Asymmetric Partisan and Ideological Evaluations from Candidate Repositioning: A Randomized Experiment ()
1. Introduction
An enduring part of American politics includes politicians repositioning from one side of an issue to another throughout their careers, often dubbed negatively as “flip-flopping”. Many salient examples exist, including George H.W. Bush saying “read my lips no new taxes”, or John Kerry’s “vote for the $87 billion [military funding bill] before” voting against it. Before flip-flopping became part of the political parlance, critics used the term “waffling” after Gerald Ford described Jimmy Carter as someone who “waffles and wiggles” on issues ( Allgeier et al., 1979 ). Repositioning has also been framed more generally—for example, Lyndon Johnson was criticized for having a “credibility gap” on the Vietnam war ( Spragens, 1980 ). Candidates are sometimes criticized as willing to “say anything to get elected” as Al Gore was by George W. Bush in 2000 ( Kartik & McAfee, 2007 ). The implication of all these examples is that repositioning on issues comes with a reputational cost, and that being consistent is valued in a democracy. The present study evaluates a candidate who repositions from a liberal to a conservative immigration position using a sample of American citizens.
Scholars have shown an interest in repositioning for some time. Downs (1957) argued that voters want consistency from candidates “only insofar as those statements serve as guides to the policies that party will carry out in office” (107). Policy positions during a campaign provide a road map of how a candidate will perform in office, and voters want clarity on how a candidate will govern. Holding a consistent position is also a way to maximize party politics in legislatures, and candidates who reposition away from their party hurt their party brand ( Snyder & Ting, 2002 ). Beyond theoretical examinations of repositioning, empirical tests can be difficult because many factors occur during a campaign and while in-office, which makes isolating the effect of the repositioning tenuous. In addition, the circumstances around each repositioning case—like the ones described in the first paragraph—have time period-specific features, which make generalizing difficult. In addition, successful politicians do not reposition randomly and instead repositioning may occur for a variety of strategic reasons like interest group pressure (e.g., Karol, 2009 ). Because of this, political behavior studies regularly use experiments to randomize repositioning and isolate its effects. Many experiments show that repositioning in-office is viewed as worse than being consistent while in-office (e.g., Hoffman & Carver, 1984 ; but also see the next section for a thorough review of the literature). The present study extends past experimental research by including a campaign control group to compare with in-office treatment conditions and asking about candidate ideology.
The present experiment uses three vignette conditions to compare a candidate for the House of Representatives who 1) holds a position on immigration during a campaign (specifically, raising the cap on asylum seekers entering the United States), 2) a candidate who holds the same position of raising the cap once in-office (consistent), and 3) a candidate who repositions to supporting the current cap once in-office. Having a condition for just the campaign position acts as a control group and allows for a direct test of whether candidates are rewarded for consistency once in-office (as opposed to just comparing consistency against repositioning in-office). This campaign control group provides leverage on whether or not citizens use campaign information to influence how they view politicians once in-office. In addition to asking candidate favorability, this study uses an outcome measure that asks about perceived ideology of the candidate in each randomized condition. This outcome measure has not been asked in previous experiments. Lastly, the present study evaluates repositioning by a respondent’s ideology and party identification.
Across all respondents, results suggest that repositioning influences favorability and perceived ideology. Candidates received a premium for holding a consistent position and were punished for repositioning. Respondents substantially updated their ideological evaluations of the candidate in the direction of the repositioning. In addition, I show that repositioning from raising the asylum seeker cap to maintaining the current cap is drastically penalized in terms of favorability by respondents who supported raising the cap (labelled in this manuscript as “liberal respondents”). However, respondents who supported lowering the cap (labelled as “conservative respondents”) only modestly rewarded the repositioned candidate with higher favorability ratings. Large differences also existed in ideological evaluations. Among liberal respondents, the candidate was perceived as much more conservative after repositioning, but conservative respondents did not update their ideological perceptions of the candidates in the conservative direction as strongly as liberals. These results suggest that conservative respondents used the liberal campaign position to form a strong prior when evaluating the candidate, which limited favorability gains from conservatives. On the other hand, liberal respondents were more sensitive to the conservative repositioning and did not reward the candidate for the initial campaign position. In the next sections, I review the literature, describe the methodology, examine the results, and conclude with a discussion of repositioning.
