Thomas Lumley

Also attested for "dude"

Looking a bit less like a eucalypt this year: Parkinson's rātā
#bloomscrolling

Quite a few sites are down and I saw someone speculating it might be cloudflare

Maybe. Or it might be a change over time

I would have said the sausage sizzle was the event that produced the sausages/bread/onions rather than the foodstuffs themselves?

"A 200-megawatt, $1 billion facility [...] capable of powering up to 80,000 homes for a day"

One day isn't very impressive for $1b

(yes, more likely they think MW are an energy unit, not a power unit)

Here's something: I'm constantly sent surveys conducted annually by companies and what puzzles me is that they ask different questions every year. They'll be like 45% of people want to quit their jobs/ are stressed/ whatever. Then we'll ask, is that unusual? What was last year's? They're like 🤷‍♂️

You've been puoned

Picture of Plato the philosopher

Picture of cut stone looking like Plato (it's the same photo)

Well...

Line chart with 6 facets, for 6 different Pacific countries. Blue line shows total poulation, green natural increase, and red net migration (usually negative).

Line chart showing net migration for 21 pacific island countries and territories.Mostly they are shaded red indicating negative net migration.

Second #rstats blog post in a series on Pacific island population issues. This post looks at net migration estimates from the UN Population Prospects. For most Pacific countries, net migration is strongly negative most of the time. freerangestats.info/blog/2025/12...

The flight does look to be a bit longer than the old AirNZ Auckland to London "direct" via LAX, which was the longest when it was discontinued

A nice reminder that "direct" and "non-stop" are antonyms, not synonyms when it comes to airlines

Registrations now open for the ViCBiostat Summer School - Feb 13-20, 2026 (Melbourne/online)!

Features courses by local & international experts in causal inference, estimands, meta-analysis and cluster RCTs

Info: www.vicbiostat.org.au/short-courses
Register: www.trybooking.com/events/landi...

The advantage of coming at midnight on a flying sleigh is that you don't shut down the local bus system at rush hour

When your country's collective ability to bullshit on the internet is extensive enough to shape AI slop.

A blog post giving a more thorough take on survey experiments and the credibility revolution: cyrussamii.com?p=4168

What sort of loss structure do you want on it? Squared or absolute error in \hat P or some sort of rank correlation?

"How can we dance while the earth is turning: mechanics in locally-flat spacetimes'
#PhysicsCrossover

Or if this is an iterative optimisation problem of finding the units with the highest true P(Y=1), then some sort of bandit optimisation thing -- Gittins index or its descendants

Weighting: it does the same job as oversampling but doesn't make the data any bigger and I think it's easier to reason about

Potentially, yes. We may not have identified specific bits of documentation that need it, though. Contact Heather (warwick.ac.uk/fac/sci/stat...) -- she will know better what the opportunities are (I'd ask her myself, but I'm travelling today)

YouTube video by xkcd's What If? Can you power your computer by typing?

Can you power your computer by typing?

What the latest What If? video in collaboration with
@minuteearth.bsky.social!

www.youtube.com/watch?v=0L44...

Kiwis are also able to be online participants, that's not only for overseas.

Heather Turner is running the event. Paul Murrell, Simon Urbanek, and me will be there from R Core.

You *do* need to register, *this week*

R Dev Days Dec 16/17 at Auckland *and remote* for people interested in contributing to the code and documentation of base R, or to infrastructure that supports such contribution.

Because timezones, there are morning and afternoon streams, join either or both
pretix.eu/r-contributo...

"I am *unsettled* that ChatGPT, fed on the plethora of articles I and others have written on the subject as well as my own chat history has reproduced a rough sketch of our arguments back to me, this is *chilling*!"

This guy has massive influence on tech policy. Wonderful.

Yeah, typo. WILLFIX

Part of the goal of our Bayesian derivation of the linear-regression sandwich estimator was to get something with better small-sample regularization behavior (www.jstor.org/stable/23362...). We haven't made a lot of progress since since, though.

Sure. You get a lot of smoothing from a constant-variance assumption. But the sample size needs to be fairly small, and (IIRC) the advantage is less impressive if you look at coverage and median length than if you look at variance

For the avoidance of doubt: does svyglm() use robust standard errors? Yes!
notstatschat.rbind.io/2025/12/01/d...

* or maybe 'Pyramids'