Conceived in Harlesden: Candidate-centred campaigning in British general elections

Recent decades have seen an increasing trend towards the personalisation of election campaigns, even in systems where candidates have few structural incentives to emphasise their personal appeal. In this paper, we build on a growing literature that points to the importance of candidate characteristics in determining electoral success. Using a dataset comprised of more than 3,700 leaflets distributed during the 2015 and 2017 general elections, we explore the conditions under which messages emphasising the personal characteristics of prospective parliamentary candidates appear in British general election campaign materials. Even when we account for party affiliation, we find that there are important contextual and individual-level factors that predict the use of can-didate-centred messaging.


Introduction
The personal attributes of prospective parliamentary candidates (PPCs) feature prominently in the campaign materials distributed in British general elections. 1 A candidate might emphasise her connections to the local community; perhaps she was born there, it is her place of long-term residence, or her children attend a local school. She might also emphasise her personal traits, such as her occupation or her experience as a local councillor, to signal her ability to provide her constituents with competent representation. And, British PPCs are not alone in behaving in such a manner. Political scientists have noted a trend towards increased personalisation within political systems around the world in recent decades (see, Dalton et al., 2002;Renwick and Pilet, 2016). The logic goes that traditional cleavages and social ties are weakening, resulting in a steady decline in partisanship (e.g., Dalton and Wattenberg, 2002). The personal traits of politicians are increasingly substituting party-focused appeals as key factors in voters' decision-making (Dalton and Wattenburg, 2002;Garzia, 2011;McAllister, 2007).
Cross-national variation in campaign personalisation is frequently attributed to the nature of the electoral system. Carey and Shugart (1995) argue that electoral laws governing access to the ballot, the transfer of votes across candidates, and the numberor the typeof votes cast, as well as the number of seats per electoral district influence the degree of candidate personalisation.
Open ballot systems, where voters have greater control over the access to the ballot and/or the ordering of the list, tend to provide incentives for candidates to cultivate a "personal vote" during the campaign. By contrast, closed ballot systems, where parties control access to the ballot and the ordering of the list, favour more party-centric campaigning, as a candidate's personal reputation has less effect on her electoral prospects. Similarly, party organisation can influence the extent to which candidates personalise their campaigning. Candidates for parties that have strong control over campaign finance and the nomination process are less likely to deviate their campaigning from party messaging (Boggild and Pedersen, 2017). But even where candidates have few structural incentives to emphasise their personal appeal, there is considerable variation in candidate-level behaviour. For example, candidates are not carbon copies of their party on policy dimensions, even in party-centred systems (e.g., Buttice and Milazzo, 2011;De Winter and Baudewyns, 2015;Lloren and Rosset, 2017). There is also considerable intra-country variation in candidates' non-policy attributessuch as their level of experience, occupation, or geographical connectionsand there is a growing body of literature that demonstrates how these attributes serve as powerful heuristics for voters. For example, a candidate's occupation background can make a candidate more (or less) attractive to voters (Campbell and Cowley, 2014;Coffe and Theiss-Morse, 2016;Horiuchi, Smith, and Yamamoto, 2018), and voters consistently prefer candidates who have a connection to the local community (e.g., Cowley, 2013;Fiva and Halse, 2016;Horiuchi, Smith, and Yamamoto, 2018;Pedersen et al., 2007;Rehfeld, 2005). Experimental evidence from the US and Denmark suggests that the personality traits of candidates, such as their perceived warmness or strength, have heterogeneous effects on their appeal to voters (Laustsen, 2017). Finally, there are now a wealth of studies that document how perceptions of candidate attractiveness influence electoral success in a variety of contexts (e.g., Berggren et al., 2010;King and Leigh, 2009;Milazzo and Mattes, 2016), including more party-centred systems such as Germany (Rosar et al., 2008) and Switzerland (Lutz, 2010).
In theory, candidate characteristics should matter little in elections to the UK House of Commons. Unlike American elections, where the use of primaries gives voters a high degree of control over access to the ballot, Britain's first-past-the-post vote system favours party control during both the campaign and the legislative process. In the absence of intra-party competition, there are few structural incentives for candidates to adopt distinct personal appeals (Carey and Shugart, 1995). However, in line with the growing comparative literature, we know that candidate characteristics do influence British voters. A number of studies show that race and ethnicity affect voters' preferences (Fisher et al., 2011;Norris et al., 1992), and Campbell and Cowley (2014) find that voters' support for various hypothetical candidates differs substantially when the candidate's occupation and localness are changed.
