You are what you read: media, identity, and community in the 2020 Belarusian uprising

ABSTRACT The movement that mobilized to oppose Alyaksandr Lukashenka in August 2020 was notable for its ability to bridge divisions of social class, geography, age, and identity. Almost uniquely among post-Soviet revolutionary movements, the Belarusians who rose up were not divided from those who did not along clearly discernible socio-demographic, ethnic, linguistic, or regional lines. They were, however, separated by one very stark barrier: the one separating the country’s two distinct media systems, one controlled by the state, and one independent. Drawing on an original survey conducted in September 2020, just as the protest movement was reaching its peak, this article finds that respondents’ choice of news media was the strongest and most consistent predictor of their political opinions. Media, then, appear to have served not merely as aggregators of and conduits for social processes generated elsewhere, but as the producers of social and political force in their own right.


Introduction
In August 2020, Alyaksandr Lukashenka's attempt to prolong his tenure as president of Belarus sparked the largest protest wave the country has seen since gaining independence from Moscow, and one of the largest in the post-Soviet space. The importance of media in this mobilization is clear. In the face of a regime that has maintained tight control over the media, independent websites and social media channels were central to the mobilizational process, for movement leaders and ordinary participants alike. Without these outlets and platforms, leaders would have been unable to get their message out, protesters would have struggled to coordinate, and reports of police brutality would have failed to galvanize the moral indignation of a remarkably broad swath of Belarusian (and global) society. Unsurprisingly, then, independent media were also among the central targets of the regime's counter-mobilizational efforts before, during, and after the protests.
And yet, for all this evident centrality, the causal role of media in political mobilization is disputed, both in the Belarusian case (see, for example, Bush 2020; Litvinenko 2020) and more broadly (see, for example, Tufekci and Wilson 2012;Freelon, Mellwain, and Clark 2016). Are media simply reflections or symptoms of underlying social forces? Do media "mediate" in the proper sense of the word, aggregating and communicating small-scale phenomena into bigger ones (McCarthy and Zald 1977; Eltantawy and Wiest 2011)? Or do media themselves produce processes and exert forces that otherwise would not have existed (Bennett and Segerberg 2012;Castells 2013)? That is the puzzle that this article seeks to help elucidate. drawing support from disparate groups and classes; and for the fact that it happened at all, in the face of what is generally recognized to be a very highly authoritarian regime (see, elsewhere in this issue, Onuch and Sasse 2022;Mateo 2022).
Lukashenka's approach to political control has focused on a strategy of "preemption" -aiming since the earliest days of post-Soviet independence to prevent the emergence of institutionally robust challengers, whether in the political parties, the press, civil society or, really, anywhere (Silitski 2005). Thus, while state-controlled media dominate -in particular, the state-owned television channels Bel1 and Bel2, and a handful of state-owned newspapers -even nominally independent media and online media suffer from a combination of direct repression, selfcensorship (exercised to ward off painful consequences for non-conformity), and commercial control (Herasimenka 2016). As Wijermars and Lokot discuss (2022), opposition-minded Belarusians have turned to the internet for the same reasons that their Russian counterparts have: as a source of alternative information, and as a communications infrastructure for constructing and framing grievances, and for coordinating mobilization (see also Herasimenka 2019; Wijermars and Lokot 2022).
The Belarusian political opposition has long been ineffective, fragmented by a combination of repression and competition for scarce, mostly externally sourced resources (Ash 2015). The ineffectiveness of political opposition, in turn, may have had three socio-political consequences. One, as Elena Korosteleva writes (2010), is that citizens, unconvinced of the prospect of ever changing the power structure, learn to cope, lending tacit acceptance (and even a kind of legitimacy) to Lukashenka's rule. Second, this motivated acquiescence to Lukashenka's power allowed a relatively weak Belarusian state to cow the opposition with a moderate expenditure of coercive effort (Way and Levitsky 2006). And third, it leaves the door open for the emergence of more effective oppositional voices specifically through the media space -a space where a community of perception and interpretation had already begun to consolidate.
