“All of Belarus has come out onto the streets”: exploring nationwide protest and the role of pre-existing social networks

ABSTRACT During moments of nationwide mass mobilization, what distinguishes the towns and cities that rise in the first week from those that do not see protest? Taking the case of nationwide protests in Belarus in August 2020, this study employs an original protest event catalogue to investigate what drives mobilization in early-rising localities. I test hypotheses in the protest literature relating to whether pre-existing social networks, or pre-election campaign rallies, influence subsequent protest mobilization. The innovative use of Telegram data demonstrates the platform’s value for social scientists studying protest. My results suggest that pre-existing social networks help drive mobilization in localities by facilitating communication, coordination, and engagement prior to protest onset, priming people to be ready when the moment of protest arrives. This article also highlights the impressive scale of nationwide mobilization in Belarus in 2020, and demonstrates that local networks were engaging in widespread opposition activity even before mass mobilization.


Introduction
When mass protest takes place, it is usually limited in geographic scope -typically occurring in capital cities, or large urban centers. However, in some cases, protests are widely dispersed across a given country. When this happens, certain towns and cities will rise early, while others will only mobilize later, once protest is established elsewhere. 1 During such moments of mass mobilization, what distinguishes these early-rising localities from those that do not see protest? Social scientists are divided as to what shapes the geography of protest. Local population size, transport and communication networks, and socio-economic factors have all been highlighted (Andrews and Biggs 2006;Brooke and Ketchley 2018). Meanwhile, some studies emphasize the role of the local political context in shaping where protest occurs (Robertson 2011;Lankina 2015;Lankina and Voznaya 2015). Other research foregrounds the centrality of social networks 2 to the mobilization process (Gould 1991;Tufekci and Wilson 2012;Onuch 2015;Jost et al. 2018;Thomson 2018;Tucker et al. 2018). The mass protests which erupted in Belarus in 2020 are one such instance of geographically dispersed protest, providing the opportunity to examine this puzzle.
From 9 August 2020, in response to extensive reports of electoral fraud, Belarusians took to the streets of towns and cities in every region of their country. Protests emerged rapidly across Belarus, with over 100 localities protesting in the first week. While the geospatial dispersion of these protests was unprecedented, not every town mobilized. In fact, some localities saw protest, while other places in their district, otherwise similar in many respects, remained quiet. Why did certain localities see protest mobilization in the first week, while others did not? Are these varying outcomes explained by differences in contextual factors, such as population, wages, or transport networks, as some scholars have suggested? Alternatively, might they be down to different political contexts, enabling preelection campaign rallies? Or, given that neighboring towns are otherwise similar in many of these respects, do pre-existing social networks, enabling communication, coordination, and engagement in advance of protest, play a role? This article seeks to answer this puzzle.
Employing an original protest event catalogue curated by the author, and utilizing multivariate logistic regression, supplemented by digital ethnography, I conduct analysis at the individual locality level for all Belarusian towns and cities with populations over 5000. 3 I deploy and test sociological theories about the role of social network ties in mobilization. In doing so, I also assess empirical propositions in public and academic discourse about the role of social network ties in facilitating (or not) recent nationwide protest in Belarus specifically. I first map the extent of regional protest from 9 to 15 August 2020. I then run multivariate logistic regressions to test for the relationship between the presence of pre-existing social networks and the presence or absence of mobilization in a locality during this period. I also test for alternative explanations centered on local political context, and control for relevant factors including population, local wages, and transport networks. Further qualitative analysis of the activity of local social networks, traced via community Telegram chats, enables me to elucidate ways in which these networks enabled communication, coordination, and engagement at the local level prior to the elections.
This article provides robust evidence to support the hypothesis that pre-existing social network ties are crucial to and drive mobilization at the local level. Specifically, I find that in localities where social networks pre-dated the start of protests on 9 August, mobilization was more likely to occur in the first week of protests. To conduct this analysis, this article focuses on social network ties captured through online (specifically Telegram-based) discussion and coordination. This approach makes this article highly innovative, as it is one of the first to use Telegram for protest event data and analysis. At the same time, the reliance on Telegram as a data source also means that the claims made here about pre-existing social networks must be cautious, as these data do not capture social network ties and social movement organization networks that may exist "offline." And thus, while we must remain cautious in our interpretation, this fact does not undermine the article's main argument, as there is a clear link between local communication, coordination and engagement in advance of the elections, and first-week mobilization. While further, offline data collection and identification of local social networks are necessary, this study is ground-breaking in its identification of geographically widespread networks engaging in opposition in advance of the 2020 protests. As such, this article will form the basis for further study of local networks in Belarus, and their role in protest.
In the sections that follow, I first map regional protest between 9 and 15 August 2020, and examine how this Belarusian case allows me to leverage key expectations in theory on the role of social networks in mobilization. I then present an analytical framework deriving theoretical expectations about how social networks may facilitate local mobilization and early-rising protest, as well as a competing hypothesis centering on local political context and pre-election opposition campaign rallies. After outlining my data sources and analytical approach, I present my results. The article finishes with a conclusion discussing my findings and highlighting avenues for further research.

