Twitter Beğeni Hilesi

Beyond excited to share that this is now out in nature! We show that despite efforts to remove overt racial bias, LLMs generate covertly racist decisions about people based on their dialect. Joint work with amazing co-authors ria / pratyusha ria kalluri, Dan Jurafsky, and Sharese King.

3+ months later, finally "under review"

TrueViews is the product of tons of work from Nicholas Stephanopoulos, Ruth Greenwood & Chris Warshaw. The hope is that it can help elected officials better learn what their constituents want – and I think it's a great tool for students, too! Check it out: trueviews.org

🚨 Job Market Paper alert! "How Much Should We Trust Modern Difference-in-Differences Estimates?" at osf.io/bqmws I assess the fit between modern DID-style estimators and real-world data features, and point to approaches for improving inference. Quick thread ⬇️

My blog post critical of this re-analysis is up: matthewbjane.com/blog-posts/blo… The re-analysis conducted by Rausch & Haidt does not follow a principled statistical approach. I re-do their re-analysis with appropriate statistical methods

I have a new paper conditionally accepted at Journal of Politics @[email protected] where Massimo Morelli and I show how important it is to have a skilled bureaucracy for politicians' ability to get things done! Check it out! lucabellodi.com/material/BM_Po…

The beginning of semester email flood is real

Me getting the R&R v. Me reading the reviews

I am increasingly convinced the *major* source of publication bias is journals (editor-referee combination), not scientists. All serious scientists would love to publish all their work, even nulls. Goes in the file drawer because authors anticipate rejection.

6/ P-Hacking: They then manually code 150 JMPs to document the relationship between p-hacking and academic placement. JMCs with marginally significant results were more likely to secure academic jobs, especially during the job market of the COVID-19 pandemic.

whoa

🚨Have you ever worried about respondents in your online panels who take many other online surveys and tasks daily? I will present a paper at #APSA2024 measuring the prevalence of these professionals and investigating their effects on the quality of online panels. Quick 🧵

This happens all too often. #NullEffectsMatter #TeamPreciseNulls

Very little polisci work using LLMs is "replicable" in the usual sense of that term. If this sounds like a problem, come see ppr w Christopher Barrie and Lexi Palmer Thurs @ APSA 12-130pm, PCC112A. We give some theory + empirics on where things stand and where we should be trying to go

where is "teaching" on the apple watch workout app?

This looks super cool!

Shameless self-promotion. I will be in Philly until this Saturday. If your department is hiring in American politics, public policy & administration, democracy, civic engagement, urban/local politics, REP, or computational social science, please feel free to reach out to me!

Devastating review of the degrowth literature (561 studies): --> 'few studies use quantitative or qualitative data...' --> [those that do] 'tend to include small samples or focus on non-representative cases' -->'large majority (almost 90%) are opinions rather than analysis'