The Belarusian protest movement that began in 2020 is notable -among a great many thingsfor being simultaneously national and local. While sparked by a national-level event and framed with reference to encompassing symbols (including the white and red flag) and thus in some respects similar to the Orange Revolution and the Euromaidan, the Belarusian movement was multi-sited, organized through a network of local groups and, in the analysis of Mischa Gabowitsch (2021), began to assemble local communities from above. In other ways, the Belarusian protests resemble both earlier Russian and Ukrainian waves: the similarities to Russia are in the makeup of what appears to have been the protest vanguard, a largely urban community of a "new class" who "use intellectual, cultural, and educational capital to produce an income and sustain privilege" (Gapova 2021, 47); the similarities to Ukraine are in the centrality of vague and ideologically noncommittal symbolism, which allows for the coalescence of constituencies who do not otherwise share views on the ends and means of governance (Artiukh 2021;Bekus 2021). Protest frames drew on everyone from Yan Kupala to John Locke and focused on the same ideas of "dignity" as figured in the Euromaidan -and as in Ukraine (or, for that matter, in Russia), both the historical references and the contemporary buzzwords centered, as Elena Gapova (2021) has argued, on a demand for the recognition of agency, both of the individual and of the group.
While the Belarusian movement is remarkable for its inclusiveness, some people were systematically excluded (though not on purpose) -and perhaps no constituency more than rural residents. These Belarusian citizens tended, by and large, not only to reject the protests and their aims, but also to reject the morality of the protesters themselves -denying them trust and ascribing to them ulterior motivations (Hervouet 2021). And this begins to point to a broader issue: the ability of the movement to reach into almost all segments of Belarusian society does not mean that all segments were equally well represented, that all messages resonated with equal vigor, or that there were no traceable dividing lines between the mobilized and un-mobilized in Belarus. The question for this paper is what role media consumption played in constructing and actuating those dividing lines.

The Belarusian media landscape
Tight control over the media -alongside the outright repression of challengers and potential challengers for power -has generally been seen as key to Lukashenka's longevity in power, allowing him to project an image of his own popularity and to keep voters' attention focused on pro-regime narratives (Marples 2006). Unlike in Russia, for example, where media control is mostly accomplished through a combination of informal pressure and incentivized self-censorship, Belarusian law itself highly restricts the freedom of the press. Thus, the 2008 media law gives the state a monopoly not only over terrestrial television, but over all reporting on political, economic, and social affairs by any online or offline media outlet . Although more than 80% of the population has access to the internet, that access runs through a state monopoly, and the state actively blocks websites (and sometimes social media networks) that it finds problematic, and prosecutes journalists and bloggers who publish challenging information (Freedom House 2020).
Lukashenka focused on media control immediately after taking office in 1994, censoring statecontrolled newspapers and replacing their editors with his own appointees, freezing independent media out of printing and distribution systems and using restrictive laws and compliant courts to hound both publishers and individual journalists (Sahm 2009). Those independent periodicals that persisted into the twenty-first century -chiefly, Narodnaya Volya (NV) and Belaruskaya Delovaya Gazeta (BDG) -struggled to compete with the subsidized print-runs of their state-owned competitors, such as Sovetskaya Belarus and Belarus Segodnya; by the end of the first decade of the new century, both NV and BDG had shut down. Independent terrestrial TV broadcasters, needless to say, were banned altogether, although Belsat found a niche in satellite broadcasting, while Euroradio clung to a piece of the radio spectrum. By the 2010s, however, the internet had become the key medium for independent journalism, with the emergence of popular oppositional websites, including Charter 97 and Tut.by. By 2014, according to the Belarusian scholar Oleg Manaev (2014), the country had evolved two distinct media systems: one, tightly controlled by the state, and serving the "common majority" of citizens, and another, hounded but resilient, serving an "advanced minority," and both reinforcing the internal cohesion of each media community (see also Manaev, Manaeva, and Yuran 2013). As the penetration increased of the internet in general and online social media networks in particular, social media became critical both to independent journalists and to civil society activists, leading to a blurring of the boundaries between the two (Pospieszna and Galus 2019).
By the time of the 2012 parliamentary elections, media crackdowns -including detentions, threats of death and violence, and the blockage of influential independent news websites -had become a familiar part of the pre-election ritual (RSF 2012). Between elections, too, authorities were active. Thus, in 2017-18 -non-election years that saw protests over taxes and jobs -authorities fined some 100 journalists (RSF. 2018b) and blocked access to Charter 97 and Belaruski Partyzan (RSF. 2018a).
