Comparing Gendered Exposure and Impact in Online Election Violence: Tunisian Political Candidates Targeted on Facebook

Abstract

Election violence is increasingly taking place online. However, we still do not know much about how such attacks affect the representation of politically marginalized groups such as women. This article develops and applies strategies for analyzing (gendered) exposure to and impacts of online attacks against political candidates. It focuses on the 2019 parliamentary election campaign in Tunisia and combines manual analysis of Tunisian candidates’ public Facebook pages with candidate interviews. We find no gendered patterns in exposure to online election violence in the Facebook data and a low general exposure to attacks. The interview data nevertheless suggests gendered perceptions and impacts of attacks, as well as a perception among the candidates that online election violence is widespread and problematic. These discrepancies highlight that we need a combination of methods and materials to capture the multifaceted nature of online election violence, and in particular those that directly link candidate exposure to impact.

Introduction

There are countless examples of how contemporary politics has moved online. Political processes and discussions are not just affected by, but also increasingly take place on, different internet platforms. Elections are of course no exception: the Internet has changed the ways in which they are administered and carried out, as well as the political communication, information, and campaigning leading up to them. The Internet has also changed the ways in which elections are manipulated and electoral misinformation spread (Persily 2017; Sinpeng, Gueorguiev, and Arugay 2020, Garnett and James 2020), and the online arena is increasingly recognized as a space where intimidation, hate-speech, and threats occur during election campaigns. In this article, we theorize and develop a research strategy for analyzing the gendered patterns of exposure to and impact of election violence on online platforms.

Election violence can be defined as a simultaneous violation of personal and electoral integrity (Bjarnegård 2018, 690). Such a broad definition acknowledges that acts of intimidation, threats, and slurs may disturb election processes. Research has established that online platforms are an increasingly common space for such abuse of politicians (Bardall 2013, Bjarnegård 2018, Bjarnegård and Zetterberg 2023; Collignon and Rüdig 2020, Håkansson 2021; Phillips, Pathé, and McEwan 2023). While it is widely accepted that such acts of psychological violence constitute violations of personal integrity of the target, their wider repercussions for electoral integrity and democratic quality are less well-known.

Research is inconclusive regarding the extent to which online election violence disproportionally targets marginalized groups of politicians, such as women. The results differ depending on the methods used. Quantitative studies of social media data, usually building on large data sets of data scraped from Twitter (now known as X), have generally found a tendency for men to be exposed to a larger extent than their women counterparts (Fuchs and Schäfer 2021; Gorrell et al. 2018, 2020; Greenwood et al. 2019; Rheault et al. 2019; Ward and McLoughlin 2020). In contrast, research that uses self-reported accounts, such as surveys or interviews, report a disproportionate targeting of politicians from marginalized groups, such as women and ethnic minorities (Collignon and Rüdig 2020; Erikson, Håkansson, and Josefsson 2023; Herrick and Thomas 2023; Håkansson, 2021; Phillips, Pathé, and McEwan 2023; Zeiter et al. 2019). We argue that these two methods should be brought together for a more accurate and holistic description of the multifaceted problem of online election violence. Our point of departure is that none of them are wrong, but that they simply shed light on different important aspects of the phenomenon: how and to what extent politicians are targeted with attacks in the online environment (exposure) and how these dynamics are perceived by politicians (impact). To assess both exposure to and impact of online election violence, we conducted two sets of analyses. To examine exposure, we used large amounts of manually coded Facebook data, and for impact we analyzed interviews with political candidates. When we refer to impact in this study, we hence refer to perceived impact building on the politicians’ own impressions of the election campaign.

This research strategy was developed and applied in an analysis of online violence in an election campaign in a relatively understudied context – Tunisia. Tunisia could, at the time of the elections that are in focus (2019), be described as an emerging democracy (Belschner 2021). ¹ Compared to consolidated democracies, the Tunisian political playing field could be described as both more unpredictable and more uneven (Belschner 2021; Hamid 2014). There is a strong tendency toward “digital authoritarianism” in the MENA (Middle East and North Africa) region, which means that online platforms, to an increasing extent, are used by powerful political players to manipulate public opinion and to spread mis/disinformation (Jones 2022). Recent research has also shown how the use of social media among political candidates is reinforcing already existing inequalities (Holm, Skhiri, and Zetterberg 2024). Moreover, Tunisia shares another common feature with emerging democracies around the world, namely the introduction of electoral gender quotas. In Tunisia, a parity law (Piscopo 2016) was installed at the national level in 2014 as part of the new constitution, which mandates all parties, coalitions, and independents to alternate between women and men candidates on their electoral lists (IPU, 2021). Tunisia’s contentious political playing field, characterized by a great variation in strength among political actors, and in which women have had a recent influx in formal politics, thus makes it a relevant context to study for increasing our understanding of the gendered aspects of online election violence.

Our analysis of a selection of Tunisian candidates’ public Facebook pages shows that the exposure to online election violence was rather low and that men and women were targeted to a similar extent. The interview study highlights the importance of type of attacks. While men candidates more strongly normalized online attacks as part of the political game, women candidates expressed particular concern about sexualized degrading talk and comments about their appearance that they received online. Moreover, in contrast to the general low exposure to personal attacks, candidates were often worried about the general occurrence and broad visibility of online election violence.

In the analysis, we pay special attention to a small sample of candidates, for whom we have both Facebook and interview data. While too small a sample to draw any definite conclusion, the possibility to directly link candidate exposure to impact in this way does not only inspire plausible interpretations but also suggests a way forward for research. Used more systematically, it will enable future studies to better capture the multifaceted nature of online election violence.

