Why Study Representation Preferences?

There are several good reasons for studying citizens’ representation preferences. For one, if we as political scientists are committed to taking seriously what our fellow citizens think—if we believe, as Sabl (Reference Sabl2015, 355) puts it, that “what ordinary people value is worth a provisional respect”—then we should also have an interest in exploring how citizens want to be represented by elected politicians. A more specific theoretical argument is available, too. This is inspired by the innovative new “constructivist” theories of political representation (major statements are found in Disch Reference Disch2021; Saward Reference Saward2010).

Unlike conventional theories of representation, constructivism foregrounds the performative and creative aspects of political representation. As Saward (Reference Saward, Castiglione and Pollak2019, 271) puts it, constructivism assumes that “representation’s political presence arises primarily by virtue of its being done—practiced, performed, claimed. Representative roles and relations gain a presence in our politics because … actors make claims to speak for others.” For constructivists, it is not elections that bring representative relationships into existence, but a “claim” by an actor to speak on behalf of a particular group. Importantly, the addressees of such claims are also thought to play an active role in co-constituting representative relationships. Indeed, for constructivists, “[a]udiences accepting (or rejecting or contesting or ignoring) representative claims is a crucial part of the dynamics of political representation” (Saward Reference Saward2010, 27).

For many constructivists, moreover, the legitimacy of representation hinges on whether a political actor’s “claim” to speak for a particular constituency is accepted, rather than, say, rejected or ignored, by that constituency (Montanaro Reference Montanaro, Castiglione and Pollak2019, 193-94; Saward Reference Saward2010, 146). Indeed, some argue that “[g]iven the burdens of judgment and the extraordinarily difficult epistemological issues involved in forging independent criteria of legitimacy” (Saward Reference Saward2010, 146), this is the most democratic way of ascertaining whether a representative relationship is legitimate (also see Wolkenstein Reference Wolkenstein2024, 283-6). As Saward (Reference Saward, Castiglione and Pollak2019, 288) suggests,

Such acceptance will only be forthcoming, however, when what politicians say and do resonates with citizens’ normative expectations about political representation, that is, how they want their representatives to present themselves and act in public. Note that conventional “selection models” of representation make similar assumptions, conjecturing that citizens seek out “good representative types” by using various personal characteristics as cues for how well politicians might represent them (see Fearon Reference Fearon, Przeworski, Stokes and Manin1999, 59; Mansbridge Reference Mansbridge2009, 381).

If this is correct, then studying how citizens would like to be represented by political actors promises to teach us a lot about why and when claims to representation may be viewed as legitimate. Indeed, as Wass and Nemčok (Reference Wass and Nemčok2020) put it, a sound understanding of citizens’ expectations of representatives and whether representatives fulfill these expectations “is important knowledge for those interested in the legitimacy of democratic processes” (333) more generally. If we treat representation as a relationship that is co-constituted by citizens and politicians, then it could even be argued that the representation preferences of citizens should influence which aspects of representation we empirically study. After all, if certain aspects are relatively unimportant to citizens, studying them will not tell us much about the legitimacy of representation.

Existing Work on Representation Preferences

Thus far, the literature on citizens’ representation preferences only makes up a small subset of the overall literature on representation (e.g., see Arnesen, Duell, and Johannesson Reference Arnesen, Duell and Johannesson2019; Wolak Reference Wolak2017; Campbell et al. Reference Campbell, Cowley, Vivyan and Wagner2019; Costa Reference Costa2021; Harden Reference Harden2015; McMurray and Parsons Reference McMurray and Parsons1965; Vivyan and Wagner Reference Vivyan and Wagner2016; Bowler Reference Bowler2017; Méndez-Lago and Martínez Reference Méndez-Lago and Martínez2002; Carman Reference Carman2006; Rosset, Giger, and Bernauer Reference Rosset, Giger and Bernauer2017; Bøggild Reference Bøggild2020; Wass and Nemčok Reference Wass and Nemčok2020). The question most often addressed in this literature is arguably whether citizens prefer either the trustee or delegate model of representation. While trustee representatives follow their own judgments, and thus are less responsive to the demands of citizens, delegates defer to their constituents’ views and are responsive to them. Most studies have found that voters prefer aspects of the delegate over the trustee model (Carman Reference Carman2006, Reference Carman2007; Barker and Carman Reference Barker and Carman2012; McMurray and Parsons Reference McMurray and Parsons1965; Rosset, Giger, and Bernauer Reference Rosset, Giger and Bernauer2017; Davidson Reference Davidson1970; Bowler Reference Bowler2017; Wolak Reference Wolak2017; Dassonneville et al. Reference Dassonneville, Blais, Sevi and Daoust2021).Footnote ¹ Some work has pitted “partisan representation” as a third model against trustee and delegate representation, demonstrating that citizens prefer both delegates and trustees over partisan representatives, who resolutely follow the party line rather than their own views or the views of their constituency (Bøggild Reference Bøggild2020). Recent studies suggest that the public generally dislikes representatives who loyally toe the party line (e.g., Campbell et al. Reference Campbell, Cowley, Vivyan and Wagner2019; Carson et al. Reference Carson, Koger, Lebo and Young2010).

