Post #919: Still waiting to see that surge

Source:  Calculated from NY Times Github data repository, data reported through 12/14/2020, or two weeks and four days after Thanksgiving day.  The apparent one-day spike in the South Central states is an artifact of change in Texas data reporting, as explained at the end of Post #915.

Well, we’ve got about three days left for the much-anticipated post-Thanksgiving COVID-19 spike to show up.  Estimates for when cases contracted at Thanksgiving would appear in the data range from T+12 (12/8/2020, my best guess  as to median lag between infection and reporting) to as late as T+21 (12/17/2020, “three weeks”, the longest lag I’ve ever seen any source suggest).

I’m sure not seeing that spike.  Yet.

This isn’t quite a done deal yet.  Possibly that spike will occur.  But with every passing day, the odds are shifting in favor of no spike.

My continued harping on this point is NOT about pointing the finger at public officials who predicted a large post-Thanksgiving spike in COVID-19.  But if we can’t say “we didn’t have a spike”, then we can’t have the conversation about why we didn’t have a spike.

And that’s a shame, given that the holidays are here and/or right around the corner, depending on what you celebrate.   Wouldn’t it be nice to know what we did right, so we could try to do that again?

I have argued that the “spike” of cases following Canadian Thanksgiving is pure fiction.  Never happened.  All that news derived from ludicrously poor data analysis (Post #916).  So we also ought to ask what the Canadians did right, while we ask what we did right.

Did people mostly stay home?  That doesn’t quite seem to square with news report that air travel was 40% of historical norms on the day before Thanksgiving.

Were travelers and socializers ultra-cautious in their behavior, thanks to that warning about a COVID-spike?  Getting tested before traveling, being scrupulous about mask use and distancing, holding dinner outdoors, and so on?  In which case, are we really just looking at the outcome of good public policy (the warning) that averted the very disaster that the public policy warned about?

Or was this really just not that big a risk in the first place?

Beats me.  But it’s now about two weeks to New Year’s Eve, and nobody’s even asking the questions.

The key public health officials who made the prediction of a post-Thanksgiving COVID surge aren’t going to be jumping at the chance to discuss why they were wrong.  They’d be just as happy if this slipped into the past without anybody noticing.  Somebody needs to hold their feet to the (toasty holiday) fire and at least try to get some answers to those questions.

Did we learn anything from this, or not?


Post #915: Still no post-Thanksgiving COVID surge

A few weeks back, it was widely (and reasonably) predicted that Thanksgiving travel and socializing would result in a spike in new COVID-19 cases. 



You don’t see much talk about that, now.  But you should.  Because that spike ought to be showing up any time now.

It’s not perfectly clear when that spike should occur, due to the lag between getting infected and having that infection counted in the data.  It takes about five days, on average, for symptoms to develop.  After that, it depends.  I’ve used a median value of 12 days total lag, both because that’s seemingly reasonable, and because that seems to be the lag from the Chinese experience in Wuhan, adjusted for more rapid American testing.  But I have seen others offer a vague “two to three weeks”, without offering any data analysis to support that.  I have not seen anyone suggesting a lag greater than three weeks.

We now have data through 12/12/2020, or two weeks and two days after Thanksgiving day.  And while the Thanksgiving holiday definitely scrambled the data reporting, the much-anticipated spike in new cases does not seem to be happening.  Yet.

Bear in mind that everybody looks at a seven-day moving average (I address this graphically at the end of the posting).  So even if a spike had begun, that data-averaging practice would blur it on all the charts you commonly see.   So it’s too early to say that we aren’t going to see that Thanksgiving spike.  Yet.  But every day that it fails to appear increases the odds in favor of no spike.

What should we expect to see if there were a Thanksgiving spike?  The signature of a Thanksgiving spike would be simultaneity.  We should expect to see a sharp upturn in cases, well above existing trends, across all the states, all at the same time.  That’s because Thanksgiving is celebrated in all states, all on the same day.  There would be some variation for the speed with which states report cases.  There would be some variation for the fraction of each state’s population that did or did not travel and meet with larger family groups.  Variation due to all the other diverse things going on in those states.  But in general, we ought to see almost everybody’s new cases departing sharply upward from trend, more-or-less all at the same time.