Our study, thereby, makes three novel contributions to the literature on election violence. First, we suggest a way forward for research on online election violence by combining methods and directly linking candidate exposure to impact. Second, the empirical analysis demonstrates the difference between exposure and impact when it comes to gendered experiences of online election violence. Third, we show how these phenomena play out in an emerging democracy where the political playing field can be described as both more unpredictable and more uneven than in most political contexts previously studied. Hence, the study contributes to our understanding of the gendered consequences of election violence, as well as to how we can capture these. This is important because online election violence can have far reaching democratic implications for the representation of marginalized groups in politics, such as women.

In the following two sections, we present and discuss previous research on - and theorize gendered exposure to and impact of – online election violence. We then discuss how we conceptualize and capture exposure to this type of violence. Thereafter, we present our research strategy and, subsequently, the findings of our two sets of analyses. Finally, we highlight how our research strategy points to a disparity in exposure and impact regarding (gendered) experiences of online election violence and suggest that future research can take this strategy further by systematically linking data on exposure with candidate perceptions.

The Multifaceted Nature of Online Election Violence

Research on the level of abuse and incivility targeting politicians online is burgeoning. It has focused on both everyday abuses directed toward politicians (Bardall et al. 2018; Fuchs and Schäfer 2021; Gorrell et al. 2018; Kuperberg 2021; Rheault et al. 2019; Ward and McLoughlin 2020), as well as on how political candidates are targeted with abuse during election periods (Gorrell et al. 2018, 2020; Greenwood et al. 2019; Stambolieva 2017). Looking closer at online abuse against politicians is pivotal, as research has shown that threats or intimidation on social media constitute the most commonly experienced acts of psychological violence among politicians in various contexts in the Global North (Collignon and Rüdig 2020; Herrick and Thomas 2021; Håkansson 2021).

Most previous research investigating exposure to online abuse against politicians builds on large datasets of scraped social media data from politicians’ official social media accounts (Gorrell et al. 2018, 2020; Greenwood et al. 2019; Ward and McLoughlin 2020). Quantitative machine learning models using large amounts of data can assess the level of abuse that politicians receive online but they are not able to distinguish between different types of violations. Most importantly, automated analysis carries a risk of capturing false positives, meaning that online content is falsely categorized as, for instance, “aggressive” or “abusive.” There are a few examples of mixed methods approaches that combine quantitative and qualitative (manual) analysis of social media data (Fuchs and Schäfer 2021; Southern and Harmer 2021), increasing validity as adjustments can be made for context-specific nuances and interaction dynamics.

Research on exposure to online political harassment that builds on social media data has concluded that the overall level of abuse against politicians online is relatively low, particularly in relation to the amount of civil messages that politicians receive online. Studies on the UK context categorize less than 10% of tweets directed to members of parliament (MPs) as abusive (Greenwood et al. 2019; Gorrell et al. 2018, 2020; Southern and Harmer 2021; Ward and McLoughlin 2020). The number was slightly higher in the North American context (Rheault et al. 2019). An important finding is that online abuse is not evenly distributed: some politicians receive more than others (Southern and Harmer 2021; Stambolieva 2017; Rheault et al. 2019).

This research is inconclusive when it comes to gender distribution of online abuse. Most studies find that men are slightly more targeted with personal attacks and threats on Twitter/X than women are (Gorrell et al. 2020; Greenwood et al. 2019; Rheault et al. 2019; Ward and McLoughlin 2020). Only one study found that women politicians were slightly more likely to be targeted on Twitter/X than men politicians, and in contrast to the other studies, this study did a manual analysis of tweets (Southern and Harmer 2021). However, one report by an organization providing election support in Zimbabwe that used machine learning but included a larger number of platforms than just Twitter/X, found that 60% of the violent posts were directed at women (Bardall et al. 2018). One common finding that stands out is that visibility affects the risk of being targeted with online abuse and it seems as if visible or prominent women are targeted more than men in similar positions, both offline and online (Rheault et al. 2019, Håkansson 2021). ² Visibility is, of course, enhanced by social media presence, as is the likelihood of being targeted on social media platforms (Theocharis et al. 2016).

In contrast to studies that focus on exposure using platform frequency categorizations, analyses that build on self-reported accounts demonstrate higher exposure in general (Zeiter et al. 2019), as well as more pronounced gendered patterns where women politicians perceive online attacks to be a greater problem than men politicians do (Collignon and Rüdig 2020; Erikson, Håkansson, and Josefsson 2023; Håkansson 2021; Zeiter et al. 2019). The impact of online political harassment may also be gendered regardless of whether it disproportionately targets women or not. Research has demonstrated that women politicians receive more personal attacks including sexist and gendered slurs, objectifying comments about their appearance, and stereotypical claims about their characteristics, while men receive more general, political, and racist abuse (Fuchs and Schäfer 2021; Gorrell et al. 2020; Southern and Harmer 2021). Hence, although the literature on exposure has shown a tendency for male politicians to be more exposed to online harassment than their female counterparts, studies focusing on impact suggest that women politicians are more affected by the attacks and more likely to adapt their behavior (Collignon and Rüdig 2020; Håkansson 2024).

