Life's a beach, and then you die

Alternative title for this post: The Sunshine Charnel House

Last week I put up a post illustrating the tremendous lag in Florida's COVID death data.

Most states take a few days to collect their data and get it posted, but after that point their older data stabilize. They're not continually updating numbers that are 10 or 20 days old.

Not Florida. They're special. It's now 25 days since I started watching their death numbers, and they're still going back and adding deaths to data that are two months old.

But they're special in another way as well, and that is simply in the amount of COVID-related death they're producing.

Along the way to illustrating that, I'll give you some pretty good evidence that VACCINES WORK, and also give a sense of how much better we might be doing with a higher vaccination rate.

Back in July I posted a comparison of the Delta surge in a collection of nine not-randomly-chosen states. (I wanted places famously having trouble, like Florida and Missouri, and the state with our highest vaccination rate - Vermont - and then through in some others, including New York, where I live, and Massachusetts, where I grew up and where I still have family.)

Overall, states with higher vaccination rates started their surges later, from a lower level, and weren't rising as fast.

That's still true, but I wanted a more comprehensive measure, that somehow gathered in both how early you started, how high you started, how high you got, and how much time you spent at a very elevated level.

What I settled on was to look at cumulative infections in a state from June 1 through August 30. I measured that by taking total infections recorded as of August 31, and subtracting  total deaths recorded as of June 1.

With that "How was your summer?" question as the main effect, I chose as my causal variable a state's level of vaccination on June 1, at the start of the mess. And I went with "fully vaccinated" rather than "at least one shot" - early in the vaccine roll-out, there was evidence that even the first shot of the mRNA vaccines did a good job of preventing infection, but that with Delta it really took both shots to get a good effect.

So: I'm looking at the relationship between:

1. What was each state's population percentage that was fully vaccinated by June 1st; and
2. What portion of each state's population got infected between then and August 31st.

And that relationship looks like this:

It certainly does appear that states with higher vaccination rates at the beginning of the summer (further to the right on the chart) also had a smaller percentage of their population test positive over the summer (lower down on the chart).

And if you fit a power function to the scatter of points, you get an R-squared of 0.627, suggesting that this single factor - how vaccinated was your state on June 1 - can explain about 63% of the variance in how many people tested positive.

The t-statistic on the vaccination variable is over 9, implying a vanishingly small chance that there is no relationship in the real world, and that the apparent relationship is just random noise falling a certain way.

And the relationship is not merely statistically significant. Its estimated size is also large enough to be meaningful. The country's overall rate of full vaccination by June 1 was 40.7%. The regression predicts that a state with that level of vaccination would see 1.43% of its population test positive over the summer. For a state with 10,000,000 people (roughly Georgia, or North Carolina, among others), that would mean about 143,000 infections over the summer.

If the state had 10 percentage points more vaccination (meaning it was at 50.7%), it's predicted positives over the summer would be 0.83% of its population. Our hypothetical state with 10,000,000 people would see only 83,000 infections - 60,000 fewer than with the average vaccination rate.

As you can see on the figure, the relationship is not linear. The lower your vax rate, the more quickly infections rise. So a state with 10 percentage points less vaccination than average (putting it at 30.7%) would expect an infection rate of 2.89%. In a state with 10,000,000 people, that's 289,000. That's 146,000 more infections than at the average vax rate.

On these charts, you might have noticed one particularly "out there" outlier.

In case you didn't, here's the second chart with the outlier highlighted.

It is 3.1 standardized residuals above its "predicted" value. The next-biggest outlier is only 1.9 standard deviations off. So Florida went into the summer with an almost average vaccination rate, but ended up really punching above its weight, with the second-highest rate of infections and an unbelievably high level given is vax status.

But that's just infections. Where Florida really excels is in COVID deaths.

Here's the chart. The horizontal axis is still percent fully vaccinated by June 1st, but the vertical is now deaths per 100,000 during June-August.

