In statistical modeling, reducible error is the uncertainty that may be reduced by discovering the relationship between inputs and outputs or discovering the function $f(x) = y$. There is also some amount of error that is irreducible (often represented by the letter $\epsilon$) that cannot be reduced. Irreducible error is due to measurement error, natural randomness, missing variables, and other factors. In practice, the irreducible error is not known and presents an upper bound on the accuracy of any model.