A second major theme in the literature on citizens’ representation preferences is whether citizens want representatives to focus their work on their district (e.g. through pork-barrel politics or constituency service) versus on national policy issues. The findings are mixed and probably depend to some extent on question wordings and research designs. While Lapinski et al. (Reference Lapinski, Levendusky, Winneg and Jamieson2016) find evidence that citizens most value legislators who represent them on “salient national issues,” Vivyan and Wagner (Reference Vivyan and Wagner2016) report that legislators who balance their efforts between national and constituency-level work gather the most citizen support on average. Doherty (Reference Doherty2013) even demonstrates that when evaluating concrete instances of legislators’ behavior, citizens actually prefer a district focus over a national focus (see also Wolak Reference Wolak2017). Preferences over the focus of representation also vary considerably by individual-level characteristics: those in economic need, with lower education, and ethnic minorities (Harden Reference Harden2015; Griffin and Flavin Reference Griffin and Flavin2011; Lapinski et al. Reference Lapinski, Levendusky, Winneg and Jamieson2016) as well as respondents with a local as opposed to cosmopolitan orientation (Vivyan and Wagner Reference Vivyan and Wagner2016) tend to prefer a district focus over a national focus.

While the literature on representation preferences has made important contributions, it also has limitations. First, many studies use conceptualizations of representation that are considered misleading or outdated by representation theorists. Take the literature on preferences for delegate versus trustee representation. Theorists have widely rejected this distinction. Rehfeld (Reference Rehfeld2009), for example, argues that three more fundamental distinctions in representation are conflated in these models: 1) whether representatives are sensitive to electoral sanctions; 2) whether they pursue the good of a group (e.g., their district) or of society as a whole; and 3) whether they rely on others’ (e.g., their constituents) or their own judgments. This also means that multiple other forms of representation are conceivable, above and beyond trustee and delegate representation (e.g. representatives who aim for the good of society but are motivated by a desire to get re-elected by their constituency), and research on representation preferences should take these forms seriously.

The literature on preferences over national versus district representation could also benefit from engaging more closely with representation theory. One key insight of recent theoretical contributions is that representation and electoral accountability may sometimes come apart (e.g., Mansbridge Reference Mansbridge2003; Saward Reference Saward2010). Representatives may represent individuals or groups from other districts that stand in no electoral relationship with them. This is relevant to the national versus district distinction, because a representative who focuses on national issues thereby also needs to focus on constituents who did not elect them (e.g., constituents residing in other districts). Asking citizens only about their preferences regarding their local MP thus risks missing something important. The theoretical framework we present in the next section integrates these insights.

A second limitation of existing work on citizens’ representation preferences is that the focus is typically on only one or two dimensions of representation (but see Harden Reference Harden2015; Wolak Reference Wolak2017). Much work implicitly restricts citizens’ representation preferences to the substantive aspects of representatives’ behavior, such as their policy positions or the constituency service they perform (see e.g., Vivyan and Wagner Reference Vivyan and Wagner2016; Lapinski et al. Reference Lapinski, Levendusky, Winneg and Jamieson2016; Griffin and Flavin Reference Griffin and Flavin2011). In contrast, other relevant aspects of representation—for instance, whether representatives resemble citizens on descriptive characteristics, how they speak about things, or whether they present themselves as rebels who defy the party line—are rarely addressed. Recent work has tried to address some of these aspects (Campbell et al. Reference Campbell, Cowley, Vivyan and Wagner2019; Costa Reference Costa2021; Arnesen, Duell, and Johannesson Reference Arnesen, Duell and Johannesson2019; Bøggild Reference Bøggild2020; Harden Reference Harden2015), but our knowledge about citizens’ preferences regarding alternative dimensions of representation is still rather limited. To address this, we employ a multidimensional understanding of representation, distinguishing six different dimensions.

Third and relatedly, existing work has rarely assessed the relative importance of different dimensions of representation to citizens. Many studies focus only on a single dimension of representation or pit two models of representation against each other (e.g., delegate versus trustee). Some research helpfully covers preferences on several dimensions, identifying individual or majority preferences on each (e.g., policy versus service focus, pork-barrel versus fair-share allocation), but their relative importance to citizens is not tested (e.g., Wolak Reference Wolak2017; Harden Reference Harden2015; Bengtsson and Wass Reference Bengtsson and Wass2010). Only some recent studies address that issue. Costa (Reference Costa2021) uses conjoint experiments to compare the relative importance of three dimensions of representation, showing that affective partisan rhetoric is less important to citizens’ representation preferences than policy congruence or constituency service. Arnesen, Duell, and Johannesson (Reference Arnesen, Duell and Johannesson2019) show that voters focus more on whether political candidates have particular policy positions they like rather than what descriptive characteristics they share with them.

Bracketing the relative importance of different dimensions of representation is problematic for two reasons. First, when preferences on each dimension are measured independently (e.g., with different survey items), respondents may simply draw inferences about other dimensions and let them shape their response. For instance, if they are asked whether they like representatives who follow their own judgment, they might also assume that such representatives are unresponsive, which they could dislike. Consequently, they may disapprove of representatives who are following their own judgment because of the unobserved confounder—the mental link to representatives’ assumed lack of responsiveness. This leads to biased estimates of citizens’ preferences.

Second, understanding how important different dimensions of representation are to citizens is crucial for the normative conclusions we draw from our findings. For example, the finding that citizens prefer delegates over trustees has vastly different implications depending on how important this distinction is to citizens. If this is what they care most about, then perhaps politicians should try to act more like delegates, and political scientists should study the workings of delegate representation. Yet if citizens actually care most about whether representatives share descriptive characteristics with them (e.g., age, gender), then practitioners and scientists should devote more attention to descriptive aspects of representation than the delegate-trustee dichotomy. To overcome this limitation of the literature, we use a conjoint experiment to directly compare the average relative importance of different dimensions of representation.

