Skip to main content

Party Elites’ Preferences in Candidates: Evidence from a Conjoint Experiment


Party elites selecting candidates are crucial for the composition of parliament. Yet, despite their pivotal position within the party, we know only little about their preferences for potential candidates and how their own backgrounds shape these preferences. This paper presents results from a conjoint experiment carried out with party delegates chosen to select the candidates for five German parties in the run-up to three state elections. Theoretical expectations derived from the principle-agent framework on delegates’ preferences in candidates are evaluated. Analyses show that delegates prefer attributes indicative of quality and socio-demographic similarity in candidates. Additionally, I show that these preferences for candidates differ between inexperienced and experienced delegates, the latter showing a stronger preference for valence attributes in candidates. These findings contribute to our understanding of the role of personal attributes of selectors for candidate selection and hold crucial implications for the composition of legislatures and long-term effects on public policy.

Candidate selection has crucial implications for the composition of parliamentary bodies. In parties that enjoy safe list positions, selection is paramount to election. In such circumstances, candidate selection and intra-party competition for nominations oftentimes supplement for the lack in inter-party competition and, thus, fulfill central democratic functions including holding incumbents accountable (to the party, that is), propelling legislative turnover and increasing descriptive representation. Especially in closed-list systems, in which voters only face “take it or leave it” options, party selectors nominating candidates hold decisive influence on who is entering parliament. They are better capable of removing corrupt incumbents than voters (Asquer et al. 2019), are better positioned to enforce gender parity (e.g., Hazan and Rahat 2010) and can with precision remove undesired candidates, as has happened, for instance, to outgoing German minister of health Andrea Fischer in 2002, when she was denied both of the promising list positions she competed for—presumably for her handling of a food safety scandal—and failed to re-enter parliament.

But what are selectors’ preferences in candidates? What kind of candidates do they like to see on party lists? And how do their own backgrounds affect these preferences? Previous studies have more or less equated nomination outcomes with selectors’ preferences for candidates, and the set of observable characteristics candidates espouse may deem reasonable to infer selectors’ preferences (e.g., Gallagher and Marsh 1988). However, cross-cutting influences from various sides, including the party leadership (e.g., Cutts et al. 2008), anticipated voter reactions (Norris and Lovenduski 1995) or local party heads (Cheng and Tavits 2011), and not last the types of candidates coming forward and—in systems employing party-lists—the need for a balanced slate in terms of gender, age or policy expertise usually dictate selectors’ decision-making and thus could mask their true preferences and bias inferences on selectors’ preference drawn from observable candidate characteristics.

Obtaining a better understanding of these preferences is not only of topical interest, but can help us to better understand whether current imbalances in nominations in terms of gender or age is due to selectors’ preferences or rather the institutional context of the selection. Understanding these preferences becomes increasingly important in cases when the nomination of candidates becomes paramount to their election to parliament. And as electorally safe nominations are usually the ones most heavily contested (e.g., Brady et al. 2007), the preferences of selectors come to bear even stronger and more consequential in these cases when they can choose among different candidates to award safe nominations, irrespective of the institutional design of the selection.

Yet, a large share of the extant literature on candidate selection focuses on the consequences of the institutional design of selectorates than on elites’ preferences. More inclusive selectorates, for instance, appear to be harmful to descriptive representation of women or other minority groups (Hazan and Rahat 2010, p. 114f.). Exclusive ones, in contrast, seem to facilitate the selection of women (e.g., Vandeleene 2014) and the de-selection of incumbent Members of Parliament (MP) (e.g., Put et al. 2015), which poses one of the greatest hurdle in obtaining legislative turnover and descriptive parliamentary representation (see Best and Cotta 2000). This literature has developed and empirically examined theoretical expectations on the link between the institutional design of selection methods and its impact on selection outcomes (e.g., Rahat et al. 2008; Rogowski and Langella 2014; Smith and Tsutsumi 2014). Yet, the preferences and personal backgrounds of the very actors involved in the selection, i.e. the selectors, have been rather overlooked despite their central role in the selection process.

To unveil some of the preferences party selectors’ have for candidate characteristics in the context of a closed-list electoral system in which voters cannot vote for specific candidates, and to further overcome the shortcomings of previous attempts to infer these preferences—including social desirability and the lack of counterfactuals—this study is employing a conjoint experiment with party delegates of five German parties in the run-up to three state elections.Footnote 1 These party delegates were chosen by local party chapters to represent them at party conventions that decide on the composition of the party’s list. Germany makes for an interesting case and allows to elicit selectors’ preferences for different types of candidates more generically than other systems for four reasons. First, selection is carried out by party selectors in contrast to primary voters. Secondly, German parties employ only minimal eligibility criteria in contrast to many other parties in Europe (see Rehmert 2020). Third, the closed-list electoral system minimizes selectors’ concerns for any single candidate’s presumed electability with voters. Selectors in German parties, thus, are less distorted by these otherwise common factors when expressing their preferences for candidates. Finally, it is not uncommon for candidates to compete in a direct match-up for promising list positions, lending credibility to the conjoint setup.

The experiment asked delegates to choose one of two hypothetical candidates competing for a promising list position based on ten different dimensions including demographics, political experience and ideology. In doing so, this study contributes to the literature on candidate selection by experimentally examining what types of candidates selectors prefer over others in a closed-list system and how their own background affects these preferences. Thus, this study adds to our understanding of nomination outcomes and the composition of parties in parliament.

Drawing on principal-agent theory, I derive theoretical expectations over what dimensions selectors form their preferences in candidates. I theorize that selectors follow two types of low-cost informational cues when assessing potential candidates: valence and homophily. Both cues are consistent with strategies to minimize agency-loss. As an observational implication I expect selectors to prefer high-valence candidates and those that espouse similar socio-demographic characteristics. Additionally, following May’s law on the curvilinear disparity of party members (May 1973), I expect that valence-based cues bear stronger on experienced selectors that are higher up in the party hierarchy than first-time selectors, due to the former’s more pragmatic and less zealous perspective.

