Mini-Publics and the Wider Public: The Perceived Legitimacy of Randomly Selecting Citizen Representatives

ABSTRACT There are two important dimensions to the membership of mini-publics that are distinct from the membership of conventional representative institutions: the selection mechanism (sortition) and the profile of the body's eligible membership (‘ordinary’ citizens). This article examines the effects of these design features on perceived legitimacy. A survey experiment in the deeply divided context of Northern Ireland finds no evidence that variation in mini-public selection features has an overall effect on perceived legitimacy, but there are important individual-level differences.


Introduction
The concept of ordinary citizens being randomly selected to make decisions on behalf of the wider public is one that is both well established and familiar to most people, but only in one form: the courtroom jury.However, the concept can also be applied to the political sphere through the use of deliberative mini-publics (see Smith, 2009;Urbinati and Warren, 2008).These democratic innovations are defined by two core features.
First, they are deliberative in that participants reach recommendations or conclusions on a particular issue 'after receiving information and engaging in a careful and open discussion' (Farrell et al. 2019: 5).Second, they are mini-publics in that the participants broadly constitute 'a representative subset of the wider population', delivered through random selection (ibid).In a narrow sense, deliberative mini-publics may be used to help elected representatives tackle specific political issues, serving a problem-solving role.In a more ambitious sense, through the inclusive participation of citizens in structured deliberation, these democratic innovations can be understood as attempts to address some of the procedural deficiencies of representative democracy.
The increasing prevalence of mini-publics around the world has been accompanied by a growing body of evidence that citizens value the experience of taking part (Jacquet, 2019;Knobloch and Gastil, 2015) and that they are capable of engaging in high-quality deliberation (Fishkin, 2009;Suiter et al. 2016).However, empirical research has overwhelmingly focused on the behaviour and attitudes of mini-public participants.Some studies have begun to address the previously neglected psychological relationship between mini-publics and non-participants.They demonstrate that demand for such initiatives is generally high, including among citizens with high levels of political engagement and high levels of dissatisfaction with representative democracy (Bedock and Pilet, 2020), as well as those who are less likely to participate in conventional politics (Neblo et al. 2010).In addition, recent studies show that mini-publics and other democratic innovations involving ordinary citizens, such as participatory budgeting, can have a positive influence on the perceived legitimacy of decision-making from the perspective of the wider public (Boulianne, 2018;Jäske, 2019).
This article focuses on how the design of one key element of a mini-publicthe selection of participantsaffects public perceptions of legitimacy.It does so in the post-conflict context of Northern Ireland, where high levels of dissatisfaction with representative institutions offer compelling grounds to experiment with democratic innovations (see, for example, Haughey, 2019), but where deep political divisions pose particular challenges in designing institutions that command widespread support (Morrow, 2005).By examining different approaches to participant recruitment for one type of mini-publica citizens' assemblythis study weighs up the relative effects of sortition (over election) and citizen representatives (over professional politicians) on perceptions of legitimacy.The results of an online experiment show that, overall, a mini-public with the purest selection processthat is, one composed exclusively of randomly selected citizensis perceived as equally legitimate as one in which members are elected, or one in which members are a combination of ordinary citizens and elected politicians.Compared to other models, the purest selection model is regarded as particularly legitimate among individuals who attach a high level of importance to political equality.The findings suggest that mini-public organisers have a degree of flexibility over design choices without sacrificing perceptions of legitimacy from the perspective of the wider public, even in deeply divided political settings.

