2
$\begingroup$

I have $n$ Poisson random variables $X_i$ each sampled from an independent Poisson distribution parametrized by $\lambda_i$. I want to compute $P(X_1+X_2+...+X_n=z|X_1+X_3=y)$. The distribution of the unconditional sum would be a Poisson, but I am not sure how to condition it.

$\endgroup$
3
  • 4
    $\begingroup$ It is equivalent to computing $P(X+Y=z|Y=y)$ when $Y$ is Poisson with parameter $2\lambda_n$ and $X$ is Poisson with parameter $(n-2)\lambda_n$ $\endgroup$
    – Xi'an
    Jul 22 at 17:40
  • $\begingroup$ I don't think so cause if the sum and one of the variables are known, the answer is easy to calculate and deterministic. If there were two free variables like $P(X+Y+Z=a|Y=y)$, then it would be similar to this llc.stat.purdue.edu/2014/41600/notes/prob1805.pdf, and the answer would be $\frac{e^{-\lambda}\lambda^{a-y}}{(a-y)!}$. However, when I have more free variables, the summations get complicated, and I am unsure about how to reduce them to a similar Poisson variable. $\endgroup$
    – ItsmeKe
    Jul 22 at 19:00
  • 2
    $\begingroup$ One proof of the statement by @Xi'an begins by observing that $Y=X_1+X_3$ and $X=X_2+X_4+\cdots+X_n$ have Poisson distributions with parameters $\lambda_1+\lambda_3$ and $\lambda_2+\lambda_4+\cdots+\lambda_n,$ respectively, and are independent. You can go further and rewrite the question as asking for $\Pr(X=z-y\mid Y=y),$ which permits you to write the answer immediately. $\endgroup$
    – whuber
    Jul 22 at 19:44
3
$\begingroup$

The sum of independent Poisson random variables is a Poisson random variable as shown here. So for $Z = X+Y$ where all are Poisson r.v., $P(Z=z)=\frac{\exp^{-\lambda}\lambda^z}{z!}$. Here, $\lambda=\lambda_x+\lambda_y$. If $P(Z)$ was conditioned with $X$, $P(Z|X)$ the solution would be deterministic as the sum of two variables is known and one of the variables is known. As per @whuber and @Xi'an suggested, $P(X_1+X_2+X_3+X_4+...+X_n=z|X_1+X_3=y)=P(X_2+X_4+X_5...+X_n=z-y)$.

So, the conditional can be reduced to an unconditional that is just a sum of r.v. which in turn is just an extension of the attached derivation. Thanks @Xi'an and @whuber, appreciate the help.

$\endgroup$
1
  • 2
    $\begingroup$ That's right: but take some care to specify that when $z\lt y,$ the chance is zero! $\endgroup$
    – whuber
    Jul 22 at 20:49

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.