My findings reveal that selectors do tend to follow cues of valence and homophily when choosing between potential candidates. However, valence-related characteristics have a stronger impact on experienced than on first-time selectors. These findings hold implications for the staffing of party’s selectorate bodies. While parties aiming for high quality candidates might want to staff their selectorate with experienced selectors differentiating between high and low quality candidates, parties aiming for greater descriptive representation might consider choosing more unexperienced selectors as many markers of quality are currently unequally distributed across social categories including gender and age. How selectorates are staffed can thus have crucial implications for parties’ parliamentary composition and, hence, public policy in the long-run. Although these preferences are premised on nominations for closed party-lists, the overall preference for high-valence candidates espoused by more senior selectors are likely to transcend electoral incentives (cf., DeFelice 1981).

Selectors’ Preferences in Candidates

What characteristics help candidates to get nominated? The extant literature is skewed towards majoritarian electoral systems with its particular set of incentives for candidate selection and usually is mute when it comes to selectors’ own backgrounds. Two main empirical approaches can be identified in this field. The first one explores the demographic and occupational backgrounds of endorsed candidates or MPs, while the second uses interviews or surveys of party elites to directly elicit their preferences over candidates.

Numerous contributions in Gallagher and Marsh (1988) base their inferences on what selectors’ are looking for in candidates on observational data of candidates. Yet, most country chapters only refer to formal requirements such as party membership, or mention the general importance of localness and the prerogatives of incumbents for (re-)nomination (Gallagher and Marsh 1988, p. 248 ff.; on the latter see also Hazan and Rahat 2010). Obviously, some of these aspects such as localness, appear to be less important under closed-list rules than more candidate-centered systems (Shugart et al. 2005). More recent studies that make also use of observational data suggest that selectors prefer competent (Besley et al. 2017) and loyal and committed candidates (Galasso and Nannicini 2015). Other studies, employing a gender focus, examine how loyal membership and holding visible party office positions help mainly men in securing nominations, but not so much women (Verge and Claveria 2016), or that men benefit more from past political experiences than women do (Franceschet and Piscopo 2014). While these aspects generally indicate the preference for electorally strong, competent and loyal candidates, these preferences may be conditional on the pool of eligibles, other characteristics, or the electoral system in place. Eventually, they do not allow to infer much about selectors’ actual preferences.

With Bochel and Denver (1983) and Norris and Lovenduski (1995) two comprehensive though somewhat dated studies employing survey data on candidate nomination processes in the majoritarian system of the United Kingdom exist, from which selectors’ preferences in candidates might be inferred. Both report that selectors prefer candidates with greater electability and evaluate other characteristics of candidates, personal as well as political ones, in light of this. That is, Labour party selectors apparently forego the chance to nominate more left-leaning candidates if they know it would cost them support among voters (Bochel and Denver 1983, p. 60). Not nominating personally desirable candidates out of concerns of their electoral reception in the district—labelled “imputed discrimination” by Norris and Lovenduski (1995, p. 107)—renders any attempt of infering selectors’ true preferences by purely examining nomination outcomes biased.

As observational data tends to fail in disentangling selectors’ true preferences from contextual confounders, so do many standard survey techniques fail in eliciting respondents’ true preferences due to social desirability, lack of counterfactuals and the evaluation of candidate characteristics only separately. Not surprisingly, recent studies have moved to conjoint experiments to answer questions related to respondents’ preferences and attitudes. Presenting respondents a choice between a pair of hypothetical candidates with varying characteristics along a set of attributes, researchers can minimize several shortcomings of conventional survey designs—including social desirability and the limit of one single treatment characteristics (see Hainmueller et al. 2014; Hainmueller and Hopkins 2014)—while simultaneously obtaining externally valid and reliable estimations of respondents preferences (see Hainmueller et al. 2014).

Accordingly, a range of recent studies employ candidate choice experiments in which voters are asked to evaluate hypothetical candidates running either in legislative (e.g., Ono and Burden 2019; Franchino and Zucchini 2015; Carnes and Lupu 2016) or presidential elections (e.g., Hainmueller et al. 2014). These studies find that voters have significant preferences over gender (e.g., Ono and Burden 2019), age (e.g., Horiuchi et al. 2018), occupational background (e.g., Carnes and Lupu 2016) and other valence related attributes of candidates (e.g., Franchino and Zucchini 2015). Focusing on voter preferences for candidates that stand to represent them in office is undoubtedly of great relevance. Yet, as candidates usually need to clear the hurdle of nomination first in most political systems, studying the preferences of those doing the nominations stands to generate new insights into selection outcomes, as it is selectors who determine whom the voters can face in the first place.

However, studies estimating party leader preferences remain rare and centered on single-member district electoral systems. Set in the US, Doherty et al. (2019) ask local party chairs of both major parties to assess which of two hypothetical candidates would be more likely to win the primary election. Yet, while we learn about how progressive party chairs think their primary electorate is, we do not learn anything about their or their primary electorate’s actual preferences in candidates. In the same vein, Broockman et al. (2019) employ conjoint experiments with local party leaders in the US, to find that they shun centrist and tend to nominate more extreme candidates due to their misperception of their electorates’ average ideological position. While these experimental studies shed some light on selectors’ preferences for candidates, the nomination for single-member district and the importance of anticipating voter reactions to candidates may limit the generalizability of these studies’ findings.

Overall, we have a rather poor understanding of what party selectors in general and those in closed-list systems in particular prefer in their candidates, what is driving their nomination choices and how their own background affect these choices. The next paragraph spells out my theoretical expectations about how selector will vote in a race of candidates for promising list positions based on their own and the characteristics of the competing candidates.

