Do Citizens Perceive Elected Politicians, Experts and Citizens as Alternative or Complementary Policy-Makers? A Study of Belgian Citizens

In the scholarly literature, studies have underlined a link between citizens' low levels of support for elected politicians and demands for a greater role of other actors such as citizens themselves or independent experts in policy-making. Yet, what remains unclear is whether such demands to increase the role of these actors are rooted in a desire to replace entirely politicians, or whether citizens and experts are perceived as complementary to elected politicians. It is precisely what we explore in this article. Using data from 2019 Belgian Election survey, we conduct a latent profile analysis to see what models of governance emerge among citizens. First, we demonstrate that while some citizens indeed perceive politicians, citizens and experts as separate governing groups, others combine support for multiple actors. Building on the typologies that emerge, we conduct two complementary analyses. In the second section, we try to analyse how these different views regarding who should govern translate into support for specific institutional reforms consultative referenda, binding referenda, assemblies of citizens and a government of experts. Our results show that, in general, citizens tend to favor the mechanisms that empower the actors they support, for instance deliberative democracy mechanisms are preferred by those who are positive about citizens as policy-makers. Finally, in the last section we examine the impact of citizens' personal characteristics (age, gender, education, employment) and political attitudes (political interest, political knowledge, political trust, left-right) on belonging to each of the latent classes identified. We determine the main socio-demographic traits and/or political attitudes that predict the likelihood of belonging to one of the seven classes. For example, we observe that people who delegate decision-making to politicians and experts share better socio-economic conditions and have higher levels of political interest and political knowledge. We close our analysis by explaining the importance of taking into account such preferences for a mix of policy-makers (citizens, experts and politicians) in broader debates on models of democracy.

In the scholarly literature, studies have underlined a link between citizens' low levels of support for elected politicians and demands for a greater role of other actors such as citizens themselves or independent experts in policy-making. Yet, what remains unclear is whether such demands to increase the role of these actors are rooted in a desire to replace entirely politicians, or whether citizens and experts are perceived as complementary to elected politicians. It is precisely what we explore in this article. Using data from 2019 Belgian Election survey, we conduct a latent profile analysis to see what models of governance emerge among citizens. First, we demonstrate that while some citizens indeed perceive politicians, citizens and experts as separate governing groups, others combine support for multiple actors. Building on the typologies that emerge, we conduct two complementary analyses. In the second section, we try to analyse how these different views regarding who should govern translate into support for specific institutional reforms consultative referenda, binding referenda, assemblies of citizens and a government of experts. Our results show that, in general, citizens tend to favor the mechanisms that empower the actors they support, for instance deliberative democracy mechanisms are preferred by those who are positive about citizens as policy-makers. Finally, in the last section we examine the impact of citizens' personal characteristics (age, gender, education, employment) and political attitudes (political interest, political knowledge, political trust, left-right) on belonging to each of the latent classes identified. We determine the main socio-demographic traits and/or political attitudes that predict the likelihood of belonging to one of the seven classes. For example, we observe that people who delegate decision-making to politicians and experts share better socio-economic conditions and have higher levels of political interest and political knowledge. We close our analysis by explaining the importance of taking into account such preferences for a mix of policy-makers (citizens, experts and politicians) in broader debates on models of democracy.

INTRODUCTION
For several decades, studies have shown that public support for elected politicians tends to be rather low (Dalton and Weldon, 2005;Marien, 2011). Some would even claim that citizens "hate" politicians (Hay, 2007;Grossman and Sauger, 2017). Previous studies have shown that citizens' low levels of trust in politicians may translate into support for a greater role of other actors in policy-making. For some, it triggers an increased public demand for a more active role of citizens in policy-making, via mechanisms of direct and deliberative democracy (Cain et al., 2003). Other scholars have singled out the desire of some citizens to empower actors such as independent experts or technocrats in shaping public policies (Hibbing and Theiss-Morse, 2005;Bertsou and Pastorella, 2017). Building on these views, recent studies have described public opinion in consolidated democracies as divided between those preferring elected politicians, to those favoring citizens or experts as core policy-makers (Bengtsson and Christensen, 2016;Gherghina and Geissel, 2017).
In this article, we build on this literature but propose a different view on citizens' evaluations of elected politicians, citizens and experts. Most previous work evaluated the three actors separately and asked citizens to declare which actor they would like to play the central role in policy-making. And citizens had to pick up one. Here, we opt for a different approach by examining more carefully whether some citizens may hold positive views toward more than one actor. An individual could, for example, hold negative evaluations toward elected politicians, while being positive toward both experts and citizens. Other citizens would remain positive about elected politicians but would at the same time be also positive about experts and/or citizens. In this article, we are especially interested in these more hybrid views on who should govern.
We then consolidate our analysis by confronting citizens' evaluation of the three sets of actors to support for institutional reforms that would empower citizens and experts (direct democracy, deliberative democracy and technocracy). Indeed, if representative democracy has to be transformed, it would require citizens' evaluation of potential policy-makers to be translated into demands for specific institutional reforms. Here again, we propose an approach that examine whether these instruments of governance are perceived as alternative or as complementary modes of decision-making. Some citizens may indeed see direct or deliberative democracy as antagonistic to technocracy or representative democracy. Yet other could be pushing for combining (or complementing) representative institutions with instruments of direct democracy, deliberative democracy or technocracy.
Using data from the 2019 Represent Belgian Election Study, 1 we study citizens' evaluations of elected politicians, citizens and experts. We combine different methodological approaches. First, we try to identify different subgroups of citizens regarding how they evaluate the three actors. In order to do it, we use Latent Profile Analysis (LPA), which allows identifying groups made 1 http://represent-project.be/ of respondent sharing the same patterns in their answers for the survey items that are relevant for our study (here, how they evaluate citizens, experts and elected politicians as policymakers). The next two steps build upon these latent profiles. We start by examining how support for various institutional reforms is associated with the latent profiles identified. We look at support for instruments of direct democracy (referenda), deliberative democracy (assemblies of citizens composed via sortition) and technocracy (government of experts). Finally, we use of multinomial regressions in order to examine whether some sociodemographic traits and political attitudes (political trust, political interest, left-right position) appear to be more associated with some of the latent profiles identified.

