How candidate characteristics affect favorability in European Parliament elections: Evidence from a conjoint experiment in Finland

Previous literature has studied voter behavior in European Parliament elections. However, it remains unclear how candidate characteristics affect favorability of EP candidates since it cannot be taken for granted that these characteristics work as in national elections. We therefore use a conjoint analysis to examine how gender, left-right ideology, issue focus, political experience, representative focus, citizenship, and attitude towards European integration affected EP candidate favorability during the 2019 campaign in Finland. Results show that while traditional candidate characteristics matter, their impact is dwarfed by effects of traits particular to the EP elections. Hence, it is imperative to observe the particularities of EP elections to understand candidate choice.


Introduction
Existing evidence shows that European Union (EU) citizens do not perceive the union as a legitimate political system (Beetham and Lord, 2014;Schmidt, 2013;). Weak links between citizens' preferences and political decision-making in the EU are a central problem in this regard (Kr€ oger and Friedrich, 2013;van der Eijk and Franklin, 1996). Previous studies have examined important aspects of this issue, including how Members of the European Parliament (MEPs) perceive their representative roles (Scully, 2005), and the attitudes of citizens towards European integration and trust in EU institutions (Hobolt and De Vries, 2016;Sa´nchez-Cuenca, 2000).
While these studies provide many insights, they fail to inform us on how different characteristics of individual candidates affect citizens' evaluations of them. This is important since the European Parliament (EP) constitutes the most direct body representing the voice of European citizens in EU decision-making. Citizens' expectations are key factors for determining whether and how people vote in the European elections, thereby affecting the democratic credentials of the EU and the composition of the EP (Franklin and Hobolt, 2015). The relationship between the voter expectations and candidate traits has been investigated at the national level Wass, 2010, 2011;von Schoultz and Wass, 2016). It cannot be taken for granted, however, that the voters' preferences are similar when it comes to candidates in EP elections. On the contrary, preferences are likely to differ since the EP elections are second-order elections (Hobolt and Wittrock, 2011;Reif and Schmitt, 1980;Schmitt, 2005;Schmitt and Teperoglou, 2015). This may imply that voters emphasize other qualities in candidates when deciding whom to vote for. There are also traits that are unique to European elections and the effects of these are virtually unknown. For example, it is possible for citizens from other member states to run as a candidate in the EP elections in their country of residence. How such features affect candidate favorability is unchartered territory.
It is therefore important to examine how certain candidate traits influence the candidate's favorability in EP elections. We do so by means of a conjoint analysis conducted in Finland during the campaign for the 2019 EP elections. Conjoint analysis makes it possible to determine the causal impact of multiple attributes for making multi-dimensional choices (Hainmueller et al., 2014) and has been used for assessing the impact of candidate attributes (Breitenstein, 2019;Carnes and Lupu, 2016;Franchino and Zucchini, 2015;Kirkland and Coppock, 2018;Marx and Schumacher, 2018). We focus on seven key candidate characteristics: gender, ideology, issue focus, political experience, focus of representation, position on EU integration, and citizenship. These attributes cover both commonly employed candidate attributes and attributes unique to the EP elections, thereby making it possible to compare their impacts.
The results suggest that traditional candidate characteristics, such as gender and ideology, still matter but that their effects may differ from what is found elsewhere. Furthermore, their effect is dwarfed by the impact of traits particular to the EP elections such as candidates' citizenship and representative focus on representing Finland in the EU. This shows that it is necessary to observe the particularities of EP elections to understand the mechanisms at play when it comes to candidate selection.

Candidate characteristics and voters in EP elections
Since direct elections for the EP were introduced in 1979, the EP has seen a constant expansion of powers and competencies with each treaty revision. While the role of the EP differs from the functions of national parliaments, there is little doubt that the EP today has genuine powers to affect legislative decisions in EU policymaking. Nevertheless, there are important institutional differences between EP elections and national elections that also affect how citizens express their preferences (Franklin and Hobolt, 2015;Hix, 2002;Hix and Marsh, 2011;Marsh, 1998;Reif and Schmitt, 1980;Schmitt and Teperoglou, 2015;van der Eijk and Franklin, 1996). Although the importance of the EP has grown over time, the role of the EP in EU decision-making differs fundamentally from what European voters are accustomed to in their domestic political systems. For example, national parties nominate candidates and arrange campaigns, but MEPs are grouped by European Party Groups, which organize the daily working of the EP. There is also no government-opposition divide within the parliament, or in election campaigns, since MEPs need to collaborate to be able to influence EU decisionmaking when bargaining with the Council of Ministers and the European Commission. Partly for this reason, EP elections are usually considered to be second-order elections where domestic rather than European issues dominate election campaigns (Hix and Marsh, 2011;Marsh, 1998;Reif and Schmitt, 1980).
