Илья - Reading log
[Recent Entries][Archive][Friends][User Info]
Nate Silver, The Signal and the Noise (2012). Nate Silver is a statistician and a poker player who is famous for accurately predicting election results. This fairly disorganized book is a popular overview of statistical prediction. Silver is a Bayesian and anti-frequentist; he thinks that a prediction based only on past data without a theory to provide prior probabilities will be inaccurate because it will overfit the past data. Meteorologists can now predict the landfall of hurricanes within a radius of 100 miles days in advance, which helped with the evacuation of New Orleans before Katrina; before modern supercomputers, the radius was 350 miles. For earthquakes, we are as much in the dark as centuries before. The reason for the difference is that atmosphere is well-understood, and earth's crust isn't. Likewise, we cannot predict which strain of swine flu or bird flu will be deadly: the 2009 flu pandemic killed fewer than 1/1000 as many people as the 1918 flu pandemic, but no one could have told this in advance. There are also chapters on marginally relevant subjects such as computer chess and poker; they distract from the main story. As told in many other books on the late-2000s financial crisis, the rating agencies mispredicted the default rate of mortgages during the housing bubble by a factor of hundreds because they were paid by the issuers, and had no incentive to be accurate.
Tags: books, statistics
Meteorologists can now predict the landfall of hurricanes within a radius of 100 miles... before modern supercomputers, the radius was 350 miles.