Informational Cues in Selecting Candidates

How and why do selectors decide and vote for or against a certain aspirant for candidacy? For selectors, nominating (personally perhaps unknown) candidates resembles a principal-agent challenge. How can they—as the principal—ensure the persons nominated—the agents—will act in their own and their party’s best interest? As in all delegation, this one too could entail conflicting interests between principals and agents, and principals’ limited information about agents. Parties or selectors can rely on two mechanisms to minimize agency loss; ex ante screening and ex post monitoring (e.g., Lupia 2003). While ex post monitoring resembles traditional understandings of democratic accountability potentially culminating in incumbent the de-selection, ex ante screening is used to evaluate candidates’ potential to disrupt or undermine the party’s agenda leading to the non-selection of obvious mismatches.

The selector’s interest is to nominate active candidates that will act on behalf of the selector and the wider party. Selectors dislike lazy and inactive candidates (see Bochel and Denver 1983) and those that go against the party too often, blurring the party’s label and thus undermining the party’s electoral performance (e.g., Greene and Haber 2015). Potential candidates, now, want to be nominated, otherwise they would not stand. They know, that selectors desire candidates that abide by the party leadership’s calls even if these go against the candidates’ own interests or beliefs. Thus, candidates have the incentive to signal obedience and loyalty, even if they already disagree with the leadership on certain topics.

For selectors, false signalling by potential candidates creates a dilemma. How can they know for certain whether any given aspirant will make a good candidate and member of parliament? Absent of institutionalized eligibility criteria by parties or otherwise meaningful information about candidates, I argue that selectors have to revert to certain low-cost informational cues about candidates’ valence and quality in evaluating potential candidates and in making their decision to select the “good type” (Fearon 1999).

Selectors can now follow two informational cues in detecting “good types”. First, they can rely on objective signals of quality and commitment to the party, which I call cues of valence. Secondly, they can follow cues of homophily, as individuals typically evaluate strangers as more likeable and more similar to oneself when they espouse similar demographics and attitudes (Hampton et al. 2018). While the first cue attempts to minimize agency-loss on the quality dimension, the second cue attempts to minimize it on an attitudinal dimension given the scarce information available.

As valence, I define those characteristics that are indicative of a candidate’s likely ability to perform as an effective and loyal legislator which are most desired by selectors (e.g., Bochel and Denver 1983; Galasso and Nannicini 2015). In other words, valence characteristics refer to overall candidate quality. Here, the easiest for selectors in terms of acquiring information at low costs is with incumbents. The latters’ performance over the course of the past legislative cycle provide selectors sufficient information to guide their decision. Not only can they easily learn about the candidate’s loyalty in voting, but also about their positions on issues debated controversially within the party and overall activity in the legislature. Given this advantage in the availability of low-cost information about a candidate, I would thus expect that selectors take the easy route and generally favor incumbents over new candidates.

Additional valence-related cues extend to other observable characteristics indicating expertise, quality or achievements in or for the party. Candidates that can prove a long track record as a party member, candidates holding a party or any elective office can credibly signal their experience in campaigns, commitment to the party or competence in the legislative realm. Finally, in mixed-member electoral systems in which candidates can compete either on one or on both tiers of the electoral system, selectors might take advantage of vetting processes and nomination outcomes at the other electoral tier that precede their nomination tasks. In most mixed-member electoral systems, nominal tier nominations precede the selection of list candidates. Selectors deciding on party-list nominations can thus take cues from nominal tier nomination outcomes. That is, in absence of information to the contrary, selectors may rely on the decision taken by co-partisan selectors in evaluating candidate quality.

The second cue selectors may rely on is socio-demographic homophily. According to social psychology, we evaluate strangers as more likeable, the more similar their demographics and attitudes are to our own (Hampton et al. 2018). Social homophily, i.e. the tendency to bond with similar individuals, has been found to stretch across the social categories of age, gender and education (Smith et al. 2014). However, selectors preferences for similar candidates is not limited to an affectional level but has rational underpinnings related to descriptive representation, too (e.g., Pitkin 1967). Not only are, say, younger selectors expected to evaluate younger candidates more positively for having similar tastes but also because they are more likely to advance policy content more relevant to younger people in general (McClean 2018). Hence, selectors may prefer candidates that exhibits similarity on characteristics salient to the selector (see also Niven 1998).

Which of these two informational cues selectors follow might be contingent on their experience as selectors. Following the law of curvilinear disparity (May 1973), I expect returning selectors to be higher up the party’s echelon relative to first-time selectors and for that reason to espouse more centrist, pragmatic and professional attitudes when it comes to candidate nomination.Footnote 2 While they might have joined the party out of ideological concerns, the time spent in and climbing up the ladder of the party organization turns experienced selectors into pragmatists that evaluate candidates more based on their value for the party brand (see also Wilson 1962; DeFelice 1981). First-time selectors, in contrast, are likely to be (yet) more driven by ideological zeal or personal convictions when it comes to candidate selection. I expect these attitudinal differences to translate into preference differentials for candidates. While experienced veteran selectors are expected to prefer candidates of greater quality or valence, first-time selectors are expected to prefer candidates based on socio-demograhics and ideology.

Research Design

In this section I first explain the case selection of Germany and why delegates of German parties were chosen for the experiment. This is followed by an introduction of the experimental design and data.

The German Case

The German system is an ideal case to elicit selectors’ true preferences over candidates unconfounded by contextual restrictions for a number of reasons. First, the electoral systems in the three states entail the combination of an equal share of seats allocated through single-member districts and closed-list proportional representation. While nominations for the nominal tier are decided at the local Kreisverband, party-lists are voted on and finalized at state-wide party delegate conventions. Each Kreisverband sends a number of delegates to these conventions depending on their membership register. In contrast to open-list or single-member district systems, selectors in closed-list systems are ceteris paribus less affected in their preference formation over candidates by their presumed electability (except perhaps for candidates on leading list positions, see Riera 2011)—i.e., by “imputed discrimination”—and more likely to follow own preferences. Secondly, nomination is carried out by party delegates in contrast to primary voters. Third, German parties have only minimal formal eligibility criteria for candidates and lack stricter ones that could eliminate a certain set of hypothetical candidates in terms of quality or party commitment. The German system thus enables us to uncover selectors’ preferences less biased by anticipating voter reactions or by a homogeneous pool of candidates that lack empirically in variation on relevant dimensions.

