Peter Ellis

The original graphic looked a bit like this. I found the 3-d perspective plot just a bit hard to follow. I think the main improvement in my own plot (post above) is just putting the two most important variables on the x and y variables while using colour for the elasticity assumption.

The Wikipedia community has built a tool that makes it possible to embed our interactive Our World in Data charts right in any Wikipedia page. I have always admired Wikipedia and all the people contributing to it. And I'm very happy that our work now can get used this way!

This is also basically the logic of Buddhist prayer flags and wheels - to automate the act of prayer.

anyone got examples of (preferably epidemiology) papers that make good use of DAGs or have really clearly presented DAGs that add to the paper? showing off your own papers is welcome/encouraged! (to be used for teaching purposes)

The UK Statistics Authority is looking for three non executives to join the Board to oversee the statistical system. 24 days a year. £15k. 5 year term apply-for-public-appointment.service.gov.uk/roles/9647?n...

Some people were making me mad on the internet so I wrote a blog post about it. On the different reasons for people have for being left wing and how people manage the psychological tensions of the fact that they come apart. sootyempiric.blogspot.com/2026/10/the-...

I've implemented quizzes in my class this year. I can see this being a next step: generating custom quizzes based on students' own written work to see if they understand what they've handed in.

He's taking a lot of heat for this in the quotes but he's definitely correct. At 1%-3% of lifetime income for an average American, this would be nice and might seem a temporary life-saver, but not make a sustained change. 10x more is when it might mean better house, better education for kids, etc.

US total crude oil stocks have fallen below 700 million barrels for the first time since the 1980s. But that's still plenty of crude oil for going on with (US refineries use about 17m barrels per day at peak throughput). freerangestats.info/fuel-crisis/...

Was pleased today that I at least was able to mention 'neglected tropical diseases' in a speech, one of the existing but, well, neglected, black marks on humanity.

Good talk and great plots from @freerangestats.info at #AEA2026 #rstats

It must be particularly complex trying to vaccinate birds that are shy.

Great keynote today by Neil Pearce for the Australasian Epidemiological Association using asthma and kidney disease as egs. Argues much that is called 'causal inference' is really causal estimation, and inference is more like multi-source, multi-method detective work eliminating suspects one by one.

What I am getting at is I believe that if there was a credible threat of getting in actual trouble when your AI does something bad that the labs would implement basic simple measures that would be highly effective at preventing those bad things.

With all this talk about noise maybe it's worth pointing out this newish paper about different sources of total survey error. It's measurement error, not non-response error docs.iza.org/dp18918.pdf

Great example

in 2020 my lab quit Microsoft Word and began writing in R Markdown, an open-source document-publishing system that facilitates a clear, permanent link between statements in text (scientific prose) and the code producing analyses and visualizations supporting those statements (scientific analyses)

A very stylish #Roman woman's shoe with a decorative pattern punched into the leather. Found in 1907 in a well in the civilian settlement of the Saalburg fort, dating first half of the 2nd century AD. Photo: Römerkastell Saalburg

Asked to rank “the most important population-related issues facing the world in the next 20 years?”, 65% put climate change in their top three, followed by aging (53%) and inequality (51%). Only 9.5% mentioned "low fertility and population decline," with no one listing it as most important.

I did the 9 books thing. But it feels very arbitrary... in the end books read at a particular moment in life are more likely to get to the 'never forget' status. myninebooks.com/g/9e7ff98277fd

Many papers I see: A global problem Search for modifiable "risk factors" A cohort study sorry I mean real-world data "Multivariate" regression Some associations Can't say cause Residual confounding Policy must change in ways having nothing to do with our findings More research is needed

I don't even know what fantasy football is.

It's difficult to find estimates of total world crude oil stocks. I did the best I could with the 38 countries that have consistently complete data in JODI. Key message - they're low, but not at crisis levels for a while. #rstats blog at freerangestats.info/blog/2026/10...

I know I'm out of tune with the reaction to this, but when I've agreed to recording by a journalist my understanding has always been that I have zero future control. I expect them to share with colleagues, edit, publish, whatever. So I don't quite get the outrage this is generating.

