OK, so I finally got time to read this post aside from making snarky Panera comment. So what it appears is that there's far too many confounding variables to really assess the data properly, but unfortunately many people may be deriving far too many conclusions than may be warranted given the information?
I absolutely agree with your statement Brian about "our side". I've become quite dismayed that many people are going down their own rabbit hole of hysteria and unfounded assumptions. I think the Dr. Ardis fiasco just shone a light on everything going on and really makes us look ridiculous to those in the mainstream press.
Unfortunately, this has caused me to react reflexively to posts coming from "our side", and I've become rather quick to disagree with posts that have come out. I think this quick reaction came through in my LNP article yesterday, which after a little bit of stewing I believe jumped the shark on a few topics. I am hoping to either release a correction either today or tomorrow and expand on my ideas.
I enjoy your writing style; it's why I'm a paying subscriber.
I disagree that you should write more in a way that the "layman" understands. People who are genuinely interested in data and facts and also understand that "our side" is being played a-plenty by those who purport to be on "our side" are able to discern who is doing the work and who is riding a moment of glory.
One measure I use is to see whose writing has the most "likes" and "comments". If you're someone with hundreds of commenters that means you are not doing the deepest thinking because if you were doing the deepest thinking you wouldn't have so many people subscribed and/or commenting.
May 9, 2022·edited May 9, 2022Liked by Brian Mowrey
It's pretty clear that there are a lot of confounders in this data, and we are limited in what we can do with it. I'm sure ONS could do a much better job, if only the results were to their liking.
However, I had an idea of a way to eliminate a lot of the confounders.
The idea is to consider only 'ever vaxed' and 'never vaxed' as the denominators, and then for each age group plot the time series of deaths for all the various categories of vaxed.
That is, for say 40-50y age group, plot the monthly data by vax status category, using the corresponding month's estimate of the quantity of at least one dose vs. no doses. The various vax category deaths as a stacked line chart, with the denominator of population with at least one dose, compared against the unvaxxed deaths, with the denominator of population never vaxxed.
This captures several things of interest:
1. It eliminates all the worries about why people didn't get the next shot. After the first shot, they are in the vaxxed camp. This way it doesn't matter if they were about to die for whatever reason. That issue remains for those first timers, but really those first timers are pretty tightly clustered in the UK data. The bulk of the vaccinations occurred in the initial rollout - I'm not sure what drove people to get a shot much later than the big rollout, but I'm guessing the decision doesn't have much to do with their health - more likely for administrative reasons, especially in the younger, more healthy groups.
2. It lets us see correlations between booster rollouts and further death, without confounders
3. It captures the cumulative risks in all the shots.
It doesn't address possible biases in the population that meets the various ONS criteria. Perhaps a lot of the vax holdouts are also not in this database. However, by including younger cohorts there should be larger groups of healthy holdouts, which should improve the quality of the data also.
Unfortunately it seems to be a bit of a pain to find the monthy vax numbers broken down by age group (unless you know of a source.. all I know of is the line charts in the UKHSA docs).. so before I embarked on this I thought I would ask if you thought this would be worthwhile...
EDIT: It looks like they provide this info in person-years.
The biggest problem is the data is unreliable. We don't really know the real extent of the damage. We know enough to know there is a problem, but no amount of careful analysis of unreliable data can produce reliable results. Obfuscating data is often just incompetence, but sometimes nefarious. Either way, we really need to fix it. That will require a change in governments. It begins in the US in January. If people get the right information, most will make the right decisions. We're in an information war, and the tide is turning.
May 9, 2022·edited May 9, 2022Liked by Brian Mowrey
Brian,
Just a feedback:
Most readers of Substacks are laypeople. And accordingly they gloss, if at all, over any detailed recitation and explanation of data minutia. I do. They just need a few well-written paragraphs and some data in simple tables or charts. Writers can do the detailed analysis "below the line" or as a footnote.
In short, most readers just want, or can only handle, executive summaries.
Ancient BS Statistics & Computer Science here. The official data is a) wrong, incomplete, confounded, inaccurate, and possibly fudged; b) not intended for analysis, only for headlines.
I find the insurance company exec statements (also cumulative funeral home anecdotes) to be the most interesting, followed by the Walgreens data. Neither dataset claims to be complete in any way, but is likely the most accurate - within their context.
May 8, 2022·edited May 8, 2022Liked by Brian Mowrey
Oh lord, at first I had no idea what you were talking about! I was about to search the internet to see if Panera is a country in Europe that I hadn't heard of! 😅😅😅
Okay, the timers are the shots, and timing of shots may be/is statistically significant. Once I got it, the Panera metaphor is actually a creative way to communicate the different statistical difficulties you are pointing out.
Thanks for this - very helpful, and I appreciated the terminology, which eligicted repeated wry smiles: 'UnTimered' and 'Just-First-Timered'... thanks again.
