Introduction
Populist parties and candidates have been successful with a variety of policy programs ranging from the radical right to centrist, post-materialist, and single-issue platforms to the radical left. Yet, from the perspective of most scholars, what unites these parties, despite their programmatic differences, is a common thin-centered populist ideology (Mudde Reference Mudde2004, Reference Mudde2007). This ideology is most tangible in populist politicians’ rhetoric in the public realm. Populist arguments are characterized by appeals to a fictitious group of ‘the people’ and strongly worded criticism of ‘conspiring elites’, which results in a Manichean worldview of politics as a constant fight between good and evil, ‘the people’ and ‘the elites’ (Mudde Reference Mudde2004). This rhetoric is commonly referred to as ‘thin populism’, which then interacts with ‘thicker’ ideologies (such as nativism or socialism) and policies (Mudde Reference Mudde2007). Not only scholars (for example, Canovan Reference Canovan1982, Reference Canovan1999; de Vreese et al. Reference de Vreese, Esser, Toril, Reinemann and Stanyer2018; Appel Reference Appel2018; Moffitt Reference Moffitt2016; Mercieca Reference Mercieca2015; Kazin Reference Kazin2016) but also many journalists (for example, Fisher Reference Fisher2017; Goldhill Reference Goldhill2017; Healy and Haberman Reference Healy and Haberman2015; Giridharadas Reference Giridharadas2016) have speculated to what extent this rhetoric might explain populists’ electoral success. Some, for example, argue populist politicians are the ‘most effective practitioners of persuasive rhetoric the political world has seen in a long time’ (Romm Reference Romm2016).
Previous research on populist persuasion has focused on populism as ideological stances rather than rhetoric (for example, Neuner and Wratil Reference Neuner and Wratil2022; Castanho Silva et al. Reference Castanho Silva, Neuner and Wratil2023), on populist narratives about societal problems (Busby et al. Reference Busby, Gubler and Hawkins2019), and a country’s past (Elçi Reference Elçi2022), or populism in the media (Bos et al. Reference Bos, van der Brug and de Vreese2013; Hameleers et al. Reference Hameleers, Bos and de Vreese2017; Hameleers and Fawzi Reference Hameleers and Fawzi2020; Sheets et al. Reference Sheets, Bos and Boomgaarden2016, Müller et al. Reference Müller, Schemer, Wettstein, Schulz, Wirz, Engesser and Wirth2017). Many of these studies are predominantly concerned with outcomes that are only indirectly linked to electoral success, like group identities (Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020), blame attribution (Hameleers et al. Reference Hameleers, Bos and de Vreese2017, Reference Hameleers, Bos and de Vreese2018), support for anti-democratic executive action (Bessen, Reference Bessen2024), voters’ perceptions of out-groups (Hameleers and Fawzi Reference Hameleers and Fawzi2020; Hameleers and Schmuck Reference Hameleers and Schmuck2017), or populist attitudes (Elçi Reference Elçi2022). Existing work either does not operationalize populism as a speech act, the outcome measured is not directly linked to populists’ electoral success, or it does not disentangle populist rhetoric from ‘thick’ ideologies (but see Dai and Kustov Reference Dai and Kustov2024). Here, we test populism as a particular type of rhetoric in a political argument, identify its effects on electorally relevant outcomes, and isolate its effect from the effect of the argument’s substantive position.
We explore two routes for populist arguments to affect electoral outcomes, one indirectly via issue preferences and one directly via candidate support. First, populist arguments could be more successful in changing voters’ issue preferences than non-populist arguments. Presenting, for example, a health care reform as the ‘will of the people’ and condemning ‘the establishment’ for long waitlists might make voters more supportive of the policy than an argument presented without such populist elements. The changed issue preferences could then translate into support for politicians who campaign on the issue. Second, populist arguments may directly make political candidates appear more appealing. A candidate who praises ‘the people’ and condemns ‘the establishment’ might simply appear more electable to voters. They might change voters’ candidate support more than non-populist arguments.
Theoretically, we combine well-established theories from the framing (for example, Entman Reference Entman1993; Chong and Druckman Reference Chong and Druckman2007) and public opinion literature (for example, Converse Reference Converse1964; Zaller Reference Zaller1992) with recent work on populist political communication that argues populist rhetoric to function as an ‘identity frame’ (Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020; Hameleers et al. Reference Hameleers, Reinemann, Schmuck, Fawzi, Reinemann, Stanyer, Aalberg, Esser and de Vreese2019; Hameleers et al. Reference Hameleers, Schmuck, Schulz, Wirz, Matthes, Bos and Andreadis2021). As such, populist rhetoric is essentially a ‘heuristic’ (Sniderman et al. Reference Sniderman1991) that ‘invites’ voters to identify with the glorified group of ‘the people’ (the ‘in-group’) while ascribing various negative qualities to ‘the elites’ (the ‘out-group’) (Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020). On this basis, we argue that different types of priors should moderate the effects of populist arguments.
First, we expect that populism will be moderated by voters’ priors on candidates (for example, Arceneaux Reference Arceneaux2008; Bisgaard and Slothuus, Reference Bisgaard and Slothuus2018; Nicholson Reference Nicholson2011; Slothuus and Bisgaard Reference Slothuus and Bisgaard2021). When voters perceive candidates as members of the in-group, the people-centric elements of populist rhetoric should be more effective. Second, we argue that the effects of populist arguments will vary conditionally on voters’ priors on political issues (for example, Carmines and Stimson Reference Carmines and Stimson1980; Bélanger and Meguid Reference Bélanger and Meguid2008; Fournier et al. Reference Fournier, Blais, Nadeau, Gidengil and Nevitte2003). We expect that populist rhetoric depicting elites as an evil out-group will work best on highly salient and familiar issues that voters are likely to perceive as actual political problems. Third, we posit that voters’ populist attitudes as an attitudinal prior increase the effect of populist rhetoric. Populist attitudes make the in-group and out-group thinking of populist rhetoric more ‘accessible’ (Chong and Druckman Reference Chong and Druckman2007).
Empirically, we conduct a survey experiment on a large and high-quality sample of the UK public (n≈9,000) to identify the causal effects of populist rhetoric. This is particularly advantageous for understanding the effects of thin populist arguments because politicians usually enmesh them with thick populist ideologies, policy platforms, and positions (Hunger and Paxton Reference Hunger and Paxton2022; Neuner and Wratil Reference Neuner and Wratil2022). We design a single-profile vignette experiment in which a hypothetical political candidate with unknown positions provides an argument on one out of nine different political issues. The arguments vary randomly in their direction (for or against) and in whether they contain populist rhetoric or not. Importantly, we elicit voters’ evaluations of the hypothetical candidates and their position on the issue before and after reading the argument, allowing us to identify the within-individual change in issue preferences (issue preference effects) and candidate evaluations (candidate support effects).
Our results reveal no difference in the average changes of issue preferences between populist versus non-populist arguments. However, the average null effect results from strong heterogeneity in conditional effects by whether respondents have a positive versus negative pre-treatment evaluation of the candidate who makes the argument. Compared to non-populist arguments, populist arguments change the issue preferences of respondents who are already inclined to vote for a candidate more but work much less among those who do not support the candidate pre-treatment. Regarding changes in candidate support, we find that candidates who make a populist argument are, on average, evaluated more negatively compared to candidates who make a non-populist argument. We also find heterogeneity conditional on populist attitudes: non-populist respondents strongly punish candidates for making populist arguments, whereas respondents with strong populist attitudes do not penalize them for using populist rhetoric. We do not find systematic evidence that the salience and familiarity of an issue moderates the effects of populist arguments.
