COVID Forecasts are Underpredicting

I used to think the forecasts were junk. In the early days of COVID, the April prediction estimated 2 million deaths, which caused a lot of people to disbelieve the "models".

But we're in a period now where the models are actually reasonably accurate in predicting the next month's worth of deaths, and in some cases underpredict.

Below I show current US deaths (as of 8/28) compared to 7/13 predictions. The light green line represents the 7-day moving average of deaths and the light green dots are the daily reported deaths. The dark green dots and dotted line represent the daily predicted deaths as of the Saturday in each week. The shaded green area represents the 95% confidence for prediction. As you can see, the 7-day moving average deaths lie within the 95% confidence intervals. Yet the predictions, generally, are underpredicting.

News about COVID cases usually focuses on either new cases or death numbers without giving the full picture. I've spent some time to compile data across multiples sites to create charts that show 1) the relationship between the 14-day lag of new cases and deaths, 2) actual vs. forecasted deaths, 3) Rt estimates, and 4) current reported case fatality rate (total deaths/positive cases).

Takeaways? The 14-day lag of new cases correlates pretty well with current deaths. It's sobering that it's expected that we'll have 18k more deaths in the next month.

This chart is updated daily at the URL: https://bryanwhiting.github.io/covid19, which contains more information about how the chart is made and how to interpret it. There's a lot of information to consider, I'm attempting to compile it into one, readable view.

What can you do? Let's be kind to each other, be understanding that the data are sobering, and be considerate by wearing a mask and practicing physical distancing. And spread the news: the forecasts are now underpredicting!

If you work with data and want to help build out the site, it's on GitHub: https://github.com/bryanwhiting/covid19.