1. Introduction
After decades studying the polarization of ideologies, attitudes, and issue opinions among citizens in democratic countries, political scientists have recently turned to examining identity-based, affective polarization: the negative feelings toward rival partisans compared to the positive feelings toward the supporters of one's own party (Iyengar et al., Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019). In the United States, scholars have noted that affective dislike of out-party supporters has been on the rise (Iyengar et al., Reference Iyengar, Sood and Yphtach2012; Webster and Abramowitz, Reference Webster and Abramowitz2017). Similar increases are not visible in other contexts (Boxell et al., Reference Boxell, Conway, Druckman and Gentzkow2020, Reference Boxell, Gentzkow and Shapiro2022), but this is in part because affective dislike of out-party supporters was always high in other countries (Gidron et al., Reference Gidron, Adams and Horne2020; Wagner, Reference Wagner2021). Hence, affective polarization is an important phenomenon across many different contexts. While affective polarization may be inherent in political competition and even have positive consequences, for example, by driving turnout (Harteveld and Wagner, Reference Harteveld and Wagner2023), it is generally seen as potentially troubling for democracies (McCoy and Somer, Reference McCoy and Somer2021).
As a result, attention has recently turned to how affective dislike of out-parties could be reduced. Some of this work examines cross-national patterns to uncover contextual factors associated with lower affective polarization (e.g., Gidron et al., Reference Gidron, Adams and Horne2020). Other work explicitly focuses on the effectiveness of specific interventions, inspired by a recent revival of interest in research on prejudice reduction (Paluck and Green, Reference Paluck and Green2009; Paluck et al., Reference Paluck, Porat, Clark and Green2021). For instance, Levendusky and Malhotra (Reference Levendusky and Malhotra2016) show that depictions of the public as moderate rather than extreme reduce affective polarization, while Levendusky (Reference Levendusky2018a) provides evidence that events that prime higher-order national identities—such as the July 4 national holiday—reduce affective polarization (though see Brandt et al., Reference Brandt, Turner-Zwinkels, Inbar and Quinn2020). Interpersonal contact, even if only imagined, can increase inter-partisan affect (Wojcieszak and Warner, Reference Wojcieszak and Warner2020; Amsalem et al., Reference Amsalem, Merkley and Loewen2022), as can highlighting certain social norms such as open-mindedness (Wojcieszak et al., Reference Wojcieszak, Winter and Yu2020). However, not all efforts to reduce group-centric attitudes prove to be effective. For instance, heightening partisan ambivalence (i.e., highlighting aspects of one's in-party one disagrees with) and self-affirmation techniques fail to robustly reduce affective polarization (Levendusky, Reference Levendusky2018b).
Our paper speaks to this literature on mitigating affective polarization by focusing on how elites interact, using the example of coalition signals. We ask how and why the willingness among party elites to govern together reduces affective dislike toward a rival party's voters. More specifically, we design a survey experimentFootnote 1 and examine situations in which partisans face the possibility of a coalition (a) between their party and a rival party and (b) between two rival parties.
Our experiment builds on and extends important prior research. Experiments in the United States show that a respectful and trusting interaction between political elites reduces affective polarization (Huddy and Yair, Reference Huddy and Yair2021). Our study complements their work by looking at the strongest form of inter-party interaction, namely a common government. Observational research by Horne et al. (Reference Horne, Adams and Gidron2023) that shows an association between co-governance experience and warmer inter-party affective polarization. Our work adds and goes beyond this contribution by implementing a survey experiment that allows us to examine voters’ reaction to specific coalition signals. To our knowledge, there is only one experimental study that looked at coalition formation and tolerance of out-party supporters (Bassan-Nygate and Weiss, Reference Bassan-Nygate and Weiss2022). This study, however, is on the unique case of an unity government in Israel and does not allow a test of the effects of coalition signals on out-partisans (i.e., on partisans of other parties than the ones signaling to govern together).
Our key expectation is that, when party elites signal that they can work and govern together, there will be a decrease in the mutual affective dislike among their partisans. In addition to examining the overall effect of coalition signals on affective distance, we also examine whether these effects are due to changes to perceived ideological distance (Fortunato and Stevenson, Reference Fortunato and Stevenson2013; Falcó-Gimeno and Muñoz, Reference Falcó-Gimeno and Muñoz2017; Hjermitslev, Reference Hjermitslev2022) and/or due to warmer personal relations between party leaders (Huddy and Yair, Reference Huddy and Yair2021). We test our expectations in an online survey experiment conducted in Austria in fall 2021. Usefully for our experimental design, this was a period when various coalitions between major parties seemed possible. Our findings indicate that cooperating with rivals reduces affective distance. While there is indicative evidence that elite signals influence inter-partisan affect via perceptions of ideological distance, we find no discernible impact of warmer personal relations.