2. Theoretical and Empirical Tests of Repositioning
Reasons exist for why it might be suboptimal for candidates to reposition after an election. Downs (1957) argued that consistent positioning from one time period to another is valuable; that “voters regard reliability as an asset” (107). Candidates who are consistent on their positioning have “integrity”—that is, “policy statements at the beginning of an election period are reasonably borne out by its actions” while in-office (108). Stimson (2015) also argued that repositioning might signal that a candidate is unprincipled about their issue positions. Voters might want consistent positioning for ideological reasons; that is, they want to accurately evaluate how close the candidate is to their own positions. Or, voters might value consistency from a valence perspective, where consistent candidates are viewed as trustworthy or competent. Taking both reasonings together, candidates are advantaged by consistency to maintain support or maximize votes in the next election. Downs’ (1957) theory was extended to include ideological labels (i.e., liberal, moderate, or conservative) as a summary of multiple policy positions that voters use to assess the distance to their own ideologies (e.g., Enelow & Hinich, 1984 ; Jessee, 2012 ). Although it is not directly tested, the implication from these models is that repositioning might also influence perceived ideology of the candidate (whether candidates are viewed as liberal or conservative after repositioning).
A candidate might also be disadvantaged by repositioning away from their party’s positions. Cox and McCubbins (2005) show that the majority party is able to control the agenda in Congress as a way to prevent opportunities for repositioning away from the party in roll call votes. For example, bills that get a vote on the floor must have a majority of the majority party’s support ( Cox & McCubbins, 2005 ). This suggests that consistency within a party is needed to ensure that the party brand is in good standing. Formal models also show that politicians who are “mavericks” and support policies that are not backed by their party are worse off than politicians who are consistently liberal or conservative ( Snyder & Ting, 2002: p. 93 ). In this model, party brands are used as a cue (similar to an ideological label) to signal policy platforms, and the signal becomes muddled when candidates hold conflicting positions with their party. In addition, experimental evidence exists that candidates might feel psychological pressure to be consistent and fulfill their campaign promises once in-office ( Corazzini et al., 2014 ; Guo, 2020 ). Fulfillment of campaign promises can be interpreted differently depending on the voter’s predispositions and the specificity of the campaign pledge ( Dupont et al., 2019 ; Guo, 2020 ). In the experiment to follow, I used specific and unambiguous question wording to make sure the candidate’s repositioning is not subject to multiple interpretations.
Repositioning and randomized experiments have a long history. The first wave of experiments that were conducted by teams of psychologists offer valuable evidence but have some drawbacks ( Allgeier et al., 1979 ; Hoffman & Carver, 1984 ; Sigelman & Sigelman, 1986 ). The very first repositioning experiment used a “stranger” as the person doing the repositioning ( Allgeier et al., 1979 ), but others used very detailed vignette about hypothetical politicians ( Sigelman & Sigelman, 1986 ). These were all small-scale experiments conducted on college students where only one study exceeded 100 students ( Sigelman & Sigelman, 1986 ). Once the vignettes were geared more toward political repositioning and not interpersonal, researchers showed decreases in favorability from repositioning as well as a reduction in valence traits like competency and trustworthiness.
The second wave of repositioning experiments was conducted by political scientists. These experiments typically used samples that were more representative of American citizens instead of student samples and used vignettes of various repositioning situations. Several of these repositioning experiments focused on the issue of war, from the Vietnam War to the second Iraq war ( Spragens, 1980 ; Levendusky & Horowitz, 2012 ; Croco & Gartner, 2014 ; Croco, 2016 ). Experiments on the Afghanistan and Iraq wars showed no penalty for hypothetical Senators who repositioned in press releases as long as they supported the respondent’s preferred position, suggesting that the electorate’s views on policy matters ( Croco & Gartner, 2014 ; Croco, 2016 ).
