The persistence of cross-cutting discussion in a politicized public sphere.

The persistence of cross-cutting discussion in a politicized public sphere - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Proc Natl Acad Sci U S A . 2025 Oct 17;122(42):e2516942122. doi: 10.1073/pnas.2516942122 Search in PMC Search in PubMed View in NLM Catalog Add to search The persistence of cross-cutting discussion in a politicized public sphere Diana C Mutz Diana C Mutz a Samuel A. Stouffer Professor of Political Science and Communication, University of Pennsylvania, Philadelphia, PA 19104 Find articles by Diana C Mutz a, 1 Author information Article notes Copyright and License information a Samuel A. Stouffer Professor of Political Science and Communication, University of Pennsylvania, Philadelphia, PA 19104 1 Email: [email protected] . Contributed by Diana C. Mutz; received June 30, 2025; accepted August 26, 2025; reviewed by Jaime Settle and Paul Sniderman This contribution is part of the special series of Inaugural Articles by members of the National Academy of Sciences elected in 2021. Series information Inaugural Article Received 2025 Jun 30; Accepted 2025 Aug 26; Issue date 2025 Oct 21. Copyright © 2025 the Author(s). Published by PNAS. This article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND) . PMC Copyright notice PMCID: PMC12557757  PMID: 41105707 Significance Changes in political discussion networks over the last quarter century convey important information about the state of the American public sphere. Previous historical comparisons have not been possible due to methodological variations in research on political networks. This study uses two virtually identical pre-election surveys, one from 1996 and the more recent from 2020. Evidence is consistent with the idea of an inherent tension between the goals of deliberative and participatory democracy. Americans’ increasingly large political networks include a greater number of like-minded discussants. This change coincides with increased political participation, declining levels of political tolerance, and less extensive understanding of the rationales behind others’ views relative to one’s own. Keywords: political networks, political tolerance, deliberation Abstract Although studies of Americans’ general discussant networks have been repeated over time, research that assesses change in the nature of Americans’ political discussion networks has yet to be conducted in nationally representative probability surveys. In this study I answer two questions about the quality of the American public sphere that have generated widespread speculation, but little evidence to date. First, how have Americans’ political discussion networks changed over the past 25 y? Second, are the consequences of these changes what one would expect based on previous theory linking Americans’ interpersonal information environments to political tolerance and political participation? I resolve competing claims suggesting that people feel less free to discuss politics, with claims suggesting instead that political discussion now permeates everyday life to a greater extent than in the past. Findings suggest widespread increases in political discussion, changes driven almost entirely by increases in like-minded political discussion partners. Surprisingly, Americans are no more or less likely to engage in conversations across lines of political difference. The predicted consequences of these fluctuations confirm an intrinsic tension between characteristics valued in democratic citizens. Political tolerance has declined significantly, along with decreased awareness of rationales for others’ relative to one’s own views. Political participation is significantly higher on average than 25 y ago. Few people reported engaging with online political discussants, despite efforts to make sure they were included in network measures. Over the last quarter century, much has changed in American politics. Some of these changes—such as rising animosity between partisans of the two major political parties—have been well documented ( 1 ). Yet we know little about how the informal, everyday public sphere has changed for ordinary American citizens. The concept of a public sphere—where citizens interact informally and talk knowledgeably about political controversies—gained popularity in the United States with the translation into English of Jurgen Habermas’s Structural Transformation of the Public Sphere in 1989 ( 2 ). Habermas’s historical account was about the rise of a public sphere in Europe and, most notably, its subsequent decline. Nonetheless, among American social scientists this concept prompted many efforts to quantify the quality of the American public sphere. The criteria for a vibrant public sphere are debated, but they generally include citizens who are actively and enthusiastically engaged in political discussion, deliberating about the controversies of the day with those of opposing and like-minded views. The purpose of a public sphere is to enrich individual opinions, mutual toleration, and political culture more generally by virtue of the diversity of thoughts and opinions exchanged. Most view this concept as an ideal type of citizenry, not as a reality in the United States or Europe at any specific point in history ( 3 ). One could never satisfy, let alone empirically study, all of the additional requirements that have been suggested for an ideal public sphere ( 4 ). Nonetheless, these ideas inspired many empirically oriented scholars to focus on the informal political conversations of everyday people in democratic countries. The 1990s may have been the peak of interest in what ordinary Americans might accomplish simply by talking informally about politics and public affairs. This enthusiasm inspired my own interest in Americans’ discussion networks and their consequences, resulting in a book entitled Hearing the Other Side: Deliberative versus Participatory Democracy ( 5 ). The raw material for this book came from political network surveys executed shortly before the 1996 presidential election, combined with experimental evidence. In the present study, I address two major questions. First, a quarter century later, what can be said about how Americans’ political discussion networks have changed? Are they fundamentally different in ways that have implications for the balance of political tolerance and participation? Second, have the earlier predictions about the correlates and consequences of people’s information environments been borne out? Any theory based on evidence from a single point in time is on relatively weak ground. But if the consequences suggested by this theory are consistent with the changes that have occurred over the past quarter century, then the theory gains credibility. As of 1996, I found little cross-cutting political conversation in most