Today’s data provided a surprise. Where to start? We just had a second day in a row showing a huge increase in the seven-day moving average of daily new COVID-19 cases per capita.
Do I really want to make anything out of that? Upon reflection, I think I do. One maxim of data analysis is that new information comes in the true surprises. Today’s figure was quite a surprise. The only question is whether it’s true, or merely an artifact of holiday data reporting.
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 12/4/2021, 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.
On the one hand
First, I didn’t expect that, based on last year’s behavior. Last Thanksgiving, the seven-day moving average data had more-or-less returned to trend by this time. And this year, because the just-prior trend was flat-ish, I didn’t expect to see a big jump in the data reported today.
Second, when I line up the first and second pandemic years in their entirety, by eye, there’s now a neat parallel between the two Thanksgiving regions. As if we’re now going to see a trend, moving forward, that’s similar to what we saw last year. As if history is going to repeat itself. Which is more-or-less what I expected, early on, given that the balance between COVID-19 infectiousness and population immunity is about where it was last year.
Third, that view is kind-of consistent with the weather. Viewed this way, the lag (or horizontal gap) between last year’s winter wave and this year’s winter wave stands at just over three weeks. Which is just at the outside of what I might have guessed, based on the much warmer weather we had this year, compared to last year. (I went through all of that in Post #1315, where I came up with a figure of two weeks. Winter is two weeks later this year, compared to last year, in the key Mountain and Midwest regions that led last year’s winter wave).
October of last year, and October of this year, from NOAA. Blue is cold, red is hot. (The November map for this year isn’t out yet.)
And speaking of weather, do you recall the ID/MT/WY anomaly from prior posts? Those states were at the forefront of last year’s winter wave. But this year, they form a sort of “hole” in the tier of states abutting Canada, in terms of new COVID-19 cases.
Turns out, it’s been unseasonably warm in that part of the country. As in, postpone-the-skiing-season warm. As in — as of yesterday — record-high-tempertures warm. Plausibly, their winter wave hasn’t really started this year because their winter hasn’t really started.
But on the other hand …
But, surely some portion of the jump is an artifact of data reporting over the Thanksgiving holiday. And that might be significantly different from last year, because the quality of data reporting deteriorated. The majority of states no longer report new data on the weekends. As of June 2021, I was complaining about it (Post #1171). As of July 2021, I had modified my data processing to “gap fill” the missing weekend data.
Sounds good, but in fact, the “deteriorated quality of data reporting” argument goes in the opposite direction. This last data point is the seven-day average of data reported from Saturday 11/27/2021 through Friday 12/3/2021. Last year, most states would have reported fresh data on that Saturday, and so caught up with their backlogs. This year, more states will have skipped Saturday reporting. For those, we’re still waiting for the full catch-up to have occurred on the following Monday.
In other words, the deterioration of data reporting means that states will “smear” the catch-up cases over a longer time period. It means that we haven’t seen the full extent of those catch-up cases yet.
Finally, it would be a heck of a coincidence if a true pickup in the trend just happened to occur at the same time as these Thanksgiving data reporting anomalies. To be clear, for sure, this isn’t due to people who were infected over Thanksgiving. Those new cases — if any — will show up some time next week, owing to the roughly two-week interval between the time infection occurs and those new cases are fully reported in the seven-day-moving-average data.
A couple of deciders
As I noted in earlier posts, there’s a lot less discretion in the hospitalization data than in the reported-cases data. You might defer a test or even fail to get a test if you’re only slightly sick. States can defer reporting of new cases. But if you’re sick enough to require hospitalization, chances are pretty good that you’re going to get hospitalized, holiday weekend or not.
The CDC hospitalization data show rising rates of new COVID-19 hospitalizations through 12/2/2021.
Source: CDC COVID data tracker accessed 12/4/2021.
Finally, if this is some broad national trend, then I can filter out the data reporting noise by focusing on those states that appear to report true numbers every day. Based on the last seven days, the following states appeared to report “true” data last Saturday and Sunday: NY, PA, OH, NJ, AZ, MD, AL, AR, HI, DE, ND.
Here’s the seven-day moving average data for those states, where you should focus on the end point, not necessarily the trend lines. With the exception of the two southernmost states (AL, HI), I’d say that the trend is up across-the-board.
The ambiguity isn’t going to resolve fully until next week.
But as of this moment, I’d say that maybe the U.S. winter wave has finally started in earnest. That’s based on the — oh, call it a week or two’s worth of normal new case growth — that showed up in the data today.
If so, the winter wave is a little over three weeks behind last year’s wave. That’s at the outer edge of what you could justify based on this year’s warmer weather, compared to last year. It’s also plausible that unseasonably warm weather explains the lack of new case growth in the ID/MT/WY area.
What can I say? Things are only perfectly clear in hindsight. If then. Today’s number was a big surprise that I can’t quite seem to wish away as an artifact of data reporting. And so, when the facts change, I change my mind.
We’ll know more next week.