In terms of country contexts, there has been an overwhelming focus on online harassment against politicians in Western contexts and/or in consolidated democracies, mostly the UK (Gorrell et al. 2020; Southern and Harmer 2021; Stambolieva 2017; Ward and McLoughlin 2020), but also Japan (Fuchs and Schäfer 2021), North America (Rheault et al. 2019), and Italy (Pacilli and Mannarini 2019). While there is a strong interest in these issues from organizations working with electoral support (Bardall et al. 2018; Greg-Obi et al. 2019; Zeiter et al. 2019), it has not been sufficiently supported by academic research. The overwhelming focus on the Global North in combination with relatively easy access to Twitter/X data for researchers has also contributed to the fact that Twitter/X is the platform that is, by far, the most studied (e.g. Fuchs and Schäfer 2021; Gorrell et al. 2018, 2020; Greenwood et al. 2019; Rheault et al. 2019; Southern and Harmer 2021; Stambolieva 2017; Ward and McLoughlin 2020), while there are much fewer studies on the level and type of online abuse against politicians on Facebook (Pacilli and Mannarini 2019). This is a potentially problematic focus considering the dominant position that Facebook has in many countries, particularly in the Global South. While Twitter/X reached 436,4 million users in 2022, Facebook had 2.91 billion monthly active users and is the world’s most active social media platform (Datareportal 2022). In Northern Africa, it is estimated that 104 million people are Facebook users (Statista 2022a), while only 8 million are Twitter/X users (Statista 2022b).

While digital media platforms such as Facebook present oppositional groups in authoritarian contexts with new opportunities to mobilize, they, to an even greater extent, provide repressive governments with efficient tools for monitoring and suppressing such opposition (Sinpeng 2020). Regarding the MENA region, it has been argued that the digital optimism during the Arab Uprisings that began in 2010 has transformed into a “digital authoritarianism.” This means that social media has become increasingly used by authoritarian governments within the region to monitor and control their populations, repress potential uprisings, manipulate public opinion, and to spread mis/disinformation (Jones 2022). Even though Tunisia emerged as the only democracy after the Arab uprisings, its most powerful political players stalled important media regulations during the transition period that would have led to a more diversified media/online sphere. This has, in turn, enabled those with power and resources to manipulate the public discourse and to spread mistrust and misinformation, and not only through online platforms (Bennett 2023). Several civil society organizations monitoring the 2019 elections also reported that the legislative election period was characterized by attacks on political opponents, disinformation, and misogyny on Facebook (DRI 2019; IRI-NDI, 2019). From an electoral integrity point of view, it is important to assess the attacks that take place in the online political arena and to understand how they affect political campaigning and representation and how these phenomena play out in an emerging democracy. By combining scraped social media data with self-reported accounts in the Tunisian context, we would get a more nuanced and context-sensitive assessment of online violations against electoral integrity

Theorizing Gendered Exposure and Impact

Frameworks for categorizing different elements of gendered political violence emphasize that gender differences can appear in patterns of exposure, forms of violence, and the impact of violence (Bardall et al. 2020). Focusing on impact implies acknowledging “the subjective meaning-making processes that occur as different audiences react to political violence” (p. 916). While we should be careful to interpret perpetrators’ motives based on the consequences of violence (Eriksson Baaz and Stern 2013), the impact and gendered interpretation of violence are important in their own right. Subjective understandings of violence are shaped by one’s positionality and experiences of vulnerability. Masculinity theory outlines how violence and aggression are valorized and form part of masculine ideals (Kimmel et al. 2005, Bjarnegård et al. 2023). Processes of normalization likely occur, so that many men are socialized into expecting violence as part of their political life. Feminist research on violence has also posited that victims of violence tend to understand a wide range of incidents as connected and aligned on a continuum (Kelly 1987, Bjarnegård 2023). In so far as gendered positionality shapes reactions to, interpretations of, and strategies for preventing political violence, it will also affect the inclusiveness of the political processes in which the violence occurs.

When politicians partaking in an election are targeted, it constitutes a simultaneous attack on personal and electoral integrity, decreasing democratic quality in several ways (Norris 2013, Bjarnegård 2018, Bjarnegård, Håkansson, and Zetterberg 2022). Studies of online abuse against politicians tend to build on the assumption that such abuse mainly limits the representative role of the politician it is directed to. While the impact on the direct target is, of course, important, online abuse against politicians during election periods has additional important components with potentially broad implications for the democratic process. When electoral politics move online, the number of actors involved is expanded (Garnett and James 2020). Online abuse thus reaches far beyond the targeted politician. This highlights the necessity of focusing on exposure as well as impact in order to capture online election violence and its potential wider repercussions for electoral integrity. While we need to map the prevalence of violent incidents, we also need to understand how the fact that these online personal attacks are visible to supporters and opponents alike affect their impact.

Gendered attacks are, thus, likely to hurt a much larger community than the individual target (Krook and Restrepo Sanín 2019, Erikson, Håkansson, and Josefsson 2023). These insights resonate particularly well with online election violence, where such impacts are likely to grow exponentially.

Conceptualizing and Capturing Exposure to Online Election Violence

Some of the inconclusive results reported in the section on previous research regarding the exposure to online election violence may also be attributed to the differences in categorization of online abuse and harassment, often depending on the focus of the study. We build on previous research on gendered political violence, as well as communication theory, to construct a coding scheme that captures online election violence and its gendered aspects. Online violence implies, by definition, psychological rather than physical violence (Bardall 2013). Psychological violence can be defined as acts that deliberately inflict, or have a likelihood of inflicting, “trauma on individuals’ mental state or emotional well-being” (Krook and Sanín 2020, 744). To capture psychological violence, we assess if the comment includes degrading or hurtful libel and rumors directed at a politician, and/or threats about physical violence, such as death threats and threats of rape, and thereby is intentionally seeking to cause harm to the individual politician and thus to the political process.

