Stephen Wild

@stephenjwild.bsky.social

545

Followers

374

Following

682

Posts

I try to put straight lines through things but usually fail. Try to be Bayesian when I can. Views my own. RT/like != endorsement.

Posts Replies Media

Kid movies can be very wise. youtu.be/bSWVi3X1ho4?...

one often meets his destiny on the road he takes to avoid ityoutu.be kung fu pandaturtle wisdom

Reposted by

Stephen Wild

A cool example of a more than 100 year old data visualization from the Romanian census @datavisfriendly.bsky.social you might like this!

Do your thing R people. I think the answer is no. My choice would be mgcv or gamlss, but I don't think they could do this either.

Mixed effects model Repeated measures Spline with one knot BUT location of the knot is different for everyone (like a random intercept of the knot) AND a variable in the data set predicts location of the knot. Can it be done (STATA or lmer)? #stats

MANOVA is terrible. That is all. psychologicalstatistics.blogspot.com/2021/08/i-wi...

if I see MANOVA in the wild, i assume a peer reviewer told the authors it would be more "efficient" or something

I Will Not Ever, NEVER Run a MANOVApsychologicalstatistics.blogspot.com I have been thinking to write a paper about MANOVA (and in particular why it should be avoided) for some time, but never got round to it. Ho...

Reposted by

Stephen Wild

New talk (with video) on methods for constructing and analyzing composite outcomes in studying effectiveness of treatments: fharrell.com/talk/cons #stat #statistics #clinicaltrials #rct

Rare Degenerative Diseases & Statistics:Methods for Analyzing Composite Patient Outcomes – Statistical Thinkingfharrell.com In this talk I’ll explain why statistical power is maximized by analyzing the rawest form of clinical trial outcome data, as opposed to analyzing patients at a single time point or reducing rich longi...

Unforgiven is a great novie. More of you should watch it. youtu.be/jRi_u4GbfIw?...

Unforgiven - "First, you got to cock it"youtu.be English Bob gets bluffed by Little Bill.

Reposted by

Stephen Wild

And here's the no-longer-forthcoming blog post

I've been using Positron as much as possible over the past week and it's been delightful! Blog post (probably) forthcoming about how I have it all set up. #rstats github.com/posit-dev/po...

Fun with Positron | Andrew Heisswww.andrewheiss.com Combine the best of RStudio and Visual Studio Code in Posit’s new Positron IDE

Reposted by

Stephen Wild

This semester I taught matching for the very first time, so I had to look up a lot of stuff. Whenever I found an answer and thought “perfect, that’s exactly what I was looking for”, it turned out to be written by @noahgreifer.bsky.social 🙏

Reposted by

Stephen Wild

🚨 New Substack Post - A secret third way: Likelihoodist statistics I provide a quick tutorial on conducting research using the Likelihoodist approach. It's step-by-step + #R code open.substack.com/pub/mzlotean... #substack #statistics #education #guide #science #research #edusky

A secret third way: Likelihoodist statisticsopen.substack.com A tutorial on designing and reporting experiments in the most objective framework

Amazing

brilliant. Do you know Lumsden (1976) who explains IRT using the flogging wall test (sticks hitting people on the head (being hit=accuracy, stick height=difficulty, wiggliness of the stick=discrimination, height=theta). Whole paper is a hilarious masterpiece psycnet.apa.org/record/1976-...

Approved methods of handling frustration are not allowed on Twitter. I think they prefer to use desks.

Reposted by

Stephen Wild

Sharing some updates on my positconf::2024 slides 🧵

👀

Rough time for the reciprocal effects crowd: Cross-lagged panel models can’t reliably establish cross-lagged effects. Need big N, need time invariant effects, & equivalent models can't distinguish whether they exist. from Bengt Muthén & Tihomir Asparouhov www.statmodel.com/download/Rec...

Answer Andrew so I too can get back up to speed

What is the current state-of-the-art for tree-based prediction models? Is it still xgboost? lightgbm? Are people still enamored with BART? My knowledge in this area is a few years out of date but I’d love to get back up to speed. #stats #machinelearning

Reposted by

Stephen Wild

Reply to

Will Lowe

Fully agree. It annoys me to see a paper where lots of work was done on the data collection/modelling, then variables were simply thrown in a blender to make an index. I am okay with sum scores if you can convince me it represents a single factor, items are roughly equally weighted, no DIF, etc.

It's probably different for you guys, but whenever I see an index being introduced I think to myself "here's some folk who would like to measure a thing but would rather not derive a measurement model for it because that would lead to intrusive thoughts about measurement validity, DIF, and all that"

Now trying to figure out how properly account for both state and national-level polls. ?×#&

Think I'm going to do this. Time to see how bad I fail.

Looks interesting. Added to the reading list!

✅ Final revisions of my chapter for The Routledge Handbook of Philosophy of Scientific Modeling (eds. T. Knuuttila, N. Carrillo & R. Koskinen). You can read the preprint here: philarchive.org/rec/AYDTPO-4 #philsci

"Bayesian prior" 😏

Yeah, I can definitely see the challenge. Especially for something like “Bayesian” which gets thrown around (and misused) a lot in other contexts.

Reposted by

Stephen Wild

Weekend #stats question: Under a Likelihoodist framework, are hypotheses always "one-tailed"/directional? Since you have to specify simple hypotheses, H0 = 0 and H1 = 2, does that mean I'm only testing against +2? or against |2| (so I can check -2 as well)?

Looking for some good books on the science and impact of climate change. Hit me up with your suggestions

Since we're discussing potential change in vote intention due to the debate, a reminder about polls and statistical power. There's an old Jackman presentation that has a nice plot showing the sample size required to detect change in voter support. I think it assumes SRS.

Think I'm going to do this. Time to see how bad I fail.

I am debating building a model to forecast the US presidential election. Any suggestions for other models to look at?

Reposted by

Stephen Wild

After seeing papers I can't find anymore discussing lme4 approximations of SEM models, I thought the major limit was constant residual sd. Trying it, I can't see how to scale random effects by an estimated par (ie factor loading) though - possible? links to papers?

This a very good summary of using survey weights in regression models

This is a blog post for helping people unfamiliar with survey weights understand their use in regression models: www.practicalsignificance.com/posts/survey...

Practical Significance - What we estimate when we estimate regression models with survey datawww.practicalsignificance.com I explain the reasoning behind why survey software estimates regression parameters and their standard errors the way it does.

Reposted by

Stephen Wild

Does anyone have a good resource to get a good understanding of survey weights? Like journal article length, not book length. These are super uncommon in psychology, but I'm on a dissertation committee where a student is using Statistics Canada data that has weights #StatsSky

Reposted by

Stephen Wild

How we got here... #stats

Oof... but also not wrong. Possibly deserved, one might say.

Reading Cosma Shalizi on LLMs.

Reposted by

Stephen Wild

Researchers be like