Darkest Winter

Journal of a Pandemic

Vol 1: April to July | Vol 2: August to September | Vol 3: October to December | Vol 1-3: 2020 | Vol 4: 2021 Endgame (pdf)
  • Stats
  • Methodology
Get state         Get county (counties ordered by new cases)

  • 2021 180-day projection
  • 2020 180-day projection
  • Summer 2020 projections
  • Spring 2020 projections
  • New case growth rate

  • Endgame
  • 2021 projections
  • Summer 2020 projections

The endgame falls into roughly 3 scenarios. First, the IHME assumption of a 3-4x understatement means half the country has been exposed to the virus and transmission is not only getting harder for the virus at the same time immunizations are underway. Second, the CDC estimate of 8x undercounting is more accurate, meaning we are clearly due for a dive as herd immunity is rapidly realized. As a final scenario – the pessimistic one – is that actual cases reflect the reporting numbers. In that final case, the immunizations will pull us out of the pandemic and not likely before the shots are given to well over half the population.

All scenarios assume that the 1,000,000 2nd vaccinations per day is already in progress.


  • Spring 2020 projections
  • Summer 2020 projections
  • Sept 2020 projections

Historical notes

  • March 19, 2020:
    • California orders lock-down.
  • March 17, 2020:
    • 37 states have closed their public schools.
    • White House advises people not to gather in groups of more than 10.
    • Defense Department gives 5 million face masks to health officials.
  • March 13, 2020: U.S. government prohibited air travel by non-U.S. citizens to the U.S. from Europe.
  • March 12, 2020: King County (Washington) limited public events to 250 attendees, and closing public schools.
  • On March 11, 2020: New York quarantined New Rochelle, NY.

These forecasts are based solely on reported JHU-CSSE data as of the date of the model, using the latest growth rates and derivatives -- starting with an 7-day averaging of new cases -- to project into the future. No epidemiological, political, or behavioral assumptions or models are incorporated, other than that the current observations as represented in the data are valid into the future with the exception that acceleration/decelerations in daily rates are capped at 2%. Thus, the forecasts do not attempt to account for possible changes in human behavior such as tightening or relaxing discipline, improved testing, or more accurate reporting. Those changes might be inferred in deviations of the actuals over time from the models.

For counties with low numbers of new cases (generally, below 5 per day), the projections are highly suspect as small changes in cases have an oversized impact on the growth rates. As well, counties with suddenly higher numbers of new cases will have accelerations over-estimated.

The method is as follows:

  • Total confirmed cases data is collected from JHU-CSSE daily.
  • New cases are computed as today's confirmed cases less yesterday's confirmed cases.
  • For each country, U.S. state, and U.S. county, the backward-looking 7-day average for new cases is computed.
  • The change in new cases (our first derivative) is computed as today's averaged new cases less yesterday's averaged new cases.
  • The change in the first derivative (our second derivative) provides the acceleration (or deceleration) of the change in new cases, and is computed as today's change in average new cases less yesterday's change in averaged new cases.
  • The model is built by taking the JHU-CSSE data on the date of the model. Then, for each day after the date of the model, multiply the total cases of the previous day by the first derivative and multiply that by the second derivative.
  • The first derivative for subsequent days is the first derivative of the previous day multiplied by the second derivative. The second derivative is not changed.

For additional details see the Methodology tab.

 

Copyright 2020 by Depth.Network, Inc. Reasonable non-profit use acceptable with attribution to Depth.Network and a link to http://www.depth.network.