It’s Not Just the Taking Part that Counts: ‘Like Me’ Perceptions Connect the Wider Public to Minipublics

Many deliberative democrats herald the potential of minipublics to help improve the quality of democratic decision-making. Yet these democratic innovations present a paradox: how can the use of minipublics be perceived as legitimate by the maxi-public when most citizens cannot participate? In this article, we address this question in the context of Lafont’s argument that minipublics amount to ‘shortcuts’ in the democratic process. We challenge this argument by hypothesising that non-participants perceive minipublics to be legitimate when they perceive minipublic participants to be like them – and when they perceive politicians to be unlike them. Similarly, we expect that the relative importance of descriptive similarity will be related to the issue in question. We test our hypotheses in the deeply divided context of Northern Ireland, where a minipublic was held on the salient and contentious question of the polity’s constitutional future. Survey evidence confirms that ‘like me’ perceptions constitute a significant predictor of minipublic legitimacy perceptions. Our results have implications for the communication of minipublic features to the broader public, for the use of minipublics alongside conventional decision-making processes, and for further empirical research.


Introduction
Deliberative minipublics have increasingly been prescribed as remedies to counteract some of the alarming ailments that afflict contemporary representative democracies (Dryzek et al. 2019;Elstub & Escobar 2019). By involving ordinary citizens directly in decision-making, the design features of these democratic innovations have the potential to make decision-making more inclusive, more deliberative, better informed and, ultimately, more legitimate (Harris 2019). However, minipublics remain relatively novel tools in the ambitious -and inherently risky -task of democratic engineering. Most of our empirical knowledge on the effect of minipublics is with respect to the relatively small number of citizens who engage with them as participants. Studies typically show that the opinions of participants are transformed, to varying degrees, after learning about and deliberating on a given topic (Farrar et al. 2010;Himmelroos & Christensen 2014;Niemeyer 2011). They also typically show that minipublic participation has a positive effect on citizens' perceived ability to participate in, and have an influence on, political decision-making (Farrell et al. 2013;Fournier et al. 2011).
Despite evidence of the potential for minipublics to enhance democratic opportunities for participants at the micro-level, there are concerns that they may exacerbate democratic problems more broadly. The argument against the use of minipublics in political decision-making has been most prominently articulated by Lafont (2019), who claims that they amount to taking 'shortcuts' in the democratic process. Since the vast majority of citizens do not -indeed, cannot -themselves participate, minipublics are expected to undermine, rather than strengthen, democracy. This claim is not empirically supported by the limited number of studies that extend their scope to non-participants; that is, the broader public. Among this emerging body of research, a key finding is that minipublics can help to foster favourable attitudes towards the political system, even among those who do not themselves participate (Boulianne 2018;Knobloch et al. 2020). However, while these studies make a helpful contribution to our understanding of the democracyenhancing function that minipublics serve, they do not provide a clear answer to a crucial question that is pertinent to Lafont's critique: why would non-participants trust minipublics and the outcomes recommended by their participants?
In this article we attempt to address this question. We begin by outlining what we understand to be the main tenets of Lafont's 'shortcut' argument, before reframing the relationship between minipublics and the broader public. Our main argument is that legitimacy perceptions of minipublics are rooted in the perceived similarity of minipublic participants to non-participants. When citizens perceive minipublic participants to be like themand when they perceive politicians to be relatively unlike them -they will be more likely to perceive minipublics as legitimate. Similarly, we expect that the relative importance of descriptive similarity will be related to the issue in question. We tested our hypotheses on a representative sample of citizens in Northern Ireland, who were presented with details of a minipublic that took place on the salient and contentious question of the polity's constitutional future. The survey results provide support for our hypotheses, suggesting that 'like me' perceptions connect non-participants to minipublics and their outputs.

Deliberative Minipublics and the Maxi-public
Deliberative minipublics embody two core features (Farrell et al. 2019). First, participants are randomly selected. This feature helps to achieve a sample that is broadly representative of the wider population (Fishkin 2009). Second, the citizens who are selected to serve in this microcosmic setting, usually anywhere between 12 and 200, then engage in a process of deliberation, involving learning (from experts and stakeholders) and facilitated group discussions (Setälä & Smith 2018). On a given issue, the combination of these features enables deliberative minipublics to produce (typically advisory) outcomes that reflect considered public opinion. Policymakers may then decide to implement these recommendations on the basis that they provide an estimation of what all citizens would have decided if they too had engaged in the process. With minipublics widely heralded as smallscale applications of deliberative democracy, as one ideal, the centrality of these twin features draws attention to a significant trade-off in the promotion of different democratic principles in institutional design. By inviting only a small number of citizens to engage in structured micro-deliberation, a minipublic necessarily excludes the vast majority of members of the wider -or maxi -public. Thus, representativeness and deliberation come at the expense of mass participation (Fishkin 2009).
