Pulling through elections by pulling the plug: Internet disruptions and electoral violence in Uganda

Garbe, Lisa Article — Published Version Pulling through elections by pulling the plug: Internet disruptions and electoral violence in Uganda Journal of Peace Research Provided in Cooperation with: WZB Berlin Social Science Center Suggested Citation: Garbe, Lisa (2024) : Pulling through elections by pulling the plug: Internet disruptions and electoral violence in Uganda, Journal of Peace Research, ISSN 1460-3578, Sage, London, Vol. 61, Iss. 5, pp. 842-857, https://doi.org/10.1177/00223433231168190 This Version is available at: https://hdl.handle.net/10419/294780.2 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ https://doi.org/10.1177/00223433231168190 Journal of Peace Research 2024, Vol. 61(5) 842 –857 © The Author(s) 2023 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/00223433231168190 journals.sagepub.com/home/jpr 1225162JPR0010.1177/00223433231168190Journal of Peace ResearchGarbe research-article2023 Regular Article Pulling through elections by pulling the plug: Internet disruptions and electoral violence in Uganda Lisa Garbe WZB Berlin Social Science Center Abstract Does increasing Internet access and use challenge authoritarian elections? I argue that Internet access provides both opposition supporters and government authorities with new means to shape electoral conduct. Opposition supporters can use the Internet to report on electoral malpractice and mobilize for support. At the same time government authorities can use the Internet to monitor antiregime sentiment prior to the elections and disrupt Internet access to selectively repress regime opponents during the elections. Studying Uganda’s 2016 presidential elections, evidence from election monitoring and survey data suggests that electoral violence is significantly higher in opposition strongholds with greater Internet access prior to the Internet disruption and is targeted specifically at voters. Insights from qualitative interviews with politicians, journalists and activists underline that the disruption of Internet access indeed hindered opposition supporters to effectively challenge electoral malpractice. Overall, the results stress the important role that Internet access can play for opposition actors in authoritarian elections. At the same time, they highlight their susceptibility to manipulation by government authorities. Keywords authoritarian elections, electoral violence, social media, sub-Saharan Africa Introduction Does Internet use challenge authoritarian elections? The number of African elections during which governments ordered the disruption or manipulation of Internet access suggests indeed that many African rulers perceive Internet use as a threat. Between 2015 and 2020 alone, one-third of all national elections in sub-Saharan Africa (SSA) were accompanied by an Internet disruption, with governments either blocking specific websites or curtailing access to entire networks (Freyburg & Garbe, 2018; Rydzak, Karanja & Opiyo, 2020). At the same time, as shown in Figure 1, levels of electoral violence were significantly higher during those African elections that were accompanied by an Internet shutdown, defined as any ‘intentional, significant disruption of electronic communication within a given area and/or affecting a predetermined group of citizens’ (Rydzak, 2018: 6), even when controlling for relevant factors, including local conflict, gross domestic product (GDP) per capita, and regime type. 1 This raises the question in what way Internet use – and its disruption – affects electoral conduct. The role of new information and communication technologies (ICT) in the conduct of authoritarian-developing elections remains poorly understood (Gagliardone, 2019). Anecdotal evidence, such as from the Ushahidi platform, which has been used to crowdsource data on incidents of violence during multiple African elections, including in Kenya 2007, Burundi 2010 and Sudan 2010, suggests that citizens and opposition actors routinely use ICT to report on electoral violence (Bunz, 2014; Okolloh, 2009; Zein, 2010). Simultaneously, governments increasingly monitor online discourse to identify Corresponding author: [email protected] 1 see Online appendix 1 for results from OLS regression models. Garbe 843 antiregime sentiment and relevant opposition activists (King, Pan & Roberts, 2013; Qin, Stro ¨mberg & Wu, 2017). Scholars have thoroughly examined the ‘offline’ determinants and effects of electoral violence – but how does Internet access, and, in turn, its disruption, affect electoral violence? I borrow from insights in political communication studies (Bennett & Segerberg, 2013; Bimber, 2017; Karekwaivanane, 2019; Mare, 2016) to argue that the use of ICT helps opposition supporters address traditional challenges in mobilizing support, such as preference falsification (Kuran, 1989) or resources constraints (McCarthy & Zald, 1977). Facing these challenges, mobile smartphones can be used to undermine electoral malpractice by enabling the opposition to expose violence or fraud and to monitor election results in real time (Baguma & Eilu, 2015). In turn, and for the same reasons, authoritarian governments are likely to manipulate or suppress online communication during contested elections. As indicated by the literature on digital repression, ICT likely help governments localize opposition actors and selectively repress regime opponents during periods of Internet disruptions (Gohdes, 2020; Xu, 2020). Given that a growing number of people rely on ICT to access information, communicate and mobilize during authoritarian elections (Karekwaivanane, 2019), it is plausible to expect a disruption of ICT