THEORETICAL FRAMEWORK
Many citizens are nowadays critical of elected politicians (Dalton, 2004;Norris, 2011;Armingeon and Guthman, 2013;Klingemann, 2013). Within this context, support for an increased role for other actors in policy-making is growing. On the one hand, a growing share of the population asked to activate the participatory instruments of democracy, in order to give more opportunities to citizens to be directly involved in the politicaldecision processes (Cain et al., 2003;Neblo et al., 2010). Other groups in society would rather call for empowering independent experts, technocrats or businessmen (Hibbing and Theiss-Morse, 2005;Bertsou and Pastorella, 2017). Building on these analyses, Bengtsson and Christensen (2016) have conceptualized the existence of three models of democracy. In the first model-the elitist model, democracy is primarily for citizens to select by means of elections the leaders who will govern. In the second model-the expertise/technocratic model-efficient leaders selected on basis of their expertise should be in charge of governing, whereas citizens' involvement should be strictly minimal. Finally, the third model-participation or pluralistic model-sees citizens' participation as central in democracy; therefore, citizens should be given a direct say in major political decisions.
Beyond the exact institutional arrangements to which these three models are attached, the central question is who should govern, or more precisely which actors do citizens perceive as having all required qualities to govern? Would it be elected politicians, citizens or independent experts? The question has already been asked in a few recent studies that used survey data. In Spain, Font and his colleagues asked respondents to declare which was the best form of decision-making. Respondents could choose between three actors, the people, experts and politicians (Font et al., 2015). In Germany, Gherghina and Geissel (2017) ran a similar study asking who should make important policy decisions. And they found that respondents were almost equally divided between those supporting elected representatives, experts and citizens.
Building on these earlier studies, we propose to analyse citizens' attitudes toward elected politicians, citizens and experts. Our general ambition is to look at citizens who would have a clear preference for one set of actors, but also to take into consideration those preferring models of governance that associate several types of actors. For example, we could imagine that some citizens would perceive both citizens and/or experts as a complement to the traditional logic of representative democracy, to come in support of elected politicians. Other citizens could be in favor of a much more reduced role of elected politicians. They want to bypass or overcome the representative logic, but they would be positive toward both citizens and experts, and not toward one of them only. In other words, there is definitely a need for richer and more complex accounts about how citizens evaluate elected politicians, citizens and experts as potential governors. Actually, this mix of actors is what we may observe already in most contemporary European democracies. Elected politicians and representative institutions remain the heart of the political system, but they often associate experts and citizens to policy-making. In most democracies, experts are invited to parliamentary hearings. Bodies of experts that are attributed a more formal role are also frequent. Regarding citizens, in many countries, referendums are institutionalized; and participatory forums of various kinds are also more and more frequently organized. It is therefore crucial to enrich the literature with analyses that take into account the possibility for elected politicians, citizens and experts to work together.
We will proceed in three steps. The first is to examine how citizens in Belgium evaluate the qualities of elected politicians, citizens and experts as potential governors. We focus on three qualities isolated as key elements in how citizens evaluate politicians: honesty, competence and capacity to apprehend societal needs (Kinder, 1986;Dalton, 2004;Seyd, 2015;Halmburger et al., 2019). Honesty is a moral trait referring to an actor's integrity and transparency, it has been previously framed as key source of trust (Bruckmüller and Methner, 2018). Competence is defined as an actor's "past political experience, ability as a statesman, comprehension of political issues, and intelligence" (Miller et al., 1986: 528). Finally, elected representatives' capacity to understand the needs of those they represent is central in theories of representation (Mansbridge, 2003). In previous studies, these traits prevail as the main criterions people use to judge or evaluate a politician (Kinder, 1986;Miller et al., 1986). On that basis, we identify how evaluations of elected politicians, citizens and experts combine in the minds of Belgian citizens, and extract the most common patterns.
Second, we examine how these evaluations of the respective qualities of elected politicians, citizens and experts translate into attitudes toward specific institutional reforms that could be introduced as a complement to traditional representative institutions. Previous studies associated directly some institutional arrangements with support for alternative (nonelected) actors. Font et al. (2015), for example, found significant correlations among Spanish citizens demanding for a more active role of citizens and support for mechanisms of direct democracy such as referenda or citizens' assemblies. Webb found the same kind of associations among British citizens (Webb, 2013). Bedock and Pilet (2020c) found that distrust in politicians was a key driver in France of support for sortition in politics. Schuck and De Vreese (2015) identified that cynicism toward politicians had an impact on support for referenda. In the same vein, a few studies have identified that citizens who were more negative about politicians were more open to reforms that would empower independent experts in policy-making (Font et al., 2015;Bedock and Pilet, 2020a). Yet, other studies have questioned these findings. First, several scholars have underlined that public support for referendums did not ascribe solely to participatory democrats. Other citizens, whose favorite policy-makers are experts or elected politicians may also believe that adding mechanisms of direct democracy to representative institutions could be positive. Indeed, they perceive referendums as a lever to keep representatives under stricter control of the people (Bengtsson and Mattila, 2009;Coffé and Michels, 2014). Recent research also demonstrated that support for consultative forms of deliberative democracy could be favored by citizens that are not especially distrustful of elected politicians or independent experts (Bedock and Pilet, 2020b). Finally, in their study of public support for technocratic governments, Bertsou and Pastorella (2017) showed that the share of citizens favorable to this model reached above 60pc in many European countries. Such high levels of support for technocratic governments are partially explained by negative evaluations of elected politicians. But support for experts goes beyond that. There are also citizens who do not fully reject elected politicians and who could still be calling for a greater role of independent experts or technocrats. These contrasting findings highlight the interest of examining in-depth how citizens' evaluation of elected politicians, experts and citizens as policy-makers associates with support for various forms of institutional reforms that would come in addition to representative institutions. In this study, we focus on three: referenda, citizen assemblies and government of experts. The rationale is that these reforms tap into three models of governance that may come as complement or alternative to representative democracy: direct democracy, deliberative democracy and technocracy. Once again, we would approach citizens' support for these three institutional reforms by examining not only public support for each of them but also how individuals may be in favor of a combination of such institutional reforms.