In addition, we also examine situations in which partisans are exposed to coalition signals between two out-parties, that is, two parties that they do not support. We suggest that reactions to such signals should be muted, limiting the overall systemic impact of coalition signals on affective polarization. However, we also show that certain citizens react to a situation where two out-parties form a coalition, specifically when one of the two parties is from the same ideological bloc and thus viewed quite positively (Reiljan and Ryan, Reference Reiljan and Ryan2021)—for example, left-wing voters evaluating coalition signals between the Social Democratic party and right-wing parties. We show that these voters reduce their affective distance toward supporters of the right-wing out-party. Hence, coalition signals can have effects on voters in the same ideological bloc, influencing attitudes beyond the narrow set of party supporters.
Thus, this paper contributes to research on how to mitigate affective polarization (Levendusky, Reference Levendusky2018a). We provide evidence on how, why, and among whom coalitions and coalition signals work to reduce affective dislike. Hence, the way in which elites talk to and treat each other has an impact on society at large. Our findings imply that politicians that want to reduce levels of partisan animosity can do so by signaling their willingness cooperate with rival parties, perhaps particularly if these are ideologically distant. Intraparty cooperation—or even just the willingness to discuss such cooperation—thus has consequences for how deeply partisan divides characterize society. Given that in our case our findings apply most clearly to the radical right, our paper also sheds light on processes of legitimization and normalization of these non-mainstream competitors (Valentim, Reference Valentim2021).
Just as important is our contribution to the literature on affective polarization in general (Iyengar et al., Reference Iyengar, Sood and Yphtach2012, Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019). Our results have implications for understanding how affective polarization differs between countries and across time, as we show when and among which citizens coalition signals formation will have an effect on affective dislike. Elite signals will mainly affect supporters of the two parties directly involved, but may also have additional, distinct effects on other partisans within the same ideological bloc. Moreover, we argue that one likely reason why elite signals matter is in part because they imply that ideological gaps as perhaps not as deep as imagined.
2. Theoretical framework
To understand why elite cooperation in general—and coalition signals specifically—may reduce affective distance between partisans, it is first important to understand relevant sources of affective distance. Overall, recent work that builds on social psychological research on social groups has highlighted the importance of perceived threat (e.g., Stephan et al., Reference Stephan, Boniecki, Ybarra, Bettencourt, Ervin, Jackson, McNatt and Renfro2002). In a multiparty system, rival parties whose political plans and projects can be seen as a danger to voters’ worldview and identities will therefore disliked more (Renström et al., Reference Renström, Bäck and Carroll2021). The perceived threat will thus depend on factors such as the ideological distance between one's own preferences and those of the party (Rogowski and Sutherland, Reference Rogowski and Sutherland2016) or the extent to which groups have solidified into stable, deep-seated social identities or even tribes (Mason, Reference Mason2018; Finkel et al., Reference Finkel, Bail, Cikara, Ditto, Iyengar, Klar, Mason, McGrath, Nyhan and Rand2020).
This also means that intergroup attitudes and affect may be responsive to factors that change the perception of threat, and elites may play an essential role in this. In political science, Zaller (Reference Zaller1992) highlighted the role that elite cues have in forming perceptions of politics. From a social psychology perspective, Hogg (Reference Hogg2001) stresses the role that group leaders play in determing group relations (see also Bassan-Nygate and Weiss, Reference Bassan-Nygate and Weiss2022; Huddy and Yair, Reference Huddy and Yair2021). While elites can of course heighten affective distance through their rhetoric, they can also reduce affective dislike by making these groups and their members appear to be less threatening. Hence, if politicians show in their rhetoric or actions that they are willing to work with other parties, this may also affect how people think about and evaluate the supporters of these parties.
There is some evidence from existing work that partisans dislike rival party supporters less in the context of mutual coalitions. Using cross-national data from the Comparative Study of Electoral Systems, Horne et al. (Reference Horne, Adams and Gidron2023) thus show that parties that are currently in government together are characterized by higher mutual affect, as are parties that have a history of governing together. Hence, it is likely that affective distance to out-party supporters is lower for parties that signal that they are willing to form a coalition. Specifically, we expect that:
H1: Citizens will have higher levels of affect toward out-partisans when their in-party signals that it is willing to form a coalition with the out-party.
2.1 The mediating role of perceptions of party ideology
H1 posits that governing together—and even just signaling the willingness to do so—reduces perceptions that the out-party and its supporters are a threat. As existing research shows, a key determinant of these threat perceptions is ideological distance: parties that are more dissimilar in terms of their political aims are a greater danger to the preferences—and even way of life—of individual citizens, and are thus disliked more (Rogowski and Sutherland, Reference Rogowski and Sutherland2016).Footnote 2 For example, running an experiment in the United States, Webster and Abramowitz (Reference Webster and Abramowitz2017) manipulated the ideological distance between a respondent and a hypothetical congressional candidate from the opposing party. The results show that respondents that were confronted with ideologically extreme candidates disliked the opposing candidate much more compared to respondents that did not receive information about the ideological standpoint (while respondents that were treated with ideologically moderate opposing candidates gave much more positive ratings). Conversely, threat perceptions should also be lower if parties are more similar ideologically.