The effect of localness has been subject of several recent studies, all showing that the closer a candidate resides to their votersand the more local they are perceived to bethe greater the electoral payoff (Arzheimer and Evans, 2012;Campbell and Cowley, 2014;Evans et al., 2017). Moreover, electoral circumstances or voters' personality traits can determine the impact of localness. Middleton (2018) finds that for candidates inheriting seats in which the incumbent MP is retiring, being local can have a negative impact on electoral performance. But for those challenging the incumbent party, having attended a local school boosts electoral performance (Middleton, 2018). Localness also appears to impress some voters more than others. Using data from the British Election Study, Collignon-Delmar and Sajuria (2018) demonstrate that voters with lower levels of education, distinct regional identities, and those with a higher level of interest in elections, are all more likely to value local candidates.
Our study builds on this growing literature by examining the extent to which messaging emphasising candidates' personal characteristics appears in the campaign materials distributed during British general elections. Given that voters value certain characteristics, we explore whether there is systematic variation in terms of when and where leaflets include candidate-centred -i.e., "personalised"messaging. We test our arguments using the largest collection of British election communications available to datemore than 3,700 leaflets distributed by PPCs during the 2015 and 2017 general elections. 2 We find that, even when we control for party affiliation, there are important contextual and individual-level factors that predict the use of personalised messaging. Our findings are significant because they provide evidence that there is variation in the personalisation of messaging that voters receive during general election campaigns.

The candidate-centred campaign in Britain: Theory and hypotheses
The absence of primaries in Britain ties candidates to their party, allowing party organisationsboth national and localto enforce a relatively high degree of discipline during both the campaign and the legislative process. 3 Thus, much of the early literature on British electoral politics emphasises the importance of national party politics, with local factors or candidates being dismissed as irrelevant (Butler and Stokes, 1974). As Cox (1987: 3) writes: "Representation in modern Britain is conceived of as almost exclusively 'national' and party-based. What might be called 'local'…is scarcely mentioned at all". However, more recent literature demonstrates that voters value certain traits in their representativeseverything from their connections to the local community (e.g., Arzheimer and Evans, 2012;Campbell and Cowley, 2014;Childs and Cowley, 2011;Johnson and Rosenblatt, 2007) to their occupational background (Campbell and Cowley, 2014).
Why should we expect election literature to emphasise the candidate? The localness, and perceived localness, of a candidate has strong effects on their popularity and electoral performance (Arzheimer and Evans, 2012;Campbell and Cowley, 2014;Evans et al., 2017). Some have argued that this is because voters anticipate distributional benefits or in-group favouritism from a local representative (Key, 1949;Fiva and Halse, 2016). Others posit that there may also be a psychological satisfaction associated with voting for candidates that are similar to them (Lewis-Beck and Rice, 1983), or that local candidates are simply better at mobilising voters in their hometown (Rice and Macht, 1987). Regardless of the mechanism, we know that personalisation and localness can be vote-winners, and therefore, there are incentives for candidates to emphasise their personal story.
Given the important role that parties play in determining candidate selection in British elections, we expect variation in the personalisation of leaflet messaging to be driven, first and foremost, by party-level factors. The design of campaign materials is often overseen by the central party organisation or designed according to a central template, so British PPCs frequently do not have full control over the format of their election communications. Candidates and local party elites may be able to adapt the content for the local environment, but the general design is often consistent across constituencies. Each party views the messaging trade-off between party and candidate differently, and through the context of their own electoral strategies and popularity.
Moreover, surveys of PPCs suggest there are differences in how the candidates from different parties perceive the purpose of their campaign. Table 1 presents the percentage of candidates from each party who stated that the purpose of a campaign was primarily to raise their own profile, as opposed to raising the profile of their party. While only a minority of candidates prioritise a personal campaign overall, it is clear that campaign motivation varies across parties within the same election, as well as within parties over time.