Americans’ political networks. Given well-established preferences for homophily, this was not unexpected. But the extent to which like-minded discussants dominated people’s networks was especially sobering. Based on theories of deliberative democracy, citizens need to talk about politics with those of opposing views in order to derive benefits such as greater political tolerance, as well as increased awareness of legitimate rationales behind opposing political perspectives. Cross-cutting discussion facilitates the kind of mutual understanding and tolerance that are important to maintaining a pluralistic democracy. In contrast, the best information environments for maximizing political participation are those dominated by like-minded political discussion. When surrounded by like-minded others, people’s views are continually reinforced and they are spurred on to take political action. Like-minded networks make political action more attractive by reducing the potential costs associated with political action, such as the uncertainty triggered by exposure to countervailing arguments, or the potential disruption of social relationships due to political disagreement. In this sense, politically heterogeneous and homogeneous information environments can be said to facilitate different, though both important democratic goals. What was noteworthy in the 1996 study was that characteristics held up as democratic ideals—enthusiastic participation and tolerance of those with opposing views—were intrinsically at odds with one another. The type of information environment that created ideal conditions for one, unintentionally undermined the other goal. Based on these findings, I concluded that there was an inherent tension or trade-off between deliberative and participative democracy. Deliberation requires sober consideration of oppositional views in mixed political company; participation requires the enthusiasm and confidence facilitated by like-minded contact. This tension is built into a political system in which all political participation is voluntary, and thus requires internal motivation ( 6 ). In the 1996 study, individual information environments were characterized using a summary measure that captured the balance of like-minded versus oppositional political viewpoints in a person’s political discussion network. Doing so ensured that people could be located on a continuum from having predominantly like-minded versus oppositional political discussant networks, independent of the size of their network. While this simplified the analyses, it also discarded potentially important information. It is possible to have few or many like-minded or oppositional political discussants within one’s network, discussants who are politically middle-of-the-road, or any combination thereof. These are network characteristics that a single summary measure cannot capture, although I did not disentangle these possibilities at the time. For purposes of historical comparison, in 2020 I executed the exact same study of Americans’ political discussion networks, using the same questions, in the same order, the same number of weeks in advance of a presidential election as in the 1996 study. In the 1990s, the term “social network” was used to refer to the people with whom one conversed, regardless of how this took place—by telephone, in person, etc. Today the term is often used to refer to social media or online networks in particular. In this study, I retain the broad original definition, making it clear to current survey respondents that their political discussants can be people they know online or off (see Methods and Materials for details). Both the 1996 and 2020 studies were random probability samples of the US public. However, because the best way to reach a random probability sample had changed in the interim, the probability sample in 2020 was gathered using both online and telephone interviews, whereas the 1996 sample was strictly sampled and interviewed by telephone. In 2020, those who could be interviewed via either mode were randomly assigned to respond either online or by telephone to enable examination of potential mode of interview effects ( SI Appendix, Appendix B ). Theoretical Expectations. Since 1996, the American public has increased substantially in what is known as “affective polarization,” that is, the extent to which partisans on both sides hold negative feelings toward the other party ( 1 ). Strong antipathy toward the “other side” is widely believed to suppress Americans’ willingness to engage in political discussion ( 7 ). In addition, today’s citizens may find it easier to make assumptions about others’ political leanings through demographics, lifestyle choices, or purchasing behavior, thus facilitating avoidance of dissent ( 8 ). On the other hand, affective polarization is also indicative of more intense political preferences, something that would logically correspond to a more politically engaged populace ( 9 ). It is thus unclear whether one would expect political discussion networks to be smaller or larger as of 2020, although most scholars have predicted smaller, more homogeneous political networks due to high levels of affective polarization ( 7 , 10 ). Without a source of historical comparison, however, this question remains unanswered. Larger networks are generally thought to increase the likelihood of cross-cutting political discussion for two reasons. First, assuming each additional political discussant has an equal probability of holding opposing views, then having more discussants increases the probability of having at least some cross-cutting discussion in the network ( 11 ). Second, when people talk to larger numbers of others, it becomes increasingly likely that at least some of them are “weak ties,” that is, people with whom one is not particularly close, and thus who are more likely to hold differing political views. This study utilizes a summary indicator of political network composition from 1996 and 2020, while also separately examining the extent of like-minded, oppositional, and neutral discussion in both years. By doing so, I shed light on which characteristics of political discussion networks matter most to outcomes of interest in American politics. Further, I examine whether the kinds of information environments that facilitate these outcomes remain in tension with one another as found in the 1996 study. The goal of this research is first and foremost to document historical change in the composition of Americans’ networks over almost a quarter century. In addition, this study advances understanding of how the political composition of people’s networks relates to highly valued democratic outcomes. Do the consequences of changes in network composition produce the outcomes anticipated by Hearing the Other Side ? If Americans are exposed to more like-minded discussion, this