Thomas Lumley

statistics. computing. genetics. choral music.
product of more than one country.
may contain nuts.

verification: https://tslumley.github.io
blog: https://notstatschat.rbind.io
group blog: https://statschat.org.nz

Last Notes

2026-06-18 08:50:25 CEST

Two new results on Staph aureus bacteremia treatment from the SNAP Bayesian adaptive platform trial.
SNAP is much larger than any previous trial in this condition, and is set up to examine multiple questions at once and to add in new questions as old ones are answered. One concern for funders is whether you will actually find anything this way, so SNAP is also important to platform trial methodology.
The reports here are comparing two standard choices of antibiotic to see if one of them is better.
For pencillin-susceptible S. aureus (which, yes, still a thing), OG penicillin is better than advanced anti-staph penicillins (lower risk of kidney injury and probably lower mortality)
https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(26)00761-0/fulltext?rss=yes
For methicillin-susceptible (but not penicillin-susceptible), cefazolin is better than the anti-staph penicillins (lower risk of kidney injury and probably lower mortality)
https://www.nejm.org/doi/full/10.1056/NEJMoa2506905

2026-06-17 08:08:30 CEST

The R Project is awarded the biennial Rousseuw Prize in Statistics
https://www.rousseeuwprize.org/2026

2025-08-02 04:17:25 CEST

#bloomscrolling
Winter magnolias
https://fediscience.org/system/media_attachments/files/114/956/756/257/147/866/original/2a6571e01694090e.jpg
https://fediscience.org/system/media_attachments/files/114/956/756/716/859/671/original/5ea7f1b040a5e57b.jpg

2025-06-01 09:13:07 CEST

- reply

KILL IT WITH FIRE!!
ahem
I suppose there are perfectly innocent Oxalis species in some places

2025-01-24 01:25:04 CET

- reply

you can use withCallingHandlers() to get at the information. In the image, I run the code inside withCallingHandlers() and print a stack trace from the handler before continuing. This doesn't save *the actual stack* like error() does. One could, but that requires deep copying of everything if you're going to continue, since the contents of the stack will change.
https://fediscience.org/system/media_attachments/files/113/880/473/849/869/779/original/3c6e7c243e11efa9.jpg

2024-12-23 02:12:23 CET

Become ungovernable
https://fediscience.org/system/media_attachments/files/113/699/473/874/708/894/original/dea9f85bf19353ba.jpg

2024-11-20 22:41:08 CET

Collinearity in regression: a quartet of dataset illustrating that collinearity diagnostics don't actually get you anywhere in either predictive or causal model selection
https://notstatschat.rbind.io/2024/11/21/collinearity-again/