The ambiguous effects of campaign negativity

When scholars refer to a political campaign engaging in negative advertising, they typically mean that the message involves the ad’s sponsor criticising their opponent (Geer Reference Geer2006). Any critical message is negative, while types of negative messages are categorised based on other dimensions of the advertisement – for example, the tone of the criticism or the specific focus of the message (Brooks and Geer Reference Brooks and Geer2007). Of course, many campaign communications contain multiple messages with many appeals to voters, both criticising the opponent and praising the sponsor; these types of campaign communications are called ‘contrast’ ads (Fowler and Ridout Reference Fowler and Ridout2013).

The goal of negative ads is clear: to create a poor impression of the attack’s target in the minds of voters, thus causing them to prefer the attacker (Kaid Reference Kaid1997) or, if the voters prefer the opponent, to at least get them to abstain from voting (Krupnikov Reference Krupnikov2011). Contrast ads might have dual benefits for sponsors by creating a favourable impression of the attacker while harming the image of the opponent. At the same time, there are concerns that the negativity and potential incivility associated with negative ads could have effects beyond candidate evaluations. The fear is that politics having a generally negative tone could lead citizens to lower their political efficacy, trust in government, and feelings about the country (Lau, Sigelman, and Rovner Reference Lau, Sigelman and Rovner2007).

The effects of negative campaigns have been widely studied for decades using both experimental methods (e.g., Brader Reference Brader2006; Haas, Hassan, and Morton Reference Haas, Hassan and Morton2020; Krupnikov and Piston Reference Krupnikov and Piston2015) and negativity in ads during actual campaigns (e.g., Freedman and Goldstein Reference Freedman and Goldstein1999; Jackson and Carsey Reference Jackson and Carsey2007; Ridout and Walter Reference Ridout and Walter2015; Walter, van der Brug, and van Praag Reference Walter, van der Brug and van Praag2014). The dependent variables studied include vote choice (e.g., Krupnikov Reference Krupnikov2012; Somer-Topcu and Weitzel Reference Somer-Topcu and Weitzel2022), turnout (e.g., Ansolabehere and Iyengar Reference Ansolabehere and Iyengar1995; Finkel and Geer Reference Finkel and Geer1998; Krasno and Green Reference Krasno and Green2008), political efficacy (e.g., Ansolabehere et al. Reference Ansolabehere, Iyengar and Simon1994; Dardis, Shen, and Edwards Reference Dardis, Shen and Edwards2008), and many more. This literature has produced a diverse range of findings (see Dowling and Krupnikov Reference Dowling and Krupnikov2016, Haselymayer Reference Haselmayer2019, Lau, Sigelman, and Rovner Reference Lau, Sigelman and Rovner2007 for comprehensive overviews). Notable US studies showing that negative messaging can decrease voter preferences for the target include Kahn and Kenney (Reference Kahn and Kenney2004), Merritt (Reference Merritt1984), and Pinkleton (Reference Pinkleton1997). However, in a succinct demonstration of the complex nature of this literature, each of these works also finds evidence that negativity harms voter evaluations of the attacker.

Additionally, more recent studies in multiparty systems show that negative messaging has the potential to harm both the target and the attacker (Duggan, Crepaz, and Kneafsey Reference Duggan, Crepaz and Kneafsey2024) and may largely benefit idle third-party candidates (Galasso, Nannicini, and Nunnari Reference Galasso, Nannicini and Nunnari2023). Some analyses have suggested that the primary outcome of negative messaging is backlash or ‘boomerang’ effects against the attacker (Garramone Reference Garramone1984; Sanders and Norris Reference Sanders and Norris2005), while others such as Kaid (Reference Kaid1997) find that negativity can damage the target without backfiring on the sponsor. Ansolabehere and Iyengar (Reference Ansolabehere and Iyengar1995) and Ansolabehere et al. (Reference Ansolabehere, Iyengar and Simon1994) have suggested that negative messaging can depress turnout (or ‘demobilise’ voters), though later works such as Brooks (Reference Brooks2006), Finkel and Geer (Reference Finkel and Geer1998), and Goldstein and Freedman (Reference Goldstein and Freedman2002) have challenged these findings.