Selecting Mini-Public Participants
While the random selection of citizen representatives can be considered innovative in the context of contemporary political systems, the practice can be traced to early forms of democracy.In ancient Athens, the widespread use of sortition was designed to help promote political equality (Owen and Smith, 2018: 421).By giving all citizens a theoretically equal chance of being selected to participate in democratic decisionmaking, sortition embodies the 'idea that all citizens are equally capable of political judgement and equally responsible for the public good' (Jacquet, 2017: 641).In the context of mini-publics, a more commonly articulated justification for the modern use of sortition is to promote descriptive representation (Fishkin, 2009).This is especially true for larger forms of mini-public, such as citizens' assemblies, which typically contain between 99 and 150 participants (Setälä and Smith, 2018: 301).For example, Thompson (2008) highlights the importance of descriptive representation in establishing the legitimacy of the pioneering British Columbia Citizens' Assembly (BCCA) on electoral reform, where a multi-stage stratified random sampling procedure was used to generate a body of citizens broadly similar to the demographic composition of the broader population (see Fournier et al., 2011: 54). 1 Similar selection procedures were applied in constructing the Ontario Citizens' Assembly, the Dutch Burgerforum, the Irish Citizens' Assembly, and citizens' assemblies on climate change in the UK and France.
From a normative perspective, we can rely on principles such as political equality and descriptive representation as objective criteria against which the 1 While sortition helps to promote descriptive representation, there are two main limitations.First, despite helping to reduce the biases observed in participatory bodies that rely purely on self-selection (Warren, 2008: 59), the mechanism still entails an element of self selection in that invited participants may still refuse to participate.Second, random selection relies on scale to successfully obtain a representative sample; stratification can help to reduce the risk that certain groups are over-or underrepresented, but this still requires the identification of relevant strata.Notably, for example, while there was stratification for gender, geography and age in the BCCA, it failed to stratify for ethnicity.Representatives from BC's Indigenous population were subsequently appointed to correct an initial imbalance arising from random selection.legitimacy of mini-public properties may be evaluated.However, if we are to better appreciate the potential for mini-publics to help reduce democratic deficits, understood as a gap between citizens' aspirations and their satisfaction with the way democracy works (see Norris, 2011), it follows that we must also pay attention to people's subjective attitudes toward these novel forms of decision-making.This requires us to consider the concept of legitimacy through an empirical lens.Research in political psychology has routinely found that the way in which decisions are reached, distinct from the substance of the decisions themselves, has an influence on perceived legitimacy -the 'belief that authorities, institutions, and social arrangements are appropriate, proper, and just' (Tyler, 2006: 376).In other words, perceived legitimacy is, at least in part, a function of procedural design.
When it comes to mini-publics, there are two salient dimensions on which their selection features differ from conventional representative institutions.First, there is the selection mechanism: sortition as opposed to election.Second, there is the profile of the body's eligible membership: 'ordinary' citizens as distinct from professional politicians.2It should be noted that mini-publics themselves may vary along these dimensions: organisers may deviate from the purest modelthe random selection of lay citizensin order to promote other goals.It is not the objective of this article to defend mini-public features according to desirable normative criteria; instead, recognising both the relative novelty of mini-publics and their potential variation in design, this article investigates the extent to which the core selection features of minipublics influence their legitimacy as perceived by the wider public.

Sortition versus election
As Farrell et al. (2019: 5) note, random selection is the design feature that separates mini-publics from other democratic bodies, producing 'a certain mindset in the room, which is very different to that resulting from a selection process governed by election' or other mechanisms.But what about the attitudes of those outside the room, in the wider public?Elections remain, overwhelmingly, the dominant mechanism by which political representatives are selected to serve in conventional legislatures.To most citizens, they provide an equal opportunity to express their political preferences and have them formally represented in decision-making; elections are, thus, understood to lay the foundations for representative democracy (Manin, 1997).Regardless of the relative merits of sortition over election (see Stone, 2016), the concept of voting for political representatives is at least familiar to virtually all citizens.This baseline familiarity might contribute to a perception that elections are the most legitimate instruments for the selection of decision-makers.For this reason, combined with a range of political factors, some real-world citizen bodies have deviated from the method of selecting mini-public members via sortition and instead relied on a conventional voting procedure.Most notably, members of the Icelandic Constitutional Assembly were selected by a national election (Bergmann, 2016). 3It differed from a normal legislative election in that most candidates were lay citizens, unaffiliated with political parties; indeed, the president and members of parliament were explicitly precluded from standing (Participedia, 2020).Thus, despite being elected, members of the Constitutional Assembly could not be described as 'politicians' in the usual sense of the term.In total, 522 individuals put themselves forward, having obtained signatures from at least 30 supporters, and 25 were elected.While they included people from 'a broad range of backgrounds,' critics alleged that, 'only previously well known individuals had been elected' (Bergmann, 2016: 22).This, together with a low level of participation from the broader electorate (turnout was just 37 percent), cast serious doubt over the representativeness of those selected and the extent to which the process promoted political equality.However, the case does at least highlight the possibility of lay citizens being selected to a body through a mechanism other than sortition.

Ordinary citizens versus mixed memberships
The Irish Constitutional Convention (2012-2014) took a different approach to moderating the potential novelty of a mini-public by blending its design with a more conventional concept.Rather than injecting familiarity through the selection mechanism, it adopted a mixed membership profile: two-thirds (66) were lay citizens selected at random from the Irish population; the remaining 33 members were professional politicians appointed by political parties (in proportion to their representation in the lower chamber of the legislature). 4The inclusion of politicians in the membership of the body was a way of 'anchoring the process in the political system, making it more likely that the convention's recommendations would receive a fair hearing' (Farrell et al. 2020: 55).This design choice appeared to have little effect on the quality of deliberation in group discussions or on the overall outcomes, but Farrell et al. (2020) do report evidence of an ideological bias among politician members.It should be noted that in the two subsequent citizens' assemblies established by the Irish government, the mixed membership model was abandoned: the 99 members in each process held since 2016 have been lay citizens (serving alongside an appointed chairperson).Perhaps the inclusion of politicians in the initial Constitutional Convention helped to secure 'buy in' from political elites for the use of mini-publics in decision-making (Suiter et al. 2016), but the effect of these different design choices on legitimacy perceptions among the broader Irish public is not clear.

