Cain Marko wrote:Any thoughts on what the new criterion was? In your view was it justified to use this new criterion?
I believe earlier they were counting only deaths in hospitals. Then the gov. of NY made a statement that deaths were being under-counted, because they were missing deaths in homes and nursing homes. So I think those were added. One fine day (15th April to be precise) they (I think) summed up all the deaths in homes and nursing homes up to that point which were deemed to be from COVID, and added them to the tally. That's what caused the spike (4X the deaths as on the previous day). Then every 4 or 5 days after that, they kept adding further sums to the tally as and when the numbers came in (lag in reporting), causing the saw-tooth graph.
I personally don't think this is justified, but that's my opinion, and so irrelevant here. The points to note are:
* That paper which you posted earlier, suggesting 93,000 deaths by Aug. 1 in the USA (which ended up becoming the official estimate for the US) - that death estimate came from a mathematical model, which was trained on data of daily deaths up to around the beginning of April, and was then used to make predictions beyond that date (first they said 60,000, then they updated to 93,000). But the training data was from the old criterion. And the actual daily death data after April 10th or so was already starting to fall far short of the prediction. According to the authors, they had already fully accounted for the lockdown in their prediction, so the shortfall can't be attributed to that.
* Then when the new criterion was adopted, the actual data fell right back on the predicted curve. This is deceptive, like I said earlier - the model was trained using only hospital deaths, so the model was only good for predicting hospital deaths, and it was over-predicting those. If the model is to be compared to hospital+home+nursing home deaths, then the model should have been trained with that data from the beginning. Else the comparison is spurious.
* The peak was already reached in terms of daily deaths around the middle of April. The adoption of the new criterion hid that fact for nearly a month.
That was in response to Mort's post, where he was surprised at the sudden steep drop-off over the last couple of days - that drop-off had already started back in April, is my point - it looks surprisingly steep now, because it was hidden for a long while.