Johan Ugander

That's exactly what sent me to the SI!

Most unusual SI/methods write-up I've seen in a while: Digging in Utah scrapbooks for hair samples to document historic lead levels. Remarkable findings, note the log-scaled y-axis.

@Grok is this true: How X’s chatbot performs as a fact-checking tool New research explores whether the chatbot might replace the crowdsourced fact-checking program – and what that might mean for getting to the truth on X

New on @indicator.media: "@grok is this true" was the single most frequent reply tagging X's AI chatbot in the six months following its launch.

Santa Clara County runoff turnout dropped 53%. Ranked choice voting is the solution. - FairVote Santa Clara County, CA's runoff for county assessor cost taxpayers an estimated $13.1 million, and turnout declined by 53%

On Dec. 30, Santa Clara County, CA held a $13 million runoff for county assessor.

Turnout dropped 53%, and the winning candidate got fewer votes in the runoff than in the general election.

All of this could be avoided w/ #RankedChoiceVoting. @calrcv.bsky.social ⬇️

I was at FB in 2011 when Google+ launched, with "circles", and all this was much-discussed! When FB added friend lists and close friends (around then), there was a push to have that feature *not* include a recommendation system, so that the lists would be uncontaminated. Forget how it played out.

Ah! That's a really good point that FB 2008 didn't have reshares or public replies yet. And yes, so the basic bsky affordances definitely drive a lot of triangle closing. I guess a better comparison is early Twitter, which added WTF in 2010: archives.iw3c2.org/www2013/proc...

The replies to this post surfaced lots of great 3rd party tools (thanks!), but I still think it's interesting that bsky, as a platform, has decided to not prioritize algorithmic triangle closing (or follow recommendations, generally). See this fig from Zignani et al. (2014), when FB introduced PYMK.

Find Your Friends on Bluesky - Bluesky We're introducing Find Friends — a contact import feature that makes it easy to find people you know on Bluesky without compromising your privacy.

Huh — bsky has no PYMK ("people you may know") rec sys, it seems? Or? It has a mobile contacts uploader (linked), but it's highly opt-in by design. Lists help some with discovery. But no triadic closure engine, it seems. It appears to be doing fine without it? bsky.social/about/blog/1...

Well, no. What reduces productivity is not AI. It's the corporates' quick trading off of experienced humans with AI without a good understanding of when, how, and by whom it should be used.

Disinformation Summer Institute (DSI) - Disinformation Summer Institute Disinformation Disinformation is false or misleading information created and spread to advance policy, political, economic or ideological goals. It is sometimes used by industries, governments, and ot...

Applications are open for the second annual Disinformation Summer Institute. This is a 4-day, intensive summer institute intended primarily for early career researchers. It will be held at IslandWood. Beautiful place! Applications short and due February 15th, 2026. More info: disinfoinstitute.org

I had to look up Lizardman's Constant: the 4% of people who will tell a pollster lizard people are taking over the world, ie the minimum answer you will get for even the most absurd question.

Fair! I've actually heard lots of mixed feels. To me its pros and cons, but I tend to tune in.

Interesting, will check out! IIRC the main thread of the podcast critique is not that the results are wrong, but more a criticism of the lack of impactful translation to policy. Episode 2 talks a lot of government "nudge units". Been a while since I listened tho.

- DeMarzo et al. (2003) "Persuasion Bias, Social Influence, and Unidimensional Opinions", QJE. academic.oup.com/qje/article-...
- DellaPosta et al. (2015) "Why Do Liberals Drink Lattes?, " AJS. www.journals.uchicago.edu/doi/10.1086/...

If you enjoy salty podcasts, If Books Could Kill has a two-parter on Thaler & Sunstein's Nudge. The first episode covers organ donation (and IIRC a bit on retirement) and other early work. Recommend both episodes (and the rest of the podcast too). Available wherever you get your podcasts.

