What to Look for in the Next Run of the Minnesota Model

Minnesota has a pretty good, if quirky, model for epidemic impacts.   It has some serious defects in the parameter values it is using.  We are promised a new version of the model this week, so we will see if those issues have been fixed.  You always have to be suspicious about political interference in these matters, since these models have been used as cover by our Governor, and politicians nationally, to justify extreme shutdowns, with all the consequent damage.  If you understand the Minnesota model, and you are tracking developments in the epidemic, such as the true number of cases, the percent that are asymptomatic or mild, and the rate of hospitalization, you would know that the model will have to show a very sharply reduced projection of deaths.  And if they follow through on what they have acknowledged is a very serious problem–the clustering of cases in nursing homes that inappropriately skews the data if not pulled out; the risk to the general population will appear to be what it is–miniscule.

If you recall the series of posts I wrote on how the Minnesota model works, look especially at that odd “detection rate” term that the modelers had initially set at 1% of all cases being known and then magically shot up to 75% on the day the model began running.  Meanwhile, state officials have continued to say their actually are probably 100 cases for every one that shows up as a positive test result.  I don’t think the ratio is that high, in part because if you applied it to cases from long-term care settings, you have more cases than there are residents in those settings.  But I do believe it is likely there are at least 10 times more cases than positive infection test results show.  Antibody survey after antibody survey and widespread infection testing in certain settings–homeless shelters, prisons, ships–supports this estimate.  If you apply that to the model, and use 10% as the detection rate, that change alone would reduce projected deaths to one-seventh of those in the last run.  That rate is the primary control valve on what the projections look like.  If they don’t change it to reflect reality, you can assume they got pressured to show higher death estimates.