Conceptualizing Six Dimensions of Representation

We focus on six different dimensions of representation that figure prominently in contemporary theoretical research on representation. The first two, substantive and descriptive representation, which were initially defined by Pitkin (Reference Pitkin1967), are possibly the most widely used conceptions of representation in empirical scholarship (see Wolkenstein and Wratil Reference Wolkenstein and Wratil2021, 864-5). The latter four conceptions of representation—surrogation, justification, personalization, and responsiveness—are derived from Wolkenstein and Wratil’s (Reference Wolkenstein and Wratil2021) recent attempt to translate the theoretical insights of leading representation theorists (esp. Mansbridge Reference Mansbridge2003; Rehfeld Reference Rehfeld2009; Saward Reference Saward2010) into operationalizable conceptions of representation. Consistent with both the theoretical literature on which we build and our empirical interest in individual citizens’ representation preferences, all of these dimensions of representation are conceptualized with individual constituent-representative relationships in mind. They are not addressed to representation at the collective level, which is typically conceived in terms of a congruence between policy and average public opinion. However, as we will explain more later, representation at the individual level is linked to collective-level representation.

We start with substantive representation, focusing on what we consider the conceptual core of Pitkin’s (Reference Pitkin1967) original definition: representation as “acting in the interest of the represented” (209). In our reading, substantive representation thus defined does not contain any stipulation about the mechanisms that lead representatives to act in accordance with the interest—or better—preferences (we avoid the controversial assumption of objective interests) of the represented. This contrasts with many popular conceptualizations of substantive representation, which conceptually link substantive representation to electoral mechanisms: representatives are thought to act in line with what constituents want in order to forestall electoral sanctions—think of the classic “delegate” conception of representation or what Mansbridge (Reference Mansbridge2003) calls “promissory representation.” Pitkin does not make this link; given that she (1967, 209) envisions representatives to “act independently” on the basis of their own “discretion and judgment,” it seems in fact that she leaves room for varying levels of substantive representation, as well as assuming that substantive representation can occur without the threat of electoral sanction (on this, also see Fearon Reference Fearon, Przeworski, Stokes and Manin1999; Mansbridge Reference Mansbridge2009).

This makes good sense. After all, representatives might do what their constituents want for reasons other than wanting to be re-elected (e.g., because they are intrinsically motivated to follow their constituents’ preferences, Mansbridge Reference Mansbridge2003 dubs this “gyroscopic” representation), or they might substantively represent certain people who cannot vote for (and hence sanction) them because they reside in a different electoral district (what we call surrogation).Footnote ² So, here is what we are empirically interested in: we seek to study how important it is for citizens that their representatives advance their policy preferences, irrespective of the mechanism(s)—electoral or not—that lead to representatives advancing an individual’s policy preferences.

We stress again that our individual-level conception of substantive representation is different from how substantive representation is often conceptualized in collective-level studies. These latter studies tend to conceptualize substantive representation in terms of policy responsiveness to average public opinion (e.g., Soroka and Wlezien Reference Soroka and Wlezien2010; Wlezien Reference Wlezien2017). At the level that we are interested in, this conceptualization of substantive representation does not make much sense, since an individual citizen’s policy preferences may well diverge from public opinion at the aggregate level. Suppose I individually prefer policy P1, but the majority of the public endorses policy P2. Suppose further that the government or majority of elected representatives are responsive to average public opinion and implement policy P2. In this case, I would not be substantively represented by the government or majority of elected representatives, although there is substantive representation as defined in the collective-level literature. In fact, I would only be substantively represented if politicians were not responsive to average public opinion. Thus, substantive representation at the individual level is not the same as substantive representation at the collective level, and we only seek to study the individual level.

The second traditional conception of representation that we focus on, descriptive representation, has also typically been linked to the work of Pitkin. For Pitkin (Reference Pitkin1967, 60), descriptive representation occurs when “a representative body is distinguished by an accurate correspondence or resemblance to what it represents.” At the individual level, the notion of descriptive representation is usually employed to describe the extent to which single representatives “look” or “are” like their constituents with respect to gender, race, class, etc. With this in mind, we want to assess how important it is for citizens that representatives actually resemble them, with a particular focus on the extent to which demands for descriptive representation vary across different groups of citizens.

Leading theorists have highlighted the importance of additional conceptions of political representation, beyond those of descriptive and substantive representation (see espescially Mansbridge Reference Mansbridge2003; Rehfeld Reference Rehfeld2009; Saward Reference Saward2010). Building on the groundwork of Wolkenstein and Wratil (Reference Wolkenstein and Wratil2021), we investigate citizens’ preferences regarding the following four additional dimensions.

First, surrogation was defined by Mansbridge (Reference Mansbridge2003, 522) as “representation by a representative with whom one has no electoral relationship—that is, a representative in another district.” In addition to Mansbridge’s notion of territorial surrogation, where the surrogate representative is the representative of a different electoral district, we also conceptualize partisan surrogation, which occurs when a constituent expects representation from a representative of a party for whom she did not vote. For instance, during the U.S. presidential campaign in 2020, Joe Biden appealed to such expectations of partisan surrogation, reaching out to Republican voters: “I’m running as a Democrat, but I will be an American president. Whether you voted for me or against me, I will represent you” (on Twitter, October 2, 2020). We seek to examine how important it is to citizens that they have direct electoral relationships with their representatives, considering both territorial and partisan surrogation.