Lastly, candidates are free to compete at nomination conventions for desired list positions allowing to model this in a conjoint setup despite having a party list in this closed-list system. Yet, how competitive are these nominations in German parties? In other words, how frequent are so-called Kampfabstimmungen in the real world?Footnote 3 Usually regional party leaderships propose ordered lists of candidates on which delegates vote rank by rank for the promising positions and in brackets for lower list positions. In the Green party, however, it has traditionally been the case that bar any proposed list candidates declare their candidacies on the spot. Yet, in all parties it is possible that anyone present at these conventions may challenge a designated nominee for their proposed list position. Unfortunately, no comprehensive statistics are available on how often nominations are challenged.

As one of the few studies, Schüttemeyer and Höhne (2019)—analysing 1.275 list positions in all major parties—find that competitive nominations are most common in the Alternative für Deutschland (AfD) (88.6% of list positions competed),Footnote 4 followed by Die LINKE (54.7%), B90/Grüne (44.5%), FDP (16.8%) and far behind the SPD (4.2%) as well as CDU (1.1%) and the CSU (none), for which list positions are practically irrelevant. At the state level, information on competitiveness is even scarcer. Yet, the order of parties presumably is the same or similar. B90/Grüne for instance saw competition for 68% of their list positions at their party convention for the very state election in Brandenburg. The FDP, in contrast, had contested nominations for roughly 20% of its list positions in the state of Saxony. Although this could imply that results from left-leaning parties such as LINKE and B90/Grüne are the most relevant and consequential for actual nomination outcomes, the analysis below suggests that differences in selectors’ preferences between parties are negligible and that preferences in candidate characteristics seem universal.Footnote 5 What it does imply, however, is that actual nomination outcomes for these parties are likely to stronger reflect selectors’ preferences conditional on the pool of candidates.

The Conjoint Experiment

To elicit selectors’ preferences I ran a conjoint experiment with party delegates of five major German parties—the CDU, SPD, FDP, B90/Grüne and LINKE—between January and July 2019 in the run-up to three state elections in Brandenburg, Saxony and Thuringia.Footnote 6 Delegates were contacted through each parties’ state-level managing director (Landesgeschäftsführung) who centrally disseminated links to the online-based survey experiment. In total, 296 delegates completed the survey, resulting in a relatively high average response rate of ca. 20% for an elite study.

In the actual experiment respondents are confronted with a hypothetical situation in which they are asked to choose between two made-up candidates that compete in a so-called Kampfabstimmung for a promising list position. Respondents were presented a brief primer about the upcoming task. It stated that two candidates are competing for a promising list positions and that they had to select whom they would vote for.Footnote 7

The candidate profiles varied randomly along the following dimensions: age, gender, school education, occupation, district nomination, legislative experience, length of party membership, ideological position within the party, holding a party office and activity in local party chapter. Table 1 list all possible feature levels for each of the ten dimensions. A few combinations of these values are either unrealistic or illogical and were therefore prohibited from appearing in the profiles.Footnote 8 This task was repeated five times by each respondent, each time with a new pair of hypothetical candidates. Through this randomization of candidate profiles, this type of experiment eliminates systematic confounders and thus enables us to infer selectors’ actual preferences in candidates. Moreover, it minimizes social desirability by respondents and hence allows to analyze my hypotheses on valence- and homophily-oriented voting more credibly (e.g., Hainmueller and Hopkins 2014).

Table 1 Dimensions of candidate characteristics

Selectors’ Preferences and Candidate Quality

Do party selectors favor candidates with objective criteria of quality? Do they follow valence-cues when evaluating them? As an observational implication we should expect selectors to prefer active, experienced and committed candidates over inactive, inexperienced, or less committed ones. Figure 1 presents marginal means (MM) with 95% confidence intervals for feature levels of those dimensions indicating candidate quality, i.e. activity, legislative experience, length of party membership, party office and having obtained a nomination at the district level.Footnote 9

MM values describe the proportion of profiles selected that exhibit a given feature level (Leeper et al. 2020). Due to the random assignment of feature levels, a value of 0.5 indicate indifference, while values below 0.5 entail rejection of an attribute level and values above favorability. Compared to the often used average marginal component effects (Hainmueller et al. 2014), MM have the advantage of not requiring the setting of a reference category for calculation and display, and are better suited for subgroup analyses—which follow below.

Overall, the findings support my expectations on valence-related cues. Delegates significantly favor candidates of greater quality and commitment to the party. In particular, candidates rarely active, relatively new to the party, lacking legislative experience, not having a party office or nominal tier nomination are rejected. Put differently, selectors prefer active, experienced, long-serving and office-holding candidates with a nomination at the district level. Valence is taken seriously, as are previous vetting outcomes from nominations at the district level. Nominal tier nominees may be prefered for an additional reason, too. Parties in mixed-member systems are believed to field candidates even in futile districts to increase the party’s voteshare in the proportional representation tier (Manow 2015). Now, lest that self-interested candidates in hopeless districts realize their unfortunate lot and campaign only half-heartedly, thereby jeopardizing the party’s overall performance, parties and selectors have the incentive to push these nominees’ campaign efforts with promising list positions.

Fig. 1
figure 1

Preferences for valence-related candidate characteristics. Note: figure shows marginal means each with 95% confidence intervals for feature levels of valence related candidate attributes

Now, is the preference of selectors for candidates of quality conditional on their experience as selector? Are more experienced selectors looking out for candidates that benefit the party brand more? Are first-time selectors, in contrast, perhaps more strongly driven by ideological or personal preferences? Figure 2, now, presents MM for candidates’ characteristics separately for first-time and veteran candidates, i.e. delegates that have been elected as a delegate for at least the second time. This splits the sample nicely into roughly two equal parts among respondents without missingness on this variable.

