Risk metrics bouncing around wildly day-to-day

[josiah-roberts]

This may be some issue in the underlying data, but for instance the Santa Clara prevelence went from 15,074 microCOVID for an average person on Jun 26 to 5,400 on the 28th. The numbers have been bouncing around like this for some time; they'll triple or quadruple for a day or two, go back down, etc.

[zerotrickpony]

Hi Josiah, thank you for flagging this. I'm also in SCC and experienced this. Unfortunately this is an unlucky artifact of stale data, and is likely working as intended though a bit surprising.

The data feed had stalled on June 12th for the last few weeks, which you could only have known by looking in the tiny text of the "Data last updated" line inside the "Details" spinner. Starting over the weekend there was a "Warning: stale data" banner. Back then the adjusted prevalence modeled out to about 1.54%, due to the rising rates and high test positivity.

Fortunately, your friendly microcoviteers unjammed the data feed over the weekend, and you can now see based on yesterday's data that the Adjusted Prevalence estimate for SCC is now 0.5% or about 3x less. The reduction is largely due to the non-rising case rate, which causes the model to remove most of its undercount factor. (It's designed to get a bit conservative on the rising edge of waves, which we have been in until recently.)

The 3x reduction in prevalence estimate corresponds pretty precisely to the change in numbers that you saw. Sorry about the whiplash.

[zerotrickpony]

Oh wait apologies, I think you're right, it's not just the stalled data nor the rising percent. Sorry for the long-winded reply.

The current (6/29/2022) case count is 7325, but when I looked yesterday (6/27/2022) it was in the 3000s. I see on Google's feed that the 7 day average has been steady around 1100 for most of June, but the daily data shows as zeros 6 days out of 7, with big spikes of cases only every few days. If microcovid is using the 7 day average, it should in theory average out to 7000ish. That's the number that agrees with Google and agrees with the county dashboards:

Google, 7DA=1011:
https://www.google.com/search?q=santa+clara+county+covid&oq=santa+&aqs=chrome.0.69i59j69i57j69i59j69i60l3.2554j0j7&sourceid=chrome&ie=UTF-8

County official website, 7DA=1012:
https://covid19.sccgov.org/dashboard-cases-and-deaths

...so either we're pulling data that disagrees with this, or the cumulative calculator is choking on the zeroes.