access to have profound consequences for the conduct of elections, resulting in increased electoral violence. In the next section, I briefly review the literature on electoral violence and develop a theoretical argument on how access to and the disruption of Internet services can affect the occurrence of electoral violence. Subsequently, I discuss Uganda’s 2016 presidential elections as a case study and present statistics drawn from original survey data (N¼2,042) and election monitoring data from polling stations (N¼238) across Ugandan constituencies. I contextualize the statistical findings by 25 semistructured face-to-face interviews with journalists, citizen activists and politicians from three opposition-leaning Ugandan districts, as well as evidence from major Ugandan newspaper outlets. The last section summarizes the findings and discusses the differential impact of an Internet disruption on opposition and government actors. Results from the statistical analyses indicate that electoral violence is higher in those constituencies with more opposition support and higher Internet access prior to the disruption. Evidence from the qualitative interviews posits that Internet disruptions can indeed affect the occurrence of electoral violence: While a disruption prevents opposition supporters from effectively using ICT to challenge electoral malpractice, at the same time, it may obscure the use of violent state repression. My study thus highlights the importance of disentangling the ways in which ICT are used by citizens and state actors at election times to better understand the effects of Internet access and its disruption on core democratic processes. Electoral violence in authoritarian regimes Today, most countries in the world, whether democratic or not, hold regular elections. Authoritarian rulers may have different motivations for holding elections, including the co-optation of elites (Boix & Svolik, 2013), party members (Magaloni, 2008) and broader groups in society (Gandhi & Przeworski, 2006); or to strengthen the regime’s legitimacy (Waterbury, 1999), to identify bases of support and opposition (Gandhi & Lust-Okar, 2009) and to foster long-term regime stabilization (Knutsen, Nygård & Wig, 2017). In any case, the incumbent ruler needs to ensure that the election result does not threaten their rule. Cheeseman & Klaas (2018: 5) thus argue that ‘the art of retaining power has become the art of electoral manipulation’. Electoral violence, defined as ‘events in which incumbent leaders and ruling party agents employ or threaten violence against the political opposition or potential voters before, during or after elections’ (Hafner-Burton, Hyde & Jablonski, 2014: 150), can be seen as the most coercive form of electoral malpractice (van Ham & Lindberg, 2015). In some cases, state security forces coerce citizens, such as during Kenya’s 2017 presidential elections, Figure 1. Election violence during African elections 2015–20 Higher scores indicate higher levels of election violence; horizontal lines show the median. Data for electoral violence (v2elintim) comes from the Varieties of Democracy (V-Dem) project (Coppedge et al., 2020) and ranges from 0 to 4; data for Internet shutdowns from #KeepItOn campaign of the civil society organization Access Now (Access Now, 2016); N country-election-years ¼76. 2journal of PEACE RESEARCH XX(X) 844 journal of Peace Research 61(5) where the police used high levels of repression against opposition protesters ‘sometimes firing live ammunition at unarmed protesters’ (Mutahi & Ruteere, 2019: 254). In other cases, unofficial organizations such as youth gangs are put in place to harass or intimidate voters. In Uganda, so-called crime preventers were installed to support the government ‘by intimidating and brutalizing opposition supporters and reducing their turnout’ (Dow, 2022: 1606). Ho ¨glund (2009) argues that electoral violence differs from political violence due to its different motives and hence needs to be studied as a phenomenon in itself. Between 2012 and 2016, a quarter of all elections worldwide and more than one-third of SSA elections involved severe violence (Cheeseman & Klaas, 2018: 95f). One of the main aims of electoral violence and intimidation is to demobilize potential non-supporters and prevent them from casting their vote (Bratton, 2008; Ho ¨glund, 2009). As such, electoral violence is usually targeted at areas with a strong opposition and well-informed voters (Bhasin & Gandhi, 2013; Rauschenbach & Paula, 2019; von Borzyskowski & Kuhn, 2020). In consequence, electoral violence can have far-reaching consequences not only for the electoral process but also for election results. In response to electoral violence, citizens may engage in collective action (Tucker, 2007), especially when they know that others do so too (Bratton, 2008). However, such efforts are scarce in electoral autocracies, where collective action involves high costs and low chances of success (Tucker, 2007). Digital forms of communicating and sharing information may help ‘solve collective action problems that have long bedeviled those traditionally shut out of mainstream politics’ (Tucker et al., 2017: 47). 