Over the last fifty years, political parties at the national, state and local levels in the United States have increasingly nominated ideologically extreme candidates.Footnote 1 During the same period, formal party leaders'Footnote 2 influence over party nominations has declined; outside groups now dominate many primaries (for example, Masket Reference Masket2009). Many scholars have suspected that these two trends are related, with parties' and formal party leaders' declining influence in primaries contributing to the declining number of party nominations going to centrist candidates (for example, Persily Reference Persily2015).
The scholars who expect party leaders to support centrists often argue that parties are ‘the sole political organizations whose primary goal is to win [general] elections’ (La Raja and Schaffner Reference La Raja and Schaffner2015; see also Hassell Reference Hassell2017, Reference Hassell2018) and that parties are more likely to win general elections when they nominate centrists (for example, Hall Reference Hall2015).Footnote 3 This line of reasoning suggests a way to decrease elite polarization: namely, by ‘enhancing the role of parties’ and their leaders in primary elections through reforms that strengthen parties' autonomy, internal organization, financial resources and power over candidate nominations (McCarty Reference McCarty and Persily2015, 143).
In this article, we add an important caveat to this line of reasoning. Political party leaders at all levels of government influence candidate emergence and primaries – not just party leaders at the state or national levels, who have been the focus of most recent work. And in local political parties, we argue, there are reasons to doubt that party leaders will act as advocates for centrist candidates.
Scholars have seldom studied the preferences of local party leaders, but they play a paramount role in the candidate entry process. In many areas, local party leaders recruit, screen and support candidates for local, state legislative and even national office (Crowder-Meyer Reference Crowder-Meyer2010; Crowder-Meyer Reference Crowder-Meyer, Green and Coffey D2011; Maisel Reference Maisel, Cohen, Fleisher and Kantor2001; Moncrief, Squire and Jewell Reference Moncrief, Squire and Jewell2001). They wield considerable influence over who runs and who wins (especially in primaries), thanks in part to recent changes in party structure and campaign finance (Hershey Reference Hershey2013). However, scholars have paid little attention to local party leaders' preferences in primaries; we do not know whether they share the affinity for moderate candidates that scholars have documented among state and national party leaders. Moreover, the political party structure in the United States is more a ‘stratarchy’ – a system in which each level has some power to act on its own – than a hierarchy in which higher levels of the party determine the behavior of lower levels (Eldersveld and Walton Reference Eldersveld and Walton2000).Footnote 4 We therefore cannot take for granted that US local party leaders will mirror the views and behaviors of their counterparts at the national and state levels.
In contrast to much of the recent work on state and national party leaders, our argument in this article is that US local party leaders – especially Republicans – should often not think that centrists have strong advantages in elections, in part because certain features of local politics may bias their perceptions of the general electorate. Compared to state and national party leaders, for instance, local party leaders – especially Republicans – are less insulated from ideologically extreme activist organizations, which actively seek to alter their perceptions of public opinion (for example, Grossmann and Hopkins Reference Grossmann and Hopkins2016; Layman et al. Reference Layman2010; Skocpol and Hertel-Fernandez Reference Skocpol and Hertel-Fernandez2016). Such activism has blossomed in recent years, especially on the political right (Blee and Creasap Reference Blee and Creasap2010; Skocpol and Hertel-Fernandez Reference Skocpol and Hertel-Fernandez2016).
Moreover, compared to state and national party leaders, local party leaders tend to have less concrete information about public opinion (like detailed polling) that would allow them to form accurate perceptions of the general electorate. Compared to state and national parties, local parties are also easier for grassroots activists to capture by installing themselves and their allies as party leaders (Skocpol and Williamson Reference Skocpol and Williamson2011). Persily (Reference Persily2014) thus worries there may be cause to ‘fear…capture [of parties] by the extremes’ (see also La Raja and Schaffner Reference La Raja and Schaffner2015, 22). Against this backdrop, we hypothesize that many local party leaders might not perceive strong electoral incentives to nominate centrists, which would in turn dampen their desire to nominate centrists or even lead them to prefer nominating extremists.
In this article, we use an original survey of county party leaders – one of the few large-sample surveys of this population ever conducted – from which we gather a methodologically diverse set of experimental and descriptive evidence. The methodologically diverse nature of our evidence means it has complementary strengths and weaknesses. Together, our findings present a rare glimpse into how local party leaders decide which candidates to back for their party's nomination. Our findings are generally consistent with the implications of our theoretical argument.
Our first finding is drawn from a conjoint experiment (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014; Teele, Kalla and Rosenbluth Reference Teele, Kalla and Rosenbluth2018) that asked party leaders to choose between hypothetical candidates with randomly assigned personal qualities, including ideology. We asked party leaders which candidates they would encourage to run, which would be more likely to win if nominated, and which would be more likely to remain loyal to the party if elected. Contrary to conclusions from research on national and state party leaders, local party leaders clearly prefer centrist candidates the least; they most often favor candidates who share the views of typical co-partisan voters. Leaders from both parties even prefer nominating extremists to nominating centrists. Intriguingly, and consistent with our argument, local party leaders seemed to favor extremists to centrists in part because they did not believe their party would face an electoral penalty for nominating an extremist and therefore did not perceive an electoral incentive to nominate centrists, contrary to some of the rosier accounts of state and federal party leaders.
This conjoint experiment also uncovered a stark partisan difference: when faced with a choice between an even more extreme candidate and a moderate, Democratic Party leaders prefer the extremist by a ratio of 2 to 1, whereas Republican party leaders prefer the extremist 10 to 1. Consistent with our theoretical argument, we show that Republican leaders believe that extremists are actually more likely to win general elections than centrists (whereas Democrats simply view extremists as not significantly more likely to lose). Republicans thus believe they ‘can have their cake and eat it, too’: nominating extremists, they believe, provides both ideological and electoral rewards.
Our second finding concerns how party leaders perceive the public, further investigating the apparent differences between how Republican and Democratic party leaders assess the electoral chances of extreme candidates. We present data from survey items that asked party leaders broad questions about how they perceive general public opinion in their districts. Consistent with the idea that Republicans are more likely than Democrats to see extremists as winning candidates, we find that both Democratic and Republican party leaders perceive public opinion in their counties as more conservative than it really is (compared to data from large public opinion surveys), a disconnect that may help explain Republicans' greater enthusiasm for extremely conservative candidates and Democrats' (relative) restraint in selecting extremely liberal candidates.
