Representative democracies are increasingly contested. Public trust in political parties and other political institutions is eroding (Claassen, Reference Claassen2020), fostering civic detachment and support for anti-democratic alternatives (Grzymala-Busse, Reference Grzymala-Busse2019). The populist wave of the past decades feeds into citizens’ frustrations with the status quo. Against this backdrop, practitioners across the democratic world are experimenting with participatory governance initiatives to mitigate these trends. These reforms can take various forms—e.g., minipublics, participatory budgets, citizens’ assemblies—and are meant to improve decision-making, foster political interest, decrease polarization, and heighten support for liberal democracy (Beauvais and Warren, Reference Beauvais and Warren2019; Cirone, Reference Cirone2019; Dryzek and others, Reference Dryzek, Bächtiger, Chambers, Cohen, Druckman, Felicetti, Fishkin, Farrell, Fung and Gutmann2019; Fishkin and others, Reference Fishkin, Siu, Diamond and Bradburn2021; Wuttke and Foos, Reference Wuttke and Foos2025). However, the ability of government-led participatory initiatives to strengthen citizen engagement remains an open question.
In this paper, we focus on citizens’ assemblies, temporary political forums in which randomly selected citizens deliberate on specific policy issues and develop policy recommendations (Suiter and others, Reference Suiter, Farrell and O’Malley2016; Macq and Jacquet, Reference Macq and Jacquet2023; Pilet and others, Reference Pilet, Bol, Vittori and Paulis2023). Existing evidence from deliberation studies suggests that participating in these types of initiative fosters decision-making skills (Morrell, Reference Morrell2005) and increases citizens’ perceived ability to effectively participate in politics (Boulianne, Reference Boulianne2019; Niemeyer and others, Reference Niemeyer, Veri, Dryzek and Bächtiger2024). More disaffected citizens are particularly optimistic about the prospects of citizens’ assemblies (Pilet and others, Reference Pilet, Bol, Vittori and Paulis2023). Unlike mass representative democracy, the scale of discussion in citizens’ assemblies amplifies the notion that individuals have a meaningful voice and that decision-makers are responsive (Fishkin and others, Reference Fishkin, Siu, Diamond and Bradburn2021). However, such effects are largely confined to assembly participants, as research on spillover effects to the broader population suggests these are limited (Van der Does and Jacquet, Reference Van der Does and Jacquet2023).
At the same time, there are concerns that the prospects of deliberation are overblown or can even lead to backlash (Sanders, Reference Sanders1997; Morrell, Reference Morrell2005). Deliberative forums and other types of participatory governance are likely to raise participants’ expectations about their ability to shape policy. Participatory governance initiatives are often not tied to specific budget lines, limiting policymakers’ ability to act on the recommendations produced. Unmet expectations can lead to frustration—potentially undermining institutional trust, political efficacy, and future civic engagement. Backlash is particularly likely with demanding participatory initiatives (Hanson and others, Reference Hanson, Kronick and Slough2025). Participants can also feel they are being instrumentalized when government-led initiatives give the false impression of public engagement but lead to limited policy change, a phenomenon colloquially known as “citizen washing” (Martins and others, Reference Martins, Hafen, Robert and Silk2024). Following Lisbon’s first edition of the local citizens’ assembly, for instance, one participant voiced frustration with the process: “We were completely used as a political marketing strategy. [...] Most people left disappointed.”Footnote 1
Crucially, lab or lab-in-the-field studies may miss these effects as they lack direct involvement from political actors and the potential for politicization by organizers or opposition parties. We leverage a natural experiment in Lisbon, Portugal, where the local government has organized citizens’ assemblies annually since 2022 to integrate residents in the policymaking process. Each year, participants meet over two weekends to deliberate on specific local policy issues, culminating in a list of recommendations for elected officials. The government then selects a subset of participants to monitor implementation in the following months. Participants are selected through a lottery, allowing us to isolate the causal effects of participation (Cirone, Reference Cirone2019). First, the local government invites citizens to register their interest in participating in the assembly. The initiative is disseminated through institutional channels and via mail. The city then contracts an independent firm to conduct the lottery and select 50 participants for each assembly. Citizens who registered but were not selected constitute the control group. Across two consecutive editions (2024 and 2025), a baseline survey with the treatment group and an endline survey with treatment and control allow us to isolate the effects of participating in the assembly on political attitudes and issue knowledge.
