The idea of deliberative democracy relies heavily on attitude transformation, here defined broadly as shifts or switches in agreement with reasons, discoursers or values underlying policies, as well as with specific policy options. An openness among deliberators to transform their attitudes, by being responsive to and recognizing merits in arguments put forward by others, is considered pivotal in the pursuit of defining a ‘common good’ or a ‘public will’ (Barabas 2004; Gutmann & Thompson 1996; Fung 2003; Mansbridge et al. 2010). Agreement among autonomous individuals who exercise communicative rationality, that is, engage in comprehensive, sincere and truthful exchange of views, is argued to possess legitimacy beyond that derived from majority-rule voting (Cohen 1989; Habermas 1984). In policy-oriented texts on deliberative planning, attitude transformation is implied in prescriptions of citizen dialogue as a means for mutual agreement (e.g., OECD 2020; SALAR 2019). In this literature, the non-coercive characteristics of deliberation are frequently contrasted with top-down policy implementation by planners and other experts; with hierarchical policy instruments such as command-and-control legislation; or with economic incentives and information campaigns that exercise influence in predefined directions.

Critique against the reliance on voluntary attitude transformation through communicative rationality, and thereby explicitly or implicitly against the ideal and practices of deliberative democracy, can be found in both democracy theory and behavioral science. Difference democrats such as Mouffe (2005) hold that political conflicts cannot be overcome through rational consensus as idealized by Habermas. Instead, they call for agonistic debate and a continuous openness to perspectives that challenge hegemonic norms and discourses. In planning theory, a stream of research warns that careless application of deliberative ideas may create democratic facades that legitimize decisions taken elsewhere, or provides already strong societal groups with arenas to exercise illegitimate power (e.g., Huxley 2000; Hillier 2003; Tewdwr-Jones & Allmendinger 1998;). Adding to these critical voices, there are ample empirical studies demonstrating how people who engage in political dialogue do not necessarily conform to communicative rationality, but instead are subject to various biases that cast shadows of doubt upon the practical feasibility of deliberation (e.g., Giljam & Jodal 2003; Rienstra & Hook 2006; Zajonc 1999).

To scrutinize whether mechanisms influencing deliberators’ attitudes in real-life settings are compatible with assumptions underlying deliberative democracy is arguably a central task for political science and planning scholars. It is a large and complex task with many entry points, some of which involve deep-going philosophical questions about rationality; autonomy and formation of will (see Rostbøll 2005). The present paper will restrict its focus to a seemingly neglected aspect of this issue, namely deliberators’ awareness of whether their own attitudes transform through dialogue.

This inquiry rests on the hypothesis that the level of deliberators’ awareness signals whether criteria of communicative rationality are fulfilled, and hence can serve as an indicator of deliberative quality. Indeed, the significance of awareness can be directly derived from central ideas in deliberative theory, including the Habermasian requirement for sincerity and authenticity of validity claims – a requirement that arguably presupposes that persons making these claims are mindful of what they believe and desire. Accordingly, Riensta and Hook (2006, p. 313) stipulate that ‘no validity claims are redeemable between communicative participants if the agent cannot access, substantiate or understand their own rationality’. The requirement for awareness can also be linked to, for example, Elster (1983), who asserts that rational formation of preferences must be intentional and autonomous; to Dryzek (2002, p. 162), who notes that ‘authenticity of deliberation requires that communication must induce reflection upon preferences in a non-coercive fashion’; and to Christman (2009), who states that the desires of an agent are authentic if s/he can reflect upon the historical process that caused them without feeling alienated from it.1

Nevertheless, deliberators’ awareness of attitude transformation has received little direct attention in literature. While the study of deliberative groups has seen significant methodological progress over recent decades, the issue is not accounted for in approaches to identify, measure and evaluate attitude transformation – the Q-methodology that focuses on how changes in discourses can be derived from changes in deliberators’ attitudes towards more particular propositions (Dryzek 1990; Dryzek 2005a); the deliberative quality index that focuses on fulfilment of procedural criteria drawing on Habermasian discourse ethics (Steenbergen et al. 2003); and the deliberative reasoning index that focuses on intersubjective consistency among deliberators regarding preferences for policies and considerations underlying these policies (Niemeyer & Dryzek 2007). Further, while recent handbooks on deliberative democracy (Bächtiger et al. 2018; Levy et al. 2018) comprehensively deal with prospects for conscious deliberation, they do not address the more specific issue of recognizing and interpreting deliberators’ self-awareness of attitude transformation. That said, even though few authors explicitly discuss the issue, both theoretical and empirical studies illuminate factors that help explain whether participants in political dialogue become aware of any attitude transformation, and, if so, whether they are willing to acknowledge this to others. The present paper is an attempt to combine findings of these studies into a heuristic model that can guide further inquiry into the issue.

