Yesterday’s post focused on the plausibility of the ONS’ stated estimates of vaccine effectiveness against death during the period of April 2021- March 2022.
Mar 10, 2023·edited Mar 10, 2023Liked by T Coddington
If the ONS were trying to discredit their other report with this one, they could scarcely have done a better job. It's nice to see what the deaths data says without a lot of pointless adjustments, so thanks for that. However, I think there is a difference in the definition for "Covid death" in that the efficacy report merely requires mention of Covid-19 on death certificate while the deaths report requires primary cause. This may or may not influence results and the changes after March (i.e., maybe it would look even worse).
Using the deaths report (primary cause), unvaxxed seemingly aren't Covid-dying very much in 2022. At a certain point you would expect that the vaxxed should actually die more, simply because the control group is "all died out" while the vaxxed are still catching up with actually getting infected and either dying or gaining natural immunity. This is a known problem in long term vaccine efficacy appraisal https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7003633/
Maybe the ONS authors behind this new report wanted to "correct" for that problem. *edit: Actually, this problem might not have had any impact on their outputs had they used the months after March, 2022, anyway, because they are measuring time-to-event (HR), so a leveling off at the end shouldn't change anything - however, again, the results might be worse using the efficacy standard of "mentions" rather than "primary" cause of death. Really it just seems like a troll.
I just wanted to point out that starting January 2022 was actually the perfect time to ascertain effectivenss as the cohort groups had finally stabilised with the vaccination campaign winding down. (Exception being the over 75's 2nd booster campaign in April-May).
Some may argue there were much fewer cases in 2022 but the test positivity figures suggest otherwise. The argument that the new omicron variants BA.2 and BA.5 brought much fewer severe cases is probably a valid.
Just that it is generally a transitory status, and by even the beginning of this period, a pretty small part of the population. And I'm lazy. Will add just to show I'm not trying to hide anything (unlike the ONS). I don't know how much we can learn from it, but will be up in a minute.
I wasn't suggesting you're hiding anything, not at all. I'm just curious about the group, because I assume a large proportion of them suffered some type of side effect that made them avoid another injection.
We can learn that the first jab is almost as effective as the second for the 60 to 69s even though it isn't even claimed to be. In other words, it's spurious, just like dose 2. The first jab is apparently injurious for the over 80s even though it is progressively beneficial the more you dose. In other words, all we observe is survivorship bias (the weakest die first, leaving the most hardy in the triple jabbed).
The single-jabbed are indeed an interesting group. Whenever I am looking at some real-world German data they seem to be either ignored or overrepresented. According to RKI data, they should constitute around 0.8% of the 60+ population, but for Covid ICU occupancy ("with" or "because of") the proportion has been hovering between 3% and 4% all the time.
Oh wow. That's quite impressive. Especially when we consider that a lot of them will have died (from any cause) since their shot. Population size should be overestimated.
I asked the head of the ONS Sarah Caul why the data started in March and didn't continue until December 2022. She answered the first part of the question saying that is when they undertook the census but ignored the second part. I replied saying thanks but can you answer the second part of my question. She ignored this. I prompted her again today but again nothing. Her silence is deafening.
Most of the convergence of rates between unvaxxed and vaxxed is due to the rates of the former falling off a cliff. Hard to imagine the boosters change the picture much. And I'm not willing to give these folks the benefit of the doubt in regards to hiding unpleasant data.
It wouldn't matter if it made a change or not - they would be spiking hypothetical booster protection into the 3rd dose buckets, so you could call the result inaccurate or intentionally pro-vax deceptive
The census population estimates also seem horribly off as can be seen on that site. There are about 600k people more in NY mid-2021 compared to mid-2020.
All of this seems so fishy.
I wish there was a better place for these things to be discussed. I installed a forum on my website (forum.pervaers.com), but I just don't have the time to set it all up and promote it. Right now it seems we are all doing a lot of work individually and would benefit greatly from more cooperation.
Very nice looking site! I'm not exactly sure which numbers you are trying to reconcile. I've found with many US data sets that it's very easy to double count values because, for example, they might have a row for every state & then a row for "United States"... they might have a row for each month, then a row for the year, then a row for entire pandemic. Generally, I try to do any aggregation myself, working with the lowest level of detail available. Wonder if this is useful to you: https://public.tableau.com/views/COVIDDeathsbyAgeandUSState/COVIDDeathsbyStateAge?:language=en-US&:display_count=n&:origin=viz_share_link
If you open my “main” tableau dashboard for US (starred on my Tableau public site), there is a tab for Sources that shows what I used for all county level data.