These findings have important implications for our understanding of political rhetoric for populist success. They suggest that engaging in populist rhetoric is a risky strategy for politicians because populist arguments can hurt their electoral viability among many voters. Populist arguments can convince supporters to change their issue preferences though, and they do not repel voters with strong populist attitudes. Hence, the results suggest that populist rhetoric is helpful for consolidating and mobilizing the pre-existing populist base but counter-productive for winning over new voters.
The Effects of Populist Rhetoric
Among scholars (for example, Canovan Reference Canovan1982, Reference Canovan1999; de Vreese et al. Reference de Vreese, Esser, Toril, Reinemann and Stanyer2018; Appel Reference Appel2018; Moffitt Reference Moffitt2016; Mercieca Reference Mercieca2015; Kazin Reference Kazin2016), as well as in public discourse (for example, Fisher Reference Fisher2017; Goldhill Reference Goldhill2017; Healy and Haberman Reference Healy and Haberman2015; Giridharadas Reference Giridharadas2016), there is a widespread conjecture that populist rhetoric is an effective rhetorical strategy. Given how widespread this suspicion is, there is relatively little research that isolates the effects of populist rhetoric from those of substantive policy positions while also testing populism as a speech act on outcomes closely related to electoral success. Here, we address this important gap in the literature.
We employ an ideational understanding of populism as a ‘thin’ ideology that can manifest in political communication. This populist ideology is formed of appeals to a fictitious and homogeneous group of ‘the people’ and frank criticism of conspiring and evil ‘elites’ and ‘the establishment’. This dichotomy then results in a Manichean worldview of politics as a constant fight between good and evil, ‘the people’ and ‘the elites’ (Mudde Reference Mudde2004). In politics, this thin populism gets combined with ‘thicker’ ideologies (such as nativism or socialism) and positions (Mudde Reference Mudde2007).Footnote 1
A set of ground-breaking experimental studies is concerned with populism in the media. Bos et al. (Reference Bos, van der Brug and de Vreese2013) conducted the first experiment on populist rhetoric, assessing the effects of populist rhetoric on items asking respondents whether they think that the respective politician will ‘comply with democratic rules’ and ‘could pose a threat to democracy’. Hameleers et al. (Reference Hameleers, Bos and de Vreese2017), Hameleers and Schmuck (Reference Hameleers and Schmuck2017), as well as Hameleers et al. (Reference Hameleers, Schmuck, Schulz, Wirz, Matthes, Bos and Andreadis2021), use changes in populist attitudes, perceptions of blame deservingness, and evaluations of the senders of the populist message as outcomes. Müller et al. (Reference Müller, Schemer, Wettstein, Schulz, Wirz, Engesser and Wirth2017) underscore these experimental results with observational evidence documenting a polarizing effect of populist news and arguing that exposure to populist messages increases populist attitudes. Hameleers and Fawzi (Reference Hameleers and Fawzi2020) show that these effects also translate into affective evaluations of out-groups. Bos et al. (Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020) extend this argument to include not only emotionalized blame attribution but also group cues against immigrants and political elites. This set of studies makes a crucial contribution by testing different communication strategies that are frequently linked to populist politicians, demonstrating that these strategies affect how voters make sense of politics. However, most of these experiments do not present populism as rhetorical statements made by politicians but rather as media frames or opinions of institutions or anonymous experts. Consequently, the outcomes addressed in these studies are mostly not directly relevant to explaining electoral behaviour.
Populist rhetoric has also attracted the attention of political behaviour scholars. Leveraging evidence from Ecuador, Bessen (Reference Bessen2024), for instance, argues that populist frames can increase the acceptability of executive actions that undermine democratic institutions. Whether this also translates into electoral support remains unclear, though. Other work overcomes this challenge: Busby and colleagues (Reference Busby, Gubler and Hawkins2019) show that, among US voters, framing policy problems in a populist manner can increase support for populist candidates. However, these authors, too, do not present populism as the rhetoric of a politician. This also applies to Andreadis et al. (Reference Andreadis, Cremonesi, Kartsounidou, Kasprowicz, Hess, Reinemann, Stanyer, Aalberg, Esser and de Vreese2019), who tested populist messages (with and without a host ideology attached) in fifteen European countries, revealing remarkable cross-country variation in their effects on voting intentions. Hameleers et al. (Reference Hameleers, Bos and de Vreese2018) do test the effects of populist messages on vote intentions but do not disentangle populist rhetoric from a positional cue. Dai and Kustov (Reference Dai and Kustov2024) do assess the electoral effects of populist rhetoric but only on a relatively small sample, which does not allow them to test for small interaction effects with, for example, voters’ prior beliefs on issues or political candidates.
Bakker and colleagues (Reference Bakker, Rooduijn and Schumacher2016) find that populism resonates strongly with people who score low on certain character traits such as agreeableness. Neuner and Wratil (Reference Neuner and Wratil2022), as well as Castanho Silva and colleagues (Reference Castanho Silva, Neuner and Wratil2023), use conjoint experiments to understand the causal effect of thick populist policies and thin populist priorities on vote intentions for political candidates. These authors find that some populist anti-immigration and pro-redistribution positions increase candidate popularity. Similarly, Bakker and colleagues (Reference Bakker, Schumacher and Mathijs Rooduijn2021) show that candidates in the American context can improve their electoral fortunes by adopting anti-establishment positions. However, these studies do either not operationalize populism as rhetoric or, at least not as the rhetoric of authentic political candidates.
Taken together, the existing evidence suggests that populist arguments might play an important role, for instance, in shifting blame, increasing the acceptability of government action, or gaining legitimacy. However, there is very little work that presents populist rhetoric as actual political rhetoric (that is, an utterance of a politician) and investigates outcomes clearly relevant for electoral competition (such as issue preferences or candidate support) while also isolating the effect of thin populism from host ideologies and positions. Our experiment makes a contribution by combining these three elements.
To understand when populist rhetoric is most effective, the logic of our experiment is to create many different situations in which populist arguments could potentially make a difference. To identify these conditions, we combine insights from the studies on populist rhetoric as an ‘identity frame’ (Hameleers et al. Reference Hameleers, Reinemann, Schmuck, Fawzi, Reinemann, Stanyer, Aalberg, Esser and de Vreese2019; Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020; Hameleers et al. Reference Hameleers, Schmuck, Schulz, Wirz, Matthes, Bos and Andreadis2021) with well-established theories from the framing literature.
