Thread by @zorinaq

Chart of age-stratified IFR based on the Spanish serosurvey of 60897 participants

Open, verifiable data & Python source included

Script: gist.github.com/mbevand/f0aa2c…
Serosurvey: mscbs.gob.es/gabinetePrensa…
Death stats: mscbs.gob.es/profesionales/…
Population pyramid: worldpopulationreview.com/countries/spai…

"Overall" IFR is 1.140% based on 26744 deaths in the country as of 5/11

IFR for "ages 0-199" is 0.798% based on 18722 deaths, which is the total reported in table 3 (mscbs.gob.es/profesionales/…) The discrepancy is because deaths with no age data are not included at all in table 3

Limitations of these calculations:

• true IFR may be HIGHER due to right censoring: some infected persons may have died after 5/11

• true IFR may be HIGHER due to underreporting of deaths (economist.com/graphic-detail…)

• true IFR may be LOWER due to false negatives in serosurvey

@gummibear737 had done similar work

https://twitter.com/gummibear737/status/1262453620735385605

He calculated a 0.734% IFR
I found 1.140%
Why?

He excluded 8k deaths. He presumes they were suspected deaths as opposed to confirmed but table 3 header explains deaths with no age data are simply removed from the table

Also he did his calcs based on deaths as of 05-May

I did mine with deaths as of 11-May (the last day the serosurvey was conducted)

Because the serosurvey was conducted between 27-April and 11-May, I believe using deaths as of 11-May more accurately minimizes right censoring

But by large the biggest factor explaining our differences is that he did not include the 8k deaths that have no age data.

If I ignore these 8k deaths, I find an IFR of 0.798% close to what he calculated (0.734%)