Finally, the last goal of the chapter is to understand what factors may differentiate among citizens holding different views regarding citizens, experts and elected politicians as potential policy-makers. Two main lines of explanation have been prevalent so far in previous studies on citizens support. The first one is that how citizens evaluate politicians, experts or citizens is related to how much resource they hold to participate politically. The more resources, the more support for empowering citizens themselves. In their study on support for referendum, Schuck and De Vreese (2015) refer to the "cognitive mobilization" hypotheses, but the same kind of logic is also found in several other studies (see Dalton, 2004;Bowler et al., 2007). Two types of resources have been examined. The first are objective resources related to individual sociodemographic profile. The main factors considered are age, gender, level of education and professional occupation. These factors have been central in studies of political participation for many years (see Brady et al., 1995). Recently, they were confirmed in studies on support for more direct forms of participation for citizens. Bedock and Pilet (2020c: 15) found in France that support for a model of democracy in which citizens would rule via referenda and assemblies randomly selected was lower among older, higher educated citizens, as well as among those with higher income. By contrast, it has appeared that less resources meant reduced support for delegating politics to elected politicians, but also to experts. In the Netherlands, Coffé and Michels (2014: 6) found that lower educated citizens were more sceptical toward elected politicians. Under a different lens, a study of British citizens revealed that age and education are negatively associated with positive evaluation of experts but also of elected politicians (Webb, 2013).
Next to objective resources, other scholars have examined subjective ones. The idea is that what matters is how competent one feels politically, and how much she is interested in politics. In Germany, Gherghina and Geissel (2017: 37) found that citizens who are more politically interested hold more positive evaluations of citizens as potential policy-makers (Gherghina and Geissel, 2017: 37). It has been confirmed in Spain by del Río et al. (2016: 93). The same kind of effect is observed when political knowledge increases Mattila, 2009: 1041-44).
The main competing line of explanation to the "cognitive mobilization hypothesis" is that support for alternatives to elected politicians is driven by political discontent, what Bowler et al. (2007) labeled the "enraged citizens" explanation (by contrasted to "engaged citizens"). Indeed, many studies have underlined that political trust was a strong correlate of citizens' evaluation of elected politicians, but also of other potential policy-makers such as experts and citizens. For instance, trust in institutions, trust in government or satisfaction with democracy are factors that have been shown to be significantly associated with citizens' views regarding who should or is entitled by citizens to be in charge of policies. Dalton (2004: 14) suggested that citizens calling for more direct participation where often showing rather low levels of political trust. Norris observed that citizens' low levels of trust vis-à-vis politicians is corelated to public support of representative institutions (1999: 20-21), which suggests a low support of elected politicians. Hooghe and Marien (2013: 145) pin-pointed that European citizens with higher political trust are more likely to take part in institutionalized forms of participation whereas citizens with lower political trust are more likely to engage in non-institutionalized forms of political participation. Similar results were drawn by Bedock and Pilet (2020a: 15) in France. In other words, evidence so far seems to indicate that lower political is often significantly associated with being more supportive of a growing role for citizens in policy-making.
In our study, we will test these two main theories and apply them to how citizens evaluate politicians, experts and citizens as policy-makers. We will also test a third factor that is not related to a specific theory but that has been found to be a correlate to preferences for models of government. This third factor is ideology, or rather citizens' political positioning on the left-right spectrum. An ideological leaning toward the left has been associated with a favorable view of citizens as policy-makers. In contrast, right-leaning citizens would be more sceptical toward citizens, and favor experts and elected politicians Mattila, 2009: 1041-44;Webb, 2013: 759;del Río et al., 2016: 93). We will test it here.

DATA AND METHODOLOGY
For our analyses, we are making use of the data collected by the 2019 Belgian Election Study coordinated by the interuniversity consortium Represent. 2 A representative sample of Belgian citizens (based on age, gender and level of education criterion) was surveyed twice, first within the 4 weeks preceding the 2019 Federal Elections, and a second time in the 2 weeks that followed Election Day. The survey was conducted online, and respondents were recruited by a private polling company (TNS Kantar). Here, we are making use of the first wave of the survey with 7,609 respondents. 3 Generally-speaking, the Belgian context appears to be very appropriate for the analyses we propose in this paper. The topic of who should govern is very salient in the Belgian context. Trust in elected politicians is among the lowest across Western Europe (Eurobarometer, 2019). Within this context, debates regarding an increased involvement of citizens and independent experts in policy-making have become more salient. Citizens' direct participation has been mostly linked to two democratic innovations: referendums and deliberative mini-publics. The 2010 Constitutional reform authorized the organization of consultative referendums at regional level, reopening debates on direct democracy. Also, four regional assemblies have tested deliberative mini-publics composed of citizens selected by lot over the last 5 years: the Walloon parliament, the Brussels parliament, the parliament of the French-speaking community, and the parliament of the German-speaking community. Finally, in 2019, several prominent politicians claimed the formation of a government of technocrats as an alternative, if government to form a government among parties would fail. 4 This context explains why the questionnaire of the Represent survey provides a large number of questions relevant for the analyses we propose to conduct. First, it contains three blocks of questions that are relevant to capture citizens' evaluations of elected politicians, citizens and independent experts as policymakers. Respondents were asked to rate to what extent the three actors are perceived as politically competent, honest/noncorrupt and able of understanding the needs of lay citizens. The exact wording of the questions is the following (see Table 1). Unfortunately, the Represent survey does not include questions allowing to determine whether respondents give more importance to one of the three traits. It would have been a very useful element in order to consolidate our analyses.
In addition to these evaluations of actors, the Represent survey also contains questions asking Belgian voters to declare to what extent they would support specific reforms, specifically direct, deliberative democracy and replacement of elected politicians by experts. The exact wording of these questions is the following. 4. Regarding the following reform, could you indicate if you are completely against, somewhat against, somewhat in favor, or totally in favor (1-4)-Experts should take the major political decisions instead of politicians.