Less glib: my ability to do research-by-graduate-student leverages skills i've built up over a decade, e.g., knowing how to "unit test" student output, considering edge cases, etc. Things like: why doesn't the plot start at y=0? Etc. Processing the output of vibe-code feels very similar. 2/2

I've often joked that as faculty I program in a high-level language called "graduate student". Having tried out Claude Code this morning, I (i) feel extremely at home, (ii) am realizing that research-by-graduate-student is perhaps the original vibe-coding. 1/2

Here are the stats for my own account. (Separately: used this script rewrite to test drive claude code for the first time, amazing.)

This is a cool/simple/elegant glimpse. And thanks for sharing the script! I think the lurker percentage is pretty comparable, but I don't have hard numbers to point to for (any vintage of) any other network.

this is an interesting descriptive statistic for bsky network [pareto principle of production, most people are rubberneckers] that is estimable thanks to atproto's openness. script to make your own is attached - make your own!.
what does this look like for other networks? cc @jugander.bsky.social

Computational Social Scientists in the Nordics, unite!
🇩🇰🇫🇮🇳🇴🇸🇪🇮🇸

The brand new Nordic Society for CSS welcomes all researchers and practitioners based in the Nordics. The Society will promote student mobility, events, and education initiatives.

Join for free: nosocss.org/join.html.

We Let AI Run Our Office Vending Machine. It Lost Hundreds of Dollars. An AI agent ran a snack operation in the WSJ newsroom. It gave away a free PlayStation, ordered a live fish—and taught us lessons about the future of AI.

Wall Street Journal got an A.I.-run vending machine for their office.

Takeaway: Highly entertaining, but financially disastrous to let a generative A.I. chatbot run your business. www.wsj.com/tech/ai/anth...

(Ignore the spurious "Original" string at the bottom; that's just a bad crop by me, I was comparing the modified and original tex tables in the same tex file)

With a disclaimer that this was just 30 minutes of hacking, generously aided by AI coding, so I might have messed up the code tweaks. But I wanted to report back and also give huge kudos for the clean replication code. 3/3

Your code repo is amazingly clean, so I was able to do straightforward code modifications and replicate your Table 6 and 7 with added mayoral cohort controls. Happy to report the law enforcement effect finding in Table 6 is largely robust to this change in model spec, 3.6% instead of 3.9%. 2/

Interesting! Reading the paper, I quickly wondered about the linear control for "year became judge", worried it would be more appropriate to have non-linear/cohort controls for the mayoral administration at appointment. NYC mayors have pendulum'ed a lot on crime over the last 30 years. 1/

A key methodological challenge in observational studies with interference between units is twofold: (1) each unit's outcome may depend on many others' treatments, and (2) treatment assignments may exhibit complex dependencies across units. We develop a general statistical framework for constructing robust causal effect estimators to address these challenges. We first show that, without restricting the patterns of interference, the standard inverse probability weighting (IPW) estimator is the only uniformly unbiased estimator when the propensity score is known. In contrast, no estimator has such a property if the propensity score is unknown. We then introduce a \emph{low-rank structure} of potential outcomes as a broad class of structural assumptions about interference. This framework encompasses common assumptions such as anonymous, nearest-neighbor, and additive interference, while flexibly allowing for more complex study-specific interference assumptions. Under this low-rank assumption, we show how to construct an unbiased weighting estimator for a large class of causal estimands. The proposed weighting estimator does not require knowledge of true propensity scores and is therefore robust to unknown treatment assignment dependencies that often exist in observational studies. If the true propensity score is known, we can obtain an unbiased estimator that is more efficient than the IPW estimator by leveraging a low-rank structure. We establish the finite sample and asymptotic properties of the proposed weighting estimator, develop a data-driven procedure to select among candidate low-rank structures, and validate our approach through simulation and empirical studies.

"Low-rank Covariate Balancing Estimators under Interference"
Always neat to see CBPS in the wild

arxiv: arxiv.org/abs/2512.13944
#statssky #causalsky

Use AI unto others as you would have them use AI unto you

I'm embarrassed to say I saw this story because it was hyped up in Yale's "Yale Today" daily campus-wide email. Linking to Dale and Kreuger (2002) as a palate cleanser:
www.jstor.org/stable/4132484