I obtained the spreadsheets from a substack, 2nd Smartest Guy in the World. There was a link at the end called Proof v.7. This included an Excel spreadsheet with text boxes, one which said the data used is available from Steve Kirsch. Steve also alluded to the substack and has defended. As stated, I pointed out the problem, but have been…
I obtained the spreadsheets from a substack, 2nd Smartest Guy in the World. There was a link at the end called Proof v.7. This included an Excel spreadsheet with text boxes, one which said the data used is available from Steve Kirsch. Steve also alluded to the substack and has defended. As stated, I pointed out the problem, but have been ignored.
May I ask who wrote the Code on the Github at NZMOH which calculated Excess Death? Where was the data matrix for the program? IMHO, the best measure of excess death is based on linear ewgression, or 2nd order polynomial best fit of the previous nine years, 2011-2019, death rates, for each cohort for each week of the year. This creates a much lower expected deaths,especially in the older cohorts, where death rates have declined rapidly in that period. I did an analysis based on annual population deaths rates for 5-year buckets cohorts, using both trend line fits and came up with over 4000 excess deaths in 2022-2023. But it would have been higher with weekly increments. And importantly, NZ, like many countries, went onto hyper vigilance from 2020-2022, keeping out circulating influenza and isolating people from risky behaviors, meaning the underlying death rate was even lower than regression trends would predict.
Yeah the "Proof v.7" spreadsheet was made by Scoops McGoo and not Kirsch: https://docs.google.com/spreadsheets/d/1BwtabtrYjvSfAKlI3o_OTUlgPN_NrzCXTsPM_QT6gKI. It restores part of my hope in humanity that someone else found his method of adjusting for average age to be incorrect. I had a long argument with Scoops about it on Twitter.
His method of categorizing people under each dose was also flawed, because he only kept people under the first dose whose final dose included in the dataset was the first dose. So it meant that all people who died while the first dose was their newest dose were included under the first dose, but the vast majority of people who didn't die while the first dose was their newest dose were not included under the first dose because they later went on to get a second dose, so it artificially inflated the mortality rate of people under the first dose. But he should've instead counted the number of person-days that people spent under the first dose until they got the second dose.
I tried applying his dose categorization method to a simulation which contained a million people and which ran for a thousand days, where all people had a constant daily likelihood of death of 1/100/365, and all people had a 1/300 likelihood of getting a new vaccine dose each day. During the first 100 days of the simulation, people whose final dose was the first dose had about 30 times higher mortality rate than the total mortality rate of all people in the simulation: sars2.net/moar.html#Simulation_which_shows_that_the_dose_categorization_method_used_by_Scoops_is_biased. So it explains why Scoops also got over 1000% excess mortality for people under the first dose in the early part of his plot.
I didn't find any code for calculating excess deaths on the GitHub account of the NZ MoH. But when I calculated a 2010-2019 linear trend in CMR for each single year of age, I projected the trend to 2020-2023, and I multiplied the projected trend for each age with the population size of the age to get the expected deaths, I got about 3,371 excess deaths in 2022-2023. So it's similar to your figure:
Brilliant. I never picked up the dose categorization error. One error was enough for me. The reason I asked about code for the github program for working out excess deaths, is because if you open the code you see a variable called 'excess deaths' . I'm not sure if that's a misleading name by the MOH, if they even wrote the code, because Stats NZ never agreed on a way to calculate excess deaths. The government Muppets just referred to OWID, which was flawed.
Good job on calculating excess deaths 2022-2023. If I'd admit exactly the same method, but I used 5-year buckets, not single years because STATS NZ report those figures for population and death rates. I applied my trend line to 2011-2019 data only, because I've been using aggregated weekly deaths for that period to compare with current deaths and that's all that was available until a few months ago, when suddenly weekly deaths back to 2000 appeared. I'll share with you some of my extrapolated death rates for the 2020-2923 period so we can compare and see why I got 4000+ excess deaths. I use Libre Office. I don't know R language.
