By Tomas Fürst at Brownstone dot org.

The 21st century is said to be the century of data. So, if Alice claims that Covid vaccines saved millions of lives but Bob says they killed millions, it should be really easy to decide who is right. Just get the data, right?

We obtained the data for the Czech Republic. I still can't believe we have it, but it is here. It is the official data obtained from a government agency through a FOIA request, and it is available for everyone to download and analyze. The data contains over 11 million rows - a single row for each Czech resident who was alive on January 1, 2020, or was born between January 1, 2020, and December 31, 2022.

For each individual, the data row contains the year of birth, sex, exact date of death from any cause (if the individual died within the three studied years) and exact dates, types, and even batch numbers of all Covid vaccines given to that individual. The cause of death is, unfortunately, not provided. To the best of our knowledge, this is the only officially released dataset that links all-cause mortality to Covid vaccination status at the level of individuals on the scale of a whole country.

Before we return to Alice and Bob, I need to tell you a bit about the Czech Republic. Everything is much more homogeneous here than an American can imagine: There are no significant ethnic minorities here. We have universal, free, and very regulated health care, so pretty much everyone gets the same care (allowing for some corruption here and there).

From the communist times, we inherited the system of compulsory "personal citizen numbers" (state-provided IDs), so everyone is very well accounted for: It is impossible to be born, or die, without the state noticing immediately.

Consequently, the Czech official data is almost exactly correct (unlike e.g. the British Office of National Statistics, which somehow manages to lose a couple of million unvaccinated Brits). In other words, this Czech dataset is so precise, clean, homogeneous, and detailed that nothing comparable will ever be available in the US. So, if answers can be found in this type of data, they will be especially apparent and irrefutable in the Czech data.

It is not completely straightforward to compute all-cause mortality (ACM) in a particular age cohort according to vaccination status. One would be tempted to count the number of deaths in that cohort and divide it by the cohort size at a particular time. But this would be incorrect because people keep shifting among the vaccination cohorts, so that their sizes keep changing.

For example, consider Auntie Betty, who entered the study on January 1, 2020, as unvaccinated. She got her first dose on March 13, 2021, went on to get the second dose on April 13, 2021, and died 25 days later. Thus, she contributed 437 person-days to the unvaccinated cohort, 31 person-days to "dose 1 only" cohort, 25 person-days to "dose 1 and 2" cohort, and one death to "dose 1 and 2" cohort. This type of breakdown must be done for every age cohort and every individual.

Only then can the number of deaths in each vaccination cohort (further stratified by age) be divided by the number of person-days spent by individuals in that cohort to get the correct value of ACM.

Further technical details are written in the original paper, but we basically carried out the procedure explained above to compute the monthly ACM rates stratified by vaccination status, sex, and age. The ACM was then compared to the expected mortality based on pre-pandemic data.

The expected mortality also needs to be computed carefully. One might be tempted to simply compare the computed ACM to pre-pandemic mortality rates (I am afraid that most authors do just this). However, this would be wrong again. Many people died during the pandemic (for various reasons) and since they are not going to die again, mortality must be expected to decrease after the pandemic.

Thus, from the pre-pandemic data, we estimated the probability of d...