2

How to generate an unbiased estimator for $e^{-\lambda}$ in Poisson distribution: $\frac{\lambda^k}{k!}{e^{-\lambda}}$

I tried: $$E[a^x]=\sum_{x=0}^\infty a^x\frac{1}{e^{\lambda}}\frac{\lambda^x}{x!}=\frac{1}{e^{\lambda}}\sum_{x=0}^\infty \frac{(a\lambda)^x}{x!}=e^{a\lambda-\lambda}=e^{\lambda(a-1)}$$ But here I cannot just let a=0. So I have to find other ways.

I prefer a deductive answer instead of a guessed one.

Furthermore, how to generate the UMVUE for it if possible? Thank you.

Tim
  • 108,699
  • 20
  • 212
  • 390
CuteCat
  • 191
  • 9
  • Possible duplicate of [Is this an unbiased estimator?](https://stats.stackexchange.com/questions/185312/is-this-an-unbiased-estimator) – Xi'an Apr 17 '19 at 07:12
  • 1
    Hint: the probability of a poisson variable equal to zero is $e^{-\lambda}$. – B.Liu Apr 17 '19 at 07:15
  • @Xi'an Sorry that doesn't solve my problem. He/She can fit (-2) in that problem, but I cannot fit 0 in my problem. – CuteCat Apr 17 '19 at 10:14
  • @B.Liu Yes. But I am still unsure about what to do next. Sorry~ Could you explain more? – CuteCat Apr 17 '19 at 10:15
  • 1
    Since $P(X_i=0) = e^{-\lambda}$ for $X_i \sim Po(\lambda)$, if you can construct an unbiased estimator for the former, then you automatically get one for the later. Can you construct one using e.g. empirical probability? Do see @Ben's answer below if you need more inspiration. – B.Liu Apr 17 '19 at 10:36
  • 1
    Related: https://stats.stackexchange.com/questions/55377/need-help-finding-umvue-for-a-poisson-distribution – B.Liu Apr 17 '19 at 10:46

1 Answers1

1

Hint: Given observed data $x_1,...,x_n$, consider the estimator of the form:

$$\widehat{e^{-\lambda}} = \sum_{k=0}^\infty w_k \Bigg[ \frac{1}{n} \sum_{i=1}^n \mathbb{I}(x_i = k) \Bigg],$$

where $w_0,w_1,w_2,...$ are a series of estimator weights corresponding to the possible outcomes of a Poisson random variable. You can see that this estimator estimates the target value as a weighted sum of the proportions of values equal to each possible outcome of a Poisson random variable.

Try to find an expression for the expected value of this estimator, and then see if there is any choice of values you could make for $w_0,w_1,w_2,...$ that would lead the expected value of this estimator to be equal to the quantity you are trying to estimate.

Ben
  • 91,027
  • 3
  • 150
  • 376