Yeah, I left for a reason. I'd go back for family reasons but not for my career unless something changes with their approach to R&D.

I'm pretty sure the only ship traffic is the inter-island ferries because it's too dangerous for other vessels to risk it 😂

Interesting point, I never really thought about the semantics and connotations. Maybe a better framing for, say, a choice experiment vs a retrospective study would be intended and observed behaviour?

You cannot convince me these people are real.

Richard McElreath -- tagged in the OP

Not sure it's exactly the same, but when dealing with censored data and no real predictors of note, I've found fitdistcens() from fitdistrplus to be ace. I tried to one-up it myself using strong priors to inform a latent variable model but it was, at best, equivalent to the fitdistcens() approach.

I was absolutely positive about 3 years ago that this was the play for AI. I'm actually surprised it took this long for one of them to start doing it.

I wouldn't downplay your field. If bad science leads to bad policy, the impact on wellbeing can be substantial, can't it? The right thing to do would be at least a letter to the editors. One shouldn't be precious about their work being critiqued; knowledge is (or should be) public domain.

I've sometimes thought you could boil most stats down to the astronaut meme, with "It's all matrix multiplication?" -- "Always was"

I think this is the first resource that made Bayes click for me, hard to overstate how important that is given the subject matter. It's excellent.

Schalke got Hitler, Arsenal got Bin Laden. The comparison isn't too bad

The sad/funny thing about the (real) CRediT taxonomy is that I am doubtful it's ever stopped undeserved authorship in a situation with real power imbalances.

Yeah, hard to watch nowadays. I reckon they're so unused to winning that they get nervous and play with mortal fear or losing the ball when it's really do-or-die.

Modelling isn't just for the computer

Well, the feet are deliberately cropped 😘

Let's just say I was savouring it by only reading a chapter every fortnight

Finally finished! Never thought I would enjoy reading a textbook front-to-back, yet I did, immensely. The best thing about the book is getting you to think of data generating processes. The flexibility of Bayes in creating scientifically valid equations is incredibly motivating. #rstats #statsky

I don't think it's controversial, just unrealistic to think we'll ever make a special rule for gambling companies. Though, at the moment, I'm developing an eye twitch from having my gmail inundated with "200 free spins" emails every 3 minutes. Gmail does nothing to address it, as you would expect!

I think of these things as more or less a declaration that they plan to commit similar atrocities in the near future.

I was wondering that. Maybe the Bayesian trial design stuff, but that's been in the works since well before the last few years. I'm actually surprised they didn't kill it, perhaps someone told them Bayes is AI

Depends on the country, the position and the university IMO. Slacks and a dress shirt would be fine in lots of Australia and NZ for research fellow and below. If in the US I'd probably bring a sports coat and maybe a tie depending on the institution and role. Asst prof and above, probably suit.

Partly agree, but IIRC even small co-pays can push some people out of accessibility to essential services, and the financial arguments against moral hazard tend to be empirically weak (in healthcare anyway). My suspicion is that they would just find another reason to shit on the poors.

To be fair, it might have been painful to watch them against OKC anyway

Have noticed that journals will often desk reject my/my colleagues papers based out of southeast Asia on the basis of "lack of generalisability." Titles and aims need to be finessed to dupe editors even to make it to peer review. No such problems when I worked in Australia! #medsky #academia

In all seriousness, it's unfathomable why editors are completely incapable of making the substantive judgements required to accept or reject based on review. Task shifting back to reviewers is a scourge.

Without getting into the weeds of defining performance, I agree. But unless the authors have a good reason for dichotomising or stepwise that isn't just because they can't fit a spline or talk to a subject matter expert, I'm still going to be pretty sceptical.

Two reasons we should publish clinical prediction models in journals:
1) it's going to be used in clinical practice and needs to be combed through, e.g. the NHS uses it, or 2) it's a genuinely novel method that solves a real clinical problem. The rest is just clogging up PubMed. #statsky #medsky

Ok, but then I also think we publish too many prediction models, and unless it's accurate, has substantial clinical utility and generalisability, I don't think it should be getting past a desk reject. If we aren't trying to make things better in peer review, then what's the point?

Ugh no matter how many times I ask, Spotify insists that actually, I LOVE Moby, and all those requests to never suggest him again are not real. Trade you?

Seems to me the clearest solution is that said crappy papers shouldn't be making it anywhere near peer review. The fact that they are suggests journals prioritise predicted "impact" over good science. Nice thread btw, agree with all of it.