Sage Journals: Discover world-class research

Abstract

Do election outcomes exacerbate affective polarization? While polarization often rises during campaigns and correlates with democratic backsliding, isolating the effect of winning or losing has proven difficult because of the need for a pre-election baseline and to generalize across multiple elections. In this study, we leverage pre- and post-election questions about partisan affect in the American National Election Study between 1996 and 2024. Our first analysis studies how respondents’ attitudes changed based on their party’s success in its bid for the White House. Our second analysis extends this to hundreds more races, applying a regression discontinuity design to attitudes after close subnational election results. Both analyses support the conclusion that the losing side drives the post-election gap in polarization, and that they do so by feeling less warmly toward their own party. In the United States, political loss may erode in-group attachment more than it fuels out-group hostility.

Keywords

affective polarization partisanship winner-loser gap regression discontinuity

Affective polarization, or the relative dislike of opposing partisans (Iyengar et al. 2012), tends to rise during election campaigns (Martin and Nai 2024). It has also been tied to support for violating democratic norms (Hernández et al. 2021; Kingzette et al. 2021). These twin findings suggest a conundrum for democracy. Does the fundamental mechanism of democracy—competitive elections—also pose a risk to its stability?

In recent years, multiple studies have analyzed trends in polarization among an election’s eventual winners and losers. To date, however, no cohesive picture has emerged. In Canada and Israel, single-case studies suggest that the winning side depolarizes (Gidron and Sheffer 2024; Sheffer 2020), while an analysis of 2020 and 2022 returns in the United States showed no effects (Fasching et al. 2024). A recent cross-national study did not account for pre-election levels of polarization, but noted that almost universally, winning sides tend to be more polarized after votes are counted (Andrews and Huang, 2024).

Studying the outcome-to-polarization link is challenging because of potential endogeneities. People tend to be more polarized when surrounded by co-partisans (Connors 2023; Warner 2023), and their perceptions of this balance likely crystallize as the election draws near (Singh and Thornton 2019). Polarization also has a mobilizing effect on behavior (Phillips 2024), making it plausible that the more polarized side is likelier to win to begin with.

In this study, we take advantage of similar questions that were asked about partisan affect in both the pre- and post-election waves of the American National Election Study between 1996 and 2020. By accounting for respondents’ pre-election levels of polarization, in-party love, and out-party hate, we can observe how these values may have shifted in the wake of victory or defeat. Importantly, we make these observations in two distinct analyses—both of which point toward largely the same conclusion. The first analysis uses respondent fixed effects in a panel model to directly compare post-election attitudes to a pre-election baseline, conditional on their party having won or lost its bid for the presidency. The second analysis applies a regression discontinuity design (RDD) to these data, observing partisan attitudes in the wake of close wins and losses for their party in 227 US House and 47 Senate races, and in 86 state-level results for president. Both analyses indicate that, in the US context, outcomes associate with a winner-loser gap. Contrary to past work, however, we find this effect is driven primarily by losers depolarizing. After the election dust settles, the losing side appears to lose faith in its own group more so than they feel additional animosity toward the out-group.

In the sections that follow, we describe how there are many plausible explanations for how outcomes may affect polarization. We then review prior work, discuss how our approach differs, and justify our focus on the US context. From there, we lay out our data and analytic strategy before presenting the results. While losing sides may indeed show less satisfaction with and weaker allegiance to democracy (Blais and Gélineau 2007; Kern et al. 2024), our results suggest that this does not happen because they are embittered at the opposition.

How May Election Outcomes Impact Affective Polarization?

Affective polarization is a combination of group-based identity and affect (Iyengar et al. 2012). For someone to be “affectively polarized,” they must identify with one group and have opinions about another, referred to as the out-group. Their level of polarization refers to the extent that they like their own group relative to the out-group.

In the context of US politics, affective polarization is studied among members of the two major parties, the Democrats and Republicans. Elections can impact polarization between them because they clarify group status, a key component of how intergroup relations work (Huddy 2004). Pre-election, there exists uncertainty about which group will be handed the reins of power; the results dispel that uncertainty. They may not be unequivocal, of course—they may produce a divided government or different regional and national results—but on average, voters will emerge with a stronger sense of where their party stands.

Social psychologists have long studied how people react when their groups rise or drop in status (Brewer and Kramer 1986), and this provides hypotheses for how partisans may respond to election outcomes. However, these expectations do not all point in the same direction, especially because election results may affect either of polarization’s components: in-party love and out-party hate. A brief review of the literature sets a clear expectation that winners grow in in-party love, and losers in out-party hate, but how the winning side will see their opponents—and how the losing side will see themselves—is less theoretically clear.