In this heuristic, as outlined in the coming sections, I will conceptualize learning as a fundamental mechanism mediating any impacts from deliberation on attitudes. These potential impacts include formation of new attitudes as well as consolidation, loosening or transformation of preexisting attitudes. Drawing on a notion of learning as meaning making based on both conscious and subconscious clues, I will assert that deliberators are unlikely to be fully aware of these impacts, or be able to attribute them to specific deliberative occasions. In literature on deliberative democracy, I discern three overall factors of relevance for the level of such awareness – one that should heighten it and two that may lower it. The awareness-supporting factor is the conscious reasoning on arguments laid out as overt and justified persuasion in accordance with communicative rationality. The hampering factors are, first, various subconscious biases inherent in social processes in general and political dialogue in particular, as described in behavioral research; and second, the potential calculated manipulation of participants that political dialogue lends itself to, as warned for by critical scholars. The extent to which these overall factors come into play during deliberation may in turn be partly controlled, or at least influenced, by design choices of those in charge of setting up dialogues.

Beyond the paper’s theoretical focus, for illustrative purposes, I also present empirical results from a minor explorative study of two Swedish citizen dialogues (in total 24 respondents) that prompted the conceptual investigation. While the samples are too small for firm inference, the observations indicate that level of discordance between revealed and stated switches in attitudes may be a viable vantage point for exploring the concepts of the heuristic model in real-life dialogue settings. More specifically, differences between the two dialogues tentatively corroborate earlier suggestions that attitude transformation is less likely in ‘hot’ dialogues around well-defined contested political issues than in dialogues characterized by broader topics and less friction among participants, but also that the awareness of any transformation might be higher in the former case.

Learning is arguably a defining feature of deliberation, and consequently ubiquitous, if not always explicit, in deliberative democracy research. This learning not only encompasses sharing or jointly creating knowledge on the subject matter, but also improved understanding of views and values of fellow deliberators (e.g., Mansbridge et al. 2012), as well as, at a meta level, the practice of deliberation itself – that is, what Owen & Graham (2015) call ‘the deliberative stance’. Still, elaborate discussions on the more specific dynamic between the interlinked processes of learning and attitude transformation are not common in deliberative democracy literature. According to Wright (2022), theory often assumes that deliberation is about debating and justifying standpoints formulated prior to deliberation, as opposed to new ideas emerging along the way. At the same time, several authors highlight how people’s stands on political issues are often ill-formed, incomplete or conflicting (Bartels 2003; Cuppen 2012; Feldman & Zaller 1992; Lavine, Johnston & Steenbergen 2012). The alleviation of such ambiguity, as well as construction of identities through formation of preferences, is seen as an important learning potential in deliberation (Fishkin 1995; Karpowitz & Mendelberg 2018; Prior & Lupia 2008; Rostbøll 2005). Using Wright’s (2022) terminology, if this potential is realized, impacts on attitudes can be attributed to ‘co-creation’ of views, rather than to persuasion based on the ‘force of the better argument’ (Habermas 1996). Connected to this reasoning, correlations presented by Luskin et al. (2002) indicate that deliberators who learn more in terms of acquisition of facts also tend to change their attitudes more, while Zhang (2019) shows that uninformed opinions are more susceptible to influence through deliberation than well-founded opinions.