OK, so I finally got time to read this post aside from making snarky Panera comment. So what it appears is that there's far too many confounding variables to really assess the data properly, but unfortunately many people may be deriving far too many conclusions than may be warranted given the information?
I absolutely agree with your statement Brian about "our side". I've become quite dismayed that many people are going down their own rabbit hole of hysteria and unfounded assumptions. I think the Dr. Ardis fiasco just shone a light on everything going on and really makes us look ridiculous to those in the mainstream press.
Unfortunately, this has caused me to react reflexively to posts coming from "our side", and I've become rather quick to disagree with posts that have come out. I think this quick reaction came through in my LNP article yesterday, which after a little bit of stewing I believe jumped the shark on a few topics. I am hoping to either release a correction either today or tomorrow and expand on my ideas.
I enjoy your writing style; it's why I'm a paying subscriber.
I disagree that you should write more in a way that the "layman" understands. People who are genuinely interested in data and facts and also understand that "our side" is being played a-plenty by those who purport to be on "our side" are able to discern who is doing the work and who is riding a moment of glory.
One measure I use is to see whose writing has the most "likes" and "comments". If you're someone with hundreds of commenters that means you are not doing the deepest thinking because if you were doing the deepest thinking you wouldn't have so many people subscribed and/or commenting.
I've never been to a Panera.
It's pretty clear that there are a lot of confounders in this data, and we are limited in what we can do with it. I'm sure ONS could do a much better job, if only the results were to their liking.
However, I had an idea of a way to eliminate a lot of the confounders.
The idea is to consider only 'ever vaxed' and 'never vaxed' as the denominators, and then for each age group plot the time series of deaths for all the various categories of vaxed.
That is, for say 40-50y age group, plot the monthly data by vax status category, using the corresponding month's estimate of the quantity of at least one dose vs. no doses. The various vax category deaths as a stacked line chart, with the denominator of population with at least one dose, compared against the unvaxxed deaths, with the denominator of population never vaxxed.
This captures several things of interest:
1. It eliminates all the worries about why people didn't get the next shot. After the first shot, they are in the vaxxed camp. This way it doesn't matter if they were about to die for whatever reason. That issue remains for those first timers, but really those first timers are pretty tightly clustered in the UK data. The bulk of the vaccinations occurred in the initial rollout - I'm not sure what drove people to get a shot much later than the big rollout, but I'm guessing the decision doesn't have much to do with their health - more likely for administrative reasons, especially in the younger, more healthy groups.
2. It lets us see correlations between booster rollouts and further death, without confounders
3. It captures the cumulative risks in all the shots.
It doesn't address possible biases in the population that meets the various ONS criteria. Perhaps a lot of the vax holdouts are also not in this database. However, by including younger cohorts there should be larger groups of healthy holdouts, which should improve the quality of the data also.
Unfortunately it seems to be a bit of a pain to find the monthy vax numbers broken down by age group (unless you know of a source.. all I know of is the line charts in the UKHSA docs).. so before I embarked on this I thought I would ask if you thought this would be worthwhile...
EDIT: It looks like they provide this info in person-years.
Not going to lie, seeing everyone post about the "Panera problem" I thought we were going to find out another soup was not made inhouse!
The biggest problem is the data is unreliable. We don't really know the real extent of the damage. We know enough to know there is a problem, but no amount of careful analysis of unreliable data can produce reliable results. Obfuscating data is often just incompetence, but sometimes nefarious. Either way, we really need to fix it. That will require a change in governments. It begins in the US in January. If people get the right information, most will make the right decisions. We're in an information war, and the tide is turning.
Brian,
Just a feedback:
Most readers of Substacks are laypeople. And accordingly they gloss, if at all, over any detailed recitation and explanation of data minutia. I do. They just need a few well-written paragraphs and some data in simple tables or charts. Writers can do the detailed analysis "below the line" or as a footnote.
In short, most readers just want, or can only handle, executive summaries.
I did a quick scan of this article.
Ancient BS Statistics & Computer Science here. The official data is a) wrong, incomplete, confounded, inaccurate, and possibly fudged; b) not intended for analysis, only for headlines.
I find the insurance company exec statements (also cumulative funeral home anecdotes) to be the most interesting, followed by the Walgreens data. Neither dataset claims to be complete in any way, but is likely the most accurate - within their context.
Oh lord, at first I had no idea what you were talking about! I was about to search the internet to see if Panera is a country in Europe that I hadn't heard of! 😅😅😅
Okay, the timers are the shots, and timing of shots may be/is statistically significant. Once I got it, the Panera metaphor is actually a creative way to communicate the different statistical difficulties you are pointing out.
Thanks for this - very helpful, and I appreciated the terminology, which eligicted repeated wry smiles: 'UnTimered' and 'Just-First-Timered'... thanks again.