Hence, one reason why coalition signals will decrease perceptions of threat is because they affect perceptions of ideological distance. We know from prior research that parties in coalition are seen as being closer together ideologically (Fortunato and Stevenson, Reference Fortunato and Stevenson2013; Adams et al., Reference Adams, Ezrow and Wlezien2016; Falcó-Gimeno and Fernandez-Vazquez, Reference Falcó-Gimeno and Fernandez-Vazquez2020; Hjermitslev, Reference Hjermitslev2022). Given the difficulty of acquiring detailed ideological information of party positions, voters use cooperation in a coalition as a heuristic for the ideological position of parties. This also extends to coalition signals, not just actual cooperation (Falcó-Gimeno and Muñoz, Reference Falcó-Gimeno and Muñoz2017).
The effect of coalition signals on perceived ideological distance may be particularly relevant for parties that are currently at the fringes of the party system, such as the radical right or left. In addition to the simple fact of their ideological radicalism, these parties are often given a “pariah” status by mainstream parties, with co-governing prohibited by a “cordon sanitaire” (Van Spanje and Van Der Brug, Reference Van Spanje and Van Der Brug2007). Elite signals could therefore work to normalize radical competitors in particular (Valentim, Reference Valentim2021).
In sum, it is likely that at least part of the effect of coalition signals should be mediated by changed perceptions of where the parties are ideologically located. We thus expect that:
H2: The effect of coalition signals on affective polarization is mediated by the perception that the parties willing to form a coalition are ideologically closer to one another.
2.2 The effect of warm personal relations between rival politicians
Affective polarization is based not just on ideological distance, but also on whether out-party politicians and voters are seen as socially distinct, distant groups. Horne et al. (Reference Horne, Adams and Gidron2023) argue that coalitions between parties also mean that the two parties interact more in public, for example when they join together to defend the government against attacks from the opposition. This can apply not only to policy attacks but also to attacks based on politician valence such as integrity or competence. In related research, Huddy and Yair (Reference Huddy and Yair2021) show that vignettes that stress warm interpersonal relations between two leading US politicians reduce affective polarization more than policy compromise. They argue that such warm intergroup relations affect norms of behavior and shape relational identities.
Affective polarization may be particularly reduced when warm social relations between elites remind voters that good personal relations with out-partisans are possible, even if there are ideological disagreements and even if social distance exists. Hence, we expect that:
H3: The effect of coalition signals on affective polarization is stronger if the parties also signal that there are warm personal relations between them, not just the basis for policy compromise.
2.3 The effect of coalition signals on other voters
All existing work on the effect of inter-party cooperation has examined the effects on supporters of the two parties that signal that they can work together. However, in multiparty systems many voters will not support either of the two parties that are signaling their willingness to cooperate. For example, if a radical-right and a moderate-right party say they could work together, how do other supporters of other parties react? Importantly, we believe that—under certain circumstances—coalition signals should also have consequences for those for whom neither party is an in-party.
To understand why this is the case, our starting point is that in many European countries not all out-parties are alike (Reiljan and Ryan, Reference Reiljan and Ryan2021; Wagner, Reference Wagner2021). Instead, party systems operate as ideological blocs (Bartolini and Mair, Reference Bartolini and Mair1990), where left-wing parties are distinct from right-wing parties. Affective polarization thus occurs across blocs as well as across parties (Kekkonen and Ylä-Anttila, Reference Kekkonen and Ylä-Anttila2021), with radical-right parties arguably forming a distinct bloc (Oesch and Rennwald, Reference Oesch and Rennwald2018). Thus, voters form distinctive affective groups not just based on parties, but also based on ideological camps. Hence, affect toward partisans within one bloc will be positive, even if not as high as toward the in-party. Interestingly, part of the reason why these ideological and affective camps may be precisely due to repeated elite signals, with party leaders indicating their willingness to cooperate across party lines or even declare pre-election pacts.
So, a first scenario may be one in which both parties sending coalition signals are from the opposing bloc. In this case, we focus on those who support neither of the parties sending out cooperative signals and see those two parties and their supporters in negative terms. In such a case, we might not expect much of a reaction when such voters are presented with cooperative signals: why should this citizen shift in the evaluation of out-party supporters just because the two disliked actors now pledge to cooperate? Thus, it seems reasonable not to expect an effect (or, if anything, an increase in affective distance).
A second scenario occurs when one of the two parties is from the opposing bloc, but the other from one's own ideological camp. In this case, we focus on citizens who belong to the same ideological bloc as one of the parties sending out cooperative signals. We expect that coalition signals between an in-bloc and an out-bloc party may influence affect levels toward the two parties, but this effect may differ for the two parties. Coalition signals with an in-bloc party will increase affect for partisans of the out-bloc party. For instance, if a Green voter hears that the Social Democrats could see themselves working with a center–right party, then supporters of the center–right party will be disliked less. In addition, there may be an additional change in affect toward supporters of the in-bloc party; in the above example, affect toward supporters of the Social Democrats would decrease, as the elite signals point to cross-bloc defection and greater ideological distance. While we test for this effect, our key expectation concerns the out-bloc partisans.
Hence, we expect thatFootnote 3:
H4a: Among citizens for whom neither party sending coalition signals is in the same ideological bloc, affective distance will not be affected by elite signals.
H4b: Among citizens for whom one party sending coalition signals is in the same ideological bloc, elite signals will reduce affective distance toward the out-party.
