Point 2 is where it's at: there are too many things we can and should ask data when making decisions - no single-mumber summary can do it all, especially without content (I believe Fisher had the same approach)!

Cameron Patrick@cameronpat.bsky.social · 13h ago

My take on p-values: 1. p-values do what they are designed to do, which is not what most people using them imagine them to do. 2. Trying to fix p-values is a trap, because no single-number summary can do what people want p-values to do! 3. Trying to explain p-values is also a trap, because...

R > Stata

Elizabeth Picciuto@epicciuto.bsky.social · 2d ago

What is your favorite example of when the cheap version of a product is legit better than the expensive one?

My 19-year-old asked me if I found my job hard. I thought it was an interesting question and I had to think about it. After discussing with a colleague, he created the best analogy. He said our kind of work is “hard like an escape room with friends”, not “hard like hard math.” I think thats right. 

my annoyance of bayesian modeling is mainly due to selection bias - I approach it when there's no other viable alternative, so i always encounter the painstaking experience of needing to iterate and refine with every fit taking upwards of 30 minutes

Q&A with @devezer.bsky.social about metascience, open science, and her work in this area.

Q & A with Berna Devezer

Critical Metascience

markrubin.substack.com

In all likelihood a favourite Bayesian beer.

#statsmeme

The ability to properly define a research question is more important today than ever before. Describe a vague research question and your data and 10 times out of 10 Claude WONT ask for any clarification before it suggests 19 different things you can *try*.

Sean Mackinnon@seanpmackinnon.bsky.social · 3y ago

Stats consulting is constantly like: Them: I want you to run (complex stats) Me: OK, what's your research question though? Them: ...A (complex stat)? Me: Research question? Them: You know, like the stats in this journal article. Something reviewers will like.

Made a detailed analysis plan for a client to run. A week later they email me results to review. They don't match our discussed plan at all. What happened? They didn't understand some of my suggestions - did they email me to get clarifications? No - they asked Claude to "fill in the blanks".

Translator Of Original ‘Minions’ Epic Poem Slams ‘Sexless, Bloodless’ Film Adaptation

Translator Of Original ‘Minions’ Epic Poem Slams ‘Sexless, Bloodless’ Film Adaptation

PHILADELPHIA—In a 4,500-word essay published in Harper’s Magazine, Emily Wilson, known for her popular translation of the Minions epic poem, blasted the latest film Thursday as a “sexless, bloodless” ...

theonion.com

I have found a great new #stats teaching hack to find new open datasets for teaching. I've been going to the Collabra: Psychology journal website, and typing in an analysis type (e.g., logistic regression) into the search bar. Both quick and high-quality data! online.ucpress.edu/collabra

New blog post! Odds ratios are...a bit weird. Here I explain two problems (interpretation, non-collapsibility) and provide some simple recommendations for improved reporting. www.the100.ci/2026/07/30/r...

Reviewer notes: Odds ratios are really odd

When psychological researchers investigate binary outcomes, they routinely report odds ratios to quantify the strength of an effect or an association. This is quite understandable given that psycholog...

the100.ci

More thoughts about the normal prior vs the student's t prior dpananos.github.io/posts/2026-0...

Adaptive Shrinkage and the Student’s t Prior – Demetri Pananos Ph.D

dpananos.github.io

I am begging you to talk to other human beings directly, and not through a Claude layer. This goes doubly when I ask you, directly, to talk to me not through a Claude layer. Please.

A client thought up some BS with Claude and sent me the full report + analysis code, all 100% written and run and interpreted by Claude. I gave a detailed response explaining why it was trash, which my client (for laughs) forwarded to Claude. Claude was not happy being criticized.

Peter's Law of Chatbot Accuracy*: "everything you read that's produced by an LLM is absolutely true, except for the rare output about which you happen have expert knowledge". *Remixed from Knoll's Law of Media Accuracy, where the exact same rules apply!

By Cory Doctorow (GPG 0xBF3D9110957E5F4C)@doctorow.pluralistic.net · last wk.

AI is a normal tech. That means there are times when it is useful and times when it is pointless or actively harmful. One rule of thumb for chatbots is that they can only provide useful information to experts who have the discernment to ignore the defective output that LLMs *always* emit. 60/

#Position 's auto format for @quarto.org files is extremely slow recently, and does *not* work well with rmd files (breaks formatting) - anyone else experiencing this? Any solutions out there?

