Kahneman, why do data scientists suck at COVID-19?

On the cognitive errors of the numerate

Here is a thing…

Hope to see you on Friday.


I highly recommend this New Yorker Radio Hour interview with Nobel prize-winning psychologist Daniel Kahneman (Annie Duke endorsement is a bonus).

That interview is about the general public’s inability to grasp exponential growth.  Kahneman asked a question in another article a few years back:

A lily-pad on a lake doubles in size every day. On day 48 it covers the whole pond.  On what day did it cover half the lake?

According to Kahneman, most people answer 24.  The correct answer is 47.

I expect people who work in quantitative modeling (statisticians, data scientists, economists, ML experts, etc.) to be able to get the right answer. Years of exams and job interview questions should have taught them to recognize problems where gut reactions are not to be trusted.

But I've observed errors of thinking within this community as well.

Why Smart People Are Stupid
  1. Why weren't the Harvard physicists and statisticians I work with at the startup I work at freaking out until the governor sounded the alarm?

  2. Why were prominent data scientists on Twitter criticizing Nassim Taleb when he was saying back in January that its good to panic and panic early?

  3. Why are statisticians trying to fit datasets on COVID-19 data when there are clear sampling issues due to lack of widespread testing. Junk in, junk out.

  4. Why do data modelers struggle with modeling interventions? They fear that if they say things will be better by June, then policy-makers might start relaxing social-distancing early, resulting in a resurgence of the illness. But causal models (models that explicitly represent interventions and their effects) exist. Why not use them?

  5. Why weren't all the epidemiologists, who know a thing or two about stats and might even have experience with diseases like Ebola, wearing masks in public back in January? Shouldn’t they be beyond feeling awkward by violating social norms?

Kahneman had something to say about this in a 2012 New Yorker article that is worth revisiting.

Free Minicourse on Causal Modeling in ML