You probably noticed that outlier, but in case you didn't, I've made it easier to see.

Florida is 3.5 standardized residuals above its predicted value, wile the next largest error is 2.3 standardized residuals.

I don't know why Florida excels in death. It could be as simple as having an older population, in which case age-adjusted mortality rates would bring its numbers more in line with the other states.

Hypotheses are welcome.

The Florida slow-roll

You've probably heard in the news about what a hard time many states are having with the Delta surge, with schools having to close, ICU's overfilled, people denied care for other health problems because there's not hospital capacity, etc.

Ground Zero of this preventable catastrophe is ... Florida.

So you go to some data site like Worldometers, and you look at how things are going in Florida and:

From here, downloaded Sep. 12, 2021

Sure, they recently reached their highest daily death rate in the course of the pandemic, but since the peak on August 24th, things have been improving incredibly fast.

Here's another screenshot, focusing in on that end of the graph:

That's fantastic improvement!

Except that it's nothing more than a delay in reporting COVID-19 death data.

I would look in on the Florida page every two to four days, and it also seemed like they had recently peaked, and then things started getting better.


So on August 24th (as it happens) I started transcribing the day-by-day death stats for Florida off of Worldometers, checking back every couple of days to transcribe a new batch.


I somewhat arbitrarily chose July 30th as my starting point, opened up a new spreadsheet, and typed in the data from July 30th to August 24th.


Like on the Worldometers chart, I calculated the 7-day average, and the graph came out like this:





You can see the death rate peaking on August 11th, then falling very impressively, from 178 all the way down to 28, in just 13 days. Remarkable progress.


But let me re-scale that, so that there will be room for the later curves I'm going to have to add:





It's the identical data, just with the vertical axis allowed to rise all the way to 350, and the horizontal axis running out through September 9th.


I went back on August 27th, and the peak was a little later, and a little higher. That downturn after August 11th turned out not to be true:




That had been a Friday. I went back two days later, on Sunday, and the first two weeks of the death curve were unchanged from Friday, but the last seven days had all been bumped up:





Two days later, the whole peak had been raised. The peak now didn't come until August 16th, at 221 deaths per day.




After that, I missed five days. When I came back, the peak hadn't been pushed any later, but it had been raised, and a lot more deaths had been added to previously recorded days after that peak.





When I gathered data on September 8th, the peak had been pushed back to August 21st, and raised somewhat.





Friday, September 10th, saw a big jump and the peak not happening until August 24th.




Today, September 12th, and went and got the numbers again. Like August 27th and 29th I had a Friday-to-Sunday pair. And like that time, most of the curve was left unchanged from Friday to Sunday, with the update only affecting the last seven days of the curve.




I fully expect that if I go back on Tuesday, the entire curve will have been lifted again. The pattern over the last two weeks suggests that, even though the reported data show a decline in deaths, the underlying reality is that they're not done increasing their death rate in Florida.


I should take a moment to address one question that will be obvious to some readers: Is this a big deal? Isn't this just a function of data taking time to compile?


Maybe.


My best argument is that I've been looking at data for many states for a while (particularly for the nine states that featured in my previous post), and none of them have this obvious a rosy scenario. They report their data, and aside from a few small updates, their data don't keep drifting higher.


Just today I started the process of a small check on my impression, by gathering today's death data for New York state, starting likewise on July 30th. In a week or so, I should be able to tell if NY is doing something like what Florida has been doing.


I doubt it will be that dramatic.


In the meantime, we can count up how many deaths Florida had failed do count on August 24th, compared to their report today.





On August 24th (indicated by vertical, black lines), there were a reported 2,907 COVID-19 deaths in Florida.


Today, not quite three weeks later, that total number of reported COVID-19 deaths from July 30th through August 24th has more than doubled, to 6,187.


That particular window (through August 24th) might be just about ready to stop getting worse.


Stay tuned.