2 In particular, the Internet provides access to new sources of information about the electoral process and potential fraud (Reuter & Szakonyi, 2013), which may influence citizens’ decision to engage in countermobilization. In turn, authoritarian governments are incentivized to block or manipulate Internet access during elections. Yet, we still know little about the effects of Internet access, and its disruption, on electoral violence. This study therefore addresses the following question: How does access to Internet services, and its disruption, affect the occurrence of electoral violence? Theoretical expectations Since the early days of the Internet, scholars are interested in the conditions under which Internet access may bolster or challenge authoritarian rule and study the effect of Internet penetration on authoritarian survival, protest behaviour, or state repression (Deibert et al., 2008; Gohdes, 2015b; Rød & Weidmann, 2015; Ruijgrok, 2017; Ruijgrok, 2020; Xu, 2020). I borrow from insights in political communication to define the ways in which Internet access, and its disruption, can affect electoral violence. First, I systematically discuss how Internet access can help opposition supporters to challenge electoral violence. Second, I argue that a disruption of previous Internet access limits opposition supporters’ ability to address electoral violence while, at the same time, it allows state actors to obscure their use of electoral violence. Challenging authoritarian elections: The role of Internet access Internet access may help citizens to address electoral violence. On the one hand, access to Internet services encourages individuals under authoritarian rule to reveal their true preferences and, by providing information about the opposition’s strength, to encourage protest against electoral violence. On the other hand, once people have decided to engage in mobilization, ICT use substantially facilitates the organization of their efforts. According to social movement theory, to become active in antiregime mobilization, citizens need to have a sense of the extent to which their grievances are shared with others (Gurr, 2015; Turner & Killian, 1957). Applied to the context of elections, voters are more likely to challenge electoral malpractice if they perceive the elections as fraudulent (Daxecker, Di Salvatore & Ruggeri, 2019) and if they know that others are also willing to stand up against it (Bratton, 2008). However, opposition elites and their supporters in authoritarian regimes are often confronted with the problem of ‘preference falsification’ (Kuran, 1989). Even if they secretly favour the opposition, citizens may deny their preferences in public due to the threat of punishment and uncertainty about broader public opinion. So-called ‘islands of separateness’ (Friedrich & Brzezinski, 1963: 279ff) – places in which people express and mobilize for their antiregime opinions – tend to be scarce in the authoritarian offline world. In the context of elections, citizens may lack information about the extent to which others share their disapproval of the regime and, hence, be reluctant to engage in mobilization challenging 2 See also Bailard (2015) and Pierskalla & Hollenbach (2013) for empirical research on the effect of mobile penetration on collective action. Garbe 3 Garbe 845 election malpractice. The alleged anonymity on the Internet can encourage individuals to share their ‘true’ preferences (Farrell, 2012), especially in places where the public sphere is heavily restricted (Chen et al., 2016). For instance, in Zimbabwe, the Facebook site run by anonymous blogger Baba Jukwa provided a crucial space for critical voices ahead of the elections in 2013 and is said to have had a significant impact on the mobilization of voters (Karekwaivanane, 2019: 54; Mare, 2016). Moreover, Internet access makes it easier for protesters to communicate and organize without the existence of formal organizations (Bennett & Segerberg, 2013; cf. Bimber, 2017). Citizens typically face a constraint of resources needed to successfully organize mobilization, including money, time and knowledge (McCarthy & Zald, 1977). Internet access reduces the costs of sharing information and can make formal organizational structures obsolete (Bennett & Segerberg, 2013; Castells, 2015). In particular social networking sites provide a platform ‘for debate and knowledge-sharing while also enabling a message to reach its targeted audience in unprecedented fashion, within seconds’ (Mutsvairo, 2016: 6). They can be used nationwide, such as Twitter for mobilization during Nigeria’s elections (Bartlett et al., 2015), or in specific local contexts, such as Whats App groups ahead of county elections in Kenya (Omanga, 2019). Original evidence from Uganda indicates that opposition supporters are more likely to use ICT to mobilize voters and report on electoral malpractice. The survey data collected for the purpose of this study (see methods section) suggests that opposition supporters are more likely to use ICT to challenge electoral malpractice and mobilize voters than citizens who do not support any of the opposition parties during Uganda’s 2016 presidential elections. Respondents supporting opposition parties declared a higher propensity to use social media to report election malpractice and to communicate their participation in the elections, even when accounting for the influence of several individual characteristics and including constituency fixed effects (see Online appendix 1 for regression tables). Importantly, mobilization taking place online can also spill over to offline engagement (Chibita, 2016). Baguma & Eilu (2015) argue that, in developing countries, particularly mobile (smart) phones provide individuals with an efficient tool for monitoring electoral malpractice. Pictures and other pieces of information can instantly be shared with broader networks, documenting incidences of violence and enabling opposition actors to send assistance to affected polling stations. During several African elections, citizens used online platforms to report on violence during elections and thereby ‘enable stakeholders at the local level to prevent or evade conflict’ (cf. Bartlett et al., 2015; Mutahi & Kimari, 2017: 20). Deterring mobilization