should drive higher levels of political participation, but less tolerance and awareness of oppositional perspectives. If they are exposed to more oppositional discussants, then tolerance and understanding of oppositional views should increase, and political participation should decline over time. Study Design. Given variations in the operational measures used to characterize people’s political discussion networks in the past, over time comparisons have seldom been on strong footing ( 12 ). Moreover, network measures are known to be highly sensitive to context and wording ( 13 ). In the 1996 study, I created an index that combined the extent to which people’s networks included like-minded, oppositional, and politically neutral discussants. Doing so simplified analyses by allowing one variable to summarize the extent to which people’s networks were dominated by like-minded versus cross-cutting discussion. In the present study I use this same basic approach to create a summary measure, which I refer to as the Index of Cross-cutting Discussion . This index ranges from those experiencing overwhelmingly cross-cutting (+3) versus like-minded (−3) discussion. * Each discussant’s political views were characterized using a four-item index that relied not only on party and candidate preference similarities and differences between the main respondent and the discussant, but also on more nuanced assessments of how much they agree or disagree on issues or have similar or differing political views ( SI Appendix, Appendix A ). In addition, I analyze these three measures separately—the Number of Like-minded , Oppositional, and Neutral Discussants . Doing so makes it possible to assess whether it is the entirety of a person’s information environment that matters most to outcomes of interest, or whether these components have differing implications for tolerance, awareness of oppositional, and like-minded viewpoints, as well as political participation. Additional variables included in these analyses are described in the Methods and Materials section. They include the Frequency of Political Discussion within each dyad, Awareness of Rationales for Like-minded and Oppositional Views , Political Participation , Voting, and Political Tolerance . Results The most striking change in Americans’ political discussant networks is their growth in size. In 1996, Americans reported an average of 1.66 political discussants in their political networks; as of 2020, that same number had increased by 22 percent, to 2.01 (t = 7.10, P < 0.001) ( 13 ). Although these numbers are not necessarily what one might obtain using other political discussant generators, they were asked consistently over time and within the same proximity to an election, when political discussion tends to increase. Due to changes in survey data collection best practices, it was important to compare the 1996 data not only to the 2020 data as a whole, but also to the respondents randomly assigned to take the survey by telephone in 2020, as was originally done in 1996. As shown in Fig. 1 , the growth in network size is even larger when comparing 1996 telephone to 2020 telephone interviews, with 2.20 political discussants per respondent in 2020, an increase of 33 percent since 1996 (t = 9.94, P < 0.001). This increase in discussion is also evident in the decreased percentage of political isolates, that is, people who could not name even one individual with whom they talked about politics. This percentage was 24% in 1996, and only 13% in 2020. Finally, the average Frequency of Discussion among those with one or more discussants also increased from 1996 to 2020 (t = 2.36, P < 0.05). Fig. 1. Open in a new tab Change over time in average size of American political discussant networks, 1996 to 2020. Note: All comparisons shown above indicate statistically significant increases over time ( P < 0.001). Politics has clearly become a more frequent topic of conversation for Americans over the last 25 y. The greater salience of politics in 2020 and the larger networks that result from it should, in theory, be a boon to the public sphere in multiple ways. If people are talking about politics with more people at a greater frequency than in 1996, their networks should include more loosely connected discussants who may embrace views unlike their own, thus encouraging cross-cutting political discussion. However, as shown in Fig. 2 , increased network size as of 2020 has not produced these benefits. Regardless of how one operationalizes the political complexion of Americans’ political discussant networks, the number of oppositional discussion partners has not increased significantly. The sheer number of individual discussants who are like-minded has increased by far the most. The number of neutral discussants declined significantly, and oppositional discussants were largely unchanged. Fig. 2. Open in a new tab Change over time in composition of political discussion networks. Note: All changes over time are statistically significant increases or decreases with the exception of the Number of Oppositional Discussants, which represents at most a marginally significant increase ( P < 0.10). Why did increasing network size fail to benefit the extent of cross-cutting discussion in Americans’ political networks? Fig. 3 illustrates the composition of people’s networks by the total size of their networks. Two differences are immediately evident. First, the size of the green, like-minded component of people’s networks has increased from 1996 to 2020 regardless of network size. Second, the width of the gray band representing neutral political discussion has declined, again, regardless of network size, as seems predictable in an environment where politics is highly salient. The red band, illustrating the extent of oppositional discussion, increases slightly, but not significantly, from single discussant to three discussant networks. Fig. 3. Open in a new tab Relationship between size of political network and Number of Like-minded, Oppositional, and Neutral Discussants. Note: Figures represent the average number of discussants of each type within networks of each size. Discussants are categorized based on details in SI Appendix, Appendix A . What this figure also demonstrates is that the proportion of oppositional discussants declines as networks become larger; larger networks disproportionately increase like-minded discussion. This pattern suggests that as politics becomes more salient, and political discussion increases, like-minded discussants increasingly dominate people’s networks, and oppositional discussion becomes a less substantial proportion of the network. For these reasons, as illustrated in Fig. 2 , the Index of Cross-cutting Discussion declined in 2020, despite larger political discussant networks and more frequent political conversations. Given rising polarization in