Descriptive statistics for these items are reported in Appendix 2 in Supplementary Material. They indicate a quite strong support among Belgian citizens for the four reforms. The mean scores of support for consultative referenda and for consultative citizens' assemblies is at or slightly below 7. Mean support for binding referenda is at 6.75 and for a government of experts is slightly below 3 (on a 1-4 scale). Finally, the Represent Election Study 2019 includes questions regarding respondents' sociodemographic traits (gender, age, education, professional occupation), political attitudes (political interest, political knowledge, left-right self-placement), and political trust (see Appendix 3 in Supplementary Material). In terms of sociodemographic traits, it shows that the sample is balanced in terms of gender (51.7% of male, 48.3% of female) and of age. There is however a slight overrepresentation of respondents with a superior non-university degree or a university degree, as well as a relative under-representation of respondents who are inactive professionally. Nevertheless, the large sample size still guarantees that the N for these categories is sufficient to run robust analyses.
In terms of general political attitudes, we have four main variables: political interest, political knowledge, self-positioning on a left/right continuum, and political trust. For political interest, responses' mean situates on 5.46 on a 0-10 scale where 10 reports high political interest. Political knowledge (0-6 scale) is based on six questions about the functioning of Belgian democracy, where one correct response translates into an extra point. The mean score for political knowledge is 3.13. Finally, the variable left-right (were 0 is left, and 10 is right) the mean value situates at 5.28.
In terms of political trust, the study contains two questions capturing citizens' trust in two institutional actors: political parties and federal parliament (Cronbach's alpha = 0.897). We have compiled these two indicators into an aggregate score using principal component analysis, which generated one principal component (eigen value = 1.81) that account for 90.68% of the variance.
Building upon this large dataset, we run analyses in three steps. The first goal of the article is to understand more accurately how Belgian citizens evaluate elected politicians, citizens and experts regarding their qualities as potential governors. We rely upon respondents' evaluation of these three actors for three qualities: honesty, competence and capacity to understand societal needs. On that basis, we make use of Latent Profile Analysis (LPA) in order to identify subgroups of respondents sharing the same type of answers on the nine survey items evaluating citizens, experts and elected politicians.
In the next two steps, we build on these various subgroups identified via the Latent Profile Analysis (LPA). First, we examine whether belonging to each group translates into specific attitudes regarding support for institutional reforms (referenda, participatory budget and government of experts).
Second, we run multivariate regressions examining whether belonging to a specific latent profile is a significant driver of support (or opposition) to the various institutional reforms examined when controlling for respondents' sociodemographic and political traits. Finally, in the last section of the article, we make use of multinomial regressions in order to detect which respondents' characteristics appear to affect significantly the likelihood of falling into the various latent profiles. We examine the impact of sociodemographic traits, general political attitudes and political trust.

DIFFERENT GROUPS OF CITIZENS WITH DIFFERENT ATTITUDES TOWARD ELECTED POLITICIANS, CITIZENS AND EXPERTS. A LATENT PROFILE ANALYSIS
In this section, we are using Latent Profile Analysis (LPA) to identify different profiles of respondents based on how they evaluate elected politicians, citizens and experts' qualities as potential policy-makers. LPA is an approach that aim to identify subgroups within the population based on their answer to a defined set of indicators (see Peugh and Xitao, 2013;Oberski, 2016 for a mathematical and detailed approach of the model) in this piece, those indicators are the nine items evaluating politicians, experts and citizens. As a specific case of finite mixture model, "the idea is that subjects fall into one of a finite number of discrete categories ("classes"), and that the classes differ with respect to values of the indicators" (Jackman, 2008: 139). The method assumes, therefore, that people can be classified with varying degrees of probabilities into specific categories that have different configural profiles of personal and/or environmental attributes (Spurk et al., 2020). It differs from classic clustering methods such as the k-means as in cluster analysis the respondent is either member of the cluster k or not (Peugh and Xitao, 2013) while with LPA class membership are treated as an unobserved categorical variable and is defined on the basis of the computation of a certain degree of probability (Spurk et al., 2020). Furthermore, LPA differs from factor analysis (FA) because LPA assume that the latent variable is categorical while FA assume that the latent variables are continuous (Vermunt and Magidson, 2004: 175).
For an illustrative purpose here under lies a generic LPA model developed in the article of Peugh and Xitao (2013).
In which µ ik and σ 2 ik represent profile-specific (k) means and variances for variable i, and π k indicates profile density, or the proportion of N participants that belong to profile k (Peugh and Xitao, 2013: 618). According to Peugh and Xitao (2013) three assumptions underlie LPA models. First, samples drawn from a heterogeneous population produce data that are a mixture of K profile-specific distributions. Second, observed y indicators are normally distributed. Third, the profile-specific mean vectors µ k are the profile-specific (k) observed variable means [see Peugh and Xitao (2013) for a more detailed discussion regarding the mathematical foundation of LPA].
The model first provides random initial estimations on profile membership. Then the maximization step produces estimates of the maximum likelihood (ML) for the conditional table and the expectation maximization (EM) uses the estimated parameters to update the predicted values for the cell of the table until the parameters converge and stop changing (Oberski, 2016).
The number of latent profiles to be extracted is to be determined by combining goodness-of-fit statistics and the researchers own judgement when examining what could be the meaning of the various subgroups extracted in relation to the theoretical framework. Practically-speaking, we have extracted 2, 3, 4, 5, 6, and so on latent profiles and compare the results. In terms of goodness-of-fit statistics, we should examine the Bayesian Information Criterion (BIC) computed as follow : BIC = [−2logL + plog(n)] which penalize non-parsimonious model. The goal is to have the lowest BIC as the smaller it is, the better the model fit the data. Yet, as we can see from Table 2, the BIC gets lower and lower every time the more latent profiles are extracted. We should therefore see when the marginal gain in BIC becomes less significant. In our case, it is when seven latent profiles are extracted. Going for eight latent profiles rather than seven would have a lower impact on the BIC's reduction. We would therefore opt for working with seven latent profiles.
We should then try to see how we could make sense theoretically of these seven latent profiles. In Figure 1 we report the mean value on each of the nine survey items for each of the seven latent profiles. We also add a line with the mean value for the overall sample.