The case of early-rising protest localities in Belarus
The question of what drives regional variation in mass mobilization is an enduring puzzle in social science, and Belarus represents a particularly fascinating case of this phenomenon. The Belarus 2020 protests are a critical case of nationwide mass mobilization, occurring in an authoritarian state, where there is limited past protest (De Vogel 2022), criticism of the government is high-risk (Bedford 2017;Freedom House 2019), and protest is punished (Herasimenka 2017). And yet, the protests of summer 2020 were widespread and cross-national from their onset, with towns and cities mobilizing from the very first day. On 9 August, when the incumbent President Alyaksandr Lukashenka claimed victory in fraudulent presidential elections, at least 19 locations saw protest, including many of Belarus's largest cities, but also smaller towns such as Lahoysk and Lyakhavichy (with respective populations of 15,000 and 10,700). Within a week of the elections, over 100 localities mobilized, contributing to a truly nationwide protest wave (see Figure 1). I found that in a minimum of 27 localities, over 1% of the local population mobilized in a single event. A further 21 localities saw at least 0.5% of their population protest (see Figure 2). 4 The most common protest tactic was "chains of solidarity," where protesters stood single-file, facing a busy road, sometimes holding red and white flags, flowers, or banners. Marches were also held, particularly where protester turnout was higher. When security forces targeted protesters, marches dispersed, and then small groups would re-gather and keep moving. Toward the end of the first week, rallies and meetings were held in some localities. Protesters gathered on central squares, outside government buildings, demanding to speak to their mayor or local officials. Nevertheless, although this wave of protest activity was unprecedented in scale, as evidenced in Figure 1, dozens of towns and cities remained quiescent, with no protest at all.
This widespread mobilization did not emerge from nowhere on election day. Awareness of and engagement with the upcoming elections was happening throughout Belarus in 2020. Using Telegram as a tool, I identified discussion and coordination around opposition to the Lukashenka regime in advance of the elections in dozens of towns and cities, pointing to the presence of pre- existing social networks throughout the country. In the weeks and months preceding 9 August, networks of local citizens and activists were discussing and engaging with political opposition activity -in some cases even organizing protest events criticizing Lukashenka's government and the repression of opposition figures. The campaign for the main opposition candidate, Sviatlana Tsikhanouskaya, also attempted to rally support throughout Belarus, organizing nationwide campaign events (Viasna 2020). Tsikhanouskaya and her team extensively toured Belarus between 19 July 2020, when her candidacy was confirmed by the Central Electoral Commission, and election day.
So, can these pre-existing networks and campaign activities in advance of the elections help explain why some localities mobilized during the first week of protest, and others did not? While there is some discernible variation in first-week mobilization among the regions of Belarus, what is of particular interest are within-region, and even within-district differences in whether towns mobilized. For example, Iwye, a 30-minute drive from Lida in western Belarus, experienced protest in the first week -but Voranava, of a similar size, and roughly the same distance from Lida, did not. And Kilmavichy and Kastsyukovichy, two towns in eastern Mahilyow region, have roughly the same population size (16,000-17,000), are situated just 36 km apart, and yet the former mobilized while the latter did not. What accounts for these differences in early-rising protest, in localities which otherwise appear similar in terms of demographics and region? In addressing this puzzle, this article not only seeks to improve our understanding of the Belarusian protests specifically, but also to contribute to the study of nationwide protests more generally.