Me to students: in this course we will not be using the ~ notation in @mc-stan.org for reasons that will be made clear later. Me, later: see when estimating the marginal likelihood we need all the normalizing constants, so be wary of LLMs that like using ~ notation. Students' assignmens: ~~~~~~~~

LPA is real and it CAN hurt you but the profiles it generates aren’t

Scary stories about teens sell. The data tells a different story. After decades studying adolescent mental health, here's what I found: This isn't an anxious generation. It's a resilient one. Let's start treating them that way. go.ted.com/candiceodgers

What we're getting wrong about teens and tech

For years, the warning has been: smartphones are destroying a generation. But developmental psychologist Candice Odgers says that decades of data on teens tells a different story — violence, alcohol u...

go.ted.com

When you find an #rstats package that does exactly what you need, but it hasn't been updated in 10 years, doesn't allow for parallel processing, and nothing is vectorized... I'm gonna have to ask an LLM agent to re-write this, hu?

📉 The AI jobs apocalypse that isn't Dario Amodei warned half of entry level office jobs could vanish. A new report finds it still isn't showing up in the labour figures. 🔗 theguardian.com/technology/2026/jul/25/ai-jobs-apocalypse-human-labor #AI #Jobs #Anthropic

The AI jobs apocalypse probably isn’t coming anytime soon

Artificial intelligence may not deliver on its promise of vast economic opportunity at a price that humanity is willing to pay

theguardian.com

i see a particular grievance regularly on here: "a paper I reviewed came back to me at a diff journal. the authors didn't address my concerns at all." seriously what are you doing accepting to review the same paper? let people breathe & have another chance. let them decide what's worth addressing.

Matt Weiner@mattweiner19.bsky.social · 2w ago

will the next reviewer have the same objection? if you revise to address the objection, will the next reviewer say "I don't see why the paper spends so much time addressing this objection that no one could possibly have"? who knows

Peer review is so random bro. Got 2 papers under review. One is clearly an excellent piece, splashy and well-argued. It's on its 3rd round of revisions with a reviewer who doesn't understand it. Other paper is shorter, "worse" (more narrowly focused). Just got the easiest first round R&R. Dice rolls

Since paper here is nested in journal, those 2 sources of variance can be IMO combined to represent between paper variability. And even w/ this, we get a generalizability score (reliability) of 0.31 for 2 judges and 0.40 for 3 judges. You'd need 11 (!) judges to get a reliability > 0.7! 1/

Alejandro Montenegro@aemonten.bsky.social · 2w ago

"Here we show, in a large post-publication peer review database, that research assessment is driven more by differences between evaluators than by difference in the evaluated research. " arxiv.org/abs/2607.09783

This should come as a surprise to no one who has ever gone through Peer Review™.

Alejandro Montenegro@aemonten.bsky.social · 2w ago

"Here we show, in a large post-publication peer review database, that research assessment is driven more by differences between evaluators than by difference in the evaluated research. " arxiv.org/abs/2607.09783

New preprint w/ Leo Tiokhin & @lakens.bsky.social! Registered Reports (RRs) are great for science b/c publication is guaranteed before results are known, reducing publication bias & QRPs. This feature supposedly also makes them great for *scientists*. But is that really true? 1/ tinyurl.com/4sy993rr

Career incentives make Registered Reports unattractive under realistic academic conditions

Registered Reports are a publication format designed to reduce publication bias by guaranteeing publication before results are known. This guarantee is considered attractive in the prevailing ‘publish...

tinyurl.com

"Results were analyzed using AI (Galton, 1885; Yule, 1897; Student, 1908; Box, 1958)" #stats

Things will get worse as more students secretly use AI to do homework. There has always been a divide between those who care about education for its own sake and those who only care about grades, but AI will worsen this divide, leaving fewer people with the ability to do causal reasoning.

The Volatile Mermaid@ohnoshetwitnt.bsky.social · 2w ago

I’m sorry what

Descriptive studies overadjusting for a dozen variables. Seemingly causal studies missing major confounders or adjusting for mediators. We recently received an R&R where the reviewer asked why we didn't adjust for confounding by a set of mediating factors in our descriptive study. #episky 2/3