Data wonkery, or what happened to that post-Thanksgiving spike?

[dreamshark]

Remember all that high-decibel hand-wringing about the thousands of people who were destined to die in the inevitable "post-Thanksgiving COVID spike?"  Well, has anybody noticed that there was no spike? In the country as a whole the exponential growth curve has continued, which is of course appalling. But a continuation of the same curve is a failure to curb the infection, not a "spike" or a "surge." I've noticed that news reports and public health press releases on the topic tend to change horses in mid-stream, suddenly deciding to highlight the post-Thanksgiving death rate rather than the case rate, or using phrases like, "Post-Thanksgiving, COVID cases continue to soar" which doesn't really say anything. Maybe there were regional spikes, but not where I live. 

In the Upper Midwest the new infection rate has plummeted since Thanksgiving. In Minnesota the case rate peaked about November 20, almost a week before Thanksgiving. Minnesota peaked at almost 9,000 new cases per day (about 7,000 if you look at the 7-day average) and is now at about 3500 less than 4 weeks later. Surprisingly, all of our neighboring states, including the famously feckless Dakotas where you can't tell anybody what to to do, show exactly the same pattern. Look at these graphs,  collected by MPR

You can pretty much skip past the rambling text, which explains at interminable length that the total number of new cases isn't as meaningful as the number of new cases per capita, makes irrelevant points about testing capacity, and so on. The important graphs are "New COVID-19 case rate in upper Midwest, COVID-19 hospitalizations in Upper Midwest, and COVID-19 positivity rate in Upper Midwest. 

If you like to analyze data trends, it's great to see all this info collected in one place where you can identify patterns. For instance, something horrific seems to have happened in Iowa, which clearly lost control of its testing program sometime in October, leading to what looks like a jaw-dropping (but probably misleading) death rate in December. The similarly high death rate in the Dakotas, however, is likely real. They seem to have been doing an adequate job of testing (based on positivity rate), but a terrible job of actually treating the avalanche of cases. 

I predict that in the next week or so there will be much discussion of how the November/December lockdown measures in Minnesota "brought the infection rate under control" as everybody argues about whether to continue those measures. Bear in mind as you read all this that the lockdown measures went into effect on the same day that the infection rate peaked and started its precipitous decline. So these policies could not have caused the decline (although they may well have accelerated it).

Comments

[quadong]

It does seem nearly impossible to connect cause and effect about any COVID statistics. Like, why did cases suddenly start rising in October, in a way that didn't match the weather (or school openings) at all? As you say, why do the current trends seem to have nothing to with supposedly important holidays? Maybe after crunching all the data someone years from now will have figured out what it all meant.

[dreamshark]

The start of the huge spike in October may have been related to college openings, which often occur later in September than K-12 schools. And possibly weakly related to colder weather. But mostly I think the much larger patterns of community spread and herd immunity overpower everything else and once contagion takes hold in a specific area there is little that can be done to contain it until it reaches some threshold of herd immunity and starts a precipitous decline.

It's hard to see these patterns when you're looking at state data, especially for huge states like Texas and California. But in the early days when the virus was striking big cities first, you could see the pattern. Typically 6-8 weeks from screaming peak back to a relatively low baseline of infection.