LucibeeHey @statsguy - what do you make of this?
https://github.com/sophieehill/sweeteners/blob/main/ReadMe.md
Full BSky thread captured here: https://skywriter.blue/pages/did:plc:wk4ks4sjxkducttvaawps2wb/post/3ly44btd47s2z
Adam Jacobs 🇺🇦@Lucibee Well my initial thought is that most studies that say [commonly consumed food or drink X] causes [common health condition Y] are bollocks, so Sophie probably has a point. But let me take a closer look.
@Lucibee OK, so she has a good point about P hacking. It's a massive red flag that the effect on cognitive decline was only observed in the under 60s and not in the over 60s. I would have expected there to be a lot more cognitive decline to measure in the over 60s, so that result just screams type I error.
@Lucibee And she's right about the confidence intervals having an obvious typo (at best) in them
@Lucibee And of course all the points she makes about correlation vs causation are entirely valid.
So yes, I think Sophie's assessment looks pretty good.
@statsguy She was also wondering about the weird multiples of 8 in the CIs. Could that be a sparsity issue?
I can't see the paper, unfortunately.
I can't see the paper, unfortunately.
@Lucibee The multiples of 8 are indeed weird.
My best guess would be that they did the original regression on a different scale which differs from the scale they reported by a factor of 8. They then took the regression coefficients and CIs to 3 dp and manually multiplied them all by 8.
The report results over 8 years of follow-up so perhaps the regression was for annual cognitive decline.
If you did that, it would explain it.
@statsguy Do they explain in the text in real terms what the βs (slopes) actually mean in terms of decreasing function? I'm wondering if they don't know themselves. And they seem so tiny.
@Lucibee I don't have access to the paper either, but from the bits of it she posts on the github page it looks like the outcome measures are z scores, so presumably they're the difference in change in z score per 8 years.
So yeah, tiny.
@statsguy It's also weird from the contribs section that the epidemiologists involved "acquired the data" but weren't involved in analysing it. 🤔
@statsguy
Sophie sent me a hint...
https://github.com/sophieehill/sweeteners/blob/main/paper/paper.pdf
Sophie sent me a hint...
https://github.com/sophieehill/sweeteners/blob/main/paper/paper.pdf