3. Coalition governments in Austria
Austria can be understood as a typical case for Western European countries with multi-party systems. At the time of our survey, five parties were represented in parliament. These were the Austrian People's Party (ÖVP), the Social Democratic Party (SPÖ), the radical-right Freedom Party (FPÖ), the Greens as well as the liberal party NEOS—The New Austria and Liberal Forum. The largest party, the ÖVP, held 39 percent of the parliamentary seats, followed by the SPÖ (22 percent), the FPÖ (16 percent), the Greens (14 percent), and the NEOS (8 percent).
Given the likely existence of pretreatment effects (Slothuus, Reference Slothuus2016), it is important to lay out the current and past patterns of coalition between these parties. The ÖVP has been in coalition governments with the SPÖ, the FPÖ, and the Greens in the past ten years. At the time of writing, the party was in government with the Greens. The prior coalition, also led by the ÖVP, was with the FPÖ, but this collapsed acrimoniously in 2019; the parties had previously already governed together in the early 2000s. Historically speaking, the so-called “Grand coalition” between the ÖVP and the SPÖ has dominated Austrian coalition governments. Out of 33 governments that were formed after World War II, 20 included both the ÖVP and the SPÖ. The SPÖ and the FPÖ have only been in government together at the national level once, in the early 1980s, before the rightward shift of the FPÖ in 1986. Since then, the SPÖ has in the run-up to elections regularly excluded the possibility of forming a national-level coalition with the party—a commitment that social democratic party elites have recently questioned, in part because doing so massively reduces the party's bargaining power. While the Greens are currently in government for the first time, the NEOS have no prior government experience. All parties have, however, been part of joint regional governments, and a wide variety of governing arrangements currently exist at that level. Hence, coalitions between all parties are plausible, even if some will be seen as more likely than others. We will discuss this in more detail below.
4. Experimental design
We base our analyses on a quantitative online survey among residents in Austria ($N = 2190$
).Footnote 4 The survey was conducted between the October 21 and November 3, 2021 by the company Marketagent.
Importantly, our experiment benefited from the uncertain political situation during the time of our fieldwork. Shortly before we ran the survey, the Austrian Chancellor Sebastian Kurz (ÖVP) had to resign from office. Kurz was involved in bribery and corruption allegations, and pressure by the junior coalition partner, the Greens, and the state governors was too high for him to remain in office. At the time of the survey, several options were being discussed in the media, ranging from a mere government reshuffle within the existing coalition to the possibility of early elections. Party elites were engaged in different negotiations with uncertain results, so that different coalition options were being discussed in the media. Thus, the survey was conducted in a setting where we could credibly discuss new government compositions in our vignette.
Our respondents were split into seven groups (six treatment groups and one control group). The treatment groups read fake newspaper articles on a potential early election and the most likely partisan composition of the next coalition government (see Figure 1). We chose to focus on three different coalition types that included the three main political parties. Thus, we designed vignettes for a potential ÖVP–SPÖ, ÖVP–FPÖ, and SPÖ–FPÖ coalition.Footnote 5 This decision provides us with a range of partisan composition while ensuring large enough sub-samples of partisans for our analyses. It allows us to test hypotheses 1–3 based on party identifiers from these three actors. Hypotheses 4a and 4b will be tested on all party identifiers whose party was not mentioned in the vignette. The most likely partisan composition of the next coalition government was supported by a quote of a political expert and a reference to former coalitions of that type either at the national or regional level. Three of the six treatment groups included additional information on a warm social relationship between the party elites involved in the informal talks. The stories revealed that the party elites got to know each other and even laughed with each other (see Appendix B for all news stories). This is meant to capture the “warm personal relations” that sometimes accompany coalition signals (Huddy and Yair, Reference Huddy and Yair2021). For the main analyses, we average across the two types of treatment in order to increase sample size in each group. At the end of the survey, participants were debriefed that they had taken part in a survey experiment, and that the news article was invented. All question texts are included in Appendix C. The balance test in Appendix D reveals no significant differences between the seven groups.
Figure 1. Example vignette. Note: One of six treatment types. Text relating to parties that was varied in each treatment highlighted in gray. The three main treatments were: ÖVP–SPÖ, ÖVP–FPÖ, and SPÖ–FPÖ. The paragraph entitled “improved personal relationships” was shown to half of the treatment groups only. Full text of vignettes (in German) is provided in Appendix B.
Data from a concurrently-run representative survey show that party supporters of each party like supporters of their own party but dislike supporters of the other two parties.Footnote 6 Hence, competition between these three parties is tripolar (Oesch and Rennwald, Reference Oesch and Rennwald2018), with each party's supporters seeing the other two as out-groups. Furthermore, those who support no party on average dislike supporters of all three parties, but supporters of Greens and NEOS tend to feel comparatively warm toward SPÖ voters. Hence, we should classify Greens and NEOS supporters as closest to the SPÖ in terms of affective aspects of party competition, so for the purposes of hypothesis 4b we examine how Green and NEOS supporters react when the SPÖ crosses bloc lines to signal cooperation with the ÖVP or FPÖ.