If a summation of this vast literature is possible, then it must be that the effects of negative campaigns are highly conditional on the types of messages and the election context (Dowling and Krupnikov Reference Dowling and Krupnikov2016). In this vein, previous work has shown that effects can be dependent on whether the message triggers an emotional reaction (Brader Reference Brader2006), whether an attack is seen as unprovoked or a retaliation (Lau and Rovner Reference Lau and Rovner2009), the ideological proximity of the target to the attacker (Ceron and d’Adda Reference Ceron and d’Adda2016), the presence of viable and ideologically proximate alternative candidates (Duggan, Crepaz, and Kneafsey Reference Duggan, Crepaz and Kneafsey2024; Mendoza, Nai, and Bos Reference Mendoza, Nai and Bos2024), as well as respondent attributes such as political ideology (Jung and Tavits Reference Jung and Tavits2021) and personality traits (Nai and Maier Reference Nai and Maier2021; Nai and Otto Reference Nai and Otto2021). Importantly for this paper, these conditional effects of negativity are themselves conditioned by the country in which the study takes place. For example, Nai et al. (2024) find that women candidates who go negative are especially likely to be ‘punished’ by voters in countries with low levels of gender equality in government. Hence, the fact that the campaign negativity literature largely examines the effects of television ads in the United States means our understanding of campaign negativity around the world is still limited despite the decades of research devoted to the topic.Footnote ⁴

These are obvious limitations for a research area where the effects are dependent on the elements of the message, the campaign context, and the media in which campaign communications are delivered. Foremost, the election systems and laws governing campaigning differ across countries, and negative campaigning is simply more common in the United States than in many other democracies (Fowler and Ridout Reference Fowler and Ridout2013; Walter Reference Walter2014). One country where voters should expect significant levels of negativity is the United Kingdom, where negativity is a common campaign tactic even if party leaders feign outrage at the negativity in British politics (Duggan and Milazzo Reference Duggan and Milazzo2023; Walter Reference Walter2014). However, the OpenElections archive suggests that the type of negativity in British campaigns is distinct from that found in US politics. Negative messaging in British politics can be considered tame by US standards, and this is particularly the case for personal attacks. For example, the seminal work on US elections by Brooks and Geer (Reference Brooks and Geer2007) uses a treatment (designed to be a realistic representation of US campaigning) that attacks a candidate based on their history of several failed marriages. Attacks on family life or other intimate personal issues are essentially non-existent in British campaigns, with personal criticism being largely focused on whether the candidate has ties to the constituency, their educational background, or their career. Additionally, Brooks and Geer (Reference Brooks and Geer2007) only find a difference between negativity about issues and traits when the tone of the attack is uncivil, but different political cultures will differ in what it takes to cross the line from civil to uncivil. As such, it is necessary to reconceptualise what constitutes personal or uncivil attacks in British politics and test the effects of these types of messages.Footnote ⁵

One similarity with the United States – but still different from many European countries – is that national politics in Britain, though not local politics in the home nations beyond England, is dominated by two parties. For these reasons, some of the lessons from the American context might apply to British politics. At the same time, there are major differences in the campaign contexts of Britain and the USA. Most notably, campaigns in Britain utilise a comparatively ‘tame’ type of negativity, take place over a shorter period, are without paid television advertising, and place parties, rather than candidates, at their centre. These differences make it difficult to ascertain whether the current negativity literature generalises to the British case.

Reconceptualising policy and personalisation for campaign leaflet communication

Our study concentrates on the most common form of campaign communication in British general elections – the election leaflet.Footnote ⁶ Election leaflets remain the most common form of campaign contact in British general elections, as political parties and their candidates spend more money on designing and distributing election leaflets than on any other campaign activity.Footnote ⁷ Moreover, there is clear evidence that voters engage with leaflets. According to the British Election Study Internet Panel (BESIP), almost 90 per cent of respondents who recalled being contacted by a political party after the 2024 General Election indicated they received at least one leaflet during the campaign. The BESIP also suggests there is a reasonable level of active voter interaction with campaign leaflets. Of the respondents who received a leaflet, 52 per cent reported that they had read a campaign leaflet, letter, text, or email during the campaign. If we look at those respondents who reported that they voted in the 2024 General Election, the figure rises to 56 per cent (Fieldhouse et al. Reference Fieldhouse, Green and Evans2021). Thus, voters not only remember receiving election leaflets, but they also recall reading them.