For Lafont (2019), the principal consequence of this compromise is a loss of democratic control over decisionmaking. By defining the ideal democratic system as one of self-government, she argues that it is not possible to be agnostic about citizens' participation in politics at the macro-level, and thus explicitly criticises the use of minipublics in the decision-making process. Since nonparticipants have not themselves engaged in a deliberative exercise of the kind involving minipublic participants, they apparently 'have no specific reason or justification' to endorse their outcomes (Lafont 2019: 99). If nonparticipants were to treat minipublic outputs as legitimate, it would amount to 'blind deference', leaving the wider public disconnected from decisions taken on its behalf (Lafont 2019: 121). Instead, she develops a participatory conception of deliberative democracy that is centred on the ideal of self-government, leaving no room for 'shortcuts' that bypass deliberation involving the mass public.
However, it does not necessarily follow from Lafont's normative point of departure that minipublics have no democratic utility within a political system's decisionmaking architecture, beyond informing wider public deliberation. Let us consider the claim that minipublics ' erode the fundamental commitment of the democratic ideal of self-government, namely, to ensure that all citizens can equally own and identify with the institutions, laws, and policies to which they are subject' (Lafont 2019: 3;emphasis added). From this benchmark it is taken that for citizens to own and identify with public policy, they must be directly involved in the deliberation that precedes its implementation. But this is a claim that invites empirical investigation. How, for example, can citizens feel a sense of ownership over policies? What does it mean, more broadly, for all members of the public be able to identify with the political system and its policies? To answer these questions, we must account for citizens' actual -or rawperceptions of decision-making.
We suggest that the crucial mechanism connecting minipublics to the wider public is a psychological attachment that develops between citizens (in the wider public) and citizen representatives (in a minipublic). This relational approach to representation is not new (see, for example, Warren 2008). Recognising the inevitability of some division of labour between representatives and the represented, MacKenzie and Warren (2012: 123) argue that such a division of labour is justified on the basis of trust. They further argue that minipublic participants are particularly capable of obtaining trust as information proxies from the broader public due to their selection process on the one hand, which results in the representation of all affected interests, and their deliberative process on the other, which allows all relevant interests to be fairly considered. Crucially, trust in minipublics may stem from perceptions of participants' similarity, in a broad sense, to the wider public, perhaps coupled with a perception that they promote a convergence of wider public interests (Warren 2008). Lafont (2019: 115) is not persuaded that these reasons are strong enough to justify citizens 'blindly' placing their trust in minipublics, contending that non-participants cannot reasonably trust the outcome of a minipublic without examining its deliberations on the issue in question. Whereas the participants in a minipublic play an active role in digesting relevant information, reflecting on the arguments and engaging in face-to-face discussions ahead of collectively recommending the best course of action for the whole public, non-participants -the vast majority of the public -are left essentially ignorant of the entire process. Blind deference, it is argued, exists 'if there is no capacity … for control' (Lafont 2019: 8). However, to assume that trust is blind downplays the first point raised by MacKenzie and Warren: that trust is created not just by the process of deliberation, but also by virtue of who participates in the process. While it is true that citizens cannot control the members of minipublics in the way that they can control representatives -through elections -it is not the case that no such capacity exists; the relationship simply operates on a very different basis.
Whereas elections create bonds of accountability and authorisation between citizens and their representatives, the mechanism of sortition may be understood to eliminate such bonds between minipublic participants and the wider public (Lafont 2019: 119). This is at least the case in a direct, substantive sense. But in another sense, sortition may be understood to create a bond between non-participants and participants. The difference is that this bond is affective, not substantive. It develops when a citizen feels that his or her interests are represented by fellow citizens in a minipublic. By extension, this bond may be broken if the citizen no longer feels represented. Framed in this way, the perception of attachment is a matter of degree -and only individual citizens can decide the extent to which they feel attached to citizen representatives in a minipublic. This is clearly very different to the control that citizens might exercise in elections, but the difference does not necessarily amount to a reduction in control. Indeed, when voters select candidates on the basis of their alignment to their own views, or sanction candidates for poor performance, it is also very much a matter of degree as to whether or not elections effectively facilitate these ends (Achen & Bartels 2016). More importantly, while citizens do not physically cast a ballot to authorise minipublic participants or hold them to account, this is not the same as an absence of control. Approaching the bonds of authorisation and accountability from a psychological perspective, there is no greater form of control than being able to determine one's feelings.