4.1 Manipulation check
As a manipulation check for our treatment, respondents were asked to evaluate the probability that the two parties in the vignette would form the next government, using an 11-point-scale.Footnote 7 T-test results presented in Appendix H show that our vignette had the expected effect on respondent perceptions: treated respondents rate the likelihood of the respective coalition more highly compared to those in the control group. Specifically, those who see the respective vignette think the coalition described is 0.39 points higher for the ÖVP–SPÖ vignette, 0.57 points higher for the ÖVP–FPÖ vignette, and 0.67 points for the SPÖ–FPÖ coalition, compared to the control group that did not see a vignette. The standard deviations of the probability perceptions are around 2.6–2.7 points, so the effect is around a quarter of a standard deviation—a substantial effect. Each of these differences is significant at the 0.05 level. Importantly, there is no clear or consistent effect on other coalition perceptions, so receiving the ÖVP–SPÖ vignette, for instance, does not make respondents think that the other two coalitions we asked about are more or less likely. The above analyses average across the presence of the warm personal relations condition, and differences between the two conditions are small in terms of the manipulation check. Overall, we take these effects as a sign that the vignette was seen as credible by respondents.
The successful manipulation of expectations shows that pre-treatment effects, while present, did not mean that this new information was ignored. At the same time, the stronger effect on coalitions involving the FPÖ is consistent with this option being the least plausible in the context of the fieldwork period. Looking at the probability ratings for each coalition type in the control group, who did not read the vignettes, we can see that respondents view an ÖVP–SPÖ coalition as most likely, followed by an ÖVP–FPÖ coalition and an SPÖ–FPÖ coalition. Parties’ co-governing histories—the long-tradition of “grand coalitions” especially in times of crisis, the quite recent, though unsuccessful ÖVP–FPÖ episode, and the SPÖ's skeptical attitude toward the FPÖ—would suggest the same order. So, among the respondents in the treatment groups, the vignette boosted the probability ratings more if the vignette referred to a less likely scenario.
4.2 Measurement
Research on affective polarization has used several different items to measure the concept, including self-reports, implicit association tests, or behavioral measures such as trust or dictator games (Iyengar et al., Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019). We opted to follow the most prominent branch in the literature and operationalize affective polarization based on a feeling thermometer and two social distance measures. A feeling thermometer asks respondents to either rate a party or a party's voters on a scale from 0 to 10, where 0 means one does not like the party/the party's voters at all and 10 means one likes the party/the party's voters a lot. In the present study, we decided to explicitly ask respondents to rate parties’ voters in order to adequately address our research question whether elite's coalition signals can reduce affective polarization in a society (Druckman and Levendusky, Reference Druckman and Levendusky2019).Footnote 8 Explicitly asking respondents to rate parties’ voters has two additional benefits: first, we make sure that respondents are thinking about the same target group when assigning their scores (Iyengar et al., Reference Iyengar, Sood and Yphtach2012, 411) and second, it fits better with the two social distance measures that have shown to correlate less with the party version of the thermometer item (Druckman and Levendusky, Reference Druckman and Levendusky2019).
With respect to the social distance measures, we asked whether a marriage between a close family member and a voter of a certain party as well as whether having a close colleague at work who is a regular voter of a certain party would bother the respondent. Both social distance items used the same scale from 0 (would not bother me at all) to 10 (would bother me a lot). Descriptive statistics can be found in Appendix E.Footnote 9
For the analysis, we create an overall like–dislike index for each party out of these three variables. Higher values mean higher affective dislike toward the party's supporters. Cronbach's alpha for this additive index is 0.76, a satisfactory score. We also present results for the feeling thermometer and the social distance measures separately. This is because the correlation between the feeling thermometer and the social distance measures turned out to be comparatively weak ($r = 0.35$
for the ÖVP, $r = 0.28$
for the SPÖ, and $r = 0.49$
for the FPÖ), so we deviate from the pre-analysis plan here by presenting additional results.
In addition, we asked respondents to place the three parties on a 0–10 left–right scale after the experiment. For our analyses, we calculate the absolute distance between the two parties in the vignette as our hypothesis refers to the perceived convergence of the two parties. The order of perceptions of left–right positions and of affective distance questions was randomized, given that responses to the two sets of questions could influence each other. We include the order of questions as a control in our experimental analyses and present separate mediation analyses in Appendix P2.
4.3 Analytical approach
Our analysis requires us to classify respondents by party identification. We identify party supporters based on their stated party identification and, if none is given, their current vote intention; both measures were included pre-treatment. Note that over 70 percent of respondents report that they feel close to a political party, and we use vote intention to classify an additional ten percent of respondents, so that we assign over 80 percent of the sample to a partisan group.
The analysis of H1-H3 requires us to compare respondent evaluations of the key out-party in the vignette with equivalent respondent evaluations in the control group. To do so, we stack the data for the control group so that we have three observations per respondent, that is, party affect toward each of the three possible out-parties. We include (1) fixed effects for all combinations of target party and party identification and (2) clustered standard errors by respondents to account for this data structure. In Appendix J, we also present our main results when controlling for basic socio-demographics (age, gender, education [5 levels], region). The socio-demographic questions were asked at the end of the questionnaire. The size and statistical significance of our key coefficients only change due to the inclusion of these controls to a very limited extent. Additional analyses, included in the pre-analysis plan but presented only in Appendix L, make use of the strength of party identification and political sophistication (an index of interest, media use, and knowledge); these were measured pre-treatment.