But before exposing the seven latent profiles, it is important to take into consideration some aspects of this analytic tool. First, LPA tends to extract more easily clusters of respondents that deviate significantly from the mean. A majority of respondents hold attitudes that are close to the mean value. Respondents that remain close to the mean value of the full sample are harder to decompose with LPA, for this reason one of the profiles corresponds to median values. Moreover, as we have seen in Table 1 (see standard deviations), the distribution of answers is not very dispersed, especially for respondents' evaluation of elected politicians. This makes it even harder to interpret the meaning of the various latent profiles that we can extract. Nevertheless, we will try to make sense of the seven latent profiles both empirically and theoretically. Empirically, we should pay attention to two elements. On the one hand, how each latent profile distinguishes itself from the other latent profiles in the absolute mean score for each set of actors. 5 For example, examine if certain groups are more or less positive toward elected politicians than the other latent profiles. On the other hand, we shall compare within for each latent profile what are the mean scores for the three different actors (citizens, elected politicians and experts). Then, theoretically, we will try to connect the latent profiles extracted to what we may find in the literature on citizens' support for democratic models based upon experts, citizens or elected politicians (Bengtsson and Christensen, 2016). Building on these elements, here are the interpretations we propose for the seven latent profiles.
-The first profile (44.6% of the sample) is composed of respondents whose answers are very close to the mean. We would refer to them as the median citizens. They hold rather low evaluations of elected politicians. They are slightly more positive about citizens and independent experts without making much distinction between the two. They form the 5 In order to do it, we provide in Appendix 3 in Supplementary Material box plots reporting the mean scores on the nine indicators for all seven latent profiles with confidence intervals. Especially for the evaluations of elected politicians, it makes it easier to grasp when a latent profile holds evaluations that are significantly more positive or negative than the other latent profiles.
largest group for the methodological reasons inherent to LPA previously exposed. -Regarding the second profile (28.4%), we find respondents who, in absolute terms are rather negative toward elected politicians (mean score between 4 and 5 out of 10), and quite positive toward citizens and experts (mean score around 6-7). Yet, in relative terms, they are a bit more positive toward elected politicians than most other profiles. And they are among the most positive profiles for citizens and experts. We may refer to them as the hybrid democrats. We may suspect them to be in favor of a model that would keep some role for elected representatives, but that would be positive to enrich policy-making with a greater role for experts and citizens. -The third profile (10.2%) is composed of respondents who tend to hold negative evaluations of all actors. For the three sets of actors, their mean scores are low in absolute terms (between 3 and 4). They are among the most negative toward elected politicians, experts and citizens. As they do not appear to hold stronger preferences for any actor, we would refer them as the apathetic citizens. -The fourth profile (5.3%) is composed of respondents who distinguish themselves in three respects. First, they are especially negative in their evaluations of citizens. Second, their evaluations of elected politicians are the more positive of all latent profiles. Third, they hold significantly more positive evaluations of independent experts than most other profiles. We could call them the stealth democrats (Hibbing and Theiss-Morse, 2005) or delegative democrats (Caluwaerts et al., 2018).
They are positive about actors to whom they could delegate policy-making, be them politicians or experts, while they are rather sceptical toward citizens. -The fifth profile (5%) is composed of respondents who especially distinguish themselves as the most positive toward both experts and citizens, while they are negative toward politicians (especially concerning their honesty and capacity to understand societal needs). We would therefore refer to respondents falling in this latent profile as the nonrepresentative citizens. We may suspect them to be in favor of a system that would center policy-making around citizens and experts, with a more limited role for elected politicians. -The sixth profile (3.3%) would be the sceptical representative citizens. They hold low evaluations of elected politicians. But they hold even lower evaluations of citizens and experts. The scores they give to these latter two sets of actors are approaching zero. They are the most negative of all latent profiles toward them. We might expect them to prefer a democracy based upon representative actors and certainly not upon citizens or experts. Yet, it does not mean that they are that positive about elected politicians. -Finally, the seventh profile (3.2%) would be the participatory democrats. They differ from the other latent profiles by being among the most positive in their evaluations of citizens. By contrast, they are the most negative toward elected politicians with very low mean scores (around 2) and the second lowest scores for experts (between 2 and 4).
Beyond their intrinsic values, these seven latent classes also confirm the added value of an approach that pays attention to the many ways public evaluations of citizens, elected politicians and experts as potential policy-makers may combine. The seven profiles extracted clearly shows that citizens' evaluations of elected politicians, independent experts and citizens should not be conceived as fully opposed alternatives. Belgian voters who hold (more) positive evaluations one of set of actors while being (more) distrustful about the two others are rare. The participatory democrats who hold positive views only about citizens are scarce (3.2%). Sceptical representative democrats who distrust both citizens and experts, while being a bit less sceptical about politicians' qualities also exist but account for 3.3% of the sample. Whereas, pure stealth or technocratic democrats who would only be positive toward experts are not identified. By contrast, there are several profiles where respondents hold positive evaluations of more than one actor. We observe the presence of citizens who are positive about both experts and politicians (delegative democrats). We also find citizens who are sceptical about politicians but positive about both citizens and experts (hybrid democrats, non-representative democrats). The largest group extracted (median citizens) are also quite positive toward both citizens and experts. These three profiles account together for 43% of the sample. Yet, these profiles are not linearly connected to one model of democracy (representative, technocratic or participatory). They would rather situate themselves in a hybrid position between the technocratic and the participatory model (Bengtsson and Christensen, 2016). And they confirm that lines between these ideal-types are blurred.
Our findings also provide some clarification to previous studies that have been highly influential in the field. For example, they can help understanding who stealth democrats would be. According to Hibbing and Theiss-Morse (2005) stealth democrats are characterized by distrust in elected politicians, support for experts, and are opposed to a model of democracy that would call for too much participation from citizens. With our study, we rather show that there are indeed citizens quite positive toward experts as policy-makers. Yet, all of them tend to be also supportive of another actor. Delegative democrats want to combine a role for experts and for politicians, and are very negative about citizens. Non-representative citizens are positive toward experts and citizens and very negative toward politicians. This later group seems to be in line with what Webb claim in his study of stealth democrats in the UK. "It is quite conceivable that political actors who hold such (stealth democratic) views would be drawn to the idea that the ordinary and presumably virtuous people should at least occasionally be able to take decision-making power out of the hands of elites by recourse to referendums" (Webb, 2013: 761).