Finally, for our third finding, we present further evidence consistent with the conjoint experiment's finding that party leaders prefer to nominate extreme candidates. This analysis uses data from a question that appeared on the paper version of the survey to examine what party leaders spontaneously say about the traits they look for in candidates for their party's nomination.Footnote 5 We find that party leaders rarely mention centrism or moderation but often spontaneously say that they prefer ideological loyalists. For example, Republican leaders mention conservatism as an ideal nominee trait over six times more often than they mention centrism.
Throughout, we show that our results are robust and generalize across partisan contexts: to counties that are evenly divided by party, to counties where each party is favored, and to counties that party chairs subjectively perceive as competitive. Weighting all our results by county size also does not affect our conclusions.
Our findings have two particularly important implications. First, to the extent that future reforms strengthen local party leaders, they may empower individuals who do not perceive the sharp trade-offs between candidates' extremity and electability that scholars have documented among state and national party leaders. Future reform proposals may therefore benefit from a more surgical approach that empowers national and state leaders who appear more supportive of centrists (Hassell Reference Hassell2018; La Raja and Schaffner Reference La Raja and Schaffner2015) while avoiding empowering local party leaders. Secondly, the findings presented here raise the possibility that local party leaders' preference for nominating extremists may already contribute to the asymmetric polarization of politicians and policy at the local, state and national levels.
Argument and existing evidence
Many political observers argue that party leaders have lost influence over primary elections in the United States, and that this development has contributed to elite polarization at all levels of government.Footnote 6 The formal leaders of party organizations can influence primary elections in many ways: with their information, resources and connections, they can recruit new candidates, direct financial and human resources to nominees they favor, and boost the fortunes of nominees they endorse (Crowder-Meyer Reference Crowder-Meyer2013; Lawless Reference Lawless2012; Masket Reference Masket2016). And because they oversee ‘the sole political organizations whose primary goal is to win [general] elections’ (La Raja and Schaffner Reference La Raja and Schaffner2015), many scholars argue that party leaders have historically used their influence to block extremist candidates, who might cost their party the general election (for example, Hall Reference Hall2015), in favor of candidates closer to the ideological center, who better balance their ideological and electoral concerns. Unfortunately, some scholars maintain, this process has broken down over the last several decades, as changes to nomination processes and campaign finance laws have reduced party leaders' influence in primaries and increased the sway of interest groups and activists (for example, Masket Reference Masket2009; Skocpol and Hertel-Fernandez Reference Skocpol and Hertel-Fernandez2016), who often back more ideologically extreme candidates. As party leaders' power over primaries has waned, scholars argue, their ability to keep extremist candidates at bay has also waned.
Scholars have argued that these conclusions have important implications for political reformers. In particular, these studies suggest that interventions that increase party leaders' influence in primary elections could help decrease elite polarization. In this view, newly empowered (or perhaps, re-empowered) party leaders will use their influence to ‘clamp down on candidates and incumbents outside the mainstream’ and throw their support behind centrists (Persily Reference Persily2015, 132), thereby ‘exercis[ing] a moderating effect’ in primaries (La Raja and Schaffner Reference La Raja and Schaffner2015; see also Barber and McCarty Reference Barber, McCarty and Persily2015; Levitsky and Ziblatt Reference Levitsky and Ziblatt2018) that flows downstream into general elections and ultimately political institutions. According to this school of thought, party officials serve as a rare bulwark against extremism and polarization that needs to be reinforced. Because of this belief, for example, La Raja and Schaffner (Reference La Raja and Schaffner2015, Ch. 6) advocate ‘a party-centered system of campaign finance’ that dramatically relaxes or eliminates spending limits on party organizations, frees party organizations to provide unlimited financial and in-kind aid to candidates, and redirects public financing to party organizations. For these scholars, more influence for party leaders could be an important way to combat growing polarization.
Most work in this school of thought has focused on national and state party leaders, and largely side-stepped local party organizations. We believe local parties are important for two reasons, however. First, in many areas, local party leaders already play an important role in recruiting, screening and supporting candidates for local, state legislative and even national office (Crowder-Meyer Reference Crowder-Meyer2010, Crowder-Meyer Reference Crowder-Meyer, Green and Coffey D2011). For example, in a survey of candidates for state legislative office, we found that over 57 per cent of candidates indicated that their local party organization was important in encouraging them to run for office (Broockman et al. Reference Broockman2015).Footnote 7 Likewise, in the survey of local party leaders we present below, 46 per cent said they ‘played a very important role in recruiting or supporting primary election candidates’ for state legislature or the US House.Footnote 8 Secondly, many reforms designed to empower national and state party leaders to play a larger role in primaries – such as campaign finance rules that advantage parties and disadvantage outside groups in primaries – may empower local leaders to play a larger role, too.Footnote 9
Unfortunately for the cause of ideological moderation, many local party leaders may have different preferences in primaries than their national and state counterparts. That is, there are good reasons to doubt that local party leaders are as firmly opposed to extreme candidates as state and national party leaders.
For one, we suspect that many local party leaders are highly exposed to forces that bias their perceptions of the general electorate, leading them to believe that the population is more open to ideologically extreme candidates.Footnote 10 In recent years, sophisticated activist networks have often worked to barrage local elites with expressions of support for extreme policies, such as via direct communication, at town halls, with protests, through partisan media (Goss Reference Goss2008; Hacker and Pierson Reference Hacker and Pierson2005; Skocpol and Hertel-Fernandez Reference Skocpol and Hertel-Fernandez2016; Skocpol and Williamson Reference Skocpol and Williamson2011). A principal aim of these tactics is to alter elites' perceptions of public opinion and of the electoral viability of polarized candidates. As leaders reflect on what kind of nominees voters prefer, they may think of the most vocal activists (Miler Reference Miler2009). Where these efforts succeed, local party leaders may tend to think voters are more ideologically extreme and may not worry that extreme nominees could cost them the general election.
Secondly, unlike state and national party leaders, many local party leaders may not have the organizational resources to offset these perceptual biases, such as detailed polling, extensive data on a large number of candidates and detailed voter files. Just as researchers with strong priors and small samples are more likely to reach extreme conclusions, local party leaders with fewer informational resources may be more likely to conclude that their voters are ideologically extreme.
Thirdly, selection could also play a role: people who believe their polarized ideologies are favored in general elections (compared to those who do not) might be more likely to agree to serve as party chairs in the first place (Thomsen Reference Thomsen2017).