In addition to psychological violence, we broaden our definition of election violence to also include certain uncivil behaviors. ³ Inspired by a conceptualization of the democratic consequences of online incivility developed by Papacharissi (2004), incivility can be defined as “the set of behaviors that threaten democracy, deny people their personal freedoms, and stereotype social groups” (Papacharissi 2004, 267). Applied to the interactions between politicians and constituents in online discussions, incivility has previously been operationalized as assigning stereotypes, acts of silencing, questioning their position as an MP or as a political candidate (e.g., imploring them to resign or otherwise leave the political sphere), threats to individual rights (such as personal freedoms and freedom to speak), and threats of physical violence (Southern and Harmer 2021). While the latter two, by definition, are included in our operationalization of psychological violence, we follow this operationalization to also capture uncivil comments that assign negative stereotypes to the candidate, and/or comments that engage in silencing practices or question the candidate’s position. To be characterized as election violence, comments should simultaneously violate the personal integrity of the political candidate on whose page it is posted, as well as the electoral integrity. This is, for instance, the case if a post is stereotyping a social group and, at the same time, targeting a candidate that is a representative of that group. Moreover, in the context of elections, it is important to distinguish election violence from impolite but legitimate political critique, which may be uncomfortable or unfair, but has a political purpose. For instance, an online rumor about a candidate’s corrupt behavior is different from a post generalizing corrupt behavior to a social group, i.e. stereotyping. Whereas the former is potentially relevant information for voters, safeguarding democracy, the latter harms democratic principles, effectively threatening a collective and limiting the opportunities for the group in question. Due to the democratic role that politicians have in combination with the very public forum that social media platforms constitute, we assess personal attacks against politicians somewhat differently from Papacharissi (2004). While she might see attacks on politicians’ personal characteristics or appearance as examples of rudeness, we characterize them as violations of personal and electoral integrity, and, thus, as instances of election violence. This is because attacks that target the personal integrity of a politician in order to affect his or her possibilities of getting elected and representing certain groups may have negative consequences for the individual candidate, broader social groups, and democracy. Therefore, we also include comments that target the candidates’ appearance and characteristics in a derogatory manner as part of election violence.

Data and Methods

In this section, we present our research strategy, which combines a statistical analysis of comments on Facebook public figure pages with in-depth interviews with a selection of candidates in the 2019 parliamentary elections. However, before doing so, we present the specific context of our study: online campaigning in Tunisia.

The Tunisian Context: An Unpredictable, Uneven, and Contentious Political Playing Field

Studying election violence on Tunisian politicians’ Facebook pages brings the analysis of online election violence to a new and highly relevant context. Compared to most analyses of consolidated democracies, the Tunisian political context is characterized by volatile party politics and political turmoil. Since the Tunisian revolution and the overthrow of President Ben Ali’s authoritarian regime in 2010-2011, Tunisia has had a fragmented party system in which new parties constantly emerge and established ones quickly can fall apart. Similar to other emerging democracies, there is also great variation in organizational strength among parties. The Islamist party Ennahda, which emerged as the main challenger to Ben Ali’s regime in the 1980s and early 1990s, has dominated the Tunisian political scene after the revolution and it is one of the few parties that can be described as resource strong (Belschner, 2021; Hamid 2014). The other larger party, the former governing party Nidâa Tounés, fell apart after internal conflicts during the run-up for the 2019 elections (Martin and Carey 2022). There is also a large number of small parties with few organizational resources (Belschner 2021). This study focuses on the online election campaign of the 2019 general elections where nearly 15,000 candidates from over 1,500 party-, coalition-, and independent lists competed for the 217 parliamentary seats (IRI-NDI 2019). Following the adoption of the electoral gender quota at the national level in 2014, every second candidate on the electoral lists needs to be a woman. This is, however, not applied to the top position; also, in the 2019 parliamentary elections, parties, coalitions, and independents were reluctant to put women on the top of their lists. This led to only 14 percent of the lists being headed by women (IRI-NDI 2019, 6). It has also been shown that lack of organizational strength makes it difficult for parties to comply with the electoral quota in the Tunisian context (Belschner 2021). Thus, only 26 percent of the elected representatives in 2019 were women (IPU 2021).

Tunisia is also a relevant context for studying online election violence because online platforms have played a central part in Tunisian politics and public debates for a long time. By the early 2000s, Tunisia had unusually well-developed telecommunications infrastructure, as well as internet markets, which enabled internet access among the population. During the revolution, social media played an important part in facilitating information diffusion and coordination among ordinary Tunisian citizens (Bennett 2023). Still, in comparison to other countries at similar levels of development in the region and beyond, a relatively large proportion of the Tunisian population (68.4%) has access to the Internet. Most of these are also Facebook subscribers (Internet World Stats 2021). In contrast, only 4% are on Twitter/X and not even 0.5% on Instagram (Statcounter GlobalStats 2021). Facebook was widely reported to play an important role during the campaign period for the 2019 elections as the main social media platform used by the candidates, parties, and lists to communicate with potential voters (DRI 2019; Elswah and Howard 2020; IRI-NDI 2019).

Collection of Facebook Comments

The statistical analysis builds on a sample of over 23,000 Facebook comments that were captured from individual candidates’ public figure pages and then coded manually. The data was purchased from Brandwatch, a company focusing on social media monitoring and analytics, and downloaded and stored in Excel-files. Data collected for each comment include: publishing date, link to comment, post type, post title, and content. Upon storage, the Excel-files containing comments have been encrypted. We focus on Facebook’s public figure pages because these Facebook pages are specifically designed for the purposes of public figures and have many features that help political candidates promote themselves in a more professional way vis-à-vis potential voters. These are therefore frequently used by politicians. Moreover, Facebook only allows for systematic data collection from public profiles/pages, thus it is not possible to collect this type of data from private profiles.

