Informal Networked Deliberation: How Mass Deliberative Democracy Really Works

2.1 Two Models of Deliberation

Informal networked deliberation diverges from the standard model of deliberation in some important ways. The standard model conceives deliberation as an organized group activity requiring all participants to be present in the same place at the same time (Young 2002, 44–5, 126; Young 1990, 233; Goodin 2008, 3; Fishkin 1995, 33; Dryzek 2000; Fishkin and Laslett 2003, 1). Co-presence is an essential feature of the standard model of deliberation, enabling all deliberators to directly listen to and speak to each other. This is what I call the ‘synchronic-group’ model of deliberation. Informal networked deliberation, on the other hand, allows deliberation to occur in a serial fashion, among subsets of people in extended networks. Although direct and face-to-face deliberation plays an important role in community-wide informal networked deliberation, in ways that will be explained below, not everyone needs to be co-present, engaging directly face-to-face with one another in the same place at the same time to ‘deliberate together’. By enabling the flow of arguments and information across the entire extended network, informal networked deliberation can make people ‘discursively present’ to one another, as the results from your deliberative exchange with one person are carried forward into your next deliberative exchanges with others. People’s claims, arguments, and information can thereby come to be widely known across the community as a whole. In this ‘serial-diachronic’ model, deliberation is seen as a protracted process that occurs both directly and indirectly (vicariously) when people act as discursive intermediaries, passing along to their network contacts the views, perspectives, arguments, or information that they have collected from their previous interactions with others. The face-to-face direct engagement of everyone in the community with everyone else is not required for everybody to have deliberated ‘together’ in this more ‘distributed’ way.

In making discursively present people who are not direct participants in our deliberative exchange, networked deliberation allows us to influence, and be influenced by, distant others (e.g., ‘a friend of a friend’) without ever coming into direct contact with them (see Coleman, Katz, and Menzel 1966; Burt 1987, 1992, 2000; Granovetter 1973 for the importance of extended networks and weak ties). Such diffusion processes, with people routinely passing along information from one of their contacts to others in their network, are and have always been ubiquitous. We now have a wealth of empirical evidence about such exchanges, making us better able to study them. But the political role and value of informal networked deliberation has long been recognized by social and political scientists. The two-step communication model introduced by Lazarsfeld and his colleagues is an early case in point (Lazarsfeld, Berelson, and Gaudet 1944; Katz and Lazarsfeld 1995; see also Huckfeldt and Sprague 1995; Huckfeldt, Johnson, and Sprague 2004a; Druckman, Levendusky, and McLain 2018). People’s core face-to-face networks have been shown to influence non-political beliefs and behaviors as well (Christakis and Fowler 2009).^[7] And to foreshadow, even if those core networks have been shown to be more homogenous than not, as we will see in Section 3, people’s extended networks are more heterogenous and characterized by more disagreement.

2.2 Deliberative Democratic Desiderata: Inclusion, Equality and Reciprocity

Deliberative democrats posit three prime desiderata as ‘success conditions’ for their proposed innovations: inclusion, equality, and reciprocity. Their great boast is that classic mini-publics serve all of these goals well. My aim in this section is to show that community-wide informal networked deliberation can do likewise.

‘Democratic’ deliberation is ordinarily understood to require inclusive deliberation. That is understood, variously, as including all members of the community, all affected parties or their representatives, or all relevant perspectives and group interests. In the standard model, what it is ‘to be included’ in deliberation is to be physically present to the deliberation—to have a seat at the table and be capable of listening to and directly addressing all other participants (Mansbridge 1999; Phillips 1995; Young 2002). Direct inclusion has thus long been the gold standard. Such a high standard of inclusion cannot be achieved in any mass deliberative democracy, however. It is simply impossible for millions of people to be present around the same table and deliberate together.

