Michael Lewis was a mistake

We need some other discourse tbh

Protein in everything

nice paper from @bingkai.bsky.social adding evidence that adjustment in RCTs is good but ML is often not better than linear regression. arxiv.org/pdf/2602.00434 The result shouldn't surprise you! The signal to noise ratio is too high to learn useful nonlinearities in all but the largest trials.

I just finished this excellent book. Strong recommend.

Oh, Bluesky. Never change.

The Schitt's Creek gifs will continue until morale improves

Jokes aside, if you don't have a will, POA, etc, go get them. Do it before you need them.

But *the* core memory is from 1993

Movies peaked in 1995

The Schitt's Creek gifs will continue until morale improves

I got some “Thank you for doing the work” comments when I posted a photo of my baby in a sling here. Bewildering (but thank you).

As a man, I demand compliments for keeping my offspring alive. THANK YOU FOR YOUR ATTENTION TO THIS MATTER

MoAr WWII discourse!

Getting elite institutions, whether academic or tech tbh, to be open to the idea that nontraditional learners are real human beings who have hopes and dreams and potential was the real work of online learning.

This has been a good series so far

Good comparison

Time to repost some causality memes #IdentificationK9

blockjack is now an R package, is online and passes sensible statistical tests... github.com/MichelNivard...

If you ever need a fast (block) Jackknife… also can be extended for WLS and its many robust cousins.. but maybe this was obvious to others, in which case never mind me!

It's incentives all the way down

Lots to say on this, but one thing recently came to my mind is the isomorphism between “wide format” & “long format” for panel datasets. The latter treats time as a variable, which confuses the issue, at least conceptually. DiD with wide format is clear and here: journals.sagepub.com/doi/10.1177/...

Some great ones shared already. I haven't seen this one shared yet.

Outside of my a loaded question, my other contribution to this debate is that we need to distinguish between 1. science as process, 2. science as community, 3. science as outcome, and 4. science as academic publishing norms/issues. We often look at 4, then use that to make inferences about 1-3.

What is the best movie that's set on a farm?

This is correct

I succumbed to the bait and I am so mad at myself.

Many people are saying

Me, every time I am presented with something I disagree with:

Beautiful

This kind of creativity is what Claude will take from us.

use() %>% pipes() %>% SomeTimes() |> mix() |> them() %>% because() |> moaR() %>% chaos() then$do <- this Then$otheR_times = this this -> finally$chaos be_careful <<- what.you wish ->> for

All I could think of when I saw that image

Sportspuck brings us together, sportspuck tears us apart. Circle of life, really.

STRAIGHT INTO MY VEINS

🌶

okay stephen you're back to being my nemesis again

I don't think the solution is to limit grades to reduce stress and competition. Maybe I am wrong.

Correct

"...we need to worry about non-classical measurement error more than we need to worry about classical measurement error."

In the same way, I don't think mixed models for small complex experiments and mixed models for longitudinal data are all that similar despite both being multivariate Gaussian regression models

The pooled IV estimator mostly misses out on recognizing the second mode. Even when the DGP is "everyone does this", the IV estimator severely under-estimates how much someone will react to a signal.

IV: is there anything it can do? 😉

Many people are saying

What R packages have the highest code quality and should serve as examples to learn from in your opinion? #rstats

Hehehe. Ottawa. Also I would put Calgary in tier 3

Imagine the whining if things had turned out differently

It appears we are discussing insurgency today. Middle-aged dads of Bluesky, now is your chance to talk about something other than WWII! 🍿