To create our sample of comments, we identified all women candidates elected to parliament in the 2019 elections who had used a public figure page during the election campaign. We then used our initial sample of elected women and matched these with elected men who also had public figure pages. Firstly, we sought to include men candidates from the same party in the same district, but if this was not feasible, we included men candidates from the same party, but from similar districts regarding size, development level, percentage of women elected in the last national elections, and the size of the two main parties (Ennahdha and Nidâa Tounés). We then matched the sample of elected candidates with non-elected men and women candidates in a similar manner. The politicians’ Facebook profiles were identified manually by using Facebook’s own search function. To make sure we had identified the right candidate’s Facebook profile, we employed several strategies. Firstly, we tried different spellings of each candidate’s name in both Arabic and Latin, and compared pictures on the Facebook account with other pictures from the campaign period. We also searched for information about the candidate’s district and party on their Facebook page, and compared this with the information about the candidate in the official list of candidates that we had obtained from the central electoral authority (Instance Supérieure Indépendante pour les Elections – ISIE). Altogether, this gave us a sample of 89 elected and non-elected candidates with public figure pages (45 women candidates and 44 men candidates).

We captured all the comments from these candidates’ official Facebook public figure pages during the three-week official election period, i.e., when campaigning is permitted in Tunisia. Because we noticed that there was a spike in the number of comments around election day (October 6), we, in addition, extended the sampling period to a week after the elections to not miss any relevant material. Among the candidates, there was great variation in how many comments they received during the official campaign period, ranging from a dozen comments to tens of thousands. To make a manual analysis of the comments feasible, as well as the amount of analyzed comments for each candidate more comparable, we took a random sample of 500 comments from the candidates that had over 500 comments in total by using the randomizer function in Excel. Our final sample contained 23,321 comments, of which men received 54 percent and women 46 percent.

Coding

The comments were manually analyzed using a coding scheme constructed to capture a range of negative comments that were defined as simultaneous violation of personal and electoral integrity. This included comments that targeted the candidates’ appearance and characteristics in a derogatory manner, comments that assigned negative stereotypes to the candidate, comments that engaged in silencing practices or questioned the candidate’s position, and psychological violence (threats about physical violence [e.g., death threats], threats about sexual violence [e.g., rape], and libel or rumors of a sexual or non-sexual nature). ⁴ All in all, this led to six variables, and all variables were measured as binary. Almost all comments were in Arabic and therefore coded by two Arabic native-speaking coders (of which one is Tunisian native-speaking). We first conducted a pilot study on a smaller sample of the material, which was followed by a debriefing with the research team before the full sample was coded. We, in addition, had continuous debriefs during the coding process as issues arose in relation to specific parts of the material. We think such debriefings increase the reliability of the coding.

After the whole sample had been coded by our Tunisian native-speaking coder, we did an intercoder reliability check on a random sample of 2200 comments, which was recoded by our second coder. This showed that 94.5 % of the recoded sample corresponded with the initial coding. Because all comments that had been incorrectly coded were found among the comments that had been categorized as election violence, our second coder then recoded this whole sample of comments in order to strengthen the reliability of the analysis.

Measurements and Statistical Analysis

We conducted two sets of statistical analyses of the Facebook data: a descriptive analysis and a regression analysis. The descriptive analysis contained two parts. First, it looked at the prevalence of election violence as a proportion of all the analyzed comments on the Facebook public figure pages. Second, it included a candidate-level descriptive analysis to understand the distribution of comments that can be categorized as election violence among the selected candidates.

The regression analysis built on these analyses to examine variation in experiencing election violence. The unit of analysis in this analysis is comments on the candidates’ public Facebook pages. Our main dependent variable is violations of personal and electoral integrity, hence election violence. To reiterate, this includes negative stereotypes related to the candidate, silencing practices, questioning the candidate’s position, degrading talk about the candidates’ personal characteristics or appearance, and psychological violence. When we present the regression analysis below, we have coded the data according to whether a comment includes any form of election violence. ⁵

Our main independent variable was gender. We coded men as 1 and women as 0. We also included incumbency (coded 1; non-incumbents are coded 0) as an independent variable. Previous research has shown that high visibility of politicians increases exposure to psychological violence, especially among women politicians (Håkansson 2021). Incumbents are likely to be more visible than candidates who were not legislators at the time of the elections. Therefore, we expect incumbency to have a direct relationship with election violence and, potentially, to moderate the relationship between gender and election violence (as men are more likely to be incumbents than women).

As our dependent variable is binary (a comment either includes election violence or not), we use logistic regression. We present both bivariate and multivariate analyses. We include robust standard errors clustered at the level of the candidate to correct for heteroscedasticity and autocorrelation problems, respectively. In the multivariate analysis, we include party fixed effects to make sure that we compare the comments of candidates representing the same political party. Otherwise, there is a risk that any observed gender differences are a result of men and women representing different parties, some of which receive more negative comments than others. Ideally, we would have liked to control for age, as young people (and perhaps young candidates) are more likely to use social media than older people (and older candidates). Unfortunately, the candidate lists in Tunisia do not include such information.

The findings we present below are robust to various model specifications. For instance, our findings are no different when we include candidates’ positions on the list (coded as their number on the ballot) in the model. This variable refers to a kind of visibility too, as Tunisia’s closed proportional-list system makes top-ranked candidates (i.e., those heading candidate lists) more visible during the campaign than their fellow partisans placed at a lower position. Moreover, the findings are not significantly different if we run a multilevel logistic regression model that explicitly addresses the hierarchical structure of the data. ⁶

Interviews

To capture the perceived impact of online election violence among the Tunisian candidates, we conducted semi-structured interviews with 19 candidates, both men and women, and both winners and losers. To have a large as possible exposure to different perspectives, we also aimed for a variation regarding political parties, districts, and list positions similar to the Facebook data. ⁷ To get in contact with relevant interviewees, we used a snowball-method, whereby we asked the candidates interviewed for contact details to new potential interviewees. ⁸ All candidates were asked similar questions about their exposure to online election violence and how this had impacted them. We asked about the extent to which negative interactions had occurred and through which online channels these mainly had taken place. We also asked which type of negative comments they had received during the election campaign, if/how these types of comments had affected their campaign and/or well-being, and if they had any ideas about who the perpetrators behind these instances of online election violence were. The research team carried out the interviews online (because of the pandemic) between November 2020 and June 2021, mostly via messaging services like WhatsApp and Facebook Messenger. All interviews were conducted orally, and lasted around 45-75 minutes. The lion’s share of the interviews was conducted in Arabic (only one was conducted in English), and they were then translated to English upon transcription. All transcripts were coded in NVivo. In the analysis, the interviews provide a deeper understanding of the results of the regression analysis of Facebook comments by allowing for an examination of the context in which violent interactions online occur from the perspective of the political candidates. This includes candidate perceptions of being targeted with different types of negative comments, which consequences such attacks have for their political work, and who they perceive to be the main perpetrators of online violence.

