Let we have $X_{1},X_{2},..,X_{n}$ as independent and identically distributed random variable from $U(\theta,\theta+1)$. Clearly, the maximum likelihood estimator of such distribution will be $[X_{(n)}-1,X_{(1)}]$. I know that maximum likelihood estimator follows the invariance property. Can any function from these two estimators will be maximum likelihood? Then, why not $\frac{X_{(1)}+X_{(n)}}{2}$ will be maximum likelihood estimator since they are functions of maximum likelihood estimators.


There is no unique MLE. So perhaps it is a matter of semantics whether the invariance property applies to quantities from the interval $[X_{(n)}-1,X_{(1)}]$ and perhaps difficult to say what invariance would mean.

For $n \ge 2,$ your proposal $T_1 = (X_{(1)}+X_{(n)})/2$ is severely biased; $E(T_1) = \theta + \frac 12.$ The midpoint $T_2$ of the interval is unbiased. (I suspect it may be what you meant to propose.) You might want to explore whether $T_2$ has smaller variance than other unbiased linear functions of the interval endpoints?

| cite | improve this answer | |
  • $\begingroup$ Perhpas look at this page. $\endgroup$ – BruceET Jan 14 '18 at 20:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.