Leaflets provide a great source of heterogeneity in campaigns across constituencies. This is true both in form – some leaflets are postcards while others are essentially short magazines – and substance. Because of this variation, campaign leaflets are an important form of campaign communication to study as scholars increasingly see the important role of campaigns within British constituencies, despite all the focus on the parties and their leadership during the campaigns (see e.g., Duggan and Milazzo Reference Duggan and Milazzo2023; Fisher Reference Fisher, Robinson and Fisher2005; Johnston et al. Reference Johnston, Pattie and Cutts2012; Milazzo, Trumm and Townsley Reference Milazzo, Trumm and Townsley2021, Shephard Reference Shephard2007).

Obviously, there are many potential forms of negativity in leaflets that we could investigate; this study looks at effects dependent on whether the negativity is focused on the personal attributes of the candidates or their policy positions. In their study of various dimensions of campaign negativity, Brooks and Geer (Reference Brooks and Geer2007) refer to this as the ‘focus’ of the message.Footnote ⁸ This is arguably an understudied dimension of negative advertising, as candidates even in party dominated systems have attempted to incorporate what Van Aelst, Sheafer, and Stanyer (Reference Van Aelst, Sheafer and Stanyer2012) refer to as the ‘privatisation’ of the candidate – that is, an increasing attention to the candidate as a person rather than the holder of a particular political office.Footnote ⁹ Indeed, in recent British elections candidates have attempted to craft their campaign messages to highlight their personal strengths (Milazzo and Townsley Reference Milazzo and Townsley2020; Trumm et al. Reference Trumm, Milazzo and Duggan2023), and parties have chosen candidates with stronger local ties to a constituency (Butler, Miori, and Ford Reference Butler, Miori and Ford2025) despite the importance of party-focused campaigning in the UK’s electoral context.

While the personalisation of political campaigns has continued to increase in the 21^st century (McAllister Reference McAllister, Dalton and Klingemann2007), individual politicians have always had incentives to present themselves in ways that suggest they are similar to their constituents or, at a minimum, understand their constituents and empathise with them (Fenno Reference Fenno1978). The introduction of television (a visual medium) and then social media (an interactive medium) has only increased the appeals centred on personal characteristics (McGregor Reference McGregor2018; Van Aelst, Sheafer, and Stanyer Reference Van Aelst, Sheafer and Stanyer2012). Members of parliament can present themselves this way when they meet with constituents; incumbents and challengers can create personalised posts on social media. But these methods only reach the small percentage of deeply involved voters who pay outsized attention to politics (Krupnikov and Ryan Reference Krupnikov and Ryan2022).

The primary means for candidates to present themselves to voters who are less engaged is through the unsolicited campaign communications that they send to large numbers of voters – i.e., their leaflets. These leaflets, however, are a relic of a campaigning era that predates the personality-driven mass media of the current age, potentially causing voters to process the personal information contained in leaflets differently than they would television ads or social media posts. An important way UK election leaflets differ from US television ads, which has been the focus of the negativity research, is that leaflets include a wide variety of messages in a single leaflet. While it makes sense in the context of US elections to study positive and negative ads separately, that does not reflect the reality of many UK election leaflets. Hence, it is important to study campaigning negativity in the context of contrast ads that attempt to highlight differences in policy and personal traits rather than simply attacking an opponent’s policies or personal traits.

Candidates may believe that a mix of positive and negative messages may shield them from the documented backlash that comes from attacks on personal traits in purely negative ads. The hope would be that positive personalisation leads to increased positive affect on the part of voters towards the candidate, and voters who have positive affect towards a candidate give those candidates more leeway (Piston et al. Reference Piston, Krupnikov and Milita2018). This logic makes sense in the context of a long US election campaign in which the effects of negative ads depend on their timing (Krupnikov Reference Krupnikov2011). If a candidate can increase positive affect via personalisation early in the campaign with a series of positive messages, there might be more leeway later in the campaign for less desirable behaviours.

This changes, however, in the shorter UK campaign context with the positive and negative messages appearing on the same leaflet. A candidate saying, ‘I am just like you’ may attract voters if repeated enough times to have its maximal effect (Lipsitz and Padilla Reference Lipsitz and Padilla2024). If that positive message is immediately followed with a statement saying, ‘My opponent is not one of us’, then the leaflet could fail to lower voters’ evaluations of the opponent while increasing the perception that the attacker is engaging in unfair campaigning. The positive messaging in the first sentence has not had enough time to become the voter’s primary association with the candidate, and thus the negative message operates the same as a purely negative message that voters might see in other campaign contexts.