Media restrictions were ramped up during the COVID-19 pandemic, as government officials accused independent journalists and social media influencers of "spreading rumors" and "sowing hysteria," leading to raids, detentions, fines, and bullying (BAZh. 2020; Belta. 2020; Nasha Niva 2020; RSF. 2020a). These restrictions bled into the traditional pre-election media crackdown, which was already underway two months or so before the August vote; by the end of July, some 40 journalists had already been arrested (RSF. 2020b). In the 11 months after the August 2020 elections, Belarusian authorities detained some 471 journalists, levied fines against 79, and beat or tortured 69, according to Reporters Without Borders (RSF. 2021c). Nearly a year after the elections, and well after most protests had subsided, authorities forced a Ryanair jet bound from Greece to Lithuania to land in Minsk, in order to detain Raman Pratasevich, the editor of the influential Telegram channel Nexta (RSF. 2021b). That same month, authorities succeeded in shuttering the most popular independent news website, Tut.by, although the remaining team quickly re-established itself as Zerkalo (RSF. 2021a).

Media, protest, and polarization
Studies of the US have provided ample evidence of the ability of news media to alter individuals' awareness of and depth of knowledge about politically salient issues, contributing to partisan sorting and polarization of patterns of media consumption (Morris 2005; DellaVigna and Kaplan 2007)including evidence that more avid consumers of Fox News, for example, may tend to reduce their consumption of newspapers, where they are more likely to encounter dissonant messages (Schroeder and Stone 2015). This shifting informational diet, moreover, is driven not only by cognitive factors (i.e. wanting to consume messages that add to or at least do not contradict one another), but also by factors of identity and emotion, as people begin to associate themselves with a particular community of perception and interpretation, and to distrust those who perceive and interpret differently (Levendusky 2013a(Levendusky , 2013b. It is not, moreover, a story of only one news channel, or only one side of the partisan divide in the US: politically sorted media ecosystems comprise networks of resonating and reinforcing messaging on both right and left (Baum and Groeling 2008). Similar partisan ecosystems have also been found outside the US, particularly in Europe, although evidence there has focused primarily on the far right (Hefft et al. 2019). There appears to be some differentiation in terms of who is affected by this sorting. While evidence is mixed about whether the media is making Americans more partisan on the whole, research suggests that it is making already partisan people more isolated from alternative views, and pushing non-partisan people out of political debate and participation (Prior 2013).
The sorting and polarizing impact of media has also been shown to occur at the community level, and not just at the individual level, as messages saturate the social milieu in which people express opinions and make decisions. The effects of partisan media have been shown to shape the views and behaviors even of those who do not read or watch such media, as messages are passed on through word of mouth (Druckman, Levendusky, and McClain 2017). Further, research in the US has shown that the prevalence in local viewership ratings versus other cable news channels of Fox News -with its "skeptical" coverage of the COVID-19 pandemic -was associated with reduced observance of "stay-at-home" guidance from local, state, and national health authorities (Ananyev, Poyker, and Tian 2021).
If most of our theory on media polarization comes from the US and Western Europe, then much of what we understand about media and protest is derived from examples closer to Belarus -particularly, large-scale protest movements in both Ukraine and Russia.
In arising to prevent the institutionalization of autocracy via the anti-democratic handover of power between Leonid Kuchma and Viktor Yanukovych, Ukraine's 2004 Orange Revolution succeeded in using symbols of culture and identity to forge a "civic" revolutionary constituency more committed to what they were mobilizing against, than to a proactive agenda for governance (Beissinger 2013). The subsequent Euromaidan in 2013-14 mobilized an even larger and broader coalition -in many ways, more diverse and disparate than the Orange coalition of a decade earlier, and with less consonance in the understanding and framing of grievances -again with reference to symbols of culture and heritage as rallying points (Kulyk 2016;Onuch and Sasse 2016). Indeed, the centrality of these symbols was sufficient not only to create support for the revolution, but to make identity itself more central to individuals' relationship to politics (Pop-Eleches and Robertson 2018).
By the time protests emerged in December 2011 to oppose Vladimir Putin's return to the Russian presidency the next year, Russia had effectively developed two competing media ecosystems: one dominated by the state and centered around television, and one independent, and centered around the internet. In 2010, Sarah Oates found that people who got more of their news online and less on television had lower levels of trust in national and regional government (though not necessarily Putin himself), higher levels of dissatisfaction, and a greater sense of their own political efficacy (Oates 2013). Researchers have found a significant -if small -relationship between the use of online social media networks like Facebook and participation in the 2011-12 "Bolotnaya" protests in Russia (Enikolopov, Makarin, and Petrova 2020). The boundary should not be overstated, however, and by the time Russia went to war with Ukraine and embroiled in geopolitical confrontation with Europe and the US, the Kremlin had learned to push its "strategic narratives" just as effectively online as offline (Oates 2016).