Hypotheses
At this point we can see that a number of design choices are available to mini-public organisers, but our understanding of the effect of these choices on public attitudes is limited.Some studies find support for the idea of ordinary citizens being more directly involved in politics, including through deliberative forums (Gastil et al. 2016;Neblo et al. 2010).Others find that awareness of mini-publics can increase political efficacy (Boulianne, 2018) and influence opinion change (Ingham and Levin, 2019), even among non-participants.However, we lack systematic evidence on people's attitudes towards the two key selection features of citizens' assemblies: the mechanism of sortition (compared to election) and members' profile as lay citizens (not professional politicians).If we take as our point of departure that a mini-public in its purest form should comprise citizens who are randomly selected (as per Farrell et al. 2019), we would expect that the further a mini-public deviates from this model, the lower its perceived legitimacy: H1: The perceived legitimacy of mini-public decision-making is highest under conditions of membership being exclusively citizens and citizens being selected by sortition, and lowest under conditions of membership being a mixture of citizens and politicians and citizens being selected by election.
One normative justification for randomly selecting mini-public participants is to promote political equality.The introduction of either elections or politicians, or both, to the selection process may weaken the perceived capacity of mini-publics to effectively promote political equality in practice.Therefore, individuals' level of attachment to the principle of political equality is expected to moderate the relationship between mini-public selection features and legitimacy perceptions:

H2:
The perceived legitimacy of a mini-public with the purest selection process will be higher among individuals who value political equality.
Drawing on descriptive representation as a further normative justification for randomly selecting citizens to serve in mini-publics, we may further expect that people who are under-represented, either through their political attitudes or group membership, will be particularly supportive of mini-publics with the purest selection features.These hypotheses will be specified in the next section which situates the present study in the political context of Northern Ireland.

The Case of Northern Ireland
Mini-publics have been held in a number of deeply divided political settings; more often than not, and perhaps contrary to expectations, they illustrate how citizens from rival groups are still capable of engaging in respectful deliberation on sensitive political issues (see O'Flynn and Caluwaerts, 2018).As in 'normal' democracies, however, we lack evidence on the attitudes of non-participants towards these processes and their design features.Yet, in such polities, including those emerging from conflict, it can be particularly challenging to design and implement decision-making arrangements that command widespread legitimacy (Morrow, 2005: 45).In Northern Ireland, with a recent experience of ethno-national conflict, political stability has proven difficult to consolidate.Unionists, who typically come from a Protestant community background, seek the preservation of Northern Ireland's place in the United Kingdom; nationalists, who typically come from a Catholic background, seek the unification of Northern Ireland with the Republic of Ireland.The Belfast/Good Friday Agreement of 1998 offered the basis of an institutional solution to conflict in the region by accommodating these two main communities, but deep divisions remain.
This challenging context offers an appropriate setting to test the hypotheses formulated in the previous section for three main reasons.First, Northern Ireland's postconflict institutions reflect an elite-centric logic, in that they emphasise power-sharing among political elites over the improvement of community relations at the grassroots level (Nordlinger, 1972).In practice, not only have elites often found themselves in regular gridlock (see Haughey 2019), but the logic of the system creates a gap between citizens and their representatives.This context opens the possibility for the use of minipublics, such as citizens' assemblies, to help politicians address issues that cause gridlock and to help strengthen people's perceptions of the legitimacy of decisionmaking more generally. 5Second, at the time of the study, Northern Ireland had no direct experience of citizens' assemblies, making the concept sufficiently novel.In other words, while advocates of mini-publics may highlight their potential to improve the quality of representative democracy, their key attributes remained untested from the perspective of the broader public.Third, the static nature of Northern Ireland's institutions raises questions for democratic representation in a post-conflict context.These institutions were created to facilitate the sharing of power between unionist and nationalist representatives, but a growing number of people, especially those growing up without any direct experience of conflict, are identifying as neither nationalist nor unionist (see Hayward and McManus, 2019).Figures 1 and 2 illustrate striking differences in the distribution of ethno-national identity in the Northern Ireland Assembly compared to the wider population: since 1998 no fewer than 88 percent of elected representatives have designated as unionist or nationalist, whereas a plurality of citizens have identified as neither since 2006.A system designed to promote inclusion, paradoxically, does not adequately accommodate this emerging moderate group through the electoral process and existing representative institutions.
[Figures 1 and 2 about here] Therefore, following the logic of descriptive representation, we may expect ethnonational ideology to moderate the relationship between the mini-public selection features and perceived legitimacy.As a result of their under-representation in the Northern Ireland Assembly, individuals who identify as neither nationalist nor unionist may be particularly supportive of mini-publics with the purest selection features, whereas those who identify as nationalist or unionist are less likely to support a minipublic selection process that rebalances their over-representation:

H3:
The perceived legitimacy of a mini-public with the purest selection process will be higher among individuals who are ethno-nationally moderate, and lower among those with stronger ethno-national positions.
Finally, community background has been a very sensitive aspect of political representation in Northern Ireland.Historically, members of the traditional minority community, Catholics, experienced systematic political inequalities in an electoral process dominated by the Protestant majority.Generally speaking, Catholics have been more supportive of electoral reform in the direction of proportional representation, perceived to be a means of redressing historic imbalances, whereas Protestants tend to be more supportive of majoritarian or plurality-based electoral systems through which the security of their status can be better maintained (see Garry, 2016).Therefore, owing to perceptions of group interest, we may expect Protestants to be less supportive of reform that disrupts the status quo; correspondingly, we may expect Catholics to be more supportive of radical reform: H4: The perceived legitimacy of a mini-public with the purest selection process will be higher among Catholics compared to Protestants.

