Consider the following experiment, commissioned by the CIA, no less, in the late 1990s. Eight experienced horserace handicappers were given suggestions for 88 different pieces of information that they might use to predict a horse's performance.
The weight carried, how often the horse had placed first in the past, the number of days since its last race, etc. They were then asked to say which five pieces of information were the most important, that is, which they would use if they could only use five, then which 10, which 20 and which 40.
Next, they were given the data from 40 races and asked to predict the winning horse. They were also asked to say how confident they felt in their prediction.