5. Analysis
We first describe the overall results of our experiment (H1) before we turn to ideological distance (H2) and warm personal relations (H3). Finally, we examine effects on respondents who support neither party mentioned in the vignette (H4a/H4b).
5.1 Effects of coalition signals on in-party voters
Our first hypothesis is that partisans will feel more positively toward out-partisans if their party signals that it might form a coalition with the out-party. Our experiment provides support for this expectation. The left panel of Figure 2 shows that, among partisans, exposure to the vignette on potential coalition formation leads to a 0.26 decrease in dislike toward out-party supporters; this effect is statistically significant at the 0.1 level and is equivalent to about 10 percent of a standard deviation. This effect is clearest and strongest for the feeling thermometer, while it is much less clearly detected in social distance evaluations.Footnote 10
Figure 2. Effect of coalition signals on affective distance. Note: 90 and 95 percent confidence intervals shown. Full results are provided in Appendix G.
The right panel of Figure 2 then presents results for each potential coalition partner. The effect is clearest for the radical-right Freedom Party and weaker (as well as statistically not significant) for the two mainstream parties; note, however, that differences between the different treatment effects are not themselves statistically significant, so we cannot reject the null hypothesis that reactions to all three parties are the same. However, there is reasonably strong evidence that affect toward all three sets of out-party supporters is reduced by coalition signals, with social distance harder to shift. Appendix I also presents more detailed results for each inparty–outparty combination, which confirms that the effect is clearest for coalitions with the Freedom Party. This appendix also shows that Freedom Party supporters are, however, not more likely to adjust their affect when their party sends coalition signals toward another party.
The effects for the radical-right Freedom Party may be stronger due to weaker pretreatment effects (Slothuus, Reference Slothuus2016), but this difference couldalso be due to the fact that coalition signals may have less of an impact when the two parties are already closer together ideologically (Falcó-Gimeno and Muñoz, Reference Falcó-Gimeno and Muñoz2017). Since social distance toward the supporters of mainstream parties is always low, ceiling effects may also play a role. Overall, the outsider, almost pariah status of the Freedom Party (Van Spanje and Van Der Brug, Reference Van Spanje and Van Der Brug2007) may explain the larger impact of signals that include that party.Footnote 11
An additional question of relevance is whether coalition signals also influence affect toward one's in-party. It is conceivable that in-party affect might decline if one's party says it might work with disliked rivals, as discussed above. However, it is also likely that in-party affect is quite robust to such information, especially if we consider party identification as a social identity. We include this exploratory analysis, not included in our pre-analysis plan, in Appendix M. There, we show that there are no effects of coalition signals on in-party affective distance, so elite signals do not have clear effects on how citizens feel about their own in-group. It takes more than coalition signals to shape respondent affect toward supporters of the favored party.
5.2 The role of ideological distance and warm personal relations
What explains the effect of coalition signals on out-party affect? In H2, we hypothesized that coalition signals reduce affective dislike by changing perceptions of ideological distance between the in- and out-party. Hence, perceptions of distance would act as a mediator between coalition signals and affective dislike. The data provide only indicative evidence that the coalition signals treatment reduces perceptions of the ideological distance between the two parties in the vignette: while the effect is in the expected direction, it is small and not statistically significant (Appendix N). We run a causal mediation analysis (Hicks and Tingley, Reference Hicks and Tingley2011; Imai et al., Reference Imai, Keele, Tingley and Yamamoto2011) to test whether our evidence is consistent with the notion that coalition signals influence affective dislike by changing perceptions of ideological distance. The left panel in Figure 3 shows that the average conditional mediated effect is a 0.05 reduction, which also amount to an estimated 17 percent of the total effect of the vignette that is thus mediated by distance perceptions; however, it is important to note that this effect is not statistically significant.Footnote 12
Figure 3. Analyses: Left–right distance and personal relations. Note: 90 and 95 percent confidence intervals shown. Left panel shows average conditional mediated effect of coalition signals on affective distance via reduced left–right distance. Full results are provided in Appendix G.
We randomized the order of the mediator and the outcome, and additional analyses show that the evidence for distance as a mediator is clearer and stronger if left–right positions are asked first (Appendix P). To the extent that coalition signals influence affect through policy perceptions, policy needs to be at the forefront of people's minds.
In addition, we expected in H3 that elite signals might have a larger effect if they also signaled warm interpersonal relations. However, the right panel of Figure 3 shows that there is little difference in the effects of coalition signals depending on whether warm personal relations are described or not. If anything, coalition signals have a stronger effect without personal connections. This is also true if we look at party-level effects (Appendix I).
5.3 Effects of coalition signals on other voters
Next, we test H4a and H4b, so whether respondents for whom neither party in the vignette is their in-party react to coalition signals. First, our general result is that coalition signals have no overall effect on affective dislike for respondents for whom neither party is the in-party. Figure 4 shows that the effect, while slightly negative, is not statistically significant. However, the right panel shows that those without an in-party in the vignette do reduce dislike toward the radical-right Freedom Party if there are coalition signals toward that party; this effect is statistically significant at the 0.1 level. This illustrates how coalition signals can reduce the outsider status of radical-right partisans, for better or for worse.
