How Right‐Wing Populist Comments Affect Online Deliberation on News Media Facebook Pages

Right‐wing populist user comments on social media are said to impair online deliberation. Right‐wing populism’s anti‐ pluralist and conflict‐centered message might hinder deliberative debates, which are characterized by reciprocity, argu‐ ments, sourcing, politeness, and civility. Although right‐wing populism has been found to foster user interaction on social media, few empirical studies have examined its impact on the scope and deliberative quality of user debates. This study focuses on debates on 10 Facebook pages of Austrian and Slovenian mass media during the so‐called “refugee crisis” of 2015–2016. Proceeding in two steps, we first analyze how right‐wing populist user comments affect the number of reply comments using a dataset of N = 281,115 Facebook comments and a validated, automated content analysis. In a second step, we use a manual, quantitative content analysis to investigate how right‐wing populist comments affect the delibera‐ tive quality of N = 1,413 reply comments. We test five hypotheses in carefully modeled regression analyses. Our findings show that right‐wing populist comments trigger replies but impair their deliberative quality. People‐centric comments decrease the probability of arguments in replies, and anti‐immigrant comments spark incivility. Countering populism fur‐ ther increases impoliteness. We discuss our findings against the backdrop of an increasingly uncivil online public sphere and populism’s ambivalent relationship with democracy.


Introduction
Right-wing populism has shaped Europe's political landscape in recent decades (Mudde, 2013). To understand its success, researchers increasingly focus on populism as a communication phenomenon (de Vreese et al., 2018). Social media has been found to be the preferred channel for populist communication Gerbaudo, 2018). A different, growing branch of research is concerned with citizens' populist attitudes (e.g., Zaslove et al., 2021). However, only few studies have examined ordinary citizens' expressions of populist views in online public spheres. Initial findings show that user-generated populism flourishes in comments sections below news stories Galpin & Trenz, 2019;Thiele, 2022a) and breaches norms of democratic communication (Hameleers, 2019). What we do not know is how these comments affect other users and discussions among them. This study narrows this research gap by asking: How do right-wing populist comments affect the number and deliberative quality of reply comments on Facebook?
Deliberation is the respectful exchange of reasons in public and is considered vital for democracy (e.g., Friess & Eilders, 2015;Habermas, 1996). At the beginning of the millennium, user comments were expected to enhance deliberation (Dahlberg, 2011). Repeated findings of incivility in comments (e.g., Coe et al., 2014;Rowe, 2015) have left little of that hope (Quandt, 2018). Scholars have argued that populism may further impair deliberation (Abts & Rummens, 2007;Waisbord, 2018). Populism is a "thin" ideology that holds that a "corrupt elite" deprives "the people" of their sovereignty (Mudde, 2004, p. 543). Its anti-pluralism and Manicheanism, so the argument goes, run contrary to an exchange of reasons (Abts & Rummens, 2007;Waisbord, 2018). Other authors have argued that populism's conflictive message could revitalize democratic debates (Laclau, 2005;Mudde & Kaltwasser, 2012). Connecting the separate strands of populist communication and online deliberation research, this study aims to contribute empirically to this normative debate.
Furthermore, we aim to disentangle the relationship between different dimensions of right-wing populism and deliberation. The right-wing variant of populism clings to the nativist idea that foreigners are threatening (Mudde, 2007, p. 156), which renders it even more problematic for liberal democracy (Sauer et al., 2018). Hameleers' (2019) qualitative content analysis of user-generated, right-wing populist content on Facebook illustrates this threat but leaves open the question of whether populism per se or right-wing ideology is the problem. Here, we differentiate between right-wing, anti-immigrant messages and populist messages that involve anti-elitism or people-centrism. Likewise, to grapple with the presumably ambivalent impact of right-wing populism on democratic debates (Canovan, 1999;Mudde & Kaltwasser, 2012), we differentiate five key aspects of deliberation, namely reciprocity, argumentation, sourcing, politeness, and civility . Using a quantitative content analysis allows us to untangle the relations between those dimensions.