Just as in everyday language and in policy pertaining to deliberation (e.g., OECD 2020), the term learning is typically surrounded by positive connotations in deliberative democracy research, signaling a better-informed basis for attitudes and their amendment. For example, in the context of deliberative polling, Luskin et al. (2002; p. 474) assert that ‘The rationale of Deliberative Polling requires that such [policy preference] changes be not merely random bouncing around, nor the outgrowth of some purely social dynamic, but rather the product of learning and reflection, of coming to see consequences for valued goals more clearly and weigh them more carefully […]’. Some deliberative democracy papers, however, acknowledge that learning per se does not necessarily impact attitudes in accordance with communicative rationality, for example, as it may derive from skewed information (Strassheim 2020). Smets and Isernia (2014) show that in relation to some issues, participants calibrate new information conveyed through deliberation with ideological predispositions, thereby enforcing their preexisting views, while in other cases, learning may instead moderate and nuance attitudes. Westwood (2015) argues that links between knowledge acquisition in deliberation and opinion changes are generally weak, making the case that persuasion through justified arguments is instead the main driver of attitude transformation.

In education theory, the question how classroom activities designed as deliberation intentionally or unintentionally may shape political attitudes has seen an increased interest in recent decades. Transactional and other learning theories that draw on pragmatism, including Dewey’s works on linkages between democracy and education (e.g., Dewey 1916), as well as on Habermasian discourse ethics, conceptualize education that encourages a plurality of perspectives as the key mechanism for learning (e.g., Englund 2000; Gutmann 1999). Acknowledging an entanglement of facts and values in such classroom deliberation, research on these theories harbors a debate on how teachers should handle the normativity that is inherent, but not necessarily overt, in education on political issues, such as sustainable development (e.g., Hellquist & Westin 2019; Jickling & Wals 2008).

Connected to the more specific issue of awareness of attitude transformation, studies of cognitive aspects of learning demonstrate how it stems from both conscious and subconscious processes involving perceptions, language, memories, emotions and motivations (e.g., Kuldas et al. 2013), and is therefore ‘explicit’ as well as ‘implicit’ (Seger 1994). In a strict sense, it can thus be asserted that it is impossible for learners to fully deduct and account for effects of a specific deliberative occasion. This assertion is further strengthened upon a closer look at the psychological origin of attitudes towards political issues, which Eagly and Chaiken (1993) describe as an interplay between beliefs, desires, actions and preexisting attitudes towards other, related issues. Learning prompted by deliberation may affect one or several of these ‘attitude components’, but this may not be directly noticeable, let alone manifested, as shifts or formation of fully developed attitudes. Instead, it can be assumed that impacts are often indirect, latent and non-linear, and therefore may be difficult to attribute to a particular deliberative occasion for both deliberators and researchers (Chambers 1996; Mackie 2006). As an example of this, Mugny and Perez (1991) describe how a dialogue on abortion did not change the position of participants opposing decriminalization, but made them more favorable towards contraceptives, even though the latter issue was not explicitly mentioned during discussions.

In sum, the reviewed literature supports a conceptualization of learning as a fundamental mechanism that mediates any impacts from deliberation on attitudes (cf. Bohner et al. 2002). These potential impacts include attitude transformation, that is, shifts or switches in position on political issues at hand, but also formation of attitudes that were previously non-existent or obscured; as well as affirmation and consolidation of preexisting attitudes, or, conversely, loosening of preexisting attitudes through increased ambiguity. Being a broad concept, learning does not necessarily confirm with communicative rationality, since both cognitive input and internal sense making can be distorted. Moreover, as learning stems from both conscious and subconscious clues, it can realistically be assumed that rather than fully aware, deliberators will be cognizant of impacts on attitudes to a lesser or greater degree, depending, for example, on to what extent they engage in ‘internal deliberation’ following dialogue with others (Goodin 2000). In the following two sections, I review literature suggesting that this level of awareness depends on to what extent overt persuasion in accordance with communicative rationality is distorted by, first, unintentional socio-psychological biases, and, second, calculated manipulation.