Data & Method
The hypotheses were tested in an online experiment in March 2018.Participants were recruited from an opt-in panel hosted by LucidTalk, a survey company based in Northern Ireland.While such convenience samples are not representative of the target population, they offer greater diversity than student samples, provide cost-effective alternatives to random probability samples, and have been found to produce substantially similar treatment effects to representative population-based samples (Mullinix et al. 2015).A total of 329 adults living in Northern Ireland took part in the present study. 6Respondents were randomly assigned to one of four experimental conditions.They were presented with a description of a citizens' assembly, varying according to the profile of decision-makers (either all citizens or an even combination of citizens and elected politicians) and the mechanism by which the citizen members are selected (via sortition or election).The four conditions are summarised in Table 1.
[Table 1 about here] Each of the short vignettes provided respondents with some necessary contextual information about the concept of a citizens' assembly and why such a body might realistically be considered for political decision-making in Northern Ireland, with every effort made to keep the text as simple and concise as possible.To help isolate participants' attitudes to the design features of the process described, distinct from their attitudes to other parts of Northern Ireland's political system, respondents were told that the decision of the citizens' assembly would be politically binding.Respondents had the opportunity to provide open-ended feedback at the end of the survey.The submitted comments did not indicate that respondents had any difficulty in imagining a citizens' assembly as a realistic possibility, but the respondents did use the opportunity to express their views about the relative desirability of certain features described; indeed, these comments underscore the very rationale for conducting this kind of study.The manipulated text is shown in italics (with bold emphasis as presented to respondents): In Northern Ireland, there are some issues on which politicians are divided over the way forward.On these issues, it has been proposed that a [version of a] citizens' assembly could decide on the way forward instead.[Group 3]: The citizens' assembly would consist of 100 ordinary members of the public.Anyone would be able to put themselves forward to be a member, apart from politicians who currently hold elected office.They would then face an election.Out of all people who put themselves forward, the 100 people with the most votes would be selected to sit on the citizens' assembly.
[Group 4]: The citizens' assembly would consist of 50 ordinary members of the public and 50 elected politicians.Any member of the public would be able to put themselves forward to be a member.They would face an election.Out of all members of the public who put themselves forward, the 50 people with the most votes would be selected to sit on the citizens' assembly.The political parties would appoint the remaining 50 membersin proportion to their seats in the Northern Ireland Assembly.
The citizens' assembly would consider the contentious issue, consider the evidence on different perspectives and different arguments, and then decide on the best way forward.
The decision would be final.In other words, the government would accept in advance the decision of the citizens' assembly and agree to implement it.
Underneath the text, respondents were presented with an infographic summarising the manipulated selection process from each condition.The purpose of the infographics was simply to summarise and reinforce the vignettes, facilitating their comprehension.
Apart from the necessary variation needed to operationalise the independent variable, the four visualisations (presented in Figure 3) were designed to be as simple and consistent as possible.
[Figures 3a, 3b, 3c and 3d about herethey could be collated to produce a single figure, Figure 3] Online surveys have a number of advantages.They facilitate the efficient randomisation of respondents to experimental conditions, maximise the practical convenience of participation, and encourage respondents to provide honest answers to questions.These benefits are accompanied by a number of trade-offs, the most significant being the researcher's relative lack of control over the experiment itself.As such, two short manipulation checks were embedded in the questionnaire before the main set of response items, asking participants to identify both the profile of decision-makers in the citizens' assembly described and the mechanism by which citizen members were selected.The vast majority of respondents in this experiment passed both manipulation checks. 7An average of 93 percent of participants correctly identified the profile of the citizens' assembly described in each condition (see Figure 4).The pass rate for this check was slightly lower for Group 2, in which 88 percent of respondents correctly identified a mixed profile of ordinary members of the public and politicians, but variation in pass rates across groups was not statistically significant.