On the winning side, the “basking in reflected glory” effect leads us to believe that winners will grow in in-party love after the election. Cialdini et al. (1976) showed that university students have a higher probability of wearing school apparel and using “we” terms after their football team wins. Miller (2009) expands this to the political realm, noting that supporters of Barack Obama kept their yard signs up for longer than John McCain supporters did after the 2008 election, and Andrews and Huang (2024) show in cross-national contexts that winners exhibit more in-party love than does the losing side. However, it is less clear what an election’s victors will think of their erstwhile opponents. On one hand, a successful campaign may remove feelings of group threat, which heavily motivate out-group hatred (Riek et al. 2006). Oc et al. (2018) observed that prior to the 2016 election, Republicans exhibited high degrees of in-party bias in a dictator game, but this dissipated in a post-election follow up study. A separate strand of research, however, indicates that being in an advantaged position can increase out-group hostility. Social dominance theory holds that high-status groups will favor discriminatory ideologies as a means of keeping the out-group in their place (Sidanius and Pratto 1999), and this appears especially true when the privileged status seems precarious, as it might after a close election. Cunningham and Platow (2007) observe that high-status groups demonstrate greater animosity when the underlying power structure is unstable, and Hornsey et al. (2003) likewise show greater animosity among the high-status group when their privilege comes amid controversy.

On the losing side, there is a stronger expectation about how feelings of hostility may increase in the wake of defeat. People naturally feel anger after experiencing relative deprivation (Guimond and Dambrun 2002; Smith et al. 2012), and hostility can also serve as a coping mechanism. Wann et al. (2005), for example, show that after losing a match, fans of a sports team are more likely to endorse anonymous aggression toward their opponents than before the match or after a victory. What is less clear is how partisans on the losing side may feel about their own group. According to Cialdini et al. (1976), “cutting off reflected failure” effects are the flip side of people’s desire to associate with successful groups, and average levels of in-group pride appear lower among low-status groups (Jost and Burgess 2000). But defeat can also induce cohesion when members feel they must band together to guard against threat, or to succeed at the next opportunity (Rubin et al. 2014; Turner et al. 1984).

Taking these insights, and applying them to the present study, it seems that winners may emerge as more polarized than losers after an election if their out-party hate increases or if the losers’ in-party love decreases. By contrast, losers may emerge more polarized than winners if the winners soften on their out-party hate or if the losing side sustains its levels of in-party affect. In previous studies about election outcomes and affective polarization, most of the movement has been described as on the winners’ side. Sheffer (2020) and Gidron and Sheffer (2024) use panel data to show that, relative to before the 2015 Canadian and 2021 Israeli elections, respectively, the winning side emerged as less polarized than the losing side because their attitudes toward the out-party softened. Looking more broadly across advanced democracies, Andrews and Huang (2024) diverge by showing that winning sides tend to be more polarized than losers after the election is over, although they do not control for pre-election levels. In the United States context, findings have varied as well. Fasching et al. (2024) study partisan affect during the 2020 and 2022 election cycles, and find no significant change between pre- and post-election levels both for Democrats nationally or for the winning side in down-ballot races. However, in an experimental study of Americans in 2023 and 2024, winners of a simulated election showed higher polarization than prior to it as a result of greater in-party love (Hamrak 2025).

These studies, and their divergent results, highlight two of the challenges to studying how outcomes impact polarization. First, it is hard to hold other factors constant when studying the impact of an election result. Affective polarization has ties to social context and political mobilization (Connors 2023; Phillips 2024), which are endogenous to election outcomes, and partisan groups likely begin the election cycle at different, preexisting levels of partisan affect. Second, it is quite possible that the effects vary across democracies. Garzia et al. (2023) show that trends in polarization have varied widely across countries over the last 30 years, and polarization itself can be operationalized differently depending on the structure of a country’s partisan politics (Tichelbaecker et al. 2023).

These challenges highlight the need for country-specific analyses that estimate the effect of winning or losing relative to a pre-election baseline. We attempt this in the United States context by leveraging the panelized (pre- and post-election) nature of the quadrennial American National Election Study, which asked questions about partisan affect both before and after presidential elections in 1996, and from 2004 to 2024. We focus on the US case because it is the largest advanced democracy, and citizens and leaders alike cite polarization as a top national concern (Pew Research Center 2023; Sharpe 2023). Applying a model with fixed effects at the individual level, the ANES data allow us to observe how people’s partisan attitudes shift after their party wins or loses the presidency. This serves as our paper’s first analysis.

However, we also take our analysis a step further. Recognizing the small number of presidential races in our 24-year series, and the way that reactions to salient events like these may be context dependent, we reevaluate our presidential-level findings with a regression discontinuity analysis of subnational races. Regression discontinuity designs (RDDs) compare the effect of events near a cutoff—here, races that end in near-victory or defeat for a party—on outcomes like citizen attitudes. The logic is that when results fall within a narrow bandwidth, units on either side of the cutoff are likely very similar. Necessarily, we expect weaker, less consistent results from an analysis like this, given the lower salience of races for US House and Senate. However, by including hundreds of elections, this analysis improves the generalizability of our findings, and its focus on close cases makes it more likely that the effect we observe is causal. The next sections describe our research designs and findings in more detail.

Polarization in the Wake of Presidential Elections

As mentioned above, our analysis relies on the American National Election Study cumulative file. In the surveys for the 1996 and 2004–2024 presidential elections, respondents filled out 0–100 pt feeling thermometer ratings for the Democratic and Republican parties before the election, and 0–10 pt dislike-like scales for the parties after the election.¹ After classifying respondents (including leaners) into parties, we transformed these metrics into in-party warmth, out-party warmth, and affective polarization (out-party warmth subtracted from in-party warmth). All partisan affect variables are scaled from 0 to 1 for ease of interpretation. All replication data and code can be found on Harvard Dataverse (https://doi.org/10.7910/DVN/XGS79K).