We focus on user debates during the so-called "refugee crisis" in 2015-2016 on Facebook pages of news media from Austria and Slovenia. These neighboring countries faced similar challenges during the crisis but diverge markedly in terms of the commenting behavior of their citizenry (European Commission, 2016, p. 452). The crisis was accompanied by a growing polarization among citizens (van der Brug & Harteveld, 2021) and a shift of public discourse towards right-wing populism (Krzyżanowski, 2018), which also influenced the positions of competing political parties (Gessler & Hunger, 2022). Focusing on the impact of right-wing populist comments on user debates complements our understanding of this crisis, as heated debates below news stories have been found to fuel audience polarization (Asker & Dinas, 2019).
Facebook is a popular tool used by media houses to publish news stories and to invite the audience to com-ment (Humprecht et al., 2020). The platform allows users to reply to other users' comments, which promotes reciprocal discussions among users (Esau et al., 2017). This structure allows us to study the impact of right-wing populism in higher-level comments on the number and quality of replies.
The empirical analysis of this study was carried out in two steps. In the first step, we conducted an automated content analysis of 281,115 Facebook comments found below posts on 10 popular Facebook pages of Austrian and Slovenian news media, analyzing the impact of right-wing populist comments on the number of replies. In the second step, we sampled 535 comments from this population, downloaded up to five replying comments, and conducted a manual, quantitative content analysis of 1,413 replies to investigate the impact of right-wing populism on deliberative quality. Our findings show that right-wing populist comments triggered an increase in the number of replies but induced a deterioration of their deliberative quality. Countering populism further increased levels of impoliteness in replies. Our findings substantiate the theorized ambivalent relationship between right-wing populism and democracy (Mudde & Kaltwasser, 2012) at the level of user debates.

Right-Wing Populist User Comments
Right-wing populism is here defined as a compound of two ideologies. On the one hand, populism is a "thin" ideology that asserts that "the people" are ruled by a "corrupt elite" and demands the restoration of the people's sovereignty (Mudde, 2004, p. 543). This ideology is "thin," in the sense that its focus is limited to these core ideas (Freeden, 1996). At the core of right-wing populism is, additionally, the nativist idea that foreign, "nonnative elements…are fundamentally threatening" (Mudde, 2007, p. 19). This additional dimension makes right-wing populism a "thicker" ideology (Krämer, 2017). Nativism is frequently articulated as opposition to immigration (Mudde, 2007, p. 19). We focus on the expression of these ideas in texts as right-wing populist content (de Vreese et al., 2018). Following previous operationalizations (Aslanidis, 2018;Wirz et al., 2018), we capture three dimensions of right-wing populist content: anti-elitist, people-centrist, and anti-immigration messages.
Various actors can disseminate right-wing populist messages (de Vreese et al., 2018). Scholars have focused on the populist communication of politicians (e.g., Ernst et al., 2019;van Kessel & Castelein, 2016) and the media (e.g., Wirz et al., 2018). This research has found that social media provides favorable opportunity structures for populism (e.g., Blassnig & Wirz, 2019;Ernst et al., 2019). Although "the people" play a crucial role in populist thought, few studies have investigated populist content from ordinary citizens. Galpin and Trenz (2019) analyzed user comments on media websites and found evidence for "participatory populism." This content analysis, however, was limited to negativity (Galpin & Trenz, 2019, p. 788). Using a more sophisticated coding scheme,  analyzed news coverage of immigration and user comments on media websites and found that populist reporting stirs populist comments.
To our best knowledge, only Hameleers (2019) analyzed populist user content against the backdrop of democratic communication. Conducting a qualitative analysis of right-wing populist Facebook community pages, Hameleers (2019) showed how this user content infringed upon democratic norms through extreme hostility and avoidance of argumentative debates. However, these findings are limited to a niche public of users who actively engage with right-wing populist community pages (Hameleers, 2019). Secondly, the qualitative approach of that study does not allow sufficient differentiation between populist and right-wing elements, as suggested by populism scholars (Rooduijn, 2019). Finally, that study did not investigate the effects of right-wing populist messages on other users. We aim to overcome these limitations by distinguishing between right-wing and populist content, by investigating its impact on replies from other users, and by analyzing comments sections on Facebook pages of news media organizations that reach a broad public.