Community-wide informal networked deliberation in the serial-diachronic mode suggests another standard of inclusion: indirect inclusion through people’s extended, overlapping networks. One is included in mass deliberation in this sense when one is connected to other people who are themselves connected to yet others. In that way, a person can be made discursively present in multitudinous serial-diachronic deliberative exchanges while participating directly in only a few of them. Ideally, we would like everyone to be indirectly connected to everyone else, such that everyone can be discursively present in everyone else’s direct deliberative exchanges. That is to say, we would ideally like everyone’s views, information and arguments to flow freely and uncorrupted across the entire community so as to reach everyone else.^[8] The structure of the community’s overall deliberative network and the way people discursively act within that network should ideally allow such opportunities.

Classic network studies are reassuring on both those scores. The classic finding was that there are at most ‘six degrees of separation’ between people: everyone is connected to everyone else through six (or fewer) network links. With the advent of networking on online platforms, that has now been reduced to 3.57 (at least among Facebook users) (Milgram 1967; Pool and Kochen 1978/9; Bhagat et al. 2016; Watts 2004; see further Tanasoca 2020, sec. 5.2.1–5.2.2). Of course, network transmission is imperfect and information gets corrupted after being passed from one person to another more than three times (Moussaïd et al. 2017, 4120). Even so, back-of-the-envelope calculations based on the above evidence (and a few additional simplifying assumptions) suggest that 70 % of Facebook users might be able to expect to transmit and to receive broadly uncorrupted information to and from fully 70 % of other Facebook users (for the calculation and the assumptions underlying it, see Tanasoca 2020, sec 5.2.3). The ideal of full democratic inclusion would of course be 100 %, but 70 % inclusion is higher than other classic indicators of democratic inclusion such as electoral turnout (IDEA 2022). And of course it is much higher than what could ever be achieved through micro-scale deliberative mini-publics.^[9]

The second normative end of democratic deliberation is reciprocity or feedback—talking and being talked to in turn, with one’s deliberative contribution eliciting a response from other deliberators (Gutmann and Thompson 1996). Deliberation is by definition a dynamic, interactive process, not merely the sum of isolated, individual discursive utterances (deliberation contrasts in that way with, for example, contributions posted on an Internet forum or message board). It relies on the capacity of individuals to prod and probe each other—on the focused back-and-forth between deliberators. No matter how democratically valuable people’s contributions might be in insolation, in the absence of any meaningful exchange people are merely talking past each other, not deliberating—and the democratic value of their deliberating together is lost. In smaller deliberating groups, it is easier to ensure this dynamic back-and-forth among participants. But what kind of discursive reciprocity could be ensured in community-wide informal networked deliberation involving millions of people?

Notice, first, that even artificially organized micro-deliberations under the supervision of a moderator have difficulty ensuring reciprocity for all participants (Gerber 2015). Some participants’ contributions end up getting more attention and evoking a stronger response from others. To some extent this is unavoidable. Because deliberation is largely a path-dependent process where people respond to, and build upon, others’ contributions, what will be said next in deliberation depends on what has previously been said. Order effects, not just the force of argument, presumably also influence the outcome of the deliberation. Thus, as a matter of structural constraint, some people’s contributions will end up ‘guiding’ and impacting the deliberation more than other people’s, eliciting thereby a greater response.

In community-wide informal networked deliberation, reciprocity is threatened not just by the large number of participants but also by the particular way in which deliberating ‘together’ happens: sequentially, across innumerable dyads or small conversational groups over a protracted period of time. Our capacity to intervene and interact dynamically in conversations to which we are only ‘discursively’, not physically present, is inevitably strictly limited by the fact that networked deliberation necessarily involves discursive intermediaries. A person’s deliberative contributions might be communicated downstream in her extended network without feedback flowing back upstream to her. After all, we often pass information from other people forward to our network contacts, but we do not always (or even all that often) communicate their responses back to our original informants.

