Peter Ellis

@freerangestats.info

Director, Statistics for Development Division, at the Pacific Community. Posts are personal views, and most likely to be about data and #rstats, maybe with bits of history, social science and philosophy thrown in. Blog is at https://freerangestats.info/.

On Thursday, I'm giving the last of the 2026 Ihaka lectures on "Spotting the weird ones" (about anomaly detection using #rstats). It's an honour to speak at something named after Ross Ihaka. It will be live-streamed for those who can't be there. www.auckland.ac.nz/en/science/a...

Ihaka Lecture Series – University of Auckland

The Ihaka Lecture Series is an annual event hosted by the Department of Statistics which presents the latest research about statistics and computer science.

auckland.ac.nz

I've open-sourced a web service container that will give you a WorldPop population count anywhere on earth given a point (100x100m), a point + radius, or a geojson polygon. It's using R2025A at the moment, but can change as needed. github.com/InternetSoci...

GitHub - InternetSociety/wpopapi: API for accessing WorldPop data

API for accessing WorldPop data. Contribute to InternetSociety/wpopapi development by creating an account on GitHub.

github.com

The chronological setting of the Odyssey really is, effectively, 'nowhen.' It is absolutely set in the past, but not in a specific, dateable past. Present tense (for Homer) stuff shows up, truly ancient s*** (for Homer) shows up. Also giants, monsters, giant monsters and straight-up divinities.

Generally this sort of data is good for extra timeliness or granularity (spatial or otherwise) but needs regular calibration to the more traditional representative sources, if it is to be used for official statistics.

Erika McEntarfer@erikamcentarfer.bsky.social · 2w ago

GS points out that (data task force member) Chetty's alternative economic indicator series has performed rather poorly since the pandemic, highlighting the challenges in using non-representative data

Graph shows the divergence between Opportunity Insights employment series and NFP

Speaking of which, I am in the market for the latest in what people regard as good introductions to Git and GitHub. Ideally a short course with some kind of rigour (online quizzes etc). One that doesn't mention graph theory but focuses on the basics for users.

Discussing whether I'm right to insist people get up to speed with GitHub before using AI to improve their code. Someone suggested I am advocating learning XXX by hand rather than computer, but it is more like insisting you get the hang of track changes before letting AI re-write your report.

Kind of staggered to come across current university course notes (for a third year course!) saying that simple random sampling requires the population to be normally distributed to work, and that non-normal distributions are why we sometimes use stratified sampling.

So it seems the only place in the southern hemisphere to see The Odyssey properly (IMAX 70mm) is in Melbourne, ironically about 300m from where I used to live in Fitzroy. How much of a compromise is it to see in a regular cinema?

That's not to say climate change is over and we can declare victory but the path to zero carbon is now extremely obvious and achievable and its mostly a matter of time (and how much damage we inflict along the way) rather than an existential danger we might not solve.

The impact of AI on jobs is uncertain. What is certain: assessment of job-creation/destruction by tech company CEOs should be heavily discounted, or ignored altogether. Tech CEOs have no insights in labor economics, and have bad incentives on this issue.

Bild

Just discovered an annoying user experience with Office/Copilot between them turning down my room booking (for my own office!) and automatically booking another - in its view, 'similar' - room instead. And there is no way to turn this behaviour off by default, only for individual meetings.

While there isn't timely, detailed info on national fuel use, if March-May petrol prices are up an average of 30% and Diesel up an average of 73%, then electronic card spending on fuel being only up 17% means that even if we don't know the proportion of fuel being sold, there is a lot less sold

Graph of change in card spending and change in fuel prices

For me the saddest part is the death of Stack Overflow, which was an internet marvel of the same sort as Wikipedia. The idea of replacing so much human interaction with each of us individually typing into solipsistic web forms and terminal prompts is depressing as hell.