All of this entails that EP elections differ from domestic elections at both national and local levels, and it can therefore not be taken for granted that voters emphasize similar criteria when they decide whom to vote for. Since there is a paucity of literature examining the reasoning of European voters, we know little about what type of candidates they prefer. This question is particularly relevant in countries where citizens can vote for a specific candidate, since the likability of candidates strongly influences who is elected. The Finnish electoral system, where voters must choose a candidate from party lists, is an ideal case to study how candidate characteristics affect voters' choices. When voters cannot directly determine the order of candidates, as is the case in closed list systems, party elites may ostensibly emphasize other criteria. However, even in this situation, it is important for parties to present likable candidates to be able to compete with other parties in elections and maximize the number of candidates elected. Furthermore, when the campaign issues are less salient for voters, as the secondorder status of EP elections implies, it seems reasonable to expect candidate traits to be even more important factors for determining how people assess the merits of their prospective representatives.
Previous literature has examined MEPs' profiles, views, attitudes, and roles (Frech, 2016;Hix, 2002;Scully and Farrell, 2003). However, little attention has been paid to how voters in EP elections perceive these characteristics. Below we outline what candidate characteristics we focus on in this study. When choosing what aspects to include, we do not apply a uniform theory of voting behavior since our aim is not to assess such theoretical models. Instead, we rely on previous studies of EP elections and conjoint analyses of candidate favorability to determine whether these can help explain candidate favorability in EP elections as well. We discuss the direct effects of the traits we include, before outlining how we assess whether these effects differ across candidates' attributes and respondents' characteristics. The first three characteristics are traits often argued to matter in national elections that are also likely to play a role in EP elections, whereas the latter four are more directly connected to the European election.
Gender is often argued to affect candidate favorability as female candidates are at a disadvantage compared to their male counterparts (Besley et al., 2017;Huddy and Terkildsen, 1993;Matland, 1994;Ono and Yamada, 2018). Studies have shown that people tend to attribute specific leadership traits to hypothetical candidates based on gender, i.e. people unwittingly make assumptions about how suitable the candidate is likely to be (Alexander and Andersen, 1993). Although Finland is relatively gender equal when it comes to political representation (47% of the candidates elected in 2019 to the Finnish parliament are women), there seems to be no reason for this pattern to be different.
Ideology is also likely to influence candidate favorability (Anderson and Singer, 2008;Jacoby, 2009;Treier and Hillygus, 2009). Although voters can be expected to prefer candidates with ideological views similar to their own, fringe political parties may have an advantage in EP elections compared to national elections due to the second-order nature of EP elections (Hix and Marsh, 2011;Marsh, 1998;Reif and Schmitt, 1980). Because of the low salience of EP elections, voters use these to express dissatisfaction with the current domestic government or to signal their preferences on a particular policy issue the main parties are ignoring (Hix and Marsh, 2011: 5). This may entail that candidates with clear ideological convictions are preferred over more centrist candidates.
The third characteristic that we include is the issue preference of the candidate, which concerns the topics the candidate promises to work on if elected, which is thought to affect how candidates fare in elections (Denver and Hands, 1990). Issue voting may be a determinant of electoral behavior since several studies show that people increasingly vote based on issues rather than deeply rooted ideological convictions (Ansolabehere et al., 2008;Bullock 2011;Macdonald et al., 1995). This development is connected to the increased salience of postmaterialist or value-driven issues over more traditional political issues in Western societies (Inglehart, 1977), which means that voters who care about such issues are more likely to choose candidates based on this.