In Russia, then, the relationship between the internet and anti-regime mobilization has remained one of affordance, rather than causation: the internet allows people to gain access to alternative information, but does not cause them to do so; the internet allows people to communicate and to solve resource mobilization tasks, but does not cause them to do so; and so on (Smyth and Oates 2015). In her work drawing together the Ukrainian and Russian cases, Tanya Lokot writes of online media fostering the development of what she calls "augmented dissent," in which "older and newer logics, offline and online activities, forms of participation and protest tactics exist simultaneously, and extend into each other in complicated ways instead of simply replacing one another" (Lokot 2021, 5). In particular, Lokot identifies five key "affordances" provided by online media, including social media, whose exploitation was key to the success of the 2014 Euromaidan: ephemerality, flexibility, visibility, scalability, and persistence. In this view, then, it is the media themselves that are the message. Bennett and Segerberg (2012) argue that while the kind of mobilization Lokot describes may outwardly resemble the kind of collective action traditionally studied by scholars of social movements, the affordances of social media have helped give rise to a new kind of "connective action," in which broadly inclusive mobilizational frames and low barriers to action encourage individuals to mobilize as an act of communication, first and foremost. Further, Natalie Fenton (2008) argues that media can contribute to political mobilization not by altering individual consciousness, so much as by fostering a collective consciousness, in which the boundaries and horizons of political possibility can be reimagined. Moreover -and, for activists, this is the point -this initially isolated space can begin to force changes to the information agenda in the broader political community (Freelon, McIlwain, and Clark 2016).
All of this leaves the question of the role of the media in mobilization somewhat ambiguous. Are media merely a source of affordances, allowing aggrieved communities -constructed independently of the media that help them communicate -to do things that they might not otherwise have been able to do? Or are media themselves at the heart of the mobilizational process, structuring the lines around which citizens divide? These are the questions this paper seeks to put to the Belarusian case.

The data
This paper draws on an original survey, conducted in Belarus in September 2020, involving 1,276 respondents. The survey was delivered online, and the sample was stratified by age, gender, and geography, in an effort to approximate national representativeness. In most categories -with the partial exception of the upper and lower age brackets -the sample is very close to national statistics; because point values are not relevant to the arguments made here, and because there is no theoretical reason to expect minor differences in age distribution to affect the relationships between key variables, the statistical results reported in this article are not weighted (see Table 1).
The survey instrument focused primarily on media consumption habits and related issues. Among the measures of interest here are: (a) To measure individuals' media consumption habits, respondents were given a list of 11 prominent media outlets and asked to indicate the frequency with which they turned to them for news coverage. If a respondent reported turning to a given outlet for news at least once a month, the variable was coded as 1; otherwise, it was coded as 0. (b) To measure individuals' preferences for independent vs state-linked media, we did not ask directly about their orientation. Instead, we counted the number of independent and prostate media outlets in their media diets and assigned a preference for independent or state media depending on which category predominated in their responses. In the data reported here, respondents who showed a preference for independent media were coded as 1; others were coded as 0. 1 (c) To measure individuals' interest in various topics, we asked about the frequency with which they discussed a range of topical categories with their social circles. In the analysis here, each topic was coded as a separate dummy variable, with a value of 1 if the respondent reported discussing it regularly with family, friends or colleagues. (d) To measure individuals' sense of identity, we asked how important it was for them to be part of or associated with the Belarusian state, Belarusian culture, Russian culture, and the union state of Belarus and Russia. Responses were coded on a five-point scale, from a strong to a weak sense of identification. (e) To measure individuals' language preferences -also closely linked to identity -we asked individuals to indicate whether they preferred to speak Belarusian, Russian, or another language in a range of settings; the emphasis in this paper is on language preferences for public settings, although the effects are largely the same for language preference at work and at home. (f) In measuring political opinions, we were barred by Belarusian legislation and the polling company from asking direct questions about support for Lukashenka or any of his policies, as well as about opinions regarding the government's political or economic performance, or the protests themselves. Instead, political sentiment here is measured in two ways: (a) Respondents were given a long list of issues -ranging from unemployment and economic growth to battling COVID-19 and respecting the rights of ethnic minorities -and asked them to indicate which ones they felt were the top three issues facing the country. Among that list, we included two that can be linked to current political sentiment: the defense of freedom of the press and expression, and the pursuit of constitutional reform. Responses were coded into two dummy variables, one for including media freedom in the top three issues, and one for constitutional reform. (b) We asked respondents to report the emotions they felt when thinking about the COVID-19 pandemic in Belarus, including a range of positive emotions (pride, hope) and negative emotions (anger, sadness, anxiety, fear). Responses were coded on a five-point scale, from most to least intense.