A Framework for Populist Persuasion
For long, public opinion scholars have questioned whether citizens hold coherent preferences that explain their voting behaviour (Converse Reference Converse1964; Campbell et al. Reference Campbell, Converse, Miller and Stokes1980; Lewis-Beck et al. Reference Lewis-Beck, Norpoth, Jacoby and Weisberg2009; Riker et al. Reference Riker1986; Zaller Reference Zaller1992). Seminal work argues that voters base their decisions on heuristics (Sniderman et al. Reference Sniderman1991), most importantly, frames and cues (Entman Reference Entman1993; Chong and Druckman Reference Chong and Druckman2007; Druckman Reference Druckman2001). Framing theory argues that by ‘select[ing] some aspects of a perceived reality and mak[ing] them more salient in a communicating text’ (Entman Reference Entman1993, 52), politicians can sway people’s opinions. Voters’ pre-dispositions (or priors) are thereby crucial for explaining the effects that different types of political arguments have (Arceneaux Reference Arceneaux2012), so that ‘particular kinds of rhetoric may have an inherent advantage over other claims because of patterns in thought […]’ they evoke (Clifford and Jerit Reference Clifford and Jerit2013, 660). But why might populist rhetoric have such an inherent advantage, on whom, and under what conditions?
Extant work conceives of populist rhetoric as an ‘identity frame’ (Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020; Hameleers et al. Reference Hameleers, Schmuck, Schulz, Wirz, Matthes, Bos and Andreadis2021). Based upon Social Identity Theory (Tajfel and Turner Reference Tajfel, Turner, Worchel and Austin1986), scholars argue that populist rhetoric functions by ‘inviting’ voters ‘to identify with the constructed in-group’ (‘the people’) while attributing various negative qualities to out-groups (that is, ‘the elites’) (Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020, 5). The dichotomy between ‘the people’ and ‘the elites’, ‘good’ and ‘bad’, constitutes a powerful heuristic to voters because the constructed out-group gets blamed for all problems that ‘the people’ experience (ibid.). Intuitively, one would expect that some people might be willing to reward politicians for such polarizing rhetoric while others shy away from candidates who use such appeals (for example, Gennaro et al. Reference Gennaro, Lecce and Morelli2019; Levy et al. Reference Levy, Razin and Young2022; Dai and Kustov Reference Dai and Kustov2022). In addition, recent observational work describes variations in the communication of populist parties conditional on issue salience (Kollberg Reference Kollberg2024) and the dynamics of electoral competition (for example, Dai and Kustov Reference Dai and Kustov2022; Breyer Reference Breyer2023; Licht et al. Reference Licht, Abou-Chadi, Barberá and Hua2024). Together, this suggests that populist rhetoric might be more effective with certain issues when being advanced by specific candidates or among some parts of the electorate.
Previous research on public opinion and candidate choice argues that the effects of a political argument are conditional on voters’ priors (Arceneaux Reference Arceneaux2012; Clifford and Jerit Reference Clifford and Jerit2013) – that is, the predispositions voters hold before they encounter the argument. To build a theoretical framework of the effects of populist arguments, we first discuss general expectations about how the effectiveness of arguments is affected by priors on candidates and issues. These general expectations could or could not be moderated by the use of populist rhetoric – we selectively allude to such possibilities below but primarily rely on an inductive approach testing moderation empirically. Second, we present domain-specific expectations that should only apply to populist arguments, these concern attitudinal priors among citizens, namely the moderating power of populist attitudes.
Priors on Candidates
Much work argues that the efficacy of elite messaging is moderated by the source of the respective cue and voters’ perceptions of it (for example, Arceneaux Reference Arceneaux2008; Bisgaard and Slothuus Reference Bisgaard and Slothuus2018; Druckman Reference Druckman2001; Lupia Reference Lupia1994; Mondak Reference Mondak1993; Nicholson Reference Nicholson2011; Slothuus and Bisgaard Reference Slothuus and Bisgaard2021). Citizens are more likely to change their issue preferences because of messages of elites they support (for example, Carsey and Layman Reference Carsey and Layman2006; Kam Reference Kam2005; Mondak Reference Mondak1993; Slothuus and de Vreese Reference Slothuus and de Vreese2010; Zaller Reference Zaller1992). An argument made by a politician who receives strong support by a voter should thus have a stronger issue preferences effect than an argument by another politician who receives less support (ceteris paribus). In parallel, one might also expect that voters are also more likely to change their support for candidates if they are already supportive of the candidate before encountering the message.
Thinking of populist rhetoric as a strategy of identity framing implies that voters, when reacting to populist arguments, have a strong incentive to maintain a positive perception of ‘the people’ while attributing blame to ‘the elites’. Populist arguments could thus be even more effective than non-populist arguments if respondents perceive the sender of the argument to belong to their respective in-group. In other words, liked politicians might be even more successful in changing their supporters’ preferences or mobilizing them when using populist rhetoric.
Priors on Political Issues
The effects of an argument are also moderated by the priors that voters have on the policy issue at hand. Such priors include information about the specific issue (for example, Lecheler et al. Reference Lecheler, de Vreese and Slothuus2009; Druckman et al. Reference Druckman, Fein and Leeper2012; Druckman and Leeper Reference Druckman and Leeper2012; Howe and Krosnick Reference Howe and Krosnick2017) as well as the ability of citizens to link arguments on an issue to their economic or moral core values (Carmines and Stimson Reference Carmines and Stimson1980; Pollock et al. Reference Pollock, Lilie and Vittes1993). For issues that voters hold strong priors on, voters will be reluctant to change their opinion based upon a single elite message, whereas they will be more inclined to update their preferences for issues they are less familiar with. Voters’ preferences on issues that are familiar and salient should be harder to manipulate than their preferences on issues that are unfamiliar and not salient (ceteris paribus).
Regarding candidate evaluations, we expect that respondents will update their evaluations more when the argument they are confronted with deals with an issue that they have strong priors on and less for an issue that they have weak priors on (ceteris paribus). On an issue where a participant does not have a strong view, there is less reason for that participant to re-evaluate a candidate who takes one versus the other position (Bélanger and Meguid Reference Bélanger and Meguid2008; Fournier et al. Reference Fournier, Blais, Nadeau, Gidengil and Nevitte2003; Howe and Krosnick Reference Howe and Krosnick2017; Rabinowitz et al. Reference Rabinowitz, Prothro, James and Jacoby1982).
The salience and familiarity of an issue might particularly matter for the effects of populist rhetoric. Attributing blame to ‘the elites’ is an important component of the populist heuristic. Clearly, this short-cut should work better for political problems that voters actually perceive as important so that they require a political solution, that is, for issues that voters have strong priors on. Busby and colleagues (Reference Busby, Gubler and Hawkins2019), for example, argue that populist rhetoric works best when voters can link the issue at hand to ‘actual widespread failures of government’ (618). Populist rhetoric should thus be more ‘applicable’ to these high-stakes issues than to other less salient and important issues.
More generally, we expect the strength and direction of these effects to be moderated by the alignment of preferences and the argument made so that respondents will update their candidate evaluations positively if the candidate makes an argument that is aligned with their issue preferences and negatively if the argument is not aligned with their preferences (ceteris paribus). Given the centrality of in-group favouritism in populist thinking (that is, we, ‘the people’, are right, political elites and opponents wrong), it is conceivable that using populist rhetoric amplifies this preference alignment effect.