The following two traits are commonly used in candidate studies of national elections but are here slightly modified to cover more unique European experiences. The first concerns the political experience of the candidate, or where the candidate has been active before running in this election. There are different perspectives on the role of this characteristic. When it comes to incumbency, incumbent candidates are usually argued to be at an advantage, also in European elections (Carson et al., 2007;Frech, 2016). Whether previous experience was gained at the local, regional, or national level may also affect candidate favorability (Kirkland and Coppock, 2018). In the current case, an important question is how experience at the European level affects candidate favorability compared to other types of political experience. Based on the idea that incumbency constitutes an advantage for candidates, we hypothesize that experience from the European level benefits a candidate.
The focus of representation, or whose interests a candidate works to represent, has also been argued to influence candidate selection (Carman, 2007;Eulau and Karps, 1977;von Schoultz and Wass, 2016). The idea is that by committing themselves to work for the interests of specific groups, candidates can unite these groups as supporters behind them. Even if, once elected, candidates are unable to fulfill their pledges, it is important to assess what citizens expect from candidates (von Schoultz and Wass, 2016). When it comes to the European level, a particularly salient question is whether candidates represent European interests or national interests (Scully and Farrell, 2003). While MEPs tend to see themselves as representatives of their home country, the preferences of citizens remain largely unknown. However, considering the second-order argument of EP elections and the findings of von Schoultz and Wass (2016), it seems likely that working for national rather than common European interests will boost candidate favorability.
The last two characteristics are only relevant in EP elections. The first concerns the issue of citizenship. A unique feature of EP elections is the possibility for EU citizens from other member states to run as candidates in the EU country where they reside permanently. It has even been suggested that elections should be pan-European, meaning EU citizens should be able to vote for candidates from any member states (Bol et al., 2016). Nevertheless, candidates who are citizens in the country of candidacy are likely to be at an advantage, since most people prefer someone with a similar background to represent them in the EP (Bol et al., 2016). It is, however, unclear, whether there are differences in the disadvantage of holding another citizenship depending on the country of origin. Some countries may be seen in a more positive light due to historical connections or contemporary relationships, while other countries have negative connotations for most people. Such predispositions are likely to affect candidate evaluations as well. We will expand on this when outlining the conjoint experiment below.
We finally include the candidate's position on integration in the EU (Hobolt and De Vries, 2016). Unsurprisingly, this aspect is often considered to be pivotal for how people vote in EP elections, since these elections are often considered to be a chance for voters to express their opinion on European integration. People are generally likely to prefer a candidate who aligns with their own view, but we expect that they prefer candidates who express a clear opinion either for or against European integration due to the second-order nature of EP elections (Hix and Marsh, 2011).
We summarize our major assumptions as follows: H1-7: Being male (1), holding clear leftist or rightist ideological views (2), emphasizing postmaterialist issues (3), having experience at the European level (4), working to represent national interests (5), being a Finnish citizen (6), and being clearly for or against European integration (7) has a positive effect on candidates' favorability.
The hypotheses pertain to the direct effects of all attributes, but do not necessarily provide the full picture, i.e. the effect of one attribute may be contingent on another attribute (Hainmueller et al., 2014: 12;Leeper et al., 2019). For example, the consequences may differ depending on whether a candidate is male or female (Horiuchi et al., 2018;Ono and Yamada, 2018), or party affiliation may determine whether negative traits such as corruption have the expected effects (Breitenstein, 2019). It is therefore important to examine potential interactions between attributes. Another possibility is that responses vary depending on the background of the respondents (Carnes and Lupu, 2016;Hainmueller et al., 2014;Horiuchi et al., 2018). This is particularly important to investigate since small groups with strong preferences may bias the average population results (Abramson et al., 2019).
It is beyond the scope of this study to include all possible interactions. We therefore focus on those where differences seem likely to occur in order to establish whether the direct effects are relatively stable across subgroups. 1 We do not formulate specific hypotheses but instead conceive this part as a robustness test of the results obtained in conjunction to H1-7.
For interdependencies between the attributes, we first examine differences in effects depending on the policy focus of the candidate, given that this aspect has been found to affect how respondents evaluate the legitimacy of decision-making (de Fine Licht, 2014). Since the role of the EU differs markedly by policy area, it seems likely that attribute effects will also differ depending on the issue. The other attribute that we include here is the candidate's position on EU integration. Other studies have found that policy positions can moderate the effects of other attributes (Franchino and Zucchini, 2015).
