Ya boi @GidMK aggregated all the existing estimates into a forest plot. Estimate: 0.79% IFR, 95% CI (0.53% - 1.05%)https://medium.com/@gidmk/what-is-the-infection-fatality-rate-of-covid-19-7f58f7c90410 …
-
-
Replying to @jamesheathers @CT_Bergstrom and
Why don't people write methods like in this post? Just add the technical stuff to a supplementary appendix for the interested parties.
1 reply 0 retweets 0 likes -
Replying to @mikejohansenmd @jamesheathers and
Only "works" when the researcher is relatively trusted, skilled, acting in good faith, and using run-of-the-mill methods. Otherwise, shoving all the important bits into an appendix hides the most important flaws. Probably will comment on
@GidMK's post later (mostly positively).2 replies 0 retweets 2 likes -
Replying to @NoahHaber @mikejohansenmd and
* "Mostly" is misleading. It's a good post, frankly much better than the average peer-reviewed published meta-analysis (MA). I have quibbles, but they are merely quibbles, and largely have to do with problems w/ MA under differing designs in general, not this implementation.
1 reply 0 retweets 2 likes -
Replying to @NoahHaber @mikejohansenmd and
Thanks! Happy to take any suggestions unless they involve using SAS
4 replies 0 retweets 0 likes -
Replying to @GidMK @mikejohansenmd and
Was gonna make a thread, but honestly probably more interesting to ask questions / give suggestions here. First thing: seems like an odd choice to include the models here, no? IFR in most of these models is more or less just an assumed number by the sim authors.
1 reply 0 retweets 1 like -
Replying to @NoahHaber @GidMK and
That's not ... entirely true for all of those in the "model" list. CEBM one is, itself, a parameter derived from MA + assumptions, so there's the weird meta-meta thing happening a bit. That's just to highlight the oddness of putting Ferguson and CEBM together, or at all.
1 reply 0 retweets 1 like -
Replying to @NoahHaber @GidMK and
Which is as good a segue as ever into one of my pet peeves in MA: I'm not sure it's a the best idea in general for any time in which you have many estimates with severely varying methods. The conditions under which MA is a reasonable estimate are almost never met in those cases.
1 reply 0 retweets 2 likes -
Replying to @NoahHaber @GidMK and
Specifically, you have to make the assumption that the net impact of the methodological changes is, in some sense (noting and avoiding the FE v RE) exogenous. That's not true in general, and SUPER not true with small numbers of extremely heterogenous studies.
1 reply 0 retweets 1 like -
Replying to @NoahHaber @GidMK and
And it's SUPER SUPER not true when there is every reason to believe that the errors and biases and so on aren't independent, but rather correlated between studies in ways that are untestable. Which is to say, meta-analysis is probably worse performing than your gut expert guess.
1 reply 0 retweets 1 like
In terms of the "model" studies - yes to be fair that's an entirely arbitrary designation and pretty much just lumping together things that didn't fit in elsewhere. To an extent, every study used a "model" of some sort
-
-
Replying to @GidMK @NoahHaber and
I think that actually speaks to your point - the lowest heterogeneity both statistical and in terms of design and analysis came from the studies from China
1 reply 0 retweets 1 like -
Replying to @GidMK @NoahHaber and
It's definitely true that the MA is, at best, a bit of a random collection, although I'm not sure my best guess is any better lol. That being said, I've tried to put some words around that in the piece. I'll think about how to better express it maybe
1 reply 0 retweets 1 like - Show replies
New conversation -
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.