Thus, at the mass level, reciprocity will largely be a distributed property among subsets of people within the network who are directly deliberating, engaging in the back-and-forth with one another at any given time. What matters is that all members of the community be part of some such subsets and that their discursive contributions elicit some response from their direct, immediate interlocutors—that, as opposed to their distant audience in their extended network. Insofar as everyone routinely engages in informal deliberation with their close contacts, distributed reciprocity will hold across the entire deliberative network.

Finally, democratic deliberation aims to ensure equal inclusion. As a key feature of ‘democratic’ deliberation, equal deliberative participation is painstakingly pursued in artificially-organized deliberative events through the efforts of trained facilitators who enforce special discursive norms. Despite all such efforts, literally equal participation is rarely achieved even there (Gerber 2015). Needless to say, if equality remains elusive even in those controlled deliberative settings, it will be difficult to achieve in informal networked deliberation occurring spontaneously ‘in the wild’. At the same time, however, we should not forget that there are many forms and currencies of equality.

At the micro ‘dyadic’ level, any form of dialogue—deliberation included—is a fundamentally cooperative activity (Grice 1975). A minimal form of equality is therefore intrinsic to any such discursive exchange. Deliberation requires the mutual, and broadly equal, engagement of both interlocutors. They both must be able to speak and listen in turn to one another. Any discursive interaction that is grossly unequal in this sense will be closer to a monologue or harangue than a dialogue. This minimal form of equality that is contained in any form of dialogue, and without which deliberation cannot exist at all, will be readily available within each dyadic, informal deliberative exchange.

At the more ‘macro’ level, whether everyone has an equal opportunity to engage in informal deliberation, depends on the very structure of the community’s deliberative network. In a ‘balanced’ network, where everyone is linked to (and hence able to listen to and be listened to by) an equal number of other people, everyone can be said to have equal opportunities to engage in deliberation and hence have ex ante equal influence over it. Of course, in the real world some people have more network links than others, which might on the face of it make inclusion in informal networked deliberation to appear unequal. But in assessing that we must take into account not only people’s direct contacts in their core network but also their indirect contacts in their extended network—all the other people (‘friends of friends’) to whom their direct interlocutors are connected. And since people with smaller core networks are almost invariably (e.g., on Twitter, 98 % of the time) linked to others with much (on Twitter, typically a thousand times) larger networks, people may be much more equal, taking account of both their core and extended networks than appears to be the case taking account of the former alone (Hodas, Kooti, and Lerman 2013; the more general phenomenon was first reported by Feld 1991). Deliberative equality (like reciprocity) could thus be achieved, not only writ small within each dyadic encounter, but also as a distributed property of myriad direct deliberative exchanges across the overlapping extended networks that constitute the community’s overall deliberative network.

3.1 Are People’s Social-qua-communicative Networks Opinion-Diverse?

Diversity of opinion and information matters in deliberation for two reasons. First, deliberation can have significant epistemic value only insofar as it can act as a pooling mechanism for the information and knowledge scattered across society. In pooling that information, and improving the individual competence of each citizen, deliberation can allow the community to reach more accurate judgements.^[12] This however can happen only insofar as new and different sources of (valid) information are included in deliberation. Second, for deliberation to be democratic it must include a concern for all other citizens, and particularly for those affected by the community’s collective decisions. The more people’s varying interests and opinions are represented in the deliberation, the more democratic that deliberation will be. Needless to say, echo chambers and filter bubbles can drastically reduce both the democratic and epistemic value of networked deliberation across the community.

The crucial question is however a brutely empirical one. Just how diverse are people’s (offline and online) social-qua-communicative networks? What exactly is the actual prevalence of echo chambers? Just how many people find themselves engaged just how exclusively in discussions with just how nearly identical others?

