Deliberative mini-publics: more representative and thus responsive?

Over the last decades, established democracies implemented different mechanisms aimed at increasing the participation of citizens in policymaking (Geissel & Newton 2012). DMPs are among those instruments that have gained prominence (Paulis et al. 2020). These bodies are made of citizens selected randomly to mirror the broader public and who deliberate together to formulate policy recommendations for representative institutions (Curato et al. 2021). According to deliberative theory, such deliberative instruments are expected to boost the legitimacy of public decisions for two main reasons (Fishkin 2009). First, because of their deliberative nature, they would generate better informed and more reasoned decisions. Second, sortition leads to a diverse composition, brings more equality into policymaking, and ensures that the interests of all parties are fairly represented (Delannoi & Dowlen 2016).

A core feature of DMPs lies, therefore, in the way they are composed. Participants are recruited via civic lottery, combined with some stratification with quotas defined on a series of sociodemographic traits (age, gender, place of residence, education), and, though more rarely, on some social or political attributes related to the issue at stake (Paulis et al. 2020). The goal is twofold. First, it aims at building a mini-public that mirrors the ‘maxi-public’ (Smith 2009). This is linked to the logic of ‘descriptive representation,’ according to which fair representation is ensured by the presence in institutions of individuals who reflect the characteristics of the different groups that co-exist in society (Lemi 2022; Mansbridge 1999; Pitkin 1967). Second, the use of random selection is associated with the willingness to offer a better representation of groups often absent or under-represented in elected institutions (Curato et al. 2021). Such a composition would have two effects on how DMPs are perceived by the wider public. First, descriptive representation allows non-participating citizens to identify with the deliberative process and trust the recommendations that are made by the participants who are people like them (Pow, van Dijk & Marien 2020). Second, the guaranteed representation of traditionally under-represented groups would boost the perceived fairness of the process (Farrell & Stone 2020; Gąsiorowska 2023).

Representativeness in mini-publics: a smoked mirror?

However, empirical research on the actual profiles of mini-public participants has shown that the principle of descriptive representation is not as easy to implement as it may seem. On the one hand, only a small part of the citizens randomly selected accept to participate, and self-selection is biased socially (Jacquet 2017). Confirming a wealth of research on political participation (Armingeon & Schädel 2015), those who accept to join tend to come from more affluent and privileged social groups, and less often from minorities and disengaged groups (Malkopoulou 2015). To counter these problems, practitioners rely on two strategies: they use quotas and stratification techniques in their sampling of the population (to ensure perfect descriptive representation), or they over-represent some minority groups in the final group of mini-public participants (taking deliberately some distances from the pure principle of descriptive representation for the sake of positive discrimination). Yet, those strategies do not fully compensate for inequalities in participation in mini-publics. First, DMP organizers can only stratify among those who have replied positively to the first invitation letter. Some groups have already dropped out before stratification, making perfect representation hard to reach on some traits like education (Visser, de Koster & van der Waal 2021). Second, stratification is in most cases only based on a few sociodemographic traits (like age, gender, occupation, education, and place of residence). Other traits, and especially political attitudes, are rarely considered. There are a few exceptions, but they remain rare (Paulis et al. 2020). Consequently, participants are often more interested in politics, more efficacious, or already engaged in other forms of participation (Fourniau 2019). This bias is even observed among potential mini-public participants from less privileged groups (Talukder & Pilet 2021).

There is another important obstacle to the expectation that policy recommendations from deliberative mini-publics could contribute to policy acceptance. Despite their increasing use and visibility, there remains a huge deficit of familiarity with these institutions among the population (Jacquet, Niessen & Reuchamps 2020), which can be problematic for their legitimacy. How could people identify with the process and its outcomes if they do not know anything about it?

To counter this familiarity problem, extant research has mostly relied upon experiments in which respondents were exposed to explicit vignettes about fictional mini-publics (see Goldberg 2021; Goldberg & Bächtiger 2022). Under such conditions of good information, it appears that people evaluate more positively public decisions when they know they have been reached through deliberative processes (Esaiasson et al. 2012/2019; Ingham & Levin 2018; Werner & Marien 2022). The strongest factor in how citizens evaluate mini-publics is how favorable they perceive the content of their policy recommendations (Arnesen 2017; Esaiasson et al. 2019; Pilet et al. 2021). But how exactly the deliberative process is organized—its procedures, its participants, the quality of the deliberation—also matters when citizens judge mini-publics (Goldberg 2021; Goldberg & Bachtiger 2022; Jacobs & Kaufmann 2021; Pow, van Dijk & Marien 2020). The inclusive and diverse composition of mini-publics, derived from random selection, is positively evaluated. The idea that mini-publics are representative of society at-large stands in contrast with representative institutions that overall tend to be rather different from the wider public in terms of sociodemographic attributes (Bartels 2008; Lefkofridi, Giger & Kissau 2012; Rosset 2016). This comparison between mini-publics and representative institutions turns in favor of the former (Werner & Marien 2022).

