Regarding the proposed removal of recovered values

[cipriancraciun]

In issue #1250 it was announced that the recovered numbers are going to be removed from future versions of the datasets due to inconsistencies in reporting or issues in collecting these values.

However I would suggest keeping these numbers at least for the countries that do report them and give a disclaimer that this numbers are inaccurate. (Perhaps move these files to a separate folder to make it even more clear.)

I think the recovered values are essential for computing the number of active cases, which helps visualize the overall progress of the pandemic.

[cipriancraciun]

I'll link to this issue other issues regarding the removal of recovered values, like:

• #1311 -- asking if perhaps the recovered values will still be available in the ARCGIS API;
• #1313 -- quoting: "unable to derive Active Case Trajectory";
• #1350 -- mentions this besides other issues;
• #1417, #1580 -- asking for alternative datasets;
• #1373, #1392, #1364, #1344, #1359, #1356, #1430, #1512, #1546 -- other duplicates;

[cipriancraciun]

I have updated my previous comment about other linked issues. It seems that more and more would pop-up as the various derived datasets are being updated.

[KeulRalf]

I'd agree. I would love to get recovered numbers where we can get those even though these are not complete for all countries. Currently Active is the most important number we report.
If some don't deliver I would prefer to get them at least for those countries data is trustable.
Thank you for everyone's effort here.

[lkounadis]

please add recovered. Its really important to know. It actually helps people feel "safer" if you get my drift

[karen2k]

Assuming people heal completely in N days at max, the number of recovered cases is equal to the delta of the confirmed cases minus deaths minus confirmed cases N days ago. You can find visualization here and play with N=14 or N=21: https://covid-19.karenishe.com/

[Akuukis]

Most of dashboards use "recovered" to calculate "active" cases. I agree with JHU that recovered is badly counted and thus better not used at all, and thus the "active" cases are not accurate too.

IMHO The best solution is to use model by Tomas Pueyo that estimates "actual" cases based on deaths not recovered - so I made a dashboard that does it: https://pandemic-estimator.net

I'm happy to chat more on this direction, ping me in a issue of your own dashboards or create one here

[cipriancraciun]

I think there are conflicting issues (due to conflicting reasons) thus a lot of "drama" on this subject. Let me explain:

• some people try just to build nice graphical representations -- thus even an estimate would be OK; (however they aren't statistics / math experts, thus they just want a "number", "any number";)
• some people try to build statistical models, and they use the recovered values as inputs or as validation; (thus they want "the" number;)

Now JHU, because they say recovery data collection is unreliable decided to just drop it from every country, which basically let both categories without any inputs.

@Akuukis can you provide us with a simple formula that one can apply in a "normal" programming language (i.e. Python, Ruby, Java, etc.), thus not "maths" programming languages (like Julia, R, etc.)

If I can implement it I'll augment it in my own dataset.

---

I've looked in the "methodology" section of the pointed calculator, and although the formulas seem "simple", they require 4 parameters which I assume are quite dependent on the country's context...

Thus I am skeptical that any "computed" recovery values are accurate enough to be useful.