Well, I was wrong. It’s no biggie, because it’s something I’m used to, but a big shout out to Witzbold and cm27874 (who also writes a substack) for challenging me and getting me to think again.
Once again, the issue was the following. The latest UKHSA data showed that with 70% of the population vaxxed we had 90% of covid deaths occurring in the vaxxed population giving a vaccine efficacy of - 285%
The thing is when we lump everyone together into a homogenous block like this (identity politics anyone?) we open ourselves to the possibility that all sorts of biases and confounders might lurk beneath that would invalidate any general conclusion we might come to if we only looked at the overall picture and didn’t look at things in finer detail.
My intuition was that there would be no way to split this data to ‘remove’ a particular cohort and be left with a good picture for the vax. I was completely wrong on that and I thought the constraint of having to maintain the overall proportions would scupper any attempt to do so.
Although it’s a bit frustrating to bugger things up, it’s also really good too. I am a bit strange, but I really love it when my intuition is Freddy Kruegered. When you see something counter-intuitive it just begs more investigation - and searching for more truth, more insight, has to be a good thing, right? (unless you’re an authoritarian tit and believe that only governments and official bodies speak the truth, and only the truth, or only they should be allowed to speak on certain matters).
I still haven’t fully understood this. The numbers all work out, but I can’t give you a simple way of seeing what’s going on at the moment. A counter-example to my hypothesis that this split couldn’t be done with positive efficacies in both groups was provided by cm27874 (is there any chance I could just call you “Charlie” instead of having to write out a string of numbers?). Charlie’s counter-example was quite extreme - but that’s fine, it was a proof of concept.
The question is what are the limits here? Are the counter-examples we can generate realistic?
Here’s one example (inspired by Charlie) that I came up with.
Although we have only one group (group B) with positive efficacy, group B is 98% of the population here. The efficacy is only 64% but it means that for 98% of the population (with these stats) we’d have the vax giving a reasonable benefit. Of course, we’re ignoring any potential harms of the vax itself here.
What would group A represent here? It would be a group of people for whom the vax is particularly dangerous (not a side effect as such - the vax makes them much more susceptible to a covid death) and covid itself about 32 times more dangerous than in group B in the unvaxxed.
Is this realistic? Is there such a group? Just because we can find a theoretical example that makes the vax look good it does not mean that this is what we actually have in the real world.
At the moment, even though it’s theoretically possible to adjust away the turd of the UKHSA figures, I’m not convinced this is what we actually have - but, as we’ve seen, my intuition is a bit wonky these days.
One other thing I noted whilst playing around with various numbers was that it’s quite a sensitive procedure. You don’t need to change things all that much to get a wildly different picture. This means that the stats we’re working with themselves have to be particularly accurate.
The real question is the range of parameters for which this Goo-beneficial split is possible. I struggled to find a counter-example initially and it was only Charlie pointing me in the right direction of where to look that helped me search in the right sort of place. There are obviously some conditions where this beneficial split is possible - I just haven’t figured those out yet.
In a couple of days I’m heading out to visit family for a week so I won’t be writing on here for a while - but, if I get time, I’ll work some more on this when I’m away and hopefully I’ll have some better answers when I return.
Let’s hope my next attempt will be more successful than my attempts to eat more healthily
The string of numbers is very important; it reminds the world of my birthday. I don't care much about anonymity; just google cm27874, and ignore the cat calming disks (I will have to write about these at some point).
Oh, and the issue with the salad is just colour perception. Fixed: https://ibb.co/xf6y37V
Keep on the good work! It often makes my day.
Well, I can't see anything wrong with your healthy salad.
As for the statistics, these days I tend to just glance over them since it's only a question of 'bad" and "worse", that manufactured infliction mercifully has 99.9% survival rate. Any remedy that can not show even 85% effectivnes in preventing/spreading is no brainer to me and, don't get me started on safety. Appreciate all your hard work R and, keep yourself properly nourished won't you?!