Narrative Battles and Games with Models, 7/13/20 Data

COVID19-hospitalizationsandprojections-071320-featured

The debate, and the data, around COVID-19 is starting to get interesting again.  States that were harder hit earlier in the pandemic are claiming that they demonstrated leadership, versus states that weren’t hit as hard early on but are now seeing increases, even though the harder-hit states still have higher cases and death rates.  Meanwhile, the debate, as presented in the news media, has fully shifted from the initial panic narrative that we had to shut everything down because we had no idea what the virus would do or how it worked to the current narrative, which strongly implies that martial-law-like policies should remain in place until the threat from the virus has completely gone away.

And then there is the complete silence around the idea that people can decide to balance this thread different, or to bring a different perspective to bear.

Some of this can be explained without recourse to political bias (although probably not all of it).  The news media, for example, has incentive to maintain the impression that the virus continues to be a desperately important story that can’t be ignored for a single day.  Thus, we get WPRI star reporter Ted Nesi tweeting that “the 10% increase in RI COVID hospitalizations since Friday (from 61 to 67) is something to keep a close eye on.”  Nesi says this “adds context to Friday’s briefing where @GovRaimondo shared a forecast warning of the potential for hospitalizations to start rising again.”

Does that really “add context”?  And if so, what is it really?

As you can see in my updated hospital projection chart, below, my model is now, indeed, predicting an increase that looks a lot like the projection from the governor in Nesi’s article, but I don’t make any assumptions at all about people’s future behavior.  Rather, my model at this moment projects an increase for three reasons that may be nothing or may arguably be positive:

  1. The decrease in the rate of new infections over a 14-day span has edged up, from 0.035 to 0.039 over five days.  That isn’t what we want to see, but it’s hardly terrifying.  This means that the number of “active” cases, per my methodology, has increased, which is a part of my hospitalization projection.  Looking more directly, we see that daily new cases have remained under 100 for four full weeks.
  2. The increase in hospitalizations is not because more people are being admitted to the hospital for COVID-19 than before.  It isn’t even because more people are being admitted to the hospital with COVID-19 (whatever reason they’re in the hospital) each day.  Those numbers have remained consistently below 10 for nearly three weeks.  Rather, the increase in hospitalizations results from the facts that (a) fewer people with COVID-19 are being discharged from the hospital each day (probably because they’re in the hospital for something else) and (b) COVID-positive deaths in the hospital have mostly been at none or one per day.  Last week, only four patients with COVID died; if the four days that had zero deaths had each had one, there would obviously be four fewer in the hospital.
  3. The numbers are relatively small, so a percentage increase over a few days can extrapolate to exaggerated increases later.

In summary, if we see an increase in hospitalizations, it may very well be an indication that fewer people are dying, which is good, and that people are in the hospital for other reasons, which isn’t an indication that COVID is overwhelming our resources.

Part of what makes human beings special is our ability to, in a sense, live in the future.  We envision a future and begin to react intellectually and emotionally as if we are already living in that moment.  This makes it a matter of grave importance to be sure that our future projections are accurate and that we know what they actually reflect.  Fomenting a vision of harmful trends may encourage people to pay closer attention to the news, but it will also lead them to respond in ways that aren’t justified.

COVID19-hospitalizationsandprojections-071320

(See here for my original methodology and here for a subsequent modification I made. A thorough explanation of the chart is included in this post.)

Projections versus actuals (date of report).

  • Cases:
    • Projection for 7/13 (from 7/8): 17,360
    • Actual for 7/13: 17,487
    • Projection for 7/14: 17,538
  • Hospitalizations:
    • Projection for 7/13 (from 7/8): 46
    • Actual for 7/13: 67
    • Projection for 7/14: 68
  • Deaths:
    • Projection for 7/13 (from 7/8): 980
    • Actual for 7/13: 984
    • Projection for 7/14: 987


  • Lou

    Who knew Woolery was one of you? I thought he was dead. What’s next, a Secretary of Love Connection cabinet position?

  • Christopher C. Reed

    Whence arises this notion that somehow we will escape a pandemic without widespread infection? When has that ever happened? As the eminent epidemiologist N. Pelosi notes, “Virus gonna do what virus gonna do.”

  • Mario

    Things seemed relatively orderly for a long time, but I don’t understand Tuesday’s numbers at all. We had a couple of days where the new case number was twice as much as I expected, but today’s 102 is more like nine times what I’d expect. It seems too big to be a random variation, I’d very much like it to be an error though. It would be nice if we had some sense of where the numbers were coming from, this could be a case where some particular population was tested in-depth rather than the usual more random sampling the number usually reflects. But it doesn’t look good at all; I’d hate to be alarmist, but trying to fit 102 into the nice, old orderly data I was using suggests that the deaths would go up by over 300 over the next couple of weeks, and I’d like that to not be the case.

    • Justin Katz

      Asked entirely out of interest in your methodology, with absolutely no judgment implied: How do you go from 102 new positive tests in a day to over 300 deaths?

      I’m curious about that even given out-dated age ratios, but as a distinct question, did you account for the fact that new cases are skewing younger?

      • Mario

        I did try to adjust the formula to take into account a lowering fatality rate, which helped (the number I ended up with was ~25% less fatal than my previous estimate), but it did very little to bridge the gap between the last few days and what I was expecting.

        OK, so the number I’m using predicts the number of new cases based on the level of daily testing (which I just estimate using an average of the last two same -day results and one 5% higher — just need to get close enough) and the overall unaware infected population. Specifically, what I try to estimate is the percent of the population captured by a particular level of testing. So, for instance, 1500 tests might capture about 1.6% of the infected population that were available on that day to be found.

        Obviously that depends on the size of the unaware infected population, which I absolutely do not know. But, if you could estimate the overall fatality, and you know the number of people who have died, you can work backwards and figure out how many people were infected two weeks ago based on the total number of deaths today, more or less (after you try to take out the people testing positive and the people leaving the infectious stage). I think you can see by now why I really don’t have a ton of confidence in my conclusions, they just look so nice sometimes.

        Anyway, the problem with 102 is that if the infected population keeps dropping the way it seems, I have this as representing 8.6% of today’s remaining infected population, which is enormous, on just 1400 tests. The easier answer is that the infected population is a lot higher today than I thought it was, which means that the deaths two weeks from now would have to be a lot higher to account for it in retrospect. So where I have just over 1000 deaths on July 28th, it would have to be something like 1150-1400 deaths to bring the new numbers fully back in line with the existing model (obviously you can make that smaller by assuming that the model was already somewhat wrong in the appropriate direction, but that’s cheating) . I just said three hundred as an estimate, it really depends on if we are lucky and start seeing some backfill deaths being revised in, or if all of this is to come. If there is going to be a death increase, but it already started and we just aren’t seeing it, it’s less of a problem than if this is all increases yet to come.

        The last time I had the population as high as 102 would suggest was June 10th when we had about 4 deaths per day, more or less, and March 30th which is not really comparable. My real concern is how, if it’s that high, it got to be that way, because things were doing so well otherwise, it looks like a sudden explosion (it would have been June 30th/July 1st, not quite the holiday).

        Obviously, I would like to point out how enormously speculative everything I’m saying is; please don’t assume this is anything more than an internet rando with too much time on his hands. It’s easy enough to go back and see how little I’m right.