Second, justification is defined as the degree to which representatives, when justifying their actions, refer to the good of the whole citizenry versus particularistic goods of some societal group(s). Justification, as understood here, is therefore about a specific contrast regarding how representatives justify, rather than about whether they justify/explain their decisions at all (e.g., Esaiasson, Gilljam, and Persson Reference Esaiasson, Gilljam and Persson2017, Reference Esaiasson, Gilljam, Persson, Narud and Esaiasson2013). This contrast tracks widely shared normative concerns in political theory and is derived from Rehfeld’s (Reference Rehfeld2009) distinction between pluralist (aiming for the good of a group) and republican (aiming for the public good) aims of representatives. According to Rehfeld, this is one of the three central distinctions underlying trustee and delegate models of representation, with the trustee aiming for the good of the whole and the delegate aiming for the good of a part. Wolkenstein and Wratil (Reference Wolkenstein and Wratil2021) reconceptualize this distinction in terms of political justification, focusing on the claims of representatives about whose good they aim to serve with their actions and policies. With regard to citizens’ attitudes, our primary concern is with studying whether citizens want representatives to explain in public speech how their policies and decisions affect and benefit society as a whole, or whether they would rather like to see them explain how their policies and decisions affect and benefit the particular socio-economic, ethnic, age, gender, etc. groups that citizens identify with.

Third, personalization refers to the extent to which representatives and constituents regard representatives as individual persons versus agents of their party. Strong forms of personalization involve representatives emancipating themselves from their party. They might present themselves as capable of making the right decisions individually, and without guidance from the party—or as resolute followers of their constituents, who are ready to go against the party line (Mansbridge Reference Mansbridge2009, 381). The key issue we want to shed light on is whether citizens prefer the “representative-as-party-member,” who acts in accordance with the party platform or the commands of their party leader—or conversely find more appealing the “representative-as-individual” who presents themself as an “independent leader,” “spokesperson of her constituents,” or even “party rebel.”

Fourth, responsiveness captures what we call, with Rehfeld (Reference Rehfeld2009), the degree to which representatives are sensitive to electoral sanctions. Though linked to substantive representation (at least in democracies), this is a distinctive aspect of representation. It is not about whether representatives follow the views of their constituents per se, but about whether they do so in order to forestall electoral sanctions. Recall that sanction-sensitivity thus understood is often assumed to be the primary mechanism that ensures that representatives act in line with the substantive policy preferences of their constituents. But since we separate substantive representation’s conceptual core from assumptions about incentive-generating mechanisms, we also study separately citizens’ attitudes about sanction-sensitivity. Specifically, we want to investigate whether citizens prefer representatives who evidently care about being re-elected (e.g., by stating it or changing course under electoral pressure) versus representatives who are unfazed by the threat of being unelected (e.g., because they prioritize “what they think is right” over short-term electoral success or failure).

As with all other dimensions of representation that we focus on, responsiveness is pitched at the level of individual representative-constituent relationships, not at the collective level, where (as mentioned earlier) “responsiveness” is often conceptualized as policy responsiveness to average public opinion. Notice, however, that there is a connection between our understanding of responsiveness and public opinion at the collective level. Effective electoral sanctioning requires that a critical mass of citizens votes in a particular way; an individual vote usually has little impact on the result. Thus, a real threat of being electorally sanctioned—to which a representative could be sensitive—only exists when public opinion at large (and not an individual constituent) finds the representative’s record wanting. Our conception of responsiveness also has an elective affinity with another key concept in the literature, namely electoral accountability. Yet while the literature on electoral accountability focuses more on whether citizens (can) hold representatives accountable in elections (e.g., Manin, Przeworski, and Stokes Reference Manin, Przeworski and Stokes1999), our concern is with how representatives behave, or should behave, in the eyes of citizens. This is exactly what Rehfeld seeks to capture by conceptualizing sanction-sensitivity as an attribute of individual representatives. For this reason, we do not use the term electoral accountability, but stick with Rehfeld’s terminology.

Two final points. First, our six dimensions of representation are thought to capture different aspects of a single unified concept of representation. This means that several—possibly even all—six dimensions can be embodied in a single action (or a string of closely connected actions) performed by a representative. Think, for example, of a “rebellious” representative who votes against their own party on a particular bill B. In virtue of voting in this way, the representative exhibits a high level of personalization. At the same time, the representative substantively represents citizens with the same policy preference (i.e., against B), while not necessarily representing all of those citizens descriptively because that policy preference is mostly held by women and the representative happens to be a man. Downstream, the representative may further use a republican justification to justify their vote on social media, and this justification may resonate more with women than with men. Likewise, the representative may have voted against B because they are responsive, that is, sensitive to electoral sanctions, and the constituents in their district are against B as well. And women who feel represented by the representative may live outside of the representative’s district, receiving surrogate representation. All of the dimensions are present in this example; they simply capture different aspects of it.

