1. Introduction
Research on affective polarization has shown that information about other people’s political leanings influences what we think of them and how we treat them. When people know that someone is a Democrat, opposes abortion, or supports Brexit, they also tend to infer other characteristics about that person. In other words, people hold stereotypes about others based on their partisan leanings or political views (Rothschild et al., Reference Rothschild, Howat, Shafranek and Busby2019; Hobolt et al., Reference Hobolt, Leeper and Tilley2021; Cassidy et al., Reference Cassidy, Hughes and Krendl2022). Knowing someone’s political leanings also affects how we treat them. People form homogeneous networks based on political views (Mutz, Reference Mutz2002; Mason, Reference Mason2018), and such networks often exacerbate polarisation (Hobolt et al., Reference Hobolt, Lawall and Tilley2024). People are also willing to discriminate against others because of their partisanship (Mason, Reference Mason2018; Kalmoe and Mason, Reference Kalmoe and Mason2022), even in social interactions devoid of political context (Gift and Gift, Reference Gift and Gift2015; Huber and Malhotra, Reference Huber and Malhotra2017). Shared partisanship – or the lack thereof – is a powerful social force (Iyengar and Westwood, Reference Iyengar and Westwood2015).
Political views and identities are, however, generally much more hidden than other important characteristics such as race, gender, weight, or age (Lee, Reference Lee2021; Wagner, Reference Wagner2024). This naturally applies to personal interactions: when using online dating profiles, for example, one may not know from first glance whether a potential partner shares one’s political identities. But it may also apply, to a more limited extent, to political figures: when reading a news headline about a political scandal concerning a less well-known politician on a social media feed, people may not know that politician’s partisan allegiance. Nevertheless, people may form expectations about the political stances of people they encounter usually based on other, more easily accessible characteristics. For example, some objects and activities are associated with certain parties and stances, such as guns, Nascar, Volvos, and craft beers (Hiaeshutter-Rice et al., Reference Hiaeshutter-Rice, Neuner and Soroka2023), as may be traits such as gender, religion, or occupation (Lerman and Sadin, Reference Lerman and Sadin2016; Ahler and Sood, Reference Ahler and Sood2018; Jones and Brewer, Reference Jones and Brewer2019; Goggin et al., Reference Goggin, Henderson and Theodoridis2020; Barber and Pope, Reference Barber and Pope2022; Titelman and Lauderdale, Reference Titelman and Lauderdale2023). Indeed, these ostensibly non-political signals may sometimes be intentional (Lee, Reference Lee2021; van der Does et al., Reference van der Does, Galesic, Dunivin and Smaldino2022).
In this paper, we examine how people project partisanship and political stances onto individuals in the absence of direct cues. We build on the long-standing literature on social projection, standard models of which predict that people expect others to be similar to themselves: people tend to universally assign their own characteristics and attitudes to others (Robbins and Krueger, Reference Robbins and Krueger2005; Krueger, Reference Krueger2007; Davis, Reference Davis2017). However, we depart from these standard models by arguing that the projection of political attitudes and identities will likely strongly depend on people’s valence evaluation of the target. Existing models of social projection suggest that this is merely weaker for low-valence targets (Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014), but we suggest that, for political evaluations, people will even engage in counter-projection, assigning disliked out-groups to targets with low or negative valence.
We expect projection to depend heavily on target valence because political debates and conflicts are characterized by deep group divisions (Mason, Reference Mason2018; Gidron et al., Reference Gidron, Adams and Horne2020; Harteveld, Reference Harteveld2021b; Kalmoe and Mason, Reference Kalmoe and Mason2022; Wagner, Reference Wagner2024) and strong moral framing that reduces conflict to a black-and-white division between good and evil (Akkerman et al., Reference Akkerman, Mudde and Zaslove2014; Garrett and Bankert, Reference Garrett and Bankert2020; Spinner-Halev and Theiss-Morse, Reference Spinner-Halev and Theiss-Morse2024). Social identity theory suggests that in-group identifiers will place a premium on preserving the in-group’s self-image (Tajfel and Turner, Reference Tajfel, Turner, Austin and Worchel1979; Marks, Reference Marks1984; Sedikides and Strube, Reference Sedikides, Strube and Zanna1997) to reduce cognitive dissonance (Aronson, Reference Aronson1969). We argue that, as a result, people will assign their own partisanship and political views to those they view positively and assign distant partisanship and political stances to those they view negatively. Put simply, citizens will likely assume that individuals they like and admire share their political views, while those they dislike and disapprove of will have opposing political views. Hence, it is likely that individuals engage in politically motivated projection.
We test our expectations using two original pre-registered experimental studies. Study 1 is a novel visual conjoint experiment (López Ortega and Radojevic, Reference López Ortega and Radojevic2025 Vecchiato and Munger, Reference Vecchiato and Munger2025) conducted in the USA and Britain. Respondents were shown images of fictional characters that vary in several characteristics, including whether they are heroes or villains. We then asked respondents to guess the characters’ likely partisanship and left-right ideological position. Our results show that respondents engage in motivated political projection: they believe that, independent of these characters’ socio-demographic characteristics, more heroic figures share their partisanship and ideological views, while more villainous figures are assigned opposing partisanship and ideological stances.
Given the stylized nature of this Study 1, Study 2 then uses a more realistic vignette experiment which, among other socio-demographic characteristics, manipulates the valence signals of a fictional political figure: respondents are presented with the scenario of a local politician who turned out to be either highly corrupt or highly virtuous. We assess whether respondents infer partisanship based on this valence signal, even though the partisan affiliation of the politician is never mentioned. Hence, we test whether political projection occurs even when respondents are given clear, easy opt-outs. Study 2 complements Study 1 by (1) providing a more externally valid (and political) setting, (2) building on a much weaker prompt for partisan projection, and (3) including a placebo comparison prompt. Together, these studies provide strong, cross-national causal evidence in support of our political projection thesis. In Study 1, projection is equally strong in the USA and Britain. Consistent with the expectations of social identity theory, both studies show that projection is particularly strong for those with more deeper-seated political identities. Moreover, we find evidence that counter-projection is in fact larger than projection, so respondents project out-party support onto “villains” more than they project in-party support onto “heroes.” We argue that this is likely because negative valence signals present a threat to the social value of the in-group, prompting counter-projection—or disidentification (Turnbull-Dugarte and López Ortega, Reference Turnbull-Dugarte and López Ortega2024)—to maintain the distinctiveness of the in-group.