Although penalties are documented in experiments, repositioning occurs regularly among presidents and legislators. In fact, some legislators are able to have long careers in-office due to repositioning to a newfound party position (e.g., Karol, 2009 ). Simply randomizing repositioning verses consistent positioning (and only those two possibilities) does not account for other factors that might mitigate the negative effects of repositioning. To that end, some experiments explore this need for external validity. For example, when a politician uses persuasive messaging to justify their reposition, they are able to persuade citizens of their new position ( Robison, 2017 ). Excessive criticisms of “flip-flopping” from unelected groups like the media might also backfire and not further reduce a politician’s favorability ( Gooch, 2022 ). Issues that have a high level of uncertainty, like an international crisis, also afford politicians some leeway in changing positions ( Levendusky & Horowitz, 2012 ). Citizens might also be more willing to accept a reposition if the original position is from many years ago ( Doherty et al., 2016 ).
Taken together, ideological proximity is crucial for understanding whether or not repositioning will be penalized. By ideological proximity, I mean the distance between the candidate and the citizen on policy positions ( Enelow & Hinich, 1984 ). Candidates and voters are proximally close when they both simultaneously hold a liberal or conservative position. Or, the proximate distance might be large if the candidate holds a liberal position and the voters hold a conservative position. Across many experiments, when the politician repositions proximally closer to a citizen, the politician does not incur a favorability penalty ( Hoffman & Carver, 1984 ; Carlson & Dolan, 1985 ; Sigelman & Sigelman, 1986 ; Croco & Gartner, 2014 ; Croco, 2016 ; Doherty et al., 2016 ). These results reflect the basic conclusions of theoretical models showing the importance of proximity ( Downs, 1957 ; Enelow & Hinich, 1984 ). The only experiment where proximity seemed less important included a hypothetical and newly developing crisis, and so respondents were more receptive to new positions as the situation changed ( Levendusky & Horowitz, 2012 ). However, these proximity experiments do not explore how Democrats and Republicans (or liberals and conservatives) might use campaign information differently to evaluate repositioning in-office, and how these subgroups might update their ideological evaluations of the candidates differently. The present study attempts to fill these gaps and contribute to this line of research.
3. Method and Data
3.1. Experimental Design
This section describes the experimental design. The survey experiment was conducted online with a sample of American citizens (more details in the “Data” subsection). This experiment was pre-registered at aspredicted.org. The exact language of each treatment condition can be found in Table 1. A candidate takes a position on the asylum seekers cap in the United States during a campaign or
![]()
Table 1. Randomized treatment conditions.
while in office. In-office treatment conditions have two options: holding the same position as during the campaign (consistent on asylum seekers) or repositioning after the election (changed positions from a liberal to a conservative position). This design allows for evaluations between statements made during the campaign and positions held once elected. Previous research on repositioning compared a consistent candidate already in-office with an inconsistent candidate in-office. Note that these treatments do not use partisan labels for the candidates. Dias and Lelkes (2021) argue that party and policy are inseparable for well-known issues like immigration, and therefore, party and policy should not be included in a treatment together. My research design heeds this recommendation. Goggin, Henderson, and Theodoridis (2020) tested this notion and showed that respondents can guess a candidate’s party from their policy positions, and so my experiment shows policy positions of the candidate without party. See the online appendix at the author’s website for the full questionnaire, balance tests, and a screenshot of the vignette from the respondents’ perspective.
3.2. Outcome Measures and Hypotheses
Respondents answered three outcome measures that tap into favorability and ideology. Favorability was asked as a feeling thermometer from 0-100 using a slider. Respondents were also asked to evaluate the candidate’s ideology on a five-point scale that included a “not sure” option. In the analysis to follow, ideology was recorded from 0-100 to match favorability and excludes “not sure”. Below are my pre-registered hypotheses. I hypothesized that the consistent candidate would be more favorable than a candidate who simply held the position in a campaign. Second, I hypothesized that the repositioned candidate would be less favorable than the campaign candidate and less favorable than the consistent candidate. These hypotheses are informed by theoretical expectations and evidence in the first and second wave of repositioning experiments—that a penalty exists for repositioning when no other information is presented. The bulk of the results section evaluations preregistered analyze by respondent’s position on asylum seeker and their party identification (which were asked pretreatment). Although these analyzes were preregistered, I made no specific hypotheses about the subgroups. To measure proximity, respondents were explained what an asylum seeker is, the current cap level, and then were asked if they prefer to raise the cap (liberal respondents), maintain the current level, or lower the cap (conservative respondents). After that question and several other demographics, respondents were randomized into a treatment condition. The survey also included an attention check to ensure that respondents were reading the questions.
