The Annex: 7.3 million congressional messages, and why the attack gets 80% more retweets

The document. “Measuring Partisanship and Representation in Online Congressional Communications”, by Michael Kistner of the University of Houston and five co-authors, in the American Political Science Review¹. It covers members of the US Congress from 2009 to 2022: about 4.7 million posts on Twitter, 2.4 million on Facebook and 184,000 official email newsletters¹. Desk arithmetic: roughly 7.3 million messages in all. On the day after a midterm, it is worth reading what the campaign looked like when nobody was campaigning.

The line. On Twitter, negative partisan attacks received 60% more likes, 80% more retweets and 79% more replies than an average post by the same politician¹. On Facebook, they drew 86% more shares and 93% more comments¹. Messages about bipartisan work or services for the local district did worse than the typical post, and so did claiming credit for something government had delivered¹.

The lines around it. Attacks made up about five to ten per cent of messages in the early years and nearly twenty per cent by the end of 2022¹. Language-based partisanship rose across all three channels from 2011¹. Measured this way, Democratic rhetoric showed no statistically significant trend over the decade, while Republican rhetoric moved strongly towards the extremes¹. The sharpest rise in Democratic attack messages came after the 2016 election; the Republican spike followed 2020¹. Messages about funding for the district or helping constituents did not keep pace¹.

The caveats. The data stop in 2022, before the rise of newer, more partisan platforms¹. Messages were classified by language models trained on a hand-coded subset; the authors checked the scores against roll-call votes, fundraising and bill sponsorship, and they lined up well¹. Engagement measures what a polarised online audience rewards, not what the electorate wants.

The constructive reading. The researchers point to separate surveys in which most voters say they prefer representatives who focus on legislative work over partisan bickering¹. They warn of a “social media feedback mirage”, in which politicians mistake likes and shares for support¹. That is a fixable error. The dull jobs that score badly online, such as answering a constituent’s letter, securing a grant, or co-signing a bill with the other side, are still what the voters who were counted this week say they want. A politician who checks the casework queue before the retweet count is reading the better metric.