[ Table 1 here] In addition to the party affiliation, the use of personalised messaging may also depend on the local popularity of the candidate's party leader. Leadership evaluations serve as an important proxy for party competence (Clarke et al., 2009), and there is ample evidence that assessments of party leaders affect vote choice (e.g. Clarke et al., 2009Clarke et al., , 2011Evans and Andersen, 2005;Graetz and McAllister, 1987;Stevens et al., 2011). If a candidate believes that her party leader is popular with her voters, then she has an incentive to emphasise her leader in her campaign communications. Doing so means she might devote less attention to her own profile, but emphasising a popular leader may prime voters to consider the leader's positive image. Meanwhile, a candidate with a locally unpopular leader will have an incentive to distance herself from the leader and build a strong, personal basis of support to strengthen her position.

H1: Leaflets will be more likely to contain personalised messages if the candidate's party leader is unpopular with her voters.
We might also expect the context of the local political environment to affect whether leaflets include personalised messages. For example, leaflets should be more likely to emphasise candidates' personal characteristics when the reward for doing is greatest, such as in marginal constituencies where small shifts in votes can alter the outcome. Given that voters favour candidates with certain traits (e.g., local connection, previous experience), candidates with these traits should be more likely to use personal appeals that emphasise these popular characteristics to get an edge over their opponent(s). There is evidence that politicians respond to the safetyor precariousness of their electoral position in their constituencies in other aspects. For example, legislators in marginal seats tend to focus on constituency service more than those in safe seats (Heitshusen et al., 2005). We would expect candidates to make efforts to 'sell' their personal story and local connections in order to win over voters. But as races become more competitive, we anticipate that the incentive to include personalised messages will be stronger given that additional votes gained are more likely to be decisive in winning the election.

H2: Electoral marginality will increase the likelihood that leaflets contain personalised messages.
In addition to marginal seats, we expect that leaflets distributed in rural areas will be more likely to emphasise personal characteristics than those distributed in urban areas. Research in Japan shows that urbanisation reduces the incentives for candidates to run personalised campaigns (Richardson, 1998). Cities tend to be more diverse in terms of both interests and demographics, and the increased heterogeneity may reduce the sense of shared local identity. As a result, promoting one's personal attributes, such as a shared connection to the community or personal traits may be less effective as a campaign strategy.
H3: Leaflets in rural constituencies will be more likely to contain personalised messages.
Finally, leaflets from candidates who are less well-known may be more likely to emphasise personal attributes. MPs spend roughly half of their time working in their constituency or working on constituency issues (Rosenblatt, 2006). British voters value such services (e.g., Cain et al., 1987;Heitshusen et al., 2005;Wagner, 2015, 2016), which allows MPs to develop a "personal vote". Indeed, there is evidence that long-standing incumbent MPs who retire take with them a larger personal vote than those with shorter tenures (Middleton, 2018). While we might expect incumbent MPs to be knownto varying extentsamong their constituents, challengers are less well-known and face the need to "introduce" themselves to their potential voters. There is long-standing evidence that by "bringing them to the attention of voters" (Jacobson, 1978: 488), campaigning is more effective for challengers than incumbents (see also Stratmann, 2005;Moon, 2006). Given this, and in order to counter incumbents' personal vote, challengers might be more likely to emphasise their personal background.
In addition, some PPCs are particularly well-known because they occupy high visibility positions (e.g., party leader) or offices (e.g., Chancellor of the Exchequer). Indeed, evidence from the BES supports the idea that such individuals are more familiar to voters than ordinary MPs. In autumn of 2014, 72 per cent of respondents were able to correctly identify the name of their MP from a list of hypothetical candidates. However, 92 per cent identified Ed Miliband as the leader of the Labour Party, 87 per cent correctly identified George Osborne as Chancellor of the Exchequer, and 85 per cent identified Theresa May as Home Secretary. 4 We expect that a PPC who has a particularly prominent profile will have less need to flag her personal characteristics, and therefore, her leaflets should be less likely to included personalised messages.
H4: Leaflets from incumbents MPs and PPCs who hold high profile offices will be less likely to contain personalised messages.