However, those latter studies suffer from two problems. First, they expose respondents to fictional mini-publics, maximizing the internal validity of the findings but reducing their external validity (Stoker 2010). Second, they provide very little information on who the mini-public participants are. Vignettes usually mention that the mini-public is constituted via sortition to ensure a fair representation of most groups in society, but they do not specify what the exact profiles of participants are (Esaiasson et al. 2019: Goldberg 2021; Ingham & Levin 2018). When they do, they would just say that mini-publics are representative of society at large on a few sociodemographic traits (Muradova & Suiter 2022; Pow 2021). And, as we have elaborated above, we know that the composition of mini-publics might remain skewed if we go beyond a few sociodemographic characteristics. In several countries that have organized mini-publics, skeptics have often pointed out the skewness of the participants’ recruitment and criticized these processes as dominated by activists and party supporters (Grandjean 2024; Rangoni, Bedock & Talukder 2023). Regarding Luxembourg’s climate assembly (KBR), this line of reasoning was developed by several press articles claiming that the process was suffering from a lack of representativeness and hence was selective, homogeneous, and not accessible to all citizens.¹

Therefore, the question that we pose here and that we aim to address via a survey experiment building on a mini-public that was taking place is what would happen in terms of outcome acceptance if citizens received accurate information about the representation of certain groups, stressing eventually some problems of imperfect representation within the process?

Descriptive representation hypothesis (H1)

The first hypothesis postulates that respondents care about the overall fairness of the composition of the mini-public. They want all groups in society to have a place within the mini-public and therefore want it to perfectly mirror their weight within society at large. We know indeed that, along with procedural fairness, participatory equality is an important principle when citizens evaluate decision-making processes and their outcomes (Christensen et al. 2023; Jäske 2019). Besides, statistical representativeness is crucial to the legitimacy of deliberative polling techniques (Fishkin 2009), as it ensures that the nature of the deliberation within the mini-public replicates what would happen with a similar deliberation with the whole population. Within this logic, the statistical representativeness of a mini-public is expected to boost outcome acceptance because this information will trigger a positive stimulus that appeals to the value of equality and the belief that all citizens have an equal chance of influencing the political process (Brown 2006; Dryzek & Niemeyer 2008; Landemore 2013).

Hypothesis 1. Information about (a) the fair representation of any group will increase the respondents’ willingness to accept the policy recommendations of a mini-public, whereas (b) the unfair representation of any group will decrease it.

Positive discrimination hypothesis (H2)

It might be that respondents do care about fair representation within mini-public, but only of certain specific groups. Especially, respondents are expected to pay attention to the fair inclusion of citizens from less affluent groups, which are traditionally less represented within political institutions. Indeed, deliberative processes are often argued to provide better representation of these groups that are marginalized in society and representative institutions, like women, citizens with lower levels of education, or members of minorities (Ryan & Smith 2014; Setäla & Smith 2018). The idea here is thus that we should not be obsessed with mini-publics being perfectly statistically representative, but rather with their capacity to address socio-political inequalities and to compensate for the under-representation of certain groups in politics (Steel et al. 2020). However, as we have explained above, it is very hard for mini-public organizers to make sure that the most vulnerable groups are fairly represented among the participants, implying that they remain often under-represented in practice. Therefore, if respondents do care about the representation of less affluent groups, they should react negatively in terms of outcome acceptance when confronted with information about the low presence of most vulnerable groups within mini-publics.

Hypothesis 2a. Information about the under-representation of politically disengaged groups will decrease the respondents’ willingness to accept the policy recommendations of the mini-public.