Following May’s law, we should expect more significant effects for experienced selectors than for inexperienced ones. In general, differences are only minor. However, apart from candidates’ activity which first-time and veteran delegates prefer equally, differences in the preference for candidate experience, length of membership, party office and nominal tier nominations do exist that underscore my expectations. While veteran delegates reject inexperienced candidates and (though insignificantly by conventional standards) prefer incumbent members of state legislatures or of a municipal council, first-time delegates only prefer member of state legislatures. As the latter are expected to be less swayed by cues of quality compared with veterans, the preference for actual incumbents is not all too surprising as it possibly is the strongest cue of candidate quality for inexperienced selectors.

Subtle differences exhist in the preferences for candidates’ party membership and office as well. Veteran delegates reject candidates with the shortest party membership and those without party office, while effects for first-time delegates tend to be insignificant despite their larger number of observations. Nominal tier nominations, in contrast, crucially set both types of delegates apart. Veteran delegates clearly favor nominal tier nominees over candidates without such a nomination.

Based on the ten dimensions in this experiment, the ideal average candidate of first-time delegates would be a 30 years old female with a high school diploma working as a farmer. She would have obtained a district nomination and is an incumbent member of a state legislature. She would have joined the party six years ago and finds herself to the left in the party, holds a party office at the village level and is often active in her local party. For veteran delegates, this ideal candidate would not differ much. She would, however, be more likely a member of a municipal council, an ideological centrist and would have a party office at the municipality level (see Fig. 7 in the Appendix).

Fig. 2
figure 2

Preferences for valence-related candidate characteristics, by delegates’ veteran status. Note: figure shows marginal means each with 95% confidence intervals for feature levels of valence related candidate attributes separately for first-time and veteran delegates

Although differences appear to be rather small on average with one clear exemption—nominal tier nominations—the direction of these differences still complies with my theoretical expectations. Yet, when we compare the relative importance of all attributes, i.e. which of the attributes have the greatest impact on the likelihood of selecting that profile, we see no differences.Footnote 10 For both veteran and first-time delegates the five attributes with the greatest utility are activity, ideology, party membership, nomination, and experience. The difference between first-time and veteran delegates will be examined in more detail below.

Now, do other demographic and ideological characteristics of delegates affect their preferences in candidates? Whether selectors follow cues of homophily when evaluating candidates is analyzed in the next section.

Homophilous Preferences in Selectors

How does the demographic and political background of delegates affect their preferences for candidates? To test whether social homophily is a driver of delegates’ preferences, Table 3 shows MM obtained from fully specified estimations on split samples.Footnote 11 The samples were split along values on demographic and ideological characteristics of delegates corresponding more or less to the dimensions used in the candidate choice experiment. All splits divide the sample in two, with cut-off values chosen substantially and with an eye to obtaining roughly balanced samples where possible. Figure 3 presents subgroup analyses on delegates’ gender, age, education and ideology.

If selectors follow cues of homophily female delegates should prefer female candidates over male and male delegates male candidates over female ones. Yet, the top left panel in Fig. 3 shows that female and male delegates prefer female over male candidates. This result is in line with findings from the majority of candidate choice experiments (see Schwarz and Coppock 2020 for an overview), but at variance with actual gender imbalances in parliaments—and with previous findings on selectors’ decisions (e.g., Norris and Lovenduski 1995; Niven 1998). It could be that in the experimental setting delegates prefer female over male candidates out of concerns for the party’s image among a public sensitized by intense recent debates on equality, or, due to internalized party rules such as gender quotas. This receives some support by party differential effects, as male and female delegates of the centre-right parties without quotas, i.e. CDU and FDP, do not favor either gender.

The upper panel to the right explores whether age is a salient dimension for homophily. Delegates are seen as either young or old, depending whether they are older than the sample average of 43.2 years or not. While selectors of both age-groups reject the oldest candidate of 65 years, we see differences for the other two age categories. While older selectors prefer candidates of the middle-category, younger selectors prefer the youngest candidates.

Fig. 3
figure 3

Homophilous preferences of delegates for candidate gender, age, education and ideology. Note: Panels show marginal means each with 95% confidence intervals based on fully specified models. Delegates are classified as old or young along the sample’s mean age of 43.2. Delegates are classified as High-Educated if they have attended university and as Low-Educated otherwise. Delegates perceiving themselves right or left of their own party are classified as Right-leaning and Left-leaning, respectively

The bottom left panel examines the educational background.Footnote 12 University-educated delegates are classified as highly educated and as low educated otherwise. In fact, highly educated delegates prefer candidates with the highest school education and reject candidates with lower school education. Low-educated delegates, in contrast, seem indifferent to candidates’ education. For high-educated selectors, education has the second highest relative importance of all attributes, for low-educated education has the second lowest relative importance. For education, homophily appears to play a big role.

The bottom right panel, now, explores whether homophilous preferences respective demand for descriptive representation extends to ideological positions within the party. Selectors were asked to place themselves and their party on a 11-point left-right scale. Selectors that perceive themselves to the right (left) of their own party are classified as right-leaning (left-leaning). Selectors that see themselves in the centre of their party are excluded. The results are clear. Left-leaning selectors favor centrist and favor strongly leftist candidates while clearly rejecting rightist candidates. Right-leaning selectors, however, are indifferent toward the extremes and prefer centrist candidates, which might be attributable to progressives’ greater tendency to participate in survey research (Bundi et al. 2018).

These findings show that selectors tend to follow cues of homophily on certain dimensions, and do prefer candidates that resemble them in terms of age, education and ideology. Homophily in selection exhist and might reproduce outgroup effects (Niven 1998).

Veteran vs. First-Time Delegates

Veteran status appears to be more important for valence cues, but how about homophily? Are first-timers more prone to nominate candidates based on socio-demographic similarity? To avoid low-n subgroup analyses I have coded for each hypothetical candidate profile a valence- and homophily score. A profile can have a value of between 0 and 5 on the valence score. Whenever the profile has one of the following attribute levels on the five valence dimensions: district nomination, member of state legislature, 6 years of party membership, member of state party board and often active in local party chapter a 1 is added to the profile’s valence score.Footnote 13 The homophily score is constructed in interaction between profile and delegate characteristics. It is based on the four socio-demographic dimensions of gender, age, education and ideology. Whenever profile and delegate characteristics are identical or similar a 1 is added to the homophily scores.Footnote 14 Candidate choice is then regressed on veteran status interacted with each of these two scores in two separate linear probability models.Footnote 15 If veterans and first-time delegates differ in their preferences, the interaction terms should show significant effects.

