Robin Blythe

Going to start calling the Ashes "The Superbowl, because the bowl the ball" to trigger my Aussie friends

Please tell me this is going in an academic paper so I can cite it

I had a *lot* of exposure to (and arguments with) behavioural economists during my Master's.

I would seriously recommend avoiding all of this stuff. I know people are horrible/antisemitic/selfish/deranged etc. I don't need to be reminded of it when I want to talk to intelligent people about science

Haha I was just reading Gigerenzer on Kahneman and Tversky, and Thinking Fast and Slow is firmly off my reading list 😬

I must admit I've only seen him in the context of being dunked on by others, but what you offer is a good endorsement. Where would you start? The Black Swan?

I keep thinking about whether I want to read him, and will eventually give him a shot, but the older I get, the less time I have for people who are confidently preaching outside their subject area. I found Pearl similarly abrasive and almost self-defeating about statistics. LMK how you find Taleb

Thanks, and I think this aligns with the thinking of my cardiologist colleagues for the most part. In NZ and Australia the thinking is that you should use risk calcs/ratios for decision-making but given LDL is an independent predictor, I wonder whether that's maybe a little myopic.

Interesting - mind elaborating on the second point? I've also heard some reasonable arguments that we should be thinking about 30-40 year risk rather than the standard 10 year ASCVD (vested interest for me as my GP in Singapore keeps trying to prescribe me statins at age 35)

Yeah. I thought it could function as good PR for them, as they're trying to build social trust around their product in light of the (frankly psychotic) marketing of ChatGPT and Grok right now.

You're worried about potentially bad methods, or more wondering what their angle might be for releasing it? Both equally valid questions

More so that I find it a tedious debate. Often social research is hard to recruit for, and I'm less interested in questions of sufficient N than I am in the utility of the findings given the size of the error terms. Matters more for stuff like clinical trials where wrong conclusions harm health

52 is a pretty reasonable sample to demonstrate the main effects. Compelling findings

That's brilliant teaching material

That's what I proposed! Got shot down I'm afraid. Propensity scores fix all data problems, apparently

Thanks for sharing -- didn't know this existed!

Machine learning based screening of potential paper mill publications in cancer research: methodological and cross sectional study Objectives To train and validate a machine learning model to distinguish paper mill publications from genuine cancer research articles, and to screen the cancer research literature to assess the preva...

Congrats to @aidybarnett.bsky.social, Baptiste and team on this important #metaresearch paper. I got to see the early analysis stages of this work and have been waiting with baited breath for it to be published:
www.bmj.com/content/392/...

I mean more in the sense that it appears to be a shorthand way of saying two variables influence each other. I don't think I've seen it taken much further than that, although as the replies to this thread show, there's some weird stuff out there.

Haha I think you're right, but I'm projecting a little here because of a looong argument I've been having with some colleagues who refuse to let the data "go to waste"

Incidentally, DAGs are a nice tool to figure out when your dataset is completely useless for your research question and you need to just stop. Unfortunately this is when some causal inference studies double down that they've found "truth"

I guess another issue might be that you might have an effect of X1 on Y1, but only because the timing of measurement had induced it. I.e. your DAG depicts a more granular time than the data are equipped to detect.

I interpret it as
X1 -> Y2,
X1 -> X2,
Y1 -> X2,
Y1 -> Y2.

There's temporality but distinguishing it in the data is tricky.

It should have clicked with me when you shared the 18-in-1 body wash, the ultimate cost-effective solution

I'm a health economist and genuinely don't recall the last time I saw one on here.

Sri Lanka is a stunning place, and the food is great. Plus, there's this amazing dense fog out at sea that reminds me of The Pale from Disco Elysium

It's more complicated but for virtually everyone else the other factors aren't worth worrying about, yeah. I think it's convenient to pretend calories aren't important because being in calorie deficit usually feels like shit

What made you think it wouldn't work?

The primary challenge is the counterfactual. Medicine interferes with Pt trajectories, as everyone is treated. We need to check whether the treatment differs between groups, then infer whether that treatment difference is meaningful and positive. Often too much noise to pick up a signal from CDSS.

Gave a fun talk last week on how we should view clinical decision support as behavioural interventions and not medical ones. A key distinction: we need to analyse (and collect!) proximal outcomes like Tx orders rather than always looking at patient outcomes to judge effectiveness #statsky #episky

As in, 2 x 0 is still 0?