the US, most observers expected that Americans were engaging with fewer oppositional political discussants. They have never been very likely to engage across lines of political difference, and that has apparently not changed. This overall pattern is consistent with the idea that cross-cutting discussion tends to be largely incidental rather than chosen. Even though Americans may feel more uncomfortable talking to those of opposing views in the 2020s, a certain amount of oppositional exposure is unavoidable. Instead of a decline in cross-cutting contact, the potential problem is that people have increased tremendously in the extent to which they talk about politics with like-minded others. Discussant networks are not zero-sum; when they increase in size, various components may increase at different rates. If what matters is what dominates the overall composition of people’s networks, then this bodes poorly for political tolerance since the proportion of networks dominated by like-minded discussion has increased, as shown in Fig. 2 . If what matters is the sheer volume of oppositional discussion, then it is no worse than a quarter century ago. Changes in the demographics of networks are not as substantial as one might expect given major demographic changes in the US. In 1996, I noted that high status groups such as males, those with high incomes and higher educational levels encountered fewer cross-cutting discussants. Status of many varieties appeared to insulate dominant groups from cross-cutting contact, giving people greater control over the composition of their networks, as well as over their physical environments. As of 2020, the only significant change to this general pattern is among women. As with other historically subordinate groups, women in 1996 had significantly greater cross-cutting than men. But as of 2020, this relationship had completely reversed, with women now reporting systematically less cross-cutting discussion than men. SI Appendix , Table S1 (cols. 2 and 6) indicates that this is due to much greater increases in the number of like-minded discussants in women’s as opposed to men’s networks. Whereas Republicans had a lower Index of Cross-cutting Discussion in 1996, Democrats have decreased since then at an even greater rate than Republicans ( SI Appendix , Table S1 ). This pattern emerges because Democrats have increased significantly more in their like-minded discussion as of 2020. Both Democrats and Republicans are less likely to have neutral political discussants in their networks than they were in 1996. This pattern could be a result of the more intensely partisan political environment in 2020, or it could result from perceiving political significance in actions that were not previously deemed indicative of a political position. If in 2020 one cannot drink coffee, eat chicken, or drive a car without somehow signaling their politics ( 14 ), then Americans may perceive fewer of their political discussants to be politically neutral, regardless of the reality. Consequences of Network Change. The 1996 study implied an intrinsic tension between various democratic goals. It suggested that declining levels of the Index of Cross-cutting Discussion would portend a decline in political tolerance, as well as fewer rationales for oppositional views relative to rationales for people’s own side of various political controversies. Given the decline in the Index of Cross-Cutting Discussion , have those predicted changes occurred? As shown on the far right of Fig. 4 , Political Tolerance has, indeed, declined significantly since 1996, by about nine percent of its 1996 level. This change represents a relatively small, but statistically significant decline (t = 3.27, P < 0.01). As predicted, Americans are now less willing to uphold the civil liberties of disliked political groups than they were in 1996. This pattern is consistent with the expectation that information environments with a lower Index of Cross-cutting Discussion facilitate less tolerance. Fig. 4. Open in a new tab Change over time in levels of political tolerance, and awareness of rationales for own and oppositional views, 1996 to 2020. Note: Error bars represent 95% CI. Political Tolerance is measured on a four-point scale shown on the right y -axis. Awareness of Rationales ( Left axis) is calculated as counts of the actual number of legitimate rationales mentioned by respondents that were later independently coded. See Methods and Materials . More importantly, using the three decomposed indicators representing the Number of Like-minded, Oppositional, and Neutral Discussants reveals new insights into why the Index of Cross-cutting Discussion is closely tied to levels of political tolerance. As expected, the Index of Cross-cutting Discussion positively predicts tolerance, and it does so equally well in both years. But as shown in SI Appendix , Table S2 , when decomposed, it is not the Number of Oppositional Discussants that predicts higher levels of Political Tolerance, so much as the Number of Like-Minded Discussants that negatively predicts tolerance. When oppositional discussants are swamped by like-minded discussants in an individual’s network, this is a net negative for political tolerance, even if the Number of Oppositional Discussants is the same. Notably, this relationship could be causal in either direction; that is, intolerant people could be more motivated to surround themselves with like-minded others, or those surrounded by like-minded others may become more intolerant since the “correct” positions seem completely obvious within their homogeneous networks. Either way, information environments that are dominated by like-minded discussants go hand in hand with political intolerance. My previous emphasis on oppositional discussion appears to have been misplaced. Tolerance has declined without any decline in oppositional discussion because it is the overall composition of the network that appears to matter, with burgeoning like-minded discussion just as important in determining that mix. Is Awareness of Rationales for Own and Opposing Views similarly related to network composition as it was in 1996? Open-ended measures were independently coded for the number of legitimate rationales that respondents offered in response to these questions. Fig. 5 plots the means over time using the combined index from three different issues areas. As shown in the Top panel of Fig. 5 , Awareness of Rationales for Own Views increased a great deal from 1996 to 2020 (t = 13.31, P < 0.001). Awareness of Oppositional Views increased marginally at most (t = 1.73, P < 0.10). People naturally have more legitimate rationales for the views they actually endorse. Consistent with the expanding number of Like-minded Political Discussants, the increase has been substantially steeper for Awareness of Rationales for Own Views. Fig. 