Attitudinal Priors
Finally, we have a set of expectations that should only apply to populist rhetoric, specifically pertaining to populist attitudes among voters. Populist attitudes represent ‘the individual-level support’ for a populist worldview (Schulz and Wuttke forth., 2). These attitudes will moderate the extent to which voters are willing to engage in the in-group versus out-group thinking inherent in populist arguments. Populist rhetoric – as the manifestation of a thin ideology in communication – might resonate in particular with voters who hold such a pre-existing and coherent set of beliefs. The causal mechanisms between populist attitudes, populist rhetoric, and outcomes related to electoral success are yet to be established though (see Neuner and Wratil Reference Neuner and Wratil2022; Castanho Silva et al. Reference Castanho Silva, Neuner and Wratil2023; Dai and Kustov Reference Dai and Kustov2024).
Important recent contributions have argued there to be significant variation in the extent to which voters hold populist attitudes (for example, Schimpf et al. Reference Schimpf, Wuttke and Schoen2024, Wuttke et al. Reference Wuttke, Schimpf and Schoen2020). We thus argue – to put it in the words of framing theory – that there is variation in the extent to which populist beliefs are ‘available’ (Chong and Druckman Reference Chong and Druckman2007) to voters when reacting to populist rhetoric. We expect that populist arguments then make these populist attitudes ‘accessible’ to voters (ibid.) and moderate the extent to which they rely upon populist identity heuristics when forming issues and candidate preferences. We expect that voters with strong populist attitudes will change their issue preferences in reaction to populist arguments more than voters with weak populist attitudes. A candidate who makes populist arguments will also be evaluated more positively by voters with pronounced populist attitudes.
These general and domain-specific expectations are summarized in Table 1. We will test them alongside the average effect of populist rhetoric on voters’ issue preferences and candidate support in the empirical section of this study.
Table 1. Hypothesized issue preference and candidate support effects for arguments in general as well as for populist arguments specifically
Research Design
Analyzing whether populist arguments follow a different logic than non-populist arguments is difficult using observational data, as the occurrence of populist arguments in the real world is usually confounded by populists taking certain policy positions or ideologies and addressing certain issues (Hunger and Paxton Reference Hunger and Paxton2022; Neuner and Wratil Reference Neuner and Wratil2022). Moreover, voters might self-select into exposure to populist arguments (for example, through news consumption). To overcome these challenges, we conduct a pre-registered single-profile, between-subjects vignette survey experiment on a nationally representative sample of voters in the UK (n = 8890), featuring thirty-six different arguments randomized across nine issues by two argument directions by populist versus non-populist argumentation. The survey was administered as part of YouGov’s standard political omnibus in the UK in March 2022.
The UK is a particularly suitable case to study the effects of populist arguments due to their widespread presence in recent political discourse (for example, Scotto et al. Reference Scotto, Sanders and Reifler2018). In particular, there is a ‘high propensity of “Mainstream Populism”’ (March Reference March2017, 283) in the UK. Both Labour and Conservatives regularly engage in populist rhetoric (for example, Alexandre-Collier Reference Alexandre-Collier2022; Bale Reference Bale2013; Watts and Bale Reference Watts and Bale2019). In the recent past, both parties were led by politicians – Jeremy Corbyn and Boris Johnson – who are widely seen as examples of populist leaders (Alexandre-Collier Reference Alexandre-Collier2022; Demata Reference Demata and Kranert2020). This allows us to credibly assign populist versus non-populist arguments to fictitious political candidates. It makes it less likely that respondents will instantly associate populist arguments with one political party, which could potentially bias results. If, instead, we conducted a similar experiment in the USA, for example, one concern would be that populist rhetoric might be perceived as more credible when coming from a GOP politician or might signal that an unlabeled politician was from the GOP. Similarly, in other European countries, respondents would foremost associate populist statements with radical left or right parties when seeing populist statements. Against this backdrop, we argue that by conducting the experiment in the UK context, we can increase the ecological validity of our findings, as presented vignettes will strike respondents as realistic. Selecting the UK as a case implies that our results are most likely to replicate in other polities that are characterized by mainstream populism (for example, Turkey or Italy).
Pre-Treatment
First, respondents are asked a three-question populist attitudes battery originally developed by Castanho Silva et al. (Reference Castanho Silva, Andreadis, Anduiza, Blanusa, Corti, Delfino, Guillem, Ruth, Spruyt, Steenbergen, Littway, Hawkins, Carlin, Littvay and Kaltwasser2019). We pick three of their suggested survey items (the one most strongly associated with each subdimension of the scale) to measure populist attitudes among voters in its conceptual breadth.Footnote 3 The order of these questions is randomized.Footnote 4
Next, participants are asked for their preferences on one randomly selected political issue from a set of nine. All issues were selected based on three different criteria: first, they were chosen to create variation in the strength of prior attitudes across issues. We rely upon previous survey-experimental work on issue stability and issue-voting to identify which issues people have strong, medium, or weak priors on (Hanretty et al. Reference Hanretty, Lauderdale and Vivyan2021). These authors have conducted a three-wave survey to estimate within-individual opinion stability for different issues. They find, for instance, that foreign aid is an issue that UK voters have very stable preferences for. We take this as evidence that respondents’ preferences on this issue are harder to change than, for example, their preferences on a subsea electricity cable to Iceland. While the former has been a salient issue in the UK, the latter is de facto absent from political competition so that people will tend to have weaker priors. The second criterion is that it is plausible for candidates from various parties to make arguments for and against each issue statement. We did purposefully not include issues that constitute the core of the policy platforms of one of the two big UK parties or are directly linked to recent populist movements in the UK. This is to ensure that we do not introduce possible biases into the experiment that could result from participants associating certain positions or arguments with one of the major parties. Finally, we also ensured that, on each issue, it is conceivable for candidates to make arguments for and against all these issue statements. The issues are listed in Table 2, including our pre-registered assessment of people’s average attitude strength on them. Respondents rate the respective issue statement on a standard five-point agree-disagree scale.Footnote 5
Table 2. Issue statements used and their pre-registered strength
Afterwards, participants are introduced to a hypothetical candidate profile and asked to state how much they would want to vote for this candidate on a five-point scale.Footnote 6 The profiles contain a brief description of the candidate and a picture to induce variation in the extent to which respondents support the candidate and to increase the ecological validity of our findings – two such profiles are presented in Figure 1. The information respondents receive in these profiles is similar to what voters might learn about a candidate, for example, from a campaign leaflet; namely, the candidate’s name, gender, age, family background, profession, political experience, and cues about their ethnicity.Footnote 7 Similar stimulus material was used, for instance, by Kirkland and Coppock (Reference Kirkland and Coppock2018) to analyze candidate attributes in non-partisan elections. Note, however, that we leave out the political positions of the candidates at this stage; they are presented during the experimental intervention. We construct the candidate profiles by randomly combining different text elements with pictures of real political candidates from regional elections in Canada. In total, the candidate profiles broadly resemble the distribution of candidates present in the UK in reality with regards to the ratio of male/female, white/minority and young/old candidates (Lamprinakou et al. Reference Lamprinakou, Morucci, Campbell and van Heerde-Hudson2017).
Figure 1. Examples of the main treatment page for both outcomes.