Figure 4. Effect of coalition signals in out-party-only vignettes. Note: 90 and 95 percent confidence interval shown. Analysis for respondents whose favored party is not in the vignette. Full results are provided in Appendix G.
To test H4a and H4b, we first need to establish key blocs in Austrian party competition. Our survey points to three poles: the center–right, the radical right and the center-left, which is made up of supporters of the SPÖ, the Greens and the left-liberal NEOS. As Appendix F shows, affect for the Social Democrats among supporters of the Greens and NEOS is thus comparatively positive; this also applies to those who do not identify with a party but place themselves at or below 4 on the left–right scale is thus comparatively positive. We therefore focus on this ideological bloc for subsample analyses.
H4a is tested in Figure 5. Here, we subset the sample to these left-wing voters, that is, those who identify with the SPÖ, Greens or NEOS or are left-wing ideologically, and examine how they react to coalition signals between the ÖVP and the FPÖ. We can see that there is no consistent effect on any measure of affect, with effects toward a reduction in affective distance, if anything. So, a willingness to compromise among non-bloc parties is clearly not viewed negatively, even though both parties are out-groups. So, our evidence favors H4a, with some indication of positive reactions even among out-bloc partisans.
Figure 5. Effect of coalition signals between ÖVP and FPÖ, left-wing subsample. Note: 90 and 95 percent confidence interval shown. Analysis for respondents with left–right position smaller or equal to 4 or who support either the SPÖ, Greens or the left-liberal NEOS parties. Full results are provided in Appendix G.
Finally, we hypothesized (H4b) that effects, if they exist, may depend on whether one of the parties sending out coalition signals is in the same ideological bloc as the respondent. We again sub-set our analysis to the center-left ideological group. For these respondents, Figure 6 shows that assessments of Social Democrat partisans remain stable regardless of whether they send coalition signals that they will work with a right-wing party. This fits our results in Appendix M that respondents do not adapt their assessment of the in-party due to coalition signals. Additional analyses also show that this stability applies to SPÖ coalitions with either the center–right ÖVP or the radical-right FPÖ (Appendix O1).
Figure 6. Effect of coalition signals in out-party-only vignettes, left-wing subsample. Note: 90 and 95 percent confidence interval shown. Analysis for respondents with left–right position smaller or equal to 4 or who support either the Greens or the left-liberal NEOS parties. Full results are provided in Appendix G.
In contrast, the right-wing parties are viewed more positively among these voters in these vignettes. Here, the effect is a 0.50 reduction in the index version of affective dislike, also around 20 percent of a standard deviation; here, the effect is statistically significant at conventional levels. Hence, we find support for our hypothesis H4b concerning ideological blocs. Additional analyses in Appendix O2 show that this effect is similar for the center–right ÖVP and the radical-right FPÖ. Moreover, left-wing supporters do not respond as clearly to coalition signals between the ÖVP and FPÖ, so the key factor is the involvement of the ideologically related Social Democrats.
6. Discussion and conclusion
The results of our experiment show that when parties signal that they can cooperate with one another, their supporters feel less affective distance toward each other. This effect was particularly clear for mainstream party supporter evaluations of the radical-right party supporters. Moreover, we found that even those who support neither party sending coalition signals react to these, mainly by increasing their affect toward the party outside of their ideological bloc.
There are three potential reasons why the effect of coalition signals was clearest for the willingness to cooperate with the radical-right Freedom Party. First, coalitions with the radical right are less frequent, leading to weaker pre-treatment effects. In other words, for the Freedom Party our experiment contained more new information for respondents, and hence they were more likely to update their affective evaluations. Second, coalition signals with the radical right may contain particularly relevant information on the ideological distance of the party to its mainstream competitors. As the radical-right party is comparatively extreme, respondents could see coalition signals as implying party moderation. Finally, coalition signals with the radical right could contain more relevant information on the acceptability of this government constellation. At the national level, the Freedom Party still has a stronger outsider status, despite recent coalitions with the party. Overall, it is thus not surprising that our treatment had the clearest effect for the radical right.
Our results concerning the reasons for why coalition signals work are indicative, but inconclusive, and call for further work. On the one hand, we found some, albeit weak, evidence that coalition signals reduce affective distance by making potential coalition partners appear ideologically closer. It is likely that elite signals matter at least in part because of the ideological signals they contain, which is unsurprising given how closely ideological and affective distance are intertwined (e.g., Rogowski and Sutherland, Reference Rogowski and Sutherland2016; Orr and Huber, Reference Orr and Huber2020). However, we found no evidence that warm personal relations help to reduce affective distance, in contrast to prior findings (Huddy and Yair, Reference Huddy and Yair2021). Our experimental design did not include a potential third aspect, namely the re-categorization of in- and out-parties as a result of coalition formation (Lupu, Reference Lupu2012; Horne et al., Reference Horne, Adams and Gidron2023). Yet, this potential additional mechanism is plausible. Overall, we would call for additional research on the mechanisms underlying responses to coalition signals, and this research needs to pay particular attention to power considerations.
