Jana Belschner analysed 875,000 Twitter exchanges during Germany's 2021 election. Here, she reveals complex patterns in online toxicity between citizens and elites. Politicians’ behaviour matters, but identity markers also shape experiences of digital political toxicity
Online abuse, hostility, incivility, or violence against politicians has received considerable attention in popular and academic discourse. Studies generally define ‘political toxicity’ as the presence of hostile or offensive language in interactions between citizens and politically active individuals.
Of course, political toxicity is certainly an unpleasant phenomenon for anyone who becomes its target. But what if it also acts as yet another source of political inequality? Does political toxicity predominantly target and affect politicians belonging to politically marginalised groups, further excluding them from online political spaces?
My colleague Linn Sandberg and I analysed every public Twitter conversation between citizens and candidates during Germany's 2021 federal election campaign (875,028 messages in total). We found evidence suggesting a complex pattern of interactive toxicity. Using a continuous, probabilistic definition of toxicity allowing that what is 'toxic' will subjectively vary for different persons, we find that politicians' behaviour substantially predicts the amount of toxicity they receive. Yet, identity markers including gender, migration background, revealed sexuality, and party affiliation shape the form of social media abuse.
This complexity matters for understanding whether digital spaces threaten or enhance political equality. Rather than treating online toxicity as either purely discriminatory or purely triggered by how politicians act online, we need to take into account that multiple mechanisms operate simultaneously.
The clearest pattern in our data concerns reciprocity. Politicians who tweet toxically themselves receive significantly more toxic replies. Right-wing party candidates, who in our sample also tweeted more toxically, faced higher levels of toxicity when citizens replied to their tweets.
When politicians move from non-toxic to highly toxic tweeting behaviour, citizens mirror their tone, with increasingly toxic responses
The responsiveness to tweeting behaviour suggests that politicians exercise meaningful agency over their online environments. The correlation is substantial: when politicians move from non-toxic to highly toxic tweeting behaviour, citizen reply toxicity increases. Citizens appear to mirror the tone politicians set. Importantly, we find that whether a politician receives a toxic reply to their tweet is determined to a larger degree by who answers than by who the politician is. Haters gonna hate, but they do so relatively equally.
While overall exposure rates show limited variation by politicians' identity, the form of toxicity politicians experience differs systematically. We subjected highly toxic replies to candidate tweets to a mixed-methods analysis. Using this method, we distinguished three broad types of attacks: on people, parties, and policies.
The first we define as attacks against the candidate’s person, identity, or appearance. While common insults like you dick or you idiot fall into this category, other attacks are more sophisticated. Examples include go raise your kids to women politicians; fucking communist / filthy nazi to members of a left or right-wing party, shut up you big fat socialist to overweight representatives; go back to your country for those born outside Germany; or comments on a man's perceived sexuality such as you gay pig. Party leaders and ministers – the most visible politicians – are most likely to receive personal attacks.
Attacks on political parties include derogatory hashtags such as #NeverAgainCDUCSU or #fuspd, but also accusations of parties' hypocrisy related to their policy positions. Interestingly, our data shows that candidates of all parties are similarly likely to receive such attacks.
Policy-related attacks express citizens’ dissatisfaction with specific policy decisions, proposals, or positions. Examples include tweets relating to labour-market policy opining that there shouldn’t be shitty mini-jobs needed to survive or, in response to German foreign policy: let us finally stop this idiotic operation. Mali should solve its own problems, not our problem.
Women, candidates with an immigration background, openly queer candidates, and culturally left-wing politicians are more likely to face attacks on their party affiliation or policy positions, regardless of their role, online behaviour, and visibility.
What happens when candidates face persistent online toxicity? We found that receiving a steady stream of toxic replies reduces a politician's subsequent tweeting activity. This is particularly the case among right-wing candidates and those who tweet toxically themselves. This could suggest a self-correcting mechanism: toxic politicians face consequences and moderate their toxic behaviour, or simply tweet less often.
Women politicians may self-censor pre-emptively. Anticipation of harassment can thus silence as effectively as actual harassment
But the interpretation depends on what we cannot observe. Surveys indicate that women politicians perceive toxicity as more threatening despite the fact that they don't receive more public toxic replies. Our research findings back this theory up. If this is indeed the case, it reveals that women politicians may self-censor pre-emptively. Anticipation of harassment can thus silence as effectively as actual harassment. Importantly, our measures capture only those who engaged and then withdrew from Twitter, missing those who never fully engaged.
Our findings resist simple narratives about online toxicity either being purely dependent on politicians’ behaviour or identity. Both dynamics operate, along with others we cannot fully observe.
Toxic communication invites toxic responses. But identity shapes experiences of digital political toxicity in ways that persist regardless of behaviour
The good news is that politicians' behavioural choices matter: toxic communication invites toxic responses. But identity shapes experiences of digital political toxicity in ways that persist regardless of behaviour. And the most concerning dynamics may occur beyond our observational reach, in private harassment and pre-emptive self-censorship.
Most fundamentally, we need epistemic humility about what observational data reveal. While such data are useful in providing a general overview of public online conversations, we must combine them with in-depth evidence garnered via interviews and surveys. This becomes even more crucial when we consider that twenty-first century politicians will find it increasingly hard to avoid social media entirely.
Citizen-elite toxicity determines who has a voice in online political spaces. It is therefore key to understanding representational processes in contemporary democracies.