As touched upon in the introduction, behavioral research has revealed a plethora of phenomena that may influence participants in political debate independent of rational reasoning on the subject matter. These include, but are not limited to, ‘group polarization’ (the tendency of like-minded deliberative groups to become even more homogeneous and move in the direction of initial predispositions due to social dynamics; Schkade, Sunstein & Hastie 2010; Sunstein 1999); ‘adoption by attribution’ (valuation of arguments based on relationship with and perceived status of the source; Tooming 2021); ‘motivated reasoning’ (valuation of arguments based on a strive to maintain group identities; Kunda 1990; Wright 2022); ‘social intuitionism’ (founding of moral judgements on intuition, with rational motivation lacking or being construed post hoc to justify the intuitive judgement; Haidt 2001; Lodge & Taber 2013); and ‘adaptive preferences’ (contingency of preferences upon limited options available; Elster 1983; Knight & Johnson 1997). Underlying these phenomena are well-known more general psychological biases, such as confirmation bias, in-group bias, observational-selection bias, framing effects and rationalization.

Results from these studies are not necessarily coherent, and therefore difficult to synthesize into general conclusions, as they depend on contexts of investigated cases. Still, the divergence in scholarly assessments of their implications on overall prospects for deliberation is striking. For example, in their influential contribution to the ‘realist’ stream of critique against deliberative democracy, Achen and Bartels (2017) assert that hopes to achieve communicative rationality in real life settings are utopian and naïve in light of biases, unstable political preferences and generally low levels of knowledge among citizens. Along similar lines, Riensta and Hook (2006, p. 314) propose that ‘[…] Habermas’s construction of communicative rationality rests upon an agent role that might only be filled in reality by a self-reflexive critical genius. Deliberative agents are assumed to be heroic in terms of informational breadth and calculative abilities, and heroic in their ability to identify, segregate and set-aside self-interest. This agent might be an individual of Habermasian proportions and Habermasian abilities, but they are no agent of modern actuality.’

From the other side of this debate, advocates of deliberative practice have responded, first, that communicative rationality represents an ideal to be approximated and not an assumption about typical human behavior (e.g., Chambers 2018; Sharon 2019). Second, some scholars highlight the potential of deliberation to instill social norms that ‘launder’ participants’ preferences and prevent them from expressing biases related to, for example, prejudice or selfishness, thereby fulfilling a ‘second-order preference’ for high quality dialogue (Goodin 1986). Third, pertaining to the more specific phenomena listed above, proponents of deliberation can invoke studies suggesting that these biases can be abated, or at least managed at acceptable levels, through careful design of deliberative interaction. For example, Imbert et al. (2020) show how misrepresentation of deliberators’ true attitudes, which may stem from adoption by attribution or motivated reasoning, can be counteracted by ensuring sufficient opportunities for expression of minority attitudes. Luskin et al. (2022) show that deliberative polls are resilient to polarization and homogenization, while Himmelroos and Christiansen (2014) and Strandberg et al. (2019) demonstrate how rules enforcing ‘deliberative norms’ in political dialogue may curb polarization also in like-minded groups. Khader (2018) suggests that cross-cultural deliberation can enable criticism of adaptive preferences and revelation of unrecognized alternative paths to well-being, while McHugh et al. (2017) propose that social intuitionism expressed as ‘moral dumbfounding’ can be countered by demanding that deliberators provide justifications for their judgements. Dryzek (2005b) discusses how the level of contestation may affect deliberators’ openness to revise their attitudes. He asserts that this is harder in ‘hot’ dialogue settings, that is, when stakes are high, discussions are directly linked to decision making, and partisans are involved. He therefore proposes that deliberation should take place in ‘colder’ contexts at a certain distance from formal decisions. Fung (2003) tentatively takes the opposite view, suggesting that hot dialogues may be more transformative as they are engaging and creative, although he also acknowledges a lack of comparative empirical studies of attitude transformation in hot and cold dialogues. In their critical account of polarization through deliberation, Schkade, Sunstein and Hastie (2010) instead argue that expectations to arrive at decisions might amplify herding biases, if deliberators feel pressed not to obstruct the process – an example of the paradoxical, subtly coercive, ‘chilling effect’ that may arise from excessive emphasis on consensus (Dahlgren 2009; Morrissey & Boswell 2020).