For respondents' characteristics, we already mentioned that voters are likely to prefer candidates with an ideological position that resembles their own. Although such differences may to some extent appear self-evident, it is important to explore whether the results also include less straightforward consequences. The other characteristic is less straightforward and concerns how the interest of respondents in EU matters affects their evaluations. Individuals can be interested in EU matters either because they oppose or support European integration, i.e. it is hard to predict how this aspect plays out. However, in line with the work of Zaller (1992), it is reasonable to expect people with a high level of interest to have more coherent and stable attitudes compared to those with a low level of interest.

Data and research design
We test the hypotheses by means of a conjoint analysis administered to a sample of the Finnish population (n ¼ 829). The data were collected 13-20 May 2019, which means that the question was relevant to respondents since the EP elections took place on 26 May 2019. All respondents were recruited via an online panel provided by Qualtrics, where the sample was representative of the Finnish electorate in terms of age, gender, and region of residence.
Finland joined the EU in 1995 and held its first European election in October 1996. This election was held in conjunction with Finnish local elections, and turnout was quite high -60.3% among Finnish voters residing in Finland. After the first election, Finland has taken part in five European elections, including the 2019 election. Turnout has varied between 31.4% (1999) and 42.7% (2019). Women have voted more actively than men have, which reflects the pattern in Finnish national elections. In recent elections, however, the turnout gap has shrunken from 4.2 percentage points in 2004 to 1.4 in the election of 2019. Compared to national parliamentary elections, turnout in the Finnish European elections has been on average 30 points lower. This has favored some of the political parties. The Green League, the Swedish People's Party, and the conservative National Coalition Party have gained a larger share of votes in EP elections than in parliamentary elections. For the European election, Finland operates as one single national constituency which elects a total of 13 MEPs (as of 2019). This differs from national parliamentary elections, where the country is currently divided into 13 constituencies. The electoral system is an open list system where the seats are calculated based on the d'Hondt formula.
Finland constitutes an interesting case for the present research purposes for several reasons. Previous literature has studied what representative roles citizens expect candidates to perform and how well these correspond to how representatives perceive their own role Wass, 2010, 2011). The results show that Finnish citizens can discern different representative roles and have consistent expectations of representatives in national elections, even if we do not know what these preferences are when it comes to EP elections or how strong the effect on candidate choice is compared to other possible traits that may influence the choice. Furthermore, the Finnish electoral system is an open list system, and, in practice, strongly candidate-centered, meaning that party leaders have no way to ensure who gets elected among the candidates on the party list (von Schoultz, 2018). The emphasis on single candidates comes from the fact that every voter needs to select one single candidate. It is not possible to merely vote for a party list. Hence, voters are accustomed to choosing between candidates when voting, which enhances the face validity of the use of conjoint analysis to examine the relationships. 2 There is an increasing pressure on countries whose electoral system is based on closed party lists to increase voters' possibility to have a say on who actually gains a seat in parliament. Therefore, analyzing candidate favorability in an open list system is relevant also in an increasing number of countries, who have moved already or are currently moving towards a semi-open party list system.
Finland also offers an interesting case, since the 2019 EP elections took place right after national elections for Parliament were held on 14 April 2019. The close proximity in time may entail a risk that EP elections become even more secondorder when voters take it as an opportunity to exact revenge for the national elections, where the incumbent government lost its majority. 3 However, for our purposes there were two major advantages: The differences between national and EP elections should be clearer to voters, and European political issues were on top of the media agenda, so that most people would be aware of the issues involved in this study.
To test our hypotheses, we use a choice-based conjoint analysis that is a relatively novel addition to the toolbox of political scientists (Hainmueller et al., 2014). This technique presents several advantages for the present purposes. In a conjoint experiment, it is possible to investigate the effects of multiple traits on candidate choice (Breitenstein, 2019;Carnes and Lupu, 2016;Franchino and Zucchini, 2015;Kirkland and Coppock, 2018;Marx and Schumacher, 2018). Conjoint analysis also makes it possible to avoid problems with social desirability bias, which often present a challenge when analyzing attitudes and preferences in surveys, including biases against female candidates (Ono and Yamada, 2018) or voting for corrupt candidates (Breitenstein, 2019). Since it is not necessary to ask respondents directly for their preferences on the given attributes, it is more likely that they will answer truthfully. Furthermore, conjoint experiments do well in approximating real-life effects by minimizing problems with external validity that are usually present in experimental research .