It was just Puerto Rico that is not counted for US deaths. Most of the data is adding up fine now. :)
Yes I saw your tableau site and I was thinking I should ask you for the data sometime. In fact, seeing your site was what triggered me to install that forum.
Integrating new datasets is not on my agenda right now, but come to think of it, you could help me out immensely here.
Do you have county level data for
- vaccinations administered to residents (the resident part is important)
- population estimates for these counties
- COVID cases
If you have that neatly organized and could prepare a package of any form that includes these, that would be huge (JSON, fixed-length text, CSV would all be fine).
It could prove the deaths of 220k people in Q3-Q4'2021. Working on this for Children's Health Defense right now. I already have evidence based on state-level data, but I imagine county-level data would be even more impressive.
It's a bit of a mess, I'm obviously no web designer... However these are based on the public use death certificate files and give you a very good idea of what happened there in 2021 (alas only on national level).
If the ONS were trying to discredit their other report with this one, they could scarcely have done a better job. It's nice to see what the deaths data says without a lot of pointless adjustments, so thanks for that. However, I think there is a difference in the definition for "Covid death" in that the efficacy report merely requires mention of Covid-19 on death certificate while the deaths report requires primary cause. This may or may not influence results and the changes after March (i.e., maybe it would look even worse).
Using the deaths report (primary cause), unvaxxed seemingly aren't Covid-dying very much in 2022. At a certain point you would expect that the vaxxed should actually die more, simply because the control group is "all died out" while the vaxxed are still catching up with actually getting infected and either dying or gaining natural immunity. This is a known problem in long term vaccine efficacy appraisal https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7003633/
Maybe the ONS authors behind this new report wanted to "correct" for that problem. *edit: Actually, this problem might not have had any impact on their outputs had they used the months after March, 2022, anyway, because they are measuring time-to-event (HR), so a leveling off at the end shouldn't change anything - however, again, the results might be worse using the efficacy standard of "mentions" rather than "primary" cause of death. Really it just seems like a troll.
Aghh.. comment window just ate my comment..
I just wanted to point out that starting January 2022 was actually the perfect time to ascertain effectivenss as the cohort groups had finally stabilised with the vaccination campaign winding down. (Exception being the over 75's 2nd booster campaign in April-May).
Some may argue there were much fewer cases in 2022 but the test positivity figures suggest otherwise. The argument that the new omicron variants BA.2 and BA.5 brought much fewer severe cases is probably a valid.
Thank you for this :)
In response to NakedEmperor's comment:
Jabs also heal disabilities. Vaccination rates correlate with declines in disabilities.
Everyone needs to get jabbed right now if they haven't done so already.
(Especially if you're vulnerable!)
EDIT: When vaccination rates are higher in the 0-64 cohort than in the 65+ cohort the correlation flips.
Is there any reason you excluded the single-jabbed from the analysis?
Just that it is generally a transitory status, and by even the beginning of this period, a pretty small part of the population. And I'm lazy. Will add just to show I'm not trying to hide anything (unlike the ONS). I don't know how much we can learn from it, but will be up in a minute.
I wasn't suggesting you're hiding anything, not at all. I'm just curious about the group, because I assume a large proportion of them suffered some type of side effect that made them avoid another injection.
I didn't think you were... but someone might if they wanted to discredit me. Just trying to anticipate any objection and be transparent. All good 🙏
We can learn that the first jab is almost as effective as the second for the 60 to 69s even though it isn't even claimed to be. In other words, it's spurious, just like dose 2. The first jab is apparently injurious for the over 80s even though it is progressively beneficial the more you dose. In other words, all we observe is survivorship bias (the weakest die first, leaving the most hardy in the triple jabbed).
Thank you. That makes perfect sense.
The single-jabbed are indeed an interesting group. Whenever I am looking at some real-world German data they seem to be either ignored or overrepresented. According to RKI data, they should constitute around 0.8% of the 60+ population, but for Covid ICU occupancy ("with" or "because of") the proportion has been hovering between 3% and 4% all the time.
Oh wow. That's quite impressive. Especially when we consider that a lot of them will have died (from any cause) since their shot. Population size should be overestimated.
I asked the head of the ONS Sarah Caul why the data started in March and didn't continue until December 2022. She answered the first part of the question saying that is when they undertook the census but ignored the second part. I replied saying thanks but can you answer the second part of my question. She ignored this. I prompted her again today but again nothing. Her silence is deafening.