Figure 4, now, plots both interactions reported in Table 3 in the Appendix. The left panels shows the marginal effects of veteran status across the range of possible valence cues. Compared to first-time delegates, veterans are less likely to nominate candidates that lack any valence-related characteristics. With increasing characteristics of valence, veteran delegates become more likely to nominate a candidate compared to first-time delegates. The number of socio-demographic similarities, however, do not affect the two types of delegates differently. Experienced selectors, thus, differ in their preferences from first-time selectors mainly in their stronger desire for candidates of greater valence.

Fig. 4
figure 4

Veteran delegate’s candidate choice contingent on number of valence and homophily cues. Note: Marginal effects obtained from linear probability models presented in Table 3 in the Appendix. Rug plot shows distribution of cases


Party selectors hold crucial sway in candidate selection. Their preferences for candidates can translate directly into who enters parliament. Examining the closed-list system of Germany and using a conjoint experiment, this study has explored selectors’ preferences in candidates, thus overcoming challenges for internal validity usually afflicting observational and traditional survey studies. Results support my expectations that selectors follow two types of low-cost cues in selection.

First, selectors follow cues of valence indicating candidates’ quality through their commitment to the party or legislative experience. Furthermore, this tendency is conditional on selectors’ experience. Experienced selectors tend to follow this cue more compared to first-time selectors. The second cue extends to preferred socio-demographics and ideological homophily. When evaluating candidates, selectors favor candidates exhibiting attributes similar to their own. Specifically, candidates’ age, education and ideology are important for selectors with similar characteristics. Somewhat surprisingly, both men and women selectors prefer women candidates. However, it is unclear on what grounds they do so—whether out of personal preferences or concerns for the party’s image.

Although mainly applying to closed-list systems, these findings have several implications beyond the German case. Experienced selectors opting for high-valence candidates and first-time selectors being less swayed by these cues is likely to hold true across different electoral systems. In fact, this difference in preferences corresponds to behaviors documented in cases with greater participatory elements, too, as for instance the US, where concerns for electability is less prevalent among amateur delegates (DeFelice 1981) and where primary voters—usually not part of the party’s higher echelons—tend to prefer more extreme candidates (e.g., Brady et al. 2007). A case in point are the US presidential primaries of 2016, in which the establishments of both major parties favored candidates less extreme or with greater experience.

A second implication pertains to the conventional wisdom that descriptive representation within selectorates will help in ameliorating demographic imbalances in candidates and legislators (Niven 1998), yet selectors’ experience may condition this relationship. Staffing selectorates with either inexperienced or experienced selectors might produce different nomination outcomes. As most signals of quality in candidates (especially incumbency) are unequally distributed across for instance gender, age and education, valence-oriented experienced selectors may impede legislative turnover and progressive public policy reform in the long-run.

Third, my findings have implications for the literature on candidate selection and representation as well as for political practioneers. My findings add to our understanding of candidate selection by showing what selectors prefer in candidates, namely signals of quality, competence and commitment to the party and how this is affected by selectors’ experience. Future studies might take the composition of selectorates in terms of experience and demographics more into account.

However, the findings of this study come with some limitations. The focus on candidates nominated for closed party-lists may obscure selectors’ differential preferences for candidates across different electoral formulas. Selectors may prefer other characteristics in other electoral circumstances or are more willing to compromise—perhaps mainly on socio-demographics—when it comes to candidate electability in systems with a stronger personal vote component. Many political parties, moreover, at least in Germany, have instituted formal (e.g., gender quota) and informal rules (priority for incumbents) that guide candidate nominations (see Reiser 2014). At this juncture, it is impossible to know whether selectors have internalized these rules as norms guiding their preferences for candidates or whether (in-) formal rules and selectors preferences have been merely codeterminous in the first place. On top of this, party lists are often drafted with an eye to balancing interests of various stake-holders in and outside the party—including regional party branches and interest groups—which might overrule the preferences of any single selector for any given list position. Moreover, as any survey-generated data, participation bias may affect results (see Bundi et al. 2018)—though arguably more the external than the internal validity thanks to the randomization of candidate profiles.


  1. Data and replication files can be found online in the Political Behavior’s Harvard Dataverse with this link:

  2. Logistic models examining veteran selector status using several covariates support the assumption that veteran status is associated with greater legislative experience, holding a party office and longer spells of party membership. All these are expected to correlate with higher positions within the party. See Table 2 in the Appendix.

  3. The term Kampfkandidatur roughly translates to “combat candidature” and describes a situation in which a nomination or party office is allocated through a competitive election, whereas usually nominees are merely confirmend by delegates.

  4. The high figures for the AfD can be explained by the party’s lack of incumbents and their maxime of applying direct democratic means.

  5. See Fig. 5 in the Appendix.

  6. The AfD selected their candidates in all-member conventions for the three state elections this study is based on. Unfortunately, while managing directors may send emails to delegates at their own behest, sending out emails to all members requires the permission of the party’s state leadership (Landesvorstand), which was not granted by any of the AfD state party branches.

  7. In German: “Im Folgenden werden Ihnen jeweils 5 hypothetische Paare von Kandidierenden präsentiert. Beide treten in einer Kampfabstimmung gegeneinander an und bewerben sich auf einen aussichtsreichen Listenplatz. Für wen würden Sie stimmen?” Emphasis as in the survey.

  8. Four combinations have been prohibited from appearing. These are secondary school and legal profession, secondary school and teacher, 7 Months and Member of State Legislature as well as 2 Years and Member of State Legislature. These restrictions are accounted for in the analysis.

