Here are a few plots that compare the predictions of the model against the actual trajectory of the pandemic so far. In these plots, we stopped calibrating the model on April 15 and let it run its course in late April and May period, during which NYC was in NY-Pause. The model seems to be doing a decent job of predicting the course of the pandemic. As the City reopens, we will need to re-estimate the R-naught value corresponding to the current level of social distancing norms. It will be interesting to see what we will predict for what is to come in late late June and the rest of the summer. All plots focus on NYC -- 5 boroughs.
 |
| Cumulative deaths |
Note that the historical data reports both deaths from Covid and probable deaths from Covid.
 |
| Cumulative infected |
The historical data only provides the known infected. One nice thing about having a model is that we can see the number of people infected and know that they are infected, as well as then umber of people infected and do not know that they are infected. The green plot shows the total number of infected people who do and do not know that they are actually infected. The predicted fraction of the infected population appears to be lower than what is anecdotally discussed.
 |
| Daily hospitalizations |
 |
| Daily deaths |
We are a bit off in terms of daily deaths, but we over-predict earlier and later deaths, whereas we under-predict peak deaths, as a result of which cumulative deaths is captured by our model reasonably well. This could be because we do not explicitly model hospital capacities.
 |
| Daily tested positive |
The number of people tested positive comes out endogenously through the inner workings of the model. In other words, we do not have something like an exogenously-specified probability of testing positive. We fixed the testing capacity at 30,000 starting on June 1. For the other days, the number of people tested is what is in the historical data.
No comments:
Post a Comment