As they point out, the funnel plot (https://www.dropbox.com/s/fsc6lgxx25tf5zb/2018-gordon-supplement.pdf?dl=0 …) is a little wacky but the trim-and-fill is basically the same.pic.twitter.com/W4BbFGIuSn
NOT how I want my forest plot to look like. All large studies close to 0, tons of small studies suspiciously close to p = .05 on the positive side. Blinded studies found smaller effects. https://jamanetwork.com/journals/jamapsychiatry/article-abstract/2680311 …pic.twitter.com/95wMdUC2LN
-
-
-
I don't put much stock in TF. Why don't we just simulate some realistic funnel plots and put some NNs to work on them?
-
What would be the ground-truth? There's no particular reason to think a few handwritten simulations capture all the heterogeneity and bias patterns, and if you do want to claim that, you don't need NNs to do the inverse inference.
-
You can introduce varying and realistic levels of true heterogeneity, publication bias, data error etc. Who's to say people have the right theory to work out a way to estimate true effect size? Who needs theory when you have NNs?
-
ie you can invert a simulator with ABC or other likelihood-free approaches. Effect size/precision is a 2D variable, there's just not much information there...
End of conversation
New conversation -
-
-
But how do these effect sizes compare to other treatments (SSRI, CBT)?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.