Problem: Let $X_i \sim Pois(\lambda)$. Find the UMVUE of the probability that $X_1$ is odd.

My attempt:

I don't think there's any obvious unbiased estimator to use conditioning. So instead I write

$$P(X_1 \text{ is odd}) = \sum_{k=0}^\infty \frac{\lambda^{2k+1}}{(2k+1)!}e^{-\lambda}$$

The complete sufficient statistic is $T(X)= \sum_{i=1}^n X_i \sim Pois(n\lambda)$, so an unbiased estimator must satisfy

$$\sum_{k=0}^\infty \delta(k)\frac{(n\lambda)^k}{k!}e^{-n\lambda} =\sum_{k=0}^\infty \frac{\lambda^{2k+1}}{(2k+1)!}e^{-\lambda}$$

But here I am stuck. I know I need to write out the power series of each exponential, but then I get left with the product of two sums which does not seem very helpful.

$$\sum_{k=0}^\infty \delta(k)\frac{(n\lambda)^k}{k!}\sum_{j=0}^\infty \frac{\lambda^j}{j!}=\sum_{j=0}^\infty \frac{(n\lambda)^j}{j!}\sum_{k=0}^\infty \frac{\lambda^{2k+1}}{(2k+1)!}$$

Using Fubinis I thiink I can write this as

$$\sum_{k=0}^\infty \sum_{j=0}^\infty \delta(k)\frac{(n\lambda)^k}{k!} \frac{\lambda^j}{j!}=\sum_{k=0}^\infty\sum_{j=0}^\infty \frac{(n\lambda)^k}{k!} \frac{\lambda^{2j+1}}{(2j+1)!}$$

This seems to suggest $\delta(t)$ is of the form $\delta(t) = \frac{t!}{(2t+1)!}$ when $t$ is odd, and 0 otherwise?

I am not confident with my attempt at all.


We have the exact expression, verifiable using the power series expansion of $e^{\lambda}$: $$\sum_{k=0}^\infty \frac{\lambda^{2k+1}}{(2k+1)!}=\frac{1}{2}(e^{\lambda}-e^{-\lambda})$$

So that reduces the probability to $$P(X_1=\text{odd})=\frac{1}{2}(1-e^{-2\lambda})=g(\lambda)\,,\text{ say }$$

Since $T\sim \mathsf{Poi}(n\lambda)$ we have $E(a^T)=e^{n\lambda(a-1)}$. This equals $e^{-2\lambda}$ for $a=1-\frac{2}{n}$.

So UMVUE of $e^{-2\lambda}$ based on a sample of $n$ observations is $$h(T)=\left(1-\frac{2}{n}\right)^T$$

This means UMVUE of $g(\lambda)$ is $$\frac{1}{2}\left[1-h(T)\right]=\frac{1}{2}\left[1-\left(1-\frac{2}{n}\right)^T\right]$$

  • $\begingroup$ $T=\sum X_i$ to clarify. $\endgroup$ – StubbornAtom May 24 at 9:53
  • $\begingroup$ Oh that makes so much more sense. Thank you! $\endgroup$ – Xiaomi May 24 at 10:05

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.