sondreus
commented
2 years ago Dear @RoyceWHowland, thanks for sending this and the kind words.
At the moment, there are no plans to harmonize our excess mortality estimates with those from government sources. However, our estimates are based on government total mortality counts, and we incorporate updates to these as rapidly as we can. The reason is that total mortality counts (i.e. the number who have died in a given period) do not depend (or if they do, less so) on a model, whereas excess mortality always does (as they need a modelled number of expected deaths in a given year).
We prefer to apply a consistent model to estimate expected deaths, and thus excess deaths, to all countries. This both ensures that a relatively robust model is used (that takes into account changes over time with being overly sensitive to recent mortality spikes or drops), and that countries can be compared over time. This is not to say that specific government's models are necessarily worse, but it could be that theirs are built for a different purpose. For instance, some may wish the expected mortality to reflect what is plausible in a "good year" (without many flu deaths, for instance), or use simpler models that are even easier to interpret (such as averages without time trends, which tends to get worse the more years use for the average).
Hope this clarifies
I make use of The Economist global excess death model for some further modelling work of my own. Excellent work! I can't say enough how grateful I am to have a resource like this available to the public. Many, many thanks!
I have been delving into understanding specific estimates, and I'm currently looking in detail at the estimates for Canada (where I live). The government releases its own excess death information through Statistics Canada, updated periodically. A new update was just dropped today. Find it here: https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=1310078401
The StatCan data table is pretty simple, basically giving a 95% CI range and estimate for total deaths and expected deaths by province, and for the country as a whole, in a weekly time series.
Looking at the total excess deaths calculated from the StatCan data and comparing it to the same time period from The Economist data, the StatCan excess deaths are higher. E.g. the StatCan cumulative excess deaths up through 2022-01-01 are 28,388, about 15% higher as compared to The Economist cumulative excess deaths up through 2022-01-03 at 24,614.
Officially reported Covid deaths in Canada stand at 30,616, meaning both excess death estimates fall below that level. Most of us looking closely at Canadian reporting believe the Covid deaths are underreported, especially in certain provinces including Ontario, British Columbia and Alberta. So if anything, the national Covid death count should be higher yet.
Looking at the exported Canada historical deaths file here in the excess deaths tracker repo, the expected_deaths column shows "TBC". (The same as all the other countries I checked.) Obviously the StatCan published expected deaths numbers are not being used.
So I guess my question(s) relate to truing up the estimated excess deaths model from The Economist vs. any "official" published numbers from a government source. Is there any intention to fold such government expected / excess death data directly into the model? I'm simply doing that in my own model for the time being...