Additionally, as part of a pre-registered exploratory analysis, we demonstrate that the political left is significantly more willing to engage in motivated political projection than the political right. This asymmetry, consistent with evidence of higher levels of negative out-group affect among the left (Ford, Reference Ford, Cowley and Ford2016) and the (increased) effect of partisan sorting on negative partisanship among the left (Hobolt et al., Reference Hobolt, Lawall and Tilley2024), is observed in both Britain and the USA. We suggest that this may be due to the higher levels of negative out-group affect among those on the left toward those on the right than that among those on the right toward the left.
Our results have at least four implications for research on affective polarization, especially as regards how political and social identities play out in everyday life. First, our research provides a different angle on how character traits and partisan identities relate. While previous research stresses that we expect partisans to have certain positive or negative traits (Carney et al., Reference Carney, Jost, Gosling and Potter2008; Johnston et al., Reference Johnston, Lavine and Federico2017; Rothschild et al., Reference Rothschild, Howat, Shafranek and Busby2019; Hobolt et al., Reference Hobolt, Leeper and Tilley2021), we show that such traits also lead to political projection of partisan identities. Second, recent work has shown that people can and do infer political views and partisan identities from demographic characteristics and lifestyle cues (Hiaeshutter-Rice et al., Reference Hiaeshutter-Rice, Neuner and Soroka2023; Titelman and Lauderdale, Reference Titelman and Lauderdale2023). Our work shows that valence assessments also provide grounds for inferring political characteristics. However, the political projection we describe differs from other kinds of inference: while the use of valence as a heuristic may partly be driven by an aim to arrive at accurate inferences, its use may also be driven by motivational purposes related to strengthening one’s own group image and sense of connectedness (Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014). Third, our research adds to work on the nature of outgroup perceptions (Ahler and Sood, Reference Ahler and Sood2018): political projection means that citizens will overestimate how politically similar they are to people they like and how politically different they are to the people they dislike. Finally, our findings are consistent with parts of existing research on affective polarisation (Iyengar and Westwood, Reference Iyengar and Westwood2015; Iyengar et al., Reference Iyengar, Konitzer and Kedin2018), which points toward out-group hate (over in-group love) as the driving force behind motivated projection. Specifically, we show that counter-projection is stronger than projection, so it is political dissimilarity in particular that will be overestimated, likely further consolidating negative out-group affect.
Political projection has likely implications for the dynamics of intergroup affect. We would expect patterns of overestimation to have potential negative consequences over time. If disliked people are assumed to be out-partisans, then partisan animosity is likely to be exacerbated via a cyclical Bayesian updating process: if negatively valenced people are categorized as out-group members, then we will also tend to associate more and more negative attributes with out-group members. In short, when “them” equates with “bad” and “bad” also equates with “them,” then a valence-based inferential model is likely to increase negative out-group affect, with potential downstream consequences for democratic cohesion. At the same time, political projection also points to a way of reducing negative affect: the flip side of frequent, easy political projection is that there are many projected political identities that can be easily corrected, and this may work to reduce affective polarization. We reflect more on the implications of our theory and findings in the conclusion.
2. Political projection
The formation of political attitudes and perceptions is strongly driven by individuals’ tendency to rely on inference, where one piece of information serves as a heuristic for other characteristics (Feldman and Conover, Reference Feldman and Conover1983). It is well-established that political ideology and partisanship serve as particularly strong informational cues: knowing someone’s political leanings generates many inferences about that person (Conover and Feldman, Reference Conover and Feldman1982). First, if told someone’s political allegiance, people develop other stereotypical beliefs about that person, for instance, concerning their personality and their background (Carney et al., Reference Carney, Jost, Gosling and Potter2008; Johnston et al., Reference Johnston, Lavine and Federico2017; Rothschild et al., Reference Rothschild, Howat, Shafranek and Busby2019). These stereotypical inferences need not be accurate, of course (Ahler and Sood, Reference Ahler and Sood2018). Second, if told someone’s political allegiance, people will often also change how much they like that person, so affective evaluations are changed by knowing people’s partisan leanings. Those who share one’s own views and political identities are viewed positively, while those with opposing political preferences and identities are viewed negatively (Mason, Reference Mason2018; Hobolt et al., Reference Hobolt, Leeper and Tilley2021). For instance, face impressions of disclosed partisans depend on whether the face evaluated shares one’s own identity (Cassidy et al., Reference Cassidy, Hughes and Krendl2022).Footnote 1 What this research has in common is that the starting point is that people know others’ political views or partisan allegiance and make inferences based on that.
However, in many contexts, we know a lot of things about people we are acquainted with long before, if ever, we find out about their political leanings. Unlike race or gender, ideology and partisanship are generally not visible personal characteristics. We will only rarely know someone’s partisanship before we first see their face, so scenarios like that manipulated by Cassidy et al. Reference Cassidy, Hughes and Krendl(2022) are exceptionally rare. An individual using an online dating application like Tinder or Grindr, for example, is unlikely to be explicitly informed of the political identity of potential romantic partners. Rather than making inferences about people based on their known politics, individuals more often infer people’s politics based on other known characteristics. We know from recent research that traits, personality, behavior, demographic characteristics, and objects can lead to inferences about political views and partisanship (Carney et al., Reference Carney, Jost, Gosling and Potter2008; Goggin et al., Reference Goggin, Henderson and Theodoridis2020; Lee, Reference Lee2021; Barber and Pope, Reference Barber and Pope2022; Hiaeshutter-Rice et al., Reference Hiaeshutter-Rice, Neuner and Soroka2023; Titelman and Lauderdale, Reference Titelman and Lauderdale2023), so people readily and frequently reach conclusions about the political characteristics of the people they meet or know, including religious Deities (Epley et al., Reference Epley, Converse, Delbosc, Monteleone and Cacioppo2009; Ross et al., Reference Ross, Lelkes and Russell2011).