Our paper engages with and develops insights relevant for three sets of literatures. First, our paper provides a new perspective into the effects of coalitions and coalition signals on voters. Previous research has shown that participating in coalitions and showing the willingness to cooperate reduce perceived ideological distance (Fortunato and Stevenson, Reference Fortunato and Stevenson2013; Adams et al., Reference Adams, Ezrow and Wlezien2016; Falcó-Gimeno and Muñnozoz, Reference Falcó-Gimeno and Muñoz2017; Falcó-Gimeno and Fernandez-Vazquez, Reference Falcó-Gimeno and Fernandez-Vazquez2020; Hjermitslev, Reference Hjermitslev2022). We show that coalition signals have effects that go beyond voter perceptions of elite politics. Instead, they also affect how voters see each other and, potentially, how they interact with each other.
Second, for the literature on affective polarization, our paper shows that inter-party cooperation can have important consequences for how party supporters view each other. This finding is consistent with experimental work by Huddy and Yair (Reference Huddy and Yair2021) and Bassan-Nygate and Weiss (Reference Bassan-Nygate and Weiss2022) as well as observational work (Horne et al., Reference Horne, Adams and Gidron2023). Our findings clearly establish that insights from political behavior (Zaller, Reference Zaller1992) and social psychology Hogg (Reference Hogg2001) apply to partisan affective polarization. Moreover, we show that the impact of elite coalition signals reaches beyond the supporters of the two parties signaling cooperation. In many party systems, there will be other citizens who nevertheless feel somewhat warmly toward one of the two parties. These voters—often part of established, consistent ideological blocs (Bartolini and Mair, Reference Bartolini and Mair1990; Kekkonen and Ylä-Anttila, Reference Kekkonen and Ylä-Anttila2021; Reiljan and Ryan, Reference Reiljan and Ryan2021)—will react in a nuanced way to the elite coalition signals: they maintain affect toward supporters of the party they liked, even though their leaders want to work with disliked rival parties; but, they at the same time become less affectively distant from supporters of the parties they disliked.
Third, our paper provides unexpected insight into competitive dynamics in contemporary European party systems, particularly in reaction to the rise of radical competitors on the right (Abou-Chadi, Reference Abou-Chadi2016; Wagner and Meyer, Reference Wagner and Meyer2017). If, as our results indicate, coalition signals toward the radical right matter, then this illustrates that elite signals can be important in mainstreaming and legitimizing non-mainstream competitors. Bischof and Wagner (Reference Bischof and Wagner2020) find that the parliamentary entry of the radical right leads to increased societal polarization. In contrast, we show elite signals can in turn reduce this polarization, for better or worse. In this way, coalition signals parallel parliamentary representation's effect on the normalization of the radical right (Valentim, Reference Valentim2021). Future work should examine in more detail how elite signals can change democratic social norms and undermine cordon sanitaire strategies (Mudde, Reference Mudde2004; Van Spanje and Van Der Brug, Reference Van Spanje and Van Der Brug2007).
Overall, elite coalition signals have a systemic effect on how citizens see each other, demonstrating the societal impact of elite cooperation. Our results therefore provide insight into when and how affective polarization can be reduced. First, at a macro level countries that witness more interparty cooperation should be characterized by lower levels of affective polarization (Gidron et al., Reference Gidron, Adams and Horne2020). More narrowly, groups of voters whose favored parties work together a lot should also be less affectively distant. Second, affective polarization in one country may wax and wane as patterns of elite cooperation change. Elite cooperation across camps may be particularly important as such signals affect citizens across the political spectrum. It may also have a particularly strong impact when parties that are ideologically distant show that they can work together. Finally, we show that elite cooperation can even have effects among those who support neither of the two parties cooperating. While the effects are nuanced, these systemic effects of elite cooperation have so far gone unnoticed.
In sum, affective polarization can pose an important challenge to liberal democracies, but elites can contain this threat: while they can exacerbate affective polarization, they can also work to reduce it. However, reducing negative affect toward undemocratic competitors is not necessarily desirable, so signaling the possibility of cooperation with outparties is not always a justifiable strategy. When and toward whom negative outpartisan affect is a danger for democracy is thus a normative question that requires urgent debate.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2023.33. To obtain replication material for this article, https://doi.org/10.7910/DVN/IFKQBE
Acknowledgements
Katrin Praprotnik conducted this research under the auspices of the Austrian Democracy Lab (ADL), a cooperation with Forum Morgen. Markus Wagner's work on this article was partly funded by the European Research Council, grant number 101044069. This paper greatly benefited from feedback from: Ivo Bantel, Ida Hjermitslev, Thomas Meyer, seminar participants at the University of Konstanz, participants in the workshop “From political to affective polarization? Experimental evidence on the sources and implications of political polarization” at the University of Barcelona, December 2021, and participants of the workshop “Affective Polarization: A Challenge for Electoral Democracy?” at Sciences Po, Paris, May 2022.
Competing interest
None.
Open access
