Politicians around the world often switch positions on policy, and perhaps, as a result, repositioning is studied frequently in political science. In the United States, politicians switched positions on the Iraq and Afghanistan wars with varying degrees of success ( Croco, 2016 ; Croco & Gartner, 2014 ). More recently, Donald Trump disavowed disruptive cuts to the federal government espoused in “Project 2025”, but subsequently implemented these executive orders in office ( Contorno & Tolan, 2025 ). In Latin America, presidential candidates campaigned on liberal welfare policies and then switched toward conservativism once elected ( Stokes, 2012 ). In Western Europe, parties also switched positions from one election to the next across a range of issues ( Adams & Somer-Topcu, 2009 ). How might these switches at the elite level affect mass politics in an age of polarization?

Experimental results find that candidate repositioning changes three sets of outcome measures. However, past research has not analyzed how repositioning might cause affective polarization. First and perhaps most notably, candidate evaluations like favorability and vote intentions change from repositioning ( Hoffman & Carver, 1984 ; Sigelman & Sigelman, 1986 ; Croco, 2016 ; Croco & Gartner, 2014 ; Doherty, Dowling, & Miller, 2016 ; Sorek, Haglin, & Geva, 2018 ; Gooch, 2022 ). Repositioning also changes the perceived ideology of candidates in the ideological direction of the new position ( Doherty et al., 2016 ; Gooch, 2023 ). Repositioning also changes a variety of valence evaluations, such as competency, trustworthiness, sincerity, and whether or not an individual is considered morally upstanding ( Gooch, 2022 ; Carlson & Dolan, 1985 ; Allgeier, Byrne, Brooks, & Revnes, 1979 ). With regard to valence, these findings suggest that repositioning alters potentially emotionally charged character evaluations and not just spatial model factors related to voting and ideology. The present study tests whether repositioning also causes affective polarization. Affective polarization is typically characterized as an emotional response to politics and, in particular, negative feelings toward the out-party. This might be especially possible when candidates reposition in a polarized direction (e.g., when a conservative candidate repositions to a more conservative position or when a liberal candidate switches to a more liberal position).

Affective polarization is typically measured using surveys. One of the most well-known measures is the feeling thermometer of the parties, broken out by out-party vs. in-party ( Iyengar, Sood, & Lelkes, 2012 ). This measure shows a growing divide among Americans where partisan affiliation influences views of self and (increasingly negative) views of the out-party. These feelings are not contained to the mass public—the activist class in American politics also demonstrates a high degree of affective polarization ( Ladd, 2018 ). Some argue that these partisan-based worldviews extend to other identities, such as ideology, friendship groups, or the type of person that someone might prefer to marry ( Iyengar, Sood, & Lelkes, 2012 ; Mason, 2018 ). Indeed, a growing divide between Democrats and Republicans can be observed with questions about the willingness to engage with the out-party in social situations. Compared to the pre-polarized era, partisans are more likely to say that they do not want to be friends with, live next to, discuss politics with, and want their kids to marry the out-party ( Enochs, 2017 ). The present study uses these canonical affective polarization questions to evaluate if repositioning is also contributing to mass-level polarization.

My hypotheses build on evidence that affective polarization is rooted in attitudes toward the elite level ( Druckman & Levendusky, 2019 ). Also, see Iyengar, Lelkes, Levendusky, Malhotra, and Westwood (2019) for a thorough review of affective polarization). Previous research has not investigated how repositioning might influence affective or social polarization. Using affective polarization survey questions including feeling thermometers, trust, and three social polarization questions, Druckman and Levendusky (2019) showed that randomized conditions that included party labels about elites increased polarization compared to a treatment condition that included labels about voters. Therefore, affective polarization might be elite-driven even though it influences views of mass-level situations like being friends or neighbors with the out-party. Repositioning is top-down information about where a candidate stands on issues, and so this study extends this notion of elite behavior driving affective polarization to a very specific type of elite-level position-taking.

My hypotheses are as stated: 1) the importance of one’s own party will increase for every repositioning group compared to the control group, 2) repositioning will cause more negative views of partisan politicians and partisan people compared to the control group, and 3) social polarization will increase from every reposition group compared to the control group. Results show that candidate repositioning only weakly influenced affective polarization. Most repositioning treatments do not cause changes to affective polarization, with two exceptions. First, a liberal candidate who repositioned to the left (polarized reposition) caused Democratic and Republican respondents to view their own party as a more important part of how they see themselves as people. Second, I found that repositioning by liberal candidates caused Democratic respondents to rate Democratic politicians lower on a feeling thermometer scale. Otherwise, I found very little connection between repositioning and feeling thermometer scales about partisan politicians and people. Lastly, I found no connection between repositioning and social polarization questions. In the online appendix, I used a variety of alternative model specifications to test the robustness of these results, but the substantive takeaway did not change. Very little connection exists between repositioning and affective polarization. These results suggest that penalties for repositioning are concentrated among the actual candidates who switch positions. These effects did not spill over to views of partisan politicians and people or social interactions. This study proceeds as follows: I discuss the data and methods, detail the results, consider alternative explanations, and then finish with a conclusion and discussion.