We find that participation in Lisbon’s citizens’ assemblies increased political knowledge and issue salience. Participants were more likely to identify the issues discussed in a given assembly as the most important local problems and to feel knowledgeable about these topics. The assemblies also increased external political efficacy: after 2 weeks co-developing policy solutions, participants view the local government as more responsive to citizens’ preferences and express greater trust in local political institutions. The effect on external efficacy is large in magnitude (41% of the standard deviation). Participation also does not impact individuals’ perceived ability to participate in politics (internal political efficacy). Our exploratory analysis of institutional trust yields smaller effects, concentrated on attitudes toward local government rather than toward political parties or politicians. Overall, we find no evidence of short-term backlash or perceived instrumentalization among participants. These patterns remain consistent across both editions of the assembly despite differences in the topics covered and are not explained by negative updating from individuals who registered to attend but were not selected.
Finally, we explored whether participation in the event shaped participants’ views about the policymaking process. The belief that policymakers fail to solve problems simply because they lack the necessary will to do so is a common source of frustration in contemporary democracies (Clifford and others, Reference Clifford, Flynn, Nyhan and Rhee2024). We posited that participation in public deliberation could make citizens more sensitive to the challenges of collective decision-making. Our evidence does not corroborate this prediction. Participants and non-participants retain equally unrealistic expectations about policymakers’ ability to unilaterally solve local problems. Importantly, all the effects reported are specific to the population of Lisbon residents who registered to participate in the assembly.
Given differential attrition between treatment and control, we complement the main analyses with a within-subjects design comparing the treatment group before and after the assemblies. Attrition in the treatment group was residual, making this design less subject to bias. However, it does not allow us to rule out that the effects are driven by something else taking place during the same 2-week period. The results are indistinguishable across identification strategies, suggesting that neither potential source of bias is driving the main findings. Still, we are unable to rule out potential imbalances between treatment and control on unobserved confounders in the between-subjects design.
Scholars have studied the prospects of public deliberation and other forms of participatory governance for decades. In recent years, interest in these questions have increased as political authorities across the world experiment with different forms of participatory governance. We contribute to this agenda by providing causal estimates from two government-led citizens’ assemblies in a major European capital, a context with high stakes and risks of politicization. The findings extend existing scholarship on participatory governance (e.g., Morrell, Reference Morrell2005; Pilet and others, Reference Pilet, Bol, Vittori and Paulis2023; Hobolt and others, Reference Hobolt, Lawall and Tilley2024; Wappenhans and others, Reference Wappenhans, Clemm, Hartmann and Klüver2024) and suggest that citizens’ assemblies—even when promoted by partisan governments—may help bolstering the link between citizens and political institutions. However, broader democratic engagement and a better understanding of the challenges of collective decision-making are likely to require more ambitious reforms.
1. The promise of deliberative democracy
Citizens in representative democracies are sovereigns “whose vocabulary is limited to two words: ‘Yes’ and ‘No”’ (Schattschneider, Reference Schattschneider1942, 52). Representative democracy, structured around political parties and agenda-setting institutions, limits citizens’ capacity to directly influence policy agendas. Traditional voting mechanisms aggregate diverse preferences through (often) binary choices, creating different social choice problems—cycling, intransitivity, and strategically manipulated outcomes. Proponents of deliberative democracy see it as a mechanism to overcome these challenges. Deliberative democracy enables participants to identify common ground and decompose complex multidimensional issues into structured sequences of decisions (List and others, Reference List, Luskin, Fishkin and McLean2013), reducing both the technical problems of preference aggregation, and citizens’ sense of alienation from policymaking.