The other facet of this story is that respondents could react negatively also when groups that are traditionally over-represented in political institutions are also over-represented within the mini-public. Indeed, past evaluations of deliberative processes have confirmed that highly citizens very interested in politics, skilled or holding firmer views on the topic are generally more present (Buzogany et al. 2022; Elstub et al. 2021; Paulis, Kies & Verhasselt 2024). We expect that learning about this sends a signal of domination and representation biases and suggests that lower profiles or critical opinions might be under-considered in the process. Hence, it should trigger a negative stimulus that will decrease the acceptance of the mini-public outcomes.

Hypothesis 2b. Information about the over-representation of groups that are politically engaged will decrease the respondents’ willingness to accept the policy recommendations of the mini-public.

Subgroup effects’ hypotheses

The first set of hypotheses was about the main effects that should apply to all respondents, irrespective of their own profiles and opinions. Yet, we also develop hypotheses postulating that not all respondents would be equally impacted by the information received about the composition of the mini-public. A first subgroup difference depends on respondents’ pre-treatment opinion about the mini-public. It builds on earlier studies on cognitive dissonance and how individuals update their attitudes when exposed to information that reinforces or contradicts their prior beliefs. The second emerges from the idea that citizens are more egotropic than sociotropic and are therefore more affected by information related to their in-group than to out-groups.

The KBR study

In 2022, the government of Luxembourg commissioned a national Citizens’ Assembly on Climate (Klima BiergerRot – KBR). Sixty citizens living or working in Luxembourg were selected and tasked with discussing the country’s current commitments on climate change and developing potential additional measures for Luxembourg’s climate policy. Although three mini-publics were organized by academics or civil society organizations before, the KBR was the first ever organized directly by national political authorities (Paulis, Kies & Verhasselt 2024). Besides, the KBR was the first deliberative process to enjoy media visibility and political impact due to the direct role played by the Luxembourg Prime minister, Xavier Bettel, in promoting this initiative and in announcing that the recommendations would be directly integrated into the discussion in parliament and in cabinet for Luxembourg strategy to fight climate change (Paulis, Kies & Verhasselt 2024). This configuration offered a unique momentum to survey citizens relatively unfamiliar with mini-publics (and hence not influenced by past experiences) and to see how their attitudes towards such a process would evolve when they receive new information about a real-life mini-public that could have direct policy consequences. Besides, this context increases the chances that respondents took seriously when they were asked, in our survey, about their willingness to accept the policy recommendations that the KBR will be making.

We thus seized the opportunity to set up a panel study of the Luxembourg population, to see in real-life conditions how public opinion was reacting to the occurrence of the KBR. Respondents were surveyed (online) three times.² First, 3025 respondents were interviewed in January–February, at the start of KBR. A second wave (N = 2250) was held in May–June 2022, in the middle of the process. The last wave was conducted in November 2022, once the recommendations were voted on, published, and discussed in parliament. Quota sampling (based on age, sex, professional occupation, region of residence, and nationality) was used to guarantee the representativeness of the sample of respondents (see Appendix 1).

The experimental manipulation and the selection of the vignettes’ attributes

At the beginning of the second wave, we embedded one experimental task (see Table 1 for a detailed summary). The choice to use wave 2 was mainly to avoid any contamination effect.³ A brief introduction informed first all the respondents about the KBR as a Climate Citizens’ Assembly composed via sortition and its advisory role for Luxembourg climate policy. Because our experiment relies on a real case of mini-public that attracted some visibility, it is worth noting that our respondents were not entirely ignorant about the KBR, even more so considering that the survey company oversaw the KBR recruitment and that some media had already reported on the event at the time (Paulis, Kies & Verhasselt 2024). About 70% of the sample reported having heard about the KBR in wave 2 (+34% increase compared to wave 1).⁴ Yet, their factual knowledge was relatively limited, with a mean score of one correct answer out of the three questions measuring more objectively the accuracy of their knowledge.⁵ Therefore, this introduction text was important for all respondents to begin with equal access to information.⁶ Then, we split the sample and randomly assigned respondents to seven experimental groups. The control group received no further information about the KBR composition, whereas each of the six other treatment groups received a vignette describing the actual statistical distribution of one specific characteristic among (a) the KBR participants and (b) the Luxembourg population (see Appendix 2 for a thorough justification of the choices made regarding the different groups/characteristics presented in each vignette). In both cases, we used real figures. Statistics for socio-demographic traits were provided by the KBR organizers and official national statistics. For political traits (political interest, satisfaction with government, and attitudes towards climate change), we used the results of the first wave of the KBR members’ and population survey conducted in parallel. To avoid ethical concerns related to deception, we did not manipulate these figures. After the exposure to this piece of information, respondents were asked a follow-up question on their willingness to accept recommendations that would be produced by a Citizens’ Assembly like the KBR, with a response scale ranging from 0 (very unlikely to accept) to 10 (very likely).