The balance of evidence we have from empirical studies and meta-analyses across the social sciences is relatively reassuring on this score. To start, however assiduously freely associating individuals seek out the comfort of identical others, their choices will inevitably be constrained by their environment and the opportunities that it offers (Huckfeldt and Sprague 1987, 120; Huckfeldt 1983; Huckfeldt and Sprague 1995; Huckfeldt, Johnson, and Sprague 2004a). Our job, school, university, neighborhood, café, gym, church, or pub we frequent all shape and constrain our social interactions (Conover, Searing, and Crewe 2002; Huckfeldt and Sprague 1995, 110, 114–15; Mondak and Mutz 2001; Mutz and Mondak 2006). And insofar as we choose those environments, those choices will be predicated, not on our political opinions, but on other non-political criteria (does this café make good coffee? what are the salary and perks of this job? which neighborhood has better schools? which gym has newer equipment?). It is far from certain (or even likely) that those with whom we bond over our profession or common hobbies will all share the same views across a wide range of political and policy matters (Huckfeldt and Sprague 1995, 17). The workplace in particular has been shown to expose people to divergent views (Mondak and Mutz 2001; Mutz and Mondak 2006).

Moreover, group homophily is usually dimension-specific, and people are likely to hold different preferences along different social dimensions (McPherson, Smith-Lovin, and Cook 2001; see also Blau and Schwartz 1984); those intersecting social circles will in turn create a heterogenous social network across all those dimensions. Our ability to associate only with like-minded people (even if we are inclined to do so) are further undermined by the facts that (a) we cannot be sure what others’ opinions are in advance, (b) we tend to misperceive others’ opinions as being closer to one’s own and (c) their positions might emerge only after we have already associated with others (Huckfeldt and Sprague 1988, 469–70; Huckfeldt and Sprague 1995, 12, 18; Levitan and Visser 2009).^[13] Importantly, even if one’s dyadic deliberative exchanges or core networks are relatively homogenous, across the entire community small levels of heterogeneity within each individual’s network aggregate to significant levels of disagreement across the community network overall. While one’s chance of being directly exposed to disagreement might be modest, the chance of indirect exposure—of being exposed to someone else who has been exposed to someone who disagrees—is substantively higher.

This seems to hold true even online. People’s ‘received tweets’ are more politically diverse and politically moderate than the ‘accounts they follow’ (showing that one can hardly isolate oneself from diverging content). Furthermore, the ‘retweets’ that one receives from the accounts that one follows are also more ideologically diverse than the tweets that are directly authored by the ‘followed accounts’ themselves (Eady et al. 2019). Retweets thus act as ‘weak ties’ (Granovetter 1973) exposing users to more diverse opinions, with the difference in ideological diversity likely to increase with each degree of separation online (retweets of retweets being more diverse than retweets and so on). Diversity of opinion is thereby fostered within people’s extended networks (see Tanasoca 2020, 161–3 for discussion).

The upshot of all of those considerations is simply this:

Disagreements over politics and policy arise even in the smallest and most closely held social groups…. few citizens are completely insulated from the interaction with others who will disagree with them. The simple fact is that disagreements occur on a regular basis, and that simple fact forces a reassessment of (1) the common wisdom suggesting that political homogeneity is the inevitable outcome of self-selection and conformity pressures within small groups, and (2) the ideals and mechanisms of political communication and influence among citizens. (Huckfeldt, Johnson, and Sprague 2004a, 98)^[14]

Some might however argue that that conventional wisdom is now outdated. The current situation, they may say, is much, much worse. Yet even newer studies confirm that citizens’ informal deliberative exchanges are still largely shaped by their environment and the degree of diversity that it harbors, rather than by their intention to seek like-minded interlocutors. A 2020 study of cross-sectional data tracking more than one hundred networks between 2008 and 2016 shows that informal deliberation is the spontaneous by-product of people’s social relationships; “political talk is more … an incidental process than a purposive one” and thus “political attitudes and identities are poor predictors of who talks politics with whom” (Minozzi et al. 2020, 136, 141.). More importantly, the 2020 American National Election Study (ANES) confirms that people’s social networks include politically differently-minded others. When asked how many of their friends or family are ‘Democrats’, 31 % of respondents responded ‘about half’ and 23.4 % responded ‘a lot’; only a small minority indicated that their networks lacked diversity by answering ‘none or almost none’ or ‘all or nearly all’. When asked how many of their friends and family are Republicans, the largest number of respondents (31.9 %) replied ‘about half’ and 20.9 % replied ‘a lot’; only a handful replied ‘all or nearly all’ (5 %) or ‘none or almost none’ (10 %) (ANES 2021a).

