jialong@columbia:~/site$cat ./work/tweeting-cheap-talk.md
home
> Work · Paper · under review

Tweeting Cheap Talk: Elites' Communication Strategies during Corruption Scandals

year:
May 2022 – Sep 2023
place:
MPSA 80th Annual Meeting, Chicago
kind:
Paper · under review

Tweeting Cheap Talk: Elites' Communication Strategies during Corruption Scandals — a co-authored paper with Jason Gainous, Kevin M. Wagner, Weiheng (Mark) Liu, and You Wu. Presented at the 80th Annual Meeting of the Midwest Political Science Association, Chicago IL, 2023. Currently under review.

The question

When an institution lands in a corruption scandal, how do its legislators communicate? Do they pivot toward substantive policy content? Do they attack? Do they go silent? Do they amplify harder on the topics they were already talking about — their "cheap talk" — as a way of signaling loyalty without committing to a position?

The method

  • Data: ~300,000 tweets from members of the Puerto Rico Legislative Assembly (majority and minority parties) across 2018–2022 — the window spanning the Ricky Rosselló chat scandal and its aftermath
  • Pipeline: scraper with configured parameters; cleaning via Pandas; Optical Character Recognition (OCR) to extract text from the images Puerto Rican legislators often post
  • Modeling: sentiment analysis and structural topic models in R, adjusting for legislator fixed effects, party, and district
  • Outcome of interest: topic concentration — did members compress around a narrower set of themes when the institution was under duress, or did they expand?

What we found

Both, depending on proximity to the scandal. Members personally named in the chats went wider and vaguer. Members adjacent to them — same party, same chamber, not named — went narrower and louder, compressing onto the topics they were already associated with. Cheap talk became cheaper, but also more intensely branded.

The paper reads like communication science but feels like political psychology. The fun of it — the reason we wrote it — is that the data captures a small, real institutional moment, and the model recovers something about it that a close-reading never could.