Online Deliberation
Comments sections, as other interactive innovations, have changed today's media logics (Klinger & Svensson, 2015). Converging the roles of content producer and consumer, user comments allow ordinary citizens to reach similar audiences as professional journalistic output (Springer et al., 2015). Commenting on the news allows users to engage in discussions and deliberative interaction (Springer et al., 2015), to influence the perceived public opinion (e.g., Eilders & Porten-Cheé, 2022), and to counter-frame news stories (Liu & McLeod, 2019). Initially, this potential raised scholars' hopes that user comments may contribute to a more inclusive, participatory, and deliberative public sphere (Dahlberg, 2011;Ruiz et al., 2011). However, comment sections have been repeatedly found to be plagued by incivility (e.g., Coe et al., 2014), and little of this optimism remains (Quandt, 2018).
Most news media organizations run pages on Facebook, which continues to be the social medium with the highest number of users (Newman et al., 2016, p. 10). On these pages, media houses post news stories and invite users to comment, hoping to increase the visibility of their stories (Singer, 2014) and guide traffic to their websites (Humprecht et al., 2020). Facebook allows commenters to reply directly to each other, which fosters reciprocal discussions (Esau et al., 2017). However, the deliberative quality of Facebook comments has been found to be lower than on news websites (Rowe, 2015).
Here, we analyze both the scope and quality of such reciprocal discussions on Facebook.
To assess the democratic quality of online discussions, scholars have turned to the concept of deliberation. Deliberation denotes "a rational, constructive, reciprocal, and respectful exchange of reasons among equal participants" (Friess et al., 2020, p. 3). Its proponents argue that deliberation yields desirable outcomes for democratic societies Habermas, 1996). However, there is little consensus about the criteria that render a debate or statement deliberative (Mutz, 2008). To arrive at a set of operationalizable criteria, we conducted a literature review of 18 recent empirical studies (see Supplementary File, Appendix A). Drawing on this review, we distilled five key dimensions of deliberation-reciprocity, argumentation, sourcing, civility, and politeness-similar to Friess et al. (2020).
Reciprocity is an interactive process in which participants listen and respond to each other (Friess & Eilders, 2015). Here, we consider the number of reply comments under a Facebook comment to reflect the scope of reciprocity. The other four criteria of deliberative communication characterize the content of a comment. Argumentation involves the provision of reasons for one's claims . These reasons can be backed up with verifiable information by making sources transparent (Stromer-Galley, 2007). Politeness and civility both characterize respectful communication . While many authors use both terms interchangeably (e.g., Coe et al., 2014), we follow Papacharissi's (2004) argument to conceptualize politeness as a matter of tone and incivility as discourse that substantially violates democratic values. Politeness can be grasped then as the absence of impoliteness, understood as an "unnecessarily disrespectful tone" (Coe et al., 2014, p. 660). Civility, by contrast, denotes messages that do not entail stereotypes, racism, violent speech, or the intent to silence others (Papacharissi, 2004, p. 274). While uncivil comments can hardly be polite, impolite comments may serve the democratic function of exposing others to different views, as shown by Rossini (2020), using different labels. Kalch and Naab (2018) demonstrate that impoliteness and incivility have a different impact on the responses of others. Such differences might be particularly relevant in a context where users confront extremely right-wing positions.
While we expect an overall low level of deliberative quality of Facebook comments, the debate surrounding the democratic implications of populism (e.g., Mudde & Kaltwasser, 2012) raises the question of its empirical impact on online deliberation.