The relevant research agenda in social psychology relates to social projection, which is “a process, or a set of processes, by which people come to expect others to be similar to themselves” (Robbins and Krueger, Reference Robbins and Krueger2005). As people know more about themselves than about others, they use the self as an anchor against which other target individuals are assessed (Krueger, Reference Krueger2007) and tend to assume that others are like them (Davis, Reference Davis2017), even if erroneously (Mullen et al., Reference Mullen, Atkins, Champion, Edwards, Hardy, Story and Vanderklok1985).
Such social projection is greater for in-group members (Clement and Krueger, Reference Clement and Krueger2002; Lerman and Sadin, Reference Lerman and Sadin2016) and other positively valenced targets (Robbins and Krueger, Reference Robbins and Krueger2005). People expect in-group members, who have positive valence, to share the good characteristics they themselves have (Castelli et al., Reference Castelli, Arcuri and Carraro2009; Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014). Conversely, we are also more likely to grant in-group membership to those with positive qualities (Leyens and Yzerbyt, Reference Leyens and Yzerbyt1992; Yzerbyt et al., Reference Yzerbyt, Leyens and Bellour1995). In contrast, projection has been found to be weak to non-existent for out-group members (Clement and Krueger, Reference Clement and Krueger2002; DiDonato et al., Reference DiDonato, Ullrich and Krueger2011). Applied to political characteristics, this implies that people will ascribe their own political views and identities to those with positive valence.
The reason why people engage in projection may be both cognitive and motivated. A cognitive account of projection would emphasize that it results from inductive reasoning or the use of heuristics, along the lines of “good targets have good characteristics” (Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014, p. 1374). If the person is good, and people who share one’s political identity are (at least in one’s own eyes) generally good, then it is a reasonable inference that the person is likely to share one’s political leanings. A motivational account would add that projection can also have other aims, beyond providing a reasonable heuristic (Yzerbyt et al., Reference Yzerbyt, Leyens and Bellour1995; Robbins and Krueger, Reference Robbins and Krueger2005; Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014). Political projection can help individuals to hold and maintain a positive image of the self (Marks, Reference Marks1984; Ames, Reference Ames2004) and reduce cognitive dissonance (Aronson, Reference Aronson1969), while protecting and enhancing the perceived positive valence of the in-group to which they belong (Tajfel, Reference Tajfel1974; Tajfel and Turner, Reference Tajfel, Turner, Austin and Worchel1979).
So far, we have treated projection as a process that is mostly about who belongs to one’s in-group. However, political projection may also be characterized by active efforts at “counter-projection,” with individuals seeing negatively valenced others not just as less like themselves, but as the opposite of themselves (Robbins and Krueger, Reference Robbins and Krueger2005; Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014; Davis, Reference Davis2017; Denning and Hodges, Reference Denning and Hodges2022; Turnbull-Dugarte and López Ortega, Reference Turnbull-Dugarte and López Ortega2024). Much in the same way that negative partisanship is shaped by out-group animosity as opposed to in-group affinity (Bankert, Reference Bankert2022; Lee et al., Reference Lee, Lelkes, Hawkins and Theodoridis2022), such counter-projection tends to occur when targets are disliked rather than simply evaluated neutrally (Machunsky et al., Reference Machunsky, Toma, Yzerbuyt and Corneille2014; Denning and Hodges, Reference Denning and Hodges2022). From a motivational perspective, counter-projection will augment the distinctiveness of the in-group compared to the negatively valenced out-group (Turner, Reference Turner1975; Turnbull-Dugarte and López Ortega, Reference Turnbull-Dugarte and López Ortega2024). In sum, people should ascribe out-group identities and views to those with negative valence.
The political realm is likely particularly conducive to the social projection conditional on valence evaluations. Politics is linked to moral reasoning and, in one account, different moral foundations (Garrett and Bankert, Reference Garrett and Bankert2020). Politics thus contains the potential for dualistic thinking where conflicts are devoid of nuance and (over-) simplified to a dichotomous divide between right and wrong (Akkerman et al., Reference Akkerman, Mudde and Zaslove2014). Partisan group conflict is also particularly strong, even compared to other deep-seated group divisions (Mason, Reference Mason2018). Moreover, politics is characterized by strong negative identities in addition to positive ones (Bankert, Reference Bankert2022; Lee et al., Reference Lee, Lelkes, Hawkins and Theodoridis2022; Lawall et al., Reference Lawall, Turnbull-Dugarte, Foos and Townsley2025; Areal, Reference Areal2024). Hence, projecting partisanship has a moral dimension, which is an area where counter-projection has been found to be particularly prominent (Denning and Hodges, Reference Denning and Hodges2022).
Hence, we expect that
(H1) Individuals will perceive positively valenced (virtuous) individuals to be members of their partisan in-group and negatively valenced (villainous) individuals to be members of the partisan out-group.
(H2) Individuals will perceive positively valenced (virtuous) individuals to be ideologically closer relative to their own ideological position than negatively valenced (villainous) individuals.