However, the benefits of deliberative democracy remain disputed. First, participation in deliberative forums exhibits a pronounced bias toward citizens with higher education and civic engagement (Neblo and others, Reference Neblo, Esterling, Kennedy, Lazer and Sokhey2010). Second, considerable evidence reveals persistent gender inequalities in both floor time and influence within deliberative settings (Grönlund and others, Reference Grönlund, Setälä and Herne2010; Karpowitz and others, Reference Karpowitz, Mendelberg and Shaker2012). Political theorists have questioned whether deliberation adequately addresses the challenges of collective decision-making. Accordingly, deliberative settings may lack both legitimate authorization and meaningful accountability, creating a disconnect in which citizens outside these forums cannot assess whether participants reflect their own preferences, values, and policy aspirations (Böker, Reference Böker2017; Lafont, Reference Lafont2019). Empirical research partially supports the authors’ concerns. While citizens express positive attitudes toward deliberative democracy in principle (Pilet and others, Reference Pilet, Bedock, Talukder and Rangoni2024), they resist granting these forums greater decision-making authority (Christensen, Reference Christensen2020; Goldberg and others, Reference Goldberg, Lindell and Bächtiger2025).
These debates question whether deliberative democracy improves decision-making. We pursue a related but distinct question: does participation in deliberative forums reshape citizens’ attitudes toward policy issues and the political process?
1.1. How citizens’ assemblies can influence participants
Empirical research on deliberative democracy examines the effects of deliberation on political knowledge, efficacy, and trust. Studies show that deliberation enhances participants’ understanding of diverse preferences and perspectives, making them more tolerant of outcomes that conflict with their own positions (Esterling and others, Reference Esterling, Neblo and Lazer2011; Boulianne, Reference Boulianne2019). Deliberation may also enrich participants’ understanding of policy issues by revealing the range of viewpoints and rationales held by fellow citizens (Eveland Jr and others, Reference Eveland, Hayes, Shah and Kwak2005; Austen-Smith and Feddersen, Reference Austen-Smith and Feddersen2006).
Regarding political efficacy, the evidence is mixed. Internal efficacy, or citizens’ confidence in their own political skills and knowledge, does not appear to increase through participation in deliberative forums (Pateman, Reference Pateman1970; Morrell, Reference Morrell2005; Minozzi and others, Reference Minozzi, Kennedy, Esterling, Neblo and Jewell2024). However, deliberation tends to enhance external efficacy: the belief that one’s voice influences policymaking (Gastil and Xenos, Reference Gastil and Xenos2010; Boulianne, Reference Boulianne2018). Similarly, deliberation may increase trust in institutions, a form of diffuse support that provides officials with a ‘reservoir of favorable attitudes’ (Easton, Reference Easton1975; Grönlund and others, Reference Grönlund, Setälä and Herne2010; Minozzi and others, Reference Minozzi, Kennedy, Esterling, Neblo and Jewell2024). (Goldberg and others, Reference Goldberg, Lindell and Bächtiger2025) find that in citizens’ assemblies, trust emerges from the contact and political engagement with fellow citizens and not with politicians. The authors show that trust increases across an interpersonal dimension and an institutional dimension.
Despite significant advances in the study of deliberative forums, it remains an open question whether the same effects may occur in government-led initiatives. The specificities of a politicized and high-stakes environment may distort effects that prior experimental work has established. Government sponsorship can raise concerns about instrumentalization, whether real or perceived (Martins and others, Reference Martins, Hafen, Robert and Silk2024). It may also shape the selection of participants and, consequently, the effects of participation (Hanson and others, Reference Hanson, Kronick and Slough2025). For instance, if only high-trust or politically engaged citizens join the lottery for government-sponsored citizens’ assemblies, participation is unlikely to further increase trust or knowledge. Real-world forums also attract substantial media attention, enabling political actors to shape participants’ experiences. Opposition parties have incentives to delegitimize such initiatives, while governing parties may oversell their impact.
2. Research design
We isolate the effects of Lisbon’s citizens’ assembly by leveraging random selection among registered participants. Conselho de Cidadãos (Citizens’ Council) is a yearly forum of deliberative democracy promoted by Lisbon’s City Hall to integrate citizens into the decision-making process. We rely on data from the 2024 and 2025 editions, held on two consecutive weekends in the spring. In 2024, selected participants discussed five local welfare policy domains: housing, immigration, homelessness, health, and intergenerational solidarity. In 2025, the themes covered were mobility, security, public space, urban hygiene, and multiculturalism. Appendix A provides additional information on both editions of the citizens’ assembly studied.
