Sign in

Vera Wilde

@verawil.de
328 followers 329 following 538 posts

Scientist (PhD), writer, risk literacy and citizen science tool builder. Nerd-of-all-trades (research methodologist). <3 FOIA, babies, clocking bias and error. Seeker of truth, especially wild.

PostsRepliesMedia
Vera Wilde @verawil.de · 29/09/2026
you're missing the overlap in the Venn, which I'm now imagining as a new superhero: addled Goose-Girl (na na na na NA)!
010
Vera Wilde @verawil.de · 29/09/2026
So the cost-benefit question is open and depends (imho) on who's on the watchlist and what they'd have done. The "associated arrests" BTP mentions — officers spotting crimes by eye while deployed — aren't LFR performance. They're policing.
010
Vera Wilde @verawil.de · 29/09/2026
£320k, 500,000+ faces scanned, 18 deployments, and live facial recognition in London train stations produced exactly 1 alert, a false positive. As someone who studies mass screening for rare problems: the specificity is stunning. But there's no sensitivity data. www.theguardian.com/technology/2...
theguardian.com
Trial of live facial recognition in London stations ends with a false positive and no arrests
Freedom of information request finds six-month trial cost £320,000, used almost 100 police hours and led to just one – incorrect – alert
130
Vera Wilde @verawil.de · 29/09/2026
30 years of risk literacy research asks: can people compute what a positive test means? LLMs do that now. The open question is whether people notice the structure when a problem arrives unlabeled. How do we test this pattern recognition without cueing it? wildetruth.substack.com/p/the-answer...
010
Vera Wilde @verawil.de · 26/09/2026
@guidobiele.bsky.social What do you think?
000
Vera Wilde @verawil.de · 25/09/2026
INSPECT-SR recommends two independent assessors. Looking for a second rater (trial statistics or perinatal clinical background welcome), with co-authorship on the paper. Replies or DMs open.
100
Vera Wilde @verawil.de · 25/09/2026
Applying this to a Cochrane review I've critiqued: Lenells et al. 2025, breastfeeding support to prevent postpartum depression. Ten RCTs; its one "may prevent" finding rests on a single small trial. Pre-registering an INSPECT-SR audit plus re-pooled analysis. wildetruth.substack.com/p/cochranes-...
wildetruth.substack.com
Cochrane's New Breastfeeding-Depression Review Misrepresents the Evidence
Cochrane said they'd ask me to review it before publication; they didn't, published it today, and it's worse than I feared
131
Vera Wilde @verawil.de · 25/09/2026
I knoooooow! 😩 But it predates my Charité affiliation. So at this rate, I should get the Charité-inspired output in the pipeline next week and have that cooking for Q4...
010
Vera Wilde @verawil.de · 25/09/2026
Dust to dust in wind we trust to be fickle Waving everywhere not without care not without home but in a ramshackle on-the-way roam This house will fall we heed a call eat the sweet harvest when it's time wave our branches sublime All is vanity just ask Hannity 🌿🍋 www.youtube.com/watch?v=5wcG...
youtube.com
Sukkot: Ecclesiastes Reimagined in Animation
YouTube video by BimBam
000
Vera Wilde @verawil.de · 25/09/2026
Preregistered at OSF, 2 Prolific pilots, materials and code at github.com/verawilde/rarity-roulette. A companion to representativeness checks like the motorcycle test: two branches of population-benchmark screening.
github.com
GitHub - verawilde/rarity-roulette: A tool to visualize the dangers of mass screening for low-prevalence problems.
A tool to visualize the dangers of mass screening for low-prevalence problems. - verawilde/rarity-roulette
001
Vera Wilde @verawil.de · 25/09/2026
Now published in *Behavior Research Methods*: An LLM canary in the online data coalmine. LLMs solve Bayesian reasoning problems way better than people do. That capability gap can flag LLM contamination in online samples. rdcu.be/4Fw8jpGpkVpa
rdcu.be
292
Reposted by Vera Wilde
Mattan S. Ben-Shachar @mattansb.msbstats.info · 24/09/2026
This quote by haunts me. From Wilcox's "Introduction to Robust Estimation and Hypothesis Testing" #stats
To begin, distributions are never normal. For some this seems obvious, hardly worth mentioning, but an aphorism given by Cram´er (1946) and attributed to the mathematician Poincar´e remains relevant: “Everyone believes in the [normal] law of errors, the experimenters because they think it is a mathematical theorem, the mathematicians because they think it is an experimental fact.”
310422
Reposted by Vera Wilde
Gordon Hodson @gordonhodsonphd.bsky.social · 03/09/2026
#AcademicSky #PsychSciSky #MetaScience Article worth reading, & sharing w/ your student researchers. Can't get this out of my head: "Bishop’s law: The better designed a study is, the more likely it is to obtain a null result" journals.sagepub.com/doi/10.1177/... Meimoun & @lakens.bsky.social
310134
Vera Wilde @verawil.de · 23/09/2026
Bonus: by Lakens's standard, essentially nothing in this literature has a sample size justification. Studies of a few dozen women reporting 80% success rates, and nobody asks what a study that size could resolve. For a near-coin-flip outcome, the obvious answer is: basically nothing.