3.3. Data
Data come from Lucid, an online survey sampling company. The total sample size in this study is 1338 and each randomized condition contains over 400 respondents. This sample size allows for detecting differences by subgroups like proximity and party identification. The survey experiment was programmed in Qualtrics, and Lucid recruited respondents and then directed them to the Qualtrics survey. The survey took no more than 15 minutes for all respondents. To provide some background on the sampling firm, Lucid is one of the largest online sampling marketplaces ( Coppock & McClellen, 2019 ). Samples are created using demographic quotas of potential respondents who are invited to take the survey, which are used to mimic representative surveys. In a study comparing Lucid samples to Amazon MTurk—a common sample in political science—Lucid came closer to the American National Election Studies’ distributions on demographics, voter registration, turnout, and party identification ( Coppock & McClellen, 2019 ). These results suggest that Lucid samples are useful for behavioral research.
4. Results
Results will be presented graphically with 95 percent confidence intervals. The analysis includes three major sections. First, I will show treatment effects using ordinary least squares regression with covariate adjusted controls for household income, education, race, and gender. Second, these treatment effects are then broken down by proximity and party identification. Lastly, I show means for the campaign condition and the repositioning condition to demonstrate marginal differences by proximity and party.
4.1. Comparisons to Campaign Condition for Treatment Effects
Figure 1 displays treatment effects relative to the campaign control group. Both outcome measures are on the same scale (0 - 100), and labels for both treatment conditions are on the X-axis. Squared points are the favorability treatment effects, where positive values indicate an increase in favorability relative to the campaign. Hollowed circles are perceived ideology treatment effects, where positive values mean liberal and negative values mean conservative relative to the
![]()
Figure 1. Candidate favorability and perceived ideology relative to the Campaign.
campaign. All points include 95 percent confidence intervals. Focusing on the left side, I find a significant increase in favorability when respondents are told that the candidate was consistent on asylum seekers once in-office. Favorability increased by 4.4 percentage points (p-value = 0.03). This demonstrates a premium for being consistent beyond simply holding the position during the campaign; something that would not have been found without the campaign control group. However, I find no difference in perceived ideology compared to the campaign where the decrease was an insignificant 0.8 percentage points (p-value = 0.69). Therefore, perceived ideology does not change from being consistent, possibly because the ideological signal of the candidate did not change after the campaign. On the right side, I show significantly negative changes when the candidate repositions in-office. Not only does favorability decrease by 5.9 percentage points (p-value = 0.003), but ideology is strongly reevaluated as more conservative, with a 25.3 percentage point change in the conservative direction. Taken together, these confirm past research showing penalties for repositioning, but they also extend those results by demonstrating a benefit for consistency. In addition, these results extend past research by showing that ideology is also strongly influenced by repositioning with a treatment effect of roughly ¼ the scale. These results will now be evaluated by subgroups.
4.2. Candidate Favorability by Proximity
Figure 2 presents the same analysis except the results are stratified by the respondent’s position on asylum seekers entering the U.S. In other words, these results are broken down by the respondent’s proximity to the candidate’s position. For labelling purposes in Figure 2, “conservative” are respondents who want to lower the cap, “liberal” are those who want to increase the cap, and “maintain” are those who are satisfied with the current cap level. Focusing on the left side, I find no significant difference relative to the campaign group for any proximity subgroup. All groups are in the positive direction, but none are independently significant. This suggests that consistency is valued regardless of proximity. That consistency might not be policy-based because I find significant results in Figure 1, but I find no difference by proximity in Figure 2. Therefore, consistency might be viewed as a valence attribute that is on a different dimension than policy. This notion is explored further in Gooch (2022) .