There are also likely to be individual differences in the extent to which people engage in political projection. The strength of in-group identification will likely play an important moderating role in the effects of group membership (Wann and Branscombe, Reference Wann and Branscombe1990; Mullin and Hogg, Reference Mullin and Hogg1998; Huddy, Reference Huddy2001). Westfall et al. (Reference Westfall, Van Boven, Chambers and Judd2015) and Mason (Reference Mason2018) show that the strength of partisanship correlates with the perceived partisan divide. Concerning social projection specifically, Crisp et al. Reference Crisp, Stathi, Turner and Husnu(2009) show that projection is greater for those with stronger in-group identities, in their case nationality (see also Riketta, Reference Riketta2005). In-group inclusion effects also increase together with the level of identification (Yzerbyt et al., Reference Yzerbyt, Leyens and Bellour1995; Castano et al., Reference Castano, Yzerbyt, Bourguignon and Seron2002). One reason for these patterns is likely to be that those with greater in-group identification also perceive a greater threat from the out-group (Stephan and Stephan, Reference Stephan, Stephan and Oskamp2000; Renström et al., Reference Renström, Bäck and Carroll2021; Cassidy et al., Reference Cassidy, Hughes and Krendl2022). Hence, we expect that
(H3) Projection and counter-projection effects will be higher among those with stronger in-group partisan attachments.
3. Study 1: Visual conjoint experiment
To assess our theoretical expectations, we first fielded an original pre-registered conjoint experiment (Study 1).Footnote 2 A conventional conjoint involves a fully randomized factorial design in which an individual respondent is exposed to numerous iterations of a forced-choice comparison between two profiles whose attribute values are fully and simultaneously randomised. In our visual conjoint experiment (López Ortega and Radojevic, Reference López Ortega and Radojevic2025; Vecchiato and Munger, Reference Vecchiato and Munger2025), profiles in each iteration were presented visually in the form of target images.
The experiment was completed by 3,200 respondents from the USA (1,600) and UK (1,600) via a representative, quota-based, sample reflecting the gender, age, education, and racial composition of each country’s population via Dynata (previously Survey Sampling International) in September 2022. Each respondent in our visual conjoint experiment completed seven forced comparisons, resulting in a total sample of 44,800 (and thus 22,400 observations per country).Footnote 3
3.1. Manipulating valence with heroes and villains
The visual target profiles we randomly presented are well-known fictional characters from popular cinematic franchises. Opting for fictional, yet widely recognizable, characters allows us to present target profiles that vary in a wide range of demographic attributes and whose villainous or heroic identity is clear, yet whose political identities are independent of those present and primed in the real world. It assesses whether individuals engage in motivated projection in a context devoid of any explicitly partisan information heuristics.
The fictional universes leveraged in our experiment—the Marvel Cinematic Universe (MCU), Disney, Harry Potter, Lord of the Rings, Game of Thrones, and Star Wars—are some of the most financially successful media franchises, enjoy widespread cross-party popularity (see Appendix Figure B.2), and, importantly, reach audiences from diverse political and demographic backgrounds (Lacina, Reference Lacina, Carnes and Goren2022).
Of note is that our virtuous and villainous characters vary on a wide array of attributes that are often used as informational heuristics regarding partisanship (Titelman and Lauderdale, Reference Titelman and Lauderdale2023) including, importantly, gender, age, accent, familiarity, as well as other non-observables associated with each franchise. Despite all being virtuous heroes, the MCU’s Ironman, Harry Potter’s Dobby the House-Elf, Disney’s Aladdin or Sleeping Beauty, and Game of Thrones’ Arya Stark are, for example, of observably distinct socio-economic backgrounds. The characters were selected in order to be balanced on these other visible characteristics. In other words, there is an equal proportion of men/women characters that are heroes/villains and an equal proportion from each franchise that are heroes/villains.
Before running Study 1, we conducted a validation test that had two objectives. First, we wished to assess to what extent characters’ objective position as hero or villain translated into equivalent valence perceptions among respondents. In other words, even though Ironman is presented in the MCU as a hero, and Scar is presented in Disney’s Lion King as a villain, are these targets perceived as such? Second, given the primary focus of the experiment on ascertaining the propensity to project one’s in-group identity or counter-project out-group identities on different targets based on their perceived valence, it was essential to assess if there were significant partisan asymmetries between who one perceives as “good” or “evil.”
We fielded the preliminary validation test among a convenience sample from the USA and Britain using Prolific. The partisan make-up and demographic composition of the respondents included in the validation are reported in Appendix B. In both country samples, 80% of respondents were made up of an equal proportion of partisan identifiers from the two main political parties, with the remaining 20% identifying as independents and/or third-party supporters.
Respondents in our validation study essentially served as a means of crowd-sourcing the validity of the primary explanatory variable. In the validation test, respondents were asked to complete a conjoint task in which they reported the extent to which they believed that the presented characters were pure heroes or pure villains. The results of the validation (N = 12,784) are reported in Appendix B. Of the 86 characters included in our visual conjoint, 85 were validated. The only exception is the “(Evil) Fairy Godmother” from Shrek. Of core importance for the validity of our design, the probability that different partisans perceive the target as positively or negatively valenced is uniform.
3.2. Outcome measure: in-group and out-group membership
There are two outcome variables in Study 1. First, we model the propensity of respondents to project their own partisan identity, or counter-project out-party identities, onto fictional characters. The forced choice component of the conjoint experiment asked respondents Which character do you think is more likely to be a [Democratic/Republican]/ [Labour/Conservative] voter? The party presented in the question (see Figure 1) was randomized between respondents but remained constant across individuals’ iterations of the conjoint task.Footnote 4 Relying on an individual’s own expressed partisanship recorded pre-treatment, we identify if the identity projected onto a character reflects that of the respondent (1) or not (0).
Examples of experimental forced comparison.
Second, we measure ideological (dis)identification based on whether respondent’s self-reported ideological identities as liberals (left) and conservatives (right) match those they assigned to the target profiles they are presented with. In simple terms, our second outcome measures if respondents projected their own ideological position onto profiles (1) or not (0).
