Like Grok and Attie, more and more social media platforms now integrate AI directly in online conversations.
But how do these AI bots actually impact political discourse?
We ran a series of experiments to find out.
(Spoiler: the news are not altogether great.)
osf.io/preprints/so...
Dream content for @reuning.bsky.social ...
🚨NEW PAPER
Going Local? Localized Rhetoric in Congressional Communications
Is congressional comms really nationalized?
Analyzing 9M+ social media posts from Members of Congress, I show that local messaging is still common & strategic. New ML + domain sharing data
www.tandfonline.com/doi/full/10....
This is my favorite project that I have worked on and I think the findings have important implications for how we think about the nationalization of politics in the US. Members are still going local in comms. I hope Congress and comms scholars find this useful.
#Polcomm @polcomm.bsky.social
In general, Presidential control, ideological extremity, electoral safety predict the use of localized messaging by Members. As does, in perhaps my favorite finding, Member seniority/freshman status - newer Members strategically engage more locally with their districts!
I analyze multiple platforms (Twitter and Facebook) and multiple modes (messages and link sharing) to get a multi-faceted look at localized messaging. This inlcudes a local/national classification of over 45,000 shared web domains. The ML text classifier also performs really well.
Centrally, the use of localized messaging varies by party based on partisan control of the White House (and congressional chamber). Governing parties use messaging on localized benefits, opposition focuses more on national political developments. This results in a visually pleasing back and forth.
🚨NEW PAPER
Going Local? Localized Rhetoric in Congressional Communications
Is congressional comms really nationalized?
Analyzing 9M+ social media posts from Members of Congress, I show that local messaging is still common & strategic. New ML + domain sharing data
www.tandfonline.com/doi/full/10....
I'm not writing that chapter, so it'll be not the worst!
There's a apsa presidential taskforce on AI which will release a report next year with teaching being a chapter
An annoyingly good idea and well executed!
I recently reviewed a paper that had not 3 but 5 reviewers on it, so seems to be some inefficiencies in the market here! Maybe Europeans just say yes more?
BJPolS abstract which discusses the relationship between politicians' communication and voter perceptions through social media content analysis between 2012 and 2022.
#OpenAccess -
Believing What Politicians Communicate: Ideological Presentation of Self and Voters’ Perceptions of Politician Ideology - https://cup.org/4gTSToM
- @hjghassell.bsky.social, @michaelheseltine.bsky.social & @reuning.bsky.social
#FirstView
🥳 Our new paper is out at the British Journal of Political Science. We find a strong link between the ideology of candidate messaging and the perceived ideology of these candidates among voters.
Evidence that voters can differentiate between candidates and are attuned to what candidates say.
And here's Facebook. Really does seem to be a Covid term!
Size matters: LLama 3.1/3.3 70b perform best, along with Qwen2/2.5 72b. Gemma2 and Phi4 stand out as strong mid-sized models. Many smaller models produce poor results. Model selection can have a huge impact on your results!
📜Preprint!📜
GPT5 replacing GPT4 effectively renders a whole range of existing GPT-reliant research unreproduceable. Open-source LLMs offer a science-friendly alternative, but which to use? I test 28 different LLMS in 7 languages so you can pick the best models for your task. osf.io/preprints/os...
Just started using atrrr for search and user post collection. Firehose would be a massive plus!
113th congress starts in '13 is my go-to shorthand!
Fresh email this morning: "an account has been created for you in ScholarOne", nooooo!
I have a new paper forthcoming at BJPS with @hjghassell.bsky.social and @michaelheseltine.bsky.social looking at how social media messaging impacts perceptions of incumbent/candidate ideology. We find messaging ideology does relate to perceptions w/overtime changes
Pre-print: osf.io/preprints/so...
Bit chicken and egg though, did researchers adjust to stargazer or did stargazer adjust its defaults to perceived research practices
My hunch would be at least some creep after the rise of stargazer as a default tabulating method, with its default of * for 0.1.
Accept it with arms wide open, Kevin...
Worked in three reps offices. Member 1 was given weekly tally of calls, never saw emails. Member 2 given no information on either. Member 3 informal updates on calls, never saw emails.
99% is just fielded by interns. Up to the individual Member whether they care to receive any of it.
An extra 128gb ram and any 16gb or 24gb Nvidia gpu
... Alternatively, an absolute tonne of gaudy strobe rgb fans
@sysilviakim.bsky.social
Coming to a JOP issue near you!
I think @jbgruber.bsky.social is the expert on this?
Science!
R&P editors' collection 2024 (2/4)
@michaelheseltine.bsky.social & B. C. von Hohenberg find that GPT-4 coding of political text is highly accurate for shorter texts, but performance drops for longer ones, suggesting consideration should be given to task complexity when using LLM-assisted coding.