Respondents were first presented with an introduction that explained that the aim of the study was to examine views on candidates in the up-coming EP elections. Following this, they were shown seven comparisons of two hypothetical candidates and each time asked to pick the one they would be most likely to vote for in the elections. 4 Each candidate profile consists of different values of the seven attributes outlined above and each attribute included several possible values, as shown in Table 1. A screenshot of how the conjoint appeared in Qualtrics is included in the Online appendix.
In the conjoint experiment, we randomized the attribute levels. Each respondent was asked to select their preferred candidate from a pairwise comparison of profiles. Since each respondent was asked to make seven comparisons, all respondents evaluated 14 profiles each. Although this may seem like a difficult task for respondents, Bansak et al. (2018) show that treatment effects remain stable even with more comparisons and attributes. Hence, there is in practice rarely a specific upper limit to the number of comparisons, but we limited the number of comparisons to seven for each respondent to avoid satisficing. 5 For each attribute, we include a 'not shown' category and respondents were instructed to interpret this as no information being available, i.e. it was for example unclear whether the candidate was a man or a woman. We use this as the reference category in all analyses, meaning the results reflect the causal effect of revealing a certain characteristic of a candidate (being either male or female) compared to this characteristic being unknown. This approach has the advantage of avoiding the problem of selecting a reference category, which in some cases may be quite arbitrary (Leeper et al., 2019) and makes it straightforward to compare the effect sizes across attributes.
For gender, we distinguish between whether the candidate is male or female, while for ideology we use a five-fold classification that allows us to distinguish extremist positions on either side of the continuum from more centrist positions. For policy focus, we included four issues, two of which are more postmaterialist (climate change and environment; human rights and democracy) and two that are of a more materialist nature (immigration; security and defense). These issues were also found to be among the issues Finns cared the most about in relation to the EU during the election (Parlemeter, 2018). For experience, we include different types of political experience from the European level to the local level, as well as a no experience category. For focus of representation, we use a modified version of the question used by von Schoultz and Wass (2016) and include four groups of people that candidates may plausibly aim to represent in EP: all people in Europe; all people in Finland; everyone who voted for the candidate and the candidate's national party. For citizenship, in addition to the possibility of the candidate being Finnish, we include nationalities that are likely to have positive and negative connotations. The attitude towards immigrants in Finland has been positive towards people from richer countries in Europe and negative towards immigrants from poorer countries in Europe (Ervasti, 2004). To take this into account, we include candidates from Denmark and Germany, since these are generally considered as prosperous countries and net contributors to the EU budget. Furthermore, we include candidates from Greece and Romania since these have been widely associated with economic hardship and targeted financial assistance by the EU. Finally, for the candidates' position on EU integration, we include three categories that describe the views of the candidates as having a position in favor of more EU integration, about the same level of EU integration, or less EU integration.
The average marginal component effects (AMCEs) can be estimated using traditional linear regression techniques even though the outcome variable is dichotomous (profile chosen or not) (Hainmueller et al., 2014: 7-9). The AMCEs indicate the average change in the probability that a profile will win support when it includes the listed attribute value instead of the baseline attribute value.
Although conjoint experiments have several advantages compared to more traditional approaches, there are also some shortcomings. The estimated AMCEs are population averages, which entails that there may be important differences across subgroups (Abramson et al., 2019;Leeper et al., 2019). For this reason, it is also important to inspect possible variations between subgroups in the population, for example between men and women or depending on the level of interest in politics since certain attributes may have stronger or weaker impacts depending on such characteristics. It is possible to analyze this by including interaction terms to discover whether the causal effect of an attribute depends on another attribute or characteristic of the respondents (Hainmueller et al., 2014). This average component interaction effect (ACIE) shows effect sizes for the different groups.
