What is Sample Size?
Larger samples give more precise averages, but slowly: the standard error shrinks with √n, so halving it takes four times as many observations. Short runs are noisy: with a 10% claim rate, ten claim-free repairs in a row still happen 34.9% of the time. A large sample does not fix observations that are not independent or a process that changes over time.
Also called n, number of observations
- n = 10 (standard deviation $75 per repair)
- Standard error $23.72
- n = 100
- $7.50
- n = 400
- $3.75
- Four times the sample halves the standard error.
- A short run, such as 20 trades, can rarely tell a small edge from none.
- Ask how many observations sit behind a win rate or an average before trusting it.
Size is not quality
Many observations from a changing process, or with clustered results, can still mislead; size helps only when observations are comparable and close to independent.