In what ways, then, can attendance to deliberators’ awareness of attitude transformation contribute to an advancement of this debate? Operating at a subconscious level per definition, to the extent biases come into play, they will tend to obscure this awareness, and, conversely, the level of awareness will reflect prevalence of biases. Putting aside the harshest critique of deliberative democracy, the implication of which is that a vast majority of initiatives will be futile or outright harmful, and instead stipulating that distortion from biases varies depending on contextual factors, eliciting awareness of attitude transformation can help assessing whether levels of such distortion are acceptable or unacceptable. In this regard, it should prove a viable complement, if not error proof, to other indicators of deliberative quality. More specifically, if observed attitude transformation in an individual deliberator seems prompted by group dynamics such as herding or polarization, while the deliberator is reluctant or unable to acknowledge the transformation, this signals deviation from communicative rationality. However, before conclusions are drawn, methodological limitations of what realistically can be elicited should be considered, as discussed below, as well as the possibility of intentional manipulation, as discussed in the next section.

Several of the biases listed above prompt deliberators’ subconscious tendencies to either align with or maintain a distance to views of other individuals or groups, disregarding rational reasoning on the subject matter. In other cases, based on the same motives, the biases may instead obstruct attitude transformation that would have taken place if founded solely on rational deliberation. Discordance between acknowledged and observed attitude transformation will be visible to the extent that biases in play during deliberation can be lessened or eliminated in the setting where deliberators are asked directly whether their attitudes have changed (such as in a research interview or in a questionnaire). Admittedly, this prerequisite will not always be possible to meet or control. For instance, illustrating what Mackie (2006) calls ‘the unchanging mind hypothesis’, that is, the frequent, but not necessarily valid, notion that deliberation rarely alters anyone’s standpoint, people might be reluctant to admit changes in views on high-stake political issues not only to fellow deliberators and researchers, but also, at the conscious level, to themselves. A main reason for this phenomenon suggested by Mackie (ibid.) is the strive for consistency, which is directed both inwards, as a means to preserve self-image, and outwards, as consistency is closely related to credibility and integrity, which are highly valued in political debate. To assess such cases, more sophisticated methods than the ones sketched here are needed. In other cases, it is conceivable that deliberators will acknowledge that their attitudes have transformed even though this was sparked by subconscious bias. Here, assessment would rely on presence, quality and timing of justifications of these transformations, in order to detect, for example, post hoc rationalization.

One additional methodological caveat, demonstrated by Goodin and Niemeyer (2003) and Niemeyer (2020), warrants mentioning. In one of very few empirical studies dealing explicitly with alignment between observed and stated attitude transformation, the authors investigated a citizens’ jury process concerning a road through a wilderness area. Goodin and Niemeyer (2003) found that observed shifts in 12 jurors’ agreement with propositions on various aspects of the road did not align with their stated attitude change.2 Revisiting the data, Niemeyer (2020) instead compared the jurors’ stated change with revealed preferences regarding i) five policy alternatives for the future of the road, including an option to close it; and ii) only the choice between closing and keeping it. While stated attitude change again did not align with change in preferences for the five options, it did so with switches between the binary close/keep options. The explanation put forward by Niemeyer is that when stating attitude transformation, deliberators may restrict their focus to a single policy choice considered of paramount importance compared to other topics addressed in dialogue.3 More study is needed to determine if such a pattern reflects unawareness of less significant transformation, or if it rather represents a response heuristic.

Deliberative initiatives that enable communicative rationality can be construed as the very opposite of manipulation (Fishkin 2010). At the same time, deliberation is susceptible to illegitimate influence through distortion of participants’ attitudes. For example, populists may take calculated advantage of the biases discussed above, by inciting in-group bias vis-a-vis perceived outgroup elites, or by strengthening support for one-sided arguments through confirmation bias, thereby exacerbating polarization. Sharon (2019) shows that concern over such corruption of attitudes has since long been present also among proponents of deliberation, and he asserts that the risk of manipulation constitutes the single most serious challenge to deliberative democracy, as it threatens its core function of transforming independent judgment of deliberators into collective self-government through reasoning.

The concept of manipulation encompasses several related phenomena, but it is not easy to pinpoint denominators that are common to all (Noggle 2022). Manipulation can be levied directly at individuals or executed indirectly through modification of their physical or institutional environments. It can be defined as action along the spectrum between overt rational and sincere persuasion and coercion, and as such, it often aims at bypassing deliberative reasoning among target groups. This can be achieved by appealing to non-conscious emotions or hiding relevant information. Manipulation may also be executed through open argumentation, then instead being characterized by intentions that are hidden from its targets, or by deceiving messages that trick targets into beliefs that the manipulator knows are flawed. Further adding to the complexity is the question whether manipulation is always morally wrong, or if there are cases where it may be legitimate, such as if outcomes favor its targets.