Factors of political opinion in Belarus
To begin to address the role of media in shaping political opinion and behavior in Belarus, the first task is to explore a fuller range of variables that might be expected to correlate with politics. As discussed above, these include, in addition to media consumption choices, socio-demographic variables, identity, and individuals' general level of interest in politics, economics, and other salient issues -all of which might either be expected to shape political opinion directly, or to shape media consumption decisions en route to shaping political opinion. Thus, as an initial step, the article presents a series of simple regression analyses. Before launching into these, however, two caveats are in order. The first is that the aim at this stage is not to assess causation, but merely to suggest what relationships may or may not be statistically meaningful, and thus worth exploring in more detail. The second is that the survey, as mentioned earlier, does not actually contain a direct measure of political opinion. To maintain the safety of both our colleagues and our respondents, we were unable to ask about people's opinions about Lukashenka, about their past votes or future voting intentions, and certainly not about participation in protests. What stands in for political opinion is whether or not the respondent believed that the freedom of the press and/ or constitutional reform were among the top three issues facing the country -a rough, if problematic, proxy of how people felt about the political contestation, coercion and violence submerging the country at the time we conducted the survey.
As controls, the analyses include: (a) Age, commonly associated with both media consumption and protest participation; (b) Gender, commonly associated with regime support and media consumption, coded as 1 for male and 2 for female; (c) Education, commonly associated with regime support and media consumption, coded into six categories, from unfinished secondary education to advanced degree; (d) Region, coded categorically for the capital city, Minsk, and each of the six oblasts (see Matteo 2022). Given the somewhat larger protest turnout in the capital and in the West, a categorical coding is preferable to a simple dummy for Minsk; and (e) Employment, coded categorically for 11 sectors, with varying degrees of domination by the state, which has been shown in Russia and other authoritarian contexts to impact political behavior (see, for example, Frye, Reuter, and Szakonyi 2014).
Presented below, then, are two sets of logistic regressions, one for each of the two dependent variables: the importance placed by respondents on media freedom and constitutional reform, respectively (see Table 2). Each set contains four models: (1) Measuring only the effects of socio-demographic factors, including region, education, employment, gender, and age; (2) Combining the effects of a preference for independent media and the socio-demographic factors, controlling for gender and age; (3) Combining the effects of a preference for independent media, the socio-demographic factors, and respondents' discursive interests, controlling for gender and age; and (4) Measuring the effects of a preference for independent media and identity constructs, controlling for gender and age.
While there are subtle differences in the results for the two different dependent variables, which will be discussed in more detail below, the general patterns are the same. In all specifications, a preference for independent media has a large and highly significant positive correlation with believing that either freedom of the press or constitutional reform are among the most pressing issues facing Belarus. Of the socio-demographic factors, region and education are also correlated when media preference is omitted from the analysis (Model 1), but these correlations decrease or -in the case of education -fall away altogether when media preference is added to the model (Models 2 and 3).
Beyond that, differences between the dependent variables begin to emerge. The only discursive interest positively correlated with an interest in freedom of the press is politics (Model 3). This same correlation is even stronger for constitutional reform, but it is accompanied by an interest in food, and a lack of interest in the economy and in sports. The differences in the effect of identity constructs is even more stark. An identitarian attachment to Belarusian culture -and a disattachment from the Belarusian state and the union state with Russia -are strongly correlated with opinions on freedom of the press, but language preference is not (Model 4). For opinions on constitutional reform, there are weak correlations with an attachment to Belarusian culture and a preference for Russian (as opposed to Belarusian).
While it is hard to know for sure, the discrepancies between the behavior of the "free press" and "constitutional reform" variables may have their roots in the ambiguity of the question itself, and of the way in which the issue of constitutional reform played out in Belarusian public debate at the time. Thus, it is possible that some oppositionally minded respondents felt that Belarus's problem was not primarily with its constitution, but with the autocrat who abused and manipulated it. At the same time, Lukashenka himself announced a constitutional reform process, which many saw as a smokescreen designed to allow him to buy time and ensure his survival in power. This, too, may have soured many opposition-minded respondents on the idea, and those who did not want to see Lukashenka go might actually have seen .024** (.008) .026*** (.009) .019*** (.005) .036*** (.008) .042*** (.008) .039*** (.009) .040*** (.006) Notes: Logit. Standard errors in parentheses. * p < 0.05; ** p < 0.01; *** p < 0.005.
constitutional reform as being to their liking. This might also help explain both why a preference for independent media yields a smaller coefficient for the constitutional reform models, and in the fullest specification of the model decreases sharply in significance. By contrast, a statement about the importance of protecting the freedom of the press is considerably less politically ambiguous.