I obtained the spreadsheets from a substack, 2nd Smartest Guy in the World. There was a link at the end called Proof v.7. This included an Excel spreadsheet with text boxes, one which said the data used is available from Steve Kirsch. Steve also alluded to the substack and has defended. As stated, I pointed out the problem, but have been ignored.
May I ask who wrote the Code on the Github at NZMOH which calculated Excess Death? Where was the data matrix for the program? IMHO, the best measure of excess death is based on linear ewgression, or 2nd order polynomial best fit of the previous nine years, 2011-2019, death rates, for each cohort for each week of the year. This creates a much lower expected deaths,especially in the older cohorts, where death rates have declined rapidly in that period. I did an analysis based on annual population deaths rates for 5-year buckets cohorts, using both trend line fits and came up with over 4000 excess deaths in 2022-2023. But it would have been higher with weekly increments. And importantly, NZ, like many countries, went onto hyper vigilance from 2020-2022, keeping out circulating influenza and isolating people from risky behaviors, meaning the underlying death rate was even lower than regression trends would predict.
Yeah the "Proof v.7" spreadsheet was made by Scoops McGoo and not Kirsch: https://docs.google.com/spreadsheets/d/1BwtabtrYjvSfAKlI3o_OTUlgPN_NrzCXTsPM_QT6gKI. It restores part of my hope in humanity that someone else found his method of adjusting for average age to be incorrect. I had a long argument with Scoops about it on Twitter.
His method of categorizing people under each dose was also flawed, because he only kept people under the first dose whose final dose included in the dataset was the first dose. So it meant that all people who died while the first dose was their newest dose were included under the first dose, but the vast majority of people who didn't die while the first dose was their newest dose were not included under the first dose because they later went on to get a second dose, so it artificially inflated the mortality rate of people under the first dose. But he should've instead counted the number of person-days that people spent under the first dose until they got the second dose.
I tried applying his dose categorization method to a simulation which contained a million people and which ran for a thousand days, where all people had a constant daily likelihood of death of 1/100/365, and all people had a 1/300 likelihood of getting a new vaccine dose each day. During the first 100 days of the simulation, people whose final dose was the first dose had about 30 times higher mortality rate than the total mortality rate of all people in the simulation: sars2.net/moar.html#Simulation_which_shows_that_the_dose_categorization_method_used_by_Scoops_is_biased. So it explains why Scoops also got over 1000% excess mortality for people under the first dose in the early part of his plot.
I didn't find any code for calculating excess deaths on the GitHub account of the NZ MoH. But when I calculated a 2010-2019 linear trend in CMR for each single year of age, I projected the trend to 2020-2023, and I multiplied the projected trend for each age with the population size of the age to get the expected deaths, I got about 3,371 excess deaths in 2022-2023. So it's similar to your figure:
> t=fread("http://sars2.net/f/nzpopdead.csv")
> a=t[,.(dead=sum(dead),pop=sum(pop,na.rm=T)),.(year,age=pmin(age,95))]
> base=a[year%in%2010:2019,.(year=unique(t$year),base=predict(lm(dead/pop~year),.(year=unique(t$year)))),age]
> a=merge(base,a)[,base:=base*pop]
> a[year>2021,sum(dead)-sum(base)]
[1] 3370.714
Brilliant. I never picked up the dose categorization error. One error was enough for me. The reason I asked about code for the github program for working out excess deaths, is because if you open the code you see a variable called 'excess deaths' . I'm not sure if that's a misleading name by the MOH, if they even wrote the code, because Stats NZ never agreed on a way to calculate excess deaths. The government Muppets just referred to OWID, which was flawed.
Good job on calculating excess deaths 2022-2023. If I'd admit exactly the same method, but I used 5-year buckets, not single years because STATS NZ report those figures for population and death rates. I applied my trend line to 2011-2019 data only, because I've been using aggregated weekly deaths for that period to compare with current deaths and that's all that was available until a few months ago, when suddenly weekly deaths back to 2000 appeared. I'll share with you some of my extrapolated death rates for the 2020-2923 period so we can compare and see why I got 4000+ excess deaths. I use Libre Office. I don't know R language.