The first part of our analysis studies how partisan affect changes in the wake of presidential elections. Recall that each year of the ANES is fielded in a panelized format, with respondents interviewed once before the election, and again afterward. In order to make before-and-after comparisons, we first structured the data into a “long” format, so that pre- and post-election responses are treated as separate observations, linked by a unique respondent number. Formatting the data in this way allowed us to specify panel models in the following manner:

{Attitude}_{i t} = β_{0} + β_{1} Election {Status}_{i t} + α_{i} + ε_{i t}

We use the equation above to explain, for person i at timepoint t, three separate dependent variables that are represented by the Attitude term. One is in-party affect, the second out-party affect, and the third affective polarization (their difference). The key independent variable is Election Status. This is a three-category variable, with the baseline set to the pre-election condition, and the other categories indicating for post-election observations whether the respondent’s party won or lost the presidency. In this way, for example, the coefficient on “post-election/won” will indicate how much the dependent variable changed relative to its pre-election value for respondents whose party won. The $α_{i}$ term in the equation refers to the individual-level fixed effect. This limits our analysis to studying within-unit variation and holds respondents’ time-invariant characteristics constant (such as demographics or attitudes that did not change between waves).

Individual fixed effects do not, however, control for variables that may have changed between waves. Our ability to account for these is limited because the ANES asks very few repeat questions in its re-interview. When we considered our lack of time-variant controls, we determined that it primarily limits our ability to distinguish between the direct and indirect (or mediated) effects of winning or losing. For example, if losing an election reduces how much someone identifies with a party, and this in turn reduces in-party love, we would not be able to observe this full causal chain. However, we do not feel this limitation keeps us from answering our research question, as our primary interest is in whether differences emerge between the winning and losing sides, more so than how.²

The results of our models are presented in Table 1. In the leftmost column, the variables indicate the different election statuses that may impact partisan affect. Pre-election values serve as the baseline, and the model is constrained by its individual fixed effects. The next column indicates post-election changes in affective polarization among winners and losers, with the losing side seeing a significant, 3.7-point drop in polarization (on a 100-point scale) compared to no significant change among the winners. The columns that follow break down the component parts of affective polarization. The middle column studies how winners and losers change differently in terms of in-party love. While both sides’ enthusiasm for their own party drops in the weeks after an election, this is especially pronounced among the losing side, whose like/dislike ratings drop off by about 3 points more than that of the winners. Both sides grow colder toward the out-party, as well, but this change is smaller and indistinguishable between the two groups.

Table 1.

Changes in partisan affect by win/loss status after presidential elections (1996–2024)

Variable	Affective polarization	In-party affect	Out- party affect
Pre-election	Ref.	Ref.	Ref.
Post-election: Lost	−0.037*** (0.003)	−0.051*** (0.003)	−0.014*** (0.002)
Post-election: Won	−0.003 (0.003)	−0.019*** (0.003)	−0.016*** (0.002)
R-squared	0.797	0.682	0.774
Observations	n = 27,349/T = 2

Note. ***p < .001. Cells contain OLS coefficients, cluster-robust SEs in parentheses. Dependent variables scaled to 0 to 1. Data are from ANES Time Series Cumulative file with preliminary 2024 data added manually. The 2000 wave is excluded as it did not include post-election affect questions. Each respondent is included as two separate observations, one each for their pre- and post-election attitudes, and the model is specified with respondent-specific fixed effects.

To observe how consistent this pattern was across the elections from 1996 to 2024, we respecified the model for each year individually and plotted the effect of each post-election condition by year and outcome in Figure 1. Although statistical significance varies by year, all seven elections show patterns that are compatible with the overall results. After each presidential race, the losing side emerges less polarized than the winners. As above, this appears primarily driven by differences in in-party love. In all seven elections, the losing side emerged with lower in-party love than the winners—this was statistically significant in five of the seven races—and with lower in-party than out-party affect relative to their own pre-election baseline. Additionally, every year but 2024, the losing side emerged with lower in-party love than it entered with.

Figure 1.

Estimated changes by year and win/loss status in presidential elections.

The main takeaway from this first analysis is that, in the wake of modern-era US presidential elections, it appears that the losing side shows lower polarization relative to winners due to lower in-party love. Before concluding, however, we pause to consider potential question-wording effects on the results. Recall that respondents used a 0–100 feeling thermometer scale in the pre-election surveys, and a slightly different, 0–10 like/dislike scale in the post-election surveys. If respondents systematically provide lower ratings on the 0–10 scale, then this may artifactually produce the lower post-election estimates we see in Table 1. Moreover, the more limited range of response options to the like/dislike question may create scale-related compression or expansion effects.

