As winter surge ends, comparing 21/22 to 20/21

Some US counties did better this year than last, some did worse, why?

I should probably have waited a few weeks before writing this post to get the full February numbers, but I’m concerned that if I wait too long between posts, inertia will kick in and I’ll stop posting. Plus, I don’t want y’all to thinking you’ve wasted your hard-earned money 😁.

In any case, here we’ll examine the differences between COVID deaths in the winter of 2020/2021 and winter of 2021/22. All areas of the US have some form of peak in the winter, so this allows us to look across the country. The specific month that COVID peaks, however, differs for different areas of the country and across years (peaks tended to be later in 21/22 vs 20/21). As in all (or nearly all) posts on US counties, COVID death data is sourced from USA Facts. The charts I will include will cover 88 US Counties with total population of roughly 115 million, or ~ 1/3 of the US population. These 88 counties were arrived in the following way:

1. Obtain list of 125 US Counties with largest populations

2. Remove Alaska & Hawaii as geographic outliers

3. Remove Florida, New Jersey, Oklahoma and Nebraska due to inconsistent COVID death reporting at this level of detail

4. Remove counties where peak 21/22 winter COVID deaths are in February as this month is not yet over

5. 88 counties remain

With these 88 counties, comparisons will be made between the peak month of winter 20/21 for each county in 20/21 (Nov ‘20 - Feb ‘21) vs the peak month of winter for each county in 21/22 (Nov ‘21 - Jan ‘22).

We’ll start by looking at how COVID deaths compared during a county’s peak 20/21 winter month and its peak 21/22 winter month:

Within each region, blue circles (left) show the COVID deaths/100k of population by county for that county’s peak 20/21 winter month, while orange circles (right) show the COVID deaths in the peak 21/22 winter month for that county. As we can see, generally speaking you see reduced COVID deaths in 21/22. If you look furthest to the right, for the West, this is particularly significant. This is mostly made up of CA counties and we’ll come back to this in a bit. Note we are using a box and whiskers plot here to try to illustrate the spread of values.

A natural next question is to look at each county to see how much COVID deaths decreased (or increased) from 20/21 to 21/22 in their respective peak months. To get a sense of the distribution of these values, we can create a histogram:

Each square represents one of the counties (colored by state) and they are put in a column to represent the % change in COVID deaths in peak winter month 21/22 vs. peak winter month 20/21. So, for example, if you look at the tallest column (Arapahoe, CO at the top), you’ll see there are 13 (of 88) counties that had a decrease in COVID deaths between 15% & 20%. It’s easy to see the vast majority of counties had decreases (negative values on x-axis). Also noticeable is the block of red squares on the far left. These are all in California. A couple of theories on why California’s decrease might be so large:

1. It may turn out that February is the worst month in terms of COVID deaths for these CA counties. Although currently there were more deaths in Jan ‘22 than Feb ‘22 (until now), it looks to me as though that might not be true by month’s end:

2. CA had a very large peak in the winter of 20/21. Statistically, this provides a larger base number from which to see a decrease, but also it might indicate that many of the most vulnerable passed away last winter and there are less vulnerable folks in the population in danger in 21/22. Both of these points are evident in a chart of CA’s COVID deaths over time:

   

Next, can we explain the difference across counties in how they performed from one winter to the next. The most obvious factor that one might explore is vaccination rates. Since vaccines were essentially not a factor in winter 20/21, does a county’s vaccination rate in winter 21/22 correlate to its relative performance? To begin, we’ll show the same histogram as above, only this time the squares are colored according to the % of their adult population that is fully vaccinated:

Vaccination rates vary from 60% to 95%, with the associated color range from dark red to dark green. While we see some low vax counties (Wayne, MI; Jackson, MO) towards the right side (higher COVID deaths in 21/22 vs. 20/21), we also see high vax counties on the right side (Queens & New York, NY). Similarly, several of the California and Texas counties are medium red (fairly low vaccination), but show strong improvement Year over Year. It would be hard to look at this histogram and make any conclusions about whether vaccination rates in a county would predict the relative number of COVID deaths in winter 21/22 vs. winter 20/21. To get a more statistically sound view of this question, a scatter plot is useful where the x-axis represents a county’s adult vaccination rate, and the y-axis represents the % change in COVID deaths for peak winter month 21/22 vs. peak winter month 20/21:

The best fit regression line is drawn, but this clearly demonstrates no statistically significant relationship (R^2 ~0, p-value=0.8). It does not appear that level of vaccination within a county tells us much about how the number of COVID deaths in that county were likely to change between the last two winters. What does explain the difference in relative COVID deaths from year to year? 🤷‍♂️. If folks have ideas (and ideally relevant data sets) for me to explore, please share in comments.

UPDATE: I think I have at least part of the answer in a new post

Answering my own question

Addendum

I leave myself open to criticism and I must admit to being less than 100% satisfied with this analysis due to the arbitrary nature of using the calendar months. This, for example, might be a problem if in one year a county saw their peak from mid-December to mid-January and in the other year the peak was almost completely within a calendar month. I began to go down the path of looking at the peak 6 weeks of 20/21 vs peak 6 weeks of 21/22 (regardless of what calendar month), but I became frustrated when I discovered that some counties seemed to “dump” a bunch of COVID deaths into a particular week, making it difficult to trust the data. My sense is that the states try to compensate and have the calendar month totals be accurate, but I may be mistaken here. I may choose to revisit this if I become comfortable that I am handling extreme data points correctly.