In contemporary US politics, we expect these kinds of patterns to be especially pronounced among Republican local party officials. At the local level, left-leaning organizations have atrophied in recent decades, while their conservative counterparts are experiencing something of a renaissance (in part thanks to strategic investments by conservative donors and organizations and a rich network of conservative media outlets that include television, internet and radio (Blee and Creasap Reference Blee and Creasap2010; Skocpol and Williamson Reference Skocpol and Williamson2011)). Consistent with these patterns, recent studies suggest that Republican state legislative officials overestimate public support for conservative positions, but that Democrats do not overestimate public support for liberal positions (Broockman and Skovron Reference Broockman and Skovron2018). Other evidence suggests that Republican elites think differently about how to succeed in general elections, believing that nominating extremists allows them to ‘fire up the base’ (Buchler Reference Buchler2015). In light of these differences, we expect Republican local party leaders to be especially likely to think that ideologically extreme candidates can win elections.
Of course, our argument about local party leaders represents a significant departure from much of the recent work on national- and state-level party leaders. This work has argued that state and national party leaders tend to prefer moderates (La Raja and Schaffner Reference La Raja and Schaffner2015; Hassell Reference Hassell2018), but these state and national leaders may be different than local party leaders: they have access to sophisticated organizational and informational resources, and they focus on a broader set of election outcomes. However, even scholars who have studied national and state party leaders have recently wondered whether their findings extend to local party leaders and expressed concern about the possibility that local parties ‘nurture…ideological activists’ (La Raja and Schaffner Reference La Raja and Schaffner2015, 22; see also Persily (Reference Persily2014)).Footnote 11 Our original studies bear out these concerns and join other evidence (for example, Buchler Reference Buchler2015) that raises doubts about whether party leaders are truly the moderating force they are depicted as, at least at the local level. Our data also supports the novel mechanism we identified as a potential contributor to local party leaders' cool reception of centrists: most local party leaders – and especially Republicans – appear to prefer nominating typical partisans and even extremists over centrists because they do not think centrists are more electable.
Data
To better understand local party leaders' support for ideologically centrist candidates, and to determine whether their perceptions of the electorate might underlie their preferences, we fielded a national survey of the chairs of the county-level (or equivalent)Footnote 12 branches of both parties in 2013.Footnote 13 We chose to study county leaders for two main reasons. First, they are often the most active local party organizations in primary elections for state and federal office (Crowder-Meyer Reference Crowder-Meyer, Green and Coffey D2011; Lawless Reference Lawless2012); over 78 per cent of the party chairs in our sample indicated that people in their county party organization had helped support a candidate in a primary for an open seat. Secondly, counties provide a well-defined sampling frame that allows us to assess the representativeness of our respondents.Footnote 14
To administer the survey, we first manually compiled contact information for 6,219 county party chairs. We gathered this information by searching the internet for the name of every county in the US together with the name of each of the two major parties. In November 2013, we sent each chair a pre-notification and then a survey invitation at his or her email and/or postal addresses. (If both were available, we attempted contact at both.) We received responses from 1,118 chairs (18 per cent), a response rate comparable to recent surveys of politicians (for example, Broockman and Skovron Reference Broockman and Skovron2018) and double that of many mass public surveys.
Representativeness
We conducted extensive checks of the representativeness of our respondents. To save space, we review them briefly here and present them in full in Appendix B. The response rates were nearly identical by region; for Republican (18.0 per cent) and Democratic chairs (17.9 per cent); and for chairs identified as male (18.2 per cent) and female (18.5 per cent). One potential concern is that only party chairs in un-competitive areas would respond. However, Appendix Figure A6 shows that the underlying partisan composition of the areas our respondents are from is fairly representative in both parties. Another possibility is that only chairs from very small counties would respond. Appendix Figure A7 shows that we received a similar response rate from counties of all sizes, and if anything, very slightly more responses from larger counties. Another potential concern is that the most ideologically polarized party leaders would select into our survey. This does not appear to be the case: we compare our survey to a previous survey that secured a very high response rate (45.5 per cent) and found a very similar distribution of reported personal ideology within both parties (Appendix Figure A10). Appendix Table A3 presents formal tests of the representativeness of our respondents' counties.
Finding 1: Chairs do not prefer centrists in a conjoint experiment
Our first piece of evidence is drawn from a conjoint experiment (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014; Teele, Kalla and Rosenbluth Reference Teele, Kalla and Rosenbluth2018). Conjoint experiments force respondents to make trade-offs between two possible choices that differ along several dimensions. Providing respondents with a forced choice allows for a statistical estimation of average revealed preferences over each dimension. Providing multiple dimensions enhances the naturalism of respondents' choices.
In our experiment, we asked county party chairs to pick which of two possible candidates they would prefer to run in their party's primary for an open seat. Our experiments began, ‘Suppose there is a primary for an open [county board/state legislative/US House]Footnote 15 seat in your county and the two individuals below are considering running.’ The survey then described ‘Candidate A’ and ‘Candidate B’ by displaying two side-by-side lists of their personal attributes. After the local party leaders viewed the candidates, we asked ‘Which of the above candidates would you be more likely to encourage to run for office?’ Appendix E gives the full language for each condition and shows how the survey appeared to respondents.
Each aspect of each candidate's biography was independently generated at random: the survey supplied each candidate's gender (signaled by first name), age, occupation, experience in the party, life circumstances, personal characteristics and political ideology. For political ideology, we described some candidates as more moderate than the typical voter in their party (for Democrats, more conservative; for Republicans, more liberal). We described other candidates as similar in ideology to typical party members, and still others as more extreme than typical party members (for Democrats, more liberal; for Republicans, more conservative).Footnote 16 We provided several traits in addition to ideology for each candidate in order to make the scenario more realistic and to assess the relative importance of ideology to chairs' judgments compared to other potentially relevant traits. All traits (ideology, gender, etc.) were independently randomized, allowing us to estimate the effects of each trait, as each trait is uncorrelated with the others by design. Each party leader was shown only one conjoint matchup (because the survey was also administered on paper, making multiple matchups impractical).
Table 1 lists the potential attributes of the hypothetical candidates. Attributes were fully randomized, with the exception of age, which was constant: the first profile was always forty-three years old and the second was always forty-seven years old. Two different sets of first names were used for the two profiles in order to ensure that no pair of candidates had the same name. Appendix Figure A11 shows how an online survey respondent would have seen the experiment.
List of random treatments in conjoint experiment
Note: each potential candidate had one attribute from the listed options for each category. Traits were independently randomized, with the exception of age, which was held constant in each profile.