We therefore include interaction terms between the relevant attributes and all other attributes to see whether the results differ depending on what issue the candidates emphasizes and the position on EU integration. To examine differences across respondents' characteristics, we rely on survey questions. For differences depending on ideology, we use a question asking respondents to place themselves on a 0-10 ideological scale (10 ¼ furthest to the right). To allow for visual interpretations, this is collapsed into three categories: Left (0-3 on the original scale), Intermediate (4-6) and Right (7-10). For interest in EU matters, we use a question where we asked respondents on their interest in political matters at the EU level and responses were given on a four-fold scale (from 'not at all interested' to 'very interested'). This was subsequently dichotomized (Low: 'Not at all/not very interested' -High: 'Somewhat/Very interested') to ease interpretation of results.
Even if we investigate relevant subgroups, it is impossible to include all potential subgroups and there may be groups of respondents where the reported AMCEs do not apply. It is therefore also necessary to be careful when interpreting findings. A positive AMCE for an attribute level does not imply that a candidate profile containing this feature is preferred by a majority of voters (Abramson et al., 2019). Instead, the AMCE indicates a causal effect of an attribute level, not a description of the level of support for a candidate possessing that attribute level (Leeper et al., 2019). For this reason, our results do not entail majority support for any of the candidates, only those candidates who possess certain characteristics may find themselves at an advantage or disadvantage compared to a candidate with a slightly different profile.
For all regression analyses, we use clustered standard errors to account for the fact that each respondent makes seven comparisons of candidates. The findings are reported with coefficient plots, as recommended by Hainmueller et al. (2014), and included in the Online appendix. When judging the relevance of the interaction terms, we do not only rely on conventional tests of significance, since these are insufficient to certify the relevance of the interaction terms (Kim and Franzese, 2007: 43-44). We therefore also consider the practical implications by assessing whether the effects have similar magnitudes and directions for different values of the moderator. To allow for a complete interpretation of the results, we also calculate marginal means, which describe the level of favorability toward profiles that have a particular feature level, ignoring all other features (Leeper et al., 2019). These are shown in the Online appendix and are referred to in the text when they are relevant for the interpretation of findings.

Empirical results
In a first step, we present the AMCEs for the seven attributes included in the conjoint experiment in Figure 1.
For gender, the effect contradicts H1 since there is a positive impact for female candidates (B ¼ 0.04, p ¼ 0.002), which according to the calculated marginal means (see the Online appendix) entails that revealing that a candidate is female improves the average favorability from 0.49 to 0.52. The corresponding finding for male candidates is small and non-significant. The differences between male and female are also significant (B ¼ 0.03, p ¼ 0.017) when using either gender as reference category instead of the 'not shown'-category. The findings for ideology and H2 also contradict our expectations since revealing that a candidate is centrist on average has a positive impact on favorability (B ¼ 0.04, p ¼ 0.021), meaning that the marginal mean increases from 0.49 when no information is available to 0.52 for a centrist candidate. For policy focus, the results are in line with H3 since candidates emphasizing postmaterialist issues are preferred by respondents. Revealing that a candidate emphasizes human rights and democracy (B ¼ 0.09, p ¼ 0.000) as well as climate change and democracy (B ¼ 0.04, p ¼ 0.019), both have positive effects on favorability, and they thereby increase the marginal mean from 0.46 to 0.55 and 0.50, respectively. The outcomes for these first three characteristics show that traditional candidate characteristics matter in EP elections as well, but their impact sometimes differs from the expectations.  When it comes to political experience, there are no significant AMCEs from either of the attribute levels when comparing them to no information being shown, which contradicts H4. For representative focus, the findings confirm H5 since revealing that a candidate will work to represent the interest of all people in Finland improves favorability compared to no information being shown (B ¼ 0.12, p ¼ 0.000). This finding entails that the marginal mean increases from 0.46 to 0.58, which means people generally select candidates working for the interests of all Finns. It is nonetheless noticeable that working to promote common European interest also increases favorability (B ¼ 0.08, p ¼ 0.000) to 0.54, which shows that this is also a viable strategy for candidates, while there are no substantial effects from representing own voters or the national party.
For citizenship, revealing that a candidate is a Finnish citizen has the strongest observable AMCE (B ¼ 0.15, p ¼ 0.000), which entails that the marginal mean increases from 0.51 to 0.66, thus supporting H6. Finnish candidates are preferred over any other nationality, but it is noteworthy that revealing a candidate to be from Greece or Romania reduces favorability, whereas being Danish or German does not affect candidate favorability. Clearly it matters where non-national candidates come from, since some nationalities are perceived in a particularly negative light.
