, and political interest. This yielded the final matched sample to which the questionnaire was administered. Imperfections in matching were addressed by post-stratification weights.

36

Relationship Between Approval and Ideological Distance We here present results from models that utilize all the data. Model 1 in Table C1 is a complete data version of the subgroup models presented in Table 2 of the main text. As one can see, the coefficients in Model 1 (Table C1) are almost the same as those in Table 2. Furthermore, adding a broad set of exogenous covariates has little effect on the coefficients of interest (see Model 2 in Table C1). In particular, we added age, education in five categories (base category: college graduates), gender (female = 1), race coded in five categories (base category: asian), and martial status coded in four categories (base cateory: married). For each variable, we coded missing responses as a separate category.

37

Table C1: Relationship between approval and ideology Model 2 .67∗∗∗ (.02) In-Party .23∗∗∗ (.01) Out-Party −.10∗∗∗ (.01) Abs. Diff. −.54∗∗∗ (.01) Abs. Diff.*In-Party .13∗∗∗ (.01) Abs. Diff.*Out-Party .06∗∗∗ (.01) Age .00∗∗∗ (.00) Educ.: HS or Less −.01∗ (.00) Educ.: Missing .02 (.03) Educ.: Post-grad .01 (.00) Educ.: Some College −.00 (.00) Female .03∗∗∗ (.00) Race: Black −.00 (.01) Race: Hispanic −.02 (.01) Race: Other −.07∗∗∗ (.02) Race: White −.04∗∗ (.01) Marital: Missing .02∗∗ (.01) Marital: Other .01∗ (.00) Marital: Single .01∗∗∗ (.00) AIC 17601.04 17365.23 BIC 17673.43 17555.25 Log Likelihood -8792.52 -8661.61 Num. obs. 62849 62849 Num. groups: a1 2 2 Variance: a1.(Intercept) .00 .00 Variance: Residual .08 .08 ∗∗∗ p < 0.001, ∗∗ p < 0.01, ∗ p < 0.05 Intercept

Model 1 .67∗∗∗ (.01) .23∗∗∗ (.01) −.09∗∗∗ (.01) −.54∗∗∗ (.01) .14∗∗∗ (.01) .06∗∗∗ (.01)

38

Appendix D: Perceptual Bias in Placement of Senators D1: Relationship Between Approval and Perceived Distance Regressing approval on perceived distance yielded coefficients that were uniformly healthier than coefficients obtained when we regressed approval on actual ideological distance. Expectedly, the difference between coefficients for actual distance and perceived distance were substantially greater for in-party representatives (a 28 point difference), than out-party (3 points) or independents (9 points). Note too that now the in-party representative furthest away from the respondent nets on average a paltry 18 points, versus a hefty 46 points (see Table 3 in the main manuscript). Table D1: Approval of candidate as a function of perceived ideological distance estimated separately among independents, partisans evaluating opposing party Senators, and partisans evaluating co-partisan Senators.

In-Party Out-Party Independents Intercept .83∗∗∗ .52∗∗∗ .63∗∗∗ Perceived distance −.62∗∗∗ −.51∗∗∗ −.63∗∗∗ N 15118 13054 3255 2 R .12 .21 .18 ∗∗∗ ∗∗ ∗ + Note — p < .001, p < .01, p < .05, p < .1

D2: Structure of Random Error in Perceptions of Senators’ Ideology Greater random error in perceptions of in-party candidates vis-‘a-vis out-party candidates may explain why association between actual ideological distance and approval is weaker when partisans assess in-party candidates versus out-party candidates. However, we find little evidence of greater absolute error in perceptions of co-partisan Senators vis-‘a-vis perceptions of opposing Senators (Mean Absolute ErrorIn-party = .192, Mean Absolute ErrorOut-party =.186). While the mean absolute error may be the same across perceptions of in- and out-party representatives, it may be the case that error is more heavily correlated with ideology of copartisan representatives than among opposing party representatives. If so is the case, it will also 39

help explain our initial set of results which suggest that ideological accountability is especially weak for the most extreme of the representatives. We find that absolute error in respondents’ beliefs about co-partisan Senators’ positions was more than twice as strongly correlated with Senator’s ideological extremity (r = .14), than respondents’ beliefs about out-party Senators (r = .06; pDiff. < .05). The positive correlation between extremity and ideology in perception of in-party candidates naturally shows up in perceptions of candidates more extreme than the respondent. We find 25% greater error in perceptions of in-party candidates who are more extreme than the respondent (Mean Absolute Error = .21) than perceptions of in-party candidates more moderate than the respondent (Mean Absolute Error = .15, p < .05).

40

Appendix E: Details About Experimental Manipulations Government Spending Some people think the government should provide fewer services even in areas such as health and education in order to reduce spending. Suppose these people are at one end of a scale, at point 1. Other people feel it is important for the government to provide many more services even if it means an increase in spending. Suppose these people are at the other end, at point 7. And, of course, some other people have opinions somewhere in between, at points 2, 3, 4, 5 or 6. Where would you place yourself on this scale or haven’t you thought much about this?

Abortion Next we would like to know about your position on the issue of abortion. Some people believe that by law, abortion should never be permitted. Suppose these people are at one end of a scale, at point 1. Others feel that by law, a woman should always be able to obtain an abortion as a matter of personal choice. Suppose these people are at the other end, at point 7. And, of course, some other people have opinions somewhere in between, at points 2, 3, 4, 5 or 6. Where would you place yourself on this scale or haven’t you thought much about this?

41

Figure E1: Screenshot of the survey experiment

42