One of the biggest reasons is also meta-analyses. One study can have an outsized impact on the literature if it has a low SE and therefore a big effect size, which ends up polluting all recommendations from that point on.

I found a partial explanation, but from my understanding you can't possibly randomize a 3:2 study with block sizes of 4

bsky.app/profile/gidm...

Ok, I have partially solved the problem! The authors ran an imbalanced 3:2 randomization.

But, uh, how? Can you randomize 3:2 with blocks of 4?

Nope the study specifies that this was the allocation before dropout.

They said permutated, which implies that they are balanced but I suppose they could've done something really odd whereby each block could be entirely made up of one group. But that seems deeply bizarre. They would've had to have 5 full blocks allocating to only 1 group!

I would guess that most peer reviewers, particularly in 2009, never check pre-registrations and are so not aware that researchers have switched their outcomes.

There are also lots of other errors in the paper, but honestly who cares that much.

And this isn't some nothing of a study. Here it is being cited in the Endocrine Society guidelines which currently recommend vitamin D supplements for children to reduce the risk of respiratory disease.

There were 18/167 Influenza A infections in the intervention and 31/167 in the control. There were 39/167 influenza B infections in intervention and 28/167 in the control.

Overall lab-confirmed influenza was 57/167 vs 59/167. About as null an effect as you can imagine.

Problem is, the researchers didn't pre-register influenza A as their outcome. They registered doctor-diagnosed influenza, and didn't specify the type.

Is that important? Well...

The paper is here: www.sciencedirect.com/science/arti...

It's in pretty much every SR/MA on the topic. Japanese researchers randomized 430 kids to either get vitamin D or a placebo in 2008. There was a massive reduction in influenza A infections in the intervention group.

Ok, two main things I noticed:

1. Most of the p-values are wrong.
2. The all means of Ct values are not the weighted means of the two groups in many cases even with potential rounding.

The averages are actually correct as far as I can tell! They're just row averages for some bizarre reason.

The ct averages though...are they even simple averages?

This is the sort of critique that makes sense, but it's not an ~error~ as such. Think more mathematical mistakes and the like.

No one got the main thing, which is that the menopause durations are both remarkably homogenous (pretty much everyone in this study hit menopause at age 49-51), impossible (they are all GRIM errors), and also impossible (SDs are too low given the range)

Yes, the menopause regularity is...remarkable. It's not quite proof of anything, but it is to my understanding very unusual!

The 2 d.p. is an average. I would say that these values are odd, because they do imply that the data was COLLECTED in non-integer form, but that's also not quite an issue.

Reporting to 2 d.p. for means is not suspicious - certainly the vitamin D values are possible.

But tell me more about menopause onset. Are these timelines believable?