But are things different when it comes to social media—to people’s online networks? Survey data from the ANES also shows that few people belong to homogenous networks online. When asked how many of their Facebook friends are Democrats, the most frequent response (36.4 %) was ‘about half’. Very few respondents replied either ‘none or almost none’ (about 5 %) or ‘all or nearly all’ (6.8 %). Similar findings emerged when asking people how many of their Facebook friends are Republicans (ANES 2021a).^[15] Even in the online environment then, politically diverse networks are very much the rule and homogenous ones are very much the exception.

3.2 Do People Talk Politics with Those with Whom they Disagree?

People being socially connected with differently-minded others is only the first step, however. That, in itself, does not tell us much about people’s propensity to actually engage in cross-cutting deliberation within those social networks. We know that people discuss politics with family and friends quite often (ANES 2017, 496–7; Conover, Searing and Crewe 2002; Jacobs, Cook, and Delli Carpini 2009). But do those interactions genuinely constitute deliberation across opinion differences?

Luckily, we have several studies focusing squarely on the questions of whether disagreements occur during informal exchanges and whether people choose to deliberate about politics with those holding divergent political or policy views. Those studies have historically shown that high proportions (78 %) of Americans report moderate to high levels of ‘contested political discussion’, containing different views or involving people with different perspectives (Conover, Searing, and Crewe 2002, 39, 40 table 6). One more recent study of over-time discussion networks also clearly showed that most Americans are exposed to disagreement through their informal deliberative exchanges about politics (Minozzi et al. 2020). Shared partisanship also appeared to be the weakest predictor of whether two people will discuss politics, and friendship the strongest one (Minozzi et al. 2020, 147). Considering that more than 80 percent of Americans have friends from the other party that is certainly good news for the prospect of informal networked deliberation (Levendusky 2023, 3, 17, 25).

Contrary to what might be popular perceptions, the 2020 American National Election survey confirms that the last few years have not much affected people’s capacity to talk about politics. When asked whether it has become easier or harder to talk about politics with family, most respondents reported it had become ‘neither easier nor harder’ (ANES 2021b).^[16] In another question, the same survey asked Americans how often they have to self-censor out of fear that someone might call them a racist, sexist, or otherwise bad person. A majority of respondents replied ‘rarely’ or ‘never’ (ANES 2021b). The responses to those two sets of questions, taken together, suggest that most people are not hesitant to voice discordant opinions within their diverse networks. Finally, when the same survey asked Americans how much political differences hurt their relationships with close family and friends, the vast majority of respondents (61 and 59 % respectively) answered ‘not at all’ (ANES 2021b).

There is a widespread tendency to think that everything has changed with the advent of social media. I shall offer evidence that, even there, informal networked deliberation is actually reasonably heterogeneous. But first, let me enter two cautions against inferring too much from online behavior. First, remember that only a small minority of Americans (23 percent) use Twitter at all and an even smaller minority uses it all that much (the top 25 percent of users generate 97 percent of content) (Odabas 2022). Second, remember that people who discuss politics on social media are disproportionately partisan and ideologically extreme (e.g., Barberà and Rivero 2015; Settle 2018; Cohn and Quealy 2019; see also McGregor 2019). Those two facts, taken together, mean that we should be wary of drawing any grand inferences about the average citizen’s deliberative networks and behavior from such evidence. And the second of those facts should lead us to expect that the online deliberative networks of heavy users of social media would be far less heterogenous than those of people who use social media less or not at all.