  9. These estimates are obtained from a fully specified model which can be found in Fig. 6 in the Appendix. This figure also contains (essentially identical) results from specifications with frequency weights.

  10. Relative attribute importance measures how much a given attribute contributes to the total utility of a profile, whereby profile utility is equal to the linear predictions based on simple OLS estimates for all attribute levels. The greater the range in the partial utility of a single attributes’ levels, the greater this attribute’s relative importance. It is calculated by computing the absolute distance between the extreme values of an attribute’s levels, which is then divided by the sum of all attributes’ utility ranges.

  11. Figures of the complete models can be found in Figs. 8 and 9 in the Appendix.

  12. Granted, education could also be seen as a valence-related attribute, indicating cognitive aptitude, but is here grouped with socio-demographic characteristics (see Smith et al. 2014).

  13. Other values on these dimensions are coded as 0 for a more conservative estimation.

  14. For gender, 1 is added when profile and delegate gender is the same. For age, 1 is added when the profile’s age is the closest to the delegate’s age. For education, university-attendance is grouped with high school diploma. For ideology, left-leaning delegates are grouped with leftist profiles, and right-leaning delegates with rightist profiles.

  15. Linear probability models are estimated to obtain average marginal effects.


  • Asquer, R., Golden, M. A., & Hamel, B. T. (2019). Corruption, party leaders, and candidate selection: Evidence from Italy. Legislative Studies Quarterly, 45(2), 291–325.

    Article  Google Scholar 

  • Besley, T., Folke, O., Persson, T., & Rickne, J. (2017). Gender quotas and the crisis of the mediocre man: Theory and evidence from Sweden. American Economic Review, 107(8), 2204–2242.

    Article  Google Scholar 

  • Best, H., & Cotta, M. (2000). Parliamentary representative in Europe 1848–2000: Legislative recruitment and careers in eleven European countries. Oxford: Oxford University Press.

    Google Scholar 

  • Bochel, J., & Denver, D. (1983). Candidate selection in the labour party: What the selectors seek. British Journal of Political Science, 13(1), 45–69.

    Article  Google Scholar 

  • Brady, D. W., Han, H., & Pope, J. C. (2007). Primary elections and candidate ideology: Out of step with the primary electorate? Legislative Studies Quarterly, 32(1), 79–105.

    Article  Google Scholar 

  • Broockman, D. E., Carnes, N., Crowder-Meyer, M., & Skovron, C. (2019). Why local party leaders don’t support nominating centrists. British Journal of Political Science.

    Article  Google Scholar 

  • Bundi, P., Varone, F., Gava, R., & Widmer, T. (2018). Self-selection and misreporting in legislative surveys. Political Science Research and Methods, 6(4), 771–789.

    Article  Google Scholar 

  • Carnes, N., & Lupu, N. (2016). Do voters dislike working-class candidates? Voter biases and the descriptive underrepresentation of the working class. American Political Science Review, 110(4), 832–844.

    Article  Google Scholar 

  • Cheng, C., & Tavits, M. (2011). Informal influences in selecting female political candidates. Political Research Quaterly, 64(2), 460–471.

    Article  Google Scholar 

  • Cutts, D., Childs, S., & Fieldhouse, E. (2008). ’This is what happens when you don’t listen’. All-women shortlist at the 2005 general election. Party Politics, 14(5), 575–595.

    Article  Google Scholar 

  • DeFelice, E. G. (1981). Separating professionalism from pragmatism: A research note on the study of political parties. American Journal of Political Science, 25(4), 796–807.

    Article  Google Scholar 

  • Doherty, D., Dowling, C. M., & Miller, M. G. (2019). Do local party chairs think women and minority candidates can win? Evidence from a conjoint experiment. Journal of Politics, 81(4), 1282–1297.

    Article  Google Scholar 

  • Fearon, J. D. (1999). Electoral accountability and the control of politicians: Selecting good types versus sanctioning poor performance. In A. Przeworski, S. C. Stokes, & B. Manin (Eds.), Democracy, accountability, and representation. Cambridge: Cambridge University Press.

    Google Scholar 

  • Franceschet, S., & Piscopo, J. M. (2014). Sustaining gendered practices? Power, parties, and elite political networks in Argentina. Comparative Political Studies, 47(1), 85–110.

    Article  Google Scholar 

  • Franchino, F., & Zucchini, F. (2015). Voting in a multi-dimensional space: A conjoint analysis employing valence and ideology attributes of candidates. Political Science Research and Methods, 3(2), 221–241.

    Article  Google Scholar 

  • Galasso, V., & Nannicini, T. (2015). So closed: Political selection in proportional systems. European Journal of Political Economy, 40(Part B), 260–273.

    Article  Google Scholar 

  • Gallagher, M., & Marsh, M. (Eds.). (1988). Candidate selection in comparative perspective: The secret garden of politics. Thousand Oaks: SAGE.

    Google Scholar 

  • Greene, Z. D., & Haber, M. (2015). The consequences of appearing divided: An analysis of party evaluations and vote choice. Electoral Studies, 37, 15–27.

    Article  Google Scholar 

  • Hainmueller, J., & Hopkins, D. J. (2014). The hidden American immigration consensus: A conjoint analysis of attitudes toward immigrants. American Journal of Political Science, 59(3), 529–548.

    Article  Google Scholar 

  • Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2014). Causal inference in conjoint analysis: Understanding multidimensional choices via stated preference experiments. Political Analysis, 22, 1–30.

    Article  Google Scholar 

  • Hainmueller, J., Hangartner, D., & Yamamoto, T. (2014). Validating vignette and conjoint survey experiments against real-world behavior. Proceedings of the National Academy of Sciences of the United States of America, 112(8), 2395–2400.

    Article  Google Scholar 

  • Hampton, A. J., Fisher, A. N., & Sprecher, S. (2018). You’re like me and I like you: Mediators of the similarity-liking link assessed before and after a getting-acquainted social interaction. Journal of Social and Personal Relationship, 36(7), 2221–2244.