Second, analytically distinguishing the six dimensions does not imply that we do not expect correlations between them, or that they cannot be empirically correlated in how representatives behave or in what citizens want from representatives. For instance, Wolkenstein and Wratil (Reference Wolkenstein and Wratil2021, 872) note that highly responsive representatives may also aim for more personalization, emancipating themselves from their party to portray themselves as spokepersons of their constituents. Similarly, citizens’ preferences on the different dimensions may be correlated. The only assumption we make here is that the dimensions are analytically separable at the theoretical level.

Research Design

Our central research questions are then: How do citizens want to be represented on each of the six dimensions of representation? And what dimensions matter more to them, relative to others, when forming their preferences about whether someone represents them?

To answer the first question, we develop and validate six new item batteries that measure citizens’ individual-level preferences on the six aforementioned dimensions of representation. This allows us to explore, on each dimension, which kind of representation the majority of citizens prefers and how preferences for representation are related to standard socio-demographic and political covariates. Note however that we do not know a priori whether citizens have coherent preferences on all our dimensions of representation at all, since they are derived from theory rather than from citizens’ own conceptions of representation. We return to this issue later.

To answer the second question, we use a paired conjoint experiment (Hainmueller, Hopkins, and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014),Footnote ³ in which respondents consider two hypothetical politicians—whose profiles vary along the six dimensions of representation—and select the one that they think would “better represent [them] in politics.” We use the distribution of responses to these comparisons to infer the importance of each dimension for determining the probability that a politician is selected as a representative. This of course means that our measures of importance are relative. That is, we cannot determine whether dimensions matter or not in any “absolute” sense, but only how important they are when compared to each other (i.e., when all are present and no others are present).

Conjoint Experimental Design

Our conjoint experiment consists of five choice tasks in which respondents consider vignette descriptions of two politicians and select who they believe better represents them in politics. After a short introduction, our respondents are asked to choose between pairs of hypothetical politicians by answering the question, “Which of these politicians do you think would better represent you in politics?” Each politician is characterized by six short sentences that correspond to the six dimensions of representation. We randomly vary the content of these sentences to manipulate the type of representation each politician offers on each dimension:

• • Substantive representation: We provide information on the politician’s position on a single political issue in a binary format (“supports” or “opposes” the policy). For each country, we selected six political issues that are a current feature of political debate, and which include both low and high salience topics (refer to online appendix section B, “Substantive,” for the set of issues). We also solicit respondents’ binary position on each of these issues pre-treatment. Our measure of substantive representation is defined by the combination of respondents’ issue positions and the positions of the hypothetical politicians. That is, we measure a hypothetical politician substantively representing a respondent when they share the same position on the issue that features in the conjoint comparison.

• • Descriptive representation: We provide information on two descriptive characteristics of the politician. We always include the gender of the two politicians in a comparison, but the second characteristic is chosen at random from ethnicity (migratory background in Germany), sexual orientation, or social class. Respondents’ characteristics on these dimensions of descriptive representation are surveyed pretreatment. Again, our measure of descriptive representation is based on the combination of respondent and politician characteristics: a politician is more descriptively representative of a respondent the more descriptive characteristics they have in common (we provide more detail later and in online appendix section G).

• • Surrogation: We provide information on the politician’s party affiliation as well as the electoral tier/district in which they were elected. Information on the party of the politician, in conjunction with the partisanship of the respondent (measured pretreatment) allows us to operationalize the concept of partisan surrogation, while information about the district in which the politician was elected allows us to operationalize the idea of territorial surrogation.

• • Justification: We provide a statement about whether a politician more regularly engages in either pluralist or republican justification. The central challenge in operationalizing pluralist justification is that the number of possible groups whose good can be invoked is large and could also vary between countries. We tackle this problem by saying that the representative justifies by referring to the good of “people like you” or a generic category of “particular groups.” The aim is to make people think of groups they identify with (“people like you”) or other groups that they can readily think of (“particular groups”).Footnote ⁴

• • Personalization: We provide a statement about how frequently the politician acts (votes, speaks, etc.) against or in line with their party.

• • Responsiveness: We provide a quote by the politician that either emphasizes or rejects sensitivity to electoral sanctions.

At the respondent level, we randomize the order (top to bottom) of the six sentences. At the level of the choice task, we randomly draw the second descriptive characteristic as well as one political issue such that both politicians in a given comparison are characterized by information on the same descriptive attributes and their positions on the same issue. At the profile level (each politician), we randomly draw the attribute levels.Footnote ⁵ In general, our attribute levels reflect binary contrasts (e.g., high versus low surrogation, high versus low personalization). However, we deviate from this structure for some dimensions (i.e., descriptive representation, where there are more than two ethnicities) and dimension-country combinations (e.g., territorial surrogation in Germany due to the two electoral tiers) to better reflect reality. For the personalization, responsiveness, and justification dimensions we use three to four different text implementations of each attribute level (e.g., three texts that indicate “high” and three that indicate “low” levels of personalization) to guard against the possibility that any finding may be due to particular word usage in a single text implementation (Blumenau and Lauderdale Reference Blumenau and Lauderdale2022; Fong and Grimmer Reference Fong and Grimmer2023).

In online appendix section B, we provide all of the different text implementations for each attribute level. In figure 1, we provide an example of one comparison seen by respondents in our experiment.

Figure 1 Example comparison of politicians from conjoint experiment

What Kind of Representation Do People Prefer?