3.3. Results: valence signals and projection
The results of our empirical test of (H1) and (H2) are visually summarized in Figure 2, which reports the marginal mean of a target’s position as hero or villain, marginalizing across individual character fixed effects, on the probability of projecting partisanship (upper panel) or ideological proximity (lower panel). The reported marginal means indicate the mean probability that an individual respondent projects their own identity onto the experimentally presented characters that are either positively (heroes) or negatively (villains) valenced. The marginal mean of partisan projection for each individual target experimentally presented is reported in Figure A.1. The effects of additional conjoint attribute values (e.g. gender, franchise, etc) are also reported in the Appendix (see Figure A.2 and Figure A.3). Given the nested and non-independent nature of observations—multiple profiles evaluated by individual respondents—estimates are computed based on respondent-level clustered standard errors.
Social projection of political identities (Study 1).
Regardless of whether we consider estimates from the USA or Britain, our results are consistent: individuals are significantly more inclined to project their own partisan identities onto heroes and those of the partisan out-group onto villains. These results are remarkably consistent across the vast catalog of experimental targets and their diverse and varied socio-demographic characteristics. In real terms, citizens are 20 percentage points more inclined to project their partisan identities onto heroes than they are to do the same for villains. Given a baseline rate of in-group partisan projection among villains of 40%, the 20-point shift in the probability of partisan affinities being projected is substantive and equates to a 50% increase.
Consistent with our hypotheses, similar politically motivated projection biases are observed in the case of ideological identities (H2). The lower panel of Figure 2 demonstrates that individuals engage in active counter-projection with negatively valenced characters, projecting an ideological identity onto these targets that is significantly distinct from their own. Conversely, respondents project congruent ideological positions onto heroes. In the USA, for example, an average citizen is likely to perceive a villainous personality as sharing the same liberal-conservative identity as themselves 39% of the time, whereas, on average, they project a matching ideological identity onto virtuous personalities 60% of the time. Similar spatially divergent projection patterns between respondents and positively valenced or negatively valenced targets are observed in Britain. The effect is, however, somewhat smaller: while in the USA, the ideological projection bias equates to 20 percentage points, this divergence is 40% smaller at 12 percentage points in Britain.
In a pre-registered test of the moderating role of the strength of party attachments, we find that the more strongly one identifies with one’s party, the more likely one is to perceive heroes as being one of “us” and villains as belonging to the other side (Figure 3). This is true of party identities as well as ideological proximity. The difference in party projection between heroes and villains among those with the weakest identity attachment is 13 percentage points, whereas among those with the strongest identities it is 21 percentage points. The difference in these differences (eight percentage-points) is statistically significant and supports (H3): partisanship strength significantly moderates political projection. Similar patterns are found looking at ideological proximity.
Projection among strong and weak partisans (Study 1).
3.4. Discussion
Our visual conjoint experiment provides strong evidence of political projection. On average, citizens assign their in-group identities onto heroes and counter-project out-group identities onto villains. Moreover, the magnitude of projection is greater for those with stronger partisan identities. These findings are consistent with social identity theory and support the notion that political projection is motivational as opposed to being simply a means of reducing cognitive costs. As we demonstrate in various robustness tests included in the Appendix, these results are not conditional on the level of knowledge or familiarity with the targets presented in the visual experiment (see Figure C.2).
Study 1 has two potential limitations. First, while our hero-villain treatment provides a novel test of projection, this level of fictional abstraction does not allow us to say how projection may play out in conventional social or political scenarios. We have strong internal validity of the psychological processes, but less external validity of its potential applications. Second, Study 1’s design encouraged individuals to engage in projection by asking respondents to make an inference about a target’s political identity. In the absence of such a prompt, do such motivated inferences about a subject’s partisan identity still occur? We turn to Study 2 to address these limitations.
4. Study 2: Vignette experiment
In Study 2, we build on the robust empirical support for our theory provided by the visual conjoint experiment via a pre-registered vignette experiment among a panel of online survey respondents (N = 1,617) in Britain sourced from Prolific in October 2023. Respondents were surveyed from a quota-based sample that reflects population parameters based on gender, age, and education (descriptive statistics reported in Appendix D.1). Study 2 addresses the two limitations of Study 1. First, Study 2 provides external validity to our thesis by testing if projection effects are observed in a political setting that seeks to replicate the valence signals that individuals may encounter in the real world (Rudolph and Hetherington, Reference Rudolph and Hetherington2021). Second, it assesses whether projection occurs with a much weaker prompt or if it only occurs when people are strongly encouraged to make such inferences.
Our experimental design manipulated respondent exposure to one of two conditions which describe the actions of a target. In one condition (negative valence), the survey vignette describes the target stealing money from a local charity. In the alternative condition (positive valence), the vignette describes the target donating money to the same local charity. In neither condition are individuals informed of the partisan affiliation of the target. Empirically, we sought to assess if, in the absence of such information, respondents project partisanship by falsely recalling partisanship in a motivated manner based on the valence signals attached to the subject.
The vignette texts are detailed in Figure 4 and were designed to maximize covariate control (Dafoe et al., Reference Dafoe, Zhang and Caughey2018). Manipulation checks, reported in Appendix D.3, confirm that our experimental manipulation of valence signals worked as intended. Several specific design considerations are worth mentioning. First, and in addition to the randomization to the negative or positive valence vignette, within each vignette, we randomized several subject attributes. The rationale for this is that, as different socio-demographic characteristics (e.g., jobs or age) are associated with certain parties—and voters are aware of these associations (Titelman and Lauderdale, Reference Titelman and Lauderdale2023)—it was important to make sure that the average treatment effect of our manipulation of valence was independent of adjacent subject characteristics, which may result in unobserved confounding inferences (Dafoe et al., Reference Dafoe, Zhang and Caughey2018). We simultaneously manipulated the following attributes of the target: name and surname, age, past occupation, the amount of money stolen/donated, the item purchased with the money (negative condition only), and the name of one of the victims. We did not pre-register any expectations on the effect of these characteristics—the values of which are summarised in appendix material—as their inclusion was solely to test for the independence of our main estimate of interest.