Against that background, it is reassuring to find that studies of activity within online networks support the view that, even there, informal networked deliberation still harbors a fair bit of diversity. For example, a study of Facebook friendships published in Science has shown not only that these friendships are ideologically diverse but also that they facilitate the transmission of divergent views (Bakshy, Messing, and Adamic 2015, 1130–1).^[17] This finding was confirmed by another more in-depth analysis of the personal networks of millions of Twitter users in Spain, Germany and the United States. That study showed not just that such networks foster political diversity (albeit less than had they been randomly created) and that people became more politically moderate over time after being exposed to that diversity (Barberà 2015). Other scholars disagree, insisting that Twitter and Facebook are echo chambers on the grounds that, on polarizing topics, the shared content is more likely than not to agree with users’ political leaning. Yet even they acknowledge one important caveat, which becomes apparent comparing Twitter and Facebook to other social media platforms like Reddit or Gab where echo chambers are absent (Cinelli et al. 2021). The difference may well be explained by the powerful feed algorithms at work on Twitter and Facebook—but if so, that means that studies showing echo chambers tell us more about what people are shown on Twitter and Facebook rather than indicating an active choice on people’s part to seek (and read) certain content. On platforms like Reddit where the algorithm feed allows users to tweak it (i.e., when people have more choice over the content they are shown), the homophilic pattern disappears (Cinelli et al. 2021, 5–6; for other problems of algorithms, see Farrell and Fourcade 2023).

Yet other recent studies claim that the ‘echo chamber’ is largely just a misdescriptive metaphor insofar as social media platforms do not act as a completely closed information environments (Geiß et al. 2021). Another common measure of people’s openness to listening to opposing views is the diversity of the content of the news feeds that they choose. Recent studies show that, when it comes to news content, very few Americans (just 4 %) actually inhabit online echo chambers.^[18]

Furthermore, there is a substantial amount of overlap between the accounts followed by Twitter users at the opposite ends of the political spectrum (Eady et al. 2019). The same to be true in the UK.^[19] Studies in Canada show that, while people do discuss politics on social media, they nonetheless rely on mainstream news outlets and informal conversations with friends when seeking information about issues they deem important (Dubois 2015). In the UK, the vast majority of respondents disagree ‘sometimes’, ‘mostly’, or ‘nearly always’ with content posted by their friends on social media (Dubois and Blank 2018; see also Messing and Westwood 2014). And contrary to what is often supposed, studies in the UK and other countries (e.g., Austria, Denmark, Spain, Germany, Norway) show that the algorithmic selection offered by search engines actually increases the diversity of people’s news diets, with self-selection leading only a small minority of highly partisan people to inhabit echo chambers (Fletcher, Robertson, and Nielson 2021; for a discussion explaining the reasons for this discrepancy, see Arguedas et al. 2022). All of those studies lead communication scholars to conclude that the actual impact of echo chambers and partisan bubbles has been overstated.^[20]

3.3 Do People Take into Account Others’ Diverging Views?

Polarization, as I said at the outset, poses a three-pronged threat to community-wide informal networked deliberation. First is the threat that people will sort themselves into bubbles of like-minded people; second is the threat that even people in diverse core networks will not speak their minds in the presence of differently-minded others. Were either the case, people would not be exposed to differing opinions. But as I have now shown, neither is true to any significant extent. What about the third threat—of hearing yet failing to take into account others’ differing opinions or diverging information?

Survey and experimental studies show that cross-cutting deliberation does indeed have an effect on people’s political views, opinions, preferences, and choices.^[21] I will detail several of those findings shortly. But first let me just point out the role that those demonstrations play in the larger dialectic of this article. The question here being addressed is whether people listen to others who disagree with them. When all of the studies I shall go on to describe show ways in which people’s political views are altered in response to being exposed to networks containing differently-minded others, what those studies are showing (over and above the specific findings I go on to discuss) is that people are indeed listening to differently-minded others. Otherwise, the various effects reported below would simply not have occurred.