Studying variation in campaign messaging
Despite the increasing focus on social media, such as Facebook and Twitter, traditional unsolicited election communicationsi.e., election leafletsremain the most common form of contact that voters have with political elites during a general election campaign. Receiving an election leaflet was the most common form of campaign contact reported by BES respondents following both the 2015 and 2017 general elections. British parties and their candidates also spend more money on designing and distributing unsolicited materials than on any other campaign activity.
According to the Electoral Commission, prior to the 2015 and 2017 general elections, political parties spent more than £15 million and £13 million, respectively, on communications that were sent to voters via post.
While candidates and parties are legally required to report how much they spent on election leaflets, they are not required to provide information about the content of their communications. Therefore, we rely on data collected from the crowdsourced website Electionleaflets.org (http://www.electionleaflets.org). The site is run by a non-partisan organisation that urges users to photograph or scan leaflets they receive, and upload them to a centralised online repository. The result is a compilation of thousands of scanned leafletsthe largest collection of British election communications in existence. We limit our data collection to include only general election leaflets distributed by candidates from Britain's six largest vote-receiving parties: the Conservative Party, the Labour Party, the Liberal Democrats, the UK Independence Party (UKIP), the Green Party, and the Scottish National Party (SNP). 5 These parties have the resources to engage in campaigning across a range of constituencies, and they have broad policy platforms, as opposed to being single-issue parties relying on niche appeals.
As we are interested in the degree to which leaflets feature candidate-centred or personalised messages, we identify all of the leaflets where the candidate is mentioned by namemore 5 We were not able acquire sufficient leaflets from the Plaid Cymru or the SNP in 2017 to perform a reliable analysis. than 80 per cent of the leaflets collected in both elections. The cost of any unsolicited materials that mention, or promote the election of, a local candidate will be counted against the candidate's election spending. Therefore, in limiting our analysis to leaflets where the candidate is mentioned by name we gain a conservative estimate of the leaflets distributed by candidates, rather than their party. 6 After these restrictions, our dataset includes 3,723 election leaflets disseminated during the 2015 and 2017 general elections. Table 2 summarises the distribution of leaflets across parties and constituencies for the two elections.
[ Table 2 here] While our dataset represents the largest collection of election leaflets available to date, we acknowledge that it is a sample of convenience. These are self-reported data; there are no incentives or institutions encouraging citizens to upload their leaflets to the Electionleaflets repository.
Furthermore, parties are not required to report how many leaflets they disseminated, which means that we are unable to determine whether our sample is representative of the total leaflets distributed in the two elections. However, if we compare our data to contact rates calculated using data from 2015 and 2017 post-election waves of the BES, we find similar patterns. Specifically, we calculate the percentage of BES respondents in the constituency who reported receiving a leaflet from a given party in the previous 4 weeks. When we compare these figures to the total number of leaflets we have for the party in each constituency, we find a positive and statistically significant correlation for all parties in both elections. 7 In other words, we have more leaflets from seats where more BES respondents reported receiving a leaflet from the given party.
We also acknowledge that our 2017 sample contains fewer leaflets and covers fewer constituencies, compared to the 2015 sample. Theresa May announced her decision on call a "snap" election on April 18 th 2017. With the election scheduled to take place less than two months later, parties and their candidates had limited time to design and disseminate election communications.
Between the long and short campaigns, candidates standing for the 2015 general election had more than twice as long to plan and execute their campaigns. Thus, it makes sense that more leaflets were distributed in 2015. Moreover, these patterns are consistent with contact rates reported in the BES; 51 per cent of respondents who reported that they were contacted by a party in 2015 indicated that they had received at least one leaflet, compared to just 36 per cent in 2017. However, even though our 2017 sample covers fewer constituencies, when we compare the set of constituencies for which we have data to those that are omitted from our sample, we find that our 2017 sample is no less representative than our 2015 sample. 8

Coding personalised messages in election communications
In order to determine whether election communications feature candidate-centred messages, we manually code additional information from each leaflet. Initially, we identify whether the leaflet includes any additional personalised messages beyond the candidate's name. Surprisingly, many leaflets do not; 58 per cent of leaflets we examined from 2017 and 51 per cent of the leaflets from 2015 included no additional personalised information apart from the candidate's name (Table 3).