Treatment conditions (Study 2).
Second, the vignette was communicated to respondents across three pages (note the {page X} indications in the vignette examples). Given our primary dependent variable is based on motivated asymmetries in recall, we divided the vignette over three pages to increase the baseline probability that a respondent may believe they missed a piece of information on one of the pages. Doing so consequently increases the difficulty of our empirical test.
4.1. Outcome measure: projection via motivated recall
After the vignette, respondents were asked recall questions about the text. Four of these asked about information that was contained in the vignette. For these questions, levels of accurate responses were high: 46% correctly answered all four questions and 82% responded correctly to three of four.
The key outcomes are two additional recall questions asking about information that was absent from the treatment. The first is our core dependent variable: What political party was [NAME] a councillor for? The response items were Labour, Conservative, “Don’t know,” or “Don’t remember seeing this information.” The order of the two parties was randomized. We included the two non-party responses to make this prompt as weak as possible. The same non-response options were included in all recall questions. They provide respondents with easy, face-saving ways of showing that they do not know the partisanship of the fictional local politician. Indeed, 84.6% of respondents chose one of these options.
If respondents selected one of the two non-response items, they were then prompted to nevertheless make an guess based on the available information. This second question is more similar to the task in Study 1. This two-part approach allows us to measure organic projection via motivated false recall as well as, similar to Study 1, the prevalence of projection when prompted.
We also included a second false recall question that served as a placebo item: How many kids did [NAME] have? In addition to five numerical options, respondents had the same two non-response items as the party question.
As in Study 1, we operationalize projection dichotomously when the inferred political identity of the target matches that of the respondent’s in-group identity (1) or that of the respondent’s out-group identity (0).
4.2. Results: real-world valence signals and (organic) projection
In Study 2, we sought to replicate the findings from Study 1 in a setting closer to the real world and, more importantly, assess if the social projection of political identities also occurs organically when encouragement to do so is very weak. The results of Study 2, summarized in Figure 5, show that political projection follows similar patterns as in Study 1 and occurs even for very weak prompts.
Modeling projecting via false recall (Study 2).
Consider the left-hand panel of Figure 5 which reports the probability that respondents in each of the treatment conditions assign—by falsely recalling in a motivated manner—their in-group political identity to the target individual. Around one in six individuals (15.4%) assigned a partisan identity to the target despite partisanship being absent from the vignette text. This effect was slightly (but not significantly) larger among strong party identifiers at 20%. As theorized, and independent of the simultaneously randomized characteristics of the target, the assignment of identities was significantly determined by projection effects. Those in the positive valence condition project their partisan identity onto the target with a probability of 0.6, whereas those in the negative valence condition do so with a probability of .26. Given this baseline of .26, a treatment effect of .34 translates into a sizeable causal increase of 131%.
Note that these effects, like those observed in Study 1, point toward counter-projection—defining ourselves by who we are not, as opposed to who we are (Turnbull-Dugarte and López Ortega, Reference Turnbull-Dugarte and López Ortega2024)—as a core driver in the politically motivated projection effects observed. Respondents exposed to the positive valence condition believe that positively valenced targets may belong to the out-group 40% of the time. In short: while they believe positively valenced individuals are more likely to belong to the in-group (.6) than the out-group (.4), they accept that a sizeable proportion of virtuous targets are not one of “us.” Conversely, however, respondents identify negatively valenced individuals as belonging to the out-group 74% of the time. The implication of this disparity signals that individuals are more prone to disidentify from and counter-project the out-group identity onto negatively valenced targets, than they are to identify with and project their own identity onto positively valenced targets.
The right-hand panel of Figure 5 models the effects of the valence treatments among all respondents, including those who did not project organically in response to the first prompt. As visualized, when individuals are explicitly asked to infer the partisanship of target individuals, valence signals condition their in-group projection and out-group counter-projection. The variation in the probability to project one’s identity onto a target is 53 percentage points higher for a positively valenced target compared to a negatively valenced target. The causal effect of positive valence equates to an increase in excess of 300% compared to the negative valence baseline of 0.22. As in the case of the spontaneous projection effects reported in the left-hand panel, the strongly prompted projection effects also indicate asymmetries between projection and counter-projection. However, this asymmetry is much smaller than that observed in the weak prompt. Respondents are more likely to counter-project out-group identities onto villains (78% of the time) than they are to project in-group identities onto heroes (75%) of the time.
In Figure 6, we demonstrate that the political projection effects we observe, for both types of prompt, are unique to salient identities. First, we show that propensity to make any false recall is significantly (p < 0.001) larger for our core (identity) measure (15.4%) than it is for the placebo item related to the number of children (12.2%). What this means is that, in addition to the motivated nature of projection in terms of its direction, false recall is itself motivational. In real terms, the three-point difference we observe in our core political item and our apolitical placebo item equates to a sizable and substantive change of 20% in the propensity to engage in recall. Note that, as we show in Appendix Table D.6, engagement in false recall is not greater among any theoretically relevant, and observable, covariates.
Modeling projection effects on placebo items (Study 2).
Second, we show that valence signals have no effect on our placebo item regarding the target’s children. Regardless of whether we consider those who provided a weakly prompted false recall response or those who responded only when strongly prompted, we find no identifiable variation in the inferred number of children that respondents associated with the target. The motivated projection that we observe in the case of political identities is, therefore, not a function of treatment exposure engendering more widespread projection across items, but rather signals concrete and politically motivated projection of political identities.
5. Exploratory analysis: partisan asymmetries
Finally, and as part of our pre-registered exploratory analysis, we examine whether political projection is similar in magnitude across partisans from the left and right. While we did not hypothesize any asymmetry between parties, the results demonstrate that not all partisans are equally prone to political projection.