For a start, notice that survey and experimental studies repeatedly show that cross-cutting deliberation makes people more ambivalent, having a moderating effect on their opinions (Huckfeldt, Johnson, and Sprague 2004a, 188–90; Levitan and Visser 2008, 2009; Mutz 2006; Visser and Mirabile 2004). Cross-cutting deliberation increases people’s expertise; it enhances people’s ability to judge candidates; and it makes people more politically tolerant (Bennett, Flickinger, and Rhine 2000; Huckfeldt, Mendez, and Osborn 2004b, 91–2; Huckfeldt, Johnson, and Sprague 2004a, 188–90; Mutz and Mondak 2006; Mutz 2006). People’s political views have also clearly been shown to evolve as the views of those in their networks change (Lazer et al. 2010).

Psychologists have also shown that the composition of people’s networks influences the strength of their attitudes. When people’s networks include a wide range of views, people harbor weaker attitudes, ones that are less durable and more uncertain. This happens at least in part because people form attitudes by social comparison, assessing the accuracy of their own views by comparing them to those of others around them (Festinger 1950; Klein 1997). That makes sense from a Bayesian perspective, of course, assuming that people take what others think as evidence. Exposure to a diversity of ‘comparative information’ increases ambivalence and makes people less certain in their views.^[22]

In both naturally occurring networks and experimentally created ones, discussions with differently-minded others make people aware of discrepancies between their own attitudes and those of others. This increases ambivalence and thereby makes people more open to persuasion (Levitan and Visser 2009; Visser and Mirabile 2004). Importantly, those shifts were found only when the attitudes they were exposed to were expressed within the context of a social network; when people were exposed to the same views outside of the social (network) context, those views made no impact (Visser and Mirabile 2004, 786). This just shows that how people are exposed to divergent information—through informal networked deliberation—matters as much or even more than the sheer fact that they are exposed to such information. The social context in which diverging information is provided explains much of the impact seen in these studies. More robust studies of real-world social networks of college freshmen confirm these conclusions. They too show that, once political disagreements with network friends are discovered, people for the most part still continue engaging with those differently-minded contacts. Political disagreements are rarely able to predict the level of perceived personal closeness with that contact (Levitan and Visser 2009, 1063–4).^[23] Neither does the strength of one’s existing attitude influence the impact disagreement has on perceived closeness (Levitan and Visser 2009, 1063–4).

There is also direct evidence that diverse networks influence members’ beliefs through informational rather than purely social mechanisms. Heterogenous networks have been shown to increase reflectivity, making people to more carefully consider and assess information. For that reason, people who are surrounded by differently-minded others exhibit more attitude change in response to strong than to weak arguments (Levitan and Visser 2008). Heterogenous networks are useful not only in promoting the dissemination of information but also in influencing how people process that information (Levitan and Visser 2008, 646).

Finally, heterogenous networks have been shown to impact people’s political choices by making them less reliant on party cues (Ekstrom et al. 2020). Among those in heterogenous networks, partisanship is a weaker predictor of candidate preferences. Interactions with differently-minded others make people more likely to choose candidates on the basis of policy positions rather than party label. People in networks containing high levels of disagreement shift their candidate preferences to align with their policy preferences, regardless of their party identification. People who are in diverse networks, and are exposed to cross-cutting deliberation, form their political views and make their political choices through a more complex, effortful reasoning rather than relying on party cues (Ekstrom et al. 2020).

In short, a raft of studies with varying designs reveal changes in the strength and direction of people’s political opinion resulting from their exposure to differently-minded others in their discursive networks. The fact that those effects were found constitutes evidence that people do indeed listen to differently-minded others when interacting with them, as the evidence presented earlier suggested.

