One of the biggest mysteries I see is Trump – and some experts – predicting 60,000 (er, 70,000) COVID deaths and setting the goal post there. This seems to tragically understate what’s going to happen, and like a horrible tactical mistake from a PR standpoint.
Oh, I know where it comes from – he’s looking at the IHME COVID model, shown here:
And seeing a 72,000 number. He then simplistically thinks that’s the projection for the pandemic and moves the goalposts to whatever it says.
He’d better put wheels on those goal posts.
That model is just for the current wave, assuming we keep up our current distancing regime forever. Take a look at the daily (above is cumulative) death projections. It projects that we experience our last covid death in early June, and then we’re done, never to experience another.
The modelers at IHME would never claim this is a reasonable prediction, for a number of reasons. Chief among them will be that we cannot possibly hope to keep up our current social distancing until the virus is stamped out. But that’s not what’s happening. Take a look at the IHME model for Georgia:
The model assumes a stay at home order, businesses, closed and that it’s not safe to open up until June 27. But they’re opening up now. The virus is nowhere near defeated, its circulating among the population that are about to start circulating among each other. This is happening all over the country, since people are confusing deaths starting to trend down with defeating the virus. Look for future IHME model updates to start extending the chart and adding to the projected death count.
I’m going to argue that death count projection is way off. Let’s do some napkin math to prove it. Take a look at daily COVID deaths in the United States:
Does it look like it’s decisively moving down? I don’t think so either. This is especially true because, as I’ve argued in another post, this trend is the result of adding together multiple states. Until recently, growth was driven primarily by New York and New Jersey. Now they’ve gotten their trend under control, so they’ve been driving the decrease in deaths since late April. But now the rural states are picking up the ball and running, driving the recent sustained death rate around 2000 per day. So it’s likely we’ll see this death rate continue for some time.
That’s 2000 dead per day. 60,000 per month if we keep as locked down as we are.
But let’s imagine that doesn’t happen, and that the down trend since 18 April continues. What does that look like?
Using admittedly simple minded linear extrapolation (this is an extremely rosy projection), over the last 14 days our death rate has dropped by 1024 from 3052 on April 18 to 2028 on May 3 – 73 per day. Let’s imagine that continues linearly until we hit zero:
That gets us to a total not of 70,000, but 90,000 deaths. That’s bad enough, but a more realistic projection would be we’d continue to drop by the same percentage per day. The average drop since our peak has been 2.33% per day, so applying that forward, we get this:
That’s 142k Total US deaths. Again, I think this projection is optimistic in several ways:
- It assumes we continue on our current trend and don’t prematurely open up any regions, causing a new exponential spike.
- It assumes the current trends continue as a uniform distribution, which we know isn’t happening and rural states are now ramping up, not down.
I want to be clear; these models can barely be called models – they are just extrapolations on current trends. The IHME guys could build vastly better models by tweaking their assumptions around social distancing and extending the curve out through January. I’m certain they have.
But having played with the numbers, my gut says 142k is more of a lowball estimate and a lower bound on my “error bars”. Reality could be much higher as states learn, one by one, how bad exponential growth can get. Remember we have yet to see any reasonably populous state really let this thing rip for more than a few weeks before they shut things down, which makes all modeling assumptions imprecise. Georgia, Florida, and others are volunteering to help us refine our death parameters.