I take this approach because I'm not all that great with stats and data - I can "see" things much clearer with algebra.
My approach wasn't really demonstrating anything new - but trying to see the same things others had commented on but in a different (and simple) way.
Anyhoo - the point of the piece was really to highlight an effect that I'd noted : within my simple vaccination schedule model the end result for VE depended on the *percentage* vaxxed if the VE was calculated with a delay that effectively shunted the recently vaccinated deaths into the unvaxxed category.
You altered the percentage vaxxed here and found you could manipulate the efficacy. In other words you're seeing a dependence of the VE on the percentage vaxxed.
The efficacy of a vaccine should, of course, not depend on whether 10% or 20% or 80% of the population have been vaccinated. My question would be whether being able to demonstrate a dependence on vax percentage is sufficient to show "datacrime" with regards to VE?
Of course, you're adding a whole new element with the numerator-denominator but this is the revelation that got my video deleted from YT. It was called How to Lie with Statistics. Someone had been arguing with me about why his Wash State vaccine data and his brother's from Arizona both showed the unvaxxed were dying faster. I went to bed on it and sat bolt upright in the middle of the night and went to the computer. Eureka! They were both clearly from the same source with just a little formatting difference. And both said that those not fully vaxxed--2 weeks out--were counted as unvaxxed. A first year stats student couldn't get away with that. But this takes it to a whole new level, Mathew. Just to post my me-too from Rumble: https://rumble.com/vucisq-how-to-lie-with-statistics-on-dr.-john-abramson-and-russell-brand.html .
It would've been better off for humanity if the psychopaths had filled all the syringes with a saline solution rather than the experimental mRNA gene therapy concoction, but then the depraved sadists couldn't derive pleasure from torturing and slaughtering millions for profit.
I made this comment under Norman’s piece just now, but relevant here also.
“Of course, for the actual covid injections as opposed to placebo, the illusion gets an additional “booster” (sorry but couldn’t resist) by virtue of the increased propensity of the injected to become infected during the period when they then become classed as unvaccinated.
And there’s a double booster if that infection then reduces the propensity to become infected later - once classed as vaccinated.”
I'm no statistician, but back in 2021, the Province of Alberta, Canada posted on its official C19 stats website (or a brief time), a colourful 'heat' chart showing the number of C 19 cases diagnosed in the province as function of time since vaccination. The overwhelming majority (like 90%) occurred within 14-21 days days post vax. Uh oh. They proceeded to remove that chart from the website post haste and from then on, simply reclassified people having received their vaccines, as 'unvaccinated' for the first 14 days after vaccination- problem solved! Of course, I kept the screen shots of these charts for posterity and really hope that someday they re -emerge in a court of law.
This is really great Mathew. Although Norman correctly uses the ONS' stated definition of an unvaccinated case as one that occurs within 21 days of a jab, it is unclear what the CDC is doing. In this CDC statement, it doesn't seem like they are throwing those cases into the unvaccinated bucket. They are calling them "partially vaccinated" and are presumably excluded from the IR from both groups:
"Partially vaccinated case: SARS-CoV-2 RNA or antigen detected in a respiratory specimen collected from a person who received at least one FDA-authorized or approved vaccine dose but did not complete a primary series ≥14 days before collection of a respiratory specimen with SARS-CoV-2 RNA or antigen detected.
Unvaccinated case: SARS-CoV-2 RNA or antigen detected in a respiratory specimen from a person who has not been verified to have received any COVID-19 vaccine doses before the specimen collection date."
Even though the numerator in the unvaxxed IR won't be as exaggerated as in the UK, dropping 14 (not 21) days of infections from the vaxxed numerator still generates a fabricated efficacy--as long as they are including the partially vaccinated (i.e. Recently Treated) in the denominator. I am not sure if they are doing that.
1) It is important to clarify that the phenomenon of each groups' population sizes changing over time is only applicable to population level data. That aspect of this bias would not occur in a cohort study which would have both group sizes be fixed. I see some people citing Fenton's research erroneously in that regard. However other aspects of these biases would still be functioning. And also, in the event of acute negative efficacy, vaccine-caused covid cases would still be censored from the vaccine group (albeit not shifted into the unvaccinated group, as would happen in population level analysis).
[Edit: Unless they are doing some kind of person-years analysis, in which case the full extent of bias would be present]
2) One aspect I recently realized is that not only do the vaccinated have ever-so-slightly higher natural immunity due to the requirement to be healthy before vaccination, but for a second reason as well. In the first two weeks after vaccination, you have a chance to develop natural immunity, without getting a counted case of covid! The unvaccinated are not afforded this opportunity. This issue becomes worse if vaccines have an acute negative efficacy. In population level data, this would in essence not just cause covid to be assigned to the unvaccinated, but also cause the natural immunity be kept for the vaccinated! In a fixed-population-size cohort study, this would cause censoring of covid cases in the vaccinated, but would not "censor" the natural immunity they gained. But as Fenton stated, if infection rate is low, these biases relating to natural immunity may not matter a ton.