Because party leaders are difficult to survey, our experiments exposed them to just one head-to-head pairing, in the hopes of discouraging attrition. Obviously, this reduces our statistical power, although as the results show below, our key predictions remain highly statistically significant.Footnote 17
Results: Who Party Leaders Prefer to Run
Figure 1a shows the share of party leaders who selected each candidate type conditional on being shown that type, with 95 per cent confidence intervals overlaid.Footnote 18 (Recall that other candidate attributes were also shown, but that these other traits are uncorrelated with the ideological types by design.) The results show that local party leaders do not prefer to nominate centrists. In fact, leaders of both parties are the least likely to prefer nominating centrist candidates: Democrats are the least likely to select a candidate more conservative than their typical party member and Republicans are the least likely to select a candidate who is more liberal (p < 0.01 for both comparisons). For example, Republicans selected a candidate more centrist (that is, more liberal) than their party in only 7 per cent of the match-ups in which such candidates were shown. Contrary to the hope that local party leaders might be a force for moderation, we find that they most prefer candidates who resemble their co-partisans and next prefer candidates even more extreme than that.
Party leaders tend to prefer more extreme candidates over centrists
Note: estimates (and 95 per cent confidence intervals) are based on hypothetical candidate conjoint experiments embedded in a national survey of county-level party chairs conducted in 2013. Panel A shows the probability that a chair selected a candidate of each type conditional on being shown a candidate of that type, regardless of the other hypothetical candidate's type. The remaining panels calculate the differences between more extreme and more moderate candidates conditional on chairs being shown candidates of both types in a single hypothetical head-to-head. ‘More conservative’ collapses candidates who were ‘much more conservative than the typical voter from your party in your county’ with candidates who were ‘somewhat more conservative than the typical voter from your party in your county’; the ‘more liberal’ category does the same symmetrically for liberal candidates.
To communicate the magnitude of these differences, we next consider the cases in which party leaders were presented with a choice between a candidate who was more centrist than their party and one who was more extreme.Footnote 19 In these cases, party leaders preferred the more extreme candidate 76 per cent of the time, or by a more than 3-to-1 margin (p < 0.01). Disaggregating the data by party reveals that this is largely driven by Republican Party chairs. Democratic Party chairs preferred extremists to centrists 63 per cent of the time (p < 0.01), but Republicans preferred extremists to centrists 91 per cent of the time, or by about 10 to 1 (p < 0.01). To illustrate the size of these differences, Figure 1b shows the ‘win margin’ of the extremist candidate in these extremist-vs.-centrist match-ups, subtracting the share of party leaders that preferred the centrist nominee from the share that supported the extremist nominee. The error bars show 95 per cent confidence intervals.
Figures 1c and 1d demonstrate the robustness and generalizability of this finding. We first examine whether our results persist in competitive counties (since party leaders who are guaranteed to win or lose regardless of the candidate they nominate may have fewer incentives to support moderates, and since competitive counties would be most likely to be affected by any reform that empowered party leaders). Figure 1c shows the results for the subset of county party chairs in objectively competitive counties, where Obama received 40–60 per cent of the two-party vote in 2012,Footnote 20 and Figure 1d shows the subset of county party chairs who subjectively perceive general elections in their area as competitive.Footnote 21 As Figures 1c and 1d illustrate, if anything, party chairs who perceive elections in their area as more likely to be up for grabs are more likely to prefer extremist nominees; Republican Party chairs in such areas prefer extremists by a margin of 15 to 1.
Appendix Figure A2Footnote 22 shows that the results remain unchanged within both Republican-leaning and Democratic-leaning counties. For example, even in Democratic-leaning counties, Republican chairs are the least likely to select centrist candidates. Our findings are also essentially unchanged when we weight by county population; they do not reflect that party chairs over-represent small counties that may tend to be conservative.Footnote 23
Finally, in Appendix Section F.2, we apply a Bonferroni correction to the full conjoint models to account for the multiple comparisons and find that the key average marginal component effects (AMCEs) remain significant. Because our main findings are generally associated with very small p-values, this correction would not change any of our main findings. For example, for the finding in this section in Figure 1b, the p-value for the comparison between Democrats' and Republicans' preference for extremists is 2.18 × 10–6, meaning we would need to have conducted 458,294 comparisons for a Bonferroni correction to render it insignificant.
Potential Mechanism: Perceptions of Candidate Electability
After party leaders selected which primary candidate they preferred, we also asked them which of the two candidates they thought would be more likely to win the general election and which would be more likely to stay loyal to the party if elected. Unsurprisingly, party leaders were significantly more likely to initially select candidates who they perceived as having these qualities, by about a 4-to-1 margin for each.Footnote 24 And, as expected, Figure A3 shows that both sides recognize that extremists are very likely to toe the party line.
More surprising are party leaders' perceptions of electability – and the partisan differences in these perceptions. Figure 2a shows that neither Democratic nor Republican chairs perceived more centrist candidate as more electable. In fact, the main pattern in both parties is that leaders saw more liberal candidates as the least electable (p < 0.01). This means that Republican chairs were most skeptical of the electoral appeal of more moderate (that is, liberal) Republicans relative to typical or extreme Republicans. Democrats were most skeptical of more extreme (that is, liberal) Democrats' electoral appeals, but still saw typical Democrats as most electable.
Do party leaders perceive extreme candidates as less electable?
Note: estimates (and 95 per cent confidence intervals) are based on hypothetical candidate conjoint experiments embedded in a national survey of county-level party chairs conducted in 2013. Panel A shows the probability that a chair selected a candidate of each type as the more electable candidate, conditional on being shown a candidate of that type and regardless of the other hypothetical candidate's type. The remaining panels calculate the differences between more extreme and more moderate candidates conditional on chairs being shown candidates of both types in a single hypothetical head to head. ‘More conservative’ collapses candidates who were ‘much more conservative than the typical voter from your party in your county’ with candidates who were ‘somewhat more conservative than the typical voter from your party in your county;’ the ‘more liberal’ category does the same symmetrically for liberal candidates.
To better illustrate this pattern, Figure 2a again focuses on cases in which party leaders faced a choice between an extremist and a centrist. Democratic chairs are slightly more likely to view centrist candidates as more electable than extremists, although this difference is not statistically significant in match-ups between the two. The picture is quite different for Republican Party chairs, who believe extremist candidates are more likely to win general elections: 75 per cent of Republican chairs indicated they thought the extremist candidate in the conjoint was more likely to win the general election than the centrist candidate, a margin of 50 per cent for the extremist candidates (p < 0.01). This difference persists for Republican chairs who work in closely divided counties, and is even larger for those who subjectively perceive elections in their areas as close.