Consider Figure 7, which visualizes the projection effects between the two primary partisan groups in the USA from the visual conjoint experiment in Study 1 (estimates for Britain are reported in Appendix Figure C.3). The upper panel reports the pairwise difference in the propensity of Democrats and Republicans to project their partisan identity (panel A) and ideological identity (panel B) onto heroes and villains. In both cases, Democrats engage in higher levels of projection than Republicans. In the case of ideological projection, where individuals’ responses are not coerced between a binary alternative, we observe little party variation with Democrats being only marginally (2 percentage points) more likely than Republicans to view heroes as ideologically approximate to them. Where the parties diverge, however, is in the extent to which they counter-project out-group identities onto villains: Democrats ideologically disidentify more from villains than they identify with heroes and do so with a probability that is 28 points larger than Republicans. As demonstrated earlier, counter-projection is greater than projection and the difference between parties appears to be a function of the left counter-projecting out-group identities onto villainous targets rather than the left projecting heroes as more approximate to themselves. These patterns are observed in both the US and British data from Study 1.
Partisan differences (Study 1).
Replicating this partisan-based subgroup heterogeneity test using the data from Study 2 provides remarkably similar results. The increased projection and counter-projection effects among Democratic and Labour partisans vis-à-vis their right-wing counterparts are not, therefore, a function of the design used in Study 1. Despite both Conservative and Labour partisans responding identically to the valence signals manipulated in Study 2 (as reported in Figure D.2, views on the villainous/heroic nature of the target were qualitatively and statistically indistinguishable), the effect of a positive signal for Labour partisans (.72) is almost five times (480%) that of Conservative partisans (.15).
The results of our exploratory partisan subgroup analyses demonstrate that in the case of both Study 1 and Study 2, and in both Britain and the USA, respondents who identify with parties associated with the liberal left are, on average, far more inclined to identify villains as out-group partisans and heroes as in-group partisans than respondents on the conservative right. The results of this pre-registered exploratory test, while not formally theorized, are congruent with Amira Reference Amira(2018) and empirical observations of the relative levels of affect between left and right-wing partisans (Ford, Reference Ford, Cowley and Ford2016, Mason, Reference Mason2018; Hobolt et al., Reference Hobolt, Lawall and Tilley2024) or liberal and conservative issue-based identities (Hobolt et al., Reference Hobolt, Leeper and Tilley2021; Wagner and Eberl, Reference Wagner and Eberl2024).Footnote 5 Observational evidence also points to the left, who are conventionally more tolerant of diverse social groups, being more politically intolerant than the right (Ford, Reference Ford, Cowley and Ford2016).Footnote 6
Indeed, assessing imbalances in the distribution of self-reported affect toward partisan out-groups in our original data from Study 1 and Study 2, we observe descriptive patterns consistent with these findings (see Figure 7.C, and 8.B).Footnote 7 We further corroborate this by looking at publicly available observational data from the British Election Study (see Appendix Table F.1) independent of our original data collection, which demonstrates that Labour partisans are more inclined to socially discriminate against Conservative voters than Conservatives are to socially discriminate against Labour voters. Given this asymmetry, and in addition to the evidence of the relatively stronger effects of counter-projection compared to projection, the moderating role of out-group affect i conditions the propensity of respondents to engage in the motivated social projection of political identities. Our exploratory results provide evidence in support of this thesis: projection is significantly larger among those with higher levels of negative out-group affect (see Figures 7 and 8). In the case of Study 1, where ideological propinquity allows for a richer comparison of counter-projection, we see that counter-projection—projecting out-group identities onto villains—is indeed what drives the divergence in projection effects between treatment conditions among those with higher levels of negative affect toward the political out-group.
Partisan differences (Study 2).
6. Conclusion
Our combined multi-study experimental designs provide strong, comparative evidence of political projection. On average, citizens assign their in-group identities onto those perceived as virtuous and counter-project out-group identities onto those perceived as villainous. Moreover, and in line with the expectations of social identity theory, we observe that the magnitude of this projection behavior is greater for those with stronger political identities as well as for those who are strongly and negatively predisposed toward the out-group. Importantly, and as we empirically test in our novel false-recall vignette experiment, this willingness to assign identities is likely not uncommon in real-world scenarios.
Together, our unique visual conjoint (Study 1) and vignette experiment (Study 2) show that a dominant feature of social projection of political characteristics is its moderation by target valence. Projection is, therefore, more (politically) motivational than cognitive. Compared to most standard domains where social projection occurs, politics appears to lend itself particularly to counter-projection. This may be related to the inherently moral nature of political debates as well as the strong group-centered nature of political conflict (Garrett and Bankert, Reference Garrett and Bankert2020). There is significant evidence of the importance of negative, as opposed to positive, political identities (Lee et al., Reference Lee, Lelkes, Hawkins and Theodoridis2022; Lawall et al., Reference Lawall, Turnbull-Dugarte, Foos and Townsley2025 Areal, Reference Areal2024) for understanding group politics and partisanship, and such negative identities may further explain the tendency toward counter-projection. In a context where affective polarisation is high, projection appears to be driven more about defining who we (and the group) are not, as opposed to who we are.
Our findings provide guidance as to when to expect political projection to occur. As Study 2 shows, people will engage in political projection as soon as they are given the opportunity to try to classify individuals along partisan lines. Transferring this to everyday contexts, we can imagine that political projection will be high in the run-up to contentious elections and during crises that divide citizens, such as Brexit, Covid-19, or indeed conflict and wars. While our empirical application focuses on partisan identities, we are confident that the theoretical assumptions of motivated projection are equally transferable to non-partisan issue identities (Hobolt et al., Reference Hobolt, Leeper and Tilley2021; Lawall et al., Reference Lawall, Turnbull-Dugarte, Foos and Townsley2025; Wagner and Eberl, Reference Wagner and Eberl2024) or even apolitical social identities. One can easily envisage, for example, how motivated projection processes could be observed in scenarios where inter-group dynamics are also salient such as between rival sport teams (Whigham, Reference Whigham2014; Lehr et al., Reference Lehr, Ferreira and Banaji2019) or indeed between antagonistic religious groups (Borgeson and Valeri, Reference Borgeson and Valeri2007).
