Great old story from the Economist about a very common statistical error. Cherry picking.
Hospital admission data from Canada shows that Leos are likely to have gastric trouble and Sagittarians are more likely to break their arms. Both results are statistically significant...if your statistical technique ignores the fact that with 24 comparisons 2-3 are likely to be significant at the 95% level due to pure randomness.
I unconsciously resisted absorbing this idea during stats training...probably because I'm usually very keen for the results of my tests to be significant. Yet when one is doing dozens of tests (as I often am) results that appear significant are often just noise.
This example hammered the point home...probably because I am very receptive to the thought that astrology is a vicious scam. Cultural context: astrology in India isn't just harmless fun. The truth is that Leos are no more likely than anyone else to have gastric trouble. And my mom's painful feet are because of poorly designed footwear, not her Virgo birth sign.
No comments:
Post a Comment