How could anyone have ever thought otherwise? One answer relates to ‘measurement issues’ within the methodology of some studies. When seeking indicators of successful deliberation, deliberative scholars typically focus exclusively on ‘belief change’. That is as understandable as it is erroneous. After all, the value of democratic deliberation is supposed to lie in changing people’s views so as to reduce disagreement, leading to less contested and more legitimate collective decisions. When studies find, as they sometimes do, that deliberation failed to change minds (or, worse, polarized opinions, increasing people’s confidence in their preexisting positions) (e.g., Gerber et al. 2014; Wojcieszak 2011), that is seen as a sign that deliberation failed—that people either did not take those exchanges seriously or they completely discounted what others were saying.

Yet even if people are Bayesians who faithfully update their beliefs in light of evidence provided by others, not every piece of evidence will necessarily lead them to completely flip their opinions (for in-depth discussion, see Tanasoca 2020, ch. 4). Beliefs come in degrees, with varying credences attached. Deliberation may decrease a person’s confidence in a certain proposition from 60 % to 55 % percent; but since the person still believes that proposition to be true (albeit with less confidence than before), that deliberative impact will pass unnoticed in studies focusing purely on directional opinion change.^[24] Even the absence of credence change does not necessarily indicate that people have not engaged in meaningful deliberation: people might end up with the same credence as before, for example, if the new evidence against the proposition perfectly balances out the new evidence in favour of it.

Deliberative influence is a better concept for capturing the fuller range of deliberative impacts than belief change conceptualized as flipped opinions. It merely requires people to consider what others say, reflecting upon it. Deliberative influence can be exercised even if people have not persuaded one another to completely change their minds.^[25] The persistence of disagreement is not proof of a failure of deliberators to listen to differently-minded others.

A second reason we might have been led to believe that people do not listen to differently-minded others might be that we have been oversold stories about ‘motivated reasoning’. Actually, that is not necessarily much of a problem (for a similar point, see Bullock 2007, 2009; see also Druckman and McGrath 2019, 114–5). First, only a small proportion of the population is even potentially prone to partisan motivated reasoning—those holding very strong partisan identities and who have strong priors (Lodge and Taber 2000; Taber and Lodge 2006).^[26] As Bertrand Russell (1928, 13) once said, “the opinions that are held with passion are always those for which no good ground exists.” It is undeniable that polarization has increased among a minority of highly partisan individuals who also hold more extreme political beliefs (Fiorina and Abrams 2008).^[27] But their polarization is largely driven by elite polarization and the increased politization of the media, not by informal networked deliberation across difference (Bolsen, Druckman and Cook 2014; Druckman, Peterson, and Slothuus 2013; Druckman, Levendusky, and McLain 2018).^[28]

Not only is there less empirical evidence of partisan motivated reasoning than might be supposed. The evidence we have (largely US-focused) is also methodologically weak (Gerber and Green 1999, 200–2; Pennycook and Rand 2021; Tappin and Gadsby 2019; Tappin, Pennycook, and Rand 2020a, 2020b).^[29] The design of those studies is in serious need of improvement, in various respects (Pennycook et al. 2021). For example, studies showing polarization and motivated reasoning assume that citizens sincerely disclose their beliefs when answering researchers’ questions. But there is no way of telling if people’s answers actually represent their beliefs about the world or are instances of ‘cheerleading’ (‘expressive responding’) or satisficing (Bullock and Lenz 2019; Krosnik 1991). Second, when people informally interact with others on an everyday basis and form beliefs over protracted periods of time—as they do in informal networked deliberation, but not in one-off, organized group deliberations—people are less prone to partisan motivated reasoning (Huckfeldt, Johnson, and Sprague 2004a). Third, partisan motivated reasoning can be countered easily through accuracy cues—that is, by making people be motivated by accuracy instead (Druckman 2012; Pennycook and Rand 2021; Zimmermann 2020).

In short, not only do we have lots of evidence (cited at the beginning of this section) that people do indeed listen to differently-minded others within their deliberative networks. We also have good reason for not believing the sort of evidence that is ordinarily offered for supposing that they do not.