Each year, Lisbon residents may register to participate in Lisbon’s citizens’ assembly, with a lottery randomly selecting final participants. Individuals can register online or physically in several places across the city. Lisbon’s City Hall promotes the initiative widely on social media, radio, and newspapers. Furthermore, the local administration sends invitation letters to a random and representative sample of the population, inviting them to register for the lottery.Footnote 2 This effort is meant to diversify the pool of residents from which participants are selected.Footnote 3
Figure 1 describes the research design across the two editions of the citizens’ assembly studied. A total of 1241 Lisbon residents registered for the lottery in 2024 and 707 in 2025. Each year, an external firm randomly selected 50 participants through a stratified lottery using quotas for gender, nationality, and parish.Footnote 4 This procedure ensures the selected group mirrors Lisbon’s demographic composition. However, the pool of participants tends to be more affluent and better educated than the average Lisbon resident. Although the lottery process aims to correct for these biases as much as possible, the selected group is not fully representative of the city’s population because the pool of citizens who register is already biased. This pattern mirrors findings documented in most citizens’ assemblies worldwide (Neblo and others, Reference Neblo, Esterling, Kennedy, Lazer and Sokhey2010; Suiter and others, Reference Suiter, Farrell and O’Malley2016; Goldberg and others, Reference Goldberg, Lindell and Bächtiger2025).
Research design, measurement, and sample sizes.
Figure 1 Long description
The process for selecting and surveying citizens in 2024 and 2025. In 2024, 1,241 citizens registered. From these, 50 were selected as participants, while 1,191 were not selected. A baseline survey was conducted with 49 citizens. The first session took place on April 6, 2024, with 50 participants, followed by a second session on April 13, 2024, with 49 participants. The endline survey included 48 treatment citizens and 485 control citizens. In 2025, 707 citizens registered, with 50 selected participants and 657 non-selected citizens. A baseline survey was conducted with 39 citizens. The first session occurred on May 24, 2025, with 39 participants and the second session on May 31, 2025, with 34 participants. The endline survey included 34 treatment citizens and 206 control citizens.
Importantly, there is differential attrition in the endline survey. Across the two editions, attrition in the control group was 62.6%, against 18.0% in the treatment group. Covariate balance tests reported in Tables C1–C3 show that treatment and control groups are balanced across sociodemographic observables.Footnote 5 However, we are not able to rule out potential imbalances on unobserved confounders. With this caveat in mind, the between-subjects design allows us to isolate the effects of participating in Lisbon’s citizens’ assemblies among residents who expressed interest in participating. We believe this estimand is substantively and practically relevant, as participatory governance initiatives inevitably rely on citizens’ willingness to take part. In the discussion section, we highlight how this feature of the design influences our ability to extrapolate from the findings.
Figure 2 illustrates different moments of the deliberative process. Over the first session, participants identified major challenges within the five policy areas on day 1, then explored potential solutions on day 2. Throughout both days, all participants were randomly assigned to five groups of ten, with each group focusing on one policy area. Group composition rotated across sessions to maximize cross-pollination of ideas and ensure all participants engaged with diverse perspectives from across the full cohort. Appendix A provides additional details about the assemblies’ structure and facilitation.
Participants deliberating in Lisbon’s citizens’ assembly.
2.1. Measures
As described in Figure 1, we surveyed participants (treatment group) each year before the first session (pooled response rate = 98.9%) and after the final session (pooled response rate = 98.8%). We administered the same endline survey to individuals who registered for the assembly but were not selected (control group) (pooled response rate = 37.4%).
We measured the effects of participation using six outcomes. Issue Salience captures whether participants identified any of the five assembly topics as the city’s most important issue (from a list of 14). Issue Knowledge is the average self-reported knowledge across the five assembly topics. As a validation check, we also measured perceived knowledge on two issues not discussed in the assemblies that serve as placebo. We measured political attitudes using 11-point Likert scales: Internal Political Efficacy (“I consider myself qualified to participate in politics”), External Political Efficacy (“The local authorities make decisions according to the will of most people”), and Green Lantern Theory (“If politicians wanted, most of the problems in the city of Lisbon could be solved easily”). This item is taken from Clifford and others (Reference Clifford, Flynn, Nyhan and Rhee2024) and inspired by work in American politics on citizens’ unrealistic expectations about the ability of presidents to unilaterally solve problems. The original authors labeled this argument “Green Lantern Theory” in reference to the superhero that can create anything with his willpower. We use it to capture participants’ sensitivity to the challenges of collective decision-making more broadly. Finally, Political Trust averages trust in three institutions: political parties, politicians, and City Hall (see Appendix B for all outcome variables).