during elections: Authoritarian use of Internet access and disruptions Internet access may help state authorities to identify areas in which counter-mobilization is likely to challenge the elections and directly target electoral violence at voters in those areas. Authoritarian rulers are affected by the problem of preference falsification, too, facing uncertainty as to which parts of the population may threaten regime survival (Boix & Svolik, 2013). Many authoritarian governments use formal and informal institutions to co-opt members of society and thereby alleviate monitoring problems (Gandhi, 2008; Gerschewski, 2013). However, co-optation becomes expensive with the increasing size of the so-called ‘winning coalition’, which is the part of society on whose support the government relies (Bueno De Mesquita et al., 2003). Instead, it can be more efficient to use targeted repression as ‘the cost to buy support from radicals can be significantly higher than that to imprison them’ (Xu, 2020: 4). Internet access has equipped governments with new means to monitor regime dissent and forestall mobilization efforts by opposition actors. For instance, King, Pan & Roberts (2013) argue that governments use information from online communication to learn about and localize antiregime efforts (cf. Qin, Stro ¨mberg & Wu, 2017). While originally applied to government behaviour in China, studying online discourse may prove particularly useful to any government in electoral autocracies, as elections present a key moment of political uncertainty in which mobilization may threaten a regime’s survival. However, as described above, when government authorities commit electoral violence, they face the risk of countermeasures precisely in those areas in which opposition supporters have access to Internet services and hence the ability to document and mobilize against violent behaviour. Governments, in turn, have better chances to commit violence without risking denunciation of their actions if Internet access is disrupted during elections. In particular, they can commit targeted acts of state violence without risking unfavourable consequences of ICT use by the opposition (Gohdes, 2015a; Kasm, 2018). Assuming that governments frequently monitor online behaviour to forestall where antiregime mobilization is likely to occur, they may implement short-time disruptions of Internet access to prevent 4journal of PEACE RESEARCH XX(X) 846 journal of Peace Research 61(5) people from monitoring their actions and hence obscure the use of state violence. By banning access to widely used online platforms, governments hinder the opposition from documenting state violence and effectively challenging their use of coercive force. I therefore expect that: Levels of electoral violence are higher in areas, in which opposition supporters had access to Internet services prior to a disruption. The case of Uganda’s 2016 presidential elections Uganda’s presidential elections in 2016 were the third multiparty elections since the National Resistance Movement (NRM) and Yoweri Museveni took power in 1986. With a reform in 2005, the government established a multiparty system and thus allowed for (minimal) party competition. Even though previous elections were not held under free and fair conditions (Levitsky & Way, 2010; Schedler, 2006), the opposition was a credible challenger to incumbent ruler Museveni. According to opinion polls before the 2016 elections, the opposition party Forum for Democratic Change’s (FDC) popularity peaked (Beardsworth, 2016). Consequently, the reelection of the incumbent president was more contested than during any of the previous elections. In the years before the election, the Internet had become an important ‘tool for social, economic, and human rights development in Uganda’ (CIPESA, 2016). Due to increased connectivity, especially through mobile phones, it also provided individuals with new means to participate in politics (Gro ¨nlund & Wakabi, 2015; Grossman, Humphreys & Sacramone-Lutz, 2014). Especially for opposition campaigners, social media appeared a relevant tool ahead of the 2016 elections as following the Public Order Management Act in 2013, anyone holding a public meeting or rally needed to inform the police, which ‘ha[d] always restricted members of the opposition from making public consultations with citizens’ (Enenu, 2016). This act was said ‘to intimidate opponents of President Yoweri Museveni’ (Biryabarema, 2016). Furthermore, the use of social media for real-time reporting became pressing due to violence during election campaigns, ‘especially against opposition candidates’ (Ssekika, 2016). For opposition parties, social media thus represented a ‘crucial’ channel to communicate and share information (Presidential, parliamentary elections in embarrassing mess, 2016). A few days ahead of the 2016 elections, the Electoral Commission (EC) announced the ban of smartphones at polling stations (Karugaba, 2016; Musisi, 2016). This ban was regarded as a strategic move to ‘limit information flows’ (Musisi, 2016). In response to the smartphone ban, opposition candidate Amama Mbabazi publicly encouraged all voters to ignore the ban and ‘go with their phones and cameras, and feel free to record anything they think is going wrong’ (as cited in Kaaya, 2016). There was already a restriction of communication during the previous presidential elections in 2011, at a time when access to the Internet was still relatively low. Back then, the government had ordered telecom operators to ban specific keywords on SMS, including ‘Egypt’, ‘bullet’ and ‘people power’ (Biryaberema, 2011). The disruption of social media on polling day was indeed the first time that Internet access was restricted nationwide at a large scale (Kembabazi, 2016). This is supported by media coverage in major Ugandan news outlets, none of which mentioned a possible Internet disruption in the months ahead of the election. 