Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

[Phys-L] pedagogical video on Simpson's paradox



You can get into trouble if you combine data sets that
were obtained under different conditions. This is relevant
to physics and epidemiology (including current events!)
and a hundred other things.

Easy-to-follow 4-minute video that gets the main point
across:

https://youtu.be/ebEkn-BiW5k

This is relevant to current events as follows: It is risky
(possibly worse than useless) to try to understand (much
less extrapolate) US nationwide coronavirus trends. You
have to look at statewide or (better) local data. We
don't have one big outbreak; we have hundreds of smaller
outbreaks, each responding to local conditions.

Local data is noisy, because the numbers are small, so
there is necessarily a lot of Poisson noise. However,
given the choice between noisy data and biased data,
biased in some unknown way, I'll take the noisy data
ever time.