3) If you are going to develop this model further, some suggestions, if you are not already planning:
a) Do it for a fixed cohort study. The bias will be less, but since those are the studies people really cite which show the cyclical pattern of waning and boosting with new vaccines, it is the context we really need to understand.
b) Have a parameter for true vaccine efficacy that we can play with, not just a placebo.
c) Include the natural immunity bias I described. Have a parameter for natural immunity efficacy. We can dig up the best estimates for that later.
d) Also have a similar parameter for negative efficacy in the first two weeks that we can play with. That could amp things up.
d) Let's dig up some best estimates for covid infection rates during a specific rollout and use those real values. Of course, true prevalence is not the same as case count, but our goal is to do what they do.
e) Let's dig up estimates for real values for vaccination rates during that rollout and use them.
f) If you want to complicate things, model two doses. Each dose has a potentially negative VE period, and a potentially positive VE period.
g) Not all cohort studies use the 2-week misclassification. Most do, but some classify correctly and still report a benefit. Some of them do report negative acute efficacy, and some do not. Consider parameterizing the misclassification length as well.
With all that, we could do a best case, worst case, and middle case examples. This will show the huge uncertainty around observational studies. I
I made some of these comments on Fenton's article as well. I doubt there is a risk of you guys duplicating work though.
[Edit: It may be possible to avoid modeling 2 doses and also get some higher background infection rates if you modeled just the 3rd dose, which perhaps did not have a huge overlap with people getting first and second doses]
A while back I tried to figure out some possible sources of error in VE by taking an algebraic approach. You commented on the article back then.
https://rudolphrigger.substack.com/p/a-fascinating-result
I take this approach because I'm not all that great with stats and data - I can "see" things much clearer with algebra.
My approach wasn't really demonstrating anything new - but trying to see the same things others had commented on but in a different (and simple) way.
Anyhoo - the point of the piece was really to highlight an effect that I'd noted : within my simple vaccination schedule model the end result for VE depended on the *percentage* vaxxed if the VE was calculated with a delay that effectively shunted the recently vaccinated deaths into the unvaxxed category.
You altered the percentage vaxxed here and found you could manipulate the efficacy. In other words you're seeing a dependence of the VE on the percentage vaxxed.
The efficacy of a vaccine should, of course, not depend on whether 10% or 20% or 80% of the population have been vaccinated. My question would be whether being able to demonstrate a dependence on vax percentage is sufficient to show "datacrime" with regards to VE?
Does your book additionally also cover the - I think other, different from this - sleight of hand that Gato Malo called the "Bayesian Data crime"?
https://boriquagato.substack.com/p/bayesian-datacrime-defining-vaccine
Probably that is already covered in earlier parts of your article series.
Of course, you're adding a whole new element with the numerator-denominator but this is the revelation that got my video deleted from YT. It was called How to Lie with Statistics. Someone had been arguing with me about why his Wash State vaccine data and his brother's from Arizona both showed the unvaxxed were dying faster. I went to bed on it and sat bolt upright in the middle of the night and went to the computer. Eureka! They were both clearly from the same source with just a little formatting difference. And both said that those not fully vaxxed--2 weeks out--were counted as unvaxxed. A first year stats student couldn't get away with that. But this takes it to a whole new level, Mathew. Just to post my me-too from Rumble: https://rumble.com/vucisq-how-to-lie-with-statistics-on-dr.-john-abramson-and-russell-brand.html .
This post here also talks about magic, but in November 2020:
https://drmalcolmkendrick.org/2020/11/10/ninety-per-cent/
I will never understand this. Long before there is "data" some people "know" what is going to happen.
The deductionist sect.
People see things, abstract a pattern, and they learn.
Much later, when they see the first signs of the pattern emerging, they warn the other monkeys: don't fall for it.
The other monkeys get all indignant: How do you know? You don't have data! You are a denialist!
It's as if knowledge was more than science.
For we numerically challenged these breakdowns are priceless thank you.
It would've been better off for humanity if the psychopaths had filled all the syringes with a saline solution rather than the experimental mRNA gene therapy concoction, but then the depraved sadists couldn't derive pleasure from torturing and slaughtering millions for profit.
I figured it out via pure intuition. But then I had no proof
Nice. If it wasn't so tragic it would be hilarious. Even still, it is kind of hilarious.
I made this comment under Norman’s piece just now, but relevant here also.
“Of course, for the actual covid injections as opposed to placebo, the illusion gets an additional “booster” (sorry but couldn’t resist) by virtue of the increased propensity of the injected to become infected during the period when they then become classed as unvaccinated.