LINKING CITIZENS' VIEWS REGARDING WHO SHOULD GOVERN AND SUPPORT FOR INSTITUTIONAL REFORMS
The findings from the latent profile analysis that we have just exposed were about respondents' evaluations of the qualities of three potential actors of policy-making: citizens, experts and elected politicians. Yet, they do not directly tap into how such preferences translate into demands for institutional reforms. It is what we propose in this section by looking at how each latent profile positions regarding four specific institutional reforms: consultative referenda, binding referenda, assemblies of citizens composed via sortition, and a government of experts.
Research on citizens' preferences for the different models of democracy has often been linked with studies on public support for instruments of citizens' participation such as referenda or assemblies of citizens sorted by lot, as well as with support for technocratic governments or for the idea of experts taking the main political decisions (Mondak, 1995;Bengtsson and Mattila, 2009;Coffé and Michels, 2014;Grönlund et al., 2014;Font et al., 2015;Schuck and De Vreese, 2015;Bertsou and Pastorella, 2017;Bedock and Pilet, 2020c). Yet again, very few studies so far have examined whether and how some citizens may perceive these instruments as to be combined. One can, for example, be in favor of a growing use of referendums and for technocratic governments, or in favor of greater citizens' participation but without getting rid of elected representatives. An illustrative example is the study of German citizens Gherghina and Geissel conducted in 2017. They show that citizens who prefer elected politicians as core decision-makers are indeed supportive of elections, but they would also like to participate in citizens' consultations. Also, they argue that German citizens who wish for a greater role of citizens in policy-making are in favor of referenda, and citizen assemblies but also remain interested in voting. Finally, those who favor experts would still like to vote in elections and support referenda.
We adopt the same kind of perspective in this section. We examine correlations between belonging to each of our seven latent profiles and attitudes toward four institutional mechanisms that are being debated in Belgium: consultative referenda, binding referenda, assemblies of citizens selected by lot and governments of experts (see Appendix 2 in Supplementary Material for descriptive statistics). We use respondents' support for each of these four mechanisms as our dependent variables. And we evaluate whether levels of support differ across the seven latent profiles.
In the four figures below, we report the mean levels of support for each one of the seven latent profiles. We also include confidence intervals. We can see that there are significant differences across latent classes for all four reforms. 6 First, in Figure 2, we report the mean score regarding support for consultative referendum. All seven latent classes are rather in favor of the mechanism (mean score > 5) but we can also observe significant differences between three clusters of latent profiles. First, three latent profiles present levels of support that do not present statistically significant differences with the median class. It is the case of the apathetical citizens, the delegative democrats, and the participatory democrats. Two latent profiles are significantly more positive about consultative referenda: the non-representative citizens and the hybrid democrats. One latent profile, the sceptical representative democrats, are significantly less favorable to consultative referenda even if the mean score of support remains slightly above the median value (5 out of 10).
In order to make sense for these differences, it is important to go back to the evaluations of citizens, experts and elected politicians by the various latent profiles. In particular, what seems to make the difference are the evaluations of both citizens and elected politicians. Holding positive evaluations of citizens is clearly a crucial element. If it is not the case, like for the sceptical representative citizens, consultative referenda are perceived with more caution. By contrast, being very positive about the capacities of citizens leads to be more supportive of consultative referenda, this is the case for the non-representative citizens. Yet, how respondents evaluate elected politicians also matters. Being positive about both citizens and elected politicians may lead to also being supportive of consultative referenda, e.g., hybrid democrats, while being positive about citizens but very negative toward elected politicians, like participatory democrats, does not lead to very strong support for consultative referenda. Respondents in this last latent profile would not be very enthusiastic about a reform that would empower citizens but leaves the final word to politicians.
The second reform considered is the introduction of binding referenda at national level (Figure 3). Compared to what we observed for consultative referenda, we see sharper differences between latent profiles. In particular, we observe two profiles that are less enthusiastic about this instrument of direct democracy: sceptical representative democrats and delegative democrats. Both latent profiles are quite sceptical about citizens' abilities as policymakers. It translates into rather negative views toward binding referendums. In that respect, they would differ from pure stealth democrats who "appear to appreciate direct democracy primarily as an instrument of control that is used as a measure of last resort" (Mohrenberg et al., 2019: 2).
By contrast, we have three groups that are more enthusiastic about binding referenda: hybrid democrats, non-representative citizens and participatory democrats. The three groups hold the most positive evaluations of citizens, which makes it logical that they are the most supportive of direct democracy. Yet, we could also stress differences between them. Participatory democrats are only positive about citizens, and not about elected politicians or experts. Non-representative citizens are positive about both citizens and experts. Hybrid democrats are also positive about citizens and experts, but are also among the least negative toward elected politicians. The diversity of these three latent profiles shows the multi-faceted nature of public support for referenda in contemporary democracies. Previous studies found indeed support for direct democracy among dissatisfied democrats and more politically engaged citizens (Bowler et al., 2007;Schuck and De Vreese, 2015). And others found a link between support for experts and for referenda (Webb, 2013;Coffé and Michels, 2014). Our findings connect with these various studies.
The third reform falls within the deliberative democracy logic and consists of setting up assemblies of citizens selected via sortition. Such assemblies are not unknown in Belgian politics. Since 2014, citizen assemblies composed by lot have been tested by four regional parliaments across the country. At the local level, several municipalities have also tested deliberative mini-publics. In total, since 2001, at least 33 citizens' assemblies including random selection of citizens took place in Belgium at the local, regional national and European level organized by state and non-state actors (Vrydagh et al., 2020) which makes it a largely used process. Belgian citizens are therefore more likely to have heard about the use of sortition in politics than citizens in many other countries.
In Figure 4, we report the mean score of support toward the creation of assemblies composed of citizens selected by lot. We see patterns that are very comparable to what was observed regarding binding referenda. Three groups are more positive toward this instrument of deliberative democracy: hybrid democrats, non-representative citizens and participatory democrats. They are the three profiles holding the most positive views toward citizens, but they differ in the evaluations of elected politicians and experts. One group is much less enthusiastic, although still slightly positive (mean score close to 5): the sceptical representative democrats. They are the ones holding the most negative evaluations of citizens. The other latent profiles are in between but remain rather positive about the mechanism (mean scores between 6 and 7) probably because it is a consultative mechanism and elected politicians would remain in charge of the final decision. It is therefore not too frightening for delegative democrats, for example, who remain more positive toward politicians or experts than toward citizens.