5. Open in a new tab Awareness of rationales for own and opposing views, 1996 to 2020. Note: Error bars indicate 95% CI. In the Upper panel, Rationales for Own Issue Positions increased significantly (t = 13.64, P < 0.001), but Awareness of Rationales for Oppositional Views increased only marginally (t = 1.73, P < 0.10). In the Lower panel, Awareness of Rationales for one’s own preferred candidate increased significantly (t = 7.21, P < 0.001), while Awareness of Rationales for the Opposing Candidate declined (t = 4.73, P < 0.001). Although the patterns shown in the Top and Bottom panels of Fig. 5 are similar, since awareness of rationales for positions on issues may vary based on how salient those issues are at the time, even asking about the most salient pre-election issues in both election years does not guarantee equal salience and comparability. For this reason, in the Lower panel of Fig. 5 , I illustrate results with more strictly comparable indicators of Awareness of Rationales for supporting one’s own and the opposing major presidential candidate. Awareness of Rationales for one’s preferred presidential candidate increases just as one would expect from the increase in like-minded discussants in people’s networks (t = 7.21, P < 0.001). But in addition, there is a slight, but still significant, decrease in Awareness of Rationales for the Opposing candidate (t = 4.73, P < 0.001), a change that would not be predicted by largely unchanged levels of discussion with those of opposing views. It is plausible in a highly polarized environment that people are less willing to admit awareness of rationales for the other side, although we still see a more theoretically consistent flat pattern when considering the index as a whole. Perhaps most important for the mutual understanding of difference in a pluralistic democracy, the difference between the number of Rationales for Own and Oppositional viewpoints has more than doubled since 1996. With declining understanding of how anyone could support the other side, and increasing confidence that one’s own candidate is superior, it is no wonder that people are frequently asking themselves, “How can anyone possibly support Candidate X? And how can they oppose policy Y?” I next used the Index of Cross-cutting Discussion as an independent variable in a pooled multivariate regression predicting the Dominance of Rationales for Own over Others’ Views, that is, the size of the gap between the number of rationales for own and opposing positions. Higher scores represent greater Dominance of Rationales for Own views over the Opposition’s. Full results are shown in SI Appendix , Table S3 . As illustrated in Fig. 6 , the more cross-cutting a person’s network is, the smaller the difference between the number of rationales for their own and others’ positions. In addition, the strength of the relationship between the Index of Cross-cutting Discussion and Dominance of Awareness of Rationales for one’s Own versus Oppositional Positions has become significantly greater as of 2020. The significant interaction between Year and cross-cutting discussion in SI Appendix , Table S3 confirms that as of 2020, the composition of people’s networks became even more closely tied to how much they see the “other side” as holding positions with relatively fewer tenable rationales supporting them. For those at the high end of the Index of Cross-cutting Discussion in Fig. 6 , the predicted Dominance of Own over Others’ Views is roughly the same as it was in 1996. But for those on the low end of this index, that is, those with considerably more like-minded networks, the difference between rationales for their own side and the other side has become far greater than it was in 1996. When there appear to be many reasons for supporting one’s own preference, and far fewer reasons for the other sides’ preferences, then the political opposition seems especially benighted. Fig. 6. Open in a new tab Estimated Relationship between index of cross-cutting discussion and dominance of rationales for own over oppositional views, 1996 and 2020. Note: Means represent the extent to which respondents have more Rationales for Own than Opposing Positions (Own-Oppositional). Results above are based on a pooled regression equation with interactions of all independent variables by year ( SI Appendix , Table S3 ). The significant interaction between Year and Index of Cross-cutting Discussion indicated that the negative relationship between the Index of Cross-cutting Discussion and perceiving fewer legitimate rationales for opposing versus own views was even stronger in 2020 than in 1996. Political participation is the final outcome of interest relevant to testing the theory of an intrinsic democratic tension between deliberative and participatory democracy. Theoretically, a declining Index of Cross-cutting Discussion and rising numbers of like-minded political discussants should bode well for levels of political participation. I use two measures to examine this prediction, a dichotomous measure of Voting, and an Index of up to six participatory activities that go beyond voting. In Fig. 7 , I illustrate the results of a multivariate logit predicting self-reported voting in respective presidential elections. The extent to which the Index of Cross-cutting Discussion predicts voting does not vary by year. But in both 1996 and 2020, a more cross-cutting discussant network predicts a lower likelihood of voting, even after controlling for network size and a host of other covariates ( SI Appendix , Table S4 ). Fig. 7. Open in a new tab Predicted probability of voting by index of cross-cutting discussion ( Left ) and by Number of Like-minded Discussants ( Right ), by year. Note: Results are based on the logit equations predicting self-reported intent to vote in the upcoming presidential election as shown in SI Appendix , Table S4 . Results for the Index of Cross-cutting Discussion were no different by year, nor were the relationship with Numbers of Like-minded, Oppositional, or Neutral Discussants. As shown on the Left , the Index of Cross-cutting Discussion significantly predicted a lower likelihood of voting; on the Right , the Number of Like-minded Discussants predicted a greater likelihood of voting across both years. As shown on the Right panel of Fig. 7 , when breaking down network characteristics into their like-minded, oppositional, and neutral components, it again appears to be the Number of Like-minded Discussants that matters most to increasing the likelihood of voting ( SI Appendix , Table S4 ), even after controlling for network size, which is known to correspond to increased likelihood of participation. The Number of Oppositional Discussants is negative but not significantly predictive, but the Number of Like-minded Discussants does, in addition to the combined index summarizing the dominance of like-minded