Read on Twitter
So you know: a thread about the Murray model everyone is talking about.
First: it can be found here:
http://COVID-19.healthdata.org
/1
First: it can be found here:
http://COVID-19.healthdata.org
/1
It’s primary purpose is to help public officials understand were resources need to go - beds, nurses, supplies, etc.
It’s giving info to govts plan for worst case scenario on those issues.
It assumes full social distancing until May 20.
/2
It’s giving info to govts plan for worst case scenario on those issues.
It assumes full social distancing until May 20.
/2
It shows data for the whole US & also for each individual state.
One thing it shows - by state - is what mandatory social distancing measures have been put into effect for each state.
/3
One thing it shows - by state - is what mandatory social distancing measures have been put into effect for each state.
/3
There are three principal charts for the US & for each state:
1. Resources needed;
2. Deaths per day;
3. Total deaths.
/4
1. Resources needed;
2. Deaths per day;
3. Total deaths.
/4
For daily & total deaths, there is a projected number & a range for each day shown in numbers & on a chart.
It’s like a hurricane map- the projected number is like the probable path of the hurricane & the range is like the cone showing the other possible paths for the storm. /5
It’s like a hurricane map- the projected number is like the probable path of the hurricane & the range is like the cone showing the other possible paths for the storm. /5
People have been saying the model is multiples off the actual numbers we’re seeing so I wanted to see for myself.
The chart & projections for resources is very different from the actuals. There’s a significant difference in beds projected as potentially needed & actually used./6
The chart & projections for resources is very different from the actuals. There’s a significant difference in beds projected as potentially needed & actually used./6
That’s a good thing in terms of how overwhelmed our health care system is in actuality versus what was projected. So irl that’s good. Not great for the model or planning perhaps, but as the purpose is planning for the worst case scenario, maybe it’s within tolerance? 
/7
/7
Where the model is NOT multiples off is the death calculations.
For today, it predicted a daily death toll of 2140 within a range of 1387 to 3475. The actual toll through 7:45 pm is 1927.
/8
For today, it predicted a daily death toll of 2140 within a range of 1387 to 3475. The actual toll through 7:45 pm is 1927.
/8
Similarly, the model projects a total death number for today of 14,678 within a range of 13,093 to 16,904. The total death toll as of today at 7:45 pm is 12,758, which is 335 deaths below the range.
/9
/9
To date, it appears that actual deaths are consistently running just slightly fewer than the low end of the projected range (500-800 fewer).
So, the death portions of the model are consistently reasonably accurate at least so far.
/10
So, the death portions of the model are consistently reasonably accurate at least so far.
/10
The model projects peak deaths in the next 9-10 days. Although it projects roughly 81,000 total deaths by August, the vast majority of them are projected to occur in this month - now in April.
/11
/11
So:
Actual deaths Mar 1 = 9.
Actual deaths Apr 1 = 4,765.
Projected for May 1 = 72,058
That’s within a possible range of 46,919 to 115,034 for May 1.
So, low end is roughly 42,000 deaths this month.
/12
Actual deaths Mar 1 = 9.
Actual deaths Apr 1 = 4,765.
Projected for May 1 = 72,058
That’s within a possible range of 46,919 to 115,034 for May 1.
So, low end is roughly 42,000 deaths this month.
/12
I’ve taken down all the projections from now to April 16th, which is when the model projects the peak. So we’ll be able to see tomorrow: how accurate today was & how accurate tomorrow is & how the model changes as we go.
/13
/13