This section details results for 1) the importance of one’s own party, 2) feeling thermometer toward partisan people, 3) feeling thermometer for partisan politicians, and 4) four social polarization measures.

Notes: Standard errors in brackets. Controls for gender, party ID, education, income, race, and self-place position on asylum seekers. ***p < 0.01, **p < 0.05, *p < 0.1.

Table 3 starts with the importance of party, and these results show one of the few instances where repositioning caused affective polarization in a coherent way. I found null results for every treatment group except for liberal candidates who repositioned in the polarized direction. The importance of one’s own party increased by 4.3 percentage point scale and is significant at a 95 percent level. In the appendix, I stratified by party and showed that both Democrats and Republicans significantly increased their party importance when a liberal candidate repositioned to the left. Moreover, a conservative candidate repositioning to a liberal position caused Democrats to reduce their party importance by 4.7 percentage point scale. A logical expectation is that repositioning to the poles would most likely increase affective polarization, and so this result suggests a connection. But, as the results will show in totality, the connection between repositioning and affective polarization is limited.

Notes: Standard errors in brackets. Controls for gender, party ID, education, income, race, and self-place position on asylum seekers. ***p < 0.01, **p < 0.05, *p < 0.1.

Table 4 shows feeling thermometers toward Republican and Democratic politicians. Results show null results across all questions for every treatment group except for when a liberal candidate repositions left (again, a polarized reposition). Feelings toward Democrats increased by 3.3 percentage point scale but are only significant at a 90% level. These null results are particularly interesting because they suggest that penalties for repositioning are concentrated with the actual politician who repositions and not partisan politicians in general. However, I do find significant results when broken down by party. Specifically, an in-party effect from repositioning. When a Democratic candidate repositions, views of Democratic politicians become significantly worse among Democratic respondents only. In the appendix, I find a similar effect among Republican respondents when a conservative candidate repositions to the left—a reduction of 8.0 percentage point scale (p-value < 0.01) in feelings toward Republican politicians. This suggests some backlash against respondents’ own party caused by repositioning away from their party’s dominant position. Lastly, this analysis was also rerun in the appendix, where the dependent variable is recorded so that respondents are evaluating their in-party vs. out-party, but those results do not differ from the main findings.

In Table 5 , I find null and contradictory results for the views of partisan people. Feelings toward Democratic people decreased by 4.2 percentage point scale (significant at a 95 percent level) when a conservative candidate repositions to the right (polarized reposition). I also find decreased feelings toward Republican people ranging from 2.5 percentage point scale to 5.0 percentage point scale when a conservative candidate is in office (fourth column). Although a handful of the changes are in the direction of increased polarization, the results are contradictory and do not point to a clear connection between repositioning and feelings toward politicians and people. For example, it is not obvious why feelings toward Republican people would go down by 5.0 percentage point scale when a conservative candidate is consistent in office. They seem to imply that simply being exposed to a conservative candidate in office reduced feelings toward Republican people.

Notes: Standard errors in brackets. Controls for gender, party ID, education, income, race, and self-place position on asylum seekers. ***p < 0.01, **p < 0.05, *p < 0.1.

Notes: Standard errors in brackets. Controls for gender, party ID, education, income, race, and self-place position on asylum seekers. ***p < 0.01, **p < 0.05, *p < 0.1.

Notes: Standard errors in brackets. Controls for gender, party ID, education, income, race, and self-place position on asylum seekers. ***p < 0.01, **p < 0.05, *p < 0.1.

Table 6 and Table 7 display the results for the social polarization question for liberal and conservative candidates, respectively. Likewise, the results are mostly null and do not point to a clear connection. However, I do find that when the conservative candidate repositions to the right (polarized reposition), respondents are less likely to say that they are willing to talk to the out-party (4.3 percentage point reduction significant at a 95 percent level). In the appendix, I find that Republicans are less likely to talk to the out party when a conservative candidate repositions to the left (reduction of 5.75 percentage point scale, p-value < 0.10) and the status quo (reduction of 5.77 percentage point scale, p-value < 0.10). But aside from those limited effects, the results are null.

Taken together, a candidate moving toward the poles had the greatest likelihood of increasing polarization. However, I did not find this to be the case in every polarized repositioning treatment, so the results are inconsistent even in the conditions where an effect would most likely be. This suggests caution when attributing flip-flopping to increased affective polarization.

Before concluding that no connection exists between repositioning and affective polarization, I consider alternative explanations that might have generated these null results. In the end, I believe this discussion bolsters my conclusion of a weak connection between repositioning and polarization. Using an excellent guide for explaining null results in survey experiments ( Kane, 2024 ), seven factors are worth discussing. Null results could exist because respondents are inattentive and did not actually read the treatments ( Berinsky, Margolis, & Sances, 2014 ). To counteract this problem, my experiment used a screener question with only one right answer. Respondents had to correctly answer a SAT-style question that stated: “Puppy is to dog, as kitten is to…” with the correct answer being a cat. Respondents were removed from the survey before randomization if they answered incorrectly. Although it is not the most sophisticated, the question was able to remove 56 inattentive respondents before the treatment assignment.