Pre-existing social network ties hypothesis
Scholars have long focused on the centrality of social network ties to the mobilization process (Snow, Zurcher, and Sheldon 1980;Gould 1991;Diani and McAdam 2003). Many studies assert that during moments of protest, participants are mobilized by their social network ties (McAdam 1986;Tufekci and Wilson 2012;Youmans and York 2012;Onuch 2015;Tucker et al. 2018). However, social networks are important not only in the moment of mass mobilization, but also form a crucial foundation in advance of mobilization, helping to explain why some localities may rise first in protest.
Pre-existing social networks, be they grassroots organizations, formal or informal social movement organizations, or simply networked publics of ordinary citizens, are an important precondition for early-rising mobilization in a locality (Gould 1991;Wackenhut 2020). This is because these networks enable communication, coordination, and engagement in advance of protest onset. Communication between members of the network allows information to flow via trusted sources, ensuring that the network stays up-to-date and informed on relevant events -including grievances triggering the outbreak of protest (Klandermans 1997;Pamela and Myers 2003). Coordination is then possible via the tools the network uses to communicate, enabling the organization of joint action in the locality when the moment of mobilization arises (Tarrow 2011). Pre-existing networks also foster engagement: network members are already involved in some kind of shared discussion or activity with others in their local area, priming them for further action (Onuch 2015). Given all this, I argue that where a pre-existing social network is present in a locality, this represents a collective who are informed and receiving up-to-date information, have the tools and ability to coordinate joint action, and are already engaged and active, maybe even having mobilized before. This puts them in an ideal position to rapidly mobilize when the moment of protest arrives -in the case of Belarus, on 9 August 2020. Hence, I anticipate the following hypothesis, that: During the first week of mobilization, protest is more likely to occur in localities where pre-existing networks are present.

Competing hypothesis: political opportunities
However, some scholars argue that geo-spatial variation in mobilization is most significantly shaped by variation in local political context -sometimes referred to as political opportunities (Jasper and Goodwin 2011;Robertson 2011;Lankina 2015). This scholarship would suggest that protest is more likely to take place in towns and cities with more favorable political opportunities for mobilization (Eisinger 1973;Tarrow 1996). While fraudulent elections, like those in Belarus, represent a political opportunity for a nation as a whole (Beissinger 2007;Tucker 2007), political opportunities can also vary across a country (Lankina 2015). For example, in localities where elite allies are present, they can provide resources and support for protesters, enabling mobilization; and where local elites are divided, protesters may take advantage of these divisions to gain the backing of competing elites (Robertson 2011;Tarrow 2011). Favorable political opportunities are particularly important for the "early risers" of protest: McAdam (1995) argues that early risers identify and capitalize on expanding opportunities, whereas later joiners are more influenced by diffusion processes. Therefore, political opportunities are particularly important for those mobilizing early-on.
In Belarus, the authoritarian state allows for limited overt political competition, so elite allies or divisions may be challenging for researchers to measure. However, the series of rallies in support of the Presidential candidate Sviatlana Tsikhanouskaya organized in advance of the elections provides the opportunity to capture some variation in local political context. The presence of a pre-election rally in a locality may represent more favorable political opportunities for later protest, as local elites permitted the event, and local residents turned out in support. Thus, in my analysis I will test for the following competing hypothesis, that During the first week of mobilization, protest is more likely to occur in localities where campaign rallies for the opposition candidate took place prior to election day.
Nevertheless, I argue that where the political context of a locality may be more favorable for protesters, this does not necessarily mean that there is a group of primed citizens ready to mobilize locally when the opportunity arises. For example, local elites may have allowed the Tsikhanouskaya team to organize a campaign event in a city, but this does not necessarily mean that local grassroots networks are present that will be ready and willing to initiate protest following the elections. As such, although favorable local political opportunities may facilitate protest, I postulate that the presence of pre-existing social networks in a locality is of greater importance, as without these no mobilization can occur.