In this section, we use descriptive results from our item batteries measuring individual-level preferences to identify what kind of representation the majority of citizens want on each dimension and how preferences vary by socio-demographic and political covariates. In the appendix, we extensively validate our item batteries. First, we use exploratory factor analyses to show that most of the items that are supposed to measure the same construct do load together empirically on a common factor and that we are able to capture preferences towards distinct dimensions of representation (for details, refer to online appendix section D). Second, we report good test score reliabilities for four of our six dimensions, with some lower consistency for our item batteries on personalization and responsiveness (refer to online appendix section E). The descriptive results for citizens’ preferences on personalization and responsiveness presented in this section should therefore be interpreted with some caution. Essentially, we cannot be sure that citizens hold coherent preferences regarding our conceptualization of these dimensions. Third, we show that respondents’ choices on the item batteries predict their choices in the conjoint experiment (refer to online appendix section K). For instance, comparing respondents with a preference against versus for personalization in the batteries, those who want more party independence also value more strongly a party-independent representative in the conjoint than others do. While results are stronger on some dimensions than others, they cross-validate our two instruments and provide some basic evidence of predictive validity of our item batteries.

Majority Preferences

To ascertain what the majority of citizens prefers on a dimension of representation, we calculate each respondent’s average answer to the items in a battery, reverse-coding items that were negatively formulated. “Strongly disagree” with a conception of representation is coded as “1” and “Strongly agree” as “5”, so that “3” represents the indifferent middle category of “Neither agree nor disagree.” For example, if a respondent’s average response for the personalization items is less than 3, they tend to prefer party loyalty over independence, while they prefer the party rebel over the loyalist if their average is above 3.Footnote ⁷

We plot kernel densities of the average preferences of citizens on each dimension and for each country in figure 2. Most evidently, this figure reveals how similar preference distributions on our batteries are across countries, as the densities in each column have a similar shape. Perhaps the most noteworthy difference is that there are more citizens in Germany who have little aversion or even a preference for surrogation than in the two other countries. This may well be reflective of the German electoral system, in which territorially-defined electoral accountability relationships are less salient due to the proportional tier and the option for representatives to run in both tiers.

Figure 2 Densities of preferences on each dimension

In terms of majority preferences, we can clearly observe that a majority of people want substantive representation as opposed to a representative who acts incongruent with their policy wishes. Depending on the country, 87%–90% are on the side of preferring substantive representation. In contrast, citizens are much more split on descriptive representation, even if clear-cut views (i.e., towards the ends of the scales) are rare. In each country, a majority tends to not want the representative to resemble them personally on descriptive characteristics. This majority is 54% of voters in Germany and up to 72% in the United Kingdom. On surrogation, there is a clear aversion towards representation by politicians elected in other districts or from parties one did not vote for, with 76%–81% preferring representation with electoral relationships when given the choice.

On justification, we find a very clear majority preference for republican justification (from 77% in the United States to 83% in the United Kingdom). Regarding personalization, the figure shows some more mass towards party independence, with 54% preferring the independence in the United States and up to 62% in Germany. Finally, and perhaps surprisingly, we find that between 71% (United Kingdom) and 83% (Germany) of citizens want their representatives to be unresponsive to electoral sanctions. When separating responsiveness from substantive representation, it is clearly substantive representation that people like—not representatives’ sensitivity to re-election incentives.

Correlates of Representation Preferences

The item batteries also allow us to investigate the individual-level covariates of representation preferences. Here, we use factor scores from a set of exploratory factor analyses as our individual-level measures of representation preferences. For each item battery, we estimate a single latent factor based on the polychoric correlation matrix of our items, and use the estimated factor scores for each respondent as our dependent variable in the following analyses.Footnote ⁸ Figures 3, 4, and 5 present estimates from a series of bivariate regression models, in which we regress the factor scores on single covariates that are plausibly related to people’s representation preferences.Footnote ⁹ In these plots, each panel represents a separate regression model, where columns indicate the dependent variable, rows indicate the covariate of interest, and the y-axis indicates the levels of the relevant covariate. The x-axis gives the effect of the relevant covariate, either as a standardized regression coefficient for continuous covariates such as “Ideology”, “Trust”, and “Democratic Satisfaction,” or as the difference between the relevant covariate level and a baseline category.

Figure 3 Correlates of factor scores: U.K. sample

Figure 4 Correlates of factor scores: U.S. sample

Figure 5 Correlates of factor scores: German sample

Several patterns emerge from these analyses that are noteworthy. First, we find that there is a pronounced relationship between ethnicity (in the United Kingdom) and race (in the United States) and citizens’ demands for descriptive representation. Compared to white people, both Black and Asian respondents in the United Kingdom and Black and Hispanic respondents in the United States are significantly more likely to favor representatives who share some of their own descriptive characteristics. We do not ask German respondents to provide details of their own ethnicity, but we also find German respondents with a migratory background are slightly more in favor of greater descriptive representation than those without such a background, but this result is not statistically significant.

Second, across all three countries, demands for personalized representation and for responsiveness correlate with respondent age. Older respondents want their representatives to differentiate themselves more from their political parties and be less concerned about electoral sanctions, while younger respondents favor representatives who act more in accordance with their party platforms and spend effort on securing their re-election. However, one concern is that our batteries on these two dimensions have relatively low consistency. Hence, citizens may not have coherent preferences on these dimensions or conceptualize them significantly differently than we do.