'Like Me' Perceptions
Our task now is to examine the nature of this psychological relationship between citizens and citizen representatives. Why would non-participants trust minipublic participants to make recommendations about political issues on their behalf? Our main explanation rests on the profile of minipublic participants as a microcosm of the wider population. From the perspective of non-participants, their perceptions of the legitimacy of the process are rooted in the perception that the participants are like them. This is a key element of descriptive or 'mirror' conceptions of representation that stress the importance of a representative body being an accurate reflection of the population (Phillips 1995;Pitkin 1967). In a narrow sense, this approach concerns similarity across demographic characteristics, such as gender, age and ethnicity. More broadly, descriptive representation can involve invisible characteristics, such as ' experiential perspectives' (Brown 2006: 211; see also Mansbridge 1999). By virtue of random selection, the participants selected to serve in a minipublic constitute a microcosm of the wider population through the distribution of visible and invisible traits.
Lafont acknowledges this 'mirror' argument and its multiple claims: participants in minipublics are 'like' nonparticipants in the sense that they are ordinary citizens, they are independently motivated, and they reflect a diverse range of interests, values and policy objectives that exist within the population. These features lead to the assumption that the participants will reach recommendations that coincide with the latent preferences of all citizens if all citizens had the opportunity to participate in the process. We should, thus, trust minipublics in the strong sense of endorsing their recommendations as our recommendations. However, Lafont ultimately argues that the mirror claim collapses when it is combined with the 'filter' of deliberation: the post-deliberative opinions of the majority of participants in the minipublic may well differ from the opinions of a majority of citizens. By learning and deliberating about a political issue, minipublic participants will behave more like ' experts' than ordinary citizens, and so the sample will no longer be a mirror of the wider public. This creates a disconnect between the raw preferences of citizens and the outcomes preferred by minipublic participants.
There are two important responses to this critique of the mirror argument. The first relates to the theoretical basis of descriptive representation; what it means to be 'similar'. For Sintomer (2013: 11) representation is less about ' acting in the name of' and more about ' acting as'. Identity representation, for example, can be illustrated by the rhetoric employed by the Occupy Movement of 2011. When demonstrators claimed to be 'the 99%', they asserted their ability to speak like the people, rather than for them (Sintomer 2013: 21). The significance of both demographic and invisible characteristics being adequately represented in a minipublic is that it forms a relationship between participants and non-participants through common ex ante experience (Warren 2008). In this way, the descriptive view of representation is not confined to a 'passive' similarity between representatives and represented, but entails an ' active' component which assumes that representatives are capable of spontaneously responding to new information and new circumstances in a way that is similar to how those represented would have responded (Brown 2006;Mansbridge 1999Mansbridge , 2019. In a minipublic, individual participants do not shoulder the responsibility of representing a certain group or trait as a delegate; rather, they are there as themselves as part a wider group selected on the basis of political equality. From the perspective of non-participants, the agency of the participants need not distort the mirror quality. On the contrary, knowing that it could have been them in the room, in possession of such agency, arguably enhances it. From this understanding of descriptive representation we formulate our first hypothesis: The more non-participants perceive minipublic participants to be like them, the more legitimate they perceive minipublics to be (H1a).
The second, and related, response to the mirror critique is that we can empirically measure the extent to which citizens perceive minipublic participants to be like them against a more familiar point of reference. In the electoral context, we have long observed that voters often express a preference for a candidate who is like them, and candidates routinely invoke the similarity of their own background and experience to those of their would-be constituents (Mansbridge 2011). This ties into one of the contemporary problems facing representative democracies: it is not just that many citizens feel as though their interests are often ignored, but that their elected representatives are insufficiently like them to adequately understand their interests. Therefore, if citizens perceive minipublics to play a legitimate role in political decision-making, it might not simply reflect their perception that minipublic participants are like them, but it might also reflect a perception that elected politicians are not. In other words, it is a relative question: they may perceive citizen representatives to reflect their identity and experiences more effectively, or at least in a different way, than elected representatives. Thus, alongside H1a we hypothesise: The less non-participants perceive politicians to be like them, the more legitimate they perceive minipublics to be (H1b).