Appendix A explores this in three ways. First, we compare the shift between scales to other survey items on partisan attitudes—in the ANES and the VOTER Panel—that also were asked before and after Election Day but that used the 100-point scale in both. We find evidence that people tend to assign more polarized ratings on the 0–10 scale, indicating that if anything, our findings are potentially understated by the different pre- and post-election question wording. Second, we identify an instance on the post-election 2020 ANES when respondents were asked to rate the presidential candidates on both the 100-point feeling thermometer and the 0–10 scale.³ This allowed us to create both linear and distributional mappings from the thermometer onto the like/dislike scale (de Jonge et al. 2014), which we then applied to the partisan affect questions. When the models in Table 1 are reproduced using these harmonized measures, their findings grow more pronounced still. Third, as a naïve and potentially more conservative approach, we simply standardized our partisan affect measures by wave to create z-scores that assume constant levels for each measure over time. There, we find complementary results: winners emerge as more polarized than losers, this is driven by changes in in-party love. However, there is one nuance, in that we see greater pride among the winners as well as a drop among the losers when using the z-scores. While we favor the interpretation from Table 1 and the other robustness checks—that the greatest action occurs in the losing party’s self-image—readers may wish to keep an open mind to the possibility that in-party love increases on the winning side as well.

Finally, although our ability to directly test mechanisms is limited, it is worth considering how election results affect post-election polarization. Hamrak’s (2025) experimental study suggests that the instantaneous effect may be primarily to increase the winning party’s pride, but here, with post-election responses collected an average of 28 days after the event, we see a drop among the losing side’s in-party affect and potentially a reversion to pre-election levels among the winners. We attempt to bridge these findings in Appendix B, by interacting the election wave variable in Table 1 with the number of days between the interview date and Election Day. In line with experimental findings, we see higher in-group love on both sides immediately after the election, but this abates as time passes, dropping more quickly for the losing side. Some part of this may be the declining salience of elections. Cross-national research, for instance, shows polarization-related attitudes to diminish as elections fade into memory (Hernandez et al. 2021). But future research may wish to study why this drop is so much greater among the losing side; possible causes may be “post-mortem” coverage of the party’s failures (e.g., Meyer and Gatterman 2025), or psychological differences in how people interpret defeat in its immediate and longer-run aftermath (Lalonde 1992).

Re-evaluating Our Findings Using Close Subnational Elections

Our first analysis demonstrates that, in the wake of modern elections for US president, the losing side’s supporters emerge as less polarized than before due to a drop in in-party love. The question of generalizability remains, however. Is this pattern intrinsic to US elections in the current era, or is it something artifactual to the seven presidential races we observed?

To study this more closely, and with more of a claim to causal identification, we apply a regression discontinuity design (RDD) to close results in US House, Senate, and state-level presidential races. RDDs are a quasi-experimental approach. Their logic is that some “treatments” depend on a cutoff—such as winning more votes than the other party—and that cases very close to it are comparable. While this design has been used extensively in political science (Hyytinen et al. 2018), it is not a silver bullet for causal inference. Cases close to a cutoff may not be generalizable to the larger population (Wing and Bello-Gomez 2018), and in elections, actors may vary systematically in their ability to place themselves above or below the cutoff (Eggers et al. 2015). Yet observations on either side do not need to be identical in every regard, so long as there is no nonlinear bunching of pre-treatment variables that may affect the analysis’s dependent variable (De la Cuesta and Imai 2016). Moreover, the fact that subnational elections are nested in a broader, national context, and that we can incorporate pre-election attitudes—something few RDDs draw on—leads us to see this design as a strong supplement to the associative analysis presented above.

Data and Bandwidth Specifications

As in the previous analysis, we draw our data from the ANES Cumulative File. Pre- and post-election measures of partisan affect are available for all presidential-year waves from 1996 to 2020 except for 2000. From the ANES file, we use information on respondents’ congressional district and state to merge House, Senate, and state-level presidential election returns compiled by the MIT Election Data and Science Lab (2022a, 2022b; 2023).⁴

The first step toward specifying an RDD is to determine the cutoffs that will be used. To reduce researcher degrees of freedom, analysts are encouraged to use formulaically determined bandwidths that optimize some component of the model fit (Brodeur et al. 2020). We employ Calonico et al.’s (2015) optimal bandwidth procedure, which selects bandwidths that minimize the mean squared error of the model, and weighs observations close to the cutpoint more heavily to reduce bias. However, this results in the inclusion of some races that fell several points away from a 50–50 result. For example, the optimal bandwidth for estimating the effect of outcomes on affective polarization in House elections is 10.2 percent, which would include cases as lopsided as a 55 to 45 percent margin. Even though observations far away from the cutoff are common in election RDDs (e.g., Hainmueller et al. 2015), and are down-weighted compared to observations closer to the cutoff, the concern remains that findings are driven by elections with wider margins that are less likely to be comparable (Eggers et al. 2015). We therefore also set more conservative thresholds at smaller, round-number intervals of 1–5 percent (Cattaneo et al. 2020). Table 2 presents the number of races and ANES sample size drawn from each type of race at the different bandwidth specifications.

Table 2.