To reiterate an earlier point, only very limited inferences can be made here. Given that all of the variables were measured at the same time -and given the necessity of approximating key variables -I cannot and do not claim that media choice or any of the other variables are causing political opinion (though that is not an implausible conclusion). Rather, the ambition is to demonstrate the relative closeness of media consumption to political opinion in comparison to other possible explanators.
The key finding, then, is this: In every specification of both sets of models, a preference for independent media emerges as the largest predictor -and indeed, the only consistent predictor apart from age -of respondents' political opinions.

Factors of media consumption in Belarus
Before delving into the question of what might be shaping media consumption -and thus having an impact on political opinion and behavior -it is worth exploring the general patterns of media consumption in Belarus. Despite the formal dominance of the state in Belarusian media -and the hounding, prosecution, blockage, and partial dismantlement of independent news outlets and their journalists -some 54.3% of survey respondents consumed primarily independent media, 29.4% consumed primarily state-linked media, and 16.3% consumed an equal measure of each (including those who consume no news media at all). Within each cohort, respondents tend to read, watch, and listen almost exclusively within their implied political preference group. Thus, 95.9% of respondents who preferred independent media reported using the oppositional news website Tut.by, 69.8% used Onliner.by, and 52.2% used Belsat, while the most popular state-linked outlet among the same cohort was the television station ONT, with 14.6%. Respondents who preferred state-linked media were, by contrast, a bit more omnivorous, perhaps reflecting the general dominance of independent media in the public discussion at the time the survey was conducted. Thus, 93.3% of those who preferred state media reported watching ONT on a regular basis, 89.0% watched Bel 1/2, and 78.5% watched STV, but 52.3% also said they read Tut.by, and 29.7% said they read Onliner.by (see Table 3).
To measure the polarization of media audiences in Belarus, I adapted a simple but effective method deployed by Fletcher, Cornia, and Nielsen (2019), who sought to understand the relative polarization of political opinion among media audiences in a set of democracies. To do that, Fletcher and colleagues surveyed people in 12 countries, asking them about their preferences for up to 30 media outlets in each country from across the political spectrum, as well as to place themselves somewhere on a traditional left-right spectrum. The median left-right placement for each media outlet's audience is then plotted along a horizontal axis, with the diameter of each media "dot" determined by the relative size of the outlet's audience. In their findings, the US came out as most polarized, followed by the UK, while Ireland and Germany were at the bottom of the list. While the method for measuring media consumption preferences used here is broadly identical (albeit with 16 media outlets, rather than 30), placing respondents on a political spectrum in Belarus is trickier. This is in part because Belarusian politics -like authoritarian politics in general, and post-Soviet politics in particular (see Brader and Tucker 2001) -does not fit neatly onto a left-right spectrum. The task is made more complicated, moreover, by the fact that polling agencies are not allowed to ask directly political questions, and so I could not measure sentiment for or against the regime, sentiment on political or economic performance, or attitudes towards the post-election protests. Instead, Figure 1 presents polarization along four different axes: (a) Belief that freedom of expression is among the three most pressing issues facing Belarus (ranging from 0 to 1); (b) Belief that constitutional reform is among the three most pressing issues facing Belarus (ranging from 0 to 1); (c) Emotional response to the government's handling of COVID-19 (an index of positive and negative emotional responses, including hope, pride, sadness, anger and anxiety); and (d) Revealed preference for state vs. independent media (calculated by subtracting the number of independent media outlets in a respondent's reported media diet from the number of state-controlled outlets).
While each of these measures is problematic on its own, taken together they can provide a picture of the degree to which the audiences of state versus independent media in Belarus are polarized in their political priorities, their interpretation of government performance, and their general media consumption habits and preferences. On all four measures, respondents were at least as polarized as -and in most cases more polarized than -American media audiences in the study by Fletcher Cornia, and Nielsen (2019). Broadly, a preference for independent media is correlated with a higher level of identification with Belarusian culture, and a lower degree of identification with the Belarusian state and the Russo-Belarusian union state. As Table 4 makes clear, however, this general pattern masks a high degree of diversity. There are thus independent media that are positively correlated with a sense of affiliation to the Belarusian state (though not the union state with Russia). While all independent media outlets are correlated with a sense of affiliation to Belarusian culture, some state-linked media are as well, while others are not. And while all state-linked media are correlated with a general preference for speaking Russian in public places, some -but not all -independent media are correlated with an expressed preference for speaking Belarusian.