In Appendix Figure A4,Footnote 25 we also show that the results remain unchanged within both Republican-leaning and Democratic-leaning counties; even in the latter, Republican chairs see centrist candidates as the least electable. In Republican-leaning counties, Democratic chairs begin to see centrists as more electable than extremists. However, Appendix Figure A2b shows that they still do not prefer nominating them in Republican-leaning counties. These results are unchanged when we weight by county population; they do not reflect the fact that party chairs over-represent small counties that may tend to be conservative.
We also asked party chairs which candidate they thought would be most likely to recruit enough volunteers and raise sufficient money. Appendix Figure A5 shows the results. Consistent with the general electability findings, Republicans think extreme candidates are the most likely to do both.
Together, these results suggest an intriguing explanation for why Republican Party chairs prefer extreme candidates for their party's nomination over centrists: unlike Democrats, Republican Party chairs across contexts appear to believe they can ‘have their cake and eat it, too’ by nominating extremists, reaping both electoral and ideological rewards.
Of course, this evidence has several important limitations. First, it assumes that party leaders can understand (or are prone to think in terms of) ideological labels. Secondly, it relies on party chairs' perceptions of their party as a benchmark against which to compare potential candidates. However, as we show in the next section, Republicans if anything overestimate the public's conservatism, meaning that even the candidates they perceive to be similar to partisan voters may in fact also be somewhat extreme relative to Republican voters. And, thirdly, this evidence does not directly assess party leaders' views about voters. If it really is the case that local Republican leaders think extremists fare well in elections, we might expect them to not just think extreme candidates are more electable, but for them to also perceive voters as more conservative than they really are. We analyze this possibility in the next section.
Finding 2: party leaders misperceive public opinion
If (as existing work concludes) party leaders at the national and state levels work to block extreme candidates in the hopes of winning general elections, could it really be the case that local Republican Party chairs see extreme candidates as more electable, and that they view the electorate differently than their Democratic counterparts? To further investigate this finding in a methodologically distinct way, we asked party leaders about their general beliefs about public opinion in their counties and their states.Footnote 26
Data
To ascertain party leaders' perceptions of public opinion, we asked them to estimate public opinion in their county and state on several issues. Specifically, we asked them ‘What percent of people living in your state would agree with the following statements?’ and ‘What percent of people living in your county would agree with the following statements?’ followed by a series of statements taken verbatim from a public opinion survey (described below). Each party chair estimated public opinion for their state and county on three randomly selected issues. We expected these boundaries to be well known to county party chairs, especially as they indicated that they are active in primaries for countywide and statewide office.
To compare party leaders' perceptions to reasonably precise estimates of reality, we asked party chairs to estimate county and state opinion on issues that had been asked in the 2012 Cooperative Congressional Election Study (CCES), a large-sample survey (Ansolabehere Reference Ansolabehere and Schaffner2013). We were therefore constrained in the kinds of issues we could ask about, as the CCES only asked its full sample about their opinions on a limited set of issues. Fortunately, the policy domains the CCES questions covered were almost all salient at all levels of government in the United States.Footnote 27 For example, the issue of immigration was not only a federal matter; states, counties and cities had salient debates over issues such as whether to act as ‘sanctuaries’ for unauthorized immigrants and whether to grant them driver's licenses. These policy domains were thus almost all relevant to the politics of all levels of government, meaning these questions are relevant to ask local party leaders regardless of the levels of government at which they were most active in supporting candidates for office.Footnote 28
Table 2 reports the text of the issue items in the 2012 CCES that were available, as well as the ideological direction of the ‘Yes’ side and whether the policy represented a status quo change in 2013. We also report weighted national mean support for each issue in the CCES.Footnote 29
Issue questions on party leader survey (from 2012 CCES)
Note: this table lists the issues on which party leaders were asked their perceptions of state- and county-level public opinion. Question wording is taken from the 2012 CCES. The second column computes weighted national mean support in the entire 2012 CCES. The third column reports whether answering ‘Yes’ to the survey item reflects taking a liberal or conservative position. The fourth column reports whether the question would have reflected a status quo change in 2012.
Empirical Strategy 1: Raw Data
Because each state and county has a relatively small number of CCES respondents, special care is required to compare party leaders' estimates of public opinion with the CCES’ estimates of true public opinion. We use two approaches that both yield similar results.
Our first approach simply compares the average party leader's perceptions to the average CCES estimate of average county-level public opinion (across the counties where party chairs responded to our survey). Our estimation strategy follows Broockman and Skovron (Reference Broockman and Skovron2018) and is formalized in Appendix Section C. The salient point about this approach is that it does not require large county-level samples to yield accurate estimates of how party leaders' perceptions of public opinion differ from the CCES estimates of public opinion on average. In brief, using the CCES data, we estimate public opinion in the average county or state – what party leaders' average perceptions would be if their perceptions were perfectly accurate. This quantity can be interpreted as ‘the expectation of county or state opinion for a party chair respondent chosen at random’. We also estimate party leaders' mean perception and compare the two quantities. We cluster the standard errors at the county level for our county analysis and at the state level for our state analysis. Note that the county analysis excludes states in which parties are not organized at the county level because the levels at which these parties are organized (parish, etc.) are not available in the CCES data: LA, AK, ND, CT and MA.
Results: Republican Party Leaders Overestimate the Public's Conservatism
Consistent with Finding 1 that Republican Party leaders perceive extreme conservatives as more electable, our public opinion perception data indicate that Republican county-level leaders perceive the public in their counties and states as more conservative than the CCES suggests that it is. Figure 3a shows our estimates of party leaders' perceptions of public opinion in their counties and our estimates from the CCES of what their average perceptions should have been if they were perfectly accurate. Table 3 shows point estimates as well as the sample size of CCES respondents. Because thousands of CCES respondents form each point estimate, these estimates are relatively precise.
Party leaders think public opinion is more conservative than it is
Note: each panel plots average public support estimated from the CCES in counties (Panel A) and states (Panel B) weighted to reflect average public opinion in the county or state of a party leader respondent chosen at random (black points with whiskers reflecting 95 per cent confidence intervals). In grey, each panel also plots the mean perception of party chairs in the survey, along with 95 per cent confidence intervals. Panels are broken up by the chairs' parties and whether responding ‘yes’ to the issue item reflects a conservative or liberal issue position. For liberal issues, mean perceptions are lower than mean public support, meaning chairs underestimate how liberal the public is. For conservative issue, perceptions are higher than mean support, showing that chairs overestimate support for conservative positions.
