The seven-day moving average of cases fell today. For the U.S. as a whole, and in many regions and states. The U.S. stands at 234 new COVID-19 cases per 100K population per day, up just 39% in the last seven days. And down from a revised 239 from yesterday.
The question of the day is: Are these turnarounds in the trend lines real (and so expected to continue), or just an artifact of holiday reporting (and so, just a one-off jiggle in the line)?
Unfortunately, my answer is that this one-day decline is mostly a data reporting artifact. New case counts may be slowing down, but there has been no true decline in new cases yet.
All this means is that the actual rate of new case growth is lower than what has been shown for the past few days. But new cases are (probably) still increasing, and will continue to increase for a few days yet.
Data source for this and other graphs of new case counts: Calculated from The New York Times. (2021). Coronavirus (Covid-19) Data in the United States. Retrieved 1/8/2022, from https://github.com/nytimes/covid-19-data.” The NY Times U.S. tracking page may be found at https://www.nytimes.com/interactive/2020/us/coronavirus-us-cases.html.
Drill-down
I would love to belabor this, but the issue is simple. Every day, I show you the seven-day moving average. Every day, that seven-day “window” slides forward. The oldest day drops out, the newest day drops in, and you get a new average.
States sometimes skip data reporting for a few days. Most do, on the weekends. Most will on holidays. Following those non-reporting days, you’ll get a big “lump” of data as they catch up with the backlog of cases that haven’t been reported.
Source: Google. Notes are mine.
As long as those “lumps” come at the same time every week, that’s not a problem. But if you end up with two lumps in a single seven-day period, then the average will jump up. In effect, the state will have crammed 9 days’ worth of cases into a seven-day window.
And so, the question is, did we see a decline today because some great “lump” of data dropped out of the seven-day window? That’s what I would call a data reporting artifact. Or did we see it because we added relatively few new cases today?
For the U.S. as a whole, the data reporting does not look that unusual. The block that passed out of the seven-day moving average today (Tuesday 1/4/2022) was not unremarkable. And the day that got added today (Tuesday 1/11/2022) was a bit lower. And so the average fell.
But if you look at some individual states, you can see that there is a data reporting artifact for several of them. And, unfortunately, there’s a whopping great one for Florida, which is one of our most populous states.
I spotted one for Rhode Island yesterday. They had clearly delayed their “lump” by one day last week, and so gave an absurdly high new case rate for the seven days ending 1/10/2022. That number has now dropped down. Like so:
Here’s how it looks day-by-day, below. They skipped reporting an extra day last week, and that dumped a lot of cases into yesterday’s average. Yesterday’s seven-day window contained not one but two lumps of data. But not today’s average.
Source: Google
Rhode Island is too small to matter much in the U.S. average. But, unfortunately, Florida does matter. And Florida did the same thing. Like so:
Source: Google
And then, if I look a little harder at all the downturns, I find the same for Illinois, Georgia, South Carolina, and several others. More than enough to have had an impact on the U.S. averages.
The upshot is that the 39% increase over the past seven days is more-or-less correct. The endpoint of the line is now in the right place, and case growth is slowing. But the finding of an absolute decline in cases today is an artifact of many states taking a reporting holiday on Monday 1/3/2022, with no matching data reporting holiday on Monday 1/10/2022.
If I had to sketch in my best estimate of the true trend, by eye, here it is. We’ve had a significant slowing in the rate of growth, but by eye, we’re not at the peak yet.
