With guidance from this newly released CDC document, federal agencies are modeling the COVID pandemic using implausibly low fatality rate.

Their "best estimate" has a symptomatic CFR of 0.4%

Their worst case scenario has CFR — not IFR — of 1%

https://www.cdc.gov/coronavirus/2019-ncov/hcp/planning-scenarios.html

Their best estimate of the CFR for people over 65 is 1.3%, and the worst case scenario is 3.2%.

I'd love to know how these estimates were obtained, given that they are being used for government planning and recommended to modelers everywhere.

h/t @RAVerBruggen, @AndreasShrugged

The more I think about it, the more this bothers me.

These numbers are so far outside of the scientific consensus that this strikes me as a devious and cynical effort to manipulate not only federal modeling but the broader scientific discourse.

People are asking me what the IFR ranges ought to be. That's tricky and depends in part on the purpose of the models for which the parameters are intended.

That said, I'd want to span the range of reasonable estimates, and I want a point estimate that in accord with such.

The CDC estimates do not seem to be doing that.

Here's one reasonable (IMO) systematic review of IFR estimates.

https://www.medrxiv.org/content/10.1101/2020.05.03.20089854v2

John Ioannidis has lower ranges here https://www.medrxiv.org/content/10.1101/2020.05.13.20101253v1.article-metrics, but I don't find them credible; see https://twitter.com/GidMK/status/1262956011872280577

Imperial College Report 23, released today, estimates IFR for all US states. Almost all of the CDC's parameter range falls well below the almost all of the probability mass in the Imperial estimates.

https://www.imperial.ac.uk/mrc-global-infectious-disease-analysis/covid-19/report-23-united-states/

Update: fixed the decimal place for the CDC "best estimate":

CNN reports on the CDC's parameters.

https://www.cnn.com/2020/05/22/health/cdc-coronavirus-estimates-symptoms-deaths/index.html