Number of races and ANES observations by RDD bandwidth

		1 percent	2 percent	3 percent	4 percent	5 percent	Optimal	All (Inc. excluded)
US house	Elections	54	104	139	191	227	600	1,782
US house	Respondents	590	1,171	1,667	2,226	2,659	7,027	20,449
US senate	Elections	11	20	32	44	47	88	167
US senate	Respondents	1,122	1,445	2,275	3,212	4,199	7,927	13,085
State-level presidential	Elections	20	37	53	68	86	147	258
State-level presidential	Respondents	2,431	4,318	5,852	6,736	8,276	12,749	20,684

Robustness Checks

Before running the 18 models implied by the bandwidths and election types above, we first compared the “close win” and “close loss” groups in each of the samples. Specifically, for each of the 18 models, we checked whether the RDD as we set it up could predict any pre-election differences between the winning and losing side on the following variables: strength of partisan identity, news interest, ideological extremity, age, gender, race, education, urbanity, income, region, and religiosity. This is especially important to test because although the ANES is nationally representative, it may not be for the subnational unit in which a close race occurred, so we need to make sure that any differences cancel out when the treatment effects are aggregated across units.

Appendix C presents the results of these balance tests. The optimal bandwidths see the most frequent differences, but this is perhaps to be expected when comparing winners and losers in races won by several points. Importantly, the magnitude of these differences on the variables most tied to polarization—strength of partisanship, news interest, and extremism—is very small, less than 1/10^th of a point on three- and four-point scales. Meanwhile, the more conservative bandwidths rarely see any differences, and those that do emerge are not consistent from one bandwidth to another.

Beyond these demographic and attitudinal balance tests, research also shows that close wins sometimes occur in districts where the incumbent party recently over- or under-performed (Eggers et al. 2015; Hall and Thompson 2025). To test this possibility, we estimated the effect of an eventual Democratic victory on prior Democratic vote share, holding constant later margin of victory and year. In line with prior work, we found close Democratic winners to perform better in previous House elections, but no difference emerged in Senate or state-level presidential returns (see Appendix D). As such, we specify our models with the lagged vote share for the respondent’s party, although notably, the results are not sensitive to this inclusion.

Final Specifications

Feeling secure that our analysis is likely to be informative, we specify our RDD using rdrobust to study the effect of a close win on post-election polarization, in-party affect, and out-party affect. The model works by estimating local linear regressions on either side of the threshold and then comparing the difference in slopes between each side of the threshold (Imbens and Lemieux 2008). We employ this estimation technique, as well as robust bias-robust estimators, which estimate local quadratic regressions instead of assuming linearity (Cattaneo et al. 2020). In addition to the win/loss indicator, the model also includes the dependent variable’s pre-election value, which creates a pre-post test of the effect of winning; the respondent’s party’s margin of victory or defeat in the election (the running variable); its lagged margin of victory/defeat; and year fixed effects. We use HC2 cluster-robust standard errors and all hypothesis tests are two-tailed.

Results

Figure 2 presents the results of RDD analyses run at two different bandwidths (3 percent and the optimal bandwidths), for three subnational election types, and on three dependent variables. A figure using all bandwidths can be found in Figure A5 in Appendix E, and figures using bias-corrected and conventional estimates can be found in Figures A3 and A4, respectively. Findings can be found in table format in Appendix F.

Figure 2.

Effect of close victory on partisan attitudes in US House, Senate, and state-level presidential races.

When interpreting the results, recall that in the prior analysis, we found the losing side to emerge from the election less polarized due to lower in-party love. Here, however, we only see the gap between winners and losers—that is, there is no comparison to a pre-election baseline—so compatible results would show a positive effect of winning on polarization and in-party love, but not on out-party affect. This mostly aligns with what we see. The results that align best with the fixed effects model are based on state-level presidential returns, where greater polarization among the winners seems to be driven exclusively by their higher post-election affect. At the other levels, the results are compatible but with caveats. Most results after close Senate elections are nonsignificant, but the results are directionally consistent with what the fixed effects analysis would expect, and significantly so when using the optimal bandwidths. House elections, however, follow a somewhat aberrant pattern. While there is evidence that winners emerge with higher in-party warmth than losers, they also emerge with lower out-party warmth, both of which appear to drive greater levels of affective polarization among winners and losers.

Just as with the individual fixed effects design, we also create z-scored versions of partisan affect measures to address the issue that pre- and post-election measures use different scales (0–100 and 0–10, respectively). Results can be found in Appendix G. Findings are consistent with that presented in the main text, and if anything, are more robust to the threshold used than when raw scores are employed.

Conclusion

In this paper, we present what in our view is the most thorough study yet on how US election outcomes impact the partisan outcomes of winning and losing sides. Our initial analysis used the pre- and post-election panelized nature of the ANES to compare respondents’ attitudes after Election Day to a pre-election baseline. It showed that respondents whose party lost the presidential race depolarized relative to the winners, and that this was primarily driven by a drop in in-party love. Then, recognizing the limits of an analysis on so few races, we tested our findings in the context of subnational results as well. Perhaps as expected given the races’ lower salience, the results of this analysis were not as uniform. The overall patterns, however, gave additional support to our conclusion that in an election’s aftermath, the losing side depolarizes primarily as a result of lower in-party affect.