[Figures 4 and 5 about here] In Groups 1, 2 and 4, over 95 percent of respondents correctly identified the selection mechanism, but the pass rate was lower for Group 3, in which 77 percent of respondents (correctly) indicated that citizen members would be chosen by an election and 23 percent (incorrectly) stated that they would be selected by sortition (see Figure 5).The higher number of respondents in Group 3 failing to correctly identify the selection mechanism appears to reflect the nature of the citizens' assembly described.Openended comments left by participants suggest that, despite being prompted about their incorrect answer, many persisted with their response that members would be randomly chosennot electedbecause this is how they thought citizens should be selected, or because they thought that the concept of electing ordinary citizens was simply not logical.These participants ignored the wording of the question, which asked how members of the citizens' assembly 'would' be selected, not how they ought to be chosen.Their responses appear to have been based on their subjective preferences for the design of a citizens' assembly, not the selection mechanism specified in the treatment.Therefore, only the 283 respondents who passed both manipulation checks are included in the analysis that follows. 8fter reading the vignette and completing the manipulation checks, respondents were asked to evaluate the process described.The full wording of the base question and response options was: 'Imagine the way in which a citizens' assembly would deal with a political issue.As a way of making a decision, to what extent do you think this process would be … fair or unfair; trustworthy or untrustworthy; democratic or undemocratic; efficient or inefficient; even-handed or discriminatory; acceptable or unacceptable; good or bad; competent or incompetent; supportable or unsupportable; credible or not credible?'Responses were recorded on a seven-point scale and combined to produce a mean score of perceived legitimacy, the outcome variable. 9The development of the scale began with the identification of the 'relevant content domain' (DeVellis, 2016: 84).Previous studies have measured the broad 'complex and multifaceted' concept of legitimacy by capturing a range of different components (Weatherford, 1992: 53).
These entail a perception of procedural fairness (Tyler, 1990;Sunshine and Tyler, 2003); trust (Miller, 1974;Sniderman, 1981;Sunshine and Tyler, 2003;Gibson et al. 2003), confidence in the process (Smith, 1981;Sunshine and Tyler, 2003); Gibson et al. 2003), perceived efficiency or value for money (Miller, 1974), perceived competence of decision-making (Miller, 1974), a belief that decision-makers are doing a good job (Hibbing and Theiss-Morse, 1995;Gibson et al. 2003), a perception of nondiscrimination (Miller, 1974;Sunshine and Tyler, 2003;Gibson et al. 2003), a basic belief that the process is democratic (Schmitt, 1983), a belief that decisions should be accepted (Esaiasson, 2010), and a perception that the process is worthy of support (Tyler, 1990;Gibson et al. 2003).The final ten-item scale is unidimensional and internally reliable (α = .97).10A one-way ANOVA will initially be used to test the overall effect of mini-public selection features on perceived legitimacy.
Finally, a number of items were included in the questionnaire to facilitate individual-level analysis.Support for the democratic principle of political equality was measured by asking respondents to indicate on a five-point scale the extent to which it is important 'that representation promotes political equality for everyone' (1 = 'Not at all important'; 5 = 'Very important').Ethno-national ideology was captured by asking respondents if they identified as (very strongly/fairly strongly) unionist or nationalist, or neither unionist nor nationalist.Responses were recoded to create a variable for ethno-national moderation (0 = very strongly unionist/nationalist; 1 = fairly strongly unionist/nationalist; 2 = neither unionist/nationalist).Respondents were also asked to indicate their community background: Catholic, Protestant, other religion or no religion, with a further category of 'prefer not to say'.Multiple regression will be used to analyse whether any of these variables moderate the effect of mini-public design on perceived legitimacy.