Operationalization of the variables and modelling strategy

Our main dependent variable is how much respondents are willing to accept (0–10 scale) the policy recommendations that will be formulated by a Citizens’ Assembly like the KBR. Here, we replicated the same wording and scaling as previous studies (e.g., Germann, Marien & Muradova 2022; Pow, van Dijk, & Marien 2020; Van Dijk & Lefevere 2022), where outcome acceptance was generally used as a larger proxy for the public legitimacy perceptions of deliberative processes. When we did the survey experiment, the KBR policy proposals were not formulated yet. Respondents are therefore not influenced by existing, specific outcomes. They expressed generically how much they consent with policy proposals from a mini-public only based on what they know about the process. The descriptive statistics presented in Figure 1 show that respondents tended to score on average in the middle of the scale and thus expressed a relatively neutral opinion on the question. Yet, the distribution suggests that both negative (outcome reluctance) and positive (outcome acceptance) stances are well present in our sample, with the latter group being slightly more populated than the former.

Figure 1: Distribution of the dependent variable.

The treatment variable is the vignette to which respondents are exposed to the composition of the Luxembourg Climate Citizens’ Assembly. The respondents of the control group have received no information regarding the profiles of the participants. These respondents only know that the KBR is composed of citizens selected randomly who are tasked to advise the Luxembourg government on climate policy. In our analyses, they will be our reference group and, as a baseline, are coded with a null value (0). Other respondents were assigned a different categorical value ranging from 1 to 6 depending on the treatment that they received. As reported earlier in Table 1, the first two treatments report either an over-representation of participants with a university degree or an under-representation of citizens with a secondary school degree as the highest diploma. The third vignette is the sole to depict a situation of fair representation and statistical representativeness, with (non-)national citizens being equally present in the KBR and the Luxembourg population. The last three vignettes are about the political profiles of KBR participants and show the over-representation of citizens with high political interest, who are positive about the government, or who hold pro-climate attitudes.

As summarized in Table 2 below, H1 (descriptive representation) can be tested by looking at the effect of all the vignettes. H2 (positive discrimination) can be approached with the first two experimental groups when it comes to sociodemographic traits (presence of lower – disengaged – and higher educated citizens – engaged), and with the last three treatment groups when it comes to political traits (engaged: politically interested, satisfied with government and pro-climate). Switching to subgroup differences, to test H3 (cognitive dissonance), we contrast the main effects of the vignettes depending on respondents’ support for mini-public outcomes before the experimental manipulation. We used their positioning on the Likert scale of the item ‘I am willing to accept the outcomes of Citizens’ Assemblies like the KBR’ in the first wave of our survey. We dichotomized this information to distinguish all the people who (strongly) agreed with the statement and had positive attitudes (= 1, 44% of the sample) from those who opted for the other options, were not willing to accept the outcomes and hence adopt more negative or neutral attitudes (= 0, 66% of the sample). To approach H4 (descriptive similarity), we need to look at the main effects of the treatments considering respondents’ in-group membership. We replicate the analysis considering whether they were exposed to the presence of their in-group (see Appendix 4 for the proportion of respondents treated with the representation of their in-group). Finally, to test the last hypothesis (H5), we conduct again the same analysis while combining the two subsampling. In terms of modelling strategy, we report OLS regression tests to interpret the effect of the treatment variable, using the control group as reference category.

Main effects (full sample)

The first set of hypotheses (H1–H2) is interested in the effect of the treatments for all respondents. The analysis reveals, however, no major impact of any of the vignettes on respondents’ generic willingness to accept the outcomes produced by deliberative mini-publics. The right-hand panel in Figure 2 reports the effect of each vignette on our dependent variable compared to the control group.⁷ For none of the vignettes do we observe a statistically significant difference between treatment and control groups.⁸ There is no direct evidence supporting our main effect hypotheses related to descriptive representation (H1) or positive discrimination (H2).

Figure 2: The main effect of the treatments (full sample analysis).