    Article  Google Scholar 

  • Hazan, R. Y., & Rahat, G. (2010). Democracy within parties: Candidate selection methods and their political consequences. Oxford: Oxford University Press.

    Book  Google Scholar 

  • Horiuchi, Y., Smith, D. M., & Yamamoto, T. (2018). Identifying voter preferences for politicians’ personal attributes: A conjoint experiment in Japan. Political Science Research and Methods, forthcoming.

  • Leeper, T. J., Hobolt, S. B., & Tilley, J. (2020). Measuring subgroup preferences in conjoint experiments. Political Analysis, 28(2), 207–221.

    Article  Google Scholar 

  • Lupia, A. (2003). Delegation and its perils. In K. Strøm, W. C. Müller, & T. Bergman (Eds.), Delegation and accountability in parliamentary democracies (pp. 33–54). Oxford: Oxford University Press.

    Chapter  Google Scholar 

  • Manow, P. (2015). Mixed rules, mixed strategies. Candidates and parties in Germany’s electoral system. Colchester: ECPR Press.

    Google Scholar 

  • May, J. D. (1973). Opinion structure of political parties: The special law of curvilinear disparity. Political Studies, 21(2), 135–151.

    Article  Google Scholar 

  • McClean, C. T. (2018). Does it matter that politicians are older than their constituents? Yes. Paper presented at the Harvard Symposium on Japanese Politics in Cambridge, 29th August.

  • Niven, D. (1998). Party elites and women candidates. Women & Politics, 19(2), 57–80.

    Article  Google Scholar 

  • Norris, P., & Lovenduski, J. (1995). Political recruitment—Gender, race and class in the British parliament. Cambridge: Cambridge University Press.

    Google Scholar 

  • Ono, Y., & Burden, B. C. (2019). The contingent effects of candidate sex on voter choice. Political Behavior, 41, 583–607.

    Article  Google Scholar 

  • Pitkin, H. F. (1967). The concept of representation. Berkeley: University of California Press.

    Book  Google Scholar 

  • Put, G.-J., Gouglas, A., & Maddens, B. (2015). Candidate selection, intraparty competition and incumbency turnover: Analysis of the Belgian lower house elections. Paper presented at the 73th Annual Conference of the Midwest Political Science Association in Chicago, April 16–19 .

  • Rahat, G., Hazan, Y., & Katz, R. S. (2008). Democracy and political parties. On the uneasy relationship between participation, competition and representation. Party Politics, 14(6), 663–683.

    Article  Google Scholar 

  • Rehmert, J. (2020). Candidacy eligibility criteria and party unity. Comparative Political Studies, 53(8), 1298–1325.

    Article  Google Scholar 

  • Reiser, M. (2014). The universe of group representation in Germany: Analysing formal and informal party rules and quotas in the process of candidate selection. International Political Science Review, 35(1), 55–66.

    Article  Google Scholar 

  • Riera, P. (2011). Chapter four: Closed party list. In J. Colomer (Ed.), Personal representation. The neglected dimension of electoral systems (pp. 55–79). Colchester: ECPR Press.

    Google Scholar 

  • Rogowski, J. C., & Langella, S. (2014). Primary systems and candidate ideology: Evidence from federal and state legislative elections. American Political Research, 43(5), 846–871.

    Article  Google Scholar 

  • Schüttemeyer, S. S., & Höhne, B. (Eds.). (2019). Durch das Nadelöhr der Demokratie. Die Kandidatenaufstellung der CDU zur Bundestagswahl 2017. Berlin: Institut für Parlamentarismusforschung.

    Google Scholar 

  • Schwarz, S., & Coppock, A.. (2020). What have we learned about gender from candidate choice experiments? A meta-analysis of 42 factorial survey experiments. Working Paper.

  • Shugart, M. S., Valdini, M. E., & Suominen, K. (2005). Looking for locals: Voter information demands and personal vote-earning attributes of legislators under proportional representation. American Journal of Political Science, 49(2), 437–449.

    Article  Google Scholar 

  • Smith, D. M., & Tsutsumi, H. (2014). Candidate selection methods and policy cohesion in parties: The impact of open recruitment in Japan. Party Politics, 22(3), 339–353.

    Article  Google Scholar 

  • Smith, J. A., McPherson, M., & Smith-Lovin, L. (2014). Social distance in the united states: Sex, race, religion, age, and education homophily among confidants, 1985 to 2004. American Sociological Review, 79(3), 432–456.

    Article  Google Scholar 

  • Vandeleene, A. (2014). Gender quotas and ’women-friendly’ candidate selection: Evidence from Belgium. Representation, 50(3), 337–349.

    Article  Google Scholar 

  • Verge, T., & Claveria, S. (2016). Gendered political resources: The case of party office. Party Politics, 24(5), 536–548.

    Article  Google Scholar 

  • Wilson, J. Q. (1962). The amateur democrat: Club politics in three cities. Chicago: University of Chicago Press.

    Google Scholar 

Download references


Open access funding provided by University of Zurich.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Jochen Rehmert.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.



See Tables 2 and 3 and Figs. 5 , 6, 7, 8, and 9.

Table 2 Logit Estimation of Veteran Delegates
Fig. 5
figure 5

Full model by party affiliation

Fig. 6
figure 6

Full model, with and without frequency weights for party affiliation. Note: Panels show marginal means each with 95% confidence intervals based on fully specified and pooled models. Right panels uses frequency weights for party affiliation based on numbers on delegates set forth in regional party constitutions

Fig. 7
figure 7

Full model by delegates’ veteran status

Fig. 8
figure 8

Full model by delegates’ gender and age

Fig. 9
figure 9

Full model by delegates’ education & ideology

Table 3 Linear probability model: valence vs. homophilous cues in candidate choice

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Rehmert, J. Party Elites’ Preferences in Candidates: Evidence from a Conjoint Experiment. Polit Behav (2020).

Download citation

  • Accepted:

  • Published:

  • DOI:


  • Candidate selection
  • Conjoint experiment
  • Party elites
  • Candidates