Comparing party leaders' perceptions of county public opinion to actual public opinion
Note: standard errors are clustered at the county level. For each issue, the table reports mean support in the county of a chair chosen at random according to the weighting procedure outlined in Appendix Section C (Actual Public Opinion), the mean perception of chairs surveyed (Elite Perception), the difference of these two quantities (Average Misperception) and the number of CCES respondents that the weighted estimates are based on. Columns to the left reflect responses from Democratic Party chairs and CCES respondents in the counties represented by those chairs; right-hand columns do the same for Republican chairs and their counties. Liberal policies are ones for which answering ‘Yes’ to the survey question reflects a liberal position, thus, negative values for average misperceptions on these issues reflect underestimating the public's support for liberal policies. The opposite is true for the conservative policies. *** p < 0.001, ** p < 0.01, * p < 0.05
On average, Republican leaders appear to underestimate public support for the liberal policies on the CCES by about 10 percentage points and to overestimate public support for the conservative policies covered by the CCES by almost 40 percentage points. For example, only 13 per cent of CCES respondents believe that ‘Laws governing the sale of firearms should be made less strict than they are’, but Republican county-level leaders perceive their counties as 67 per cent supportive. However, the CCES evidence indicates that about 37 per cent of people in a typical county supported same-sex marriage in 2013,Footnote 30 but the typical Republican county-level leader perceived county support at 27 per cent. Democrats do not consistently overestimate voter liberalism, and indeed if anything appear to overestimate voter conservatism as well.
Figure 3b and Table 4 report the results for party leaders' estimates of state opinion. The results are similar: Republicans overestimate state support for conservative policies and underestimate support for liberal policies.
Comparing party leaders' perceptions of state-wide public opinion to actual public opinion
Note: standard errors are clustered at the state level. For each issue, the table reports mean support in the state of a chair chosen at random according to the weighting procedure outlined in Appendix Section C (Actual Public Opinion), the mean perception of chairs surveyed (Elite Perception), the difference of these two quantities (Average Misperception) and the number of CCES respondents that the weighted estimates are based on. Columns to the left reflect responses from Democratic Party chairs and CCES respondents in the states represented by those chairs; right-hand columns do the same for Republican chairs and their states. Liberal policies are ones for which answering ‘Yes’ to the survey question reflects a liberal position, thus, negative values for average misperceptions on these issues reflect underestimating the public's support for liberal policies. The opposite is true for the conservative policies. *** p < 0.001, ** p < 0.01, * p < 0.05
These results are robust when we limit the estimates of true public opinion to opinions among voters only; voter mean opinion is typically within 1 percentage point of overall mean opinion reported in Table 2, with the largest difference being a 3-percentage-point difference on the religious exemption issue. This difference is nearly an order of magnitude smaller than the magnitude of their average misperceptions, which confirms that our findings are not a result of elites thinking only about voters rather than all residents.
Perceptions of State-Level Opinion: Multilevel Regression and Post-stratification
As a robustness check, we also used multilevel regression and post-stratification (MRP) to estimate true public opinion in each state and compared these state-level MRP estimates to party leaders' perceptions of their state.Footnote 31 MRP uses individual-level survey data and demographic information about the districts from the US Census to construct state-level estimates of support for each issue (Lax and Phillips Reference Lax and Phillips2009; Warshaw and Rodden Reference Warshaw and Rodden2012; Park, Gelman and Bafumi Reference Park, Gelman and Bafumi2004). Our MRP procedure first fits multilevel models to the responses to each issue question from the 2012 CCES. Each model returns estimated effects for demographic and geographic predictors. We then use the estimates from the multilevel model to estimate support for various demographic cells, identified by age, race, education, gender and state. Finally, using data from the US Census' American Community Survey, we weight those cells by their frequency in each state. The result is an estimate of the percent of each state that supports each issue. We then dyadically compare these estimates to party leaders' perceptions. For states with sufficiently large samples, MRP is designed so that the results rely very little on MRP's demographic weighting. Appendix D provides further details.
Figure 4 presents the MRP results. The x-axis on each graph shows the MRP estimate of state support and the y-axis shows party leaders' estimates of state support. If party leaders were perfectly accurate, we would expect their responses to follow the black line, which shows the line y = x. However, the results from the MRP estimates match what we saw in the weighted raw data: Republican Party leaders consistently overestimate support for conservative policy positions, whereas Democrats do not do the same with liberal policy positions.
Comparing party chairs' perceptions of state opinion to MRP estimates of state opinion
Notes: Democratic chairs' estimates are in solid lines; Republican chairs' estimates are shown in dashed lines. The x-axis on each graph shows the MRP estimate of state support and the y-axis shows party leaders' estimate of state support. The black line, which shows the line y = x. Each dot represents one chair's estimates. The lines show loess smoothed local averages. Liberal policies are ones for which answering ‘yes’ to the survey question reflects taking a liberal position, the opposite is true for conservative policies.
These MRP results help alleviate three possible concerns about the findings from the weighted raw data. First, one possible concern with the raw data is that innumeracy leads party leaders to simply answer 50 per cent (or some other amount) by default. However, Figure 4 shows that most party chairs do not answer at any particular threshold, and that there is a clear correlation between the truth and their answers – albeit offset by a large intercept shift in the case of Republicans, due to their overestimation of state conservatism. Secondly, one might worry that party chairs are simply loath to admit their party's ‘side’ is not favored by the majority. However, it is clear from Figure 4 that Republican Party chairs still overestimate conservatism even when their side is favored; for example, in states where same-sex marriage does not receive majority public support, Republican Party chairs still estimate its support to be much lower than it is. Finally, one might worry that the results are driven by Republicans in one particular kind of state, such as the majority of states that lean rightward. However, Figure 4 makes clear that the misperceptions are consistent across states.
Discussion of Finding 2
These results help reinforce the findings from our conjoint experiment (Finding 1). Our conjoint experiment found that Democratic chairs were more likely to prefer nominating a centrist candidate and less likely to prefer nominating an extreme candidate than their Republican counterparts. The conjoint experiment's further results found that this may be due to the fact that Democrats and Republicans both see more conservative candidates as more electable than more liberal candidates. This likely pushes Democrats to be more enthusiastic about nominating centrists (who are more conservative than typical Democrats) and Republicans to be more enthusiastic about nominating extremists (who are more conservative than typical Republicans). Our findings in this section about how party leaders perceive public opinion represent methodologically distinct evidence that is consistent with the findings from the conjoint experiment. Both Democratic and Republican chairs perceive a general public that is more conservative than the survey evidence suggests. Moreover, Republicans dramatically overestimate the public's conservatism.