Beyond the candidate's name, we identify two types of personalised messages. First, we look at whether a leaflet mentions any connections the candidate has to the constituency or the local area -i.e., their 'local ties'. In both elections, approximately one third of the leaflets we observed contained at least one message that emphasised the candidate's ties to the local area. Many candidates highlight the constituency as their place of residence and/or work. For example, the 2015 Conservative PPC for Wrexham stated "Born in Wrexham, I live here with my wife and children and work locally", while a leaflet from the Lib Dem candidate for Fareham pointed out that "Matthew lives and works in South East Hampshire and has many direct links with Fareham". Candidates may choose to emphasise the length of their local ties as a means to differentiate themselves from opponents who were more recent transplants to the area. The 2017 Green candidate for Brent Central took this to extremes, going so far as to state "I am a life long [sic] Brent resident, conceived in Harlesden, born in Kilburn, grew up in Queens Park and now reside Willesden". PPCs can also emphasise local connections in more creative ways, such as highlighting their support of the local football team. In 2017, the Labour PPC for Cambridge emphasised that he was a season ticket holder for Cambridge United FC, while the Conservative candidate for North Norfolk stated that on Saturdays he could be found at Carrow Roadhome of the local team, Norwich City.
Second, we identify whether the leaflet mentions any traits or characteristics associated with the candidate, such as their experience, occupation, education, etc. Such messages were more commonroughly 40 per cent of the leaflets in our dataset contained at least one message regarding the candidate's personal attributes. Frequently, candidates make reference to their occupation/occupational experience or their previous experience in politics. The 2015 Green candidate from Exeter, for example, stressed her qualifications as an "experienced business advisor", while the 2017 UKIP PPC for South Thanet mentioned his years of service with the army, his experience in marketing, as well as the fact that he is an ordained minister. Incumbent MPs may choose to reference their more relevant experience as the local representative and/or roles they have held within the government. In a 2017 leaflet, Phillip Hammond -the Conservative PPC for Runnymede and Weybridge -stated, "Since being elected to Parliament over 20 years ago, Philip has represented you as a backbench MP, an Opposition spokesman, and a Cabinet Minster". Alternatively, PPCs may try to create a connection with voters by mentioning their family. For example, the 2017 Lib Dem PPC for North East Hertfordshire, stressed that having a young family meant that "she understands the challenges that real people face".
[ Table 3 here] It is important to note that the coding of messages is not mutually-exclusive. Many candidates reference both their local ties and provide additional personalised information about their qualifications. The percentage of leaflets containing both types of personalised was 26 per cent in 2015 and 29 per cent in 2017. Moreover, the messages themselves may emphasise both the candidate's traits and local connections simultaneously. For example, a leaflet distributed by the 2015 SNP PPC for Aberdeen North stated, "Kristy Blackman is an active local councillor". In this case, the candidate was a local councillor in an area covered by the constituency. Therefore, this message would be coded as emphasising both the candidate's local connections and her experiencei.e., her traits. Similarly, a 2017 Lib Dem leaflet in Erewash contained the following message, "Martin is the Chair of Governors at Friesland School in Sandiacre where he has served as governor since 2002." The candidate is emphasising both his occupation and his local connections, as the school in question is located within the constituency.

Predicting candidate emphasis in British campaign material
The easiest way to compare the personalisation of candidates' campaigns would be to take the percentage of each candidate's leaflets that emphasise their traits and/or local connections. However, in our case, such estimates are unlikely to be reliable for two reasons. First, as we state above, our data are a sample of convenience. Therefore, it is possible that candidates distributed additional leaflets that were not uploaded to the leaflet repository, and we cannot be sure that we have an accurate sample of personalised vs. non-personalised leaflets. Second, our sample is unbalanced across constituencies. Given the wide variance in how many leaflets are reported, relying on candidate-level percentages would almost certainly be misleading. Consider a constituency where we have two Conservative leaflets reported, one of which contained a personalised message. In this case, we would record that 50 per cent of Conservative leaflets in this constituency contained personalised material. However, given that we only observe two Conservative leaflets, our level of confidence as to the 'true' level of personalisation of the candidate's campaign would be very low.
