Thread by @ArielKarlinsky

@IHME_UW just released a report estimating excess deaths at about 2x total COVID deaths in the world.

The details of the analysis are here: healthdata.org/special-analys…

While important and prob true, I'm left with quite a few questions about this particular estimate.

1/

Their data comes form "56 countries and 198 subnational units have reported either weekly or monthly deaths from all causes for parts of 2020 and for prior years. " but where is it? The References listed are scant and include a few NSOs.

2/

After estimating excess deaths with some ensemble models for the countries they do have, they project it using various covariates to other countries. But which are projections and which are actual data? unclear.

3/

After estimating excess deaths they correct it for other non-covid related deaths (such as opioid deaths) and compute a ratio between their estimate and reported COVID deaths - similar to @hippopedoid and mine's "Undercount Ratio".

But some of these make no sense to me.

4/

For the US they find COVID-related Excess Deaths to be at 905,289 compared to 574,043 reported.
Our estimate of excess at about 605k deaths up to Mar 30th.
How they arrive at such large figure?, especially since some of their other figures match quite closely with ours.

5/

For Japan, where we find NEGATIVE excess mortality, they find actual deaths at over 100k as opposed to reported ~11k.

from just the raw data for Japan, I just don't see how. 2020 and 2021 deaths are completely within baseline/predicted.

6/

For France and Germany they find ratios of 1.3 to 1.45, while in our estimates excess deaths are LOWER than COVID deaths, especially after the first wave.

7/

For India they estimate 654,395 deaths compared to 221,181 reported, but data from India is very hard to come by as I have been in contact with researchers and journalists and prob. the best we have right now are data from Mumbai, Nagpur and Ahmedabad.

8/

They do not cite (and do not use?) #WorldMortality, or @HMDatabase's STMF, or refer to the various repos who use our data and brought the issue to the world's attention such as @OurWorldInData , @FinancialTimes, @TheEconomist , The Imperial College and more.

9/

Any thoughts on these estimates? @jburnmurdoch @J_CD_T @CT_Bergstrom @GidMK @muradbanaji @ojwatson92

They now include on the left-hand side a CSV file with sources. They have 47 countries listed there.
NO DATA listed for: India, Iran, Egypt, Peru, Indonesia, North Macedonia, Belarus and maybe more (these are just the countries shown in the "detailed analysis" tables).

10 /

Also, some of the sources are pretty outdated by now. Japan's listed latest data is September 2020 (we already have Feb 2021), Ecuador monthly 2020 (we have weekly deaths in 2021) etc.

Iran, Egypt, Peru and more are available at #WorldMortality
github.com/akarlinsky/wor…

11 /