Results
Across the four experimental conditions, Figure 6 shows little difference in the extent to which respondents perceived each model of citizens' assembly to be legitimate.The two sortition-based models, both with citizens only (M = 3.66, SD = 1.95) and mixed members (M = 3.67, SD = 1.59), received the highest average legitimacy scoreseach in the upper half of the 0-7 scale.The two election-based models consisting of citizens only (M = 3.26, SD = 1.38) and mixed members (M = 3.21, SD = 1.62) received slightly lower scores, falling below the midpoint.However, overall, a one-way ANOVA reveals no significant main effect (F(3, 278) = 1.05, p = .37).In other words, the mean legitimacy scores for each experimental condition are not significantly different from one another.Thus, H1 is rejected.
[Figure 6 about here] In the absence of an overall effect, however, it is still important to consider whether there are relevant individual-level variations (Mutz, 2011: 98).Our hypothesised moderating effects are formally tested using multiple linear regression, the results of which are presented in Table 2.In each model the dependent variable is the perceived legitimacy of the proposed citizens' assembly.The experimental conditions are included as dummy variables, with the randomly-selected citizens-only condition as the reference category.The additional independent variables considered are the perceived importance of political equality, ethno-national ideology, and community background.
Starting with the role of political equality, we observe a statistically significant effect of the perceived importance of this democratic principle on the perceived legitimacy of a mini-public with the purest selection process.Here, as shown in the second column, a one-unit increase in the perceived importance of political equality (on a 1-5 scale) is associated with a 0.8-unit increase in perceived legitimacy.This effect is not significantly different for the other condition featuring a citizens' assembly with a citizen-only membership.However, the effect differs in the other two conditions: individuals who attach greater importance to political equality report lower levels of perceived legitimacy for mixed-member citizens' assemblies compared to a citizens' assembly with the purest selection process.These interaction effects reach statistical significance, although it is only for the mixed-member citizens' assembly selected by election that the significant difference holds in the full model (in Column 5), providing partial support for H2. 11  Drawing more closely on the political context of Northern Ireland, two further moderating effects are considered.Turning to ethno-national ideology, we see that higher levels of ideological moderation are associated with higher levels of perceived legitimacy for the citizens' assembly with the purest selection process.A one-unit increase in ideological moderation is associated with a 0.6-unit increase in the perceived legitimacy of a prototypical citizens' assembly.The third column in Table 2 demonstrates that this effect is not uniform across all treatment groups.In the experimental condition featuring a mixed-member citizens' assembly selected by sortition, higher levels of ethno-national moderation were associated with lower levels of perceived legitimacy compared to the reference category.However, this moderating effect is only significant at the p < .10 level and fails to reach statistical significance in the full model, leaving us without strong or consistent evidence to support H3.
Similarly, the regression analysis leads us to reject H4.Compared to Protestant respondents, mean legitimacy scores among Catholic respondents were over a unit higher for the citizens' assembly selected by the purest selection process, but there were no significant differences in this effect for the other experimental conditions.
[Figure 7 about here] 11 The full model excludes respondents who do not identify with a Catholic or Protestant background.Therefore, the interaction effects of the perceived importance of political equality and ethno-national moderation are first tested separately (in Models 2 and 3) on the full sample that includes respondents with another or no religion.
It is possible that the lack of a main effect suggests that respondents simply do not care about mini-publics, regardless of how they are designed.Therefore, as a check on the robustness of respondents' perceptions of (hypothetical) citizens' assembly decisionmaking, all were asked to imagine the way in which the (real-world) Northern Ireland Assembly makes decisions and to evaluate its decision-making according to the same ten items on which they had evaluated the legitimacy of each respective version of citizens' assembly.At the time of the experiment in March 2018, it should be noted that the Northern Ireland Assembly had not been sitting for over a year; it collapsed along with the Northern Ireland Executive in January 2017 and was not restored until three years later (see Haughey, 2019).Therefore, for the purposes of their evaluations, respondents were to imagine the way in which the Northern Ireland Assembly makes decisions 'when it is functioning.'In all four conditions, the mean legitimacy score for decision-making by the Northern Ireland Assembly was lower than the mean score for each model of citizens' assembly.The dysfunctional nature of the Northern Ireland Assembly at the time of the experiment is likely to have contributed to particularly negative evaluations, even though respondents were asked to ground their evaluations of the Assembly on the basis of its normal operation and not on the basis of its then state of dormancy.In reality it may have been difficult for many respondents to make this distinction.However, it is precisely the prevalence of such negative perceptions of representative institutions, especially during times of dysfunction, that arguably strengthens the case for the wider use of mini-publics in decision-making, among other possible reforms.
Therefore, while differences in the profile of citizens' assembly members and the selection method of citizen members had no main effect on the perceived legitimacy of citizens' assembly decision-making, Figure 7 demonstrates that respondents are far from indifferent to different types of decision-making process.Compared to their evaluation of the Northern Ireland Assembly, respondents are, on average, favourably disposed to decision-making by citizens' assemblies selected in different ways.