The Impact of Right-Wing Populist Comments on Online Deliberation
Populism has an ambivalent relationship with democracy (Mudde & Kaltwasser, 2012). On the one hand, demanding the implementation of the people's will is inherently democratic (e.g., Canovan, 1999) and may mobilize parts of the society that feel misrepresented by mainstream politics (Mudde & Kaltwasser, 2012, p. 21). However, populism's crude majoritarianism and anti-pluralism threaten liberal democracy (Canovan, 1999;Mudde & Kaltwasser, 2012, p. 21). This threat, stemming from a neglect of minority rights, is arguably even more severe in populism's right-wing variant (Mudde, 2007, p. 156;Sauer et al., 2018). Democratic debates are likely to suffer from expressions of this Manichean worldview, often voiced aggressively by populists (Waisbord, 2018). We expect that this democratic ambivalence is reflected in the impact that right-wing populist user comments have on online deliberation.
Regarding the scope of reciprocity, we expect a mobilizing effect of right-wing populist comments. Previous research has found that right-wing populist messages on Facebook trigger user interactions (Bobba, 2018;Jost et al., 2020). Blassnig and Wirz (2019) found that this effect is driven by activating a populist schema in like-minded users. At the same time, we expect that the conflict-centered messages of right-wing populism might provoke objections from opposing users. These effects should hold for populist and right-wing content alike. As such, we propose the following hypothesis: H1: Right-wing populist user comments receive more reply comments than other comments.
Regarding the quality of deliberation in reply comments, we expect a deteriorating impact of right-wing populism. Populism's construction of "the people" as a homogeneous group that is oppressed by a "corrupt elite" is moralistic (Mudde, 2004, p. 544). Moralization makes argument-based objections rather pointless (Hameleers, 2019;Mudde & Kaltwasser, 2012). Additionally, populist messages claim to be an immediate expression of the "vox populi" (Canovan, 1999, p. 14). Within the populist logic, this makes further arguments from supporters unnecessary (Abts & Rummens, 2007;Krämer, 2017). If this reasoning is correct, we should find this impeding effect for populist messages and less so for anti-immigration messages: H2: Populist user comments decrease the probability of arguments in reply comments.
Populism delegitimizes not only political elites, but also journalists (Egelhofer et al., 2021) and scientific experts (Mede & Schäfer, 2020). This aversion to expert knowledge aligns it with a sprawling post-truth politics (Waisbord, 2018). We argue that the rejection of established sources of knowledge may discourage replying users from referring to such sources: H3: Populist user comments decrease the probability of the provision of sources in reply comments.
Hypotheses 2 and 3 have focused on effects of populist messages. Both anti-immigrant and anti-elitist messages, however, might raise levels of incivility among reply comments. Survey research has shown that socially undesirable statements are withheld if the respondent fears being sanctioned (Krumpal, 2013). Uncivil statements, such as stereotypes, racism, or approval of violence (Papacharissi, 2004), fall into this category. We argue that right-wing populist comments may signal to like-minded users that the risk of sanctions is low, thus raising their readiness to express uncivil opinions (Keum & Miller, 2018): H4: Right-wing populist user comments increase the probability of uncivil reply comments.
Similar contagion effects have been observed for impolite user comments (Song et al., 2022). Right-wing populist comments are characterized by their harsh tone (Hameleers, 2019). We expect that this rudeness might spill over to reply comments: H5: Right-wing populist user comments increase the probability of impolite reply comments.
In addition to these hypotheses, we want to know what happens when users counter populist or anti-immigrant comments. Friess et al. (2020) have shown that civic interventions against hate speech in online comments can improve the deliberative quality of debates. On the other hand, disagreement in comments sections has been linked to increased levels of impoliteness (Rossini, 2021). Since these findings do not suggest a clear hypothesis in either direction, we ask the following additional research question: RQ: How does countering right-wing populism affect the deliberative quality of reply comments?

Research Design
To test our claims, we conducted two content analyses of user comments on Facebook pages of Austrian and Slovenian news media. We chose the timeframe of July 2015 to August 2016, which covers the so-called "refugee crisis," as we expected many right-wing populist comments and heated debates in this context . The arrival of millions of refugees in the wake of the Syrian war attracted enormous media attention (Greussing & Boomgaarden, 2017). After an initial phase of welcoming by volunteers, especially in Austria, the right-wing demand for stricter border controls became increasingly prevalent. This culminated in the closure of the "Balkan route" in 2016, in which Austria and Slovenia took the lead (Gruber, 2017;Vezovnik, 2018). In both neighboring countries, right-wing populist mobilization surged in the aftermath of the crisis (Bodlos & Plescia, 2018;Pajnik & Šori, 2021;Thiele et al., 2021).
