change % positive calculation to rolling window

[andrewsu]

Consider changing % positive calculation to rolling window (e.g., over a 7-day window). As it is, it looks like % positive is based on cumulative totals of total tests and positive tests. Leads to nice smooth curves, but perhaps is a bit too robust to recent changes.

[flaneuse]

My opinion: I wouldn't expect any systematic periodicity to the percent positive data (unlike raw case counts), and it's pretty smooth past mid-April, so I would leave it as a daily value.

[flaneuse]

With the change from cumulative positivity to new positive cases per day, calculate 7 day rolling average for positivity. Also add in rolling averages for other metrics:

1. calculate testing_positivity on the backend =

     d["testing_positivity"] = d.testing_positiveIncrease
             ? d.testing_positiveIncrease / d.testing_totalTestResultsIncrease
             : 0;

2. Calculate 7-day rolling average (+/- 3 days, except at the edges) of testing_positivity as testing_positivity_rolling
3. Calculate 7-day rolling average (+/- 3 days, except at the edges) of testing_hospitalizedIncrease as testing_hospitalizedIncrease
4. Calculate 7-day rolling average (+/- 3 days, except at the edges) of testing_totalTestResultsIncrease as testing_totalTestResultsIncrease_rolling

[flaneuse]

Word of caution on testing_positivity_rolling: don't average the percent positivity, but rather calculate the ratio of sums. as in:

DON'T: mean(testing_positivity) DO: sum(testing_positiveIncrease)/sum(testing_totalTestResultsIncrease)