Right now, the observations are a running total, which doesn't lend itself to time-series analyses.
Derive the day-by-day value (e.g. 10 people recovered that day for a specific jurisdiction) by subtracting the running total of the previous interval from the current interval.
This will allow us to answer questions like "how many people recovered in the last 2 weeks in this area" and start computing confirmed, deaths, and recovered indices; compute rate of recovery, etc; and do benchmarking across jurisdictions for arbitrary time intervals (e.g. last month, last 3 days, etc.)
I also think this will make the continuous aggregates more useful.
Right now, the observations are a running total, which doesn't lend itself to time-series analyses.
Derive the day-by-day value (e.g. 10 people recovered that day for a specific jurisdiction) by subtracting the running total of the previous interval from the current interval.
This will allow us to answer questions like "how many people recovered in the last 2 weeks in this area" and start computing confirmed, deaths, and recovered indices; compute rate of recovery, etc; and do benchmarking across jurisdictions for arbitrary time intervals (e.g. last month, last 3 days, etc.)
I also think this will make the continuous aggregates more useful.
cc @avthars @coolasspuppy @mfreed