And there’s a double booster if that infection then reduces the propensity to become infected later - once classed as vaccinated.”
https://wherearethenumbers.substack.com/p/the-illusion-of-vaccine-efficacy/comment/15698818
Excellent work. Truly excellent.
Mathew, you’re cheeky.
Nice easy summary to understand, many wished that they got the saline
I'm no statistician, but back in 2021, the Province of Alberta, Canada posted on its official C19 stats website (or a brief time), a colourful 'heat' chart showing the number of C 19 cases diagnosed in the province as function of time since vaccination. The overwhelming majority (like 90%) occurred within 14-21 days days post vax. Uh oh. They proceeded to remove that chart from the website post haste and from then on, simply reclassified people having received their vaccines, as 'unvaccinated' for the first 14 days after vaccination- problem solved! Of course, I kept the screen shots of these charts for posterity and really hope that someday they re -emerge in a court of law.
Kudos! It's like a primer on how to lie with statistical brilliance! Thank you. God bless you. Amen.
This is really great Mathew. Although Norman correctly uses the ONS' stated definition of an unvaccinated case as one that occurs within 21 days of a jab, it is unclear what the CDC is doing. In this CDC statement, it doesn't seem like they are throwing those cases into the unvaccinated bucket. They are calling them "partially vaccinated" and are presumably excluded from the IR from both groups:
"Partially vaccinated case: SARS-CoV-2 RNA or antigen detected in a respiratory specimen collected from a person who received at least one FDA-authorized or approved vaccine dose but did not complete a primary series ≥14 days before collection of a respiratory specimen with SARS-CoV-2 RNA or antigen detected.
Unvaccinated case: SARS-CoV-2 RNA or antigen detected in a respiratory specimen from a person who has not been verified to have received any COVID-19 vaccine doses before the specimen collection date."
https://www.cdc.gov/coronavirus/2019-ncov/php/hd-breakthrough.html
Even though the numerator in the unvaxxed IR won't be as exaggerated as in the UK, dropping 14 (not 21) days of infections from the vaxxed numerator still generates a fabricated efficacy--as long as they are including the partially vaccinated (i.e. Recently Treated) in the denominator. I am not sure if they are doing that.
1) It is important to clarify that the phenomenon of each groups' population sizes changing over time is only applicable to population level data. That aspect of this bias would not occur in a cohort study which would have both group sizes be fixed. I see some people citing Fenton's research erroneously in that regard. However other aspects of these biases would still be functioning. And also, in the event of acute negative efficacy, vaccine-caused covid cases would still be censored from the vaccine group (albeit not shifted into the unvaccinated group, as would happen in population level analysis).
[Edit: Unless they are doing some kind of person-years analysis, in which case the full extent of bias would be present]
2) One aspect I recently realized is that not only do the vaccinated have ever-so-slightly higher natural immunity due to the requirement to be healthy before vaccination, but for a second reason as well. In the first two weeks after vaccination, you have a chance to develop natural immunity, without getting a counted case of covid! The unvaccinated are not afforded this opportunity. This issue becomes worse if vaccines have an acute negative efficacy. In population level data, this would in essence not just cause covid to be assigned to the unvaccinated, but also cause the natural immunity be kept for the vaccinated! In a fixed-population-size cohort study, this would cause censoring of covid cases in the vaccinated, but would not "censor" the natural immunity they gained. But as Fenton stated, if infection rate is low, these biases relating to natural immunity may not matter a ton.
3) If you are going to develop this model further, some suggestions, if you are not already planning:
a) Do it for a fixed cohort study. The bias will be less, but since those are the studies people really cite which show the cyclical pattern of waning and boosting with new vaccines, it is the context we really need to understand.
b) Have a parameter for true vaccine efficacy that we can play with, not just a placebo.
c) Include the natural immunity bias I described. Have a parameter for natural immunity efficacy. We can dig up the best estimates for that later.
d) Also have a similar parameter for negative efficacy in the first two weeks that we can play with. That could amp things up.
d) Let's dig up some best estimates for covid infection rates during a specific rollout and use those real values. Of course, true prevalence is not the same as case count, but our goal is to do what they do.
e) Let's dig up estimates for real values for vaccination rates during that rollout and use them.
f) If you want to complicate things, model two doses. Each dose has a potentially negative VE period, and a potentially positive VE period.
g) Not all cohort studies use the 2-week misclassification. Most do, but some classify correctly and still report a benefit. Some of them do report negative acute efficacy, and some do not. Consider parameterizing the misclassification length as well.
With all that, we could do a best case, worst case, and middle case examples. This will show the huge uncertainty around observational studies. I
I made some of these comments on Fenton's article as well. I doubt there is a risk of you guys duplicating work though.
[Edit: It may be possible to avoid modeling 2 doses and also get some higher background infection rates if you modeled just the 3rd dose, which perhaps did not have a huge overlap with people getting first and second doses]