The evidence that outcomes lead to depolarization should provide some solace to scholars concerned about the impact of polarization. While campaign activity may increase partisan animosity (Fasching et al. 2024), tensions appear to drop when the election ends, at least on the losing side. This pattern is consistent with when people distance themselves from their lower-status group memberships (Axt et al. 2018), and suggests that even if the losing side has weaker trust in the electoral process (Kern et al. 2024), it is not because they are embittered at the opposition.

These results in mind, however, our study has some limitations. While the consistent results between our analyses is encouraging, our limited ability to account for time-variant factors in the fixed effect analysis, and the ever-present concern of unobserved imbalance around the RDD cutoffs, highlight the difficulty of distinguishing between causal and associative relationships with observational data. We feel that the study offers a relatively clear account of how election outcomes affect polarization in the end, but experimental evidence like that presented in Hamrak (2025) can speak more to the immediate, causal effects. Happily, results from these studies pair together reasonably well. Both show that winners emerge more polarized and that in-party love is the driving mechanism, and the analysis in Appendix B supports the possibility that winners are prideful immediately after the election, while losers are dejected in the months that follow. Additional research may seek to explain why the different short- and long-run effects of elections appear different.

Another limitation is that our findings only engage American elections, which means comparisons with other countries can only occur indirectly. In headline figures, our findings dovetail with those from other countries. Cross-nationally, winners tend to be more affectively polarized than losers (Andrews and Huang, 2024). But this masks variation in post-election patterns across countries (e.g., Gidron and Sheffer 2024; Sheffer 2020), as well as the fact that American voters display this tendency particularly strongly (Andrews and Huang, 2024; Hamrak, 2025). We offer up some explanations for why this may be. First, the American public are not necessarily unusual in their levels of affective polarization, but the growth of affective polarization is unusually strong in American society (Adams et al. 2020). Since levels of affective polarization are more in flux, intervening events may impact American voters more than other voters (c.f., Fasching et al. 2024). Second, compared to other countries, the US is unusual in how well partisanship neatly captures political conflict. Multi-party systems often feature coalition bloc patterns of affective polarization (Tichelbaecker et al. 2023), and such coalitions can be more fragile and membership subject to post-election dynamics than partisan identities in the US. By contrast, winner-takes-all elections may more clearly settle political conflict. Third, the US is not the only single member district system, but it is unusually two-party (Chhibber and Kollman 1998). Such a setup may make victories more decisive, and thus make the “treatment” of winning or losing feel stronger than in other countries. Future work, however, would do well to disentangle why the presence and intensity of losers’ depolarization varies so considerably cross-nationally.

One final item of intrigue is the fact that findings for House elections differ from other kinds of elections. Those who see their party win House elections emerge with greater in-party warmth than those who see their party lose, in line with other findings, but also emerge with lower out-party warmth. While our research design cannot disentangle whether winning or losing is driving this effect, we offer two possibilities for why this is the case. The first is theoretical. Except for the least populous states, House elections represent a considerably more localized form of representation than Senators, who represent states, or a nationally representative president. Even as rising polarization limits the extent to which voters reward legislators for constituency work (Dancey et al. 2024), losers may be motivated to put their partisan feelings aside to some extent to attain a sense of access to the local representative they have (Frederick 2007). The second possibility is methodological. While the amount of respondents across election types is fairly consistent, there is far more variation in party electoral performance for House elections than Senate or presidential elections. It could be that similar findings would emerge for Senate or presidential elections should the number of elections used in analysis increase. Future work would do well to unpack the distinctiveness of House elections further.

Ultimately, if anything, these limitations serve to highlight that this study enters a fruitful area of research. Moreover, it does so by offering relatively clear findings over a 24-year period in the world’s largest advanced democracy—and one of its most polarized (Garzia et al. 2023). By comparing post-election attitudes to a pre-election baseline, and approaching the question from multiple analytic angles, our findings can at least somewhat allay concerns that US elections contribute to polarization any more than temporarily.

Supplemental Material

Supplemental Material - Election Outcomes and Affective Polarization in the United States

Supplemental Material for Election Outcomes and Affective Polarization in the United States by Joseph B. Phillips, Seth B. Warner in Political Research Quarterly.

Footnotes

ORCID iDs

Joseph B. Phillips

Seth B. Warner

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

Data and replication code are available on Harvard Dataverse by Phillips and Warner (2025) ().

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Adams

Gidron

Horne

W. T

. 2020. American Affective Polarization in Comparative Perspective. Cambridge University Press.

Andrews

Huang

Y.-S

. 2024. “Winners, Losers, and Affective Polarization.” Party Politics 31 (5): 841–53.

Axt

J. R.

Moran

Bar-Anan

. 2018. “Simultaneous Ingroup and Outgroup Favoritism in Implicit Social Cognition.” Journal of Experimental Social Psychology 79 (1): 275–89.

Blais

Gélineau

. 2007. “Winning, Losing and Satisfaction with Democracy.” Political Studies 55 (2): 425–41.

Brewer

M. B.

Kramer

R. M

. 1986. “Choice Behavior in Social Dilemmas: Effects of Social Identity, Group Size, and Decision Framing.” Journal of Personality and Social Psychology 50 (3): 543–49.