In deliberation, participants may be subject to manipulation from outside actors, from other participants, or through calculated choices by those designing and facilitating dialogue. Providing examples of the last possibility, Ellul (1962/2021) describes dialogue practices in Mao’s China that functioned to influence participants in predefined ways. As opposed to top-down propaganda, these practices relied on horizontal mechanisms of establishing and reinforcing certain attitudes among participants, supervised by discussion leaders who directed conversation towards topics and conclusions that were acceptable from the state ideology standpoint. He and Warren (2011) assert that such ‘authoritative deliberation’ is still prevalent in Chinese governance, the main purpose of which is to legitimize already taken decisions or provide information that increases top-down policy implementation capacity without empowering participants to have an independent say. Corresponding phenomena are by no means absent in western contexts, as discussed in several studies (e.g., Boussaguet 2016; Franzén, Hertting & Törn 2016; Polletta 2014; Swyngedouw 2018; Walker et al. 2015).

Responding to the concerns about manipulation, some scholars have suggested that deliberation should be restricted to ‘constitutional’ aspects of governance, that is, those pertaining to fundamental principles underlying other legislation and executive and administrative power, as they may be less prone to corruption and more engaging to the average citizen (Ackerman 1991; see further in Sharon 2019). Other proponents of deliberation instead trust the ability of deliberative practice to withstand manipulation, or correct calculated distortion of attitudes, and they can invoke several empirical studies in support of this view (Dryzek et al. 2019). For example, Niemeyer (2004; 2011) shows how mini-publics may realign attitudes distorted by symbolic politics with deliberators’ true preferences for policy outcomes, while Fishkin (2018) demonstrates that support for populist policy proposals may decline following deliberation. In terms of deliberative designs that may help curbing manipulation, Niemeyer and Jennstål (2018) suggest that the focus of deliberative mini-publics should not be to recommend or decide on specific policies, but to explicate and scrutinize reasons underlying relevant policy options, thereby exposing attempted manipulation. They further emphasize the safeguarding of a ‘deliberative stance’ as a key antidote to manipulation among participants in dialogue, much along the same lines as described in relation to unintentional biases in the previous section.

Central to this deliberative stance is the rejection of ‘strategic rationality’, that is, deception, threats of sanction, or other tactics that treat fellow deliberators as means to other ends than reasoned agreement (Dawood 2013; Habermas 1984). At the same time, the line between communicative rationality and strategic rationality is not always clear-cut in practice. Blau (2019) convincingly argues that certain features attributed to strategic rationality in theory on deliberation are not always true, including fixed ends and self-interest. Conversely, communicative rationality may involve self-interest as well as certain strategic means-to-end considerations, for example, regarding which linguistic means best convey validity claims. Instead, Blau asserts that the two key features distinguishing strategic and communicative rationality is that the latter requires, first, sincerity of validity claims; and, second, that any agreement with these validity claims is voluntary and autonomous. Of relevance to this assertion is the scholarly debate on to what extent rhetoric should be allowed in political dialogue (Chambers 2009). Difference democrats, such as Young (1997), have argued that deliberative demands on rational communication are technocratic, dull and excluding. They suggest that that a broader communicative repertoire, including humor, word play, figures of speech and storytelling, may increase engagement, reflection, and involvement of stakeholders with different styles of expression, including groups not used to arguing as prescribed by theory. This view has been adopted, at least in part, also by some proponents of deliberation, including Dryzek (2002), who even mentions gossip (p. 1) as an acceptable form of deliberative communication, although he also warns that rhetoric may coerce participants by manipulating their emotions. Going one step further, Casullo (2020) conceptualizes deliberation as a public performance where bodily communication should be more acknowledged. Thereby, she arguably stretches current definitions, while also highlighting the potential for populist modes of persuasion to influence deliberators.