Discussion & Conclusion
The process by which mini-publics are selected is a defining feature of these democratic innovations (Farrell et al. 2019).In their purest form, all members of mini-publics are 'ordinary' citizens who are selected at random from the wider public.The profile of decision-makerslay citizens, as opposed to professional politiciansand the mechanism used to select themsortition, as opposed to electionare both intended to promote democratic principles such as political equality and descriptive representation.In theory, all citizens stand a roughly equal chance of being selected to participate in a mini-public, and the final composition of participants should broadly mirror the demographic and attitudinal profile of the population at large.And yet the two key dimensions of mini-public selection remain conceptually novel to most people.
For this reason, organisers of mini-publics may wish to retain more conventional features of existing democratic institutions, such as selecting lay citizen members via an electoral process, or by engineering a membership profile that includes both lay citizens and professional politicians.This article offers no evidence that, overall, such deviations from the purest design features have a significant influence on the perceived legitimacy of a citizens' assembly.Of course, accepting the null hypothesis in this case does not prove the absence of the relationship, and so we must treat this finding with appropriate caution when considering its implications for mini-public design.
One possible implication is that organisers are afforded a certain degree of flexibility in the design of mini-public selection procedures, commensurate to contextual demands.If, for example, a government commissions a citizens' assembly or another form of mini-public in the face of wider scepticism among political elites, the inclusion of politicians as members of the body themselveswith all political parties proportionally representedmay help to secure the endorsement of the process and its outcomes from the political establishment.Such a scenario is imaginable in Northern Ireland, where there is an uneven degree of enthusiasm for the concept of a citizens' assembly among political parties; unionist politicians have been somewhat wary of the idea (Donnelly, 2015).However, while this mixed membership configuration certainly helped to generate 'buy-in' from political parties in the Irish context (Suiter et al. 2016), it does risk changing the dynamic within mini-publics in other settings, particularly if partisan divisions contributed to the perceived need for a mini-public in the first place. 12eanwhile, if sortition were deemed by the commissioning body to be too radical a departure from conventional selection mechanisms, then members could be selected by an election without apparently undermining public perceptions of the minipublic's legitimacy, all else equal.However, this dimension of selection is perhaps more sensitive to external variables.Sortition, by definition, produces a predictable outcome: a sample of the population that is representative on certain parameters to within a recognised (but low) margin of error.In other words, if a random sampling procedure is implemented rigorously and transparently, the outcome is purely a function of this mathematical procedure.Elections also have many predictable qualities, but such predictability relies on political parties providing structure to competition between candidates.Without parties to provide a sense of order, the process of selecting fellow citizens may become too complicated for voters to feel like they can meaningfully engage, as the experience of Iceland illustrates (Bergmann, 2016: 23).
Besides questions of practicality and political context, the design of a process for selecting members of a mini-public should also be informed by underlying democratic principles.The individual-level results of the experiment provide evidence that a mini-public with the purest selection process is more positively associated with the promotion of political equality, compared to mixed-member designs.This suggests that the profile of participants matters more than the selection mechanismat least to the citizens who most value political equality.This finding is interesting in light of a possible criticism of the experiment: that the information provided in the vignettes in the sortition-based conditions risked positively priming respondents.Contextual details were included, such as random selection being intended to generate a 'cross-section' of the population, as it is unlikely that respondents would otherwise have been familiar with the operation of sortition and its basic effect of generating a representative sample of the wider population.In the end, this information does not seem to have made a difference to participants' responses.A further study would thus be useful to clarify people's baseline awareness of the principles underlying random selection, disaggregating the specific elements that are related to its perceived promotion of goals such as political equality and descriptive representation, and how these differ from key elements of an electoral method of selection.
In many respects, the key point of mini-publics is precisely to be different to other elements of the political systemnot to present a challenge to other processes and institutions, but to help the overall system to function better.From this perspective, the general lack of moderating effects for ethno-national ideology and community background offers encouraging potential for the application of citizens' assemblies to deeply divided polities.Now, it should be noted that legitimacy perceptions were generally higher among Catholics and ethno-nationally moderate individuals compared to Protestants and individuals with stronger ethno-national positions, but these did not vary significantly according to experimental condition.Therefore, it appears that minipublic perceptions in general are particularly high among citizens who are underrepresented by conventional institutions on a salient political dimension, and among members of a group that has traditionally been in the minority.Mini-public organisers should be mindful of this potential variation in legitimacy perceptions at the individuallevel, such as by developing communication strategies that target those most likely to be sceptical of the entire concept, but it does not appear that such variation is significantly influenced by design choices in the selection of mini-publicsat least in terms of the features considered in this particular study.
If the motivation behind establishing a mini-public is to try and strengthen democratic decision-making, by virtue of its selection procedure promoting principles such as political equality and descriptive representation in a way that other institutions do not, there are compelling reasons to the adopt model with the purest selection features, even in a deeply divided place.And yet, decision-making by each of the four models of citizens' assembly considered in this analysis was consistently perceived to be more legitimate than decision-making by the Northern Ireland Assembly.Regardless of variation in the design of their selection processes, this finding in itself highlights the potential for citizen-based decision-making to play a constructive role alongside conventional representative institutions.