Brodeur

Cook

Heyes

. 2020. “Methods Matter: P-Hacking and Publication Bias in Causal Analysis in Economics.” The American Economic Review 110 (11): 3634–60.

Calonico

Cattaneo

M. D.

Titiunik

. 2015. “Rdrobust: An R Package for Robust Nonparametric Inference in Regression-Discontinuity Designs.” The R Journal 7 (1): 38–51.

Cattaneo

M. D.

Titiunik

Vazquez-Bare

. 2020. “Analysis of Regression-Discontinuity Designs with Multiple Cutoffs or Multiple Scores.” STATA Journal: Promoting communications on statistics and Stata 20 (4): 866–91.

Chhibber

Kollman

. 1998. “Party Aggregation and the Number of Parties in India and the United States.” American Political Science Review 92 (2): 329–41.

10.

Cialdini

R. B.

Borden

R. J.

Thorne

Walker

M. R.

Freeman

Sloan

L. R

. 1976. “Basking in Reflected Glory: Three (Football) Field Studies.” Journal of Personality and Social Psychology 34 (3): 366–75.

11.

Connors

E. C

. 2023. “Social Desirability and Affective Polarization.” Public Opinion Quarterly 87 (4): 911–34.

12.

Cunningham

Platow

M. J

. 2007. “On Helping Lower Status Out‐Groups: The Nature of the Help and the Stability of the Intergroup Status Hierarchy.” Asian Journal of Social Psychology 10 (4): 258–64.

13.

Dancey

Henderson

Sheagley

. 2024. “The Personal Vote in a Polarized Era.” American Journal of Political Science 68 (4): 1479–97.

14.

De Jonge

Veenhoven

Arenda

. 2014. “Homogenizing Responses to Different Survey Questions on the Same Topic: Proposal of a Scale Homogenization Method Using a Reference Distribution.” Social Indicators Research 117 (1): 275–300.

15.

De la Cuesta

Imai

. 2016. “Misunderstandings About the Regression Discontinuity Design in the Study of Close Elections.” Annual Review of Political Science 19 (1): 375–96.

16.

Eggers

A. C.

Fowler

Hainmueller

Hall

A. B.

Snyder

J. M

Jr. . 2015. “On the Validity of the Regression Discontinuity Design for Estimating Electoral Effects: New Evidence from over 40,000 Close Races.” American Journal of Political Science 59 (1): 259–74.

17.

Fasching

Iyengar

Lelkes

Westwood

S. J

. 2024. “Persistent Polarization: the Unexpected Durability of Political Animosity Around US Elections.” Science Advances 10 (36): eadm9198.

18.

Frederick

. 2007. “Constituency Population and Representation in the U.S. House.” American Politics Research 36 (3): 358–81.

19.

Garzia

da Silva

F. F.

Maye

. 2023. “Affective Polarization in Comparative and Longitudinal Perspective.” Public Opinion Quarterly 87 (1): 219–31.

20.

Gidron

Sheffer

. 2024. “Differentiating the Sources of Post-election Partisan Affect Warming.” European Journal of Political Research 63 (3): 1155–74.

21.

Guimond

Dambrun

. 2002. “When Prosperity Breeds Intergroup Hostility: the Effects of Relative Deprivation and Relative Gratification on Prejudice.” Personality and Social Psychology Bulletin 28 (7): 900–12.

22.

Hainmueller

Hall

A. B.

Snyder

J. M

Jr. . 2015. “Assessing the External Validity of Election RD Estimates: An Investigation of the Incumbency Advantage.” The Journal of Politics 77 (3): 707–20.

23.

Hall

A. B.

Thompson

D. M.

2025. “Who Punishes Extremist Nominees? Candidate Ideology and Turning Out the Base in US Elections.” Working Paper. https://www.andrewbenjaminhall.com/hall_thompson_revised.pdf

24.

Hamrak

2025. “Identity After the Ballot: How Winning and Losing Impact Partisan Attachments and Affective Polarization.” Electoral Studies 97: 102967.

25.

Hernandez

Anduiza

Rico

. 2021. “Affective Polarization and the Salience of Elections.” Electoral Studies 6 (1): 102203.

26.

Hornsey

M. J.

Spears

Cremers

Hogg

M. A

. 2003. “Relations Between High and Low Power Groups: the Importance of Legitimacy.” Personality and Social Psychology Bulletin 29 (2): 216–27.

27.

Huddy

2004. “Contrasting Theoretical Approaches to Intergroup Relations.” Political Psychology 25 (6): 947–67.

28.

Hyytinen

Meriläinen

Saarimaa

Toivanen

Tukiainen

. 2018. “Public Employees as Politicians: Evidence from Close Elections.” American Political Science Review 112 (1): 68–81.

29.

Imbens

G.W.

Lemieux

. 2008. “Regression Discontinuity Designs: A Guide to Practice.” Journal of Econometrics 142 (2): 615–35.

30.

Iyengar

Sood

Lelkes

. 2012. “Affect, Not Ideology: A Social Identity Perspective on Polarization.” Public Opinion Quarterly 76 (3): 405–31.

31.