WOMBAT2026 is next month! See robjhyndman.com/hyndsight/wo... for the details. #rstats

this was a great learning experience. This is my 3rd "refactor this into different framework" type project. And LLMs kick ass because it is a highly constrained problem. They are not designing architecture nor solving novel problems. they are casting one thing into a related thing. They do this well

The Pope! How many models does *he* have?

I am looking forward to contributing to this event by talking about the challenges and opportunities of small area estimation methods, based on experience to date with real data in Pacific Island countries. This is an online workshop on 5 November, registration details in the attached image.

The world transition to renewable power seems to be going very well. Basically once the tech became cost-effective, a whole lot of old debates became irrelevant.

A new version of the nonprobsvy package in R, and an article in JSS to boot! There is no better tool for the analysis of data from non-probability surveys, implementing all of the latest advances in this literature. If you analyze data from NP surveys, please read: www.jstatsoft.org/article/view...

Is there a term for effective altruism that is still interested in eg evidence for different development interventions and maximising ethical outcomes Peter Singer style, and hasn't gone insane through longtermism and AI obsession?

The Facial Expressions of Tethys by artist Julius von Bismarck. A suspended high-sea buoy that moves in real time. Sensors on a twin buoy out in the Atlantic Ocean send live satellite data to motors inside the museum. This makes the indoor buoy copy the exact, real-time movements of the ocean waves.

The story in this thread is extraordinary and as far as I can tell, accurate.

A small #rstats blog post following up on last week's on sample size and the central limit theorem. How well does BCa bootstrap do in comparison to older methods for a confidence interval for the mean from a skewed population? Better, but not a magic fix. freerangestats.info/blog/2026/09...

It seems to me the AI companies could improve both their image and revenue if they poured billions of investment into working with experts on ethical, effective and people-friendly applications in a field like education, rather than solving maths problems and generating unneeded celebrities.

Wrote in a work email "we won't be sending anyone to the World Data Forum this year, partly because of budget, mostly because Riyadh is under a 'reconsider need to travel' advisory due to missile and drone attacks in the area", which seems kind of easy as management decisions go.

🚨 Announcing OpenCensus, a new project with @ezrakarger.bsky.social : We are transcribing the historical US censuses and releasing them as open data 2.6 million people so far in 1940, browse here: opencensus.us

Inspired by a @f2harrell.bsky.social comment, I did some #rstats simulations about how big a sample needs to be for the central limit theorem to kick in and the sample mean be normally distributed for inference purposes. Yes, with skewed data the answer is "lots". freerangestats.info/blog/2026/09...

What do we call this phenomenon, where we are assailed with the near-identical one line summary of something multiple times. Not to pick on Reuters at all, this is a much bigger systemic thing. Sometimes we can go through 5 or 6 layers of Russian dolls of the same sentence over and over again.

Is there an interesting history about why the "responsible" in a RACI matrix (responsible, accountable, consulted, informed) as used by management consultants is so different from other traditions? In RACI, "responsible" means "does the work", rather than being responsible for decision or outcomes.

📊 #rstats #dataviz My ggCheysson package was just re-submitted to CRAN 🤞 @infowetrust.com Brings the graphic style of Émile Cheysson's Albums de Statistique Graphique (1880s) to ggplot2: authentic color palettes, hatching patterns, and hand-drawn fonts. 🔗https://friendly.github.io/ggCheysson/

I think there's actually quite an easy and obvious way to solve the collective action problem of how to make the AI companies stop conducting risky research, and it's to make it very clear that they're legally liable for anything their models do. www.politico.com/news/2026/09...

aleximas.substack.com/p/will-ai-so... The best blog ever written on AI and growth, by a clear margin. By @aleximas.bsky.social

This "going rogue" framing is nonsense. These are still just text analysis models. It's openAI that is "going rogue" by giving a computer program complete autonomy to make decisions without human oversight. That's a danger whether it's a fancy new LLM or an old-fashioned SQL database.