Contextual controls
Several additional contextual factors may also influence regional variation in mobilization, and should be controlled for in my analysis. First, transport networks enable information, resources and people, and thus, mobilization, to travel across space, and should be accounted for (Hobsbawm and George 1969;Aidt, Leon, and Satchell 2017;Brooke and Ketchley 2018). Consequently, I control for the presence of strong transport networks in a city. Second, numerous studies argue that the main participants of pro-democracy protests are wealthier, middle-class citizens (Norris 2002;Daron and Robinson 2006;Ekiert 2010). They have more resources (time, money, social capital) enabling them to participate in politics, including protest (Norris 2002), and also a greater sense of political efficacy, which empowers and motivates them to participate in collective action (Paulsen 1991;Chen and Suen 2017). So, localities with more middle-class citizens may be more likely to mobilize, and I control for this in my analysis by considering average local incomes. Incomes do not perfectly capture the middle class, which also entails better access to education (Rosenfeld 2017). Nonetheless, income is the best available measure for this study due to an absence of statistics for local education in Belarus.
I must also consider that the population distribution across a country may influence subnational variation in mobilization. A large population increases the likelihood of local protest: greater human, financial, and organizational resources are available (Bob and McCarthy 2007); the locality is more inter-connected to other places, providing channels of information and mobilization (Pamela and Myers 2003); and the costs and benefits associated with mobilization are more favorable, as more potential protesters are available (Oberschall 1973). Hence, I will control for population in my analysis.
Finally, in robustness checks I will run models to verify whether there is any significant difference between localities that mobilized before and after mass repression and the first protester death on 10 August. Robustness checks will also examine whether, similarly, there is any significant difference between localities that mobilized during and after the internet blackout of 9-12 August.

Data and operationalization
To test the theoretical expectations outlined above, I use my original protest event catalogue for Belarus; Telegram data capturing the activity of local social networks, and opposition campaign events; and data on local populations and wages from the National Statistical Committee of Belarus. My unit of analysis is the individual locality level. I include all Belarusian towns and cities with populations over 5000 (n = 132), regardless of whether they mobilized. I use multivariate logistic regressions to test the above hypotheses regarding the early-rising mobilization of localities, social network ties, and campaign rallies, while holding relevant contextual variables constant. This quantitative approach is supplemented by qualitative research on the activity of local social networks prior to and during the initial week of mobilization. The following sections outline my variables and analytical approach in more detail.

The dependent variable: local protest event
The dependent variable of interest is a binary, indicating the presence (or absence) of protest in a locality from 9 to 15 August 2020 -the first week of post-election protests. Localities where protest occurred are coded as 1 (see Table 1). My analysis is limited to first-week mobilization in order to examine what factors motivate early-rising localities specifically. Focusing on first-week protest also reduces the number of additional potential explanatory variables, as processes of cross-location diffusion become more prevalent over time. A protest event is defined as a public gathering of two or more people, expressing claims or grievances relating to falsified elections and/or the beating of protesters.
Coding of this variable is based on my original protest event catalogue, which relies on data from a channel on the Telegram messenger app. Telegram provides the best available data on regional protest in Belarus during initial mobilization. It is widely acknowledged that news outlets underreport protest events outside large urban centers (Hutter 2014). Moreover, state censorship means Belarusian media were not reporting accurately on the protests. Independent journalists covering regional events were also limited by repression (threatened and actual), and the challenge of tracking of all the places rapidly mobilizing. In contrast, the Russian-language Telegram channel Nexta Live provided a continuous stream of information on protest locations from election day (Herasimenka 2021). Nexta Live was the largest Telegram channel used by protesters, with over 1 million subscribers at its peak (Hurska 2020). Moderators shared near real-time updates of protest events across Belarus. They received this information from protesters and fact-checked posts before sharing them (Kingsley 2020). Thousands of messages were shared from 9 to 15 August, so I took a two-stage, semi-automated approach to processing this data. Using keyword-based searches informed by Weidmann and Geelmuyden Rød (2019), I used R to identify all messages referring to protests in Belarusian towns and cities. This automated round of coding identified protest dates, locations, and repertoires. In a subsequent round of hand-coding, I read the messages and viewed attached multimedia content, verifying the results of automated coding, and adding further information on protest participants and turnout. The resulting event catalogue was then used to code my dependent variable, as described above.