A second alternative explanation for null results is failing to vary the independent variable in a way that signals the construct to respondents ( Kane, 2024 ). The point of repositioning treatments is to signal a change in ideological direction, and so perceived candidate ideology works as an effective manipulation check. In this case, I do believe the independent variable varied correctly because repositioning treatments alter the perceived ideology of candidates ( Gooch, 2023 ). For example, when a candidate is repositioned in the conservative direction, they are perceived as more conservative by respondents relative to a control group. This is true for both Democratic and Republican respondents.

Another possible explanation for the null results is pre-treating respondents. This occurs when respondents receive information about the treatment condition from exposure to politics before taking the survey ( Kane, 2024 ). An example might be treatment conditions that use well-known candidates like sitting presidents. In this case, most respondents already have a large store of considerations about a sitting president, making any additional effect from a treatment unlikely. I think this is a possibility because I used a salient issue (immigration) where Democratic candidates in particular have felt pressure to reposition toward conservative due to increased media attention paid to border security. Moreover, politicians reposition frequently throughout their careers, and it typically generates news coverage. On the other hand, I used a hypothetical candidate without party labels, which is meant to remove pre-treatment bias. Although I am not overly concerned with pre-treatment bias, I do think it is possible that changes in polarization caused by repositioning might have already occurred in the real world.

Another possibility is insufficient statistical power. On the one hand, social science quantitative research has a history of being underpowered ( Esarey, 2017 ), and that might apply to this survey experiment as well. On the other hand, I think it is worth noting that significant results occur frequently with this specific experimental design and repositioning treatments in general. Repositioning routinely changed favorability, ideology, and valence in survey experiments. If it was purely an insufficient amount of data, I would find null results across a range of outcome measures and not just affective polarization. The effect sizes for favorability and valence are roughly the same size as significant results from these affective polarization questions. In the tables presented above, significant results required a four-percentage point change or more, which is not unreasonably high for a survey experiment. However, to detect smaller effects on polarization, the sample size in this study may be limiting.

Yet another possibility is a poor measurement of the dependent variable, and a remedy is to create indexes for multi-question dependent variables ( Kane, 2024 ). In the online appendix, I combined the four social polarization questions into an index but found consistently null results. I then stratified the index models by partisan only, strong partisan only, and non-Hispanic whites only. But again, I find null results for the index. Relatedly, I also recorded the data as in-party vs. out-party, but the results did not differ in substantial ways from those in the main results section. Lastly, I do not think ceiling or floor effects drive the results because I used several 0-100 thermometers, where most averages are toward the center. For example, the social polarization index has a mean of 45.3 out of 100. Also, see constants in the regression tables above, which are all in the middle range of the scale. After considering these seven alternative explanations, I conclude that the null results found in this study are indeed null.

I find very little evidence that candidate repositioning causes affective polarization. The only consistent pattern was several polarized repositioning treatments (A liberal moving left and a conservative moving right). But this did not occur in every polarized repositioning treatment. When I found significant changes across the larger spectrum of repositioning conditions, the changes to affective polarization were incoherent or contradictory. For example, sometimes consistent and repositioning treatments changed polarization in similar ways (e.g., they both increased or decreased polarization instead of going in the opposite direction). Still, most treatment conditions showed no significant difference relative to a control group. Therefore, the connection between candidate repositioning and affective polarization is weak at best and mostly contained in treatment conditions where a candidate moves toward the ideological poles.

However, these findings are credible because of the experimental design and the steps taken to minimize alternative explanations. These results are made possible by a comprehensive research design that accounts for all types of repositioning from both liberal and conservative candidates. The study also used a control group so that in-office conditions were compared to campaign conditions. Elsewhere, repositioning asylum seekers is shown to influence favorability, ideology, and valence ( Gooch, 2022 ; Gooch, 2023 ). These dynamic effects on candidate perceptions do not extend to commonly used measures of affective polarization.

Not every aspect of American politics connects to affective polarization. Affective polarization seems to influence the public, and credible research exists that confirms this connection. However, the entirety of political behavior and how voters view candidates are not purely driven by affective polarization. For example, experimental evidence exists that affective polarization does not impede views of democracy and accountability when centered on members of Congress ( Broockman, Kalla, & Westwood, 2023 ). Moreover, an experiment that randomized national security threats from China—which might ostensibly create a “rally around the flag” effect for Democrats and Republicans—failed to reduce affective polarization ( Yeung & Xi, 2025 ). There is perhaps an urge to view all of American politics through group attachment and emotional responses, but the evidence here (and elsewhere in the policy representation literature) shows that other factors are still important to the voting calculus. Policy positioning still matters in ways originally conceived through the Downsian model of candidate competition that emphasizes ideology and valence. Taken together, this study suggests caution in connecting “flip-flopping” with polarization.

The author thanks participants at MPSA and APSA in 2024 for helpful comments, particularly for suggesting alternative model specifications. The author also thanks Rowan’s College of Humanities and Social Sciences for funding the survey experiment.