3 Uganda’s 2016 presidential elections provide an ‘easy’ case (Seawright & Gerring, 2008) to better understand ICT use and digital repression at election times for three main reasons. First, most Internet disruptions on election day take place in electoral autocracies, such as Uganda (Letsa, 2019) – that is, regimes that allow for minimum multiparty competition without granting free and fair de facto multiparty elections (Lu ¨hrmann, Lindberg & Tannenberg, 2017). Second, Uganda ranks among the third of countries with the highest levels of electoral violence 4 and irregularities (see Online appendix 2); therefore, it represents a group of SSA countries notoriously affected by electoral malpractice. Third, regarding digitalization, Uganda is representative of many SSA countries, where mobile Internet connectivity substantially increased since 2014, with more than onethird of the country’s population online (Bahia & Suardi, 2019). Therefore, evidence from the Ugandan case generates useful knowledge about how Internet disruptions may affect electoral processes in other electoral autocracies in the region. 3 I systematically searched articles from major newspapers in Uganda using the information research tool Factiva (keywords: ‘internet shutdown OR internet outage OR internet blackout OR social media blocking OR social media shutdown OR social media blackout’; newspapers: New Vision, Daily Monitor, The Red Pepper, The East African, East African Business Week, The Observer; dates: 01.01.2016–07.02.2016). The search resulted in 0 articles. 4 During the 2016 elections, electoral violence was ‘most visibl[e] as repeated harassment and arrests of opposition politicians and supporters’ (Sjo ¨gren, 2018: 57) Garbe 5 Garbe 847 Research design Empirical strategy I apply a cross-sectional research design to examine the occurrence of electoral violence across constituencies during Uganda’s 2016 elections combining survey and election monitoring data. As there is no geographic variation in the disruption of Internet access, I use Internet access prior to the disruption as a proxy of how severely people were affected by the sudden loss of Internet access (limitations of this approach are discussed following the results). I estimate the effect of the interaction between the proportion of opposition supporters and Internet access on electoral violence. Specifically, I estimate the extent to which local Internet access and the proportion of opposition supporters influence the probability of electoral violence in a sample of 195 polling stations across 70 constituencies. 5 Descriptive statistics for the main indicators included in the analyses are provided in Table I. To contextualize the findings, I interviewed 25 opposition and government politicians, activists and journalists in three electoral districts – Kampala, Gulu and Kitgum – in October and November 2018 (see Online appendix 5). Those districts tend to be oppositionleaning but vary with regard to their Internet coverage from high (Kampala) to intermediate (Gulu) and low (Kitgum) connectivity (GSMA Intelligence, 2019). In addition, I use insights from media reports of major news outlets in Uganda, including the Daily Monitor,New Vision and the Observer, two weeks before and after the elections. 6 Dependent variables ‘Electoral violence’ is operationalized with six indicators based on items from election monitoring data collected by the Citizens’ Election Observers Network–Uganda (CEON-U). Election observers recorded different forms of electoral violence during the polling process and during vote counting on a checklist (see Online appendix 6). For each point of measurement, I combine these indicators into one variable for violence (with the presence of any form of violence or intimidation at a polling station ¼1). I further use each individual indicator as separate outcome (e.g. the presence of violence directed at voters ¼1). Independent variables: Opposition strength and access to Internet For Internet connectivity and opposition strength, I aggregate original data from a regionally representative survey conducted in April 2019 by the research institute Research World International Ltd covering 2,042 respondents across 82 constituencies in Uganda using a multistage stratified random sampling design. The survey was conducted face-to-face with Ugandan nationals aged at least 18 years and is representative at the level of regions. I included three questions in the survey covering Internet access, use and political preferences (see Online appendix 7). The survey, conducted three years after the elections, relies on retrospective self-reporting that might be subject to memory bias, among other possible bias (Stone et al., 2007: 12). Reliability assessments using data from the Afrobarometer and the World Bank suggest minimal bias (see Online appendix 8). 7 The first key independent variable ‘opposition support’ is based on party preferences ahead of the 2016 elections. Data was coded binomially, with support for the ruling party NRM or support for no specific party Table I. Descriptive statistics Variable Min–Max M(SD) N Polling station level violence (t1) 0–1 0.35 (0.48) 195 violence (t2) 0–1 0.35 (0.48) 195 opposition votes 2011 (in %) 0–90 38 (20) 195 Constituency level opposition support (in %) 0–55 0.23 (0.14) 70 internet access (in %) 0–67 0.27 (0.17) 70 average economic status [0;2] 0–0.82 0.16 (0.14) 70 average educational level [0;8] 1.31–4.69 2.53 (0.64) 70 urban population (in %) 0–100 44 (42) 70 M¼mean; SD ¼standard deviation; the combined indicator violence is coded with 1 if violence committed against or by any of the actors listed in Online appendix 3 is observed. 