Nevertheless, even if the differences with the control group are not statistically significant, the results still go in the expected direction. First, receiving the vignette on fair representation (non-nationals) boosts slightly policy acceptance. As shown in the left-hand panel of Figure 2, the mean score for this treatment is the highest of all groups (6.3) and slightly higher than the control group (6.2). Second, learning about the unfair representation of a group has a small (but non-significant) negative effect on outcome acceptance. The negative effect is the strongest for the vignette about the underrepresentation of participants with a lower level of formal education among KBR participants.

Our take on these first results is that the maxi public does not seem to care much about the mini-public composition to accept the outcomes. This is a non-finding that questions the relevance of descriptive representation as the main linkage mechanism between the maxi and the mini-public, or even whether it is really (only) via the inclusive nature of the process that mini-publics will be able to improve the acceptance of the public policies. At the same time, on a more positive note, employing quotas to make sure that traditionally under-represented groups are included within mini-publics might not harm their perceived legitimacy within the wider public, and might even be important normatively (Smith 2009).

Subgroup effects (subsamples)

Our second set of hypotheses (H3–5) aimed to explore the effect of our treatments on certain subgroups of respondents.⁹

H3: Cognitive dissonance

Because our experiment provided new information about the mini-public to panel survey respondents, we have argued via H3 that it might lead respondents to adapt their opinion when they are in a situation of cognitive dissonance, that is, when the message conveyed by the treatment conflicts with their position towards mini-public before that they participated in the experiment. To approach this, we took advantage of the panel structure of our survey and looked at respondents’ attitudes in the first wave. We have therefore split the sample and replicated the same OLS regression distinguishing those who were initially positive towards mini-public and ready to accept the outcomes of these processes (the ‘participatory enthusiasts’) from those who were not and adopted a more skeptical stance (reluctant or undecided/neutral, the ‘participatory skeptics’). The results are exposed in Figure 3.

Figure 3: Effect of the vignettes on participatory skeptics and participatory enthusiasts (subsample analysis based on pre-experiment attitudes).

Regarding participatory skeptics (grey squares in Figure 3), receiving the vignette about the fair representation (of non-nationals) significantly increases outcome acceptance, by 0.5 on average, turning from a 5.9 mean in the control group to a 6.4 in the treatment group (p = .031). It provides full support for H3a. It contrasts with the other vignettes that report on the unfair representation of some groups, which does not change the (already negative) attitudes of participatory skeptics. Those vignettes simply confirm their initial attitudes.

We can then look at the effect of the vignettes on the ‘participatory enthusiasts’ (black dots in Figure 3), that is, respondents who expressed positive evaluations of the mini-public in the first wave of our panel study. Findings reported in Figure 3 show that the exposure to information showing the unfair representation of some groups within KBR negatively impacts their generic willingness to support the outcomes from the mini-public. It is the case when they learned that lower educated citizens are under-represented (p = .023), that government supporters are over-represented (p = .040), or that citizens with pro-climate attitudes are over-represented (p = .026). It confirms H3b and the idea that unfair representation is especially problematic for those who found mini-publics a priori legitimate.

These findings suggest that mini-public composition matters. Not achieving a fair representation of different groups (defined both demographically and politically) could reduce the support for mini-publics among participatory enthusiasts. And achieving it could reduce the skepticism of citizens who are not initially positive about this form of democratic innovation. The results confirm earlier studies showing that non-participants’ legitimacy perceptions of mini-publics increase when they are maximally representative and inclusive (Goldberg 2021). It is therefore crucial to strengthen the communication of mini-publics about their recruitment methods and the profiles that are enrolled (Devaney et al. 2020). Insisting transparently on how mini-publics seek to reach representativeness in recruitment may be an essential strategy to convince citizens less favorable to these participatory processes to embrace their outcomes. And very importantly, our study shows that this fair representation is not only about the sociodemographic traits of participants but also about their political profile, something that is rarely integrated in the recruitment strategies of mini-publics (Paulis et al. 2020).

H4: Descriptive similarity

H4 proposes to look for divergent treatment effects depending on whether vignettes contain information about the respondents’ in-group. We thus differentiated respondents exposed to the presence of their in-group from those exposed to an out-group, to which they do not belong. We thus split the sample depending on the congruence between respondents’ own profile and the group about which they were informed in the vignette (see Appendix 3 for wording and operationalization). As for the previous subsample analysis, we replicated twice the same OLS regression to assess the effect of the treatment variable within each of the two groups. The findings are presented in Figure 4.