Maybe everyone can give her a prompt https://twitter.com/NakedEmperorUK/status/1634128291077533702?s=20
I called out the ONS as well... https://twitter.com/VanVoorheesVII/status/1634237402884521996?s=20
If you have not seen - https://twitter.com/SarahCaul_ONS/status/1635226615587442688
A reason like this really should have been included in the methods to begin with. But it makes sense if true
Most of the convergence of rates between unvaxxed and vaxxed is due to the rates of the former falling off a cliff. Hard to imagine the boosters change the picture much. And I'm not willing to give these folks the benefit of the doubt in regards to hiding unpleasant data.
It wouldn't matter if it made a change or not - they would be spiking hypothetical booster protection into the 3rd dose buckets, so you could call the result inaccurate or intentionally pro-vax deceptive
Huh. I was thinking about it wrong. Including the additional boosters would presumably only make 3rd dose (no booster) look worse.
Hey TC!
You are very familiar with the US data, right?
Can you explain to me why there are 75k+ deaths missing in the 85+ cohort on national level between 2015 and 2022?
When I add up all the states, this is what I get.
I have to use 3 datasets:
1.) Weekly_Counts_of_Deaths_by_Jurisdiction_and_Age (contains only age-stratified data, no totals)
2+3) Weekly_Provisional_Counts_of_Deaths_by_State_and_Select_Causes__2020-2023 + Weekly_Counts_of_Deaths_by_State_and_Select_Causes__2014-2019 (contains "All Cause" totals).
I made this yesterday btw: https://usa.pervaers.com (yet to fix some of those age groups)
The census population estimates also seem horribly off as can be seen on that site. There are about 600k people more in NY mid-2021 compared to mid-2020.
All of this seems so fishy.
I wish there was a better place for these things to be discussed. I installed a forum on my website (forum.pervaers.com), but I just don't have the time to set it all up and promote it. Right now it seems we are all doing a lot of work individually and would benefit greatly from more cooperation.
Very nice looking site! I'm not exactly sure which numbers you are trying to reconcile. I've found with many US data sets that it's very easy to double count values because, for example, they might have a row for every state & then a row for "United States"... they might have a row for each month, then a row for the year, then a row for entire pandemic. Generally, I try to do any aggregation myself, working with the lowest level of detail available. Wonder if this is useful to you: https://public.tableau.com/views/COVIDDeathsbyAgeandUSState/COVIDDeathsbyStateAge?:language=en-US&:display_count=n&:origin=viz_share_link
which used this as primary data source:
https://data.cdc.gov/NCHS/Provisional-COVID-19-Deaths-by-Sex-and-Age/9bhg-hcku
If you open my “main” tableau dashboard for US (starred on my Tableau public site), there is a tab for Sources that shows what I used for all county level data.
Yeah, but all of this is a bit overwhelming right now.
I have 3000 lines of code, tons of datasets, lots of things that need my attention. I have been meaning to write articles about a lot of it also...
I was just hoping you could fix something up in a few minutes that I can just use. Otherwise I'll keep using state level data for this.
I worked 20h yesterday (on the US data) and around 12-14h on average days, because there is just so much to do, so I have to prioritize a lot.
It was just Puerto Rico that is not counted for US deaths. Most of the data is adding up fine now. :)
Yes I saw your tableau site and I was thinking I should ask you for the data sometime. In fact, seeing your site was what triggered me to install that forum.
Integrating new datasets is not on my agenda right now, but come to think of it, you could help me out immensely here.
Do you have county level data for
- vaccinations administered to residents (the resident part is important)
- population estimates for these counties
- COVID cases
If you have that neatly organized and could prepare a package of any form that includes these, that would be huge (JSON, fixed-length text, CSV would all be fine).
It could prove the deaths of 220k people in Q3-Q4'2021. Working on this for Children's Health Defense right now. I already have evidence based on state-level data, but I imagine county-level data would be even more impressive.
To give you an idea what it is about:
https://substack.pervaers.com/USA_Misc/Midsummer%20Massacre.png
Oh and if you ever need an overview of causes of deaths based on death certificates, check this out:
https://cdcsucks.pervaers.com/
It's a bit of a mess, I'm obviously no web designer... However these are based on the public use death certificate files and give you a very good idea of what happened there in 2021 (alas only on national level).