Key independent variable: pre-existing social networks
To identify pre-existing social networks in Belarus, I also turned to Telegram as an innovative data source. Local social networks were using community Telegram chats as a tool for discussion and coordination prior to 9 August. Tracing the existence of these chats thus makes visible social networks communicating about and coordinating resistance to the Lukashenka regime, which are otherwise challenging to identify. The activity in these Telegram chats provides evidence of local social networks discussing politics, organizing around the elections, and even engaging in protest events in the weeks and months before 9 August.
To identify relevant chats, I used the crowdsourced list of regional Telegram chats on the website dze.chat as a starting point. I then identified additional chats missing from this database by searching location names using Telegram's native search function on the desktop app. As I am interested in pre-existing social networks, I excluded any chat created after 9 August. Most remaining chats were created in April-July, although some even pre-dated 2020. To ensure I was capturing social networks relating to resistance to the Lukashenka regime, I also excluded any chats where discussions did not center around activities of the political opposition or criticism of Lukashenka. This focus was usually reflected in the chat names, combining the locality with references to Tsikhanouskaya's "Country for Life" campaign, or the "97%" protest slogan -e.g. "Braslaw for Life" or "Orsha 97%." My list of Telegram chats is not exhaustive, as some may have been deleted prior to data collection (October 2020), but nonetheless represents an extensive map of pre-existing networks in Belarus pre-dating 9 August (see Figure 3). Using the data described above, I created an independent variable for "pre-existing social network," coded as 1 where a social network was identified in the locality prior to election day (see Table 2).

Competing independent variable: local campaign events
To identify localities where campaign events in support of Tsikhanouskaya took place prior to the elections, I also used Telegram data. Campaign events were published and promoted on the official channel of Tsikhanouskaya's campaign, "Country for Life". 5 Two types of events occurred: rallies  where Tsikhanouskaya was present, often with Maria Kalesnikava 6 and Veranika Tsapkala 7 ; and rallies that were organized in support of Tsikhanouskaya and featured "trusted" team members speaking on her behalf. Campaign events took place from 19 July 2020, when Tsikhanouskaya's candidacy was officially confirmed, and election day.
To capture these events, I used the same technique described above for the compilation of my protest event catalogue. I downloaded all messages between 19 July and 8 August 2020 from the "Country for Life" Telegram channel, used R to identify messages referring to campaign events, and then conducted a round of hand-coding to verify my findings. My results indicated that campaign events were held in 67 Belarusian localities in my dataset. I created an independent variable for opposition campaign events using these data, coded as 1 where at least one campaign event for Tsikhanouskaya was held in the locality (see Table 3).

Controls and robustness checks
To control for the fact that protest is more likely in towns and cities with strong transport networks, I used data from the Belarus Railways website to create a binary variable for each locality, where 1 indicates the presence of an active railway station. To control for income, I use a continuous variable of the local nominal average monthly wage in rubles, sourced from the National Statistical Committee of Belarus (2020). These data are available for towns of national significance and rayons (districts), so where town-level data were not available, I used the value for the relevant rayon. Finally, I include a control variable for population, using the log of population for each city, again using data from the National Statistical Committee of Belarus (2020).
I also run robustness checks to investigate the possibility that some localities were mobilized by mass repression and the first protester deaths on 10 August. To do so, I run my models with an alternate dependent variable, where all localities that mobilized prior to 11 August are coded as 1, localities which mobilized during the remainder of the first week are coded as 0, and those which did not mobilize are excluded. Similarly, to check whether there was a difference between localities that mobilized during and after the internet blackout, I run my models with a second, alternate dependent variable, where all localities mobilizing 9-12 August are coded as 1, localities which mobilized in the remainder of the week are coded as 0, and remaining localities are excluded.