Because the candidates interviewed were varied in a similar manner as the candidates whose Facebook pages we analyzed, some of the interviewees (eight, to be exact) overlapped with the Facebook data. To allow for a direct comparison between level of exposure to online election violence and its impact on the candidates, we conducted an in-depth analysis of the eight interviews with candidates who were also included in the statistical analysis of comments on the candidates’ public Facebook pages. This is a small sample that cannot provide conclusive accounts, but given the importance of directly linking exposure to impact, we consider it a promising example of how research on online violations of electoral integrity should be conducted in future studies.

Analyses

Facebook Comments

Descriptive statistics clearly demonstrate that, just like research conducted in the Global North, the number of violent comments is very few. Only 339 (or less than 1.5 percent) of the more than 23,000 comments on the Facebook pages were coded as violent. Derogatory comments about the candidates’ personal characteristics were the most common type of violent act: 220 (or 65 percent) of the violent comments were of this type.

As a second descriptive analysis of our matched sample, we looked at the distribution of violent comments among the candidates. It could be that very few candidates received most of the violent comments, or that these types of comments were more equally dispersed between all candidates. This could have important consequences for how exposure to attacks is perceived by the candidates themselves. When analyzing exposure to online attacks at the candidate level, we find that almost half (46%) of the 89 candidates in our sample received no violent comments at all. Almost as many (38%, or 34 of 89 candidates) received 1-10 violent comments, and approximately 12 percent of the candidates in our sample received over half of the violent comments. Hence, the violent comments are, on the one hand, very unequally dispersed between the candidates, and, at the same time, most of the candidates in our sample received at least a few instances of online violent attacks during the campaign. Moreover, there are few gender differences when it comes to the number of violent comments received by individual candidates (Figure 1). Hence, our candidate-level descriptive analysis suggests that men and women candidates are exposed to online election violence to a similar extent. ⁹

Figure 1. Exposure to online election violence at the candidate level.

Next, we moved on to a regression analysis to get a more comprehensive picture of the variation in exposure to online election violence among candidates in Tunisia. More specifically, we looked at the extent to which candidate sex (and incumbency) mattered for the exposure of such violations.

Model 1 in Table 1 shows the bivariate relationship between gender and exposure to violent comments on Facebook. As expected from the descriptive analysis, it shows no gender differences: men candidates are slightly more exposed, but the relationship is not statistically significant.

Table 1. Candidate sex and exposure to online election violence.

Notes: Logistic regression coefficients. Fixed effects (FE) for political parties. Robust standard errors clustered on individuals in parenthesis.

*** p <.01; ** p <.05; * p <.1.

Moving to our multivariate model, in which we also included incumbency as well as fixed effects for party (in order to compare the comments targeting candidates from the same political party), Model 2 in Table 1 shows that gender is still not statistically significant. However, incumbents are significantly more likely than newcomers to receive a violent comment on their Facebook page. Thus, within a political party’s pool of candidates, those who were already parliamentarians at the time of the elections were more exposed than others. ¹⁰

Overall, our analyses suggest that Tunisian political candidates with public Facebook pages to a very large extent receive non-violent feedback from their constituents. The violent comments are, however, unequally dispersed among the candidates and a small proportion of the candidates receives most of these comments. The gender of the candidate does not seem to matter, but the visibility of the candidate does: incumbents are more likely to be the targets of online election violence.

Candidate Interviews

Comparing the statistical analysis with the interviews corroborated the finding that visibility is a more important factor than gender for being exposed to attacks. However, while no gendered patterns were discerned in the Facebook data, the interviews nuance this result in at least two important ways. First, it was more common that men candidates described online attacks as a natural part of the political game that they tolerated as such. Second, the psychological consequences of online election violence seemed to be more severe for women than for men candidates. When directly linking the Facebook data to the smaller sample of the eight candidates whom we also interviewed, it is revealed that candidates that received no or very few uncivil comments on their own public Facebook page were also affected by online violence targeting other candidates on Facebook, or that targeted themselves but in other online spaces.

Exposure and Candidate Visibility

Regarding the extent to which the candidates claimed to have experienced online election violence, there was a clear difference between well-known candidates, and those candidates who were less well known to the public. Well-known candidates, such as party leaders or other well-known Tunisian politicians, who had a relatively large number of Facebook followers on their public figure pages or private profiles, commonly stated that they had received large amounts of libel and defamation on their own Facebook profiles, or via private messages (e.g., interviews W6, M8, M12, W13, W10). These experiences were similar among both men and women candidates. To be targeted online with defamation, libel or threats was also commonly understood as a consequence of their visibility vis-à-vis the public. As one man candidate put it:

You do not attack those who are unknown. I am attacked because I am a “famous face”, someone who has held several positions. I was a member of the Constituent Assembly, I have been in the government, I have been a Member of Parliament for a few years now, I am present in the media… And when you are famous, you are a target for attacks (interview M8).

