COVID Winter Predictions; Delta+/AY.4.2; Prediction Follies, Part 1

We’re going to undermine. Grab your shovel so we can dig under the modeling establishment.

(NPR 10/22) People wonder if they should keep calm and carry on in the face of delta plus variant,  and

(NPR 9/22) Is The Worst Over? Models Predict A Steady Decline In COVID Cases Through March. This is the product of the COVID-19 Scenario Modeling Hub, who apply techniques of data fusion to provide hopefully better predictions than any one study. As of 11/23/21,  their predictions of cases/week for 3/05/22 are:

  • Childhood Vaccination, No Variant: 63,383.61,  approx 9000/day
  • No Childhood Vaccination, No Variant: 87,682.39, approx 12500/day
  • Childhood Vaccination, New Variant: 329,215.58, approx 40,000/day
  • No Childhood Vaccination, New Variant: 467,507.85, approx 67,000 day.

Numbers like 63,383.61, implying accuracy in great excess of actual, would dock a student points in a physics exam.  These are not real numbers; two digits of precision are more than enough. Let’s call it a culture clash.

This is one of those occasions where it is constructive to undermine. To allow the predictions to remain unchallenged invites possible public health whiplash, as happened with (CNN, 2021/05/14) The CDC says masks are no longer needed inside or out if you’re fully vaccinated. By now, we should realize that COVID has more fake-out moves than the Harlem Globetrotters.

The predictive techniques of epidemiology were developed, tested, and refined on familiar diseases:

  • Formerly endemic diseases of childhood: measles, mumps, chickenpox,  pertussis, diphtheria, polio.
  • Epidemic/endemic respiratory infections, primarily influenza, which has  known rhythms spanning decades.
  • Tuberculosis, which remains a slow moving, almost silent, almost indolent scourge of mankind, with prediction  horizons of years, not weeks or months.
  • Sexually transmitted diseases, formerly called “social diseases”: syphilis, chlamydia, HIV/AIDS.
  • Endemic common cold, caused by multiple  virus families, 200 viruses in total.

The prediction problem for each of the above has a defining feature:

  • Childhood diseases. Before COVID, throughout the U.S., the environment of elementary public school education  changed little year-on-year. U.S. children experienced an environment homogeneous compared to adults. Hence measles R_o is a stable number.
  • Influenza is far more livable than COVID. Except for 1918, epidemics and pandemics have not caused large scale behavior modification.  The defining feature is resemblance of prevailing strains to previous ones.  This single feature is enough to make prediction accuracy poor.
  • New tuberculosis infections occur in about 1% of the world population/year. This means that, except for clusters with special features, such as prisons, prediction can be as simple as extrapolation.
  • Sexually transmitted diseases, in the U.S. with  a few exceptions, have a single voluntary factor. This makes them amenable to sociological study.
  • The endemic common cold, predictably present, is the future of COVID. The arrival of that future cannot be predicted.

Chaos. These simplicities are missing with COVID, which modifies individual behavior unpredictably; people wear masks or don’t, go to parties and bars, or don’t; take off their masks to chat, or don’t; hide in their abodes or eat out. They adhere, scoff, and change their minds. They respond to the weather, the season, holiday, or what their guru or political god, or media told them to do. They veer between fear,  confidence, and affront.  With this variation, the Black Swan event is the norm. Steady-as-you go is rare.

Before COVID, how close did you stand while conversing? From Culture Crossing Guide, Israel:

Israelis usually stand close to one another while talking. One to two feet is normal. It can be considered rude to back up or away from someone while they are speaking. People speak at closer than an arms distance. They may touch while speaking,

Personal space by country: (WAPO) What ‘personal space’ looks like around the world.  Culture counts.

If a vaccine is, say, 95% effective in a culture with one assumption of personal space, can it be assumed to be as effective with a smaller space? Quoting from (CNN) Boo-Boo: How safe is it for vaccinated people to return to in-person work? An expert weighs in,

  • Type A: 8 college students jammed into a payphone booth. Everybody in every group gets Delta.
  • Type B: 8 employees doing phone sales out of a 10×10 room in a converted residence. 9 out of 10 Type B groups (small room)  get 6 or more Delta cases.
  • Type C:  Skeleton crew of 8 in a large ventilated newsroom with forced air HEPA filtration. 3 out of 10 Type C groups (newsroom) get 3 or more Delta cases.

Personal space equates to proximity, which relates to exposure threshold in a way not known with any precision. Now suppose, unknown to  disease modelers, our personal space has expanded with social distancing. Since this opposes cultural tradition, it will reverse, around Thanksgiving, Christmas, New Year, and whenever the present looks bright. Israelis, with their close conviviality, may have given us a glimpse: (CNBC) Israel doubles down on booster shots as daily Covid cases set new record.

Waning immunity may not be the whole story; it may combine with dynamically changing, culturally mediated exposure to cause vaccine failure.

Takeaway: COVID-19 is a unique combination of airborne transmission and social factors that have resulted in:

  • Modeling errors, due to complex social factors, that are impossible to rectify.
  • Chaos, where the Black Swan drives the epidemic, yet defies statistics.
  • Models which have no track records beyond very short time horizons.
  • Delta+/AY.4.2 left as a question, which should tell you something: We have to see what it does before we can tell you what it will do.

Next, a little napkin calculation.


















Leave a Reply