A refinement of the 'like me' argument is that the trait or quality most important to citizens' legitimacy perceptions will depend on the issue at stake. For example, gender representation might be more important for discussions on the topic of abortion; age might be more salient if the topic relates to pension policy; social class might play a more important role on the topic of social security; and so on. This is, again, based on the premise that descriptive representation on a certain dimension (the profile of the representative) will facilitate substantive political representation (what interests the representative defends, and what decisions he or she makes) (Castiglione & Pollak 2019). In other words, descriptive representation is not confined to the characteristics of (citizen) representatives alone, but is also linked to the presence of one's interests in the deliberation (Mansbridge 1999;Phillips 1995). What will matter to non-participants is that there are enough participants who are like them on the particular trait that is relevant to the minipublic's task, allowing them to be more confident of these participants producing a collective outcome that has, at least, considered their interests and is in the broader public interest. Therefore, we further hypothesise: When non-participants are aware of the issue at stake, they will consider it more important that minipublic participants are like them on the trait most relevant to the issue (H2).

Method Case selection
We empirically tested our hypotheses in the political context of Northern Ireland. As a polity marked by deep divisions, people's perceptions of political decisionmaking are particularly sensitive in this setting. Unionists, typically with a Protestant religious background, support Northern Ireland remaining in the United Kingdom and have traditionally been the majority group. Nationalists, typically with a Catholic religious background, support Northern Ireland leaving the United Kingdom to unify with the Republic of Ireland; they have traditionally been in the minority. After the creation of Northern Ireland and its devolved institutions in 1921, unionists were perpetually in government for five decades, leaving nationalists excluded. This experience of majority rule helped to create the conditions for ethno-national conflict in 1969 (see O'Leary 2019). The Belfast Agreement of 1998 helped bring an end to violence; pivotal to making this a reality was the establishment of power-sharing institutions, requiring both unionist and nationalist political parties to be represented in government (McGarry & O'Leary 2006).
As well as accommodating the two groups through inclusive decision-making arrangements at the elite level, the Agreement also outlined a mechanism for changing Northern Ireland's constitutional status: a popular referendum in which a majority of voters supported leaving the UK to join the Republic of Ireland. 1 With the UK's departure from the European Union, itself a major constitutional change with potentially significant consequences for Northern Ireland, the possibility of a united Ireland (and Northern Ireland rejoining the European Union) has become a more prominent fixture on the political agenda, presenting a challenge to the stability of powersharing. 2 However, the debate on constitutional change remains largely focused on broad principles: beyond the binary question of change versus the status quo, little attention has been devoted to what ' change' might (or ought to) constitute. Moreover, in the wake of the Belfast Agreement, a declining proportion of the population define themselves as unionist or nationalist, leaving many citizens who might be open to persuasion on the best way forward (Hayward & McManus 2019).

Minipublic stimulus
Against this backdrop, a deliberative minipublic was held on the subject of Northern Ireland's constitutional future in March 2019. Forty-nine participants, broadly representative of the Northern Ireland population, were recruited by an independent survey company to attend the one-day event at a hotel in central Belfast. It was organised as part of an academic project led by researchers at Queen's University Belfast. By design, there was no media presence or subsequent coverage. The purpose of the minipublic was to consider not just whether or not Northern Ireland should leave the UK to join a united Ireland, but the different forms that an alternative to the constitutional status quo could take. After listening to expert presentations on the different options, the participants engaged in facilitated group discussions. At the start and end of the day, the participants completed a questionnaire, asking them to indicate the extent to which they supported or opposed the various constitutional options. Participants' collective responses constituted the minipublic outcome.

Survey design
A short time after the minipublic was held, we administered a survey to a representative sample of the population via computer-assisted web interviewing (CAWI). Quota sampling was used to recruit 1,018 participants. 3 None of the respondents in the survey were participants in the minipublic (as verified by a screening question). For H1a and H1b, our dependent variable of interest is the perceived legitimacy of minipublics. In the survey, we operationalised this in three ways. First, alongside their attitudes towards different aspects of political decision-making, respondents were asked about their general attitude towards citizens' assemblies. Some basic background information was provided. As the text below shows, respondents expressed their view on advisory citizens' assemblies, whose recommendations would ultimately be up to politicians to approve or reject. Attitudes were measured on an 11-point scale where 0 meant 'strongly against' and 10 meant 'strongly in favour' (M = 6.7; SD = 2.9): In general, are you against or in favour of an advisory citizens' assembly on important issues affecting Northern Ireland? This means that about 30 to 50 citizens are selected at random, making sure that they broadly reflect the wider population on gender, age, social class, community background, and so on. They come together and discuss a certain issue with each other. They then deliver a recommendation for the Northern Ireland Assembly.