Citizens only
Mixed members

Selection Mechanism
Sortition    Across the four conditions of the experiment on mini-public selection processes, there were no significant differences in the extent to which respondents perceived each model of citizens' assembly to be legitimate (F(3, 328) = 1.92, p = .13).However, based on these results it is not possible to accept the null hypothesis with an adequate degree of confidence.The risk of a Type II error is high due to the low percentage of respondents passing the two manipulation checks embedded in the questionnaire; thus, it is not sufficiently clear that the manipulations themselves had no significant effect on the outcome measure.In fact, as Table S1 shows, only a minority of respondents correctly identified an important aspect of the manipulation in three of the conditions.

List of Figures
The pass rates vary significantly across the four groups (χ 2 (3, N = 332) = 31.510,p < .01).It was only in the prototypical condition that a majority of respondents, 65 percent, passed the manipulation checks. [

Table S1: Percentage distribution of respondents passing or failing manipulation checks in initial study
It is worth recalling that the manipulation checks for this experiment consisted of two separate questions with binary response options.First, respondents were asked about the profile of the citizens' assembly described: were the members "100 ordinary members of the public," or were they "50 members of the public plus 50 politicians"?
Figure S1 shows that a majority of respondents in each condition correctly reported the profile of citizens' assembly members.However, the variation in the distribution of correct responses across the four conditions suggests variation in the complexity associated with different models of citizens' assembly.In the two conditions describing a citizens' assembly comprising exclusively citizens, a large majority were able to recognise that the members were '100 ordinary members of the public.'In the other conditions describing "a version of a citizens' assembly" (emphasis added) comprising a mixture of politicians and ordinary citizens, a significant minority of respondents failed to recognise that members would be '50 members of the public plus 50 politicians.'This is not to suggest that respondents in Groups 1 and 3 were more diligent in their engagement with the stimulus material than respondents in Groups 2 and 4; rather, it suggests that the concept of a citizens' assembly in which members are exclusively citizens is easier to process than the concept of a citizens' assembly that includes politicians.
Second, respondents were asked about the selection mechanism: how would the ordinary citizens be selected?Figure S2 shows that, across all conditions, most respondents identified that these members would be 'Randomly chosen (in the same way that juries are selected).'While this was the correct response for Groups 1 and 2, this was incorrect for Groups 3 and 4. In the latter two conditions, only a minority of respondents correctly recognised that citizens would be selected 'By an election.' Respondents appear to be more cognitively receptive to the concept of a citizens' assembly featuring randomly selected citizens as opposed to elected citizens.Thus, taken together, the pass rates for the two sets of manipulation checks suggest that people are more likely to recognise a citizens' assembly with a prototypical mode of selection, whereby its members are exclusively citizens who are randomly selected.Selection processes involving the election of citizen members and the appointment of politicians in a hybrid model of membership appear to be more challenging for respondents to apprehend in an online survey environment.
At this point, there are three possible responses to the results of the experiment.
In the first instance, it is possible to simply accept the null hypothesis in spite of the low, and uneven, proportion of respondents failing basic manipulation checks across the experimental conditions.The risk of a Type II error is, however, excessively high.
A more conservative strategy suggested by Wilson et al. (2010: 64) would be to 'analyse the data on the basis of the responses of the participants to the manipulation check.'The authors note that it is "common practice across the social sciences" to drop subjects after a post-treatment manipulation check (ibid).By excluding participants failing the manipulation checks, internal analysis of the data could test the original hypotheses with respect only to those successfully responding to the treatment.In this experiment, dropping responses of the participants who failed the manipulation checks yields a significant direct effect of the treatment.We must keep in mind the significant variation in pass rates across the conditions; in other words, the retention or dropping of cases was non-random.With two conflicting sets of conclusions, we appear to be confronted with a choice between accepting the null hypothesis (risking a Type II error) and rejecting the null hypothesis (risking a Type I error).However, a final, and more radical, strategy would be to repeat the experiment with some critical amendments to its design.Aronow et al. (2016: 8) 'stress the importance of manipulations that are sufficiently clear so as to minimise the necessity to remove subjects based on a lack of comprehension.' It is this latter strategy that was pursued in the study presented in the main article, bar three significant refinements.First, in addition to reading a text-based vignette describing a model of a citizens' assembly, on the same screen (underneath the text) respondents were presented with an infographic summarising the manipulated selection process from each condition.The purpose of the infographics was simply to summarise and reinforce the vignettes, facilitating their comprehension.Apart from the necessary variation needed to operationalise the independent variable, the four visualisations (presented in the article) were designed to be as simple and consistent as possible, minimising the risk of confounding effects.For example, the citizens and politicians depicted lack any specific attributes, and the colours used to distinguish the two types of citizens' assembly member lacked any obvious connotations.
Second, the placement of the manipulation checks was modified.Instead of completing the manipulation checks after answering a series of survey items measuring the outcome variable, the manipulation checks were placed before all other questionnaire itemsimmediately after the manipulations themselves.There is a tradeoff from this reordering.By requiring participants to complete manipulation checks before asking them for their subjective evaluations, there is a risk that their responses will be mediated by act of completing the task (Kidd, 1976).However, if the process of completing the checks serves to reinforce the manipulations, this should not be a disadvantage.
Finally, if a respondent failed to answer either of the checks correctly, he or she was invited to return to the previous page containing the stimulus.The ability to provide respondents with real-time feedback is a benefit of conducting the experiment online.
After receiving the prompt, respondents could still proceed to the questionnaire, but their failure to pass the manipulation check(s) would be recorded, facilitating post hoc exclusion.
Given the much higher pass rates of the manipulation checks in the study reported in the main article, these modifications appear to have served their intended purpose.

[
Group 1]: The citizens' assembly would consist of 100 ordinary members of the public.Members would be randomly chosen in the same way that legal juries are selected.A random sample of 100 citizens would be a cross-section of the Northern Ireland population in terms of age, gender, social class and religious and community background.[Group 2]: The citizens' assembly would consist of 50 ordinary members of the public and 50 elected politicians.The ordinary members of the public would be randomly chosen in the same way that legal juries are selected.A random sample of 50 citizens would be a cross-section of the Northern Ireland population in terms of age, gender, social class and religious and community background.The political parties would appoint the remaining 50 membersin proportion to their seats in the Northern Ireland Assembly.

Figure S1 :Figure S2 :
Figure S1: Responses to the first manipulation check: 'Who would be the members of the citizens' assembly?'

Table 1 :
Summary of experimental conditions

Table 2 :
Predictors of perceived mini-public legitimacy

Table A2
presents a one-way ANOVA with all respondents; TableA3presents a one-way ANOVA for only those respondents who passed the manipulation checks (as reported in the main study).Dropping respondents who failed the manipulation checks does not alter the substantive results.

Table A2 :
One-way ANOVA with all respondents

Table A3 :
One-way ANOVA excluding respondents who failed manipulation checks

SUPPLEMENTAL ONLINE MATERIAL Details of Initial Study An
initial online experiment was conducted in early 2018, broadly mirroring the design of the subsequent study presented in the article.332 adults living in Northern Ireland were recruited to complete an online survey between 19 th February and 19 th March 2018.They received an email invitation to participate as members of an online opt-in panel hosted by Opinium, a major UK survey company.