covid icon indicating copy to clipboard operation
covid copied to clipboard

CI ranges for "doubling/halving times" don't appear to include point estimates.

Open cdowd opened this issue 3 years ago • 2 comments

See (for example): https://epiforecasts.io/covid/posts/subnational/united-states/maryland/

Doubling/halving time (days): -17 (29 – -6.9)

This looks to be reporting a CI of [-6.9,29], which does not include the point estimate of -17. It appears that the true CI that should be reported is (-Inf,-6.9] U [29,Inf), and that the naive view of the "reported" CI is actually a 5% CI.

This mostly looks like a typographical issue, arising from a denominator which isn't behaving nicely.

cdowd avatar Feb 17 '21 23:02 cdowd

Hi @cdowd,

Thanks for this. As you note this is because the doubling time is a slightly odd metric in that when it inverts to the halving time it goes through infinity. If you read the CI with that in mind it is technically correct but I agree confusing.

I'm very open on other ways to represent this. Could you just clarify your suggestion with a few examples?

(I've opened an issue in the package we use to produce these estimates as any change needs to be made there).

seabbs avatar Feb 18 '21 14:02 seabbs

I'm afraid I don't have any suggestions much better than using standard set union notation. E.g. as I wrote above

(-Inf,-6.9] U [29,Inf).

I suppose an alternative would be to split out doubling and halving times separately, e.g.

CI: Halving time (-Inf,-6.9], Doubling time [29,Inf).

If those are both unpalatable, a simple solution would be to throw an asterisk on the CI being given, and note that it goes through infinity. E.g.

Doubling/Halving time: -17 (29 - -6.9)† † Intervals where the order is reversed (e.g. 1 - -1 instead of -1 - 1), include everything except what is between the numbers shown.

This would perhaps be the simplest of the solutions, as it can be tacked on purely at the end and only need affect the doubling/halving time CI calculations. Though the adjacent dash and minus symbols are still somewhat confusing. Not sure how easy it would be to replace the dash with the word "to" (in both the text and the examples of the footnote), but that would then completely resolve things.

cdowd avatar Feb 18 '21 23:02 cdowd