Analytical approach
To examine the relationship between regional protest and pre-existing social networks, I primarily use multivariate logistic regression analysis. Logistic regressions rather than OLS were chosen because the dependent variable (presence or absence of first-week protest) is binary. 8 I first run a model with contextual variables only (population, wage, transport networks), before running models with my two key independent variables included separately, and finally together. As the log odds produced by logistic regressions are tricky to interpret, I report average marginal effects. Average marginal effects represent the average estimated likelihood of the dependent variable increasing from 0 to 1, when the independent variable of interest is increased by one unit, and while all other variables are held equal. For example, because my independent variable of pre-existing social networks is binary, the marginal effect represents the increase in the likelihood that protest takes place in a locality where I identify a local social network (1), compared to no network (0), and all other factors remain the same. For my variable of wage, which is scaled in units of 100 ruble increments, the marginal effect represents the increase in the likelihood that protest takes place in a town for each 100 ruble increase in the average local wage. The findings of my logistic regressions are supplemented by qualitative analysis of the ways in which pre-existing social networks were communicating, coordinating, and engaging in opposition activity, as identified using transcripts of Telegram community chats. I only used transcripts of chats that were public, meaning that they could be found via Telegram's search function, and anyone could view chat messages.

Results
My primary expectation, that during the first week of mobilization, protest is more likely to occur in localities where pre-existing networks are present, is confirmed in my analysis. The effect of local campaign events upon first-week mobilization is, however, not found to be significant, allowing me to reject the competing hypothesis identified in the literature that during the first week of mobilization, protest is more likely to occur in localities where campaign rallies for the opposition candidate took place prior to election day. Table 4 presents the results of my statistical analysis, which are also visualised in Figure 4.

Pre-existing social network ties hypothesis
My results indicate that when all other variables are held equal, the presence of a pre-existing social network increases the estimated likelihood of a locality mobilizing during the first week of protests by 22%. This relationship is strongly statistically significant at the 99% confidence level. Although it is not possible to directly infer causality and the mechanisms of the association from this result alone, the theoretical assumptions about social networks outlined above, as well as my findings about the activity of these chats (detailed below), do suggest that pre-existing networks contributed to the mobilization of towns and cities across Belarus during the first week of protest. Moreover, as my research analyzes the effect of social networks that existed prior to the onset of protest, the direction of any causal relationship would involve these networks influencing subsequent protest occurrence, and not the reverse.

Political opportunities (campaign event) hypothesis
However, contrary to the competing hypothesis identified in the literature, I found no significant association between the presence of Tsikhanouskaya campaign events in a locality in the run-up to elections, and subsequent first-week mobilization. Essentially, towns and cities where the Tsikhanouskaya team held a campaign event in the three weeks prior to the elections were no more likely to mobilize in the first week after the elections than localities where no event took place. This finding suggests that in Belarus, regional-level variation in political context is not necessarily helpful in explaining why some localities mobilized in the first week, but not others. Nevertheless, I acknowledge that political opportunities are a complex and nuanced concept. Future research exploring other facets of local political opportunities -such as the repressive tendences of the local police force, may produce different findings. It is also important to note that the highly personalized nature of Lukashenka's authoritarian regime means that political competition at the local level is limited in Belarus (Tsaryk 2018). Other country contexts with greater variation in local political context may find that local political opportunities are important to, or even compound the effect of pre-existing social network ties upon, cross-country variation in mobilization.

Controls: transport networks, wages, and population
I note that my control variables do not all align with local mobilization in the manner anticipated. I find no significant association between regional average monthly wages and mobilization, suggesting that localities with wealthier or poorer (on average) populations were no more likely to mobilize. This result contributes to recent findings on the role of economic grievances: one survey suggests many Belarusians have financial concerns (Langbein et al. 2021), but a protester survey indicates socio-economic concerns were not a key protest driver (Onuch and Sasse 2022). The presence of a railway station, representing strong transport networks, was also not significantly associated with first-week mobilization. This may indicate that in today's technologically advanced and highly connected era, transport infrastructures no longer play a crucial role in spreading information about protest, as they did in the past. Finally, as anticipated, there is a strong positive association between population size and first-week mobilization. However, accounting for the presence of pre- existing social networks absorbs some of this effect. This implies that some of the negative impact of a locality having a smaller population may be mitigated by the presence of pre-existing social networks. Finally, in robustness checks, I observed no significant difference between localities that mobilized before and after the onset of mass repression on 10 August, or between localities that mobilized before and after the lifting of the internet blackout. Nevertheless, future studies are needed to probe the effect of repression upon early-rising local mobilization in greater detail.