In terms of interests, respondents who reported that they frequently discussed politics and the economy with their friends, family members, colleagues, and social-media interlocutors were more likely to prefer independent media as were, though to a much smaller degree, respondents who frequently discuss family, children, and romance. Respondents primarily interested in work, career, and personal finance, on the other hand, were slightly more likely to prefer state-linked media. But aside from politics and economics, there are no striking correlations between individuals' media preferences and their discursive interests. There are somewhat clearer socio-demographic correlates of Belarusians' media preferences. Younger respondents and respondents with higher levels of educational attainment were much more likely than older ones to prefer independent media, and vice versa. Geographically, independent media were most popular in Minsk (62.8%) and Brest Oblast (52.5%), and least popular in Vitebsk Oblast (39.9%) and Mogilev Oblast (41.3%). A preference for independent media predominated in every employment sector, regardless of the level of state involvement. With the exception of respondents aged 55 and above, however, a preference for independent media outstripped a preference for state media in every socio-demographic cohort in the survey.
We have, then, a picture of two distinct media ecosystems in Belarus and a large degree of polarization and sorting, such that most people who consume mostly independent media consume little or no state media, and vice versa. And while the political divide between the two media ecosystems is clear, most other dividing lines are not. Despite patterns of propensity -around age, education, an interest in politics and, to some extent, geography -there are no categories into which independent Belarusian media fail to reach, and there is indeed only one (people aged 55+) in which they do not outstrip state media. It is, meanwhile, also true that there are no categories from which state-linked media are entirely or even mostly shut out. Moreover, there is no identity construct covered by the survey that cannot find a home somewhere in the independent -and currently oppositional -media landscape. (In this respect, the obverse is not true: state media do not cover the entirety of the identitarian waterfront in Belarus, as their reported readership/ viewership lacks citizens who feel only weak connections to the state.) This is not, then, a picture reminiscent of the left/right divide that characterizes media polarization in the US, or the subtler identitarian divides that characterize media consumption in Ukraine. It is, in some ways, more reminiscent of the structure of the media landscape in Russia -a similarly authoritarian country with state-dominated media -but with the caveat that the relative audience shares of state and independent media in Russia are roughly reversed. To understand what (if anything) this means, we will need to look more closely at how media and other factors come together to shape political opinion.

Analysis
We have already seen that media preference is among the factors most consistently and powerfully correlated with political opinion in Belarus. We have also seen that age, education and, to a less consistent extent geography, are important predictors of both politics and media consumption (while gender correlates with media consumption, but not politics). This presents an important challenge to inference: with cross-sectional data, we struggle to determine the extent to which media choice is independently shaping political opinion, altering the relationship between other variables and political opinion, or merely standing in for those other variables in our observations.
To test the independent impact of media choice, I deploy the simple mediation model developed by Hayes (2009Hayes ( , 2018. Mediation analysis allows us to see the extent to which a respondent's preference for state versus independent media intervenes between other factors that might cause political opinion -such as gender, age, geography, and identity -and political opinion. More precisely, it allows us to test for a specific kind of intervention, in which the prior variable (say, age) first affects media choice, and then media choice affects political opinion, while controlling both for the direct effect of age on political opinion, and for a range of other potential causal variables. The mediation effect is distinct from a moderation effect, which does not involve a causal relationship between, in this case, age and media choice. Because we can see a strong correlation between age and media choice, for example, moderation is inappropriate. Mediation is also different from an interaction effect, which allows for a causal relationship between age and media choice but does not presuppose the direction of causality. Because the relationship between age and media choice can only run in one direction, "naïve" interaction effects are also inappropriate.
On a more technical note, the Hayes "Process" model for mediation corrects for shortcomings in earlier approaches (Baron and Kenny 1986;Sobel 1986) in two important ways: first, it measures the indirect, "mediated" effect directly, rather than inferring it from other measurements; and second, it uses boostrapping (in this analysis, with 5,000 iterations) to avoid the requirement for a normal distribution of the mediating variable, providing a more robust confidence interval (in this analysis, 95%). One drawback, however, is the inability to estimate the "total effect" -nominally, the direct effect plus the indirect effect -when the dependent variable is dichotomous, as it is in this analysis. Thus, while the general magnitude of the total effect can be inferred from the direct and indirect effects presented here, it cannot be measured.
