Bad News Travels Well ꟷ Incentives for Candidates to Go Negative on Twitter

Does Twitter campaigning incentivize negativity? Since the 2016 US presidential election, Twitter has risen to prominence and notoriety as a medium of uncivility and personal attack. A great deal of the debate is related to selective exposure and so called echo chambers, in which consumers self-select into ideologically homogeneous communities. Empirical findings remain mixed and while this demand argument has gained great attention, the supply side is just as important. We argue that the logic of Twitter attention interacts with audience effects and incentivizes negative tone. Following the mediatization literature politicians adapt to the medium by changing their tone depending on the feedback they receive. Following the echo chamber literature we argue that this mediatization effect is conditional on the follower structure. Previous research shows that more emotional and negative tone leads to more likes and retweets, increasing visibility for other users and journalists. This feedback is directly observable for politicians: they know which Tweet got a lot of attention. Since candidates aim to get the attention of potential voters in the broader public, they are incentivized to increase reach by going negative. These incentives vary between politicians and parties and interact with audience structure. On Twitter, candidates have self-selected followers who receive their messages regularly. For less known candidates with little media attention, followership functions like an amplifier and political audience at the same time. The smaller the audience, the stronger is the incentive to go negative. While ideologically homogeneous followers send clear signals about the tone demanded, mixed followerships send mixed signals. This makes candidates with heterogeneous followers cautious and refraining from extreme language. The more ideologically heterogeneous the followership, the weaker is the incentive to got negative. We test these hypotheses on a new data set of Twitter activity of candidates on three levels of elections in which German political parties compete: state, federal and (as soon as complete) European. We measure the tweet success depending on polarity, applying dictionary as well as machine learning based sentiment analysis. We find that tweets with strong emotionality are more likely to be liked which can serve as a feedback to politicians. We also find that retweets of negative tweets are more likely, creating positive feedback for negativity, criticism and polarization. To test the effect of the audience, we estimate ideal point distribution of candidates' followers using item response models of follower-friends relationships. We find that smaller, ideologically more homogeneous and extreme followerships are more likely to react to negative tone, creating even stronger incentives for candidates speaking to 'echo chambers'.