Evidence from other contexts suggests that negativity containing attacks on the personal life of a candidate can affect voters differently when compared with other types of attacks (Brooks and Geer Reference Brooks and Geer2007; Kahn and Kenney Reference Kahn and Kenney1999; Nai and Maier Reference Nai and Maier2021). Further, while voters will expect candidates to debate their policy positions, messages related to personal characteristics of the candidates may be perceived as inappropriate or irrelevant (Fridkin and Kenny Reference Fridkin and Kenney2011). This may be especially true in the British context, where negative attacks are typically focused on parties and their leadership and where it is seen as more negative to mention a candidate by their name rather than their party (Milazzo and Ryan Reference Milazzo and Ryan2024). This leads us to our first hypothesis:

The fact that debates in British politics focus on parties as a whole or party leaders in particular leads to a greater risk that personal messages will be seen as more negative. The candidate who attacks their opponent’s personal characteristics might damage themselves more than they damage their opponent. Backlash against the sponsors of negative ads is well documented (Galasso, Nannicini, and Nunnari Reference Galasso, Nannicini and Nunnari2023; Roese and Sande Reference Roese and Sande1993; Walter and van der Eijk Reference Walter and van der Eijk2019; but see Nai and Maier Reference Nai and Maier2021 for a discussion of how backlash effects are not uniform among voters). Our treatments, like leaflets in British politics generally, do not attack the candidate on personal dimensions that would be considered completely out of bounds – for example, we do not discuss intimate details of the candidate’s life. At the same time, voters may lower their evaluation of the attacking candidate because they believe the candidate is not focused on issues that affect the voter’s lives.

On the other hand, anything a candidate mentions sets the campaign agenda and primes the voters to decide based on that dimension (Milita et al. Reference Milita, Simas and Ryan2017). As a result, if a candidate contrasts their personal characteristics with their opponent’s characteristics, then the voters may decide that the election should be decided based on the personal characteristics of the candidates. And if the attempted personalisation by the leaflet’s sponsor is effective, then the voters should have a more positive impression of the sponsor than the opponent. This may shield the sponsor from backlash as voters give candidates greater leeway when they have positive affect towards a candidate (Piston et al. Reference Piston, Krupnikov and Milita2018).

But this is where the specific medium of the leaflet matters because there is reason to doubt the personalisation of leaflets will have the desired effect of inoculating the sponsor from criticism. Personalisation is meant to aid politicians by creating a ‘parasocial’ relationship between the politician and voters (McGregor Reference McGregor2018). Voters are supposed to feel that they identify with politicians they have never actually met. This may work with television, which actively engages all the senses (Druckman Reference Druckman2003; Graber Reference Graber1990), or social media because the candidates can interact with voters (Kenski, Kim, and Jones-Jang Reference Kenski, Kim and Jones-Jang2022; McLaughlin and Macafee Reference McLaughlin and Macafee2019). The major concern campaign professionals have about direct mail appeals (like leaflets) in the US case is that they are less engaging when compared with activities such as face-to-face canvassing (Gerber and Green Reference Gerber and Green2000; see also Townsley Reference Townsley2025 for discussion of such effects outside the US). Hence, creating this parasocial relationship is less likely with leaflets than with other media. As a result, candidates may not be able to shield themselves from the backlash of going negative even if they attempt to create positive feelings via personalisation.

In sum, candidates might use tactics such as personalisation and contrast messaging – appeals that include negative messages about their opponents – to create a positive impression of themselves paired with a negative impression of their opponent. The risk with these tactics has always been that voters not only lower evaluations of the opponent due to the content of the messages, but also the attacker, because the attacker is seen as engaging in unfair campaigning (Kahn and Kenney Reference Kahn and Kenney2004). The risk is especially high with personalised messages because many voters find the discussion of those types of traits irrelevant to the politician’s job (Fridkin and Kenney Reference Fridkin and Kenney2011). The positive parasocial relationship that personalisation is meant to create with voters might allow the candidate more leeway (McGregor Reference McGregor2018; Piston et al. Reference Piston, Krupnikov and Milita2018). But the ubiquitous form of media in British elections, the campaign leaflet, lacks the vividness and affordances to create the type of parasocial relationship and warm feelings that would shield them from the backlash. This discussion suggests two hypotheses:

As we discussed earlier, the literature on the persuasive power of partisan leaflets is limited. As a result, there have not been any studies that look at variables that might condition the effects of negative messaging. A key moderating variable that has been ignored is the partisan match between the ad’s sponsor and the voter receiving the message. In fact, Doherty and Adler (Reference Doherty and Scott Adler2014), in one of the only studies to look at how partisan leaflets affect voter’s perceptions of candidates, focused on voters who were not affiliated with any political party.

