Open mkadunc opened 3 years ago
@mkadunc what do I do with this?
@mkadunc @lukarenko to be honest, I don't know what I'm supposed to do here. The code for Prirast PO (the 3 disabled pages of that chart) already calculates the growth based on active cases (not cumulative cases). Am I misunderstanding things? Also, the Explainer page (see comment above) uses a fixed 7 days doubling value.
@breki problem is that 3rd chart - Prirast v dnevih:
This is calculated only taking into account new confirmed cases per day, but not taking into account that some will also recover that day. So the growth is faster that it actually is.
Idea of this chart by @pkese was: to show how small growth contribute to exponential growth, by showing first in days and when the reader has relation of doubles in X days, show what 7 day growth (as example) could look like from current numbers of today
@breki problem is that 3rd chart - Prirast v dnevih:
This is calculated only taking into account new confirmed cases per day, but not taking into account that some will also recover that day. So the growth is faster that it actually is.
OK, but how does this relate to @mkadunc description:
To remedy the issue, we should compute the weekly growth factor using 7d sums today and a week ago. It would be smooth enough and still relatively current information ("only" 1 week old data point). Formula: growth = cases[0-6]/cases[7-13] - 1 , this produces percentage growth rate - if it's above 0, it's bad (+66% growth in a week); if it's below 0, it's OK (-20% growth in a week ~ 20% decrease).
Idea of this chart by @pkese was: to show how small growth contribute to exponential growth, by showing first in days and when the reader has relation of doubles in X days, show what 7 day growth (as example) could look like from current numbers of today
Yes, I understand the intention, but I think it would be more useful to show the values for the actual doubling rate, not the example 7-day one. Also, the number of projected hospitalizations (2528 in 4 weeks) is misleading, since it's way beyond the capacity - maybe we should show the number of people that would not get the place in hospitals instead?
Yes, I understand the intention, but I think it would be more useful to show the values for the actual doubling rate, not the example 7-day one.
I agree on this point — making it closer to reality would make it more understandable. We could also do the first panel with "1 week ago", next panel with "now", and then use the same (or similar) factors in subsequent panels...
Also, the number of projected hospitalizations (2528 in 4 weeks) is misleading, since it's way beyond the capacity - maybe we should show the number of people that would not get the place in hospitals instead?
We could just call it "people requiring hospitalization" - it will also be more accurate if we use admitted to hospital instead of active hospitalized
@breki problem is that 3rd chart - Prirast v dnevih: This is calculated only taking into account new confirmed cases per day, but not taking into account that some will also recover that day. So the growth is faster that it actually is.
Growth in new cases is what we should be looking at - using active only blurs the growth factor and shifts the window more into the past. Same with active hospitalizations and ICUs — both of those are lagging
OK, but how does this relate to @mkadunc description:
To remedy the issue, we should compute the weekly growth factor using 7d sums today and a week ago. It would be smooth enough and still relatively current information ("only" 1 week old data point). Formula: growth = cases[0-6]/cases[7-13] - 1 , this produces percentage growth rate - if it's above 0, it's bad (+66% growth in a week); if it's below 0, it's OK (-20% growth in a week ~ 20% decrease).
Related: I think the growth rate plot is wrong - there should be regions when weekly growth rate was negative - e.g. 29.3. to 20.5., 4.7. - 5.8
It would be good to show the growth factor (of cases, maybe new hospitalizations and new ICU admissions) - we used to have it on "Prirast PO", but it was computed against cumulative cases and became useless, so we turned it off.
To remedy the issue, we should compute the weekly growth factor using 7d sums today and a week ago. It would be smooth enough and still relatively current information ("only" 1 week old data point). Formula:
growth = cases[0-6]/cases[7-13] - 1
, this produces percentage growth rate - if it's above 0, it's bad (+66% growth in a week); if it's below 0, it's OK (-20% growth in a week ~ 20% decrease).