Finally, the outcome contradicts H7, since candidates who favor about the same level of integration are preferred over no information being revealed (B ¼ 0.06, p ¼ 0.000), whereas there are no discernible AMCEs for candidate profiles promoting either more or less integration.  The question is whether these effects differ across candidates' attributes and respondents' characteristics. In Figure 2, we show differences depending on the candidate's policy focus and in Figure 3, we show the same for position on EU integration. All figures showing ACIEs exclude the 'not shown' category to clarify the substantial results. The estimates and significance levels for all interaction terms are presented in the Online appendix. We limit our discussion to the most notable findings.
Both a visual inspection of Figures 2 and 3 and the statistical tests obtained from the regression analysis indicate that, on most accounts, there are no major differences.
For candidates' policy focus (Figure 2), there are five significant interaction terms at a conventional p < 0.05 level, four of which concern the position on EU politics. However, when inspecting the substantial implications in Figure 2(a), the ACIEs are in all instances rather small and almost always non-significant. The only exception is for candidates emphasizing immigration, where revealing candidates to prefer less integration reduces favorability (B ¼ 0.08; p ¼ 0.011), while preferring the same level of integration increases favorability (B ¼ 0.06, p ¼ 0.047).
For candidate position on EU integration (Figure 3), there are as mentioned significant interaction terms with policy focus, but also with other variables, most notably the candidate' citizenship. However, the practical implications are again limited. The most consistent interaction appears for Finnish candidates, where all three interaction terms are significant. Nevertheless, the ACIEs are in all three cases strongly positive regardless of the candidates' position on integration.  Hence, the general pattern is that the results are stable across candidates' attributes. Figure 4 shows the findings for differences depending on the respondent's ideology, while Figure 5 shows them for interest in EU matters.  The overall picture is again that the findings are stable. For the ideological position of the respondents (Figure 4), we see the expected clear example of a subgroup difference since left-wing respondents favor left-wing candidates and do not choose candidates who are to the right. Similarly, this pattern is reversed for right-wing voters and candidates, whereas the ideological effects are small for the intermediate group of respondents. However, besides this somewhat selfevident outcome, we find that the results are consistent across the ideological subgroups of respondents. The main exception is for policy focus, where emphasizing postmaterialist issues has a positive ACIE among leftist voters, while it is reversed among rightist voters where emphasizing materialist issues has a positive impact. This shows that, as might be expected, certain issues appeal more to leftist voters and others more to rightist voters and indicates that the postmaterialist dimension is correlated with the traditional left-right ideological dimension in the Finnish electorate.
For interest in EU matters ( Figure 5), there are also visible differences for policy focus, 6 where emphasizing either of the four issues has positive effects on favorability among the highly interested, whereas it is only emphasizing human rights and democracy that will give a candidate the upper edge among those with low interest in EU matters. However, it is only for security and defense where the interaction term is significant at the 5% level (p ¼ 0.018) as compared to climate change (p ¼ 0.078). The strengths of some other results also differ slightly but remain in the same direction regardless of the level of interest. Based on these findings, and although there are some differences across the ideological spectrum, the results remain stable across respondents' ideology and the level of interest in EU matters.

Conclusions
Our results have several important implications for candidate favorability in EP elections. In the following, we outline the main findings and, where possible, compare results to the outcome of the 2019 EP elections in Finland to see whether the findings match reality .
First, the effects of candidate characteristics particular to EP elections, such as citizenship and representative focus, dwarfed the impact of more traditional characteristics such as gender (Horiuchi et al., 2018;Ono and Yamada, 2018) and political experience (Carson et al., 2007;Kirkland and Coppock, 2018). Even when the traditional characteristics did matter, their implications often contradicted the expectations. This should not be interpreted as a sign that traditional characteristics are irrelevant for EP candidates, but it shows that it is imperative to consider the special nature of EP elections when determining what attributes affect the popularity of candidates.