Third, there is an especially clear relationship between levels of political interest and demand for substantive representation. While it is unsurprising that respondents who have greater interest in politics are more likely to favor representatives who advance policies of which they approve, the consistency of this relationship across the three countries in our sample is striking. We also find that more politically-engaged respondents are also more likely to be in favor of representatives who provide republican, rather than pluralist, justifications.

Fourth, we also find that there is a negative relationship between satisfaction with democracy and demand for personalized representation in all three of the countries in our data. One explanation for this relationship is that those citizens who are dissatisfied with democracy are also likely to be opposed to the partisan institutions that are often the most visible actors in democratic politics, and therefore are more likely to favour representatives who find ways to distinguish themselves from their parties.

Which Dimensions of Representation Matter Most to Voters?

We now describe the analysis strategy and report the results for our conjoint experiment, which aims to measure the relative importance of the different dimensions of representation to voters’ judgements about which types of politicians better represent them in politics.

Conjoint Analysis Results

We plot the AMCEs of all our attributes for each country in figure 6. This figure demonstrates that the most important dimension—in our comparison of six dimensions—for whether people think that someone represents them in politics is substantive representation. A politician who shares the respondent’s view on a policy issue is between 13 and 20 percentage points (depending on the country) more likely to be selected as the better representative than a politician who does not share the respondent’s view. The next two most important dimensions are partisan surrogation and personalization. Citizens are clearly more likely to see a politician from their favored party as their representative than a politician from another party. Politicians from parties other than the one the respondent voted for are between 11 and 15 percentage points less likely to be selected as the better representative compared to politicians from the respondent’s party. Citizens choices are also influenced by whether the politician is party-independent, with loyalist politicians having much lower selection probabilities. On average, politicians with high personalization are approximately 8–11 percentage points more likely to be selected as representatives compared to politicians who toe the party line.

Figure 6 Average Marginal Component Effects

In contrast, we find much smaller and more mixed effects for the other dimensions of representation. Descriptive representation only makes a small difference in the United Kingdom and Germany but has essentially no effect on citizens’ aggregate choices in the United States.Footnote ¹² The magnitude of the effect of descriptive congruence is also reasonably modest, at least when averaging across all voters. In the United Kingdom, where we see the largest effects on this dimension, respondents who share both descriptive characteristics with a hypothetical politician are nearly 5 percentage points more likely to select that politician as their representative than a politician with whom they share no characteristics. Territorial surrogation is a small liability in the United States and the United Kingdom, but the pattern is less clear in Germany, while pointing in the same direction. Republican justification is a small asset in the United States but plays no role in the United Kingdom or in Germany. Finally, responsiveness makes no difference whatsoever in any of our countries. On average, across all voters, politicians cannot increase their appeal as representatives by promising that their behaviour will be (un)responsive to electoral sanctions.

In online appendix section J, we present a series of conditional AMCE estimates for each of the dimensions. First, we investigate key interaction effects between the dimensions. We show that there is no difference in the relative importance of the other dimensions of representation conditional on whether a politician shares the policy preference of a respondent, i.e. provides substantive representation. We also show that there are only negligible differences in the AMCEs depending on whether the representative is from the party a respondent voted for.

Second, we demonstrate that the importance of substantive representation is moderated by self-reported salience that respondents ascribe to an issue. In particular, we show that when an issue is highly salient to a respondent, a politician offering substantive representation on that issue increases their selection probability by between 16 and 24 percentage points depending on the country, whereas on low salience issues the effect of substantive representation is between 13 and 16 percentage points.

Third, and importantly, we investigate heterogeneity in the effects of descriptive representation across different groups of voters and show that—although descriptive representation has only modest relative importance across all voters—shared descriptive characteristics are more important for voters of historically marginalized groups. For voters of color (in the United Kingdom and the United States), descriptive congruence increases the probability of selection by between 4 and 10 percentage points, whereas the effects for white voters are half the size. Similarly, for voters from migrant backgrounds in Germany, the effect of descriptive representation is nearly 7 percentage points, whereas it is just 1 percentage point for voters from non-migrant backgrounds. The picture is similar for gender, where descriptive congruence increases the probability of selection among female but not male voters (especially, in the United Kingdom and the United States). However, we find no evidence of conditional effects by voter sexuality. Despite this relatively higher importance of descriptive representation for historically marginalized groups, the dimension is still only the fourth most important for them (like for the public at large), except for people with racial/ethnic minority or migratory background, who rank it in third place. For them, it is about as important as personalization.

Our conjoint results also allow us to contrast the preferences of people on each dimension we elicited through the item batteries with the relative importance of these dimensions in the conjoint. First, we saw that a large majority of people in all countries prefer strong over weak substantive representation and this dimension is also important to them when deciding trade-offs between the dimensions in the conjoint. Second, with regard to personalization, people were relatively split between preferring party independence or loyalty, with only small majorities preferring rebellious politicians. However, the conjoint shows that party rebels are more likely than loyalists to be selected as good representatives. This suggests that the group of people supporting party independence care more intensely about personalization than those who favor loyalist representatives. Third, while a large majority generally does not prefer representatives to be surrogates from other districts or parties, the conjoint reveals that this aversion is only important with regard to partisan surrogation. Politicians from the respondent’s party that have been elected in other districts only have a marginally lower probability of being seen as a good representative (e.g., 1-3 percentage points) than district representatives. Fourth, in comparison to the other dimensions, descriptive representation is only the fourth most important dimension, but clearly plays a larger—still not a top—role in the selection decisions of certain groups. Fifth, despite large majorities preferring republican over pluralist justification, this dimension is of relatively little importance when citizens select between different hypothetical representatives. Finally, the majority of citizens do not want their representatives to be sensitive to electoral sanctions, but on average people do not care about this aspect of representation when they have to assess trade-offs between the dimensions.