010
Vera Wilde @verawil.de · 23/09/2026
Faithfully swallowing a sugar pill nearly halved your death rate in the Coronary Drug Project. Except obviously it didn't. That's the artifact my favorite baby-sex-diet paper is built on. wildetruth.substack.com/p/gestation-...
140
Vera Wilde @verawil.de · 21/09/2026
Also I have to fix them myself even tho running them thru an LLM would obviously work. It's part of the process, like with coding or thinking... To read it again, think about it again, smooth the edges... Good note-taking is a really helpful part of my learning/thinking process!
000
Vera Wilde @verawil.de · 21/09/2026
Anybody else notice your notes have more typos since using LLMs? Because they don't matter for that audience... I have gotten sloppier! Seems like a low-level example of deskilling. Normally I don't believe the hype about the AI-pocolypse, but typos. Ew.
130
Vera Wilde @verawil.de · 16/09/2026
See also this on quantum indeterminacy: wildetruth.substack.com/p/book-revie...
wildetruth.substack.com
Book Reviews: Free Agents and Determined
Disciples of the secular religion of science argue over competing idols, clinging to false security against the existential terror of epistemic uncertainty
000
Vera Wilde @verawil.de · 16/09/2026
The soul is real is really tied with a rope they thought about confounds for real this is dope! But that leap from statistical distinguishability to soul that is not what "nonlocal" means. Eat the fresh plums out of the ice-box. This is how quantum rolls: unknown, unknowable, the cheeky fox.
110
Vera Wilde @verawil.de · 16/09/2026
So I did not successfully teach the kids proper marching this morning, but I did stumble on this amazing marching band pop classic playlist and we will be dance partying again to this soon. open.spotify.com/playlist/1Nu...
open.spotify.com
The Best Marching Band Songs
All is welcome so long as it is a piece played by an instrumental band, examples include military bands and university bands.
130
Reposted by Vera Wilde
Jeremy Labrecque @jeremylabrecque.bsky.social · 15/09/2026
Relationship inference: what ifn’t
162
Vera Wilde @verawil.de · 14/09/2026
Replicating the Institute for Replication post be like
000
Vera Wilde @verawil.de · 14/09/2026
I invented a Bayesian motorcycle. Or so Reviewer 1 suggested. Behavior Research Methods has accepted the paper: Bayesian reasoning problems as a capability-gap test for LLM contamination in online samples. 57% of pilot 2 got 5/5, vs. a ~24% human ceiling. wildetruth.substack.com/p/i-invented...
010
Vera Wilde @verawil.de · 08/09/2026
It is not particularly windy in Berlin today, but here is a poem about wind. www.poetryfoundation.org/poetrymagazi...
poetryfoundation.org
Wind
Like a wind asleep against stones as light as the air from a volcano’s mouth. Coves clear to the sea floor, wind grass and new lambs, donkeys eating…
030
Vera Wilde @verawil.de · 07/09/2026
New in Internet Policy Review: "Chat Control: One Law, Two Problems": policyreview.info/articles/new... Known-material scanning has an answer sheet, making its errors measurable. Searching for uncatalogued material has none — and none can be supplied. wildetruth.substack.com/p/one-law-tw...
000
Vera Wilde @verawil.de · 04/09/2026
Challah continues to please. Practice round challah for Rosh Hashanah next week. With jumbo raisins and less egg wash. I think next time it will have cinnamon because you can't have raisin bread without cinnamon?
010
Vera Wilde @verawil.de · 01/09/2026
P.S. I study this failure for a living and still built it into my own experiment without noticing. ("This is fine" meme art here from Tyler Comrie / *The Atlantic*: www.theatlantic.com/culture/arch...)
000
Vera Wilde @verawil.de · 01/09/2026
Van Gogh copied Hiroshige line for line before he could paint like that. Most people can't see the same problem in a new context until someone points. That's why we keep falling for the same screening disaster. Also, you have to read Gick & Holyoak 1983! wildetruth.substack.com/p/how-to-ste...
100
Vera Wilde @verawil.de · 28/08/2026
d0ca5882c2be48b6bfd277f830f0c394c3eb8ca88c59e5e8898be29d7c27ec77
000
Vera Wilde @verawil.de · 26/08/2026
More on Good–Turing estimation of unknown stuff. Rarity is common! Two methods read the same table: Good-Turing estimates the mass of the unseen, Chao bounds the count. Neither can tell you the shape of the tail. And that's structural, not a data problem. wildetruth.substack.com/p/good-turin...
001
Vera Wilde @verawil.de · 26/08/2026
Because menopause.
