Stuart Ritchie 🇺🇦 Profile picture
Research Comms @AnthropicAI
4 subscribers
Aug 28, 2023 • 6 tweets • 2 min read
This morning’s big media splash about ultra-processed food is based, as far as I can tell, on a couple of conference presentations. Unless you were in the room you can’t actually check the quality of any of these studies (and even then, you probably just saw a few slides!).


Image
Image
Image
Image
This happens every so often: a conference attracts big media attention even though no findings are available for non-attendees to check. Quite a coup for the conference PR people but not good for anyone’s understanding of the science.
Jun 6, 2023 • 13 tweets • 5 min read
Lots of coverage today of the UK study that says breastfed kids are more likely to get good exam results.

This is the study: adc.bmj.com/content/early/…

I can't possibly write another article about this because you'll think I'm obsessed, so here's a Twitter thread about it instead: Image Here's the headline claim as written in the press release (eurekalert.org/news-releases/…) and reported in loads of news articles.

Breastfeeding for 12 months = 39% more likely to get an A/A* for English and Maths; 25% less likely to fail English.

So does breastfeeding = smarter kids? Image
May 13, 2023 • 13 tweets • 3 min read
On further reflection I've decided to delete this tweet. "V imperfect" is nowhere near strong enough to describe this preprint ().

I really dropped the ball by not criticising it more deeply - sorry! I re-read it, and some criticisms follow: https://t.co/oP4PcWBYJJmedrxiv.org/content/10.110…
It's interesting that the features they used ("personal email address" + "hospital affiliation" + in a subset "cites other papers with those features") are common to so many fake studies. But the reason they had such a high false-pos. rate is that lots of non-fakes have them too.
May 9, 2023 • 18 tweets • 5 min read
🧵A little thread with some observations on the new "gender inequality linked to brain thinning in women" study in PNAS. pnas.org/doi/epdf/10.10…

This is what I'd have said if I'd reviewed the paper: The argument here is that a country's level of gender equality is related to (heavily implied: causes) brain sex differences, specifically a thinner cortex in women in more unequal countries. But only in the right half of the brain (and one tiny area in the left): Image
May 8, 2023 • 8 tweets • 4 min read
Have you heard the idea that you should eat "30 different kinds of plant per week" for optimal gut health? It's the latest big thing on nutrition websites. I wrote about it today: inews.co.uk/news/rule-eat-… As far as I can tell, it traces back to this paper, which is a sort of "citizen-science" project where people filled in a questionnaire and sent in faecal samples for lab analysis (yes I know, quite grim): journals.asm.org/doi/epdf/10.11…
Apr 2, 2023 • 4 tweets • 1 min read
Remember the "hungry judges" study, which has now been included in two huge bestselling books by Daniel Kahneman (among many other places)?

Now we have another relevant study, and it reminds us that the original one was just too good to be true: inews.co.uk/news/do-judges… I'm far from the first to say this (see article above for refs), but the original result was never plausible. The judges *always* drop to ~0% positive judgements before a break, and then *always* jump back to 65% after they've eaten?

Something went very wrong here!
Jan 25, 2023 • 4 tweets • 2 min read
I'm really impressed by the research on semaglutide for obesity. 15% weight loss on average is an amazing result, and it comes from several trials.

But, as I explain in this article, I'm *not* impressed by their new psychology study... inews.co.uk/news/health/se… These are a lot of questions, asked multiple times, on which they do a lot of analyses. And then they say "analyses were not controlled for multiple comparisons".

Why not just... correct them? Is it because there wouldn't be much left if you did the proper correction?
Jan 24, 2023 • 5 tweets • 2 min read
Anti-ageing research is super-important, so it’s a real shame it’s subject to so much hype.

Here’s my article on the two recent studies about life extension and age-reversal in mice, and why we shouldn’t get TOO excited just yet: inews.co.uk/news/science/s… Something I've noticed about these kinds of papers is that they're very short on statistical detail. You get graphs like this, and it just says "Two-tailed unpaired t-test; ** p<0.05". No table, no SD info. You'd have to scrape the data off the image if you wanted to check it. Image
Aug 10, 2022 • 6 tweets • 2 min read
“Reflexively defend all humanities papers when they’re criticised” is a really bad heuristic that leads you to some terrible places! Here's another example. I don't think you had to "dig" very far, given the *first sentence of the abstract* of the paper in question contained the phrase in the screenshot below!
Jul 10, 2022 • 5 tweets • 2 min read
Oh my god, I can’t handle it, I can’t take it But seriously, some of the responses to my breastfeeding article have been jaw-dropping. “It’s bad to write long articles querying the evidence.” “This is stopping people getting breastfeeding support”. “You must be being funded by the bottlefeeding industry.” 🤪
Mar 15, 2022 • 14 tweets • 3 min read
Last night I got trapped in a toilet. Here’s a thread that explains how: I’m in Dubai at a conference. After a very enjoyable dinner I came back to my hotel and thought: “I’ll have a bath!”. I always have a shower at home, never a bath, so why not?
Jan 25, 2022 • 23 tweets • 8 min read
Thread! Everyone's talking about this new study showing cash transfers "speed up brain activity" in babies (pnas.org/content/pnas/1…).