Importantly, these differing perceptions suggest that on many issues on which political scientists would expect extremely conservative candidates to take positions that are out of step with public opinion, Republican leaders appear more likely to expect such candidates would be in-step. These differing perceptions are squarely in line with the finding that Republican local party leaders expect very conservative candidates to perform better in general elections than political science conventional wisdom predicts.Footnote 32
Two caveats should be noted. First, it is always possible that other phenomena explain these inequalities – for example, that party leaders rationalize their own more conservative views by projecting them onto the public. Secondly, our data do not allow us to test whether the individual party leaders who choose more extreme candidates are also more likely to misperceive their constituents. In the conjoint experiment (Finding 1), each party chair was shown just one pair of candidates at random, and when asked to perceive public opinion (for Finding 2), party chairs were also shown just a few opinion questions at random, which leaves us with too few cases on any one question to reliably determine whether party chairs who prefer extremists misperceive public opinion. Future research might ask party leaders to evaluate more pairs of hypothetical candidates or answer more questions about their perceptions of public opinion, but as a first cut at this topic, we opted to keep our survey as brief as possible. As a result, we cannot say whether individual party chairs who prefer extremists also misperceive their constituents, but rather, that party chairs as a group both prefer extremists to centrists – especially Republican chairs – and tend to perceive their constituents as more conservative than they really are.
With this said, both of our first two studies relied on explicitly prompting party chairs for their beliefs about the electorate's ideological composition. Our next piece of evidence revisits Finding 1 – that party chairs prefer extreme candidates over centrists in a conjoint – and asks whether local party chairs spontaneously think of centrism or moderation as a desirable quantity when ideology and issues are not explicitly primed.
Finding 3: when describing ideal candidates, chairs rarely mention centrism or moderation
To provide another robustness check on our findings that local party leaders are not pushing their parties to nominate more centrist candidates, on the paper version of the surveyFootnote 33 we asked party leaders an open-ended question: ‘In an ideal world, what personal qualities would you like all of your party's political candidates to have? Please list as many as you would like.’Footnote 34 If local party leaders really prefer more centrist candidates in an effort to win general elections, we expected them to say so. If, however, they consciously prefer more polarized candidates, candidate ideology should be a ‘top of mind’ consideration. That is, if our theory is right, local party leaders' preferences for ideologically extreme candidates should not just be evident in what they do in experiments; they should also manifest in what they say in response to direct questions.
Of the 232 party leaders who completed paper copies of the survey with the conjoint, 84 per cent listed at least one characteristic in response to our question about ideal candidates.Footnote 35 After research assistants blind to the hypotheses of the study grouped these responses into 36 categories,Footnote 36 the data revealed two patterns consistent with our argument. First, only about 9 per cent of local party leaders named centrism or moderation as ideal qualities. Moderation is not irrelevant – some subset of party leaders named it as important – but nine out of ten party leaders did not mention centrism when we asked them to think of their ideal candidates.
Far more named some variant of the concept of ideological or party loyalty. Figure 5 plots the frequency of each type of response category that our research assistants identified, which included four separate characteristics related to ideological loyalty (printed in all caps in the figure): conservative, liberal, loyal to the party and loyal to the constitution. Individually, these traits each made up small percentages of responses, like moderation. Considered together, however – that is, assuming that these four categories are essentially different ways of expressing the notion of ‘ideological loyalty to my party' – these traits made up a strikingly large percentage of responses. Fully 29 per cent of party leaders mentioned one or more of these partisan ideologies as an ideal candidate trait, more than three times the number that mentioned ideological moderation or centrism (difference in proportions p < 0.001).Footnote 37
Percentage of party chairs that mentioned various traits in response to an open-ended question about the ‘ideal’ candidate
Notes: each bar represents the percent of candidates who volunteered each response to an open-ended question about ideal candidate traits. Traits listed in all capitals are partisan or ideological qualities.
The second pattern is that Republican elites mentioned ideological loyalty far more than Democratic elites. Figure 6 plots the percentage of leaders in each party that mentioned each of the five most common traits as well as the percentage that mentioned any of the ideological responses we identified. Republican Party chairs were twice as likely as Democrats to mention ideology (p < 0.001) – the largest interparty difference by far. Appendix Table A1 shows that this difference remains the same size when controlling for 2012 Obama vote share; the results are not driven by the presence of Republicans in disproportionately conservative counties. Appendix Figure A1 replicates the raw data in Figure 5, dividing the sample by party; we find that Republicans mention the category of ‘conservative’ four times as often as ‘moderate’ (29 per cent of the time vs. only 7 per cent of the time).
Republican and Democratic chairs were more likely to mention ideological loyalty than moderation
Notes: percentage values correspond to the share of open-ended responses that mentioned each quality. p-values correspond to difference of means tests between how often chairs of each party mention the quality. The within-party differences in mentioning loyalty vs. moderation were also statistically significant.
These findings mirror other findings that Republican elites place a high premium on ideological loyalty (Grossmann and Hopkins Reference Grossmann and Hopkins2016), but are at odds with any hopes that local Republican local party leaders might seek out centrist candidates in order to win elections. Indeed, while chairs of both parties were more likely to spontaneously mention ideological orthodoxy as a desirable quality than moderation or centrism, this differed by party: Democrats were twice as likely to spontaneously mention liberalism than centrism or moderation as desirable, but Republicans were nearly six times as likely to mention conservatism than centrism or moderation.
In summary, our results are again inconsistent with the idea that local party chairs seek out centrist candidates. These findings should be regarded as more tentative because of the smaller sample size and the open-ended nature of the question. However, it is encouraging that a different methodological approach suggests a finding consistent with Finding 1 from the conjoint: centrism and moderation rarely appear to come to the top of party leaders' minds when they think about their ideal nominees.
Discussion: having their cake and eating it, too
As elite polarization has continued to grow at all levels of government in the United States, scholars and political scientists have begun to ask how reformers might reduce it. Recently, scholars have considered one intriguing possibility: that reforms empowering formal political party leaders might reduce polarization, as these leaders might be more likely to favor nominating centrists in hopes that their parties will perform better in general elections. Evidence has accumulated about the typically moderate candidate preferences of national and state party leaders in primaries (for example, La Raja and Schaffner Reference La Raja and Schaffner2015; Hassell Reference Hassell2018; Persily Reference Persily2015). However, many potential reforms to empower national and state party leaders would also further increase the power of local party leaders – a group that is already influential in many primary elections at all levels of government, yet about whom we know little.