Qualitative results: the mechanisms of social network ties
While the above quantitative results demonstrate a robust and significant relationship between preexisting social networks and early-rising mobilization in Belarusian towns and cities, my qualitative results highlight the mechanisms that I am likely capturing. Observation of community Telegram chats provides ample evidence of communication, coordination, and engagement within local social networks both in advance of and during the first week of protest.

Communication
In the months and weeks prior to the elections, local networks via community Telegram chats were engaging in lively discussion of current affairs, the upcoming elections, and criticism of Lukashenka. People shared their frustrations with life in Belarus, and sometimes their hopes for the future. Community Telegram chats also proved a useful tool for sharing information about the presidential elections, such as videos, livestreams, posts and messages about opposition candidates, their programs for reforms, and campaign events. Discussions of the elections often included slogans that later became well known to the wider world during protests, such as "Long live Belarus," "Stop the Cockroach," and "We Believe, We Can Do It, We Will Win!" There were also frequent calls for people to share information beyond the network present in the chat, and speak with their families and friends about "what is happening in the country." Following the elections, this communication continued and increased in frequency in many instances. Local networks were sharing information about the election results as they were released and discussing their falsification. The emerging protests and repression in Minsk and other cities were also a common thread of discussion. Original messages, photos, and videos were shared, providing information about local events or events elsewhere that had either been witnessed firsthand or reported to the sharer by someone they knew. Messages, photos, and videos from other Telegram chats, both national and regional, were also shared, providing information about events unfolding across the country.

Coordination
However, local networks did not limit themselves to the discussion of events and sharing of information. Their usage of Telegram chats reveals extensive attempts to organize around the elections and support political opposition figures. In the weeks preceding 9 August, discussions of tactics were commonplace as people debated how best to undermine the regime. Some interactions became heated during debates about whether it was better to vote or boycott the elections altogether, useful or dangerous to protest on the streets, and whether Belarusians would be ready demonstrate following the elections. Lists of action points were shared, prompting people to do what they could in their community, such as sign petitions, call local news stations, hang flyers and flags, sign up to "initiative groups" coordinating protest action, or become a local election observer. In-person meet ups and Zoom conversations were organized to enable locals to get to know one another and exchange views. In some localities, events in support of opposition candidates were organized and promoted, as well as protests and flashmobs following the arrests of activists. People were encouraged to be ready to take to their town square on election night, and to stay home from work on strike on 10 August after Lukashenka would claim victory.
From election day onwards, these patterns of coordination continued. In multiple Telegram chats, it was possible to observe local networks organizing protest events in their town or city in response to the developing situation in the country. These protests were organized in different ways. In some cases, one user simply published details about when and where to protest, calling on others to join. In other instances, a decision about when and where to protest was reached via discussion amongst chat members. Sometimes people were told what to wear and bring with them, and what to do at the protest, or how to react in case of repression.

Engagement
Local Telegram chats also provide evidence that this communication and coordination amongst local networks translated into engagement and action prior to election day. Petitions and crowdfunding campaigns shared in the chats were signed and donated to. Flyers were printed, drawn, posted, and handed out. People turned out to meet-ups, protests, and events organized in support of opposition figures before 9 August, and discussed their impressions afterwards on their local chat. When some locals traveled to attend events in other locations, they offered a ride to others who might want to attend.
Therefore, it is of little surprise that all this discussion and activity resulted in mobilization on election day and the subsequent week. The above evidence from Telegram chats demonstrates that local networks were well informed about the situation unfolding in Belarus, had established trusted sources of information, and knew what was likely to happen on election day. They were already prepared to take some form of action on election night and during the following days, possessed the tools to organize this joint action, and in some instances had already mobilized around the elections in the preceding weeks. Thus, local social networks were primed and ready to take immediate action when, as expected, Lukashenka claimed victory on 9 August.
This identification of such widespread social networks and opposition activity before the summer 2020 protests is in itself a major finding for research on protest in Belarus. However, potential directions for future research could include efforts to better distinguish who are the participants of these local pre-existing networks -for example, are they seasoned activists, or grassroots groups? The nature of the Telegram chats I used to capture the activity of these social networks means that unfortunately scant information is available about the nature of the networks and their members. Additional qualitative study could uncover more information about local activist networks and social movement organizations, and their role in facilitating local protest. The analysis of data from other social media platforms could also reveal variation in how digital platforms help to capture social network ties.