The two countries have similar media systems (Herrero et al., 2017). However, in a European-wide comparison in 2015, Austrians were the most active online commenters (52%), while Slovenes (20%) exhibit low levels of commenting activity (European Commission, 2016, p. 452). In 2015, Facebook was the most widely used social medium (Newman et al., 2016, p. 10).
The Facebook data analyzed here has a nested structure. Mass media outlets operate Facebook pages and share news items as posts. Users can comment on these posts. We call this first level of comments "parent comments." On a second level, users can respond to comments in "reply comments." Our analysis proceeds in two steps. In Step 1, we analyze the effect of right-wing populism in parent comments on the number of replies using a large-N design and a computational content analysis. In Step 2, we analyze the deliberative quality of the content of replies to a small subsample of parent comments using a quantitative, manual content analysis.
We downloaded all publicly accessible Facebook posts from each page in the timeframe of July 2015 to August 2016 using the Facebook Graph API and Facepager (Jünger & Keyling, 2020). For each of the 7,658 posts, we downloaded up to 500 anonymized user comments, resulting in a sample of N = 281,115 parent comments, which constitutes our sample for Step 1 of our analysis. For Step 2, we narrowed down the population of parent comments to comments on posts about migration that received at least one reply. To detect the topic of migration, we used two validated dictionaries (see Supplementary File, Appendix B). Next, we applied a preliminary version of our automated measurements described below to ensure a sufficient representation of right-wing populist comments. We drew two stratified random samples of 300 parent comments per country, oversampling highly populist and anti-immigrant comments. For each parent comment, we then downloaded up to five replies, following Ziegele et al. (2020, p. 874). After dropping empty observations, N = 1,413 replies and 535 parent comments were analyzed in Step 2.

Variables
The dependent variable in Step 1 is the number of replies attracted by each analyzed parent comment, as returned from the Facebook API. By this number, we operationalize the scope of reciprocal discussion among users.
The explanatory variables in Step 1 are populist and anti-immigration content in parent comments. Following Aslanidis' (2018) argument that expressions of populism are best understood as a matter of degree, we measure both as continuous variables, applying a computational content analysis called distributed dictionary representation (DDR; Garten et al., 2018). This method combines dictionaries with word vectors. Dictionaries measure concepts by counting keywords, but struggle to arrive at exhaustive word lists (Rauh, 2018). The DDR method circumvents this problem by representing a short list of expressive keywords as word vectors (Garten et al., 2018). Word vectors are learned by neural networks and claim to represent the semantic similarity of words (e.g., Bojanowski et al., 2017). The vector representations of all words in a dictionary are averaged into one dictionary representation. The same is done for each document. The DDR method then computes the cosine similarity between the average dictionary vector and each document vector. This results in a measure ranging from −1 to +1 that provides a crude indicator for how strongly the concept is represented in each document (Garten et al., 2018).
We used our R-package dictvectoR (Thiele, 2022b) to apply the DDR method and to systematically develop concept dictionaries. The development process is documented in detail in the Supplementary File, Appendix D. Two language-specific fasttext word-vector models (Bojanowski et al., 2017) were trained on our corpora. To optimize and validate our measurements, we tested how well they predicted the binary, human coding obtained in Step 2 (see Supplementary File, Appendix C). Two-thirds of the sample coded for Step 2 were used for optimization, the remaining third for validation. Table 1 reports the validation scores Recall, indicating the proportion of relevant documents predicted correctly, Precision, the share of correct hits in all predictions, and their harmonic mean F1 (Stryker et al., 2006). The concepts of anti-immigration and populism were measured separately for each language. The short dictionaries align with the authentic language used in user-generated content and reflect equivalent dimensions. However, they also reflect country-specific discourses. In the DDR method, the average representation of all dictionary words is decisive. Therefore, it is not necessary that all terms be in themselves anti-immigrant or populist (e.g., "politicians"). Moreover, the method captures documents that resemble the combined meaning of the dictionary words without matching them exactly. Given the satisfactory F1 scores between .69 and .76, we consider the measurements good approximations for right-wing populist content. All DDR measures were standardized and mean-centered at the country level.