To avoid making misleading inferences, we opt instead to take the leaflet as the unit of analysis. In other words, rather than modelling the relative personalisation of candidates' campaigns, we assess the conditions under which leaflets are more/less likely to include personalised messages. For each election, we estimate three models where the dependent variable is coded '1' if the leaflet contains at least one reference to the candidate's local connections (model 1), the candidate's traits (model 2), or both types of personalised messages (model 3), and '0' otherwise. Table 4 summarises the results of the three models for each election. In each case, the coefficients represent the effect that the characteristic has on the likelihood that a leaflet include a personalised message of the given type. Robust standard errors are given in parentheses.
[ Table 4 here] Given the importance of parties in British election campaigns, we expect that there will be systematic differences in leaflet personalisation across parties. Table 5 presents the predicted effect of minimum/maximum change in each of the variables on the probability that a leaflet will reference the candidate's local ties, their traits, or both types of messages. In 2015, there were clear differences across parties. As Labour is the reference category, we see that leaflets from all other parties were more likely to contain references to the candidate's local connections, traits, and both types of messages. The contrast between Labour and Conservative leaflets is particularly striking; the difference in the likelihood that a leaflet from a Conservative PPC will reference the candidate's local ties, traits, and both types of messages is +19 points, +30 points and +25 points, respectively. Note, however, that we observe few meaningful differences across parties in 2017; with the exception of Conservative leaflets, which are 8 points less likely to contain both types of personalised messages, there are no meaningful differences between Labour leaflets and the leaflets of any other party. Thus, our findings confirm that strategies regarding personalisation are likely to vary across elections, even within parties.
[ Table 5 here] The predicted values in Table 4 also provide some support for H1that leaflets will be more likely to include personalised messages if the candidate's party leader is unpopular with her voters. To evaluate the effect of the local popularity of the candidate's party leader, we use multilevel regression and poststratification (MRP) to calculate constituency-level estimates of leader likeability. 9 We then create a variable, Unpopular leader, which is coded '1' if the likeability in the candidate's leader in the constituency is lower than the median likeability for all leaders, and '0' if the leader's local likeability is above the median. While we find no effect for leader popularity for 2017, we do find modest effects in our 2015 models. Leaflets from candidates with locally unpopular leaders are +7 points more likely to reference the candidate's local connections, +9 points more likely to mention her personal traits, and +4 points more likely to reference both types of messages. 9 While the 2015 and 2017 post-election waves of the BES cover all 632 constituencies in Britain and include an average of just under 50 respondents per constituency, the BES is not designed to be representative at the constituency level. MRP allows us to obtain more reliable estimates of constituency-level public opinion using a combination of nationally-representative survey data and census data aggregated by constituency (see, e.g., Hanretty et al, 2018;Park et al., 2004). Our With respect to contextual factors, we find that, contrary to H2, personalised messages are more likely to appear in leaflets distributed in safe seats. 10 In 2015, leaflets that were highly safe in the previous election were +10 points more likely to emphasise local ties, +21 points more likely to include a message about the candidate's traits, and +9 points more likely to include both types of messages. In 2017, marginality had no effect on the likelihood that a leaflet would emphasise the candidate's traits, but leaflets in highly safe seats were considerably more likely to emphasise the local ties (+19 points) and both types of messages (+24 points). In addition, we find evidence that leaflets distributed in rural -or 'county'constituencies are more likely to reference the candidate's traits (H3), though the effects are more modest than those of electoral marginality. 11 We also find evidence that leaflets from certain types of candidates are more likely to include personalised messages. We take a simple measure of incumbencythe candidate is either an incumbent or she is notand find that, consistent with H4, leaflets from incumbents are significantly less likely to contain highly personalised messages. In both elections, the difference in the likelihood that a leaflet from an incumbent MP will include a personalised message is roughly -10 points across all types of messages, with incumbency having the largest effect on likelihood that a leaflet will emphasise the candidate's traits (-15 points in 2015 and -14 points in 2017).