To move a step further and capture people's general trust in citizens as decision-makers, as distinct from the idea of ordinary citizens making recommendations for politicians to consider, we further asked respondents to indicate on a five-point Likert scale the extent to which they agreed or disagreed with the statement, 'I trust ordinary citizens to make political decisions that are in the public interest.' On this scale a value of 1 meant 'totally disagree' and 5 meant 'totally agree' (M = 3.6; SD = 1.1). 4 After providing these general views, respondents were then introduced to some brief background material on the minipublic that took place on the topic of Northern Ireland's constitutional future. For simplicity, the minipublic was described as a ' citizens' assembly'. 5 To promote engagement, the information outlining the process was summarised in an infographic (Figure 1). The outcome itself was presented to respondents in a graphic form (Figure 2), prefaced with some context: 6 Figure 1: Infographic presented to survey respondents on the minipublic.
In the citizens' assembly, the participants listened to expert presentations and discussed the issues in small groups. They considered these options. Participants of the citizens' assembly were asked to indicate the extent to which they were opposed to or in favour of each option on a scale from 1 (strongly opposed) to 7 (strongly in favour). The average score for each option is shown below (in Figure 2).
After presenting respondents with this outcome, we captured the third measure of our dependent variable: outcome acceptance. Here, they were simply asked: 'On a scale of 0-10, how willing are you to accept the outcome of the citizens' assembly?' On this 11-point scale, 0 represented 'not at all willing' and 10 represented ' completely willing' (M = 5.8; SD = 3.0).
Our key independent variable of interest is the perce ption that minipublic participants are similar to non-participants. After respondents were presented with back ground information on the minipublic (Figure 1), but before they were presented with the outcome (Figure 2), they were asked to think about the participants and indicate on a scale from 0-10 the extent to which they perceived the following statements to be accurate or inaccurate: 'The participants of the citizens' assembly are people like me'; 'The participants of the citizens' assembly have similar experiences to me'; and 'The participants of the citizens' assembly have a similar background to me'. Factor analysis show that responses load on one factor; factor loadings are all above .84; Cronbach's alpha is .92. Therefore these are combined to create a single scale (M = 6.1; SD = 2.3). For comparison, respondents were also asked the same questions with respect to politicians in Northern Ireland. Factor analysis on responses to these items also reveals one factor, with factor loadings all above .80; Cronbach's alpha is .89. Therefore, these three items were also combined into a single variable (M = 2.5; SD = 2.3).
We controlled for a number of alternative possible explanations of minipublic legitimacy. Those who are dissatisfied with the performance of representative democracy may be more likely to support the use of decision-making instruments that directly involve ordinary citizens (Bowler et al. 2007). Therefore, we controlled for the extent to which respondents are satisfied or dissatisfied with the way democracy works in Northern Ireland on an 11-point scale (0 = 'not at all satisfied; 10 = ' completely satisfied'; M = 2.5; SD = 2.6). Drawing on the 'new politics' and cognitive mobilization theses, those who have higher levels of political interest may be more likely to support democratic innovations (Inglehart 1997). We controlled for this using a basic measure of political interest (0 = 'not at all interested'; 10 = 'very interested'; M = 6.0; SD = 3.0). Previous research has found that anticipation of a particular outcome and thinking that you are in the majority are significant drivers of support for referendums (Werner 2020). Therefore, we controlled for majority perceptions by asking respondents to state their level of agreement with the following statement on a five-point Likert scale: 'The views that I have are generally shared by the people of Northern Ireland' (1 = 'totally disagree'; 5 = 'totally agree'; M = 3.2; SD = 1.0). In our analysis of outcome acceptance we also controlled for outcome favourability based on the extent to which respondents oppose or support Northern Ireland remaining in the UK (1 = 'strongly opposed'; 7 = 'strongly in favour'; M = 5.2; SD = 2.2).
Finally, to test H2, we captured respondents' priorities for minipublic composition (our final dependent variable), operationalised as the extent to which they thought it was important that members of different groups were represented in a minipublic considering the specific issue of Northern Ireland's constitutional future. This was asked after presenting the context of the minipublic (Figure 1) but before respondents were informed of the outcome (Figure 2): When thinking about the participants in the citizens' assembly that was asked to make a recommendation on the constitutional future of Northern Ireland, on a scale of 0-10 where 0 means not at all and 10 means very, how important is it to you that the citizens' assembly included the following groups in society?
