The title picture is a couple of graphs that I put together last night after getting the daily Florida COVID case and fatalities data. I have been keeping track of the data for almost 2 month on a national scale, and this is the first foray into looking at particular state.
Prologue — My Model
In most respects what I am doing is “retrospective”, since I certainly rely on a body of “real” data to establish what is happening. My earliest work on this was looking strictly at exponential growth, but about 2 weeks ago I made a mental breakthrough with how to use a spreadsheet to work with a logistic function based model. Perhaps a week ago I had the idea to create a 3 stage composite model in the hopes of plotting a curve that suggested the “apex peak” of fatalities and deaths.
An early attempt to use a single stage logistic model was suggestive, but obviously was already failing to provide any future guidance…
At this point — roughly a week ago, the single stage model was actually providing a decent (if not great) fit to the data. I played with superimposing 3 separate models and came up with each stage having somewhat different characteristics, and starting roughly a month apart. I suspected the 3rd stage would be a “slow burn” compared to the 1st two stages.
Eventually I was able to get this chart:
Reservations about the Florida Chart
I have some reservations about what the FL chart is saying — deaths are to spike without a really obvious source in the case predictions. But I just put this together last night, so there’s that. But the fatalities chart fell together VERY quickly using almost the same parameters as the national model, but with scaled down maximum values in the 3 stages. That was a bit spooky — I did not expect that I would have to tweak that so little.
The cases chart took a few iterations of playing with the parameters to get it look as it does. But still the 3 stage model appears to capture something that i going on. What that seems to be is the 3rd stage, as I have suspected, is just getting revved up.
So I am reverting back to may meteorological training, and looking at the trends. Since deaths lag cases, I will concentrate on the case curves — let’s not harp on the quality or completeness of the data, please. It is what it is. I think what trends we see are real, and I have always treated these case numbers are a very serious lower bound.
Making the Case for a 3rd Wave
Removing the 3rd wave from my Florida case model yields a curve that still fits pretty well, but on day 81 (5/10/2020) predicts about 100 cases per day. By themselves these first 2 waves estimate 34K cases, and as of yesterday Florida had about 1500 more cases. So my 3rd stage NEEDS at least that to catch up, and the 1st two are not completely done, so at a minimum we need to look at at least 3K cases in this stage. At this point the parameter of this stage are very speculative. Again , let’s dwell on the qualitative aspects more than the actual numbers. Looking out to a farther time-line (through 7/31/2020), I plotted 3 cases = 3K in a third stage (nothing much), 15K (meaning we are about 2/3 way through the total number of expected cases), and 30K (a whopper 3rd wave that will be impossible to ignore):
Please pardon the lack of state identification in the chart — I rushed this together last night and did not save these — I copied them out of the document…
Anyway, note how each of these scenarios alter the 5/10/2020 predictions...I will get back to that.
But let’s think about this...without the third model the curve goes to zero just abut the end of May, but that under-predicts where we are today. SO THERE IS GOING TO BE A THIRD WAVE. And if the state is “opening back up” today, there are obviously cases still around, and if anything we’ve seen in the last 4 months matters, there are asymptomatic cases walking around in the state.
Maybe my timelines get changed, but there will be another spike in cases in May. This is not over.
Getting a handle on the next scene of Act I
May 10 looks like the day where I can start to evaluate what the 3rd wave looks like. But this is scene 3 of Act I. Act II comes in the fall, and will include the flu. It’s inconceivable that COVID goes away — almost every other pandemic since the 18th century has lasted 12-24 months.
For Florida we see a simple guide to at least estimate the magnitude of the coming 3rd wave — if on 5/10 we’re seemingly averaging about 100 cases a day, the coming wave will be minimal. If it’s about 200 or more then we’re in for a another resurgent month of cases. Of course how long the 3 wave is can affect the numbers, the entire model maybe be shot down by next Sunday.
For my national model, taking out the 3rd wave already shows a hint of what the simple 1 stage model was hinting at on April 27. Working the model the same way I get the same type of relationship I see for Florida:
If national cases in a week seem to be in the 5K range a day (yeah, riiiiiight) then the 3rd wave is minimal. This looks very unlikely.
If national cases are about 10K at the end of next week (maybe we have to wait until about 5/12 to account for the “weekend lull”) then we’re looking at a 3rd wave that is about as big as the 1st two, but likely spread out and not quite as bad a peak.
If national cases are hovering about 15K a day then the 3rd wave would appear to be larger than the either of the first two.