Affective polarization in the US is mirrored by increasing partisan warfare, incivility
and verbal attacks on the opposing party in social media. Traditional theories of
negative campaigning offer few answers to this development, as negativity on social
media is not induced by competitive inter-party elections, but negative partisanship. We
argue that elite affective polarization on social media is a result of nationalized politics
in safe electoral districts, which creates incentives for legislators to polarize affectively
in order to prevent primary challenges. We present a novel text-based spatial measure of
affective polarization and partisan warfare in social media. We separate affective from issue
polarization in social media, based on >1,200,000 Twitter messages of 411 representatives
between 2017 and 2020. We demonstrate it is a strong predictor of conventional ideology as
measured using roll call votes. We show that safe, non-competitive districts induce affective
communication. The effect of district safety is intermediated by the presence of a primary
challenger