Now, returning to the inquiry of the present study, in what ways is deliberators’ awareness of attitude transformation relevant for concerns about manipulation? As evident from the reviewed literature, manipulation of deliberators can be either hidden or overt. In the latter case, if deliberators are faced with explicit but false claims regarding outcomes of policy choices under discussion, or if they are openly threatened with sanctions if not aligning their views to those of the manipulator, they will in most cases be aware of these influences. Then, potential manipulation can be assessed based on, for example, discrepancy between preferred and likely outcomes of their policy choices, in line with the methods applied by Niemeyer (2004; 2011), or, more generally, critical assessment of how deliberators motivate their transformation, if at all. The latter approach should be viable also if deliberators acknowledge that their attitudes have changed, but are unable to identify the underlying reasons, as the means of influence have been hidden by a manipulator. If deliberators do not at all acknowledge their own attitude transformations as observed or revealed by a researcher, and these transformations correspond to intentional choices by those designing dialogue regarding, for example, the framing of the topic or dominant views among recruited participants, this may indicate hidden manipulation. Lastly, attention to deliberators’ awareness of attitude transformation could be useful also at the micro-level, when addressing the thorny question of acceptable rhetoric in deliberation. Manipulative elements disguised as rhetoric might be indicated by such things as difficulties among participants to provide rationales for attitude transformations that appear to stem from intuitive or affective reactions to arguments (cf. Manin 1987), or by transformations that appear guided by distorted interpretations of intentions and commitments underlying arguments, to the extent this can be reliable assessed by the researcher.

Figure 1 attempts to give an overview of key concepts and mechanisms outlined in the literature reviewed above. Naturally, it represents a considerable simplification in relation to actual intricate interplays between attitudes and conscious and subconscious influencing factors during deliberation. As denoted by the dashed lines, factors influencing attitudes associated with higher and lower levels of awareness, respectively, are interlinked. For example, overt persuasion can induce conscious learning through reasoning, whereas both subconscious learning and hidden manipulation might spark biases that are also subconscious. In addition to serving as a vantage point for further inquiry into awareness of attitude transformation, the heuristic symbolizes that choices of those designing and facilitating deliberation affect how the factors come into play.

To introduce a few tangible, if tentative, ideas on how deliberative designs can influence awareness of attitude transformation, and on how this awareness can inform assessment of deliberative quality, the remainder of this paper is devoted to a brief account of two minor Swedish municipal citizen dialogues. Although not its main focus, the study of these dialogues elicited attitude transformation as changes in agreement with propositions regarding the topics, as well participants’ self-assessment of their attitude transformation. The literature review above was prompted by a curious difference in the alignment of these revealed and stated attitude transformations between the two dialogues.

The small sample sizes, especially in Svenljunga, call for caution when interpreting attitude transformation in individual respondents. Still, the overall patterns point to a few avenues worth exploring in future studies. First, from a methodological perspective, that stated attitude transformation aligns with revealed switches but not with revealed shifts among respondents in Svenljunga suggests that this distinction carries meaning. It might complement Niemeyer’s (2020) proposal that respondents focus on prominent policy choices when asked to self-assess attitude transformation; it is possible that they also primarily think of transformations from agreement to disagreement or vice versa. At the same time, in Fagersta alignment was lacking also for switches, suggesting other confounding factors there.

One possibility is that the difference in alignment between the municipalities is linked to lower initial knowledge and less firm opinions on the broader issue of integration, compared to on the more specific issue of rural schools. If so, the higher rate of revealed attitude switches, the lower rate of stated attitude transformation, and the poor alignment between these two parameters in Fagersta could be attributed to less certain and/or conscious á priori views, obscuring subjective experiences of attitude transformation and perhaps substituting them for notions of learning and increased clarity. However, the ‘I don’t know’ option available for each proposition should be able to control for this, at least partly. Instead, the lower initial share of ‘I don’t know’ answers in Fagersta, and the fact that this share did not decrease, does not fit this explanation.