Treatment – Operationalization of Populist Arguments
Next, participants are shown an argument made by the respective candidate for which we randomize whether the respective argument is populist or not, as well as its direction (for/against). All arguments contain populism’s three core elements: people-centrism, anti-elitism, and a Manichean worldview (Mudde Reference Mudde2004, Reference Mudde2007). One of the key challenges when testing populist rhetoric in a survey experiment is to develop arguments that are sufficiently populist while also being realistic and not revealing the purpose of the study. We read speeches and campaign materials from politicians of all British parties to identify and include the most common forms of populist arguments present in political discourse across the party spectrum.
People-centrism – operationalized through references to ‘British’, ‘hard-working’, ‘ordinary’, and ‘honest people’ – is extremely common across the entire party spectrum in the UK (March Reference March2017). Demata (Reference Demata and Kranert2020) shows that terms such as ‘British’ and ‘Britain’, which one might associate with the political right, are also among the most used terms in the discourse of the Labour Party and their former leader, Jeremy Corbyn. The author concludes that the combination of populism and nationalism is an important manifestation of populism in British politics on both sides of the political spectrum. Hence, taking heed of the UK context, our populist arguments contain plausibly realistic people-centric appeals with references to ‘the British people’. This choice also reflects Halikiopoulou and colleagues’ (Reference Halikiopoulou, Kyriaki and Vasilopoulou2012) finding that nationalism is the ‘common denominator’ that ‘cuts party lines’ between the radical left and the radical right in many European countries (504).
Similarly, UK parties also display striking similarities in their use of anti-elitist rhetoric: Jeremy Corbyn, for instance, used to speak about ‘taking down the establishment’, while Nigel Farage (former Brexit and UKIP party leader) frequently said that his party ‘[does] not trust the establishment’ (Hyde Reference Hyde2019), and Boris Johnson engaged in an ‘anti-parliament narrative’, in which ‘parliamentarians were not […] portrayed as representatives but as enemies of the people’ (Alexandre-Collier Reference Alexandre-Collier2022, 538). Even prime ministers did not shy away from using anti-elite rhetoric, like Theresa May, when she attacked ‘international elites’ and ‘citizens of nowhere’ (May Reference May2018), or Liz Truss, who created the image of an abstract, evil elite formed by ‘militant unions, vested interests, […] talking heads, [and] Brexit deniers […]’ (Truss Reference Truss2022). In combination with the people-centric elements of our arguments, these narratives form what populism scholars refer to as a ‘Manichean worldview’, that is, an understanding of politics as a struggle between good and evil, ‘the people’ versus ‘the elites’.
In the experiment, we present ‘elites’, such as ‘politicians’ or ‘bureaucrats’, as ‘international’, ‘useless’, and ‘out-of-touch’ – terms frequently used on both sides of the political spectrum in the UK. Not least since Brexit, scholars have pointed out an existing ‘Nationalist-Globalist policy divide’ in the UK. Scotto et al. (Reference Scotto, Sanders and Reifler2018), for instance, argue that ‘nationalist viewpoints, when juxtaposed against Globalist outlooks, are salient […], encompass left-right economic concerns and may portend a new era in British political culture’ (38). Thus, the anti-elite components of our populist arguments are – again – designed to increase the ecological validity of our findings. However, we acknowledge that our focus on ecological validity for the UK context may negatively affect another form of external validity, namely the generalizability of our findings to other country cases, where the manifestations of populist rhetoric may differ.Footnote 8
We incorporate the populist elements in the first and third parts of the candidate’s statement. The second part is kept constant across rhetorical types, providing a clear statement as to whether the candidate is making the case for or against something (i.e. the direction of the argument). Our stimulus material for the NHS surcharge issue is presented in Table 3; the material for all issues is in Supplemental Materials C.
Table 3. Example arguments presented to respondents on the NHS surcharge. Texts for all arguments are in Supplemental Materials C
Post-Treatment
After the treatment, respondents are again asked for their opinion on the respective issue and prompted to evaluate the candidate in the light of the argument made. These questions use the same wording and scales as the pre-treatment questions. We randomize whether respondents are first asked about their opinion on the issue or their opinion on the candidate. While asked these questions, respondents could still see the candidate profiles and the arguments. Screenshots of two candidates and the two outcome questions are displayed in Figure 1.
Trade-Offs of the Experimental Design
Our design is optimized to isolate potential direct and generalized effects of populist rhetoric, given the constraints of the survey context. This comes with some limitations. First, we randomize and marginalize over different descriptive attributes of the candidates who make populist v. non-populist arguments and nine different issues on which the argument is made. This is an important feature of the design that increases the generalizability of our findings (Clifford et al. Reference Clifford, Leeper and Rainey2024; Tappin Reference Tappin2022; Fong and Grimmer Reference Fong and Grimmer2023; Blumenau and Lauderdale Reference Blumenau and Lauderdale2024), but it prevents us from investigating highly specific interaction effects, such as whether populist rhetoric is especially powerful for a single issue (for example, immigration).Footnote 9
Second, given the limited effectiveness of treatments in a survey-experimental context, we also decided against including real-world candidates, party labels, or highly contentious issues in the experiment. Such elements would potentially have been associated with pre-treatment attitudes that are too strong, preventing us from observing the general treatment effects of populist rhetoric in a one-shot experiment. For instance, it seems implausible that respondents would change their evaluation of a familiar candidate from a real party or their preferences on an extremely salient issue based upon a single fictitious vignette treatment administered in a survey. Relatedly, we do not use a classical vote intention item as a dependent variable for candidate re-evaluations but a graded response on the propensity to vote for this ‘kind of candidate’. Inducing a change in vote intention in a partisan election is generally implausible in the survey context. Similarly, Coppock argues that small effect sizes in survey experiments make sense because ‘if persuasive effects were much bigger, wild swings in attitudes would be commonplace and people would be continually changing their minds depending on the latest advertisement they saw’ (2023, 5). Like most other political communication experiments, we thus rely upon the logic that small effects observed in a survey experiment will accumulate over time (see Coppock et al. Reference Coppock, Ekins and Kirby2018 for a detailed discussion).
From our perspective, the benefits of experimentally isolating the effects of populist rhetoric from the factors that it is enmeshed with in real life outweigh the potential disadvantages in terms of external validity that stem from using hypothetical candidates without party labels. While political scientists could assess the combined effects of populist arguments, positions, and party labels on certain issues observationally, it is impossible to disentangle them outside an experiment. We therefore see the isolation of populist rhetoric from its potential confounders as a strength of this research design rather than a shortcoming.
Estimation Strategy
We elicit voters’ candidate evaluations and their position on the issue before and after seeing the candidate’s statement to assess the effect of the populist argument (pre-/post-design). We thus have two outcome variables: first, we are interested in the effects of populist arguments on the within-individual change in issue preferences from pre- to post-treatment. Second, we are interested in the effects of populist arguments on the within-individual change in candidate support from pre- to post-treatment. For both outcomes, we are interested in the relative effect of populist statements in comparison to non-populist statements, holding constant the substantive issue the argument deals with and the side of the issue the argument advocates for. The statement as main treatment thereby randomly varies in whether it contains populist rhetoric and in the candidate’s position on the issue.Footnote 10
Importantly, our experimental design is not well-suited for a direct and formal comparison of effect sizes between the two pathways. This is because the baseline for the changes in issue preferences is preferences on existing issues, whereas the baseline for the changes in candidate support is opinions about hypothetical candidates with unknown political positions. We would consequently expect that respondents’ candidate evaluations are easier to change than their issue preferences. This implies that the issue preference effects of populist rhetoric we find should be taken particularly seriously. By contrast, one ought to be more cautious when interpreting our results regarding candidate support effects, as these might be larger than in reality.