The literature on partisan motivated reasoning suggests that voters will often process information through the lens of their allegiance to a given political party (Bolsen, Druckman, and Cook Reference Bolsen, Druckman and Cook2014; Leeper and Slothuus Reference Leeper and Slothuus2014). Partisan-motivated reasoning is often implicitly discussed as if it is biasing individuals away from an objective truth, but it also prevents voters from being overwhelmed by the constant flow of messages they receive (Cotter, Lodge, and Vidigal Reference Cotter, Lodge, Vidigal, Suhay, Gofman and Trechsel2020). It does not make sense for copartisans to suddenly turn on a candidate just because they hear something negative about the candidate, as processing information in this way would result in an inability to ever make a decision. This is especially true with biographical information since most would view these messages as irrelevant even in a non-partisan contest. As a result, when the target of the attack is a copartisan, there is absolutely no reason to believe they will be moved by personal attacks.

A similar logic applies to the attacking candidate. The norm in British politics has been to attack parties and leaders instead of particular candidates. It is obviously a violation of this norm to attack a candidate based on their personal traits and their backgrounds. As a result, this is why most voters will lower their favourability towards the attacker. On the other hand, copartisans of the attacker will be motivated to support the tactics of a candidate who shares their partisan identity (Claassen and Ensley Reference Claassen and Ensley2016; Claassen, Ensley, and Ryan Reference Claassen, Ensley and Ryan2025). In fact, Redlawsk (Reference Redlawsk2002) shows that partisan-motivated reasoners engage in so much counterarguing that they actually increase the favourability of candidates from their party when they receive negative information about them. In the British context, the negative campaigning on personal information may be such a norm violation that favourability is not increased but is simply stable. Our expectations, therefore, are more aligned with previous work showing that voters perceive attacks by their preferred candidates as less negative than when attacks are carried out by other parties (Haselmeyer, Hirsch, and Jenny Reference Haselmayer, Hirsch and Jenny2020). That is, we expect that attacker favourability will be unaffected for both attacker and target copartisans, as effects are largely baked in for these respondents.Footnote ¹⁰ As such, we need to moderate our previous expectations, as not all voters will respond to the personal attacks in the same way.

Exemplar leaflets used in the experimental design

To ensure that our leaflet represents a realistic example voters might expect to receive in a general election campaign, we rely on the OpenElections project (www.openelections.co.uk) – an original dataset of content-coded leaflets from British general elections between 2010 and 2024. With respect to candidate-focused negativity, we observe two types of messages: messages where an opposing candidate is referred to more generally and those where the opponent is mentioned by name. Whilst ‘naming and shaming’ was more common during the 2010 General Election campaign in the aftermath of the 2009 expenses scandal, targeting an opposing candidate without using their name remains the most common approach to candidate-on-candidate negativity that we observed in the leaflets across the 4 general elections. Therefore, we use this approach as a model for our vignettes. In Figure 1A–H, we present examples of how candidates target their opponents in their communications. These examples illustrate how candidates use both personal and issue-based messages to try to undermine their opponent(s).

Figure 1. Examples of candidate negativity from actual election leaflets. Treatments in our experiment most closely match example b.

As evidenced by Figure 1, we note a persistent attack type in the leaflet archive that attempts to paint some candidates as out-of-touch elites. Many of these attacks draw on personal information such as attending elite schools or universities (e.g., Eton and Oxford), being career politicians or bankers, being absent from the constituency, and/or having no local ties to the constituency. Such barrages are intended to appeal to the social identity of voters and to capitalise on feelings of animosity towards the out-group member – the targeted politician (Huber Reference Huber2022). This attack type may be well designed to achieve its intended goal; Campbell and Cowley (Reference Campbell and Cowley2014) find that British voters may prefer candidates that do not have a university degree and that candidates living outside the constituency are at a significant disadvantage. We design the personal attacks in our treatment to closely resemble these real-world examplesFootnote ¹³ .

As noted earlier, these examples may not be considered overtly negative by US standards, but they constitute a large class of personal attack types in British politics. Similarly, we draw on the leaflet archive to present our policy-based attacks. In keeping with our interest in constituency-level campaigns, we choose issues that are local and ideologically neutral (i.e., we would expect broad support for the highlighted issues in both the electorate and amongst the candidates). Accordingly, our vignettes were designed to closely approximate how voters would encounter leaflets in the real world.Footnote ¹⁴ This approach provides the experimental design with an element of mundane realism, which improves our ability to generalise findings from the experiment (Kinder and Palfrey Reference Kinder and Palfrey1993).

