One of the big criticisms of public health advisors during the pandemic has been the reliance on models that have been inaccurate. While models are inexact by their very nature – they are estimates rather than detailed forecasts – the variation from a forecast of 2 million dead down to 60,000 and then back up can be frustrating. Worse, it breeds distrust of those making the recommendations. Obviously, we need a better model.
For years, I’ve advocated using polling averages rather than obsessing over any single poll. Using averages smooths out the curve and minimizes the effect of outliers that give extremely high or low results. Now, researchers at the University of Massachusetts Amherst are recommending the same strategy for examining Coronavirus models.
The model produced by biostatistician Nicholas Reich and his colleagues takes an “ensemble” of models and combines them into one forecast. Reich’s work builds on similar forecasts that he has created to forecast the effect of influenza epidemics over the past four years, reports NPR, since flu models seem to resemble the spread of COVID-19 better than models based on other diseases. Reich’s model forecasts a death toll of 110,000 by June 6, three weeks away.
With the current US death toll standing at 86,600, Reich’s forecast seems plausible. The US daily deaths seem to be slowly trending down with 1,858 reported yesterday. Multiplying that by 21 days until June 6 would yield a toll of 39,018 deaths if there is no decrease. Adding that number to 86,600 would equal 125,618 deaths without accounting for the declining curve. This would be about 15 percent higher than Reich’s estimate.
The model of models currently includes eight separate forecasts. The IHME and Imperial War College estimates are among the models that Reich samples. These individual estimates range from a high of 120,000 to a low of 103,000.
Model forecasts are not static. The output varies based on new information about the spread of the virus, changed assumptions such as a different levels of social interaction, and adjustments to correct for previous errors.
It should be noted that previous estimates have been on the low side as well as too high. Critics frequently cite problems with the original estimate of 2 million dead, but few mention that a subsequent estimate of 60,000 dead has been far too low.
Reich and his team are working to make the model even more accurate by weighting the various models that are components of his estimate.
“Model accuracy is one thing that we’re tracking and we’re hoping to release some information over the next few weeks,” says Reich. “We’ve been sort of building the car as we’re driving it at 90 miles an hour down the highway. And we’re learning as we go.”
Another problem is that Reich’s model only extends four weeks into the future, partly because that’s as far as some of the component models go. The Coronavirus won’t end that quickly but no one knows whether a second wave of the virus will follow later this summer or fall. The death toll is certain to continue to rise after June 6, but no one knows by how much.