And I certainly don’t see that.  Yet.  What I see looks mostly like continuation of recent existing trends.  But you can look at the data (below) and see for yourself.  These are data through 12/12/2020, or 16 days after Thanksgiving day.

I note that California’s departure from trend occurred well before Thanksgiving-generated cases would have entered their data.  I casually have attributed that to a very dry November in Southern California.

One problem with all the graphs above is that they are seven-day moving averages.  Everybody does that to even out the large daily transient variations in case counts.  But that comes at the expense of having data that respond sluggishly to changes.  My final data point (12/12/2020) actually represents data with an average age of around 12/8/2020 – 12/9/2020.

Averaging across time is not the only way you can get rid of unwanted daily variation in each state’s reported cases.  You can simply average up different sets of states, day by day.  That will also get rid of some random daily fluctuations in counts, but will let any simultaneous changes be visible.  (For example, the “blip” that Thanksgiving put into the data report should remain visible, because those state-level “blips” all occurred on the same day).

So let’s try it that way, showing the daily count of new cases (not the seven-day moving average), but showing the population-weighted average within six regions.  And because the numbers still jump around from day to day, let me also show a three-day moving average of that.  That gives me some smoothing, and the last data point is, on average, just one day older than the raw daily data.

The regions here match the graphs above:  New England/Mid-Atlantic, South Atlantic, South Central, Midwest, Mountain, and Pacific.

Across the regions, the only thing that even looks like a spike, with the right timing, is in the South Central region (gray line)– and that turns out to be an artifact of data reporting.  On December 11, it appears that Texas did what North Dakota (Post #912) had some a few days earlier — added probable cases (presumably based on antigen testing) to their case counts.

On Dec. 11, probable case counts were added, and the dashboard was reconfigured to improve performance. For more information about probable cases, ...

Source:  Texas Department of State Health Services COVID-19 dashboard.

Far as I can tell, the Thanksgiving-related spike in COVID-19 cases hasn’t occurred.  Yet.

This does not mean that the advice to avoid Thanksgiving travel and gatherings was bad advice at the time it was given.  It was completely reasonable advice.  And we may yet see that spike in the next few days.  But if we don’t see it, that’s news, and that needs to be widely reported.

Post #898: Quarantining your college student rationally. Or, should I lock my daughter in her room while I go grocery shopping?

This post is motivated by the need to bring my daughter back from college next week.  What I was wondering is, should we all be wearing masks in the car?  But more generally, what’s the standard protocol, quarantine-wise, for returning college students?

Seems like a fairly straightforward question.  Given that there are going to be millions of college students returning from campus to home in the next few weeks, it seems like there ought to be be some standard answer to that question.

Sure seems like it.  Ought to be.  But there ain’t.  Let me summarize what I found.

When I do the math, under the circumstances I face, the likelihood that my daughter is going to give me a COVID-19 infection is 1-in-30,000.  Over the same period, the likelihood that I would just pick one up, as an average member of the community, is 1-in-93. 

So, to answer the question in the title, it makes no sense to lock up my college-age daughter, while I continue to go grocery shopping. 

Unless that’s to protect her, from the risk of COVID-19 infection that I might be bringing home.

Want do the quick-and-dirty calculation for your own returning college student?  Based on the assumptions below (the student tests negative for COVID-19 and doesn’t pick up an infection while traveling home), the 1-in-X odds of  your student transmitting infection to you, X = 11*campus enrollment / new campus COVID cases in the last two weeks.  If they don’t have a negative COVID-19 test, then replace the factor of 11 with a factor of 3.

Continue reading Post #898: Quarantining your college student rationally. Or, should I lock my daughter in her room while I go grocery shopping?