Experimental stimuli

Our experiment used a 2 x 4 design, in which each respondent was exposed to a vignette composed of one leaflet image from the current MP and one leaflet from the non-incumbent challenger. While the seat is always a Labour/Conservative battleground, we vary the party of the incumbent and the challenger. Finally, as previous research suggests that incumbent MPs are less likely to employ negative messages (Duggan and Milazzo Reference Duggan and Milazzo2023),Footnote ¹⁵ the leaflet image for the incumbent MP is always neutral – i.e., it includes no negative or positive statements. On the other hand, the challenger leaflet employs four possible frames:

1. 1) a control (positive) frame, where the challenger focuses on the merits of his own policy and personal attributes;

2. 2) a contrast (negative) policy frame, where the challenger targets his opponent’s policy positions in addition to the positive mentions of their own policy and personal attributes;

3. 3) a contrast (negative) personal frame, where the challenger targets his opponent’s personal attributes in addition to the positive mentions of their own policy and personal attributes;

4. 4) a contrast (negative) personal/policy frame, where the challenger targets both his opponent’s policy and personal attributes in addition to the positive mentions of their own policy and personal attributes.

The policy positions and personal attributes highlighted/attacked are kept constant across the treatmentsFootnote ¹⁶ . By holding both the neutral target frame and the positive elements from the attacker message constant across vignettes, we can isolate and cleanly estimate treatment effects for each of our three negativity frames while presenting the message in a way that closely hues to real contrast negative messages used in British election leaflets. As we stated earlier, having the treatment messages incorporate both positive (about the sponsor) and negative (about the opponents) statements allows us to test the way personalised politics is increasingly practised; an attempt to create a positive image of the personal life of the sponsor while then attempting to lower the image of the challenger.

The result is 8 possible treatments – four possible frames for a Conservative attacker (with a Labour target) and four possible frames for a Labour attacker (with a Conservative target). The full design allows us to estimate the effect of policy-based attacks, personal attacks, and both personal and policy attacks compared to a message that contains no attacks, which in turn allows us to test our hypotheses regarding the impact of different types of attacks. Additionally, we compare each of the three treatment conditions to each other. This allows us greater insight into whether personal attacks have a different impact than policy attacks (e.g., comparing the treatment containing both attack types to the two treatments containing only one attack type provides a further test on whether personal attacks are perceived differently and have different effects). We also split the analysis by the relationship of the respondent’s partisanship with the candidates’ partisanship. That is, we analyse the results separately for situations in which (1) the attacker and respondent are copartisans; (2) the target and respondent are copartisans; (3) the respondent does not support any party; (4) the respondent supports a party other than the Conservative or Labour parties.

After being allocated to a treatment and viewing the corresponding images, the respondent was asked to indicate on a scale from 0 (very unfavourable) to 10 (very favourable) how they feel about each candidate, with the incumbent candidate in the treatment being presented first. They were also asked to rate the negativity of each candidate’s leaflet on a scale from 0 (very negative) to 10 (very positive). The exact question wording is available in the online Supplementary Appendix (S1). To explore how the effects are moderated by partisanship, we asked respondents to indicate which party they most closely identified with.Footnote ¹⁷ Following the work of Somer-Topcu and Weitzel (Reference Somer-Topcu and Weitzel2022), our partisanship variable contains four categories: attacker copartisan (N = 418), target copartisan (N = 383), out-party (N = 262), and non-partisan (N = 351).

Our choice to present the incumbent frame as neutral while including positive attributes in the challenger control frame may have some drawbacks; most notably, it may create a large discrepancy between the control group values for our measures of attacker and target favourability. However, we find no such issue in the distribution of these variables in our control group.Footnote ¹⁸

The survey experiment was conducted by YouGov on 28 June 2022, with 1,664 respondents allocated across each of the 8 treatments. ^{Footnote 19} The median time taken for a respondent to complete the survey was 12.5 minutes (lower quartile = 9.36 minutes, upper quartile = 16.77 minutes)Footnote ²⁰ . Table 1 shows the number of respondents assigned to each treatment group (see the Supplementary Appendix (S1) for full details on the stimuli used for each treatment group).

Table 1. Allocation of respondents by treatment group