010
Vera Wilde @verawil.de · 23/08/2026
Weekend cooking continues with gluten-free pumpkin pies by popular request. Older offspring dug up the canned pumpkin puree Grandma brought ages ago, made demands. Poured much vanilla extract, also from Grandma. In our defense, it's already chilly and gray here this morning.
000
Vera Wilde @verawil.de · 22/08/2026
Can confirm the wetter-hands-for-less-sticky-dough trick also works on the gf dinner roll version, which is then amenable to cookie cutter fun.
020
Vera Wilde @verawil.de · 21/08/2026
Controversial hot take: bread is good. Also, the Internet told me to make the gluten free dough and my hands **wetter** to make it less sticky so it would braid, which is objectively unhinged advice. It worked. So I guess now I have to trust the Internet about everything forever. [Nomnomnom.]
120
Vera Wilde @verawil.de · 18/08/2026
Back when I forayed into online dating, I learned you reallydid have to assume the guy was hiding a wife and three kids under his abstract so.
020
Vera Wilde @verawil.de · 15/08/2026
Thank you! That's the compact version of the thing I spent 1,200 words circling. And the externally-invalid-values caveat is a part I left implicit. The estimand you can write down cleanly is often defined over a region nobody occupies. Controlling the co-treatments makes the answer useless!
010
Vera Wilde @verawil.de · 14/08/2026
On Rose: he's right that the causes of incidence hide in the homogeneity. He's also pre-DAG, all classification pathway, and never counts the borderliners he names. Still seven pages, still essential. saludcomunitaria.wordpress.com/wp-content/u...
saludcomunitaria.wordpress.com
000
Vera Wilde @verawil.de · 14/08/2026
Readings on transfer problems across epidemiology, screening, language, causal inference, and learning. h/t @stevensenior.bsky.social for Geoffrey Rose and @benkawam.bsky.social for Anna Wierzbicka. wildetruth.substack.com/p/science-vi...
150
Vera Wilde @verawil.de · 10/08/2026
www.youtube.com/watch?v=2Uz0...
youtube.com
Gam Zu L'Tova!
YouTube video by Jewish Songs To Inspire
010
Vera Wilde @verawil.de · 09/08/2026
Shouldn't have. The correction removed a spurious benefit (statins slowing diabetes progression), it didn't find a harm. Lévesque et al. 2010 BMJ. Though the fact that the skeptics are mostly named by their opponents tells you how that debate runs.
020
Reposted by Vera Wilde
Julia M. Rohrer @dingdingpeng.the100.ci · 08/08/2026
New blog post in which we discover the secret to immortality: accounting errors. www.the100.ci/2026/08/08/m...
the100.ci
My Immortal Time Bias
Preamble: This shortie is my attempt to come up with an example that makes the problem underlying immortal time bias as obvious as possible.  In Japan, people who reach the age of 100 are handed a co...
128031
Vera Wilde @verawil.de · 08/08/2026
Also worth adding: ITB doesn't only inflate, it can flip signs. Statins and diabetes progression went from HR 0.74 to 1.97 once immortal person-time was classified correctly.
130
Vera Wilde @verawil.de · 08/08/2026
Screening research is where this gets expensive: observational colonoscopy studies showed for years that it prevented distal cancers better than proximal, read as biology. Braitmaier et al.'s target trial emulation suggests most of that gap was time-related bias, and it had shaped guidelines.
150
Vera Wilde @verawil.de · 07/08/2026
Feeling pretty good about my Führungszeugnis!
020
Reposted by Vera Wilde
Jamie Cummins @jamiecummins.bsky.social · 03/08/2026
Have you built an LLM-based research tool and had to stumble your way through figuring out how to validate it? Me too! So I wrote a guide on how to systematically approach this. Preprint here: osf.io/preprints/ps...
07214
Vera Wilde @verawil.de · 05/08/2026
So Murphy's statutory framework (which is good!) governs the pool. But we also need upstream authorization standards — require proponents to show PPV, FPR, resource absorption before the indiscriminate search runs. wildetruth.substack.com/p/the-pool-looks-closed-from-here
000
Vera Wilde @verawil.de · 05/08/2026
Within-pool error rates are FDRs on already-filtered populations. They can't characterize population-level accuracy. This is Chat Control! Chat Control 1.0: ~48% FPR. CC 2.0 projections: ~555 false positives per true positive. The pipeline is structurally broken before anyone gets biased.
100
Vera Wilde @verawil.de · 05/08/2026
Lindsey, Hertwig & Gigerenzer showed this for DNA databases: the bigger the upstream search, the more coincidental matches, the lower the posterior probability of guilt given a hit. The pool feels closed. It isn't necessarily. pure.mpg.de/rest/items/item_2101705/component/file_2101704/content
pure.mpg.de
100
Vera Wilde @verawil.de · 05/08/2026
Murphy's "closed-universe searches" (Cal L Rev 2026) fills a real gap in 4A doctrine. But the "almost certainly the perp is in the pool" assumption is load-bearing — and it's doubly confounded by base-rate reasoning. californialawreview.org/print/closed-universe-searches
100