Figure 4: Effect of the vignettes depending on whether they contain information about respondents’ in- or out-group (subsample analysis).

This exercise did not confirm our expectations. The vignette on fair representation does not produce significantly more acceptance when it mentions the respondents’ in-group (black circles in Figure 4). H4a is therefore not supported. Turning to H4b, our results do not support either that the logic of descriptive similarity would matter more for respondents who belong to a group that is less engaged in politics. If this was the case, non-nationals should then have reacted also positively when introduced to the vignette about fair representation (of nationals and hence of non-nationals), or low educated would have reacted negatively to their under-representation. And we do not observe these patterns.

As the results presented in the right-hand panel of Figure 3 indicate, what we found instead is a significant impact of one vignette for respondents exposed to information about an outgroup. Higher educated respondents who learned that the lower educated were under-represented show significantly lower levels of outcome acceptance. There is a statistically significant difference of almost one point (.94) in comparison to the control group (p = .005). The unfair representation of one group traditionally discriminated in politics (lower levels of formal education) is mainly a source of concern among people who are not discriminated themselves (higher levels of formal education). This finding is surprising because politically engaged individuals usually exhibit more preference for their in-groups and bias against out-groups than do members of politically disengaged groups (Dasgupta 2004). To interpret this, one may return to positive discrimination and argue that citizens from more affluent groups have no real incentive to care about their own representation in mini-publics, as they know that they will be in any case represented within such participatory processes. Hence, they may pay more attention to the inclusion of less privileged groups because it is normatively desirable for the democratic legitimacy of the process and the recommendations that will be produced. The presence of this group of low educated ensures more cognitive and social diversity, which may be important for them to accept the outcomes of this kind of process. This finding points in the direction of an altruistic/sociotropic effect towards disengaged citizens coming from the politically engaged.

H5: Cognitive dissonance X descriptive similarity

Finally, the last expectation (H5) combined both the respondents’ initial attitudes towards mini-public and the in-group presence. We thus replicated the analysis while subsampling on these two aspects.¹⁰ The results are presented in the right-hand panel of Figure 5. Looking at the ‘participatory skeptics,’ we found support for H5a. Learning about the in-group representation makes a statistical difference only when the vignette refers to a situation of fair representation. It means that what particularly matters for these skeptical citizens to accept mini-public outcomes is that people like them are not only present in the process but in proportion that makes it fair and representative of the whole population.

Figure 5: Effect of the vignettes on participatory enthusiasts and participatory skeptics, depending on whether the vignette is about their in-group or an out-group (subsample analysis).

We found some support also for H5b when looking at the impact of the vignette on the over-representation of citizens with pro-climate attitudes within KBR. We did not observe any significant effect among the participatory skeptics, but well among the participatory enthusiasts. Especially, respondents who are participatory enthusiasts and climate skeptics turn less willing to accept mini-public outcomes in general when they learned that most KBR participants hold pro-climate views. The vignette does not directly mention the share of their in-group (climate skeptics) within the KBR but they can infer from the large share of pro-climate participants that climate skeptics are under-represented. This situation triggers cognitive dissonance. We have climate skeptics who probably trusted the mini-public to make all views present but then who learned that it was not the case (pro-climate citizens are clearly over-represented) and consequently became more reluctant to accept the outcomes. This finding is crucial not so much for support to mini-publics in general, but for the debates about this form of democratic innovation as a contribution to develop more widely accepted solutions to fight climate change. If climate assemblies (that are multiplying across the globe) fail to attract participants with both pro and anti-climate change views, they are less likely to generate policy support among those who do want to act strongly to fight climate change.

Finally, we have a last significant finding supporting H5b and confirming what we have already found above in Figure 4. Participatory enthusiasts who are highly educated become less willing to accept mini-public outcomes when they learn that there are too many engaged citizens like them among the KBR participants. It translates the opposite of an egotropic effect. They are affected by the representation of citizens unlike them. It was already visible in Figure 4 but becomes even clearer in Figure 5 as the effect is especially strong among highly educated participatory enthusiasts.

The interesting pattern, which we did not observe in Figure 4, is that the same logic seems also to apply to citizens who are participatory enthusiasts and politically interested or supportive of the Luxembourg government. From Figure 5, we can note a significant and negative effect on outcome acceptance for those respondents specifically. Even if the information that they received is positive regarding the representation of their ingroup, they seem to be unhappy with it and would probably like more political diversity among KBR participants.