Reposted by Tim Morris
This semester I taught matching for the very first time, so I had to look up a lot of stuff. Whenever I found an answer and thought “perfect, that’s exactly what I was looking for”, it turned out to be written by @noahgreifer.bsky.social 🙏
0 replies
4 reposts
25 likes
Observed effect sizes from your study is what we’re talking about
1 replies
0 reposts
1 likes
You’re right. Post hoc power is a nonsense concept. The notion of power is useful for planning but not once the results are in.
I like this paper by Andrew Althouse.
www.journalofsurgicalresearch.com/article/S002...
1 replies
0 reposts
5 likes
Ronaldo might start crying soon folks
0 replies
0 reposts
0 likes
Unhelpful, sorry. Do you know where those spikes are from? Is it a visualisation artefact, e.g. using 4 bins for every 5 possible values of sensor stress, so that every fourth bin includes two integer values instead of one and that makes it spike?
2 replies
0 reposts
1 likes
2 replies
0 reposts
4 likes
Enjoying that NaN at the bottom right
0 replies
3 reposts
15 likes
I don’t know how to break this to you but… Max Parmar is about as frequentist as they come
0 replies
0 reposts
0 likes
Thinking about this more, upstrap approach seems to achieve the same thing as simply analysing the interim data but replacing n_interim with n_target when calculating in the SE.
1 replies
0 reposts
2 likes
I think you aggregate over the whole lot right? Because power will be a frequency over all upstrap samples. Have I misunderstood?
1 replies
0 reposts
0 likes
Is doing this at interim analysis somehow better because we get to make the decision about whether we want to see the bigger trial? Or is ignoring inherent noisiness of interim estimates still problematic?
I really haven't thought much about it so these are actual questions.
4/
1 replies
0 reposts
2 likes
We tut when people interpret their studies as 'result wasn't significant but with a bigger trial it would have been' because a bigger trial would not just tell us the same thing.
3/
1 replies
0 reposts
1 likes
This seems close in spirit to plugging in interim estimates to your original power calculation and finding out power given your originally-planned sample size. Upstrap adds a bit of sophistication by allowing for model misspecification.
2/
1 replies
0 reposts
1 likes
Just skimming over this paper on 'upstrapping' for interim analyses of trials. Idea involves repeated simulation where you resample-with-replacement n (full trial size) from <n (interim sample size) and look at result.
Not convinced.
1/
trialsjournal.biomedcentral.com/articles/10....
1 replies
1 reposts
1 likes
Nice. I suggest grappling publicly with the notion that what you do may not matter. That one got some good engagement.
1 replies
0 reposts
2 likes
Erupted only has one r doesn’t it oh dear oh dear
0 replies
0 reposts
0 likes
Oh no, has this errupted somewhere again?
2 replies
0 reposts
1 likes
Haven’t heard from them in a while… or perhaps ever. Are they active?
1 replies
0 reposts
1 likes
Reminder of our (unconstructive) paper on the marginality principle.
tldr: the scale on which you happened to measure a variable does not justify how you include it in a regression model.
0 replies
0 reposts
5 likes
If you want to be a marginality extremist you have to not understand that it’s just a syntactic distinction, Will. I don’t make the rules.
You’re a true marginality believer if you disagree with the BPM example and think that modelling beats and minutes separately is better.
1 replies
0 reposts
2 likes
Oh no, marginality extremists! 😜
onlinelibrary.wiley.com/doi/10.1002/...
1 replies
1 reposts
3 likes
Exactly. The way to make bluesky flop is to do a mastodon and make the whole thing about twitter. The way to make it work is to make sure it’s a vaiable alternative when twitter implodes.
0 replies
0 reposts
2 likes
Reposted by Tim Morris
Tip:
We aren't comparing the effect of the treatment (tx) to the effect of placebo.
We are *defining* the effect as the *difference* in outcome between those allocated to standard of care (SoC) + tx to those allocated to SoC + placebo, where placebo is used for blinding.
3 replies
15 reposts
39 likes
Putting an s on the end of supermarket names is a cockney hallmark. Your kids will be in awe once you tell them that nugget.
1 replies
0 reposts
1 likes
You’re in! This is shaping up the be a helluva night
1 replies
0 reposts
1 likes
Indeed. At this point what’s one more!
0 replies
0 reposts
4 likes
In network meta-analysis, do ‘transitivity’ and ‘consistency’ have different meanings?
I’m reading something on transitivity and it seems to be the same concept. Am I missing something?
2 replies
1 reposts
0 likes
I still think it would be amazing to go out drinking with whoever picks out stuff for the middle aisles in Lidl and Aldi
1 replies
0 reposts
4 likes
My unit director Max Parmar is kicking off LSHTM’s annual ‘Perspectives on Statistics in Medicine’ lecture
0 replies
0 reposts
1 likes
Oh me pick me I know this one!
There’s a constant union betwixt the cause – UNIMPRESSIVE ATE DESPITE STRONGLY HELD HUNCHES – and effect – I THINK I’LL FIND IT’S A NOT MORE COMPLICATED THAN ATE CAPTURES
0 replies
0 reposts
3 likes
Another banger Julia!
Interesting point on 'starts from the opposite notion that causal effects are the same for everyone'. Seems similar in experimental design literature. Often 'assuming this model [where all individual causal effects are identical], then the optimal design…'
1 replies
0 reposts
2 likes
Thanks Anne, that makes sense. Superb paper title, bravo! I'll have a read.
1 replies
0 reposts
1 likes
Curious about 'their connection to substantive hypotheses is usually ill-defined'. Do you mean this empirically – researchers do tests that don't target what they think – or another way?
1 replies
0 reposts
0 likes
Once heard someone describe multiple outputation as defunct-when-born but would have been life-changing had it been invented 40 years earlier when computers were less powerful.
onlinelibrary.wiley.com/doi/abs/10.1...
0 replies
1 reposts
3 likes
‘Top-2% cited, yet not suspiciously prolific if we define the latter as more prolific than me’
0 replies
0 reposts
0 likes
It’s like going back in time 3–4 years!
Wonderful little JPAI special about ‘top-2% cited scientists’ under ‘Contributors and sources’ 🙃
www.sciencedirect.com/science/arti...
1 replies
0 reposts
6 likes
I was selfishly disappointed by child-free day. After initially being filled with wonder, I realised I’d completely misunderstood and it’s for the people who are also child-free on the other 364 days.
1 replies
0 reposts
3 likes
Perfect! The date for Estimands Day seems important right? Ideally when there are often conferences on… and a bit before Assumptions Day?
1 replies
0 reposts
2 likes
Great! Can we just name the date or is there something official you have to do to get the ‘___ day’ designation?
1 replies
0 reposts
1 likes