I am working on this problem and am stumped. Can anyone take a look at it?

$X_1, \ldots ,X_n$ are distributed Bernoulli$(p)$ where $p$ is unknown. Consider the parametric function $\tau(p)=(p+(1-p)e^3)^2$.

1.) Find a suitable unbiased estimator of $\tau(p)$.

2.) Since the complete sufficient statistic is $\sum_{i=1}^n X_i$, use the Lehman-Scheffe theorem and evaluate $E_p[T|U=u]$.

There were two ways I looked at it. Immediately apparent is that the MGF of Bernoulli looks a lot like it, taken at $t=3$. But the problem there is that $p$ and $(1-p)$ are switched around. So I wasn't sure where to go from there.

The other way I looked at it is distributing it out, and I got $p^2+2p(1-p)e^3+(1-p)^2e^6$

I could find an unbiased estimator of this, but that conditional expectation looks nasty...could anyone start me out? I really have no clue about this problem.


1 Answer 1


Since the probabilities are flipped you can look instead at $1 - X_i$. An unbiased estimator then could be

$$ T = e^{3(1 - X_1) + 3(1 - X_2)} $$

since we have

\begin{align} \text{E}(T) &= \text{E}(e^{3(1 - X_1) + 3(1 - X_2)}) \\ &= \text{E}(e^{3(1 - X_1)})^2 \\ &= (p + (1 - p) e^3)^2. \end{align}

Finding the conditional expectation is the challenging part but it's not impossible. We just use the fact that conditional on $S = \sum_{i=1}^{n} X_i$ all $\binom{n}{S}$ combinations are equally likely, and consider how they affect $X_1$ and $X_2$.

There are three cases to consider: when both $X_1$ and $X_2$ are zero, when both are one, and when one is zero and the other one. There are $\binom{n-2}{S}$ combinations corresponding to $X_1 = 0$ and $X_2 = 0$, $\binom{n-2}{S-2}$ corresponding to $X_1 = 1$ and $X_2 = 1$, and $2 \binom{n - 2}{S - 1}$ consistent with $X_1 = 1$ and $X_2 = 0$ or $X_1 = 0$ and $X_2 = 1$. Collecting everything together we get

\begin{align} \text{E}(T \mid S) &= \frac{\binom{n-2}{S} e^6 + \binom{n-2}{S-2} + 2 \binom{n - 2}{S - 1} e^3}{\binom{n}{S}} \end{align}

where we set $\binom{n - 2}{k} = 0$ if $k < 0$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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