While several of the more well-known candidates said they had experienced a large amount of online violence, other candidates had not experienced many negative interactions at all. For example, two of candidates who were not elected to parliament stated that they had not received any negative interactions online at all on their Facebook pages (interviews M3, W19). The man candidate that had not received any negative comments said that this was clearly a consequence of his activity on Facebook having been very limited, and that, overall, he had had few interactions with potential voters online (interview M3).

Gendered Impacts of Online Election Violence

While we found no gendered patterns in exposure according to the statistical analysis of Facebook data, the interviews suggested that the impact differed between men and women candidates. It was fairly common among the candidates interviewed to describe online election violence as part of the “political game”; it was something you had to live with being a politician. This perception of online attacks could partly be understood considering that the candidates primarily perceived those behind defamation, libel, and threats to be political opponents - in some cases, competitors from their own party, but in most cases, from other political parties. This was a general perception among all of the interviewees: both men and women candidates and candidates from all political parties. As one woman candidate from the Ennahdha party explained:

It is either a political opponent, hence a candidate from a competing party or people who have something against Ennahdha. They do not have anything against one personally, but against the party itself. So again, either someone within the framework of political competition between the parties or someone outside the parties has a historical or ideological hatred of the party (Interview W6).

However, although both men and women candidates stated that online attacks were part of the political game, this idea was brought up more strongly by the men candidates. Only one of the men candidates (a non-elected man) claimed that the comments had affected him and led him to shut down his Facebook profile after being accused of infidelity (interview M16). All the other (elected) men claimed that these kinds of incidents were, rather, a normal part of the political game; these were attacks that you needed to accept and to live with if you wanted to be a politician. As two of the men candidates put it:

You have to accept the rules of the game…. This is what happens when you run in an election or any other kind of “fight.” Because what is an election? It is a fight between opponents about who will persuade most of the electorate (interview M7).

For me, it is part of the risks that you take when you do what we do (politics). If you are part of the public sphere, you have to accept the attacks (interview M8).

While the men politicians, to a large extent, thus normalized incidents of online election violence, the perceptions of the impact of online election violence differed to a larger extent among the women politicians. One woman candidate elected to parliament, for example, first stated that she had received a lot of defamation. Later in the interview, she, however, claimed that she “had not taken them into account” (interview W13). Another woman candidate stated that she had experienced several attempts to sabotage her campaign (online) but that this was something “normal”, something that she expected from the campaign (interview W6). It was, however, more common among the women candidates to find different strategies to not let the derogatory comments “get to you”:

During the campaign, I have just kept working and stayed in contact with the voters. I have not taken this [derogatory comments] into account at all (interview W14).

Honestly, it has not affected me. It is true that it hurts, I’m still new [in politics]: that all eyes are on you and that everyone is so focused on you was something new to me. But God gave me strength. For a while, I just read them and moved on. I didn’t even answer, I saved my energy to work more and use the time left in the campaign to introduce our program and myself to the people (interview W17).

There were also accounts among the women candidates that the derogatory comments did, indeed, cause psychological harm to them and that they struggled with how to handle this (even though it was viewed as part of politics). As two of the women candidates explained:

I am affected, because after all you are a human being. I’m usually positive and optimistic but sometimes, when I read some comments, I can feel some kind of anxiety. But I usually gather myself afterwards and move on, stronger. (…) Nowadays I prefer not to read [the comments], I only read what is about a problem that I could help with and ignore the rest (interview W10).

Of course, it can be mentally difficult. It’s normal, we are ultimately human, it can affect us psychologically sometimes… But in the end it is something that contributes to your political personality (interview W6).

The differences in impact between men and women candidates suggested by the interviews could be related to the type of defamation or libel they had experienced. While men more commonly stated that they had been accused of corruption, sexual infidelity, or other types of libel that attacked their character (in a typical male gendered way), some women stated that online attacks against them had included sexualized or modified images of them, sexualized degrading talk, as well as attacks against their appearance because they were not perceived as feminine enough (e.g., interviews W10, W14). These accounts suggest that to be attacked in a more policy related and/or male gendered way in contrast to a more personal and explicitly female gendered way could have important consequences for how such attacks are perceived by targets. While it is important to keep in mind that these results only build on a few accounts, the discrepancy in how men and women politicians have been attacked is in line with the findings of previous research. Several studies show that men politicians more frequently are attacked for policy related reasons while women politicians are facing much more gender-based personal attacks on social media, including offensive gendered slurs that can be categorized as hate speech, as well as sexualized attacks (Erikson, Håkansson, and Josefsson 2023; Gorrell et al. 2020; Phillips, Pathé, and McEwan 2023; Ward and McLoughlin 2020). Our findings, thus, suggest that the gendered differences in online attacks against politicians can also have important gendered consequences, and that this phenomenon is in need of further investigation. While there are a few studies on the gendered consequences of online attacks showing that women candidates are more likely than men to adjust their campaign strategies (Collignon and Rüdig 2020) and withdraw from political debates online (Håkansson 2024), no study has, hitherto, explicitly linked the form of attack with its subsequent consequences and analyzed to what extent this phenomenon is gendered.

Directly Linking Facebook Data to Candidate Perceptions

To understand how exposure to online attacks is directly related to their impact, we compared the accounts of the eight candidates that were included both in the statistical analysis and in the selection of interviewees (5 women and 3 men). Among these candidates, there was one candidate who had received over 10 violent comments, whereas about half of them had received just a few or no such comments at all, respectively. It was, however, clear from the interviews that no matter how many online attacks the candidates had been exposed to on their own Facebook pages, all of them perceived online violations to personal and electoral integrity to be a real and growing problem.