2.2. Analysis
We isolate participation effects using two complementary designs. First, a between-subjects design compares randomly selected participants with non-participants after each assembly. These estimates come from linear models with treatment assignment as the main predictor (
$N$ = 773).Footnote 6 Second, a within-subjects design compares attitudes in the treatment group before and after each assembly. These estimates come from linear models with individual fixed-effects and a post-treatment indicator (
$N$ = 162). The within-subjects approach tells us whether participants’ attitudes shifted during the assembly period but cannot rule out confounding from concurrent events during those 2 weeks. We emphasize between-subjects estimates in our discussion because randomization eliminates omitted variable bias. However, given differential attrition in the endline survey, the within-subjects estimates serve as an important robustness check. All models include year fixed effects to account for systematic differences across editions.Footnote 7
In both designs, we estimate local average treatment effects among compliers. As described in Figure 1, there is noncompliance in the treatment group in both editions of the citizens’ assembly. While noncompliance was relatively trivial in 2024 (2%), it was more expressive in 2025 (32%).Footnote 8 We find no demographic imbalances between subjects in the treatment group who participated in the baseline survey and those who completed the endline survey after participating in all the sessions in each edition (see Tables C1–C3).
3. Results
Figure 3 reports treatment effects on political attitudes and knowledge. For comparability, all outcomes are standardized. Participants were substantially more likely to identify assembly topics as the city’s most important issues (0.39, SE = 0.11)—an effect equivalent to 39% of a standard deviation. The within-subjects estimate is somewhat smaller (0.22, SE = 0.11) but statistically consistent with the between-subjects result.
The assemblies also increased perceived knowledge about the topics discussed (0.25, SE = 0.12). Critically, this effect is specific to assembly topics: we find no corresponding increase in perceived knowledge about issues not discussed during deliberation (0.05, SE = 0.11), as shown in the final panel of Figure 3. This pattern rules out a generalized confidence boost and suggests that engagement with specific policy issues drives the knowledge gains. Together, these findings indicate that deliberation increased both participants’ familiarity with the topics covered and their perceived importance.
Effects of participating in citizens’ assembly on issue attitudes.
Figure 4 reports effects on political efficacy, attitudes toward collective decision-making, and political trust. We find no evidence that participation increases internal political efficacy—subjects’ perceived ability to participate in politics (
$-$0.16, SE = 0.12). This null result likely reflects ceiling effects: over half the control group (54.6%) selected the two highest points on the internal efficacy scale, suggesting that citizens who self-select as potential participants for the assemblies already possessed very high internal efficacy owing to their greater affluence and education. This result is consistent with prior empirical work in different settings (Morrell, Reference Morrell2005; Minozzi and others, Reference Minozzi, Kennedy, Esterling, Neblo and Jewell2024).
Effects of participating in citizens’ assembly on political attitudes.
By contrast, participation substantially increases external political efficacy (0.41, SE = 0.12). Participants expressed stronger beliefs that local authorities make decisions according to citizens’ preferences—an effect equivalent to 41% of a standard deviation. This result is substantively large and assuages concerns that government-led initiatives produce perceptions of instrumentalization or citizen washing. The positive effect cannot be attributed to disappointment in the control group following non-selection; the within-subjects estimate (0.31, SE = 0.11) is also positive and indistinguishable from the between-subjects estimate.
We also explore whether deliberation helps participants recognize the challenges of collective decision-making. A common misunderstanding in the general public is that policymakers fail to solve problems simply because they lack the necessary motivation to do so (“Green Lantern Theory”) (Clifford and others, Reference Clifford, Flynn, Nyhan and Rhee2024). This belief is pervasive in our samples. Among the control group, 74% of respondents agree with the statement that “if politicians wanted, most of the problems in the city of Lisbon could be solved easily.”Footnote 9 However, we find no evidence that participation in the citizens’ assemblies corrects inflated expectations about the policymaking process. The coefficient is small and statistically insignificant (0.08, SE = 0.12). If anything, the coefficient signal suggests that participants may be slightly more likely to agree with the statement.