over oppositional discussants represented in the network. To further confirm that network composition predicts political participation, I examined which discussant network characteristics best predicted other forms of Political Participation in 2020. As shown in SI Appendix , Table S5 , the Index of Cross-cutting Discussion negatively predicts political activities beyond voting, as it did in 1996. But when broken down into its three components, it is once again the Number of Like-minded Discussants that corresponds to a higher likelihood of participation rather than cross-cutting discussion that suppresses it. Discussion Since these measures are only available for two specific points in time, how confident can one be that this is truly a shift that has occurred over time, and not one or more outliers that are unrepresentative of longer-term change? People’s networks tend to evolve very slowly. Changes are prompted by moving to a new location, a new job, or some other life-changing event such as having children. But given that mobility in the United States has been declining steadily since the turn of the century, † it is less likely that these aggregate-level data points tap unusual points in time. The exception is obviously the impact of COVID19 on Americans’ interaction, a highly unusual period of time with ramifications for human-to-human interactions. Notably, COVID’s impact was to restrict people’s interactions with others rather than to encourage more of them, thus working against the finding of larger political discussant networks in 2020. Moreover, if one restricts these networks to discussants who are not family members or relatives—the kind of discussants more likely to be avoided during COVID—we find the same pattern: the size of political discussant networks increased significantly, and the Index of Cross-cutting Discussion became significantly more negative from 1996 to 2020 (t = 4.53, P < 0.001). Past studies of networks have been criticized for failing to include people one may meet or converse with online ( 10 ). In this study, the 2020 political discussant generator made explicit that respondents’ political discussants could come from online by repeatedly mentioning this in the name generator. Nonetheless, only two percent of political discussants in 2020 came from social media or other online sources. This finding is consistent with previous research suggesting that a small percentage of social media users generate most of the online political discussion, ‡ and that a relatively small percentage of Americans actually engage in political discussion online ( 15 – 18 ), although a great many may view others’ content. The unforeseen low level of participation in online political discussion cautions against the increasingly prevalent tendency to generalize findings from online (and especially social media) interactions to political networks more generally ( 19 , 20 ); conflating online political discussion with all political discussion can lead to a mischaracterization of Americans’ everyday political discussions. Finally, it is worth noting that in this study, unlike in 1996, I have not included experiments to confirm specific causal directions between the characteristics of people’s information environments and their political behavior. I have no immediate reason to believe the same relationships no longer hold. But more importantly, causality is not central to the implications of these findings. To the extent that a highly participative democratic culture does not easily coexist with one in which people are tolerant of political difference, then it is essential to consider the ideal balance between the two. Conclusion The American public sphere has changed in ways that defy easy categorization. Relative to a quarter century ago, Americans talk more frequently with a larger number of political discussants. This augurs well for an enthusiastic and politically active populace. And indeed, other sources confirm that rates of voting and some forms of participation have increased. According to the US Elections project, in 1996 turnout was 52 percent, compared to over 66 percent in 2020. § The American National Election Studies likewise shows increases since 1996 in various campaign activities such as the proportion of Americans who have tried to persuade someone to change their vote, or who have given money to a campaign. ¶ However, the ubiquitousness of politics in Americans’ everyday lives has gone hand in hand with substantial increases in like-minded political discussion, increases that swamp the relatively small and stable Number of Oppositional Discussants. By using an index that combined the extent of various kinds of discussion, I fundamentally overlooked the importance of like-minded political discussion in the 1996 study. Cross-cutting discussion was the central focus of that project, and thus many scholars, myself included, assumed that the patterns I observed were being driven by the extent to which people engaged with political discussants of opposing views. More like-minded discussion does not necessitate less oppositional discussion. With the benefit of historical variation, and decomposition of both old and new surveys into measures of the extent of different kinds of discussion, what appears to matter is the balance of like-minded and cross-cutting discussion in a person’s environment. For example, just as women have increased in like-minded discussion over the past quarter century, they have also voted in greater proportions. But their tolerance of oppositional views has declined, just as the theory would suggest. External sources validating the overall decline in political tolerance are more limited than for participation, perhaps reflecting the greater emphasis placed on participation as a sign of democratic health. Observers may also implicitly assume “that the pursuit of one component of the democratic ideal can never involve the endangering of another,” ( 21 ) an assumption that these results call into question. Other studies evaluating political tolerance rely on General Social Survey data ( 22 ), but scholars have questioned these results since they primarily assess tolerance of left-leaning groups. Americans currently cite right-leaning groups as the most threatening, and thus most difficult to tolerate ( 23 ). Notably, many efforts by foundations and civic organizations to increase cross-cutting political discussion in the mass public have been based on the implied premise that more cross-cutting contact in the mass public could stem the tide of rising polarization and violations of democratic norms. Holding all else constant, increasing cross-cutting discussion could improve the overall mix of like-minded versus cross-cutting discussion. But all else is unlikely to be held constant. Americans dislike engaging in difficult cross-cutting