Similarly, leaflets from candidates with particularly prominent profiles are less likely to include personalised messages. Prominent office holder is coded "1" if the PPC was a party leader and/or they occupied one of the high profile offices of state or their official opposition shadow counter-10 Marginality is measured using the difference in the proportion of the vote received by the first and second place candidates in the previous election. 11 While there are no clear criteria for differentiating between "borough" and "county" constituencies, according to the boundary commission "where constituencies contain more than a small rural element they should normally be designated as county constituencies, otherwise they should be designated as borough constituencies" (Boundary Commission for England, 2007). part (i.e., Chancellor of the Exchequer, Secretary of State for Foreign and Commonwealth Affairs, and Secretary of State for the Home Department) at the start of the short campaign, and '0' otherwise. 12 In 2015, the difference in the likelihood that a leaflets from a prominent candidates would emphasise their personal traits or both their personal traits and their local connections was -18 points and -10 points. We find no meaningful differences between prominent and non-prominent office holders with regards to candidate traits in 2017, but the leaflet of a prominent candidate was -21 points less likely to feature a message about the candidate local ties, and -22 points less likely to mention both the candidate's ties and traits. Finally, we also explore the extent to which there is a relationship between gender and use of personalised messaging. While we had no a priori expectations regarding the effect, we find evidence that leaflets from female PPCs were less likely to feature personalised messages. Future research may wish to consider this relationship in further detail.

Conclusion
In this paper, we use the largest dataset of British election leaflets available to date to explore variation in the personalisation of candidates' campaign materials during the 2015 and 2017 general elections. With regards to party-level differences, our analyses suggest that the decision to emphasise candidate's personal attributes varies both across parties and within parties over time. We find evidence of clear differences across parties in 2015, with leaflets from the Labour party being least likely to contain messages about a candidate local ties and/or traits. By contrast, we find few systematic differences between parties in 2017a fact that may be due to the unexpected na- In addition to party affiliation, we also find evidence that there is systematic variation in terms of whenand whereleaflets include more personalised messages. Contrary to our expectations, we find that leaflets distributed in safe seats are more likely to include personalised messages. However, when combined with our findings regarding incumbency, it is easy to see why this might be the case. Incumbency confers a quality advantage in terms of both appeal and name recognition (Cain, Ferejohn, and Fiorina, 1987;Erikson, 1971). Our findings support the idea that incumbents have sufficient profile that they need not emphasise their personal strengthseither because voters are already aware of their non-policy attributes, or because incumbents have established a personal vote sufficient to compensate for any weaknesses in their personal profile.
Non-incumbent challengers, particularly those competing in a safe seat against a strong incumbent, are at a quality disadvantage in terms of incumbency, and will, therefore, have a strong incentive to emphasise any non-policy advantages they may have in an effort to close the quality "gap" between themselves and their incumbent opponent. Indeed, Middleton (2018) has shown that PPCs challenging incumbents benefit electorally if they have a local connection.
Taken as a whole, our findings suggest that voters in certain types of seats are more likely to received personalised campaign materials. Thus, we provide further evidence that voters are exposed to varying levels of personalisation during the campaign (see also, Milazzo and Hammond, 2017). Voters who receive a higher volume of personalised material, may be more likelyi.e., "primed"to use this information (Krosnick and Kinder, 1990;Druckman, 2004;Takens et al., 2015). If so, we would expect that voters who receive more personalised information will be more likely to rely on candidate attributes when deciding on how to cast their ballot. Thus, our analyses provide a useful addition to our understanding of variation in the nature of campaign messages. Our analysis also provides the basis for further study of how variation in messaging affects the choices citizens make in British general elections.
That being said, we stress that these analyses should be taken as a "first cut" atrather than a definitive analysis ofvariation in candidate-centred messaging. There is considerable variation in the ways that PPCs emphasise their personal attributes and it is likely that certain types of messages resonate more strongly with voters. Future researchers should consider the nature of candidate-centred messaging in more detail to explore not only how the use of different types of personalised messages varies across parties and candidates, but also how variation in the personalisation of messaging included in campaign communications contributes to electoral success.   Notes. Logistic regression models, where coefficients relate to likelihood that a leaflet will include the given type of personalised message. Robust standard errors are given in parentheses. * p<0.05 ** p<0.01. Notes. Figures represent the effect of a minimum/maximum shift on the probability that a leaflet will include a personalised messaging of the given type. Predicted effects for coefficients that were not statistically significant have been omitted.