Finally, Britain and the USA are relatively polarized contexts with small party systems (Denning and Hodges, Reference Denning and Hodges2022), although comparative evidence shows that levels of affective polarization are similar in other European countries (Reiljan, Reference Reiljan2020; Wagner, Reference Wagner2021). Nevertheless, it would be important to extend this work to other countries with larger party systems (Wagner, Reference Wagner2021; Harteveld, Reference Harteveld2021a) or those with more complex, less stable party systems where cooperation and compromise are more common (Horne et al., Reference Horne, Adams and Gidron2023) and partisan identities might be less salient. Yet, existing evidence indicates that similar polarizing, identity-based dynamics might be found in multiparty contexts as well (Huddy et al., Reference Huddy, Bankert and Davies2018).
Our novel design in Study 2 demonstrated that individuals spontaneously associate certain characteristics with partisan and political identities: one in six respondents engaged in projection—and more importantly negatively focused counter-projection—despite being given clear face-saving opt-outs. There is no non-arbitrary benchmark against which we can consider whether the level of organic political projection that we identify is politically significant. A positive interpretation of this level of projection is that it is reassuringly low: the modal citizen does not project when unprompted to do so, and, as a result, the potential knock-on effects related to inter-group dynamics and stereotypes may well be limited. A more pessimistic interpretation, however, is that one in six people, while far from a majority, is not a trivial or negligible proportion. One important task for future research is to uncover the role of projection in varying information environments. On the one hand, in many contexts, people may have more valid heuristics for inferring partisanship and political views, so political projection may not always be based on our simple valence heuristic. On the other hand, our design allowed for clear opt-outs in order to isolate projection without nudges to do so. Moreover, many social scenarios in the real world, and in particular during periods of heightened affective polarisation, actively encourage political group-based sorting and political projection. Indeed, existing work (e.g., Lee, Reference Lee2021) indicates that such inferences are frequent, and prevalent trends of social sorting along political identities (Mason, Reference Mason2018; Harteveld, Reference Harteveld2021b) may reinforce the accuracy of these inferences. Overall, more research is needed to assess when and where identity projection processes take place.
Such political projection is potentially worrying when it leads to a feedback loop. It is well established that knowing that someone you know is one of “them” makes you think less of them (Mason, Reference Mason2018; Gidron et al., Reference Gidron, Adams and Horne2020; Hobolt et al., Reference Hobolt, Leeper and Tilley2021). But we empirically show that people assume someone they think less of must also be one of “them.” This may well mean that we will then tend to link the positive traits and actions we observe in people we like with the in-party and negative ones with the out-party. In other words, the social projection of political identities may well result in a self-fulfilling prophecy whereby a Bayesian updating process reinforces out-group biases. Given that such deep-seated negative group-based political affect has consequences for political campaigning (Lawall et al., Reference Lawall, Turnbull-Dugarte, Foos and Townsley2025) and may, as some argue, ultimately lead to willingness to engage in political violence (Kalmoe and Mason, Reference Kalmoe and Mason2022), the potential implications of the self-reinforcing nature of political projection effects are likely far from trivial. We see this as an urgent area for future research.
At the same time, our results may well offer a potential path forward toward reducing affective dislike of out-party supporters (Levendusky, Reference Levendusky2018). We show that there is a strong tendency toward political projection. However, such projection will likely overshoot, so that there will be greater diversity among in- and out-party individuals than we expect (Ahler and Sood, Reference Ahler and Sood2018). This can be used to reduce out-group dislike. For one, individuals could be encouraged to rethink tendencies to engage in projection. Measures to correct misconceptions, for which there is already encouraging evidence (Ahler and Sood, Reference Ahler and Sood2018; Mernyk et al., Reference Mernyk, Pink, Druckman and Willer2022; Voelkel et al., Reference Voelkel, Chu, Stagnaro, Mernyk, Redekopp, Pink, Druckman, Rand and Willer2023), are also likely to succeed as reality will always tend to be more complex and nuanced than our biases predict. To understand and to reduce intergroup tension based on political identities, it is therefore important to know that people tend to engage in motivated political projection on people whose political leanings are not known.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2025.10.
Replication materials
Data and replication code are available via the Political Science Research & Method Dataverse here: https://doi.org/10.7910/DVN/X2WKUZ.
Acknowledgements
We acknowledge the support from the University of Southampton for financing this research. Markus acknowledges financial support from European Research Council (grant 101044069). The paper benefited from participant comments at seminars hosted at Nuffield College at the University of Oxford, Royal Holloway University of London, Stockholm University, the University of Manchester, the University of Southampton, the University of Strathclyde, and Vrije Universiteit Amsterdam. We are indebted to Alberto López Ortega, Alexander Dalheimer, Ben Ansell, Chris Hanretty, Daniel Devine, Elena Heinz, Federica Genovese, Jack Bailey, Jack Blumenau, James Tilley, Joseph Noonan, Julie Hassing Nielson, Kaat Smets, Kåre Vernby, Katharina Lawall, Lawrence McKay, Lotte Hargrave, Lucy Barnes, Manuel Sola, Maria Sobolewska, Mariken van der Velden, Ming Boyer, Rachel Bernhard, Rob Johns, Ronja Sczepanski, Sara Hobolt, Sarah Wagner, Sascha Riaz, Semih Çakir, Stefanie Reher, Tarik Abou-Chadi, Viktor Valgardsson, Will Jennings, and Zac Greene for their detailed comments and feedback on earlier iterations of the paper.
Open access
