An unbiased estimator is a sample function:
[tex]
Z_n = f(X_1, \ldots, X_n)[/tex]
such that, for an
i.i.d. sample with a parameter of the distribution
θ that we are trying to estimate, has the property:
[tex]
\mathrm{E}\left[Z_n \right] = \theta[/tex]
If this does not hold for a finite
n, but is true as [itex]n \rightarrow \infty[/itex], then we say that the estimator is
asymptotically unbiased.
In general, if the function
f is some non-polynomial function, it is very hard to check the bias of the estimator. If, on the other hand, the estimator is a (symmetric) polynomial of degree
p (
pth moment), we may use some rules for the expectation values. For example, the mean:
[tex]
\bar{X}_n \equiv \frac{1}{n} \, \sum_{k = 1}^{n}{X_k}[/tex]
has the property:
[tex]
\mathrm{E} \left[\bar{X}_n \right] = \frac{1}{n} \, \sum_{k = 1}^{n}{\mathrm{E} \left[ X_k \right]} = E \left[ X \right][/tex]
is the unbiased estimator of the mathematical expectation of the random variable
X.