In this article, we reconsider common assumptions about party leaders' perceptions of the electorate, arguing that local party leaders might not see strong incentives to nominate centrist candidates and therefore might not prefer to nominate them. We considered several reasons why this may be the case. For example, local party leaders may be especially likely to be surrounded by like-minded and polarized individuals, such as the bevy of grassroots organizations – especially on the political right – that have attempted to use well-organized and high-visibility events to press their demands on political leaders (for example, Skocpol and Williamson Reference Skocpol and Williamson2011). Many local party leaders and their organizations do not have the same resources, professionalization or breadth of experience that their state and national counterparts have, making them more susceptible to overgeneralizing based on the opinions of co-partisans and activists when thinking about the general electorate.
In a series of original studies, we collected considerable evidence that is consistent with our expectations about local party leaders' preferences and perceptions. Our results suggested that Republican local party leaders are especially unlikely to favor nominating centrists. This appears to be because Republican county-level party leaders believe extreme conservatives are more electable than centrists. In this way, Republican Party leaders appear to believe that nominating extremists allows them to ‘have their cake and eat it, too’ – winning more votes in general elections by offering voters the opportunity to select party loyalists. Their Democratic counterparts see centrists as only slightly more electable than extremists, and so still do not favor nominating them on balance. It may well be the case that the formal leaders of local parties are less enthusiastic about typical partisan candidates and extremists than other local party activists like donors and interest group leaders. But our evidence shows that many local party leaders still remain enthusiastic about non-centrists: they do not appear to believe that nominating centrists will help their parties win. Even if local party leaders face a trade-off between nominating electable centrists and less-electable loyalists in reality, many of them do not perceive one.
Of course, our studies have several limitations, and we would welcome future research that addressed them. First, our studies use survey data and hypothetical candidate experiments, rather than data on how local party leaders actually behave or the choices they make when the stakes are real. It could always be the case that party leaders are less willing to take a chance on an extreme candidate in surveys or a hypothetical decision than they would be in real life. Although our survey-based approach allowed us to conduct experiments with a high degree of internal validity and better capture key theoretical mechanisms, observational data on how local party leaders recruit candidates in practice would clearly complement our approach.Footnote 38
A second area in which future research could build on our own is to better understand why party leaders have the perceptions of the electorate that they do. We established a new empirical finding about local party leaders (their preferences in primaries) and a likely contributing mechanism (their perceptions of the electorates' preferences), but the mechanism for this mechanism (why party leaders maintain these perceptions) is less clear. Although we had several reasons to expect Republican party leaders to overestimate the electorate's conservatism, more research is needed to understand how important each of these are. For example, it could be instructive to understand how party chairs' misperceptions of public opinion vary according to their experiences both prior to becoming chair and once in office.Footnote 39 Consistent with one reason we expected Republicans to maintain such beliefs, however, we did find that Republicans thought extreme candidates would be more likely to secure support from activists, since they would be better able to recruit volunteers and raise money than centrists.
Thirdly, our research took as its starting point prior studies that suggest that national and state party leaders prefer to nominate centrists, but our data do not allow us to directly compare county-level party chairs to national and state party chairs using a common survey instrument. We hope that future research will be able to do so.
Fourthly, our research suggests that party leaders are more likely to prefer to nominate extremists, and that they are more likely to view voters as ideologically conservative (leading Republicans to strongly favor extremists and Democrats to temper their preferences for extremists due to concerns about electability), but our data do not allow us to directly ascertain whether individual leaders who prefer extreme candidates are more likely to misperceive public opinion. We hope this point will be taken up in future research.
Relatedly, our data do not allow us to test an intriguing possibility – that Democrats' misperceptions of public opinion may offset their preference for ideologically extreme candidates – that is, the ‘extremist’ candidates who Democrats prefer may in fact be moderate relative to actual public opinion. Because the conjoint experiment (Finding 1) asked party chairs about candidates who were more or less extreme than the typical co-partisan and our public opinion perception items (Finding 2) asked party chairs about the views of the typical voter, we cannot be sure, but we hope future research will use a common scale when asking party leaders about the kinds of candidates they support and their perceptions of public opinion.
Finally, our evidence was collected at a single point in time. Future research could help us understand whether our findings are time-bound. For example, our theory would predict that if liberal grassroots groups became as active and organized in pressuring local party organizations as conservative grassroots groups have been, party chairs' perceptions of their incentives might shift. It is too soon to tell whether groups like Indivisible that have sprung up in reaction to Donald Trump's election will succeed to the extent conservative groups have, but future studies that track this activity and measure party chairs’ perceptions again could prove informative about mechanisms. Our work can serve as a point of comparison for any such future research that will allow it to uncover any such changes over time.
Our findings also suggest two intriguing possibilities with regard to the literature on asymmetric polarization (for example, Mann and Ornstein Reference Mann and Ornstein2013). First, our findings that Democratic Party leaders seem less sanguine about extremists' electoral prospects than Republicans suggest a new mechanism that may underpin asymmetric polarization at the local, state and national levels. Since Republican local party leaders believe they can ‘have their cake and eat it, too’ when considering potential nominees, they appear to advocate nominating extremists – which may contribute to the rightward movement of Republican nominees. In addition, our data also suggest an intriguing potential strategy for reducing polarization: consistent with recent field experiments (Butler and Nickerson Reference Butler and Nickerson2011), supplying local party leaders with more reliable information about public opinion and their incentives might change their perceptions and reduce their support for extremists. If local party leaders came to believe they were undermining their party's electoral prospects, they might be less likely to favor nominating extremists. This hypothesis is ripe for future research. Most broadly, our results underscore the importance of not taking the incentives political actors perceive for granted; just as scholars of public opinion theorize about and study how voters subjectively perceive the political world, scholars of elite decision making should seek to carefully understand not just the strategic realities political elites face, but also how they perceive them (for example, Bowler, Donovan and Karp Reference Bowler, Donovan and Karp2006; Broockman and Skovron Reference Broockman and Skovron2018; Miler Reference Miler2009).
Supplementary material
Replication data sets can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/L5PBPL and online appendices at: https://doi.org/10.1017/S0007123419000309.
Acknowledgements
The authors are grateful for advice and feedback from Skip Lupia, Nick Valentino, Jeremy Gelman, Ted Brader, Seth Masket, and several anonymous reviewers.