5 Electoral violence was recorded twice on polling day, resulting in two observations for each polling station. I therefore use a single indicator of violence for both moments in time (N¼390). An excerpt of the data can be found in Online appendix 4. 6 I systematically selected articles from major newspapers in Uganda using the information research tool Factiva (keywords: ‘elections AND internet OR ICT OR social media OR mobile OR phones’; newspapers: New Vision, Daily Monitor, The Red Pepper, The East African, East African Business Week, The Observer; date: 03.02.2016-05.03.2016). The search resulted in 185 articles. 7 Data from the Afrobarometer is only representative at the national level and can therefore not be used as alternative data source to assess Internet access and opposition support across constituencies. 6journal of PEACE RESEARCH XX(X) 848 journal of Peace Research 61(5) (coded with 0) and support for any opposition parties (coded with 1). 8 The second key independent variable ‘internet access’ assesses whether people had Internet access before the 2016 elections (access coded with 1 versus no access coded with 0). I calculate the proportion of individuals with Internet access and opposition support, respectively, per constituency. Controls I account for four additional factors that might be systematically related to the level of violence, namely ‘average economic status’, ‘average educational level’, the share of ‘urban population’ per constituency and the share of opposition votes per polling station in 2011. I provide more details about the control variables in Online appendix 9. In addition, I rerun all models using population size as alternative control variable (Online appendix 10.1). Methods I use generalized linear models (GLM) to estimate the effect of the interaction term of opposition support and Internet access on the likelihood of electoral violence. For the combined indicator of violence, an intraclass correlation (ICC) suggests that around 12% of the total variance in election violence is accounted for by the constituency clustering (ICC ¼0.12). Therefore, I include constituency as random intercept to account for within-constituency effects that might occur due to varying ethnic and political legacies, among others (Stegmueller, 2011). 9 For the models using the individual violence indicators as outcome, I do not include random intercepts as the ICC suggests little variance at the level of constituencies (all ICCs except for two indicators < 0.05). Electoral violence was recorded twice on polling day, in the morning and in the afternoon. I include time fixed effects to account for temporal correlation in all models. For the multilevel models, the assumption of linearity between the logit of the outcome and predictors in the model is fulfilled, and there is no indication of multicollinearity in any of the models (all VIFs < 5). As survey data is only representative at the level of regions, but not at the level of constituencies, I additionally run all models reweighting the independent variables based on census data (see Online appendix 10.2). Empirical analysis Statistical results Evidence from the election monitoring data underlines the variation in electoral violence across constituencies. Figure 2 indicates that violence during polling (t1) and during vote counting (t2) were not equally present across the country. Several places with a high number of polling stations affected by violent incidences during polling, such as Chua or Chekwii, remained free from observed violence during vote counting. Overall, the mean share of polling stations affected by violence during polling per constituency (M¼0.33, SD ¼0.34) was similarly high as the mean share of polling stations affected by violence during vote counting (M¼0.34, SD ¼0.35). Results from the GLMMs indicate that electoral violence is higher in areas with increasing Internet access and increasing opposition support (see Figure 3). The model comparisons reveal a positive significant effect of the interaction between the proportion of opposition supporters and Internet access on the likelihood of electoral violence ( 2 (1) ¼6.18, p¼0.01) (see Appendix 2 for regression table). That is, the occurrence of violence at a polling station depends on the interplay of strength of opposition support and Internet access in a constituency. The results further indicate that violence more likely occurs in areas with a lower socio-economic status (B¼-3.81, SE ¼1.52). To assess whether these findings are driven by dynamics in Uganda’s most populated district, I exclude observations from constituencies located in the capital district Kampala (N¼32). In this model, the interaction effect is even more pronounced ( 2 (1) ¼6.7, p< 0.01; see Appendix 3). Figure 4 provides an overview of the marginal effects for the individual indicators of violence. The results indicate that the interaction term has a significant positive effect on violence against voters (OR ¼1.8, 95% CI[1.05, 3.24], p¼0.04), but not on violence against polling agents or election officials (both p> 0.05). I computed the model-implied probabilities for violence against voters assuming opposition support 1 SD above the mean both for areas with high Internet access (1 SD above the mean) and for areas with low Internet access (1 SD below the mean) while holding all other predictors at their mean level. This yielded an expected probability of violence against voters of 10% for areas with high and 2% for areas with low Internet access. Overall, the results 8 I apply a conservative approach and focus on opposition supporters versus government supporters and non-partisan citizens; African countries are characterized by low partisan attachment to a political party (Kuenzi & Lambright, 2011: 779). 