The groups were identified from the information provided by respondents on their sex (women, men), social class (people from a working-class background, people from a middle-class background), age (younger people, older people) and ethno-national identity (people from a unionist background; people from a nationalist background; people from neither a unionist nor a nationalist background). 7 Each of these characteristics constituted our independent variables.

Perceived minipublic legitimacy
We tested our first set of hypotheses using OLS regression, as presented in Table 1. 8 As expected, we see that 'like me' perceptions of participants are associated with more positive general attitudes towards citizens' assemblies. The more non-participants perceive minipublic participants to be like them, the more legitimate they perceive minipublics to be. A one-unit increase in the perception that participants were like them (on an 11-point scale) is associated with a significantly higher level of support by 0.7 points (also on an 11-point scale), controlling for other possible explanations (in Model B). In contrast, we observe a significant negative relationship between our second independent variable and citizens' assembly support in the full model (but it is only marginally significant when introduced separately; see Table A5 in the Appendix). The less respondents perceive politicians to be similar to them, the higher their level of support for citizens' assemblies in general.
Our second measure of perceived legitimacy is the extent to which people generally trust citizens as decision-makers.

Trust in citizens as decision-makers (in public interest)
Outcome acceptance

Independent variables
Like  Again, in line with H1a, we observe in Models C and D a highly significant, positive relationship between the perceived similarity of citizen representatives and legitimacy perceptions. The more people perceived citizens' assembly participants to be like them, the higher the general level of trust in citizens to make political decisions in the public interest. Similarly, in line with H1b, the less respondents perceived politicians to be like them, the higher their level of trust in citizens as decision-makers. Our final test of H1a and H1b operationalised legitimacy perceptions as outcome acceptance. This time, respondents were aware both that a citizens' assembly had been held in Northern Ireland on the topic of its constitutional future, and of the overall post-deliberative preferences held by the participants -demonstrating a preference for Northern Ireland remaining in the UK. As with the other models presented in Table 1, there is continued support for H1a: perceptions of similarity explain perceptions of legitimacy. Crucially, respondents were significantly more likely to accept the overall outcome of the citizens' assembly when they perceived the participants to be like them even when controlling for outcome favourability (in Model F). This time, however, there is no support for H1b.
At this point, we can highlight some general observations across the six models. First, there is a clear pattern when it comes to the effect of 'like me' perceptions on the dependent variable, however operationalised. Not only is it a consistently significant predictor, but it also explains a high proportion of the variance across each of the models. This is especially true for attitudes towards citizens' assemblies in general, for which 'like me' perceptions of the participants produce an adjusted R 2 value of .27. Second, 'like me' perceptions of politicians have very little explanatory power. 9 Finally, and related to the previous two points, it is worth noting that many of the control variables lack much explanatory value. This is particularly striking given that (dis)satisfaction with democracy and political interest are often considered 'standard' drivers of citizens' attitudes towards minipublics. Here, they have no significant effect either way, underscoring that 'like me' perceptions constitute a unique and innovative predictor of minipublic legitimacy perceptions.

Priorities for minipublic composition
Having explored the relationship between the perceived similarity of decision-makers and the perceived legitimacy of minipublics, we now turn to the relative importance that group representation plays in a specific issue context. To test H2, we compared the extent to which respondents think it is important that the citizens' assembly on Northern Ireland's constitutional future should include participants from certain groups. The ANOVA results are reported in Table 2.
The first striking finding is that the inclusion of participants from all listed groups is considered to be relatively important by everyone, both within and between traits. Within traits, this is even true if these ordinary citizens belong to the ' opposing camp' on the issue at stake. For example, unionists think that it is rather important that nationalists are present (M = 7.8; SD = 2.9), nationalists think that it is rather important to include unionists (M = 7.9; SD = 2.9), and those who are neither unionist nor nationalist still attach relatively high importance to the representation of unionists (M = 7.1; SD = 3.3) and nationalists (M = 7.2; SD = 3.2). In each group, from Table 2 we see that the importance attached to including participants of different genders, age groups, social classes and ethno-national identities is roughly even. Second, while the differences might be relatively small in magnitude, respondents still prioritise the inclusion of minipublic participants who are like them on ethno-national identity -the salient trait in this particular context. In other words, unionists and nationalists still place a higher level of importance on the representation of participants who are unionist and nationalist respectively; these differences are statistically significant. This finding is consistent with the 'like me' argument, providing evidence in support of H2. 10 In contrast, the pattern is different for other traits. Gender plays no significant role: male and female respondents are just as likely to perceive the representation of men and women as being important. Turning to age, there are no significant differences in the perceived importance of including young participants across respondents in different age groups. In contrast, the importance given to including older participants is different for the various age groups. The perceived importance of including older participants is highest among older respondents (55+) and lowest among the youngest respondents (18-34 years old). Finally, for social class, we also see a mixed pattern. Both working-and middle-class respondents highly value the inclusion of working-class participants in the minipublic, but there are significant differences between each group in the perceived importance of including participants from a middle-class background.
