ish

[ish @Osteichthypfc]

@RichardHanania Are you 100% sure you've named the right woman? No doubt at all?

[ish @Osteichthypfc]

@ZKForTre @emukamel The set of statistical tools/approaches that biologists typically use appears to be quite limited compared to what is actually out there.
Anyway if they don't want to do FDR they can just prespecify a small set of comparisons of interest beforehand, preferably publicly.

[ish @Osteichthypfc]

@seanluomdphd @sunsopeningband I recall interacting with you about causal inference previously, which you appeared to have quite a poor understanding of. It seems like you still understand it quite poorly.

[ish @Osteichthypfc]

@5_utr Please stop writing this stuff with AI. If you want other people to read it demonstrate you've taken the effort to write it yourself.

[ish @Osteichthypfc]

@whyjohngalt @MetricReject You know the number of publications doesn't tell you much about how much someone's research is actually worth?

[ish @Osteichthypfc]

@loganthompson According to pangram this is entirely AI generated </3

[ish @Osteichthypfc]

@joefrancis505 Actually that's more "what". I don't know why, I suppose a mix of incentives, low methodological competence, and clinicians not being trained to do research, but doing it anyway

[ish @Osteichthypfc]

@joefrancis505 Lots of selective outcome reporting in clinical trials, bad statistical methods, very bad causal inference methods, people don't follow reporting guidelines despite saying they do, outcome measures aren't validated well, fraud in less good journals

[ish @Osteichthypfc]

@joefrancis505 I don't think this is to do with politics though. There's a lot of issues with medical research, and people just don't have incentives to fix them and most people in medical research have little methodological competence. Doesn't have much to do with being right or left wing.

[ish @Osteichthypfc]

@joefrancis505 The peer reviewers noticed a few issues but they missed a lot too. If you're publishing articles as important as this one you don't really want peer reviewers to be missing a lot of issues! But yes the blame is more with the editors

[ish @Osteichthypfc]

@StanfordMed The study says "our results do not address causality", and then they go on to make causal claims in the press release. Very typical.

[ish @Osteichthypfc]

@JarrydBartle Doesn't seem like very good reasoning. A low cronbach's alpha doesn't actually tell you whether or not that scale is doing what you want it to be doing

[ish @Osteichthypfc]

@mdonnino @TomKindlon Not to be rude, but methodologically I think this is a mess. It's terrible for patients that they don't get higher quality research, and that should be pointed out.

[ish @Osteichthypfc]

@mdonnino @TomKindlon There's also other issues with this trial, like the highly non-specific inclusion criteria, appearing to include people on the basis of their approach to the treatment, and I was unable to find good evidence for the content validity of the primary outcome in long covid

[ish @Osteichthypfc]

@mdonnino @TomKindlon This doesn't really matter but stable differences in how participants answer over time don't bias the trial due to randomisation. In expectation that becomes balanced between the arms. That's just adjusted for in order to increase precision. Doesn't help bias in the trial

[ish @Osteichthypfc]

@mdonnino @TomKindlon It sounds like you're just going on vibes instead of actually having a serious understanding of the methodological judgements that you're making! There is literature on this you know, indicating that, yes, these biases can be real

[ish @Osteichthypfc]

@mdonnino @TomKindlon A mixed model would not do that. It can control for stable differences in how participants use questionnaires but not for biases like what we are discussing.

And "getting better" is a very subjective thing to judge so actually, yes, that is extremely easy to imagine.