9 Visual inspection of the distribution of all best linear unbiased predictors (BLUPs) suggests no violation of the assumption of Gaussian distribution. Garbe 7 Garbe 849 suggest that the probability of violence against voters increases with increasing opposition support and Internet access. Second, the interaction term has a significant positive effect on the presence of the police (OR ¼1.43, 95% CI[1.07, 1.92], p¼0.01), the army (OR ¼3.73, 95% CI[1.18, 15.9], p¼0.04) and crime preventers (OR ¼1.93, 95% CI[1.27, 3.02], p¼0.00). In other words, with an increasing share of opposition supporters and Internet access, the probability of the unauthorized presence of the police, the army and crime preventers increases. Online appendix 4 provides an overview of all models. Understanding the role of the Internet disruption Evidence from 25 qualitative interviews with Ugandan politicians, activists and journalists conducted in 2018 emphasizes the potential of ICT use to address electoral malpractice, yet also highlights how an Internet disruption hinders the opposition to realize this potential. In particular, insights from the interviews outline: (1) the ways in which the opposition prepared for the elections using social media; (2) suspected repression by government authorities targeted at the opposition; and (3) likely consequences of the Internet disruption Figure 2. Levels of electoral violence in Uganda Percentage of polling stations affected by violence at county level. Data for county boundaries comes from the Humanitarian Data Exchange (2018). Counties with missing values are left blank. 10 Presence of violence was assessed during polling (t1, left panel) and vote counting (t2, right panel) on polling day. Figure 3. Effects of interaction term on violence (combined indicator) Plot shows marginal effects at the mean of opposition support conditioned by Internet access, i.e. with all control variables in the model held constant, with 90% confidence intervals. The predicted values are grouped for Internet access one and a half standard deviations below (¼0.04, dotted line) and above (¼0.57, bold line) the mean (¼0.31). 10 The Ugandan Bureau of Statistics originally provides data. The shapefiles provide data for county boundaries in 2006. Due to the frequent changes in administrative boundaries in Uganda (Grossman & Lewis, 2014), some areas may appear in grey even though election monitoring data is available. 8journal of PEACE RESEARCH XX(X) 856 journal of Peace Research 61(5) Karugaba, Mary (2016) EC bans use of mobile phones at polling stations. New Vision, 12 February. Kasm, Saeb (2018) Redefining publics: Mosireen, state crime and the rise of a digital public sphere. State Crime Journal 7(1): 100–140. Kembabazi, Gloria (2016) An open internet versus social media shutdown. APC, 23 May. (https://www.apc.org/ en/blog/open-internet-versus-social-media-shutdown). King, Gary; Jennifer Pan & Margaret E Roberts (2013) How censorship in China allows government criticism but silences collective expression. American Political Science Review 107(2): 326–343. Knutsen, Carl Henrik; Håvard Mokleiv Nygård & Tore Wig (2017) Autocratic elections: Stabilizing tool or force for change? World Politics 69(1): 98–143. Kuenzi, Michelle & Gina MS Lambright (2011) Who votes in Africa? An examination of electoral participation in 10 African countries. Party Politics 17(6): 767–799. Kuran, Timur (1989) Sparks and prairie fires: A theory of unanticipated political revolution. Public Choice 61(1): 41–74. Letsa, Natalie Wenzell (2019) The political geography of electoral autocracies: The influence of party strongholds on political beliefs in Africa. Electoral Studies 60(August): 102047. Levitsky, Steven & Lucan A Way (2010) Competitive Authoritarianism: Hybrid Regimes After the Cold War. Cambridge: Cambridge University Press. Lu ¨hrmann, Anna; Staffan Lindberg & Marcus Tannenberg (2017) Regimes in the World (RIW): A robust regime type measure based on V-Dem. Working Papers 2017:47, University of Gothenburg (https://ssrn.com/abstract¼ 2971869). Magaloni, Beatriz (2008) Credible power-sharing and the longevity of authoritarian rule. Comparative Political Studies 41(4-5): 715–741. Mare, Admire (2016) Baba Jukwa and the digital repertoires of connective action in a ‘competitive authoritarian regime’: The case of Zimbabwe. 45–68. Martin, Jason A (2014) Mobile media and political participation: Defining and developing an emerging field. Mobile Media & Communication 2(2): 173–195. McCarthy, John D & Mayer N Zald (1977) Resource mobilization and social movements: A partial theory. American Journal of Sociology 82(6): 1212–1241. Musisi, Frederic (2016) Queries as electoral commission bans phones at polling stations. Daily Monitor, 14 February. Mutahi, Patrick & Brian Kimari (2017) The impact of social media and digital technology on electoral violence in Kenya. IDS Working paper #493 (https://opendocs.ids. ac.uk/opendocs/bitstream/handle/20.500.12413/13159/ Wp493_Online.pdf?sequence¼287). Mutahi, Patrick & Mutuma Ruteere (2019) Violence, security and the policing of Kenya’s 2017 elections. Journal of Eastern African Studies 13(2): 253–271. Mutsvairo, Bruce (2016) Dovetailing desires for democracy with new ICTs’ potentiality as platform for activism. 3–23. Nisbet, Erik C; Elizabeth Stoycheff & Katy E Pearce (2012) Internet use and democratic demands: A multinational, multilevel model of internet use and citizen attitudes about democracy. Journal of Communication 62(2): 249–265. Observer, the (2016a) FDC to crack regime arrogance & impunity - Mugisha Muntu. 