### US deaths, February 23, 2021

The latest model run was run February 6 using data through midnight, Feb. 5, 2021,
and is shown as a solid red line.
The dashed red line shows the future values of this model run.

**Upper-bound model post-Thanksgiving**

Because of the likelihood of underreporting deaths during the Christmas and New Year's holidays,
I used an *upper-bound* least-squares scoring function for the curve fitting run.
Basically the actual squared error was used if the model under-predicted the data point, but
only 25% of the distance was used if the model over-predicted the data point. This modification
was only used for data after November 16th, to allow for missing data from Thanksgiving on.

This change in scoring causes the fitted curve to skim the top of the data points, and to
ignore the valleys.

The R-squared measure is still reported using standard least-squares (in this case 0.9633).

**Pre-Christmas model**

The pre-Christmas model run on 12/24/20 using data through midnight 12/23/20 is shown
as a green line, solid through the day of the run and dashed going forward.
This model used the standard least-squares measure of fit. Note that the R-squared
values reported are only for the data being fit... I don't update the scores for future
data.

Last updated 23-Feb-2021 by fuzzy@lazytoad.com,
from Lazy Toad Ranch

Web Site contents Copyright © 2000-2021 by Michael Mauldin