3 ms·

Can someone explain how they come up with the "expected deaths" number? The article doesn't explain how this number is calculated or which assumptions were made

by kesor 5y ago

Can someone explain how they come up with the "expected deaths" number? The article doesn't explain how this number is calculated or which assumptions were made to come up with the number. Is this a moving average of earlier years? Does it compensate for the increase or decrease in demographics? What other aspects go into this "expectation" that might be relevant?

mcculley 5y ago
I wondered the same thing. This is why I built some tooling around visualizing the body count: https://mcculley.github.io/VisualizingObservedDeaths/ https://mcculley.github.io/VisualizingObservedDeaths/ It turns out that in many places there is a very clear difference between 2017-2019 versus 2020-2021. There is no way the population increased enough to account for the extra deaths.
AnimalMuppet 5y ago
Can I explain what they did? No. Can I give an idea of how to do it? I think so. Way back in the day, one of the tables in the CRC Handbook was the "American Experience Actuarial Table". It said, for every age from 1 (or 0?) to 100, what fraction of people of that age one would statistically expect to die in America. You take a modern version of that data for the country in question, (perhaps per gender, which IIRC the CRC table did not account for) and throw it at the age distribution in that country. That gives you a fair idea what to expect. In doing so, you assume that the racial profile of the population didn't change much in that two-year span (or if it did, that it did so according to previous trends), unless you have actual data saying otherwise. Is it perfect? No. But it's fairly reasonable. Again, note my opening disclaimer: I don't know that this is how this study did it.
NickNameNick 5y ago
Its there at the bottom: >A previous version of this page used a five-year average of deaths in a given region to calculate a baseline for excess deaths. The page now uses a statistical model for each region, which predicts the number of deaths we might normally have expected in 2020. The model fits a linear trend to years, to adjust from long-term increases or decreases in deaths, and a fixed effect for each week or month.

by kesor 5y ago

mcculley 5y ago

I wondered the same thing. This is why I built some tooling around visualizing the body count: https://mcculley.github.io/VisualizingObservedDeaths/ https://mcculley.github.io/VisualizingObservedDeaths/ It turns out that in many places there is a very clear difference between 2017-2019 versus 2020-2021. There is no way the population increased enough to account for the extra deaths.

AnimalMuppet 5y ago

Can I explain what they did? No. Can I give an idea of how to do it? I think so. Way back in the day, one of the tables in the CRC Handbook was the "American Experience Actuarial Table". It said, for every age from 1 (or 0?) to 100, what fraction of people of that age one would statistically expect to die in America. You take a modern version of that data for the country in question, (perhaps per gender, which IIRC the CRC table did not account for) and throw it at the age distribution in that country. That gives you a fair idea what to expect. In doing so, you assume that the racial profile of the population didn't change much in that two-year span (or if it did, that it did so according to previous trends), unless you have actual data saying otherwise. Is it perfect? No. But it's fairly reasonable. Again, note my opening disclaimer: I don't know that this is how this study did it.

NickNameNick 5y ago

Its there at the bottom: >A previous version of this page used a five-year average of deaths in a given region to calculate a baseline for excess deaths. The page now uses a statistical model for each region, which predicts the number of deaths we might normally have expected in 2020. The model fits a linear trend to years, to adjust from long-term increases or decreases in deaths, and a fixed effect for each week or month.