It must be a big deal, because it was announced with a "Breaking News" tweet by no less than the New York Times: A lot of scientists, and university press officers, would KILL to have one of their papers announced in a breaking news tweet by the NYT. To get such attention, the research must be really robust... right?
Aug 13, 2021 • 8 tweets • 2 min read
Sorry, but there's no way the absolutely massive IQ difference in this study is real. An average IQ of 100 for kids born before the pandemic compared to an average of 79 during it? No - sorry, just too huge an effect to be plausible. medrxiv.org/content/10.110… Indicator that you should be wary of this paper: the authors write in causal terms ("direct evidence of the developmental impact of the COVID-19 pandemic"), when it's an observational (non-causal) design. They even use both "observational" and "impact of" in the title 🥴.
Jun 23, 2021 • 5 tweets • 2 min read
It's blindingly, crashingly obvious that almost everyone obsessing about ivermectin online is either grossly naĂŻve or a crank.

Doesn't mean the drug doesn't work (though I'd argue the evidence does not look good)!

But let's be real: grossly naïve or a crank. I think @GidMK does a great job here: if you look at the highest-quality studies (there are a LOT of low-quality ones!), there's nothing going on. Again, doesn't mean a study won't come along and show it works! But the sheer attention this gets is bizarre. gidmk.medium.com/does-ivermecti…
Jan 1, 2021 • 6 tweets • 3 min read
I meant to do this yesterday: a thread of my favourite stuff I wrote on COVID in 2020!

1) Written in a flash of rage at psychologists (and psychology writers) who told everyone they were worrying far too much, even as the deaths ticked into the thousands: unherd.com/2020/03/dont-t… 2) In which I recount the sad case of John Ioannidis, a former "hero" of science reform who stuffed up badly on COVID for still-unexplained reasons: unherd.com/2020/06/why-th…
Dec 30, 2020 • 7 tweets • 2 min read
A related (and underdiscussed) lesson is that some people just absolutely love conformity. What else could explain all the doctors/academics getting behind the 100% definite "masks don't work" stuff in Feb/March? They're not daft. They knew there wasn't a good case for that. The "evidence has changed since then" excuse can't explain it, because... it hasn't, really. It's not like a solid RCT has come along that definitively shows they work: a combination of theory and modelling strongly implies they're useful.
Oct 26, 2020 • 5 tweets • 2 min read
Really enjoyed writing this piece, on COVID-19 and the need to convey uncertainty, with my pal @MWStory! unherd.com/2020/10/how-th… Something we didn't mention in the article was the phenomenon of people who'd vehemently & harshly criticised the use of masks who now have mask display pics and/or mask emojis in their Twitter names. Funny old world.
Aug 16, 2020 • 11 tweets • 3 min read
Thread: Finally got around to reading this - a really brilliant investigative book that shows convincingly that the super-famous “On Being Sane in Insane Places” study was probably fraudulent. Image You know the one: where Stanford’s David Rosenhan sent entirely healthy “pseudopatients” to psychiatric hospitals in the early 1970s, and had them complain of auditory hallucinations. They were all admitted, and even though they behaved 100% normally afterwards...
Jul 10, 2020 • 9 tweets • 3 min read
Delighted to have the lead book review in The Spectator! And that's despite the fact it's really critical (essentially all the book reviews I've written for the Speccie are really critical, so it's only fair). A few responses in the thread below... spectator.co.uk/article/how-fa… There seems to be a fundamental contradiction in the review. On the one hand he says it's naïve to be surprised/shocked at the amount of scientific fraud/bias: everyone should know it happens constantly. But he also argues that replication rates are really high (92%!).
Mar 13, 2020 • 15 tweets • 4 min read
Long thread -> This is a great BBC Newsnight interview with Prof. Graham Medley about the modelling of the coronavirus spread and herd immunity. I reckon the "act as if you have the virus" advice is genius and should be widely shared: Later in the Newsnight episode, there was this chart, showing that the UK has chosen not to enact any of the measures (like school closures, etc) used in other countries to try to slow the epidemic.
Jun 20, 2018 • 4 tweets • 4 min read