Almost as predictable as the tides, many people were expecting to see a rise in COVID cases in states with laxer public health efforts. I started to see the rise in the Florida data at least 2 weeks ago a little bit. That hint a numbers person gets after looking at these curves for over a year…
Last week I said in one morning’s APR that I was close to doing some modeling and then I did. I left the Florida model I made alone since Florida reports sometimes….err, irregularly. The national model I started to tweak yesterday, adding a week to the curve fitting data my projections are based on. After a week, my Florida model looks good enough to continue to leave alone —
I look for 7-10 days of decent performance from my models — only recently during the quick decline of cases earlier this year was I able to get to that point of skill with the modeling calculations. It’s also a bit of luck, and dependent on a solid signal of a rise or fall to get a decent guess from this method…
(Note — the model error is calculated for only the days that I have initial reported data — for Florida above that is 6/19 through 7/3. Since I force my models to match the first day the average error omits the first data point.)
As I was tweaking my national model yesterday I took the numbers from 6/21 on Worldometers with a little grain of salt, seeing that Florida was missing 7/8. I took the most recent 7 day moving average for Florida as a guess to use as my final data point, using 16,300 as my guess where the 7 day moving average would end up for July 8th. It ended a bit higher (~16,800). The 3 models I calculated ended up having a ridiculously tight range of values for the next 10 days...the models diverge a bit after that. This one might need another week of data to get a meaningful variation to the model slopes, as right now I really could use any one of these (or the average) and see just about the same error.
Sometimes having really low error is NOT a good thing? Anyway, leaving just the average model for the 18 days of initial data and the next week and a half shows the curve matching better between the average model and the reported data:
This will be an interesting/stringent test of my methods...By mid-week I should be able to see if I’m in the ball park or grossly in error. Good error calculations for a set of starting curves really don’t mean a whole lot.
Finally, last week I was suspecting that Florida is leading the way in this surge, and numerically that might be true — at an average of nearly 3K new cases a day we are pumping about 1 out of every 6 new cases out of our viral engine...much higher than the ~6% of national population. But the national trend is strong to see only a couple days after it starts in Florida, the rest of the USA a few days after that. Florida displayed a marked increase in the new case slope a bit before the rest of the country, but it seems that increase in slope is now happening across the country:
The 3 surges started about the same time — that could have been driven by the southeast USA where vaccinations are lagging. But the pace of the surge increased in Florida about June 29th, and for the rest country and the country as a whole that increase in pace to the surge doesn’t happen for almost a week. I think that means the rest of the country is starting to see the Delta Surge in earnest.
Stay tuned, stay safe...