Christoffer H. Dausgaard

After 8️⃣ years as Social Media Editor of @wepsocial.bsky.social I am very happy to announce that @chdausgaard.bsky.social takes over from 1 January 2026 on. 🥳

It was a great pleasure to be part of @wepsocial.bsky.social journey & I thank the editors for the trust they have placed in me. 😊

Excited to be presenting my and Frederik's paper on elite rhetoric and group-party linkages in a few hours at the TaDa Spring Speaker Series! Tune in at the link below.

Next week, the TaDa Speaker Series is thrilled to host
@chdausgaard.bsky.social
from the University of Copenhagen presenting new work with @fghjorth.bsky.social on Elite Rhetoric and the Running Tally of Party-Group Linkages.

📅 April 16, 2025 | 🕔 17:00 Berlin Time
🔗 Subscribe: tada.cool

🚨 New piece in TIME: AI progress hasn't stalled — it's just become invisible to most people. 🚨

I used to think that AI slowed down a lot in 2024, but I now think I was wrong. Instead, there's a widening gap between AI's public face and its true capabilities. 🧵

In sum, our findings suggest that party elites have more agency in shaping group linkages than often thought - even in the short term.

Full paper link: fghjorth.github.io/workingpaper...

This relationship is robust to various specifications and is driven by group appeals to especially religious, class and age groups. Further, we find that while purely ‘symbolic’ appeals matter, ‘substantive’ appeals that mention policy are more effective by an order of magnitude.

We find that voters appear highly attuned to how party elites talk about different social groups. Adding a single net positive appeal to a given group on a daily basis over the course of 3 months improves a perceived group-party linkage by ~6 points on average.

To test this, we develop a novel automated approach that uses language models to measure group appeals observationally. Using UK data, we connect citizens’ perceived group linkages in surveys to party elites’ group appeals in parliamentary speech spanning three decades.

It’s often assumed that politicians’ use of group appeals -- valenced references to social groups -- can shape party reputations. We test this assumption, arguing that voters keep ‘running tallies’ of elite rhetoric to continuously update views of whose interests parties represent.

In a new working paper, @fghjorth.bsky.social and I show that party elites have significant power to shape group linkages through their rhetoric, suggesting that such linkages are more dynamic and elite-driven than suggested by predominant structural accounts. 🧵

It is super tough to cut down about 40% of 482 submissions for the EPSA behavior section. But I've finished the first evaluations.

Some thoughts, if you are interested:

The Best Time I Pretended I Hadn’t Heard of Slavoj Žižek One weird trick to frustrate the hell out of a Marxist bro

I also haven't heard of Zizek. Or Hegel.

Remember best practice in a power analysis is to plan not for the effect you expect, but for the smallest effect size of interest. The type 2 error is a conditional inference: If there is an effect I care about, I would have X probability to detect it. lakens.github.io/statistical_...

In causal inference there are row people:

"I care about effects, defined in terms of cases, and consider their potential outcomes" (population 🎯, mechanism 🤷🏻)

and column people:

"I care about mechanisms, which relate variables, and how they behave under interventions" (mechanism 🎯, population 🤷🏻)

Great to see this out!

A screenshot of Causal Inference in R: Causal Diagrams

A screenshot of a DAG from Causal Inference in R showing two open pathways in an analysis a scientist must account for to get the correct answer

A screenshot of a DAG from Causal Inference in R showing how to describe feedback loops in causal diagrams

We've made a major update to the causal diagrams chapter of Causal Inference in R! Check it out to learn about DAGs! I think there's a lot in this chapter that is undercovered in other sources. #rstats #causalinference @lucystats.bsky.social @travisgerke.bsky.social www.r-causal.org/chapters/05-...

Over the course of my ERC project we collected panel surveys in BE, DE, ES, GR, IRE and PT in 2019-2021 on pol behaviour including voting in national and EP elections. Recently, those data have been deposited and are available here. dataverse.harvard.edu/dataverse/MA...

Oooh yes definitely. I really like this one: journals.sagepub.com/doi/10.1177/...

I’ll also send you a working paper!

Kæmpe tillykke! 🎉

Special election results tonight were ambiguous, with Republicans overperforming partisan lean in 3 and Democrats overperforming in 4. Overall for the cycle, though, Dems are still overperforming bigly. docs.google.com/spreadsheets...

🤔
Model = parsimonious representation of a DGP intended to be refined by learning from data

Theory = parsimonious representation of a DGP without such empirical refinement intention, perhaps for normative analysis or to spell out plausible counterfactuals

New: Rs question the integrity of House races when their candidates lose at much higher rates than Ds, but candidate race and margin of loss have no measurable effect (w/@johncarey.bsky.social, B. Fogarty, @jasonreifler.bsky.social, and great Dartmouth students)
sites.dartmouth.edu/nyhan/files/...

Everyone knows about retrospective (economic) voting.
We asked who wins the votes that the incumbent loses. New parties? Established oppositions?

Get one out of 50 copies of our @wepsocial.bsky.social article for free.
www.tandfonline.com/eprint/GRD7C...

It definitely does. Hard to infer results from a forced conjoint if the real world choice situation includes an option to abstain.

Luckily this is not hard to solve design-wise!

A 🧵 on Israeli strategy or lack thereof.

Israel didn’t have a strategy for this war because Israel had no idea this war was coming. Netanyahu had a strategy: avoid war, weaken moderate Palestinians and contain Hamas. But it collapsed, leaving Israel confused and at war. 1/

Important stuff. Most conjoints force subjects to choose. This can distort results in unpredictable ways.

An image of doomguy from the video game DOOM fending off demons labeled with things like "is this adjusted for inflation?", "this is from a survey of only 7000 people", "median numbers are distorted by the top 1%", "does inflation include food, energy, and housing?", "this data comes from the government, can you really trust them?","just look outside, this is obviously untrue"

POV: you posted a chart of median real [insert data here] on the internet in 2023

This poll of Muslims in Britain, which is now being widely reported, is complete garbage. It is a self-selection social media poll, no better than the Daily Express "reader polls" showing "98% support for Brexit" or similar: x.com/MuslimCensus...

BREAKING: The GOP has decided to nominate rainfall as the new House Speaker.