Marvin Stecker

Ich würde sagen, da gehts steil bergauf, und dann um die Kurve, und wieder runter bei dir? Glückwunsch, und hab einen ganz tollen Start da!

🧵on my new paper "Synthetic personas distort the structure of human belief systems" w Roberto Cerina I'm v excited about...

🚨 Do synthetic samples look like human samples?

We compare 28 LLMs to the 2024 General Social Survey (GSS) to find out + develop host of diagnostics...

What, a content analysis dataset that you could, for example, combine with survey data and manifesto data? 😱
As part of AUTNES, we have data for the six weeks preceding Austria's last election in 2024, enabling research on topics such as media agendas, campaign dynamics, and opinion formation.

Picture of the three volumes of the Elgar Encyclopedia of Political Communication stacked one on the other. https://www.elgaronline.com/display/book/9781035301447/9781035301447.xml

CCL contributions to the new #ElgarEncyclopedia of #PoliticalCommunication, edited by
@alessandronai.bsky.social, @maxgroemping.bsky.social, &
@dominiquewirz.bsky.social! 🤝
CCL authors include @hosseinkermani.bsky.social, @marvins.bsky.social, @fabiennelind.bsky.social, & @jamoeberl.bsky.social 📝

Misinformation research has a causality problem: lab experiments are limited; observational studies confounded.

We used causal inference on 9.9M tweets, quantifying effects in the wild while blocking backdoor paths.

Does misinfo get higher engagement? Are following discussions more emotional? 🧵

Excited to share that together with @malojan.bsky.social we’ve published a research note in #partypolitics introducing PartySOME, a comprehensive dataset on political parties’ social media activity. A thread 🧵 journals.sagepub.com/doi/10.1177/...

TIL that data.table in R natively provides you with an option to perform multi-step per-row operations, and without even neeeding to prompt it, gives you a beautiful (speaking utilitarian, not aesthetical) progress bar.

Have a multilingual cross-country text analysis project? We just published a paper in @computationalcommunication.org with advice for validating your empirics! 👉 doi.org/10.5117/CCR2....

Currently in FirstView: “Moral Foundation Measurements Fail to Converge on Multilingual Party Manifestos.” @marvins.bsky.social and @fhopp.bsky.social examine 810 party manifestos in 4 languages to study how different instruments lead to different results for extracting moral foundations.

I wish my vacations were as eventful as the journey I had between the song title and the first 20 seconds of music.

I didn't have Rosalia meets Björk meets Yves Tumor on my timeline for this year. Then I got to the German chorus.

Really excited that after several years of working on it, revising, and getting lots of rejections, our paper (w/ Rune Stubager & Mads Thau) on citizens’ responses to group appeals is *finally* out @ejprjournal.bsky.social

www.cambridge.org/core/journal...

A 🧵 on our findings...

Picture of the committee, from left to right. Annie, Nicola, Azade, Eva, and Hajo.

Proud to congratulate you, @azadekakavand.bsky.social, on your PhD defense –
“Technology Affordances, Social Media Platforms & the Networked Far Right.”
A remarkable achievement that combines years of hard work & building of skills that you now use at @bredowinstitut.bsky.social. Missing u already🥳

Group photo: The CCL team standing together outdoors with the mountains around Dienten an Hochkönig in Salzburg in the background, smiling at the camera.

Ski lift: A line of orange ski lift chairs stretching up a green slope surrounded by dense forest to get up the mountain.

Hiking trail: The CCL team hiking down a narrow trail across a sunlit mountain slope with autumn-colored vegetation.

Mountain view: A wide view of mountain peaks and ridges around Dienten am Hochkönig in Salzburg, with low-hanging clouds drifting between the valleys under a blue sky.

Every strong team needs time to pause, reset, and grow together. #TeamRetreat #DientenAmHochkönig

After an hour’s delay, you’ll get 25%, with two hours, 50% back from Deutsche Bahn. Doesn’t make up for the inconvenience, but it’s at least something…

Ever asked yourself how to detect and extract social groups from texts with computational social science? @haukelicht.bsky.social and me have a solution for you out at @bjpols.bsky.social. You can also find the pre-trained models on huggingface!

Danke, danke! Hat sich ein Non-Diss Projekt mal ausgezahlt :D

Thanks!

What are our takeaways?
- Genuinely great breadth of tool development!
- These don’t agree, however, when applied to the same text
- Rather than criticising any single tool, we hope this is a benchmark and motivation to develop a more holistic, transparent and reflective approach to study morality

Regression table for all foundations on English translated manifestos, with the different effect directions for each measurement instrument. They often don't converge.

So what about it? Predicting parties’ use of moral foundations from their ideology, the measurements would lead you to different conclusions. Sometimes this concerns only significance, but with others, you will even get different effect directions.

Correlations between original language and English-translated measurements. For CCR with different types of embeddings, these are rather high, as well as for DDR, whereas the MFD has noticeably lower values.

We also compared the measurements on the original-language and English-translated manifestos. Dictionaries seem the hardest to translate. With all embedding-based measurements, their outputs are more similar, which is overall encouraging, though we also find some lower correlations here.

A heatmap of Kendall's correlations between all applied tools, with mostly rather low correlations.

What did we find? Considering the same English-language text, the measurements don’t really agree on how much a certain moral foundation is present. Most correlations between approaches are low, max 0.2 and 0.3.

A table that shows the tools we analysed in the study. We considered the MFD and MFD2 dictionaries, the eMFD, and also applied DDR, CCR and MoralBERT models. Links and citations for all are in the paper.

What tools?
We looked at the most common, off-the-shelf methods to measure different moral foundations (such as care, sanctity) in texts.
Where did we look?
All 800+ manifestos from the Manifesto Project in Spanish, English, Dutch or German, and their English translations.

New publication, out in Political Analysis:

There is an increasing array of tools to measure facets of morality in political language. But while they ostensibly measure the same concept, do they actually?

I and @fhopp.bsky.social set out to see what happens.

🚨 Neue Veröffentlichung! 🚨

Im sogenannten “Election Report” schauen wir uns die Nationalratswahl 2024 nochmal aus wissenschaftlicher Perspektive an und bereiten sie mit aktuellen Daten auf 🇦🇹🗳️🤝

📄 Lest hier: doi.org/10.1080/0140...

If you've recently wondered "What happened in Austria's last federal election" (and a lot happened!), we've got you covered:

I've never felt such accomplishments as my Docker scripts running on another persons machine using only my horrible readme file to guide them.

Really happy to see my work with @dweisstanner.bsky.social published with @bjpols.bsky.social. In this letter, we look at symbolic class signalling through cultural consumption and how effective it is across voters of different parties.

A thread 👇 (1/n)

#OpenAccess - cup.org/3GSCIKV

Thinking about using #CSS methods to study #racism, #stereotypes or #hate speech in text? 📐

👉 Check out my first dissertation paper co-authored by @fabiennelind.bsky.social and @hajoboo.bsky.social just published in Annals of the ICA! @icahdq.bsky.social 🥳

🔗 doi.org/10.1093/annc...

This is the only score I shall ever share because it was easy and I suck at this game
and my Fantasy-F1 predictions were absolutely rubbish and don’t we all deserve a tiny bit of happiness sometimes
maybe God or fate acts through puzzle apps

Connections
Puzzle #756
🟩🟩🟩🟩
🟨🟨🟨🟨
🟦🟦🟦🟦
🟪🟪🟪🟪