2 March (https://observer. ug/news-headlines/42910-fdc-to-crack-regime-arrogance- impunity-mugisha-muntu). Observer, the (2016b) Presidential, parliamentary elections in embarrassing mess. 19 February. (https://observer.ug/ news-headlines/42678-presidential-parliamentary-elec tions-in-embarrassing-mess). Okolloh, Ory (2009) Ushahidi, or ‘testimony’: Web 2.0 tools for crowdsourcing crisis information. Participatory Learning and Action 59(1): 65–70. Oluka, Benon Herbert (2016) Cash squeeze, intimidation rock FDC vote-guarding plans. Observer, 10 February (https://www.observer.ug/news-headlines/42524-cash- squeeze-intimidation-rock-fdc-vote-guarding-plans-2). Omanga, Duncan (2019) WhatsApp as ‘digital publics’: The Nakuru Analysts and the evolution of participation in county governance in Kenya. Journal of Eastern African Studies 13(1): 175–191. Pierskalla, Jan H & Florian M Hollenbach (2013) Technology and collective action: The effect of cell phone coverage on political violence in Africa. American Political Science Review 107(2): 207–224. Privacy International (2015) For God and my President: State Surveillance in Uganda. London: Privacy International (https://privacyinternational.org/sites/default/files/2017- 12/Uganda_Report_1.pdf). Qin, Bei; David Stro ¨mberg & Wu Yanhui (2017) Why does China allow freer social media? Protests versus surveillance and propaganda. Journal of Economic Perspectives 31(1): 117–140. Rauschenbach, Mascha & Katrin Paula (2019) Intimidating voters with violence and mobilizing them with clientelism. Journal of Peace Research 56(5): 682–696. Reuter, Ora John & David Szakonyi (2013) Online social media and political awareness in authoritarian regimes. British Journal of Political Science 45(1): 29–51. Rød, Espen Geelmuyden & Nils B Weidmann (2015) Empowering activists or autocrats? The internet in authoritarian regimes. Journal of Peace Research 52(3): 338–351. Ruijgrok, Kris (2017) From the web to the streets: Internet and protests under authoritarian regimes. Democratization 24(3): 498–520. Ruijgrok, Kris (2020) Illusion of control: How internet use generates anti-regime sentiment in authoritarian regimes. Contemporary Politics 27(3): 247–270. Rydzak, Jan (2018) Disconnected. A Human Rights-Based Approach to Network Disruptions. Washington, DC: Global Network Initiative. Garbe 15 Garbe 857 Rydzak, Jan; Moses Karanja & Nicholas Opiyo (2020) Internet shutdowns in Africa| Dissent does not die in darkness: Network shutdowns and collective action in African countries. International Journal of Communication 14: 4265–4286. Schedler, Andreas (2006) Electoral Authoritarianism. The Dynamics of Unfree Competition. London: Lynne Rienner. Seawright, Jason & John Gerring (2008) Case selection techniques in case study research: A menu of qualitative and quantitative Options. Political Research Quarterly 61(2): 294–308. Sjo ¨gren, Anders (2018) Wielding the stick again: The rise and fall and rise of state violence during presidential elections in Uganda. In: Mimmi So ¨derberg Kovacs & Jesper Bjarnesen (eds) Violence in African Elections: Between Democracy and Big Man politics. London: Zed, 47–66. Ssekika, Edward (2016) Kiggundu: only EC to announce results. Observer, 3 February (https://observer.ug/newsheadlines/42407-kiggundu-only-ec-to-announce-results). Stegmueller, Daniel (2011) Apples and oranges? The problem of equivalence in comparative research. Political Analysis 19(4): 471–487. Stone, Arthur; Saul Shiffman, Audie Atienza & Linda Nebeling, eds. (2007) The Science of Real-Time Data Capture: Self-Reports in Health Research. Oxford: Oxford University Press. Sutherland, Ewan (2019) Law and order in Africa–policing social networks. Paper presented at European Conference on African Studies (ECAS) 2019 (https://papers.ssrn.com/ sol3/papers.cfm?abstract_id¼3402365). Tucker, Joshua A (2007) Enough! Electoral fraud, collective action problems, and post-Communist Colored Revolutions. Perspectives on Politics 5(3): 535–551. Tucker, Joshua A; Yannis Theocharis, Margaret E Roberts & Pablo Barbera ´(2017) From liberation to turmoil: Social media and democracy. Journal of Democracy 28(4): 46–59. Turner, Ralph H & Lewis M Killian (1957) Collective Behavior. 3. Hoboken, NJ: Prentice-Hall Englewood Cliffs. van Ham, Carolien & Staffan I Lindberg (2015) From sticks to carrots. Electoral manipulation in Africa, 1986–2012. Government and Opposition 50(3): 521–548. von Borzyskowski, Inken & Patrick M Kuhn (2020) Dangerously informed: Voter information and pre-electoral violence in Africa. Journal of Peace Research 57(1): 15–29. Waterbury, John (1999) Fortuitous byproducts. In: Lisa Anderson (ed.) Transitions to Democracy. New York: Columbia University Press, 261–283. Weidmann, Nils B & Espen Geelmuyden Rød (2019) The Internet and Political Protest in Autocracies. Oxford: Oxford University Press. Xu, Xu (2020) To repress or to co-opt? Authoritarian control in the age of digital surveillance. American Journal of Political Science 65(2): 309–325. Young, Lauren E (2020) Who dissents? Self-efficacy and opposition action after state-sponsored election violence. Journal of peace research 57(1): 62–76. Zein, Fareed (2010) Using Ushahidi to Monitor the Sudanese Elections (https://www.ushahidi.com/blog/2010/04/26/ using-ushahidi-to-monitor-the-sudanese-elections). Zeitzoff, Thomas (2017) How social media is changing conflict. Journal of Conflict Resolution 61(9): 1970–1991. LISA GARBE, PhD in Political Science (University of St. Gallen, 2021), Postdoc at the WZB Berlin Social Science Center; her research addresses challenges of digitalization for political participation. 16 journal of PEACE RESEARCH XX(X)