The result for ideology seemingly contradicted the traditional description of EP elections as second-order elections (Hobolt and Wittrock, 2011;Reif and Schmitt, 1980;Schmitt, 2005;Schmitt and Teperoglou, 2015), which should entail an advantage for candidates on the fringes of the ideological spectrum. We found, however, that these candidates were on average viewed less favorably while holding a centrist position had a positive effect on candidate favorability independent of the ideological position of the respondent (Leeper et al., 2019). A similar outcome was obtained for views on European integration, where candidates emphasizing the status quo were viewed more favorably than candidates that were either for or against more integration. This may to some extent be particular for Finnish politics, where consensual and centrist parties are the norm. Nevertheless, it shows that extremist or populist candidates challenging the establishment are not necessarily at an advantage in EP elections. This was also evident in the Finnish EP election results, where the main anti-integration party, the right-wing populist The Finns Party, did worse than expected and received 13.8% of the votes, resulting in only two seats. The Finns Party received 17.5% in the national elections a few weeks earlier, showing that being extreme or anti-integration did not benefit them in the EP elections. The successful parties in terms of vote gains emphasized postmaterialist issues, such as human rights, welfare, and combating climate change over more materialist issues, such as economic growth and immigration threats (Christensen and Svensson La Rosa, 2019). This again shows that the findings from the conjoint experiment align well with the outcome of the elections in this case.
We generally found that candidates' political substance (issue focus, ideology, representative focus, EU integration) mattered more than their background (gender, political representation). Hence, candidate favorability is not predetermined by the background of candidates. Furthermore, female candidates were at an advantage over their male counterparts thus contradicting research showing the opposite (Besley et al. 2017;Huddy and Terkildsen, 1993;Matland, 1994;Ono and Yamada, 2018). While this result should not be overinterpreted, it is worth noting that 7 of the 13 MEPs elected in Finland were indeed female, showing that female candidates may have been at an advantage, at least in this election. It should also be remembered that women's representation is somewhat higher in the EP than in national parliaments (Xydias, 2016). Whether this is evidence of more general patterns of gender equality for political candidates remains to be seen.
Candidates' citizenship provides an exception to this pattern, since it clearly mattered to respondents. Being a Finnish citizen had a strong positive impact on favorability compared to other EU nationalities. While this might be expected, there were only negative effects for coming from Greece or Romania, whereas being a Danish or German citizen had no effect on favorability. This perhaps seems unsurprising, but it is worth noting that this had the strongest impact on favorability of all factors included. There is still a need to examine in more detail how such traits particular to EP elections affect the favorability of candidates.
Notably, the findings are relatively stable across both candidates' attributes and respondents' characteristics. Except for self-evident differences, such as being a left-leaning candidate had a positive effect among left-leaning respondents and a negative effect among right-wing respondents, we found no general trends for substantial differences depending on the attributes of the candidate or the characteristics of the respondents. Although we were unable to explore all potential subgroups (Abramson et al., 2019;Leeper et al., 2019), this stability boosts our confidence that the results reflect genuine causal effects rather than methodological artefacts.
These findings do not come without limitations. In addition to the already mentioned potential differences across subgroups, there is a need to examine the impact of other potentially relevant candidate characteristics not included in this study, such as partisanship, ethnicity, and class belonging (Carnes and Lupu, 2016;Kirkland and Coppock, 2018). This is particularly relevant for conjoint experiments, where it has been highlighted that it is necessary to include all relevant aspects to achieve reliable estimates of causal effects (Abramson et al., 2019). Another potential avenue for future research is to establish whether similar findings can be obtained outside of Finland. In particular, it remains to be seen whether the same effects exist where candidate traits are less central. collection process since respondents with low interest in EU matters or those who oppose European integration would be likely to always avoid picking a candidate, and thereby offering little information on their preferences for individual attributes. 5. Some combinations of attribute levels may be unrealistic or even logically impossible, which makes it necessary to exclude such combinations to increase authenticity (Hainmueller et al., 2014: 20). A better option is usually to avoid making such restrictions, either by formulating levels that can be plausibly combined or by ignoring that some are unrealistic in practice, which is what we opted for here. 6. There is also a significant interaction effect for candidates where the representative focus is on own party (B ¼ 0.06, p ¼ 0.035). This entails that there is a tendency for representing one's own party to have a negative effect among people with low interest and a positive effect among people with high interest, most likely because more of the highly interested are themselves active in political parties, leading them to prefer candidates who will represent these. However, the effects are not significant in either group, which is why not too much emphasis should be put on this result.