jgehrcke
commented
3 years ago Hey! Thanks for the detailed feedback Sarah!
First of all, I want to say thank you for all of the work you've done in making these data available in a more user-friendly format.
Thank you for the kind words!
Confusingly, the "Meldedatum" field refers to the date on which the case was first reported
There certainly was a time when I didn't know about this, but in the meantime this has been known to me for a couple of months :). It's buried in the README of this repository where I write
Note: there is a systematic difference between the RKI data-based death rate curve and the Risklayer-based death rate curve. Both curves are wrong, and yet both curves are legit. The incidents of death that we learn about today may have happened days or weeks in the past. Neither curve attempts to show the exact time of death (sadly! :-)) The RKI curve, in fact, is based on the point in time when each corresponding COVID-19 case that led to death was registered in the first place ("Meldedatum" of the corresponding case). The Risklayer data set to my knowledge pretends as if the incidents of death we learn about today happened yesterday. While this is not true, the resulting curve is a little more intuitive.
Could this be what is causing the discrepancies?
This is unrelated to #227, if that's what you're referring to here :). But no worries.
(Apologies if you're already aware of the somewhat weird way Meldedatum is defined.)
No worries at all! Truly appreciate every pair of eyeballs on all this, and people joining the discussion :)
the only way to tell when a death occurred is using the "Datenstand" field on the day the death first appears in the dataset
This is indeed an interesting approach, but then again this field could carry a wrong/misleading date, too. You probably agree that it's a little unintuitive (and sad) that the reporting chain and after all the RKI does not simply and explicitly track the 'day of death'. And maybe they track this even, but the ArcGIS databases don't expose it. Or we are to stupid to find it :-).
As I wrote in the quoted paragraph above, from my point of view the Risklayer data-based evolution of deaths appears to be somewhat more useful.
In any case, please please report back when you find more interesting detail, and especially when you find ways to remove ambiguity / get more clarity about the date-of-death based on RKI data.
sarahckramer
Hi there!
First of all, I want to say thank you for all of the work you've done in making these data available in a more user-friendly format.
I have also been exploring/using the RKI COVID-19 data in my work, and wanted to alert you to a possible error in your calculation of daily deaths (perhaps related to issue #227). For example, "deaths_rki_by_state.csv" has the first deaths occurring on March 4, even though the first deaths from COVID in Germany did not occur until the 9th.
I'm not sure if this is why, but I know that I originally made the error of assuming that the field "Meldedatum" referred to the date on with the death was reported, and I wonder if you have been making the same assumption. Confusingly, the "Meldedatum" field refers to the date on which the case was first reported, and the only way to tell when a death occurred is using the "Datenstand" field on the day the death first appears in the dataset - see comments here: https://www.arcgis.com/home/item.html?id=f10774f1c63e40168479a1feb6c7ca74, for example: "Das Meldedatum ist immer das Datum, wann dem zuständigem Gesundheitsamt der Fall bekannt gemacht worden ist. Sollte der Fall am 1.5.2020 positivt getestet werden und der Fall wird am 1.5.2020 dem Gesundheitsamt bekannt gemacht, so ist das Meldedatum der 1.5.2020. Sollte der Fall im Nachgang, z.B. am 7.5.2020 versterben, so wird das Meldedatum nicht geändert und bezieht sich immernoch auf den 1.5.2020." Could this be what is causing the discrepancies? (Apologies if you're already aware of the somewhat weird way Meldedatum is defined.)
Best, Sarah