Wednesday 27 October 2010

Survivor bias

Lasse Lien at Organizations and Markets gives a nice example of Survivor Bias: WW2 ed.
One fascinating anecdote is how these pioneers used data on damage from German air defense fire. The RAF collected large amounts of data on exactly where returning aircraft had received damage. The intuitive recommendation would be to reinforce the aircrafts were the data indicated they took the most damage. However, realizing that they only had data from surviving aircraft, the OR-group under leadership of Patrick Blackett recommended that they reinforce the aircraft in the sections where no damage was recorded in the data.
Nice. What you would really like to know is what damage did the planes that didn't return suffer.

No comments: