If you take 12 hundred people, which i think walton did, then basically you're left with virtually no noise. It goes down by the square root of n, where n is e number of observations. This is completely different from when you're looking at errors that are made on frent cases. So if you have an underwriter who overestimates one risk and underestimates the other risk, these errors don't cancel out. He's made two errors, and both of them are independently costly.
Imagine that two doctors in the same city give different diagnoses to identical patients. Now imagine that the same doctor making a different decision depending on whether it is morning or afternoon, or Monday rather than Wednesday. This is an example of noise: variability in judgments that should be identical.
Shermer speaks with Nobel Prize winning psychologist and economist Daniel Kahneman about the detrimental effects of noise and what we can do to reduce both noise and bias, and make better decisions in: medicine, law, economic forecasting, forensic science, bail, child protection, strategy, performance reviews, and personnel selection.