political discussions ( 24 ), and all but a relatively small proportion say they want to avoid it. Nonetheless, cross-cutting discussants are no less common now than they were 25 y ago, suggesting that most oppositional political discussion is inadvertent. As a result, it would be wrong to blame the mass public for our current predicament. In fact, people’s networks might more easily be improved by reducing the extent to which people engage in like-minded political discussions. Given that the extent of oppositional discussion does not depend much on network size, reducing the salience of politics in everyday life could be beneficial in altering the balance between like-minded and oppositional discussion. That balance does, indeed, appear to be linked to more polarized sentiments, although the causal direction between the balance in network composition and affective polarization cannot be evinced by these data. Larger political networks do not offer a solution to the problems of contemporary American political culture. In fact, the greater salience of politics in people’s everyday lives is likely to be part of the problem. Although political scientists commonly use voter turnout as a measure of the health of democracy, focusing on participation may divert attention from other important ends. If larger networks contribute to even greater dominance of like-minded over oppositional political discussion, then maximizing political involvement may come at a cost that most Americans would prefer not to pay. Since all forms of political participation are voluntary in the US, motivations such as anger and enmity can be helpful in motivating interest and participation, but at the expense of a more tolerant society. Mandatory voting laws might ease some of this tension ( 6 ). Regardless, intractable division will not sustain a pluralistic democracy. Given that very few people today see American democracy as doing well, the fatigued American citizenry might welcome shorter campaign seasons, and less emphasis on performing politics on a day-to-day basis. Scholars once pined for more impassioned partisans ( 25 ), but the current saturation of people’s everyday lives with politics suggests it may be time to consider other models of what an ideal American citizen should look like. As Talisse suggests, “Democracy is overdone when it is enacted in ways that crowd out other social goods that are necessary for democracy to thrive” ( 26 ). The intrinsic tension between the need for participation and tolerance of oppositional views may produce conditions inhospitable to a pluralistic democracy. In other words, more democracy may not be the cure for what currently ails American democracy. Citizens are already engaging in more political dialogue with others, but doing so has only rendered their environments more dominated by like-minded views. As most observers might have expected, political intolerance has increased in the mass public over the past 25 y. But these findings suggest that an increasingly politically isolated and balkanized mass public is unlikely to be the cause of increases in intolerance. Everyday Americans are currently no more isolated from those with opposing political views than they ever were. Just as Samuel Stouffer and others argued long ago ( 27 , 28 ), it may be political elites who served as the “carriers of the creed;” that is, elites enforced democratic norms even when the bulk of the mass public did not support them, and indeed never has. When elites cease to uphold democratic norms, one can expect democratic backsliding with the mass public following their lead. To put it bluntly, the current crisis of democracy in the US is not due to citizens who are no longer fulfilling their civic obligations and thus need to be coaxed or incentivized into better communication with out-partisans. Support for democratic norms has waned because political elites have failed to do their jobs in upholding these important democratic principles. Methods and Materials Data. Two nationally representative probability surveys were fielded 24 y apart. A quarter century was chosen as the rough time frame, but in order to hold constant the proximity of the network interviews to a presidential election (a hypothesized cause of greater political discussion), a multiple of four was necessary. Unlike the GSS network battery, both probability surveys asked specifically about Americans’ political discussants, using identical network name generators. However, as shown in italics below, two minor variations made it clear to respondents in the 2020 interview that people known online could and should be mentioned in 2020. Details for the 1996 study can be found in Mutz, D.C., Hearing the Other Side: Deliberative versus Participatory Democracy, with raw data available in the Harvard Dataverse: 10.7910/DVN/JRR3RX . In 2020, the National Opinion Research Center at the University of Chicago gathered the more recent national survey data using address-based sampling, which in today’s environment is a superior sampling technique to RDD, which was standard in 1996. The survey included the same questions in the same order, and both were random probability samples. Nonetheless, differences in sampling methods as well as response rates could alter sample composition. No weights were used in any of the analyses presented here. Political Discussant Name Generator From time to time, people discuss government, elections, and politics with other people. We’d like to know the first names or just the initials of people you talk with about these matters. These people might be from your family, from work, from the neighborhood, [from the internet,] from some other organization you belong to, or they might be from somewhere else. Who is the person you’ve talked with most about politics [either in person or online] ? [I’m/We’re] just asking for names or initials to keep things straight. Aside from this person, who is the person you’ve talked with most about politics [either in person or online] ? In all other respects, the two surveys remained the same. The same survey questions were asked in the same order using the same question wordings as in the 1996 survey. Respondents were asked about up to three political discussants with identical prompts in 1996 and 2020. Because best practices have changed, and interview mode can affect responses, it was necessary to embed an experiment in the 2020 survey to ensure results were not due to mode of interview effects. A random half of those 2020 respondents who could be interviewed via online surveys were interviewed online, while the other randomly selected half was interviewed by telephone, just as in 1996. This experiment allowed conclusions holding mode of interview constant when comparing the 1996 and 2020 telephone respondents. IRB Review. Due to minimal risk to participants, this study