$X_1, \dots, X_n$ iis geometric: $P(X=x) = (1-p)^{x}p$, $x=0,1,2, \dots$

My Attempt:

$T=\sum_{i=1}^n X_i$ is a sufficient statistic

$W= \begin{cases}1 & X_1= 0,\\ 0 & X_1\neq 0\end{cases}$ W is an unbiased estimator of $p$

To find UMVUE, \begin{align} E[W|T=t] &= \frac{P(X_1 = 0, T=t)}{P(T=t)}\\[5pt] &= \frac{P(X_1 = 0)P(X_1+\cdots +X_n=t)}{P(T=t)}\\[5pt] \end{align} Can somebody please help me expand this step. It's confusing whether I should use $t-1$ or $t-2$ in the combination part of negative binomial pdf.


Since $X_j$ is the number of failures preceding the first success for each $j$, $T=\sum\limits_{j=1}^n X_j$ is the number of failures before the $n$th success. Therefore pmf of $T$ is



\begin{align} E\left[W\mid T=t\right]&=\frac{P\left[X_1=0,\sum\limits_{i=2}^n X_i=t\right]}{P(T=t)} \\&=\frac{P(X_1=0)P\left[\sum\limits_{j=1}^{n-1} X_j=t\right]}{P(T=t)}\qquad,\,\small j=i-1 \end{align}

So the '$t$' remains as it is; it is a matter of '$n$' and '$n-1$' in the pmf of negative binomial.

  • $\begingroup$ Is the answer, $\frac{1}{1+\frac{\sum_{i=1}^n X_i}{n-1}}$ $\endgroup$ – Harry Apr 8 '19 at 11:13
  • $\begingroup$ That is what I get, yes. $\endgroup$ – StubbornAtom Apr 8 '19 at 13:34
  • $\begingroup$ I also tried solving this question using the fact that geometric distribution belongs to exponential family and therefore $\sum_{i=1}^n X_i$ is complete sufficient. Taking \begin{align}E(\sum_{i=1}^n X_i) &= \frac{n(1-p)}{p} \end{align} from this I got the umvue of $p$ as $\frac{1}{X\bar+1}$. Is this method wrong? $\endgroup$ – Harry Apr 9 '19 at 2:33
  • $\begingroup$ That looks like a 'method of moment' estimator, found by equating sample mean with population mean. Do you think UMVUE is the same? $\endgroup$ – StubbornAtom Apr 9 '19 at 20:29
  • 1
    $\begingroup$ @ZeroPancakes No, $E(W)=P(X_1=0)=p$. $\endgroup$ – StubbornAtom May 10 '20 at 6:46

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.