Finally, in Figure 5, we report mean scores toward delegating political decision-making to a government of experts in Belgium. Here, overall support is less strong across respondents. The scale is between 1 and 4. One and two express (strong or moderate) opposition to a government of experts. Three and four express (moderate or strong) support. Only one latent profile is clearly positive: non-representative democrats, who were very positive toward experts and citizens, while holding negative evaluations of politicians. The two other profiles that have positive evaluations of experts, delegative democrats and hybrid democrats, are more moderately supportive of a government of experts instead of a government of elected politicians. It could be explained by the fact that these two latent profiles were not too negative toward politicians, even if they were positive toward experts.
Finally, we may observe that the three latent profiles with the lowest mean scores of support are the three that are holding the most negative evaluations of experts' qualities as potential policymakers. It is the case for sceptical representative democrats who were less negatives toward elected politicians than experts or citizens, participatory democrats who were only positive toward citizens, and apathetic citizens who were negative about all three actors. 7 Having analyzed successively support for four institutional reforms, we can now try to make some conclusions on what our results brings to the scholarly debate on institutional reforms that would complement, bypass or overcome representative institutions. And what we observe is that the various profiles tend to be in favor of a mix of instruments. It confirms once again the added value of a methodological approach that gives space to citizens who may hold hybrid views regarding the role of citizens, elected politicians and experts in policy-making. The most striking example are non-representative citizens and hybrid democrats. They are strongly in favor of all four instruments. It shows that they would prefer a model of democracy where both experts and citizens play a greater role. The same could be said about apathetical citizens and median citizens who are, although moderately, favorable to all four reforms. Delegative democrats also show a mixed profile. They support consultative referenda and consultative citizens assemblies, as well as governments of experts. They are only more negative toward binding referenda. Finally, only two profiles seem to be in favor of one unilinear logic. First, participatory democrats only want instruments increasing citizens' participation (referenda and citizens' assemblies composed by lot), while they oppose a government of experts. Second, sceptical representative democrats are the most negative for all four instruments as they move away from the representative logic.

WHO BELONGS TO THE DIFFERENT PROFILES OF CITIZENS?
The question that we would like to pose in this section is whether the seven latent profiles that we have identified are inherently different in their profile, both socio-demographically and politically. In order to test for it, we have run multivariate analyses accounting for the effect of a series of individuallevel variables on the likelihood of belonging to one of the seven latent profiles. These variables are age, gender, level of education, professional situation (we focus on blue-collars and unemployed), region of residence, satisfaction with income, political interest, political knowledge, left-right self-placement and political trust.
These independent variables were included in a multinomial logistic regression for which the "median citizens" latent profile was the reference category. In Table 3, we report the results for the six other latent profiles. For each, we show the effect of each independent variable on belonging to that specific latent profile compared to belonging to the "median citizens." In order to facilitate interpretation, we present the relative risk ratio (RRR) for each independent variable rather than the raw coefficients. A relative risk ratio superior to 1 means that an increase of one unit in the independent variable increases the probability of belonging to the latent profile analyzed rather than to the "median citizens" latent profile by the value of the RRR. A relative risk ratio inferior to one means that one unit increase in the independent variable reduces the likelihood of belonging to the latent profile examined.
We are not going to focus on the variables that are significant for each and every profile. Yet, it should be stressed that in each case, several sociodemographic and political factors appear to be significantly associated with belonging to the latent profile considered. These findings already confirm that the seven subgroups of attitudes toward elected politicians, experts and citizens that we have identified are not purely random. They are deeply anchored socially, economically and politically. In particular, it appears that six factors play a significant role across several latent profiles: gender, satisfaction with income, education, region of residence, political interest, political knowledge and political trust. By contrast, other factors such as age, employment, or the position on the left-right spectrum seem to play a more reduced role.
Instead of detailing the effect of each variable for each latent profile, we will rather pay more attention to some findings that are relevant across several latent profiles, and that can be connected with earlier studies on public preferences regarding democracy. A first finding to be underlined within this logic concerns the effect of respondents' satisfaction with their income. Previous studies showed that citizens who were better off socio-economically support the status quo are and more reluctant regarding reforms that would reduce the role of elected politicians (Ceka and Magalhaes, 2019;Bedock and Pilet, 2020a). Our findings show a slightly different pattern. Citizens more satisfied with their income tend indeed to be less often found among participatory democrats and apathetic citizens who are very negative toward elected politicians (and experts), and only positive about citizens. Yet, we also have some findings that indicate that some more economically privileged citizens may be found in latent profiles holding more positive evaluations of citizens and experts, and who would support reforms toward direct democracy, deliberative democracy or technocracy. It is particularly the case with non-representative citizens. It is also, though to a lesser extent, the case for hybrid democrats who would like to combine the representative logic with a greater role for both citizens and experts.
In the same vein, our approach provides a more nuanced view regarding the effect of education on citizens democratic preferences. Previous studies present various, and not always congruent, findings. Coffé and Michels (2014) found that lower levels of education led to more negative evaluations of elected politicians, but also to more positive attitudes toward experts (Coffé and Michels, 2014). Other studies found that higher educated citizens would be over-represented among citizens calling for direct and deliberative democracy (Bengtsson and Mattila, 2009;Webb, 2013;Dalton, 2017). And some demonstrated that direct democracy is more supported by lower educated citizens (Schuck and De Vreese, 2015).
Our findings confirm that the relationship between education and preferences regarding who should govern is not linear. In Figure 6 we have plotted for each latent profile (except median citizens) the predicted probabilities of being within this profile for each three levels of education (primary education, secondary education, higher education). The only latent profile for which we have an upward line is for delegative democrats, meaning that this latent profile is more often associated with higher educated respondents. They are indeed the most positive in their evaluations of elected politicians, and they are the least favorable to all four institutional reforms. In parallel, we observe two profiles with downward lines: participatory democrats and non-representative citizens. For these three profiles, the effect of education is possible to interpret. Higher educated citizens are more present among those sceptical toward citizens' participation, and lower educated citizens are more present among two models that call for a greater role for citizens in policy-making. Such findings would indicate that education does not lead to a more citizen-centered models of governance like authors such as Dalton (2017) claim. Rather, supporters of such models are more often lower educated citizens. Yet, such a bold claim does not hold when we look at other latent profiles-hybrid democrats, apathic citizens, sceptical representative citizens-for which the effect of education is not linear. Moreover, it indicates that education can predict part of citizens' attitudes toward citizen-centered models of governance but patterns are less clear when it comes to support for experts or elected politicians.
