This past month brought a deadly example of one of our favorite quotations about data: “Statistics are like political prisoners: torture them long enough and they will say anything.” Normally we are big fans of anyone who uses Excel to solve problems and find insight. But White House Economic Advisor Kevin Hassett’s recent use of the trendline and display functions in Excel is a perfect example of how not to use it.
Sadly, his blunder has had deadly consequences.
Hassett apparently took daily data on U.S. COVID-19 deaths and used a standard function in Excel to create a chart. That’s it up above. The cubit fit curve that Excel automatically fitted to the data (the red dotted line) implied that COVID-19 deaths would fall to zero by about May 25th. Which is considerably rosier than the predictions of models used by Johns Hopkins, the University of Washington, and The New York Times.
After Hassett showed this falsely favorable forecast to the White House, it promptly abandoned all other models and adopted Hassett’s. The better to bolster the President’s push to reduce COVID-19 restrictions as soon as possible, and a classic example of confirmation bias. Which helps to explain why there is so much confusion and uncertainty about lifting stay-at-home orders in the U.S. The White House is making decisions based on an erroneous “model,” while many governors are using actual models developed by actual scientists.
(For more on this administration’s magical measurement thinking, see “President Donald Trump Is the Measurement Menace of the Month.”)
As will be obvious by the time you read this, the U.S. death rate will not be at zero by Memorial Day, as Hassett predicted. The lesson for all of us is that, although Excel is a wonderful tool for analyzing measurement data, make sure you understand the consequences and facts behind your data. And you know the difference between designing a mathematical model and fitting a curve to some data.
As Jason Furman, the former head of the CEA, tweeted in response to Hassett’s curve, “Faux expertise is even worse than ignorance… To the degree this crowded out input from genuine experts in the conversation and confused other participants into thinking that CEA or other economists had any sort of real or valid model of the epidemic it is really & truly terrible.”
So, for blatantly ignoring science, facts, and the most basic elements of statistics, we name Kevin Hassett our Measurement Menace of the Month.
Runner-up prize goes to Georgia’s data-shuffling COVID-19 chart
We can’t ignore an almost equally maniacal use of COVID-19 data brought to our attention by our friend Josh Bernoff in his Without Bullshit blog. Here’s a chart of COVID-19 cases by county, published by the State of Georgia:
At first glance, it looks like the trend for COVID-19 cases is down — but look carefully at the dates along the bottom. The staffer that created this chart moved the data around to make it look like the trends were getting better. Here is a detail of the bottom of the chart:
Nowhere in anyone’s use of charts ever does May 1st come before April 30th, or does May 3rd come after May 5th.
Normally our Measurement Menaces of the Month make mistakes with minor consequences. But our current COVID-19 crisis means that proper attention to the use of data is a matter of life and death. As Josh Bernoff says, if you make decisions based on either this chart or on Kevin Hassett’s “model,” people will die. ∞