As a control in Step 1, we included a variable indicating whether a post addressed migration to account for an effect of issue salience. We also controlled for Step 2, we conducted a manual, quantitative content analysis. One author constructed a codebook (Supplementary File, Appendix C) inspired by previous research Friess et al., 2020). Two authors conducted the coding. Extensive training ensured reliable coding, which was tested on 234 translated comments and measured by Krippendorff's alpha. All categories but "positioning" were coded for parent and reply comments.
The dependent variables in Step 2 are four binary indicators for the quality of deliberation in responses. Argumentation ( = .75, n = 234) was coded if the comment provided reasons for its claims . Sourcing ( = .90) was coded if the comment referred to hyperlinks or external sources of knowledge (Marzinkowski & Engelmann, 2022). Incivility ( = .71) was coded if a comment dehumanized others, used stereotypes, sexism, or racism, supported violence , or silenced others (Oz et al., 2018). We considered a comment impolite ( = .81) if it included name-calling, vulgarity, sarcasm, depreciation, or shouting .
The main explanatory variables indicate three dimensions of right-wing populism in parent comments. People-centric ( = .74) messages invoke the people as a virtuous, homogeneous, or victimized group or stress the people's will (Aslanidis, 2018;. Anti-elitism ( = .73) was coded if a comment discredited or blamed power holders (Aslanidis, 2018;. Anti-immigration ( = .81) was coded when comments opposed immigration or considered it a threat to security, economy, or culture (Callens & Meuleman, 2017).
To answer the question regarding the effect of countering right-wing populism, we coded if the reply comment agreed, disagreed, or was neutral towards the parent comment (positioning = .74, n = 138; Marzinkowski & Engelmann, 2022). We then constructed a binary variable for countering populism, indicating for each reply whether any of the preceding replies disagreed with a people-centric or anti-elitist parent comment without using these discourses themselves. Countering anti-immigration was constructed analogously. As control variables, we included the respective indicator for deliberative quality on the parent comment level and the length of the parent comment in characters. Summary statistics are reported in the Supplemenatry File, Appendix E.

Model Specifications
We tested our hypotheses in carefully constructed regression models. In Step 1, the dependent variables are count variables, so we fitted negative binomial regression models. As our automated measurements of populism and anti-immigration are language-specific, we fitted separate models for each county. The data have a nested structure, with comments nested in posts and posts nested in accounts. We accounted for the two levels (post and accounts) using multilevel models. For Step 2, we ran four logistic regression models, one for each binary indicator of deliberative quality in reply comments. As the number of observations per level was very limited, we accounted for the nested structure of the data on the Facebook page level by including dummy variables for each page-account. This cancels out between-group effects on this level (Bell et al., 2019).

Results
Right-wing populism was found in the comments of all analyzed media Facebook pages. Differences across media types were small, but mostly significant (see Supplementary File, Appendix F). Comments on public broadcasters' pages were the most strongly populist and anti-immigrant in both countries. Surprisingly, we found that tabloid newspapers attracted the least populist and anti-immigrant comments in Slovenia and scored only second in Austria. In Step 1, we focused on the number of replies per parent comment. Parent comments received .7 (SD = 3.4, Max = 248) replies on average in Austria and .5 (SD = 2.4, Max = 127) in Slovenia. The results from two multilevel, negative binomial regression models show that the degree of both anti-immigration and populism increased the number of reply comments significantly. These effects were significant in both countries. The regression tables are documented in the Supplementary File, Appendix G. Figure 1 visualizes the effects as incidence rate ratios, which are the exponentiated -coefficients. Positive effects are indicated by incidence rate ratios values above 1 and negative effects by values below 1. An increase in anti-immigration in parent comments by 1 SD increased the expected count of reply comments by a factor of 1.5 in Austria and by 1.4 in Slovenia (Blassnig et al., 2019, p. 640). Populism had a similar, significant positive impact. These findings support H1 and indicate that right-wing populist content triggers user discussions. These effects are significant, even when controlling for a salience effect of the topic of migration. Interestingly, the topic of migration was associated with an increased number of replies in Austria but a decreased number of replies in Slovenia. Looking at the other control variables, we see that tagging users and comment length were associated with an increased number of responses. Comments that reacted to dated posts received fewer responses in Austria.