Our estimation strategy follows the recommendation by Gerber and Green (Reference Gerber and Green2012) to model the post-treatment outcome with the pre-treatment outcome being a control variable in a simple regression analysis. This approach produces more precise estimates than a simple ‘difference in means’ estimator that neglects information about the pre-treatment outcome as well as the ‘difference in differences’ estimator that uses the difference between post-treatment and pre-treatment outcome as a dependent variable. Repeated measures designs like ours are rarely used in survey experiments, as some scholars worry about their propensity to create demand effects or consistency pressures. However, Clifford et al. (Reference Clifford, Sheagley and Piston2021) find very little evidence that these fears are valid, concluding that ‘conventional wisdom has been too conservative’ (1061) in regard to survey experiments using repeated measures. By including the pre-treatment outcome as a right-hand-side variable, we enable the regression models to flexibly determine their importance for predicting the outcomes (that is, we do not constrain their regression parameter to ‘1’ as the ‘difference in differences’ estimator would). This is particularly important given that our outcome variables are measured on limited scales, where some regression to the mean is inevitable as pre-treatment outcomes at the scale extremes cannot become more extreme post-treatment. We refer to this strategy as analyzing average ‘changes’ and ‘shifts’. What we mean by this are changes, net of the general pattern of regression to the mean, that we observe by conditioning on the pre-treatment measures of the same variable.Footnote 11 Since we include arguments for nine different issues in this experiment, the observations in our data are clustered. We present jackknifed standard errors by policy issues to account for potential issue-level heterogeneity.Footnote 12
Results
Effects on Issue Preferences
Following the pre-registration, we begin by examining whether our survey prompts provoked the general responses we predicted, which serves as a manipulation check. From pre- to post-treatment, respondents’ issue preferences, on average, changed by 0.49 units on the five-point scale in either direction. Participants did not simply repeat their pre-treatment responses. Throughout this section, we report the effects on z-standardized outcome variables in all tables and figures.
We begin by testing the general expectations on the effects of arguments. Figure 2 presents the predicted changes in issue preferences conditional on pre-treatment candidate support (top) (priors on candidates) and attitude strength (bottom) (priors on issues). It demonstrates that higher pre-treatment candidate support causes larger changes in issue preferences (holding the argument direction fixed at 1) (see Table 6, Supplemental Materials A). However, we find no evidence that the expected and pre-registered strength of attitudes of respondents (weak, medium, or strong priors) affects changes in issue preferences (bottom panel Figure 2, see Table 7 in Supplemental Materials A).
Figure 2. Predicted changes in issue preferences conditional on pre-treatment candidate support and attitude strength across treatment and control. The argument direction is fixed at 1. Predictions of post-treatment issue preference based on models 6 and 7 in Supplemental Materials A; 95 per cent confidence intervals as shaded areas, constructed from standard errors jackknifed by policy issue.
Next, we consider the ability of populist rhetoric to change respondents’ issue preferences. First, we do not find a statistically significant difference between populist and non-populist arguments. Averaging across all issues, all candidates and both directions, we observe that populist arguments are equally successful as non-populist arguments in changing respondents’ issue preferences. In Table 4 below, the Argument Direction coefficient captures the baseline effect for non-populist arguments, and the interaction effect of Populist Treatment*Direction reveals the additional effect of an argument being populist rather than non-populist. This interaction coefficient is close to zero and statistically insignificant: populist and non-populist arguments have similar effects on issue preferences.Footnote 13
Table 4. The issue preference effect of populist arguments
Standard errors jackknifed by policy issues in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
Second, we assess whether the populist argument effect varies conditional on pre-treatment candidate support. We want to assess whether candidates who are supported by respondents pre-treatment are particularly successful in changing their issue preferences when employing populist arguments rather than equally liked candidates who make non-populist arguments. In light of our theory, one could think of pre-treatment candidate support as a proxy measure for seeing the candidate as a member of the respective in-group.Footnote 14 Indeed, we find evidence for strong heterogeneity conditional on pre-treatment candidate support (see the top panel of Figure 3, Supplemental Materials Table 8). Populist arguments made by candidates who receive strong pre-treatment support change respondents’ issue preferences more than non-populist arguments made by equally supported candidates. At the same time, populist arguments made by candidates who receive little pre-treatment support are less persuasive than non-populist arguments. By using populist rhetoric, candidates appear to particularly persuade their own electoral base – that is, voters who already support them – whereas they are less persuasive among voters who do not already support them, compared to using non-populist rhetoric (see Table 8 in Supplemental Materials A). Hence, populist arguments are polarizing as they have more varied effects depending on how voters perceive the politician who makes the populist argument. This suggests that the power of populist arguments lies in mobilizing and appealing to their own supporters rather than persuading other voters.
Figure 3. The marginal effect of an argument being populist conditional on pre-treatment candidate support and populist attitudes. The argument direction is fixed at 1. Marginal effects based on models 8 and 12 in Supplemental Materials A; 95 per cent confidence intervals as shaded areas, constructed from standard errors jackknifed by policy issue.
Third, we turn towards our domain-specific expectations and the role of attitudinal priors. We assess the heterogeneity in treatment effects based on populist attitudes among voters, which we measure through the pre-treatment populist attitudes battery (see bottom panel of Figure 3). To do so, we interact the average score of the populist attitudes items with the treatment. We do not observe any heterogeneity conditional on the presence of populist attitudes in voters (see Table 12, Supplemental Materials A). Even among voters who tend to share populist views, populist arguments are not more persuasive when marginalizing over different levels of pretreatment candidate support.Footnote 15 This suggests that the persuasive power of populist arguments is primarily a function of voters’ relationship to and perception of the politician making the argument rather than varying attitudinal priors to respond to populist rhetoric across the population.
Fourth, we inspect whether the effects of populist arguments on issue preferences are moderated by the strength of voters’ priors on the respective issue. Holding all else constant, we do not find evidence that the persuasion effects of populist arguments vary conditional on the strength of priors (strong, medium, or weak as pre-registered) that respondents have on the respective issue (see Table 10, Supplemental Materials A).
It is worth recalling that the core idea of our analysis – which we also pre-registered – is to marginalize across different issues. Nevertheless, we also explore whether the effects of populist arguments differ between issues. The results of this exploration (shown in Table 11 and Figure 5, Supplemental Materials A) demonstrate that while there is some variation between issues, there are no clear patterns of issue-specific effects. While the differences between treatment conditions are a little larger for the NHS/Immigration issue than for some of the others, the between-issue differences are small in magnitude. Moreover, in our interpretation of the results, there are no clear patterns, which would indicate a specific mechanism according to which populist arguments work clearly better for specific types of issues or one specific issue. For instance, issues that one might associate more with the political right – like immigration and knife crime – have very different effects.Footnote 16
Figure 4. Predicted changes in candidate support conditional on preference alignment (argument direction *pre-treatment issue preferences) and attitude strength. Predictions of post-treatment candidate evaluations based on models 14 and 15 in Supplemental Materials A; 95 per cent confidence intervals as shaded areas, constructed from standard errors jackknifed by policy issue.