was judged to be exempt by the University of Pennsylvania IRB under Protocol number # 843673. All data were made anonymous and untraceable before the research firms delivered it to the authors. Since informed consent already had been administered by Amerispeak before this study, no additional consent was administered. Supplementary Material Appendix 01 (PDF) pnas.2516942122.sapp.pdf (585.5KB, pdf) Acknowledgments I thank Roseline Gray, Tyler Leigh, and Alex Tolkin for their meticulous coding skills. Two reviewers also greatly improved the manuscript’s focus. Russell Sage Foundation, Grant # G-2004-22929 (D.C.M.) Institute for Study of Citizens and Politics, University of Pennsylvania (D.C.M.). Author contributions D.C.M. designed research; performed research; analyzed data; and wrote the paper. Competing interests The author declares no competing interest. Footnotes Reviewers: J.S., William & Mary; and P.S., Stanford University. * Four questions characterized the perceived politics of each discussant. Although five questions per discussant were included in both the 1996 and 2020 surveys, in order to have a meaningful 0 point, I dropped the indicator used in the 1996 book that did not allow this. Indicators formed a highly reliable index when comprised of these four items ( SI Appendix, Appendix A ), thus suggesting little if any loss of information. † https://www2.census.gov/programs-surveys/demo/visualizations/geographic-mobility/acs-1yr/residence.pdf . ‡ https://www.pewresearch.org/short-reads/2021/05/04/70-of-u-s-social-media-users-never-or-rarely-post-or-share-about-political-social-issues/ . § https://election.lab.ufl.edu/national-turnout-rates-graph/ . ¶ https://electionstudies.org/data-tools/anes-guide/ . Data, Materials, and Software Availability Survey data have been deposited in Harvard Dataverse ( https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/JQONLR ) ( 29 ). Previously published data were used for this work ( 10.7910/DVN/JRR3RX ). Supporting Information References 1. Iyengar S., Lelkes Y., Levendusky M., The origins and consequences of affective polarization in the United States. Annu. Rev. Polit. Sci. 22, 129–146 (2019). [ Google Scholar ] 2. Habermas J., The Structural Tranformation of the Public Sphere (The MIT Press, Cambridge, MA, 1989). [ Google Scholar ] 3. Schudson M., “Was there ever a public sphere? If so, when?” in Habermas and the Public Sphere, Calhoun C., Ed. (MIT Press, Cambridge, MA, 1992), pp. 143–163. [ Google Scholar ] 4. Mutz D. C., Is deliberative democracy a falsifiable theory? Annu. Rev. Polit. Sci. 11, 521–538 (2008). [ Google Scholar ] 5. Mutz D. C., Hearing the Other Side: Deliberative versus Participatory Democracy (Cambridge University Press, New York, 2006). [ Google Scholar ] 6. Elliott K. J., Democracy for Busy People (University of Chicago Press, Chicago, 2023). [ Google Scholar ] 7. Gibson J. L., Sutherland J. L., Keeping your mouth shut: Spiraling self-censorship in the United States. Polit. Sci. Q. 138, 361–376 (2023). [ Google Scholar ] 8. Carlson T. N., Settle J. E., What Goes without Saying (Cambridge, New York, 2022). [ Google Scholar ] 9. Ahn C., Mutz D. C., The effects of polarized evaluations on political participation: Does hating the other side motivate voters? Public Opin. Q. 87, 243–266 (2023). [ Google Scholar ] 10. Lee B., Bearman P., Political isolation in America. Network Sci. 8, 333–355 (2020). [ Google Scholar ] 11. Robert Huckfeldt R., Johnson P. E., Sprague J. D., Political Disagreement: The Survival of Diverse Opinions within Communication Networks (Cambridge, New York, 2004). [ Google Scholar ] 12. Mutz D. C., “Political networks in a polarized age” in Nobel Volume on Networks, Andersson T., Edling C., Eds. (Oxford University Press, New York, Forthcoming). [ Google Scholar ] 13. Paik A., Sanchagrin K., Social isolation in America: An artifact. Am. Sociol. Rev. 78, 339–360 (2013). [ Google Scholar ] 14. Hetherington M., Weiler J., Prius Or Pickup?: How the Answers to Four Simple Questions Explain America’s Great Divide (Houghton Mifflin, New York, 2018). [ Google Scholar ] 15. Settle J. E., Frenemies: How Social Media Polarizes America (Cambridge University Press, New York, 2018). [ Google Scholar ] 16. Guess A. M., (Almost) Everything in moderation: New evidence on Americans’ online media diets. Am. J. Polit. Sci. 65, 1007–1022 (2021). [ Google Scholar ] 17. Nyhan B., et al. , Like-minded sources on Facebook are prevalent but not polarizing. Nature 620, 137–144 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Kruse L. M., Norris D. R., Flinchum J. R., Social media as a public sphere? Politics on social media. Sociol. Q. 59, 62–84 (2017). [ Google Scholar ] 19. Yi-Fan Su L., et al. , Analyzing public sentiments online: Combining human- and computer-based content analysis. Information Commun. Soc. 20, 406–427 (2017). [ Google Scholar ] 20. Nithyanand R., Schaffner B., Gill P., Online political discourse in the Trump era. arXiv [Preprint] (2017). https://arxiv.org/abs/1711.05303 (Accessed 9 September 2025). 21. Talisse R., Civic Solitude: Why Democracy Needs Distance (Oxford University Press, New York, 2024), p. ix. [ Google Scholar ] 22. Chong D., Citrin J., Levy M., The realignment of political tolerance in the United States. Perspect. Polit. 22, 131–152 (2024). [ Google Scholar ] 23. Mutz D. C., Freedom of speech in the post-Floyd era: Public support for political tolerance. Ann. Am. Acad. Polit. Soc. Sci. 708, 184–205 (2023). [ Google Scholar ] 24. Settle J. E., Carlson T. N., Opting out of political discussions. Polit. Commun. 36, 476–496 (2019). [ Google Scholar ] 25. Muirhead R., A defense of party spirit. Perspect. Polit. 23, 713–727 (2006). [ Google Scholar ] 26. Talisse R., Overdoing Democracy: Why We Must Put Democracy in its Place (Oxford University Press, New York, 2019), p. 25. [ Google Scholar ] 27. McClosky H., Consensus and ideology in American politics. Am. Polit. Sci. Rev. 58, 361–382 (1964). [ Google Scholar ] 28. Stouffer S. A., Communism, Conformity and Civil Liberties: A Cross-section of the Nation Speaks Its Mind (Transaction, Piscataway, NJ, 1955). [ Google Scholar ] 29. Mutz D. C., Replication data for “The persistence of cross-cutting discussion in a politicized public sphere.” Harvard Dataverse. 10.7910/DVN/JQONLR. Deposited 7 September 2025. [ DOI ] [ PMC free article ] [ PubMed ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Appendix 01 (PDF) pnas.2516942122.sapp.pdf (585.5KB, pdf) Data Availability Statement Survey data have been deposited in Harvard Dataverse ( https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/JQONLR ) ( 29 ). Previously published data were used for this work ( 10.7910/DVN/JRR3RX ). Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences ACTIONS View on publisher site PDF (524.5 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top