In the second step of our analysis, we focused on the deliberative quality in reply comments. In all analyzed parent and reply comments, right-wing populist messages were significantly more often impolite (90%) and uncivil (38%) than other comments (54%/3%). Surprisingly, right-wing populist comments coincided more often with arguments (32% vs. 17%). Sourcing was equally rare (3%) in both categories. We ran four logistic regression models using the binary indicators for deliberative quality in replies as dependent variables (Table 2).
We found a significant negative effect of peoplecentric parent comments on argumentation (Model 3) and a weakly significant positive influence of antiimmigration on incivility (Model 5). The effects are visualized as average predicted probabilities in Figure 2. A people-centric parent comment decreased the probability of a response including an argument from 25% to 18%, holding all other variables at their observed values and averaging across all predictions. Anti-immigrant parent comments, in turn, increased the probability of uncivil responses from 13% to 19%. These findings support H2 and H4. However, the effect of anti-immigration on incivility disappears when controlling for anti-immigration in the reply, as a closer analysis shows, which is not presented here. We discuss this finding in the conclusion. Contrary to our expectations, we found a weakly significant, negative effect of anti-elitism on incivility.  Notes: *** p < 0.001, ** p < 0.01, * p < 0.05; (R) reply; (P) parent comments. Model 4 did not find any meaningful predictors for sourcing. Impoliteness in responses (Model 6) was best predicted by impoliteness in the parent comment. We found analogous contagion effects for argumentation and incivility in Models 3 and 5. None of these findings let us reject the null hypotheses against H3 and H5. We included two variables to answer our research question regarding the impact of countering right-wing populism on deliberation. Interestingly, we found that countering populist content significantly increased the probability of impoliteness in subsequent replies (Model 6). Figure 3 visualizes this effect, showing that previous countering increased the predicted probability of impoliteness in replies from 68% to 80%.

Discussion and Conclusion
This study set out to investigate the impact of rightwing populist user comments on online deliberation. Proceeding in two steps using a computational and a manual content analysis, we analyzed the impact of right-wing populist comments on Facebook pages of Austrian and Slovenian news media during the "refugee crisis" of 2015-2016 on the number and deliberative quality of replies. Our findings show that populist and anti-immigrant comments increased the scope of replies but impaired their deliberative quality. This evidence empirically underlines the ambivalent relationship between right-wing populism and democracy (Mudde & Kaltwasser, 2012) at the level of user debates and points to differential effects of right-wing populist communication.
Both populist and anti-immigrant messages sparked discussions among users in our sample. This confirms previously identified mobilization effects of right-wing populist content on social media (Blassnig & Wirz, 2019;Jost et al., 2020). This might be driven by activating a cognitive schema in like-minded users (Blassnig & Wirz, 2019) or by provoking replies from opponents. High levels of reciprocal user discussions seem desirable from the viewpoint of deliberation theorists (e.g., Friess & Eilders, 2015). At the same time, this engagement increases the visibility of right-wing populist content (Singer, 2014). According to the spiral-of-silence theory (Eilders & Porten-Cheé, 2022), this can lead to an overestimation in perceived public opinion with subsequent consequences for the statements of others.