Repositioning on the right side of Figure 2, however, shows strong and consistent patterns by proximity. Among conservatives, favorability for the candidate increased by 11.3 percentage points (p-value = 0.001) when the candidate repositioned. Those who want to maintain the current cap were unmoved by consistency or repositioning as demonstrated by the null results, which suggests they might be less interested in the issue to begin with. Among liberals, favorability drops relative to the campaign with a decrease of 36.2 percentage points (p-value < 0.001) after repositioning. The difference between conservative and liberal is stark: a 47.5 percentage point swing in favorability. Taken together, consistency might be interpreted similarly regardless of a person’s position, but repositioning effects are driven by proximity.
![]()
Figure 2. Candidate favorability relative to the campaign by position on Asylum Seekers.
4.3. Candidate Ideology by Proximity and Party
What about ideology? Figure 3 shows ideological changes compared to the campaign position by proximity. Similar to the treatment effects found in Figure 1, I find no difference in ideology when the candidate is consistent in-office. But, when the candidate repositioned in-office, all groups reevaluated the candidate as being more conservative. Those who want to maintain the cap have the smallest
treatment effect, but I think the more interesting comparisons are between conservatives and liberals. Although the difference is not significant at a 95 percent level, liberal respondents have a larger treatment effect in the conservative direction than conservative respondents.
This treatment effect analysis, however, masks important differences in the initial evaluations of the candidates. Figure 4 shows overall means for ideology (not treatment effects) during the campaign and while repositioning. Conservative respondents viewed the candidate as more liberal than liberal respondents during the campaign. These differences are significant at a 95 percent level.
![]()
Figure 3. Candidate ideology relative to the campaign by position on Asylum Seekers.
![]()
Figure 4. Means and confidence intervals for perceived ideology during the campaign compared to repositioning in-office by proximity.
Conservative respondents were also much less favorable of the candidate in the campaign condition, which makes sense because the campaign position was liberal. Therefore, even though the treatment effects are indistinguishable in Figure 3 between conservatives and liberals, their starting place is different (Figure 4). Because conservative respondents still think the candidate is liberal to moderate on average after repositioning, the candidate’s appeal to this group might be limited. In other words, conservatives might be cuing on the ideological signal of the campaign position even after repositioning. Liberal respondents, on the other hand, initially supported the candidate at high rates during the campaign, but then drastically change their support to unfavorable (Figure 2). This loss of support is accompanied by a massive change in perceived ideology of the candidate.
Taken together, repositioning on the issue of asylum seekers can lose a candidate’s proximate base (liberals) without offsetting those losses from those who are proximally close to the reposition (conservative). Support among conservatives does increase, but not as much as the amount lost among liberals. This might suggest that conservatives are more sensitive to the initially liberal campaign position, and liberals are more sensitive to the conservative repositioning. This notion is explored further by party identification.
Figure 5 and Figure 6 replicate the previous two figures on proximity but instead use party identification as the subgroup. As noted earlier, candidate party identification was not included in the experiment because a salient issue like immigration also signals party and therefore is not necessary information to include ( Dias & Lelkes, 2021 ; Goggin, Henderson, & Theodoridis, 2020 ). Figure 5 shows differences from the campaign condition by party. Again, I find no difference by party for the consistent in-office treatment. This continues to suggest
![]()
Figure 5. Candidate ideology relative to the campaign by Party Identification.
![]()
Figure 6. Means and confidence intervals for perceived ideology during the campaign compared to repositioning in-office by Party Identification.
that it is a valence trait regardless of party or proximity. On the right side of Figure 5, repositioning in-office shows differences by party. Changes in ideology are largest among Democrats, and that difference is significantly larger than changes in ideology among Republicans. This demonstrates that Democrats especially think the candidate does not adhere to the dominate ideology of their party (liberal) after repositioning. Republicans updated in the conservative direction, but the change is smaller. In other words, respondents might be cuing on different pieces of information based on their party identification. Democrats updated candidate ideology because of the conservative reposition, but Republicans might still be hesitant because of the liberal campaign position.