Although only a few of the interviews overlapped with the Facebook data, we used them to discern two possible explanations for these seemingly contradictory accounts. First, candidates who had not experienced any or very few attacks on their own Facebook accounts nevertheless often talked about how they had witnessed other candidates, often more visible candidates within their own party, be frequently exposed to such attacks on their respective Facebook pages (e.g., interviews W19, M15, see also W14). One candidate that had “only” received one violent comment directed at himself nevertheless claimed that he had experienced “verbal violence” online: “It was rarely expressed on my own (Facebook) page, there has never been anything abusive there. But I saw some verbal violence on other pages and in comments directed at other candidates” (interview M15). Second, attacks that targeted the candidates were carried out in other online spaces than their own Facebook accounts. Several of the candidates stated that, although they had not received any violent comments on their own public Facebook pages, such attacks were often carried out on fake Facebook pages created only for the purpose of intimidating and defaming candidates (e.g., interviews W1, M15). Such fake profiles were mostly created for well-known candidates that had a chance to win in the election (interviews W10, M15). Moreover, some of the candidates claimed that attacks were carried out on other larger Facebook pages, such as the party’s main page in the district. One women candidate had not received any violent comments on her own Facebook page, but still claimed that she had been targeted with defamation and slurs: “It was not on my own page. There are well known (Facebook) pages (of the party) that belongs to the district. When the (party’s) lists are being published there are some people that shows support, but others are starting to insult (you) and to sabotage” (interview W17). Lastly, attacks were carried out in more private forums, such as messengers through WhatsApp or Facebook messenger (e.g., interviews W6, M 8). These accounts suggest that online election violence has repercussions beyond the direct targets, but also that we need to focus on other types of online venues for election violence than the candidates’ own social media accounts.

Concluding Discussion

In this paper, we have developed and applied a mixed-methods approach to investigate (gendered) online election violence targeting political candidates in the 2019 Tunisian parliamentary elections. To investigate exposure to online attacks, we analyzed comments posted on Tunisian candidates’ public Facebook pages during the campaign period, and to examine their impact, we explored candidates’ own perceptions of online attacks during the election campaigns. In line with research on exposure to online attacks against politicians in consolidated democracies, we found a relatively low general exposure to attacks but that some politicians received considerably more abuse than others (Greenwood et al. 2019; Gorrell et al. 2018, 2020; Southern and Harmer 2021; Ward and McLoughlin 2020; Rheault et al. 2019). In contrast to these studies, however, we find no gendered patterns in the Facebook data, even when considering the visibility of the candidates. Yet, the analysis of (perceived) impact paints a different picture: most candidates perceived online election violence as a widespread and growing problem and (a few) women candidates perceived its impact as psychologically harmful. The combination of methods, and their distinct results, demonstrate that we need to focus on both exposure and impact in order to capture and understand the democratic implications of online election violence.

Above all, the discrepancy between assessments of exposure and impact may shed light on the fact that a large amount of positive or neutral Facebook content does not balance out the violations against personal and electoral integrity that even a few negative and damaging posts can cause. This is also suggested by the analysis of the interviews with the small number of candidates for whom we also had overlapping Facebook data. It showed that even the candidates who received very few or no such comments still perceived that they were affected by the few comments they received or by online attacks against others. The public nature of Facebook implies that candidates are exposed to online election violence, even when it is not directed at them personally. Moreover, the online attacks against the Tunisian politicians often took place on Facebook pages other than their own.

While our research strategy shows the importance of combining different data and methods when studying online election violence, there are also some steps that need to be taken to improve the strategy. The discrepancy between the low level of violent comments found within the sample of Facebook comments and the (gendered) perceptions of online attacks captured through the interviews points to the need for additional material and methods. First, we need to know more about how candidates are affected by the occurrence of online violations of electoral integrity during an election, no matter which candidate those attacks target. Second, we need to extend our studies to types of online spaces other than the candidates’ own social media accounts, which hitherto have been the main empirical focus in studies of online attacks against politicians. This may be even more important in contexts characterized by contested elections and contentious politics, such as Tunisia, where the political playing field is (relatively) uneven and where it is more likely that powerful political players try to manipulate elections from unofficial social media accounts, including attacks on political opponents. Third, considering the gendered impact of attacks suggested by the interviews, it is important to further investigate how the type of attack one is exposed to is related to its impact. This includes the extent to which online election violence is constructed as a normal part of the political game by those (male) candidates that themselves are part of the political norm, and how attacks against women are more personal, including sexualized imagery, as well as attacks on their appearance. Continuing the small-scale study that we did here, directly linking online exposure to impact is a promising way forward. A limitation with our study, however, is the under-exploitation of this direct link in the design. To improve this research design, it would be beneficial to start with the quantitative analysis of social media data, and then draw a number of relevant politicians to interview from that data.

Lastly, while there were no clear gendered patterns in violations of electoral integrity on Facebook, incumbents were more likely to be targeted. The interviews corroborated these results, as the candidates themselves saw online violence as something that comes with political visibility. However, in contrast to previous research showing that visible women are targeted with political violence to a larger extent than visible men (Håkansson 2021; Herrick and Thomas 2022; Rheault et al. 2019), we found no gender differences in exposure among visible Tunisian politicians neither in the quantitative nor in the qualitative analysis. It is, however, important to note that we measure visibility differently from previous studies, through incumbency in the quantitative analysis and online presence/activity in the qualitative part. Research showing that visible women politicians are more targeted than their male counterparts have previously focused on position in the political hierarchy and media visibility (Håkansson 2021), legislative professionalism (Herrick and Thomas 2022), or online visibility in terms of number of followers on their social media accounts (Rheault et al. 2019). Our results, thus, suggest that that it can be important for future research to both broaden the view on what visibility is, and to be specific about what type of visibility is studied in relation to gendered exposure and impact of online election violence.