Indeed, we found some worrisome consequences. People-centric comments decreased the readiness of users to present arguments in replies. This effect might be caused by populism's claim to be an immediate expression of the people's will (Canovan, 1999;Waisbord, 2018), which renders any supporting arguments unnecessary (Krämer, 2017) and counterarguments pointless (Hameleers, 2019).
Anti-immigrant comments more often entailed uncivil replies. Arguably, these messages lowered the bar for like-minded users to express racist or violent views (Keum & Miller, 2018). Anecdotal evidence supports this view. One Austrian commenter demanded to "castrate this scum with a vise," responding to a comment that called for the deportation of an asylum-seeker and alleged sex offender. We found a similar example in the Slovenian corpus. This shows that right-wing populism can "normalize" (Wodak, 2021) incivility and racism also in online debates. However, we note that this is primarily a contagion effect of anti-immigrant content and partly driven by our overlapping operationalization of anti-immigrant and uncivil statements, which followed Papacharissi's (2004) concept. Regardless of this, the effect seems problematic for democratic debates. Future studies, however, should aim to delineate the two concepts more clearly.
We did not find an impact of right-wing populist content on sourcing or impoliteness in replies. What our findings show, however, is that debates in which one reply countered populism escalated in terms of impoliteness. This is illustrated by the case of a commenter who countered the populist claim that the "clowns in government" would not care about "Austrians who cannot  Figure 3. Average predicted probabilities of impoliteness in replies. Note: 95% confidence intervals. afford heating" by hinting soberly at the public "heating cost allowance"; this comment then faced a variety of insults, ranging from "do-gooder" to "bullshit." This finding adds to previous evidence that disagreement in online discussions fosters impoliteness (Rossini, 2021). Unfortunately, we did not find that civic interventions improve deliberative quality, as Friess et al. (2020) did.
Media organizations face conflicting incentives to restrict right-wing populist comments. On the one hand, media houses profit commercially from high levels of user interaction on Facebook, as this increases the visibility of news stories (Singer, 2014). Accepting right-wing populism as a driving force for user engagement, however, might come at a cost. It may not only diminish the quality of online debates, as shown here, but could even backfire commercially, as low standards of online debates have shown to inhibit users from commenting (Springer et al., 2015). Here, we did not find evidence that right-wing populist comments were given preferential treatment from media houses for commercial reasons. Instead, we found that populist commenters are particularly attracted by publicly funded broadcasters. This is remarkable, since public broadcasters are a noted foe of populist politicians (Egelhofer et al., 2021), and populist communication is often associated with tabloid journalism Mazzoleni, 2008). Future research should inspect the media preferences of populist commenters more systematically.
Our study comes with several limitations. Firstly, the focus on the highly polarized context of the "refugee crisis" is likely to have affected our findings. In particular, the finding that countering populism fueled impoliteness should be scrutinized in a less-polarized setting. We advise future research to broaden the thematic focus and to consider a coding scheme that better captures the complexity of positions and references in comments. Secondly, our findings are limited to the two countries Austria and Slovenia. Furthermore, the language differences constitute hurdles for our computational content analyses, both for capturing right-wing populism and the topic of migration, which could not be overcome satisfactorily. Future research is encouraged to tackle such methodological challenges in comparative studies of a larger scale. The same holds, thirdly, for our focus on the platform Facebook. Fourthly, the samples used in Step 2 served a primarily exploratory objective and cannot claim representativeness. Our findings should be substantiated using a more systematic sample. Finally, we suggest that the mechanisms underlying these observed relations be more fully explored in experimental studies.
In sum, this study connected the separate strands of research on populist communication and online deliberation, substantiating the ambivalent impact of right-wing populism on democratic debates, and contributed methodologically to the growing interest in computational content analysis.

Tjaša
Turnšek is a PhD candidate in communication science at the University of Ljubljana and a research assistant at the Peace Institute, Ljubljana. Her MA thesis "Critical Discourse Analysis: Media Representation of the Refugee" was awarded with the Professor Klinar Fund Award. Previously, she was project manager in an NGO that focused on awareness campaigns. Her research focuses on right-wing and media populism and the ownership structures of Slovenian media.