On April 11, I critiqued the University of Washington’s IHME model’s treatment of five similar states in the Upper Midwest. I pointed out that IHME was projecting far higher per capita death rates for Iowa, North Dakota and South Dakota than for Minnesota and Wisconsin, based on zero empirical evidence. I surmised that IHME must be jacking up the projections for those three states because, unlike Minnesota and Wisconsin, they did not have shutdown orders in effect–notwithstanding the unambiguous assertion on the IHME web site that its projections “assume[e] full social distancing through May 2020.”
As of April 11, IHME projected the following fatality totals for the Upper Midwestern states, through August. All of these numbers had been revised on April 10:
South Dakota: 356
North Dakota: 369
On April 13, IHME again revised all of its U.S. projections. For some reason, its Upper Midwestern predictions changed radically. IHME now projects these fatality totals:
South Dakota: 181
North Dakota: 32
I would love to see the alleged mathematical model that, in the space of three days, caused the South Dakota projection to decline by 49% and the North Dakota projection by a whopping 91%. None of these numbers have anything to do with experience, as North Dakota and South Dakota have so far reported (per IHME) 6 deaths apiece.
Maybe someone at IHME read my post and decided to adjust predictions that were obviously unfounded, if not ridiculous. Or maybe the model was tweaked in some way to give radically different results. There is also no apparent reason why IHME increased Minnesota’s death total by 48%. As Scott has pointed out, Minnesota currently has 79 deaths (70 as of April 13, when IHME issued its new numbers) with more than two-thirds occurring in nursing homes and assisted living facilities.
One way or another, public officials should not be basing major policy decisions, affecting millions of people, on the IHME model, or any other model. A model is not evidence, it is a hypothesis. The hypothesis is tested by experience. To my knowledge, no one has come up with a COVID-19 model that has been verified by experience. This is typical of past epidemic modeling, too.