The high favourability towards including participants who do not have the same trait(s) as oneself does not mean that non-participants want 'just anyone' to be included. As a point of reference, we asked respondents to indicate the relative importance of including politicians in the minipublic. 11 Notably, in contrast to other traits presented in Table 2, the respondents attached much less importance to including politicians (M = 4.3; SD = 3.3). Even though the perceived likeness of politicians has very low predictive power in explaining minipublic attitudes (in the earlier analysis presented in Table 1), non-participants have little appetite for the inclusion of politicians in the minipublic. In respondents' eyes, specific traits seem to matter less, as long as it is ' ordinary citizens' who are included. In other words, it is the ' ordinariness' of minipublic participants that appears most fundamental to shaping perceptions of similarity.

Discussion and Conclusion
Our empirical analysis yields three main findings. First, 'like me' perceptions constitute a significant predictor of a variety of minipublic legitimacy perceptions: general support for advisory citizens' assemblies, trust in ordinary citizens as decision-makers, and accepting the outcome of a citizens' assembly on a highly salient political issue. A perception that participants are similar to non-participants offers unique explanatory power. Second, and related, the less respondents perceived politicians to be like them, the higher their support for citizens' assemblies and the higher their level of trust in citizens as decision-makers. However, while statistically significant, this variable had little explanatory power. Third, we find some evidence that respondents prioritise the inclusion of minipublic participants who are similar to them on a series of traits, with the most consistent pattern emerging for the trait that is most closely connected to the issue at stake. But at the same time, the more striking overall pattern is that the inclusion of participants from different groups was considered to be relatively important across all respondents, regardless of their own profile. The key feature was that the participants should be ' ordinary' citizens.
Deliberative democrats have long argued that the legitimacy of minipublics rests, in part, on them being a microcosm of the broader population (Fishkin 2009). This article offers empirical evidence that this is the case from the perspective of citizens themselves. Minipublics, by definition, are exclusive bodies. They necessarily recruit only a sample of citizens, leaving most of the maxi-public unable to participate in the structured process of deliberation that helps characterise these democratic innovations. However, this does not mean Unweighted sample. Standard deviations in parentheses. + p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001. Because of skewed distributions for the dependent variable for each group, the robust F-statistic is based on the Brown-Forsythe test (Tomarken & Serlin 1986). The alternative calculation of a robust F-statistic developed by Welch did not provide substantially different results, generating confidence that the ANOVA results hold despite occasional violations of the assumption of homogeneity of variances.
that minipublics are necessarily detached from the maxipublic. On the contrary, non-participants appear capable of forming a psychological attachment to minipublics through a perception that the participants are like them. This in turn helps to explain the perceived legitimacy of minipublics and their outcomes. It is a bond that is created by the individual in his or her own mind; in this way it differs greatly from conventional bonds of representation which rest on the power of citizens to select and sanction politicians, but this difference does not diminish its contemporary significance. Indeed, recent research points to fundamental challenges facing representative democracies, with widespread popular perceptions of a gap between the public and politicians. Citizens perceive politicians, and politics more broadly, to be ' out of touch' or ' disconnected' from public life (see Dommett & Temple 2019;Grill 2007). In contrast, if greater use is made of representatives as ordinary citizens in a minipublic, members of the wider public may consider them more likely to share similar experiences and perspectives, leaving them in a better (if not perfect) position to represent their interests, compared to politicians attempting to do so alone. As Sintomer (2013) puts it, this does not require minipublic participants to act 'for' the public; what matters is that they are capable of acting 'like' and speaking 'like' other citizens. Therefore, rather than accepting Lafont's characterisation of minipublics as 'shortcuts' that undermine democracy, we instead contend that they can help take conventional representative democratic processes down a more scenic route. This comes with a number of important practical implications for democratic design. The first relates to communication and transparency: non-participants can only generate a perception that minipublic participants are 'like them' if they are aware of the minipublic and its selection method in the first place. By extension, this emphasises the importance of making the wider public aware of a minipublic's activities if it is to help build trust in the broader political system (Boulianne 2018). Second, while it is clear that people value minipublics due to their membership comprising ' ordinary citizens', it is less clear how minipublics should operate alongside existing decision-making processes involving politicians. Even though our results imply that minipublics can play a legitimate role in making more or less binding decisions, it is more plausible to envisage minipublics playing a more minimal role as trusted information proxies for other citizens, such as ahead of referendums (MacKenzie & Warren 2012), or supplementing representative decisionmaking under certain conditions (Kuyper & Wolkenstein 2019;Bächtiger & Goldberg 2020).
