reader anonymous Correlations between suspected death, hospitalization, and ICU are shown here postal Specifically, in the first diff specification, the usage continues:
ΔdeceasedTon = 0.00045 + 0.276Δlhospt-21 + 0.270Δ Facet-14
Adj R2 = 0.07, SER = 0.048, DW = 1.69, N = 529, bold indicates significance at 5% msl, using HAC robust standard errors.
As a criticism, he wrote:
1. Omicron is a very different variant from Delta (or alpha, beta), especially with regard to case-by-case effects. So there is some real uncertainty about training models on early Covid.
2. In particular, the model is trained using only the lags observed in the US (good, since we are expecting the US, but you don’t get insights from other countries that Omicron is sweeping. It seems that this information should influence the Make Bayesian predictions somehow.
3. The recent (DEC) data may be heavily polluted by Delta, but this is changing very quickly to Omicron. So DEC may not drive JAN.
4. Even with the exact same Covid, the population is different.
It turns out that the actual number of deaths (7-day trailing moving average) exceeds the forecast of the first variance norm.
figure 1: Covid deaths, 7-day moving average (black) and log-based prediction OLS regression lagged 7-day moving average of hospitalizations and ICU deaths (red), and log-based first difference (green) (see above) , and the actual 7-day moving average on January 28, as reported by The New York Times on January 29, 2022. Sources: Our World in Data, accessed Jan. 16, 2022, NYT, Jan. 29, 2022, author’s calculations. [Update of Figure 4 in this post]
The average daily death rate of 2529.2 was 8% higher than the point estimate of 2342, but within the 95% prediction interval.