Finally, our study also brings interesting insights regarding the effect of political trust. In the literature, the dominant hypothesis is that support for other actors than politicians-be them citizens or experts-is primarily found among respondents who distrust elected representatives (Dalton, 2004;Hibbing and Theiss-Morse, 2005;Bowler et al., 2007;Schuck and De Vreese, 2015;Bedock and Pilet, 2020b,c). What we find is slightly different and not straightforward (see Figure 7). We can confirm that lower political trust is associated with the three latent profiles that are the most negative in their evaluations of elected politicians: apathetic citizens, participatory democrats and sceptical representative citizens (downward lines). Yet, the latter group shows that being distrustful toward politicians would not systematically translate into support for alternative actors, be they citizens or experts. Sceptical representative citizens hold very negative evaluations of both citizens and experts. They are also the least enthusiastic toward referenda, citizens' assemblies and government of experts. Another interesting element from our CONCLUSION Studies on citizens preferences regarding who should governed have been multiplying over the last decade. The main models that have been isolated are defined primarily when it comes to who should govern: elected representatives, independent experts or technocrats, or citizens themselves. And answers to this question have been shown to determine how citizens support or oppose to institutional mechanisms such as referendums, citizens' assemblies or technocratic governments that could complement or replace the traditional representative model. Our study contributes to this body of literature and proposes to contrast between citizens pushing for a model dominated by one single actor and those calling for models of governance that combine several actors.
Using data from the 2019 Belgian Election RepResent survey, we conducted a latent profile analysis (LPA) and extracted seven latent profiles of democratic preferences. In their majority, these profiles demonstrate that most citizens hold positive evaluations of more than one actor. Actually, very few citizens belong to a profile holding particularly positive evaluations of one single actor. It is the case of sceptical representative citizens (3.3% of the sample) who express higher levels of support only toward politicians, and of participatory democrats (3.2%) who have a positive perception of citizens in policy-making, but a negative one of politicians or experts. By contrast, the other profiles combine support for different actors and would favor the presence of more than one actor to govern. It is the case of hybrid democrats who evaluate positively both experts and citizens as policy-makers, of delegative democrats who prefer elected representatives and experts to be in charge of politics, but do not wish for citizens to be involved in political processes, and of non-representative citizens who prefer citizens and experts over politicians We have identified thus three groups that we could call hybrid because they are positive toward more than one actor. The fifth profile are the sceptical representative democrats who tend to be quite negative toward elected politicians; yet they are even more strongly negative toward citizens and experts. Finally, our sixth profile, apathetic citizens, shares the lowest evaluations for all three set of actors.
The second step in our analyses has been to examine how these latent profiles translate into support for specific institutional reforms. We tested support for instruments of direct democracy (consultative and binding referenda), deliberative democracy (consultative citizen assemblies) and technocracy (governments of experts). And our findings have here as well-confirmed that many citizens perceived theses institutional mechanisms as complementary rather than antagonistic. Few citizens would really call for abandoning the representative logic and to replace it entirely with another model of democracy, be it centered around citizens or experts. Citizens want to complement representative democracy, rather than bypassing it. And it is observable with several latent profiles of citizens pushing for a combined introduction of referendums, citizens' assemblies and some forms of technocratic cabinets into the Belgian institutional architecture.
Finally, our study has confirmed previous studies, which had shown that there were strong socio-economic and political determinants in shaping citizens' views toward how government should be organized. In particular, three variables appear to play an overarching role in differentiating among citizens' latent profiles. First, we confirmed earlier studies which found more support for models strongly empowering citizens-participatory democrats-among the less well-off socio-economically. But we can mitigate this conclusion as we also found that a better economic position can still be associated with positive evaluations of citizens as governors, at the condition that these citizens work in association with other actors such as elected politicians or experts. Second, education has also appeared to be a crucial variable in shaping democratic preferences. Yet, what appears is that the relationship between this factor and democratic is not linear. Being less educated does have an effect on being less supportive of elected politicians, but it may lead to varying attitudes toward the two other actors: citizens and experts. Among the lower educated, we find citizens who would want citizens alone to govern; others who want citizens and experts to work hand in hand; and also some pushing for experts and citizens to work in collaboration with elected politicians. And third, our study has confirmed that lower political trust is the fertile ground to support a growing role of other actors than elected politicians. But again, it does not have a linear and univocal impact on what other actors are positively evaluated.
These elements contribute to a call for nuanced views in studying citizens' democratic preferences. Yet, we shall admit that they pave the way for more research on the topic much more than the provide final answers. In that respect, we should acknowledge some shortcomings in our study. First, despite identifying the nuances in citizens' evaluation of different actors (LPA) we cannot discern clearly what is their preferred model of governance. In particular, our study infers from respondents' positive evaluations of several actors that many citizens want mixed models associating several potential governors. But what exact shape would this mixed model take remains unclear. Goldberg et al. (2020) recent study does one step in this direction by asking German citizens to evaluate governance models that associate politicians, experts and citizens. It is probably an example to be followed. Also, we have little understanding about citizens' arguments for supporting one or two actors over the other(s). Indeed, it is not always crystal clear why one would prefer one or two set of actors over others and why they would favor one model over another. It might be interesting to address these issues using a qualitative approach. In that respect, one final element for future studies could also be to fine-graine the dimensions on which citizens are evaluating elected politicians, experts and citizens as policymakers. Here, we use questions tapping into how respondents evaluate honesty, competence and capacity of understand societal needs of the three actors. It could be interesting to look at more dimensions. We also assume that the three dimensions are given an equal value by respondents for each of the three actors. But it may very well be that some respondents give more importance to competence than to honesty or capacity to understand societal needs, for example. Future studies could try to address these questions.