The results of the mediation analysis are presented below. In Table 5, we have the correlations of the key independent variables identified earlier (age, gender, geography, education, and identity) both with the mediating variable (media choice) and the two independent variables (preferences for freedom of the press and constitutional reform). 2 As before, age, gender, region, and education all correlate with media choice, while age and education correlate with both measures of political opinion, and region correlates with opinions on freedom of the press and expression but not constitutional reform; the core measure of identity -the sense of belonging (or non-belonging) to the Belarusian state -correlates with none of the variables. Table 6 presents the mediation analysis as such, repeating the direct effect of each of the variables on both aspects of political opinion, and the bootstrapped indirect effect, as mediated through media choice. (Each row in the table represents a separate mediation analysis, with the variable in the first column as the independent variable, and the other variables used as covariates.) The mediation effect exerted by media choice is significant and sizeable across all combinations of independent and dependent variables, with the exception of identity; in some cases -in particular gender and geography -the mediated relationship is significant, while the direct relationship is not. Unfortunately, the dichotomous nature of the dependent variables means that a reliable calculation of the total effect of the models is impossible, as is an estimation of the proportion of the effect that is mediated. A rough calculation, nonetheless, suggests that between one-third to two-thirds of the effect of the independent variables on political opinions is mediated through media choice. All caveats considered, the overall implication of the mediation analysis is clear enough: in addition to the independent causal effects identified earlier, media choice significantly alters and strengthens the effect of causally prior variables on political opinion in Belarus.

Discussion and conclusions
The fact that media matter in Belarusian politics is not in dispute. Indeed, as Lukashenka prepared for his re-election, he and his government worked hard not only to disrupt the political opposition per se, but to disrupt the country's independent media ecosystem. This included arrests, raids, site blockages and, for a period of time, even the wholesale shutdown of most internet service provision. If he hoped that would turn people away from independent media and back towards the state's own news media, however, he was mistaken. Some 53% of our survey respondents indicated that they had significantly changed their media consumption habits over the three months prior to the survey. The biggest winners from that process were Telegram and Tut.by, while the biggest losers were ONT and Bel 1/2. "We have a lot of internet portals that have been either banned or blocked, so they're hard to get to," said "Vadim," one of the Belarusian citizens we interviewed in-depth alongside the survey. "We have a lot of those. And they're [my] main sources of information." And "Vadim" was not an outlier. None of the 24 interviewees reported turning towards more state and television news. Indeed, most made a point of telling us that they had turned away from state media: • "I ignore state media as a matter of principle. No state newspapers, internet, or TV. It wasn't always like that, but it is now. Because I've . . . had enough." • "After the elections, I started ignoring any and all state media, as a matter of principle." • "I won't look at any pro-government media as a matter of principle." In sum, then, the clearest and most consistent dividing line in the Belarusian uprising of 2020 was associated not with citizens' socio-economic or demographic status, not with their senses of interests and identity, but with their choice of media -a choice that itself does not seem to be clearly and consistently associated with socio-economic, demographic, interest or identitarian factors. That, in turn, suggests that even in authoritarian contexts such as Belarus analysts should perhaps pay closer attention to the independent agency of media and media consumption choices, an agency political scientists more frequently tend to associate with democracies. This is not to say that Tut.by is the Fox News of Belarus, but it is to say thatmuch as the division between left-wing and right-wing media ecosystems has become a durable and autonomous structural element in American politics -so has the division between independent and state-linked media ecosystems become a durable and autonomous structural element in Belarusian politics.

Notes
1. While the reduction of this measure to a dummy variable involves a loss of granularity, it is in many ways the cleanest measure of media preference. Using nominal counts -while providing for more variation and granularity -would conflate both media preference and the volume of media use (i.e. the number of outlets consumed) into a single variable. Using a dummy variable avoids the potentially problematic assumption that a respondent who consumes more media is politically different from one who consumes less. 2. All of the variables are defined and measured in the same way as in Table 2, with the exception of media choice.
Because mediation analysis cannot accommodate a dichotomous mediating variable, it was necessary to use a broader measure -in this case, the number of state-controlled media outlets in a respondent's diet minus the number of independent outlets. A negative number thus reflects a preference, on balance, for independent media, while a positive number reflects a preference for state media.