search

micb25 / dka

Statistische Analyse und Visualisierung der täglichen Diagnoseschlüssel der deutschen COVID-19 Tracing-App (Corona-Warn-App).
https://micb25.github.io/dka/
GNU General Public License v3.0
91 stars 5 forks source link

Add graph for keys per user #14

Closed daimpi closed 4 years ago

daimpi commented 4 years ago

Each user can at max upload 13 keys. It seems that recently there has been a strong divergence in the number of users uploading keys vs. actual number keys uploaded: 22.08: 86 people vs 639 keys: 7.4 keys/person 20.08: 91 people vs 948 keys: 10.4 keys/person

My guess would be that maybe some ppl who got tested and wanted to receive their result via app only installed it once they got tested and therefore only very few keys will be submitted.

This information might also be interesting in context of the discussion on how to improve CWA efficiency/uptake in the wishlist repo: https://github.com/corona-warn-app/cwa-wishlist/issues/150.

That's why I thought it would be nice to have a graph visualizing the keys/person development 🙂.

ndegendogo commented 4 years ago

As far as I understand, a positive-tested person uploads her (up to 14) keys in two batches: first batch is up to 13 keys, for the last 13 days (as far as keys are available), but excluding today's key. And then, on the next day, the remaining key.

Maybe this asymmetric split explains the observed divergence in ratio.

daimpi commented 4 years ago

Afaik the German CWA (in contrast to the Swiss app) currently doesn't have "last day uploads" integrated: and therefore is actually limited to a maximum of 13 keys/person.

ndegendogo commented 4 years ago

Wow - I was not aware that this is not implemented. What a pity. In my understanding the last day would be much more interesting than the old key from 14 days ago.

daimpi commented 4 years ago

I think the devs have it on their todo list: https://github.com/corona-warn-app/cwa-app-android/issues/678#issuecomment-650760871 🙂

But if you go to get a test and self-isolate while waiting for the results, you shouldn't have much contact on the upload day anyway I guess…

ndegendogo commented 4 years ago

Well - I am thinking of all the travellors returning home from their holidays. They get a test at the border. They ride home, maybe using public transport. And then their fridge is empty, so they do some shopping. And they start quarantaine after they get their result, not before. Exactly these contacts will not show up in manual contact tracing. Here the app could be a big benefit.

daimpi commented 4 years ago

Hmm… I think the rules are actually that you have to quarantine (at least if come from a high-risk area) until you get the result. But yes, some ppl will not adhere to this. But even if they come back on day 1, get tested and then go shopping, if they get their results the next day, everyone they met on day one will get a warning.

But in principle I agree that it would be better if also the encounters on the upload day would receive a warning.

daimpi commented 4 years ago

I just realized that https://ctt.pfstr.de/ is already showing "keys per user".

Given this new info, I don't think this function needs to be duplicated here, so unless you really want to implement such graphs feel free to close this issue @micb25 🙂.

daimpi commented 4 years ago

In the meantime I found out that the numbers here

22.08: 86 people vs 639 keys: 7.4 keys/person

seem to be distorted by some kind of problem with the packages on this day. I reported this issue in over at @janpf's repo: https://github.com/janpf/ctt/issues/4 🙂.

micb25 commented 4 years ago

Given this new info, I don't think this function needs to be duplicated here, so unless you really want to implement such graphs feel free to close this issue @micb25 slightly_smiling_face.

I like the way how @janpf solved it with an interactive plot. Since I'm using non-interactive gnuplot graphs, I might add an "average key length per day" graph in the future. But you're right, this is not really necessary.

seem to be distorted by some kind of problem with the packages on this day. I reported this issue in over at @janpf's repo: janpf/ctt#4 slightly_smiling_face.

Does the difference originate from the different way of analyzing the CWA packages (in @janpf's repo: daily packages; in mine: hourly packages)?

daimpi commented 4 years ago

Does the difference originate from the different way of analyzing the CWA packages (in @janpf's repo: daily packages; in mine: hourly packages)?

Nope, he's actually also using hourly data for the graphs. There seems to be an issue with the hour 17 package that day (it seems to be missing TRL 6 keys): https://github.com/corona-warn-app/cwa-server/issues/723