In summary, we do find evidence for a link between prior candidate approval and the ability of populist arguments to change issue preferences. Candidates who are supported pre-treatment are more persuasive when making an argument in a populist way, while disliked candidates are particularly unsuccessful in changing respondents’ preferences when talking in populist terms. Based on the idea of populism as an identity frame, one could thus argue that populist rhetoric works primarily among voters who perceive the respective candidate to be a member of their respective in-group.Footnote 17 While we do not find much evidence that would suggest that candidates can persuade large parts of the electorate by engaging in populist rhetoric, they might be able to build support for their issue positions by persuading those who already view them favourably.
Effects on Candidate Support
The second direct route for populist arguments to influence candidates’ electoral performance is by changing a candidate’s appeal. Whereas populist arguments might not be particularly successful in changing voters’ preferences about issues, they might affect how voters view the politician making a populist argument. We conceptualize this route as candidate support effects. Averaging across treatment and control conditions, we observe that respondents change their opinion on a candidate by 0.75 units on the five-point scale in either direction. This average movement between pre-treatment and post-treatment is fairly large.
We begin again by testing the expectations regarding the general effects of arguments. We find that candidates who make an argument that is aligned with respondents’ preferences get re-evaluated positively, while candidates who make arguments that are not aligned with respondents’ preferences get penalized (top panel Figure 4, see Supplemental Materials Table 14). This also serves as a type of manipulation check demonstrating that respondents plausibly link candidates with issue statements and update their candidate preferences in a sensible manner. Regarding the influence of issue familiarity and salience, we find no evidence that arguments on issues that respondents have strong priors for make respondents change their candidate evaluations more than arguments on issues that respondents have medium or weak priors for (bottom panel Figure 4, Supplemental Materials Table 15). This null finding mirrors the null finding on the role of priors on issues for the issue preference effects in the previous section. In parallel to our analysis of effects on issue preferences, we also test whether the effects of arguments on candidate support are moderated by pre-treatment approval of the respective candidate. We do not find evidence for this mechanism (Supplemental Materials Table 9).Footnote 18
Figure 5. The marginal effects for candidates making populist arguments relative to candidates making non-populist arguments conditional on the populist attitudes of respondents. Results based on model 19 in Supplemental Materials A; 95 per cent confidence intervals as shaded areas, constructed from standard errors jackknifed by policy issue.
Our main interest is to assess whether populist arguments cause systematically different changes in candidate support than non-populist arguments. Averaging across all conditions, we find that candidates who make a populist argument are re-evaluated more negatively than candidates who make a non-populist argument. The Populist Treatment coefficient in Table 5 shows that while this effect is relatively small (−0.06 SD), it is significant. This demonstrates that populist arguments are a risky strategy for candidates and that employing such rhetoric can hurt candidates on average. The majority of voters thus seem to disapprove of candidates using the in-group versus out-group logic inherent to populist rhetoric.
Table 5. Candidate support effect of populist arguments
Standard errors jackknifed by policy issues in parentheses
*p < 0.05, **p < 0.01, ***p < 0.001
While the results from some previous studies have indicated that populist rhetoric may have null effects on average and only work with some subgroups of voters (Bakker et al. Reference Bakker, Rooduijn and Schumacher2016, Reference Bakker, Schumacher and Mathijs Rooduijn2021; Bos et al. Reference Bos, van der Brug and de Vreese2013), in our design, the average effect is significantly negative. This resonates with recent work on the populist supply side (Dai and Kustov Reference Dai and Kustov2022; Gennaro et al. Reference Gennaro, Lecce and Morelli2019; Levy et al. Reference Levy, Razin and Young2022), which argues that some parts of the electorate are demobilized by populist rhetoric because they shy away from candidates who engage in such rhetoric. Our results provide further evidence for such a mechanism, showing that populist arguments can backfire electorally in the population at large.
Next, we turn towards moderation by priors on issues. First, we explore whether this average negative effect of populist rhetoric varies conditional on the alignment between the position advocated for in the argument and the pre-treatment preferences. One might think that voters are willing to tolerate populist arguments if they propose a policy that they agree with. Or, vice versa, they might be particularly willing to penalize a candidate who uses populist arguments to advocate for a policy that they do not agree with. However, we do not find evidence for such an interaction. This indicates that populist rhetoric is indeed a potentially disadvantageous strategy even under otherwise favourable conditions (see Table 16, Supplemental Materials A). Second, we assess whether the effect of populist arguments on candidate support is moderated by whether respondents typically hold strong or weak attitudes on the issues. As for changes in issue preferences, we do not find any evidence for heterogeneity by pre-registered attitude strength (see Table 17, Supplemental Materials A).
We again explore potential between-issue differences (see Table 18 and Figure 6, Supplemental Materials A). And, while we again see some variation, in our view, no clear pattern emerges that would point towards a specific mechanism through which populist rhetoric works in combination with a certain set of issues. For instance, while the effects of populist rhetoric seem to be relatively smaller for the most technical issues tested (food subsidies and a subsea electricity cable), there is no evidence which would suggest that populist rhetoric clearly outperforms non-populist rhetoric on a certain set of issues.
Finally, we turn towards our domain-specific expectations regarding attitudinal priors. We analyze heterogeneity in treatment effects conditional on respondents’ populist attitudes. Given that populist arguments – on average – were shown to have negative candidate re-evaluation effects, we seek to understand whether these effects vary between different segments of the electorate. We find that the effect of populist arguments on respondents’ candidate re-evaluation is indeed strongly moderated by the extent to which the respective respondent holds populist beliefs (see Table 19, Supplemental Materials A).Footnote 19 Candidates who make populist arguments get penalized by respondents with weak populist attitudes, whereas respondents with pronounced populist attitudes do not penalize candidates for making populist arguments (see Figure 5).Footnote 20 And, while voters with weak populist attitudes clearly and substantially penalize candidates who make populist arguments, for voters with strong populist attitudes, some uncertainty remains whether the effect of populist rhetoric is positive or only not negative.
Conclusion
The core finding of this survey experiment is that the effects of populist arguments on issue preferences are limited and conditional. On average, across the various conditions tested, populist arguments are not more effective in changing issue preferences than non-populist arguments. However, this finding masks strong heterogeneity: we find that the effects of populist arguments vary substantially by candidate support. Candidates who voters approve of are the most successful in changing respondents’ preferences by talking populist; candidates with little support are the least successful. Building on the literature on populist rhetoric as an identity frame (Bos et al. Reference Bos, Schemer, Corbu, Hameleers, Ioannis, Schulz, Schmuck, Reinemann and Fawzi2020; Hameleers et al. Reference Hameleers, Schmuck, Schulz, Wirz, Matthes, Bos and Andreadis2021; Hameleers et al. Reference Hameleers, Reinemann, Schmuck, Fawzi, Reinemann, Stanyer, Aalberg, Esser and de Vreese2019), we find that populist rhetoric has different effects depending on whether voters identify themselves with the candidate who makes the populist argument. In other words, populist arguments are polarizing along the lines of candidate support. Importantly, these effects are strong, even in the absence of party labels or real political candidates. Given the relevance of partisan identities (for example, Nicholson Reference Nicholson2011; Bisgaard and Slothuus Reference Bisgaard and Slothuus2018; Slothuus and Bisgaard Reference Slothuus and Bisgaard2021), it is well conceivable that these effects are even more pronounced in the real world. Overall, these results suggest that the power of populist arguments is to appeal to supporters rather than to persuade other voters.
