0
$\begingroup$

Suppose $X$ has a Poisson distribution with mean (and therefore variance) $\lambda$. Using Excel to explore properties of the distribution of $X^2$ with some small integer values of $\lambda$ I found:

Squared Poisson examples

The values of $E[X]$ are consistent with the formula (which was given in an answer to this question):

$$E[X^2] = \lambda^2 + \lambda$$

Trying to find a formula to fit the values of $Var[X^2]$, I came up with:

$$Var[X^2] = 4\lambda^3+6\lambda^2+\lambda$$

Is this formula generally valid, and if so how can it be proved?

$\endgroup$
1
$\begingroup$

We can use the fact that

$$Var(X^2)= \mathbb{E}[X^4]-\mathbb{E}[X^2]^2$$

You can explicit $\mathbb{E}[X^4]$ either through the definition or using the moment-generating function of a Poisson-distributed random variable. I will use the latter. The moment-generating function $M_X(t)$ of a Poisson-distributed random variable $X$ is

$$M_X(t)=e^{\lambda(e^t-1)}$$

Now, $$\mathbb{E}[X^4]=\frac{d^4}{dt^4}M_X(0)=|\lambda e^{\lambda(e^t-1)+t}(\lambda^3 e^{3t}+6\lambda^2 e^{2t}+7\lambda e^t+1)|_{t=0}=\lambda^4+6\lambda^3+7\lambda^2+\lambda$$

$$\mathbb{E}[X^2]=\frac{d^2}{dt^2}M_X(0)=|\lambda e^{\lambda(e^t-1)+t}(\lambda e^t+1)|_{t=0}=\lambda^2+\lambda \implies \mathbb{E}[X^2]^2=\lambda^4+ 2\lambda^3+\lambda^2$$

We can now determine the variance of $X^2$.

$$Var(X^2)=\mathbb{E}[X^4]-\mathbb{E}[X^2]^2=\lambda^4+6\lambda^3+7\lambda^2+\lambda-\lambda^4-2\lambda^3-\lambda^2=\color{red}{4\lambda^3+6\lambda^2+\lambda}$$

$\endgroup$
  • $\begingroup$ Why did you downvote my answer? $\endgroup$ – Lonidard May 22 '15 at 13:49
  • $\begingroup$ Isn't it basically a rephrasing of the two other answers? $\endgroup$ – Clement C. May 22 '15 at 13:51
  • $\begingroup$ It simply took me a couple of minutes more to write the answer because I wanted to get more in depth than you did. Since this is a quite basic question, I thought showing some more work could have helped the OP. I believe Stackexchange is a place where people help each other, not a rep rush for who answers the question first. $\endgroup$ – Lonidard May 22 '15 at 14:01
  • $\begingroup$ It is helpful, thank you ... but it's also fair to note that the more concise answers by Clement C. and Jack D'Aurizio were posted first. $\endgroup$ – Adam Bailey May 22 '15 at 14:56
  • $\begingroup$ Yes, indeed. I don't see a correlation with downvoting, though. $\endgroup$ – Lonidard May 22 '15 at 15:17
3
$\begingroup$

Hint: you can rewrite $$ \operatorname{Var}[X^2] = \mathbb{E}[X^4] - \mathbb{E}[X^2]^2 $$ and use the known expressions of the (raw) moments of a Poisson-distributed random variable (or recompute them — you already can get $\mathbb{E}[X^2]$ from the variance and the expectation).

$\endgroup$
2
$\begingroup$

The moment generating function of a Poisson random variable $X$ with mean $\lambda$ is given by: $$ f(t) = \mathbb{E}[e^{tX}] = e^{\lambda(e^t-1)},\tag{1}$$ hence:

$$\begin{eqnarray*}\text{Var}[X^2]=\mathbb{E}[X^4]-\mathbb{E}[X^2]^2 &=& f^{(4)}(0)-f''(0)^2\\&=&(\lambda +7 \lambda ^2+6 \lambda ^3+\lambda ^4)-(\lambda+\lambda^2)^2\tag{2}\end{eqnarray*}$$ proving your claim.

$\endgroup$

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.