AZGrizFan wrote: ↑Wed Apr 22, 2020 6:46 pm
JohnStOnge wrote: ↑Wed Apr 22, 2020 6:27 pm
No, I am not cherry picking data points. I looked at South Carolina because it appeared that South Carolina was under discussion. And I used the first run of the model because it reasonably would be considered to be the one most subject to error at this point.
The April 1 run of the model did miss on Alabama and Tennessee in terms of cumulative cases at this point. It did not miss on the national number. I am not going to go through all 50 States to see how it did in terms of number of States within the uncertainty intervals and number of States outside of it. However, bear in mind that if the uncertainty intervals are like 95% confidence intervals the expectation is that the true value will be outside of the 95% confidence interval in 1 in each 20 instances over the long term.
What would you suggest as an alternative for trying to anticipate impacts?
Well, considering how shitty their models were (and the decisions that were made BASED on those models), you might as well have thrown darts at a fucking dart board.
First, let me say that you are correct in saying that when it comes by State by State projections the April 1 model run did not do well. I went ahead and looked at it. I compared the State by State model projections for yesterday, April 22, to the Worldometer death counts. In 32 of 51 instances (51 total for 50 states plus DC), the deaths reported by Worldometers are outside of the uncertainty interval for the jurisdiction.
That is tempered somewhat by the fact that the current counts are not final estimates. A number of the cases in which the Worldometer case count is outside of the uncertainty intervals on the low side will be such that the final estimate will be within the interval. However, there were also quite a few misses o the high side and the final estimates will just make the case numbers even higher.
At the same time, it's not like throwing darts at a dart board. If you go through them you can see that the general nature of the situation is basically correct. Like for Nebraska the actual case count as of yesterday was 42 and the April 1 model run interval is 102 through 245. And for New York the actual case count as of yesterday was 20,354 and the April 1 model run interval is 9,594 through 19,478. The model was wrong in both instances in terms of the count being outside of the uncertainty interval. But the basic nature of the situation is consistent with the model in terms of comparing what's been going on in those two States.
There are some instances in which that is not the case. Alabama and Tennessee are two such instances. Tennessee more than Alabama.
Meanwhile on the national level the model did very well. The Worldometers death count for the United States as of yesterday 47,659. The April 1 United States death total projection for as of yesterday is point estimate 50,443 and uncertainty interval 30,142 through 71,915.
I assure you, throwing darts at a dartboard or just guessing would not work as well. And I think you really know that.