Jost

J. T.

Burgess

. 2000. “Attitudinal Ambivalence and the Conflict Between Group and System Justification Motives in Low Status Groups.” Personality and Social Psychology Bulletin 26 (3): 293–305.

32.

Kern

Marien

Muradova

. 2024. “The Effect of Accumulated Losses on Perceptions of Democratic Legitimacy.” European Political Science Review 16 (3): 465–82.

33.

Kingzette

Druckman

J. N.

Klar

Krupnikov

Levendusky

Ryan

J. B

. 2021. “How Affective Polarization Undermines Support for Democratic Norms.” Public Opinion Quarterly 85 (2): 663–77.

34.

Lalonde

R. N.

1992. “The Dynamics of Group Differentiation in the Face of Defeat.” Personality and Social Psychology Bulletin 18 (3): 336–42.

35.

Martin

Nai

. 2024. “Deepening the Rift: Negative Campaigning Fosters Affective Polarization in Multiparty Elections.” Electoral Studies 87: 102745.

36.

Meyer

T. M.

Gatterman

. 2025. “Does Media Framing of Election Results Affect Whether Voters Perceive Parties as Election Winners or Losers?” Political Studies. https://doi.org/10.177/00323217251350742

37.

Miller

C. B.

2009. “Yes We Did! Basking in Reflected Glory and Cutting off Reflected Failure in the 2008 Presidential Election.” Analyses of Social Issues and Public Policy 9 (1): 283–96.

38.

MIT Election Data and Science Lab . 2022a. U.S. House 1976-2022, V13. Harvard Dataverse.

39.

MIT Election Data and Science Lab . 2022b. U.S. Senate Statewide 1976-2020, V13. Harvard Dataverse.

40.

MIT Election Data and Science Lab . 2023. U.S. President 1976-2022, V13. Harvard Dataverse.

41.

Moore

Bashshur

M. R

. 2018. “When the Tables are Turned: The Effects of the 2016 US Presidential Election on In-Group Favoritism and Out-Group Hostility.” PLoS One 13 (5): e0197848.

42.

Pew Research Center . 2023. “Americans’ Feelings About Politics, Polarization and the Tone of Political Discourse.” https://www.pewresearch.org/politics/2023/09/19/americans-feelings-about-politics-polarization-and-the-tone-of-political-discourse/

43.

Phillips

J. B.

2024. “Affective Polarization and Habits of Political Participation.” Electoral Studies 87: 102733.

44.

Phillips

Warner

. 2025. “Replication Data for: Election Outcomes and Affective Polarization in the United States”, Harvard Dataverse, V1, UNF:6:yQsdoz0xNs2qrOLCD8CQhw== [fileUNF], https://doi.org/10.7910/DVN/XGS79K

45.

Riek

B. M.

Mania

E. W.

Gaertner

S. L

. 2006. “Intergroup Threat and Outgroup Attitudes: A Meta-Analytic Review.” Personality and Social Psychology Review 10 (4): 336–53.

46.

Rubin

Badea

Jetten

. 2014. “Low Status Groups Show In-Group Favoritism to Compensate for Their Low Status and Compete for Higher Status.” Group Processes & Intergroup Relations 17 (5): 563–76.

47.

Sharpe

2023. “New Umass Amherst Poll of Nearly 300 Former Members of Congress Shows Rising Concerns About Threats of Political Violence.” https://www.umass.edu/news/article/new-umass-amherst-poll-nearly-300-former-members-congress-shows-rising-concerns-about

48.

Sheffer

2020. “Partisan In-Group Bias Before and After Elections.” Electoral Studies 67: 102191.

49.

Sidanius

Pratto

. 1999. “Social Dominance Theory.” In Handbook of Theories of Social Psychology, edited by Van Lange

P. A. M.

Kruglanski

Arie W.

Higgins

E. T.

, 1144. Sage Publications.

50.

Singh

S. P.

Thornton

J. R

. 2019. “Elections Activate Partisanship Across Countries.” American Political Science Review 113 (1): 248–53.

51.

Smith

H. J.

Pettigrew

T. F.

Pippin

G. M.

Bialosiewicz

. 2012. “Relative Deprivation: A Theoretical and Meta-Analytic Review.” Personality and Social Psychology Review 16 (3): 203–32.

52.

Tichelbaecker

Gidron

Horne

Adams

. 2023. “What Do We Measure When We Measure Affective Polarization Across Countries?” Public Opinion Quarterly 87 (3): 803–15.

53.

Turner

J. C.

Hogg

M. A.

Turner

P. J.

Smith

P. M

. 1984. “Failure and Defeat as Determinants of Group Cohesiveness.” British Journal of Social Psychology 23 (2): 97–111.

54.

Wann

D. L.

Haynes

McLean

Pullen

. 2005. “Sport Team Identification and Willingness to Consider Anonymous Acts of Hostile Aggression.” Aggressive Behavior 29 (5): 406–13.

55.

Warner

2023. The Politics of Polarized Places. Doctoral dissertation, Penn State University.

56.

Wing

Bello-Gomez

R. A

. 2018. “Regression Discontinuity and Beyond: Options for Studying External Validity in an Internally Valid Design.” American Journal of Evaluation 39 (1): 91–108.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.87 MB