Stats For Bios

I'm very excited about this one! First continental-scale, high-res mapping of AC. AC protects against extreme heat but also causes urban heat islands. And until Amber Vayani's analysis here, we just didn't know much about AC distribution at all. www.medrxiv.org/content/10.6... #AC #DHS #HeatWave

Any downloaded data must have a README (somewhere very obvious) that says how it was obtained and the url etc.

(Additional tip: within the data directory ALWAYS have a subdirectory for raw_data and a subdirectory for processed_data. The raw data should never ever get edited, only read, no exceptions ever. The processed data can be deleted and regenerated as required.)

How to use the disaggregation R package for multi-year data. Blog post by my recent predoc fellow, Sam Bano. timcdlucas.github.io/Combining-te... #RStats #rspatial #spatial #malaria

The predictions are available as a raster here. I'm not sure what standard recommendations are. These are fine to look at and have a play with. Maybe you'd want to wait for peer-review before using in a final, full analysis? figshare.com/articles/dat... #AC #DHS #HeatWave

PhD student for role of air conditioning on health. Health effects, urban heat islands, and what role AC will have in it all over the next 50 years. Funded. Some places for international students, but even more competitive. centa.ac.uk/studentship/...

'A plausible paper can now be produced in a day, and the process of producing it does not necessarily enhance knowledge, build a resource, train a student, or advance any other goal of biology research' 'we could use this surge to acknowledge that paper counts are meaningless'

Some reflections on AI in math. I found it helpful to simply list out the various parts of what a mathematician does, why math is valuable, and where the funding comes from. From there, what happens if AI gets more capable? youtu.be/LKrwhjOI8Nc

Always loved @3blue1brown.com. This is a video about maths and ai but I think a lot relates to science and ai. It's a great video, I need to think about it for a while, I need to get around to writing my blog post on ai and science. m.youtube.com/watch?v=LKrw...

Mixed-Frequency Regression Model for Short-Term Environmental Exposure-Response Modelling: A Simulation Study https://www.medrxiv.org/content/10.64898/2026.06.24.26356336v1

Holy cow, this is *amazing* and I highly recommend it to anyone who works with environmental exposure data!

Disagapp: A Shiny app to facilitate reproducible disaggregation regression analyses https://www.medrxiv.org/content/10.64898/2026.09.09.26362604v1

Just... look at it. Not all down to vaccinations of course. But still, what a thing.

Why would an animal weighing less than a teaspoon bother flying 150 miles in a night? We have fitted almost 100 with 'airtags for bats' to find out. When one flies near an iPhone, its position is pinged to us. Look what Mike and Mira did last night! @batconservation.bsky.social @sussex.ac.uk

If you fit models with glmmTMB(), gratia 📦 has gained support for these model objects, including glmmTMB's nascent capability for fitting penalized spline terms using mgcv's s() You'll need the GitHub versions of both gratia & glmmTMB to try this out #RStats #statistics github.com/glmmTMB/glmm...

🚨 After 20 years of service, the R-Forge code-sharing platform will be retired. #rstats Timeline: - Until end of 2026: Normal operations. - Jan 2027: Switch to read-only, but SVN continues to work. - Jan 2028: SVN also read-only. Reason: The underlying FusionForge software is no longer maintained.

New preprint from my student Carolina Molina-Tabares. Different to other preprints I post! She evaluated a portable qPCR workflow for rapid on-site detection of Ae. aegypti under field conditions. Compared to benchtop Roche LightCycler 96. Tested in Kenya. www.researchsquare.com/article/rs-1...

If you were planning to send me a chatbot-written email or post a chatbot-written comment, just send or post your prompt. That’s enough. You can keep the slop to yourself. statmodeling.stat.columbia.edu/2026/10/02/i...

I have absolutely wondered about this. Amazing.

Oh, great. TIL #RStats sprintf("%03s", x) differ on macOS and Linux; Linux: > sprintf("x=%02s", c("1", "2", "a")) [1] "x= 1" "x= 2" "x= a" macOS: > sprintf("x=%02s", c("1", "2", "a")) [1] "x=01" "x=02" "x=0a" Personally, I'd argue padding strings with zeros should not be an error. Oh well...

Why does windows default to not showing file extensions? So stupid. And a really nice way to make teaching into R more difficult. #RStats

Jobs galore at @nhm-london.bsky.social @dissco.bsky.social - jobs.nhm.ac.uk/Job/JobDetai... - jobs.nhm.ac.uk/Job/JobDetai... - jobs.nhm.ac.uk/Job/JobDetai... - jobs.nhm.ac.uk/Job/JobDetai... - jobs.nhm.ac.uk/Job/JobDetai... including Heads of Digitisation and Technical Innovation

A humbling reminder from Carl Sagan: Even if you read 1 book a week for your entire life, you'd only scratch the surface of human knowledge.

Release of CmdStan 2.40, including new Stan Math, core Stan, and Stanc, is now available! Blog post with new features blog.mc-stan.org/2026/09/16/r... Highlights - integrate_1d_double_exponential, a variadic version of old integrate_1d - integrate_1d_gauss_kronrod, a different quadrature rule ...

Reminder that "wiggly" is absolutely a technical statistical term. No discussion needed.

Preprint shinyapp by @simonsmart.bsky.social here. User-friendly, reproducible software for high-resolution predictions from aggregate polygon case counts (or other data). Once you're done, click a button to get a Rmd file out. #RStats #DiseaseMapping #malaria www.medrxiv.org/content/10.6...

This paper is 20 years old but keeps being as relevant today as back then :/ doi.org/10.1111/j.13... #statistics #ecopubs

Modeling non-response is an example I like to use when arguing that "descriptive" statistics also require causal inference. To describe the population, you have to model what causes the sample. Similar to occupancy modeling in ecology in many ways.

My bar for genAI has been for a while "I'll consider it useful when it can fill in a form for me". Copilot, with a lot of prompting and nonsense, filled 90% of a form for me. That's it, I'm in.

The YouTube android app is just so much worse than the mobile webpage. And I can't delete it.

I'm looking for doctoral students, postdocs and research fellows to work with on Bayesian Workflow. Flexible starting time

Say I've got a train dataset and a test data set. I fit a linear model to the train set and evaluate within-sample error. Then I fit the linear model to the train set and evaluate with the test set (i.e. cross validation). The 2 models are identical and equally overfit. The metric is different.

I've probably moaned about this before. Cross validation doesn't control overfitting (your model is still just as overfit). It (tries to) estimates predictive performance without being tricked by overfitting. Big difference.