Finally, we find suggestive evidence that participation improves institutional trust (0.19, SE = 0.12, 
$p$ = 0.08). This composite measure combines attitudes toward three institutions: City Hall, political parties, and politicians. When analyzed separately, the positive effect is concentrated in trust toward City Hall (see Tables D5 and D10), with negligible effects on attitudes toward parties or politicians more broadly.
4. Discussion
Across the world, governments are experimenting with participatory governance and deliberative democracy. These initiatives draw on a well-established literature documenting the benefits of deliberation. However, considerably less work examines their effects in natural, government-led contexts where backlash and politicization are more likely. We leverage the random selection of citizens who registered to participate in two editions of Lisbon’s citizens’ assembly to explore the effects of deliberation in government-led settings. Participating in the assemblies increased issue salience, perceived political knowledge, and external political efficacy.
Exploratory analysis of trust in City Hall suggests an increase among participants in the citizens’ assembly. Effect sizes are meaningful, ranging from 39% to 44% of a standard deviation, and effects reflect deliberative influence rather than frustration-driven responses among non-selected registrants. However, participation did not shift perceived ability to engage in politics (internal efficacy), beliefs about the challenges of collective decision-making, or broader institutional trust. Overall, the findings demonstrate that deliberative forums can foster more informed citizens and strengthen ties between residents and local government, even when organized by partisan authorities and subject to media politicization. However, these findings suggest that broader institutional trust and internal efficacy require complementary interventions beyond deliberative forums designed at the municipal level.
In the Lisbon’s citizens’ assembly, participants deliberate and then vote to select the three leading solutions for each of the topics covered in a given edition. A small group of participants is then chosen to monitor how the local executive implements these proposals. Although the recommendations are non-binding, they carry symbolic weight, lending legitimacy to the city’s deliberative democracy initiative—an instrument the executive has positioned as central to strengthening local democracy. To date, however, City Hall has implemented only one recommendation from the 2024 edition, suggesting that despite meaningful efforts to create spaces for citizen participation, translating those outcomes into policy remains a significant challenge.Footnote 10
Our findings extend the literature on deliberation, participatory governance, and public opinion. However, four caveats warrant attention. First, we isolate participation effects among politically engaged citizens who self-selected into the lottery. This limits generalizability to the broader population. However, participatory governance is time-intensive and, absent structural changes, these initiatives will likely continue attracting primarily civic-minded citizens. In this respect, our sample reflects the typical participant pool in assemblies worldwide. Engaged citizens also hold higher expectations about government (Hanson and others, Reference Hanson, Kronick and Slough2025), which can be frustrated when initiatives underdeliver. Whether citizens’ assemblies generate spillover, diffusion, and scaling effects from participants to the broader public remains an open question, an area warranting future investigation (Van der Does and Jacquet, Reference Van der Does and Jacquet2023). We observe no evidence of short-term backlash despite the assembly’s politicized context, though effects beyond this period remain unknown. Second, while our findings demonstrate meaningful effects among participants, the risk of disappointment remains legitimate. What has been termed ‘citizen washing’—where deliberative processes create the impression of genuine influence without translating into substantive policy change—constitutes a real concern. Our data provide no evidence of short-term backlash among this self-selected cohort of engaged citizens. However, future research should track participants’ attitudes longitudinally to determine whether initial gains persist and whether the gap between deliberative engagement and actual policy influence generates disenchantment over time, particularly as participants’ expectations confront political reality. Third, we cannot identify which assembly elements produced the observed effects or how long they persist beyond 1 week post-event. Future research could experimentally manipulate assembly features or track participants over extended periods. Finally, whether these results generalize to other participatory formats—such as participatory budgeting or issue conventions—remains an open question.
Declaration of interests
None.
Funding statement
Mariana C. Duarte was supported by the Portuguese Foundation for Science and Technology under the grant 2022.05057.PTDC.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2026.10130.
Open access
