Rich Morris

That’s what drove me to always declare the method!

plot provided by @Chris_said

Trend line plot showing major depressive episodes in US teenagers peaked in 2021 and have declined since then (Source data: National Survey on Drug Use and Health)

Major depressive episodes among US teenagers peaked in 2021 and have been declining since. More evidence challenging simplistic theories of teen mental health

I did not plan to read all 240 posts on observational evidence today. I did not WANT to read all 240 posts on observational evidence today. What did you do today? Sigh…

The trouble with being an #rstats celebrity is that everything you post about statistics will be taken veeeery seriously :-P

that seems like an odd design for an RCT (would it really be an RCT?). Do you have any real world examples in mind or were you just offering a thought experiment to challenge my (our?) assumptions?

If we distinguish along those lines, then ANCOVA is always preferred when we have a clear expectation the groups do not differ at baseline (eg RCT). Where I'm unclear is the observational case when groups may differ and we want to decide whether to equate differences with ANCOVA (or matching etc)

this would also help me understand @f2harrell.bsky.social position on this issue better. (Frank have you ever written anything on Lord's paradox I could read somewhere?)

(I actually thought the paper in the op was really helpful but found the more subtle logical inferences too difficult to follow)

I think someone posted earlier that medicine is concerned with the expected effect of an intervention in the population (so ANCOVA), while education/etc is concerned with compensatory effects among subgroups (so change scores). Would like to hear if that is explained somewhere

is there an easy distinction between the questions to help understand the differences here?

e.g., Does strength training compensate for differences between men and women? vs does training increase strength in men and women?

I'm trying to understand the wrinkles you mention here. Taking your off-the-cuff eg, using ANCOVA with baseline strength would potentially introduce regression-to-the-mean (RTM) effects right? Because ANCOVA would compare post score strength in weak men to strong women... and so produce a biased ATE

That's a good point. However so far that has been mostly unsuccessful hasn't it? I recall it broke grok and they had roll back the changes...

I feel like we are in a golden age of AI right now, like when google first appeared and you could get your search results without adverts. How long before AI results start incorporating ads?

how on earth does anyone search up for details for useR! ??? Google just returns a list of results about user groups etc...

Steve And Fatima Deadbeat · Wax Poetic For This Our Great Resolve · Song · 2018

"Either politics throws open the doors, or it won’t be long before they’re kicked in by angry citizens as they smash every last stick of furniture and walk out with the chandelier of power" -this was released two years before Jan 6th

open.spotify.com/track/4Sw9Xa...

Further evidence of my assertion that most of the useful information on the internet is still text. Just need a browser which presents text

Yeah so there's no way anyone is coming here for World Cup or Olympics. if you can bag up the son of one of the most famous boxers to ever live, you can do that to anyone.

protip for Gen Z working with older colleagues: if discussing data or studies which occurred entirely after the year 2000, you can refer to them as *21st Century*. It will sound futuristic and sci-fi to anyone who grew up on Buck Rogers, Flash Gordon (i.e., Gen X, Boomers) - very impressive!

A brief history of medical statistics and its impact on reproducibility. You have to admit that medical research is a strange bird.

statsepi.substack.com/p/a-brief-hi...

This is a neat summary of the history of applied statistics I’ve read

We constantly empower people that have no interest in the truth whatsoever and I struggle to understand what can be done about that.

New blog post:
You can outsource the grunt work to an LLM, not expertise

brodrigues.co/posts/2025-0...

The debate feels like a tv melodrama. Will they, won’t they? Oh they do! At last!! Oh no the old boyfriend has come back!!! What will happen next? Stay tuned readers

Add this to my reading list: “For “random intercept” models, adding the group-level averages of all included variables as regressors (or a variety of equivalent procedures such as group-wise centering) relieves the bias otherwise induced by this regularization”

Spoke to Jen a few weeks ago because I am also trying to pivot out of academia. Can confirm Jen is a great person to help you focus on what’s important (to anyone else who is in a similar situation). She would be a great asset to any community building enterprise

After two weeks, I'm finally done!

In this post, I explain different approaches for solving linear regression in R: directly, using QR, singular value and Cholesky decompositions, and do some benchmarking for comparison with in-built approaches.

thomvolker.github.io/blog/2506_re...

I fear it is only a matter of time before he starts posting on #EconSky and any other niche expert domain he thinks he has credibility in because VP

A lawsuit in Rockland County, NY, alleges election irregularities in the 2024 Presidential and Senate races.

The plaintiffs claim statistical anomalies and uncounted votes.

But do these claims hold up?

I analyzed the data and I don't think they do: christophertkenny.com/posts/2025-0...

I’m happy to hear that argument. The use case I’m arguing for is to examine whether a series of trends are changing together. Each trend is incomparable on scale (eg some have a true zero, some don’t) but it is still valuable to understand if their changepoints coincide.