Another possibility relates to the distinction between hot and cold deliberation discussed above. Formally, both dialogues were advisory, with final decisions resting with the municipal assemblies. At the same time, observations suggest that stakes were perceived as higher in Svenljunga, as evident both in the debate during the dialogue and in reports on ongoing protests in affected villages. Further, in Svenljunga, politicians with diverging views were present, while a corresponding partisan tension was lacking in Fagersta. The lower rate of switches and persistent attitude heterogeneity in Svenljunga can possibly be ascribed to a more contested and well-defined policy choice. Further, the observations hint at a possibility that in hot dialogues, pressure on participants from opposing arguments is obvious enough to make attitude switches conscious, compensating for or overriding mechanisms that obscure attitude transformation or discourage participants to admit it to themselves or others. Conversely, in colder dialogues, this pressure may be less obvious and attitude transformation less conscious.

Lastly, the framing of the dialogue topics can be considered. In Fagersta, the direction of the most frequent attitude switches indicate that at group level, participants became more positive towards prospects of successful integration built on a shared responsibility between immigrants and natives. This notion arguable adheres to the call for more inclusion in the dialogue name. Further, the ambition to identify ‘measures to promote integration and inclusion’ implied an instrumental rather than ideological orientation, without pointing to any particular tension, as opposed to the ambition in Svenljunga, which conveyed an anticipation that the conflict around rural schools would be activated. More specifically, the ambition in Fagersta left limited room for one prominent standpoint in the broader political debate on immigration in Sweden – namely the far right wing claim that the main problem is not qualities of integration measures but rather the volume and/or culture of immigrants received. As recognized by civil servants and politicians following the dialogue, this was a likely reason why the Swedish Democrats declined to participate. Further, it can be speculated that it made the few participants carrying anti-immigration attitudes reluctant to voice those, while also, perhaps partly subliminally, contributing to a more positive outlook on prospects for integration among others. The fact that subgroups were instructed to reach consensus on concrete measures might have contributed to homogenization of views, in line with Dahlgren’s (2009) suggestion above (see also Schkade, Sunstein & Hastie’s 2010), while also subconsciously influencing values underlying these measures, corresponding to the observations in Mugny and Perez (1991).

It should be emphasized that given the effort put into identifying diverse perspectives prior to the dialogue in Fagersta, there are no indications of intentions among civil servants and politicians to subdue tensions. Rather, the apparent incoherence between the normative framing and the ambitious preparations is somewhat puzzling. At the same time, also a dialogue with practice-oriented and normative framing related to integration in a Swedish context should be able to accommodate certain conflict, and it is likely that factors other than those discussed here also influenced how discussions unfolded. Among these, the conscious decision by participants to not include the ‘winners and losers’ perspective in the topics to be addressed was probably significant. When asked about this omission, involved civil servants suggested conflict avoidance as an explanation – a phenomenon since long identified as an obstacle to deliberation (e.g., Himmelroos et al. 2017; Ulbig & Funk 1999).

To reconnect with the purpose of the paper, the reviewed literature as well as the investigated citizen dialogues demonstrate the complexity of influences on deliberators’ attitudes in real-life settings. These influences are challenging to discern for a researcher, and no doubt also for the deliberators themselves. In other words, it is a tall order for them to be aware not only if their attitudes transform during deliberation, but also why and how, as prescribed in theory. While citizen attitudes are subject to manipulation and sub-conscious biases in all realms of society, this is arguably particularly disturbing in contexts where the ideal mechanism for political decisions, if unattainable in its purest form, is autonomous and rational reasoning based on overt persuasion.

At the same time, there is credible evidence in support of the argument that possibilities to curb manipulation and biases are better in deliberation than in many other contexts. When and to what extent these distortions invalidate the legitimacy attributed to communicative rationality thus appears as an empirical question, and increased scrutiny of deliberators’ awareness of attitude transformation may provide parts of the answer. Therefore, it should be added as a component in research evaluating deliberative quality with more sophisticated methods than that applied here. More specifically, while existing empirical studies on different biases and manipulation in deliberation typically deal with single distorting factors in isolation, the literature as summarized in the heuristic model presented here suggests that they are interlinked. Approaches looking into their combined impacts would thus be valuable, albeit no doubt challenging, as they could inform tricky choices that practitioners who design deliberation are faced with, including the delimitation of discourses represented (Dryzek & Niemeyer 2008) and the appropriate level of tension between these (Esterling, Fung & Lee 2015).