At the same time, the results with respect to candidate support indicate that candidates who make populist arguments get penalized by most respondents. On average, voters in our sample disapprove of candidates using the in-group versus out-group narrative inherent in populist rhetoric. We observe important variation in these effects conditional upon the extent to which the respective respondent holds populist attitudes: candidates making a populist argument are punished by respondents with weak populist attitudes, while those with strong populist attitudes do not penalize them. This demonstrates that populist arguments are certainly not a one-size-fits-all solution for candidates to improve their electoral fortunes, they are rather useful for appealing to subsets of the electorate at the risk of alienating the majority.
These results make important contributions to our understanding of populist rhetoric. First, while previous work has found populist arguments to matter for a variety of outcomes that potentially may or may not have downstream effects on electoral behaviour, we contribute by showing a more direct link between populist arguments and electorally relevant outcomes. Theoretically, this lends credence to the idea that populist arguments essentially constitute an identity frame that can serve as an important heuristic (Sniderman et al. Reference Sniderman1991) to voters when evaluating candidates and issues.
Second, by showing that populist attitudes moderate the appeal of populist arguments, we also contribute to the literature on the effects of populist attitudes on voters (for example, Wuttke et al. Reference Wuttke, Schimpf and Schoen2020; Schimpf et al. Reference Schimpf, Wuttke and Schoen2024; Akkerman et al. Reference Akkerman, Mudde and Zaslove2014; Castanho Silva et al. Reference Castanho Silva, Andreadis, Anduiza, Blanusa, Corti, Delfino, Guillem, Ruth, Spruyt, Steenbergen, Littway, Hawkins, Carlin, Littvay and Kaltwasser2019, Reference Castanho Silva, Jungkunz, Helbling and Littvay2020). The results of this experiment demonstrate that populist attitudes of citizens matter for candidate evaluation – but not issue persuasion. Existing experimental work is inconclusive about the role that populist attitudes play in moderating the effects of populist rhetoric on populist voting (Dai and Kustov Reference Dai and Kustov2024; Neuner and Wratil Reference Neuner and Wratil2022; Castanho Silva et al. Reference Castanho Silva, Neuner and Wratil2023). While we find strong moderation, we nevertheless do not find that populist arguments would ever be particularly effective in increasing candidate support – the effects are mostly negative and only indistinguishable from zero among the most populist segments of the electorate. On a more theoretical level, this finding also helps understand why certain political arguments might resonate more among certain groups of voters than others (Arceneaux Reference Arceneaux2012; Clifford and Jerit Reference Clifford and Jerit2013).
Third, our results also matter for observational research on the presence of populist rhetoric in elite discourse. Recent work has argued that politicians engage in populist rhetoric selectively (Dai and Kustov Reference Dai and Kustov2022) due to the (assumed) serious risks associated with such rhetoric (Gennaro et al. Reference Gennaro, Lecce and Morelli2019; Levy et al. Reference Levy, Razin and Young2022). We show that and where such risks exist and that politicians ought to be careful when and how they engage in populist arguments.
It is, of course, important to consider the scope conditions following from our case selection. We expect that the effects observed here are most likely to replicate in countries that – like the UK – are characterized by a relatively large proportion of populist voters and the presence of mainstream parties relying on populist rhetoric. These criteria apply to countries like Turkey and Italy, where large parties are widely agreed to be populist and where a large share of the electorate holds populist attitudes (see Erisen et al. Reference Erisen, Guidi, Martini, Toprakkiran, Isernia and Littvay2021). In countries like Germany, where voters will inevitably associate populist rhetoric with particular (radical) political parties, the effects might look different.
Another important limitation that our research design has in common with most other communication experiments is that we test the effect of populist rhetoric without any parallel or subsequent competition with other forms of rhetoric. We know that such competing frames can change effects, though (Chong and Druckman Reference Chong and Druckman2007, Reference Chong and Druckman2007a). And, while it was crucial to isolate the effects of populist rhetoric, it is undoubtedly an important avenue for future research to investigate the effects of populist arguments when they are competing with other arguments in the public realm.
The results of our experiment indicate two promising avenues for future research to develop a more precise theoretical framework of how populist rhetoric matters. First, our finding that populist arguments are more successful in changing issue preferences when pre-treatment candidate support is strong points towards the importance of the relationship between the sender and the recipient of a populist message. Much research has pointed out that populist parties are characterized by ‘charismatic’ political leaders (for example, Kitschelt Reference Kitschelt1995; Betz Reference Betz, Betz and Immerfall1998; Taggart Reference Taggart2000; Moffitt Reference Moffitt2016). Systematic empirical evidence demonstrating that this alleged charisma translates into electoral support is, so far, missing, suggesting that the effects of charisma on electoral support for populists are most likely to be indirect (van der Brug and Mughan Reference Van der Brug and Mughan2007). The results presented here suggest that the persuasive effects of populist arguments might be one route through which the charisma of populist politicians translates into electoral support. Against this backdrop, future research could try to understand better how different attributes of the sender (for example, measuring charisma more directly) make populist messages more or less effective.
Second, one core advantage of our experimental design is that we are able to marginalize the effects of populist rhetoric over several policy issues. Yet, we have paid limited attention to differences in the effectiveness of populist rhetoric between issues beyond their familiarity and salience with the public. One could, for instance, also suspect that the ownership of a specific issue by political parties and candidates could be a crucial moderator. Can candidates and parties exploit their perceived competence advantages better when using populist arguments? Or, vice versa, can they compensate for their lack of perceived competence by talking in populist terms? Carefully amended experimental designs are necessary to test these and other hypotheses about the role of policy issues for the effectiveness of populist rhetoric (for example, they may have to include the party label or information on party competence). Future work should investigate the role of policy issues for populist rhetoric in more depth, theoretically and empirically.
Supplementary materials
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123425000201
Data availability statement
Replication data for this article can be found at: https://doi.org/10.7910/DVN/M7AFDM.
Acknowledgements
We are grateful to Alexander Coppock for his advice on the experimental design of this study. Sara Hobolt, John Griffin, Ann-Kristin Kölln, Julia Leschke, Moritz Marbach, Fabian Neuner, and the members of UCL’s research cluster on political behaviour provided valuable feedback on previous versions of this article. This paper was presented at MPSA 2022, EPSA 2022, and the Nordic Political Behaviour Workshop 2022.
Financial support
This work was supported by UCL’s Department Research Fund (MK, BL, CW); the Economic and Social Research Council (MK, ES/P000592/1); the Studienstiftung des deutschen Volkes (MK); and the Einstein Stiftung Berlin (MK, EVF-BUA-2022-691).
Competing interests
None.
Ethical standards
No deception of participants was involved in this study, nor was there a risk of harm to participants. No personal data were collected by the authors. As such, it was deemed exempt from further review under the University College London Research Ethics Committee’s exemption category 4 after departmental review. Participants were compensated for their participation via YouGov’s standard panel reward procedures.
Open access