Figure 6 shows the means and confidence intervals for perceived ideology in the campaign compared to repositioning by party identification. This analysis is meant to show that the starting place on perceived ideology differs by party. I find a similar pattern as Figure 4. Republicans viewed the candidate as more liberal on average during the campaign and when repositioning in-office. This might be a result of the initial campaign position being liberal, and that acts an anchor to their evaluations. Even when the candidate repositions to a conservative position, Republican evaluations are 57.6 out of 100; on average, the candidate is still viewed as moderate to slightly liberal. Democrats, however, viewed the candidate as more conservative compared to Republicans, and on average, the repositioned candidate is now viewed as 29.1 (conservative on average) among Democrats. This suggests that the Democrats are evaluating more based on the repositioning to conservative and not the initially liberal campaign position. And Republicans are doing the opposite. Both proximity and party subgroups suggest asymmetric evaluations of the candidate from the campaign to repositioning in-office.
In summary, results showed a reduction in support for repositioning and an increase in support for remaining consistent on asylum seekers. However, these results were also different by party and proximity, where Democrats and liberals severely reduced support after repositioning away from a liberal policy. Republicans and conservative increased support but not as drastically. This difference by party and proximity might be driven by their ideological perceptions of the candidate, where Democrats and liberals viewed the candidate as more conservative than Republicans and conservatives.
5. Conclusion and Discussion
This paper evaluates repositioning with special attention to proximity and party identification. I showed that candidates who are consistent in-office received higher favorability rating than during a campaign. This consistency premium occurs across all respondents on average regardless of proximity or party. Candidates received lower favorability evaluations if they repositioned in-office. However, this overall treatment effect masks important differences by proximity, where those who support the reposition (conservatives) increased support for the candidate and those who supported the original position (liberals) drove the reduction in favorability. Conservatives increased their favorability of the candidate by 10 percentage points, which might seem like a lot, but that increase is roughly ¼ of the magnitude of the decrease among liberals. Liberals reacted much stronger than conservatives due to the cued ideology of the reposition. On the other hand, conservative might not support the candidate because of the liberal position held during the campaign. The campaign position might be less meaningful to liberals after knowing about the reposition.
Respondents updated their perceived ideology of the candidate in the direction of the repositioning, but again, important differences exist by proximity and party that cannot be observed with just a treatment effect. I find a similarly asymmetric pattern in perceived ideology. Liberals/Democrats reacted much more strongly to the repositioning than conservatives/Republicans. Conservatives/Republicans think the candidate is more liberal than liberals/Democrats, possibly because conservatives/Republicans were cuing off the initial campaign position (which was liberal). This demonstrates how a respondent’s proximity and party identification shape their perceived ideology of the candidate. Importantly, this also shows why a campaign control group is needed for repositioning experiments because that information is used to evaluate a candidate once in-office.
These results speak to the broader dynamics at play among representatives who switch policy positions or go against the dominant ideology of their party. Politicians with heterodox positions for their party, for example Joe Manchin or Krysten Sinema, received strong, negative reactions from liberal/Democratic voters when they break from the Democratic party on policy. My results suggest that breaking from the Democratic party will result in liberal voters updating their ideological perceptions of the candidate strongly in the conservative direction. While conservatives will still view these representatives as liberal on average, resulting in more losses among their party than gains among the out-party. Obviously, these representatives break from their party because of the constituencies within their respective states. But, if we consider how they are viewed from the perspective of an “average liberal” or “average conservative” across the U.S., then my results speak to how these representatives might be penalized for repositioning at a national-level. Moreover, Joe Biden has broken from his liberal campaign promises on immigration, and my results might suggest that he might have more to lose among his base than possible gains among conservative/Republican voters.
Acknowledgements
I thank the campaigns and information panel at Southern Political Science Association’s annual conference for helpful comments and suggestions.