Conclusion
In this article I have addressed the puzzle of what distinguishes early-rising protest localities from those that do not see protest, during moments of nationwide mass mobilization. Taking the case of the August 2020 post-election protests in Belarus, I argued that pre-existing social networks were likely to be key to early-rising mobilization, and also investigated a competing hypothesis centered on variation in local political context, represented by pre-election opposition campaign events. My results enabled me to confirm the former hypothesis, and to reject the later. Overall, my findings make three key contributions relating to the study of protest in Belarus and research on mass mobilization in general.
First, I have demonstrated the impressive extent of nationwide mobilization during the first week of the Belarus 2020 protests. Mobilization took place in over 100 localities and was not concentrated in a particular region, or clustered around the capital city of Minsk. In over two dozen locations, at least 1% of the local population mobilized. My research has also highlighted that prior to the onset of these protests, opposition to the Lukashenka regime was already well established across Belarus. Protest did not emerge out of the blue on election day -rather, local networks were already engaged and organizing in the preceding weeks and months. These findings challenge some of our understandings about civil society and activism in the country, often viewed as weakened by decades of authoritarian rule.
Second, I have provided both quantitative and qualitative evidence that these pre-existing social networks played a significant role in driving early-rising, geographically dispersed protest. Using Telegram as a tool to capture their activity, I have shown that these networks facilitated communication, coordination, and engagement in towns and cities prior to the elections. This activity prepared and motivated people to be ready to mobilize from 9 August. These results have implications for research on mass mobilization in general, as well as protest in Belarus, as they suggest that local social networks help drive mobilization in "early-riser" protest localitieseven in contexts where civil society has been consistently repressed, and the costs of protest are high.
Third, this study has demonstrated that Telegram is not only a useful tool for people looking to communicate and engage in protest activity, but is also valuable for social scientists looking to study protest engagement and mobilization. Here I have used Telegram both as a source for otherwise difficult-to-collect protest event data on regional mobilization, and also as a means to trace the activity of the social networks engaging in protest. Of course, this reliance on Telegram data also underpins some of the limitations of this study. The claims made here about pre-existing social networks must be interpreted with care, as these data do not capture social network ties and social movement organization networks that may exist "offline." Future qualitative work and detailed protest tracing will help to uncover more details about pre-existing social networks in Belarus and their role in early-rising and local mobilization.
This work highlights the need to look closely at the actors involved in the early stages of mobilization during cases of nationwide mass protest. What may seem like spontaneous mobilization driven by social media, as was widely reported about Belarus, may in fact be at least partially initiated and coordinated by pre-existing social networks at the local level. And, this may be the case even under highly repressive regimes, where those who engage in dissent face great risk. The implications of this study suggest that scholars of protest should not forget to consider the role of pre-existing social networks in mass mobilization, even when such networks are not obvious; and those looking to support democratizing movements should consider how they can strengthen and support grassroots social networks and communities at the regional level.

Notes
1. This process is often referred to as diffusion. 2. "Social networks" is a broad term that can encompass formal and informal social movement organizations, grassroots activist groups, and networked publics. 3. Analysis is limited to localities with a population of over 5000 (n = 132), due to the additional resources and time required to gather data for all settlements in Belarus. The findings of this article should thus not be considered applicable to localities with populations under 5000. However, over 75% of the population of Belarus is covered by the localities in my dataset. 4. Data on protester numbers are notoriously difficult to collect, so these figures are approximate and represent very conservative estimates. 5. This channel was the most reliable available source of information about campaign events. The campaign website was offline during data collection, and other social media accounts did not publicize all campaign events. 6. Campaign chief for the imprisoned presidential candidate Viktar Babaryka. 7. Wife of Valery Tsapkala, former presidential candidate who was denied registration and fled Belarus following threats to his life. 8. Robustness checks using OLS models returned very similar results. ""