Implications for the Study of Representation

Our results have important implications for how we should study representation empirically, and how we should think about it theoretically. First, while much recent work on representation has addressed the question of descriptive representation (notably the descriptive representation of women, e.g., Wängnerud Reference Wängnerud2009; Betz, Fortunato, and O’Brien Reference Betz, Fortunato and O’Brien2021), we find that, in the aggregate, citizens put comparatively little weight on this dimension. It is primarily historically marginalized groups, in particular ethnic/racial minorties and people with migratory background, who care more about it. These results challenge previous findings that ethnic majority groups (e.g., whites in the United States) also significantly care about descriptive representation (Gay Reference Gay2002; Harden Reference Harden2015), while simultaneously underlining that descriptive congruence is valued by historically marginalized groups (e.g., Bengtsson and Wass Reference Bengtsson and Wass2010; Schildkraut Reference Schildkraut2013). Crucially, our results also show that for marginalized groups descriptive representation can be more than simply a cue for substantive representation (as some have argued it is for voters at large, see Arnesen, Duell, and Johannesson Reference Arnesen, Duell and Johannesson2019), since our conjoint design controls for substantive representation and four other dimensions of representation (see also Hayes and Hibbing Reference Hayes and Hibbing2017).

Our results on descriptive representation therefore suggest that a research agenda that takes the citizens’ perspective more seriously should devote genuine attention to the descriptive representation of women, ethnic minorities and potentially further groups. At the same time, our results show that descriptive representation is not a great representational concern of the public at large. This contrasts with findings that descriptive representation positively affects outcomes such as perceived decision legitimacy and fairness (e.g., Arnesen and Peters Reference Arnesen and Peters2018; Clayton, O’Brien, and Piscopo Reference Clayton, O’Brien and Piscopo2019; Hayes and Hibbing Reference Hayes and Hibbing2017), democratic process evaluations (e.g., Karp and Banducci Reference Karp and Banducci2008), or political efficacy (e.g., Stauffer Reference Stauffer2021) in the general electorate.

Second, we find that citizens generally strongly care about whether a representative toes or defies the party line—what we call personalization. The majority prefers party independence, but citizens are divided here. These findings underline the relevance of recent work on personalization (e.g., Slapin et al. Reference Slapin, Kirkland, Lazzaro, Leslie and O’Grady2018; Campbell et al. Reference Campbell, Cowley, Vivyan and Wagner2019; Bøggild Reference Bøggild2020; Carson et al. Reference Carson, Koger, Lebo and Young2010; Kam Reference Kam2009) and suggest devoting more attention to this dimension of representation. For instance, there still exists little work on how exactly representatives try to “personalize” their communications and self-presentations (e.g., during campaigns), and how citizens react to them.

Third, our results on surrogation show that while most citizens prefer a representative from their territorial district, they ascribe relatively little importance to this dimension when considered alongside other dimensions of representation. Many citizens are ready to see someone as their representative who was not elected in their constituency or state. In turn, citizens clearly oppose partisan surrogation and put a great deal of weight on the party of their representative when choosing between politicians. This finding is relevant to many standard quantitative research designs that elicit state or district opinion and relate them to the behavior of the local elected representative—who will for a large fraction of citizens not be from the party they voted for. These designs are bracketing surrogate relationships that likely make up a substantial share of the representative relationships that matter to citizens. Paying more attention to these non-electoral representative relationships should be a priority in future research. The same goes for potentially conflicting preferences on the dimension of partisan surrogation: note that citizens’ aversion to the latter conflicts with our results on personalization, since it suggests that the representative with the highest support is someone from one’s preferred party who still defies the party line (refer to figure A4 in online appendix section J). Citizens seem to value the party label while also wanting their politician to resist strong party discipline, but we do not know why.

Fourth, we show that the familiar finding that citizens prefer delegate-style representation over trustee representation may actually be due to confounding based on conceptual conflation. Once we break up delegate representation into substantive representation, responsiveness to voters’ electoral sanctions, and pluralist justification style, we can see that it is substantive representation and not the responsiveness to electoral sanctions that matters to voters. Moreover, majorities in all countries actually prefer a representative who does not focus upon their re-election prospects at the next election (see also Bøggild Reference Bøggild2016). The fact that our battery items on responsiveness have lower consistency may even underline that citizens hold no strong, coherent preferences towards this conception of representation. This suggests that the extensive treatment of substantive representation in the empirical literature is justified. Yet it is doubtful that more attention should be devoted to understanding how politicians forestall electoral sanctions (e.g., Stimson, MacKuen, and Erikson Reference Stimson, MacKuen and Erikson1995; Erikson, MacKuen, and Stimson Reference Erikson, MacKuen and Stimson2002; Canes-Wrone and Shotts Reference Canes-Wrone and Shotts2004).

Finally, our results show that representation preferences vary by key socio-demographic and political variables at the individual level, but are rather uniform across countries. This suggests that a research agenda starting from the citizens’ perspective must devote considerable attention to how different groups want to be represented. This promises to usefully inform both theory and empirics.