Private channel session 197
20 messages over 1h 17m, 2020-07-01 – 2020-07-01.
A “conversation” here is an activity session — a run of messages with under 60 minutes of silence inside it. The channel had no native conversation boundaries.
-
From Bette's paper "A two-sided Fisher's exact test compares the counts in the pre-onset period to the counts after the two-week delay period and provides a p- value against the null hypothesis that the fraction of D614 vs G614sequences did not change". Really? This got through peer review? @Eddie Holmes, I'm pretty darn sure you must have flagged this - that's not how that test works. Or is she just using this as an inclusion criteria?
-
And counts are just numbers of sequences in gisaid?
-
Yes. Solid stuff.
-
I said the whole analysis was crap and needed to be redone. I never saw any response to this.
-
I submitted my review on June 12th (I only reviewed the revision). Given the speed of all this - and they wanted totake a look in 48 hours - it was clear that Cell were going to take it whatever I said. I've not looked at the paper - Ican't stand to - but I assume they've not made any of the revisions I suggested.
-
Okay, finally managed to read the full (updated) paper. The analyses are fucking crap and the statistics a completejoke. I'm really surprised this is going through - no way it passed the Eddie smell test
-
The only good thing is that they're not pushing the "it's transmitting better" narrative. I'm not saying that couldn't betrue, all I'm saying is that the data and analyses do not allow for that conclusion. I think Marc might be commentingon the paper for NYT... That could be fun.. :wink:
-
If the Editor is completely on the side of the authors what can you do? Their's peer reviewer and then ignoring peer review. Clearly Cell wanted it accepted ASAP because they were worried about being scooped and wanted thecitations. Here's the 1st major comment from my review "The statistical analysis of the D614G variant is flawed. I'm not saying that the conclusion is wrong, but the analysis is."
-
I'll use those grader quotes in my next review.
-
Our analysis of the UK data seems to suggest a higher rate of G growth but it is really hard. We consistently get ahigher rate but the significance depends on the details of the analyses. The effect size is small but not insignificant(if true). On the other hand there is not much you can do with the knowledge.
-
Do you have a visual of any of this? I'm kinda struggling with how a difference like this might even look like. The previous R0 estimates I saw for the two clades from Trevor's group essentially looked identical
-
Please don't show this to anyone just yet but here is an analysis by Erik Volz and we are backing this up on the same data with different models. [shared file(s): image.png]
-
Each dot is in an independent introduction into the UK.
-
And how's growth rate calculated?
-
But basically this is not significant under most conservative conditions.
-
This is using Erik's skygrowth model.
-
But we are now doing a full BEAST run with a hierarchical model on the growth rates which may give us more power.
-
This is well above my understanding, but isn't it possible to estimate R0 (or Re) pretty reliably given the amount ofdata you have from the UK? In Erik's model is it possible to adjust for introduction time differences? I assume lockdown measures would affect these differently.
-
Yes they have individual trajectories up and down but are not conditioned on a particular lockdown date. Our BEAST estimates have two exponential phases with different growth rates (which can be negative) before and after atransition time. They all share the same transition time which is after lockdown (but allows for some lag). For the initial growth all the Ds have a rate and all the Gs have a rate and we estimate them jointly to give us power.
-
Ah, nice - I like that model. Will be interesting to see the results from that - I really can't make up my mind of what Ifind is more likely here