Our goal has been to establish whether or not an affective bond can connect non-participants and minipublic participants. While a cross-sectional survey facilitated this initial goal, our findings invite further investigation of the nature and strength of these bonds, such as whether or not certain types of citizens are more likely to hold 'like me' perceptions, and whether these bonds can be reflective as well as affective. For example, a survey experiment could examine the extent to which 'like me' perceptions might be influenced by exposure to different normative arguments (including Lafont's) or to variations in minipublic composition. Meanwhile, a longitudinal study in the wake of a minipublic could capture the durability of nonparticipants' perceptions of participants, including how these perceptions might be influenced by new information or new signals that emerge over time. Finally, research in other contexts will shed light on the generalisability of our present findings.
While the deeply divided case of Northern Ireland arguably offered a 'hard' test for our hypotheses, the absence of a devolved government for over two years at the time of the study may have created a particularly deep sense of frustration with the political system. Having controlled for satisfaction with the way democracy works in our analysis, it is still possible that the political context will have left some citizens feeling more positively disposed towards novel forms of decision-making. On the other hand, it is precisely in these challenging contexts that minipublics might reasonably play a targeted role in helping to restore trust in the broader political system. As this article shows, such citizens' assemblies do have the potential to play a legitimate role in decision-making, as explained, at least in part, by perceptions that their participants will be like non-participants. It will be through sustained empirical research that we can help discern whether minipublics at some point create undesirable shortcuts, and under which conditions, or whether they can continue to add value to democratic systems.

Notes
1 This mechanism is known as the 'principle of consent'. The Secretary of State for Northern Ireland is obliged to call a referendum if he or she considers it likely that a majority of those voting would support Northern Ireland leaving the UK to become part of a united Ireland. A referendum in the Republic of Ireland would also be required. 2 The Northern Ireland Executive and Assembly collapsed in January 2017 due to a range of significant differences among the main unionist and nationalist parties. A new power-sharing government was formed in January 2020. 3 Fieldwork was conducted between 8 August and 12 September 2019. See Table A1 for details on the composition of the sample. 4 In our survey, we also asked respondents to indicate their level of agreement with the statement, 'I trust ordinary citizens to make good political decisions.' Given that these two items on trust in ordinary citizens as decision-makers are highly correlated (r = .73) and yield highly similar results, we only focus on one item. 5 According to the criteria outlined by Setälä and Smith (2018), a citizens' assembly typically has 99-150 participants. However, the use of the term ' citizens' assembly' was used in this case to facilitate understanding from the perspective of survey respondents. Similarly, while minipublic participants were described as being 'randomly selected' in Figure 2, despite in practice being recruited via quota sampling, this terminology was not used in a statistical sense but rather to distinguish the recruitment method from forums involving self-selection. 6 Participants in the minipublic were not forced to choose a preference between the three options. Instead, as Figure 2 summarises, they indicated their post-deliberative level of support for each option. The status quo option (of Northern Ireland remaining in the UK) received the highest level of support. 7 For social class, we categorise respondents according to standard ABC1 (middle class) and C2DE (working class) classifications based on occupational background. 8 We use unweighted data in the analysis that follows.
We replicated the analysis with weights for age, sex and community background; the results were substantively similar and are available on request. 9 See Tables A5-A8 in the Appendix. 10 The pattern is more complicated when considering the group identifying as 'neither' nationalist nor unionist. When asked about the importance of including 'neither' participants, the only significant difference (p < 0.05) that emerges in Games-Howell pairwise comparisons is between 'neither' respondents who find this more important than unionist respondents (but not significantly more so than nationalist respondents). 11 While minipublics typically comprise lay citizens as participants, some design choices deviate from the norm to include politicians within the membership (see Farrell et al. 2020).

Additional Files
The additional files for this article can be found as follows: •