0
$\begingroup$

Suppose $X,Y$ are independent chi-squared distributed random variables with $m,n$ degrees of freedom, $X \sim \chi^2(m)$ and $Y \sim \chi^2(n)$. What is the distribution of

$$ Z = \frac{1}{m} X + \frac{1}{n} Y $$.

Is there a closed formula or an approximation? Is there a generalisation to a sum of more than two terms?

$\endgroup$
1
$\begingroup$

The Chi-squared distribution $\chi^2(\nu)$ is the the Gamma distribution ${\cal G}(\nu/2,1/2)$. Then $\frac{1}{m}X \sim {\cal G}(m/2,m/2)$ and $\frac{1}{n}Y \sim {\cal G}(n/2,n/2)$. These are Gamma distributions with different rate parameters (the second parameter) when $m \neq n$, and there is no closed-form formula (or formula in terms of elementary functions) for the convolution (sum of two independent) of these distributions.

I don't know whether there is a known approximation but I would not be surprised this question has already been adressed here. However, depending on what you exactly want, the Welch-Satterthwaite approximation might be an answer.

$\endgroup$
  • $\begingroup$ It would help if you could clarify whether you apply the "shape-scale" or the "shape-rate" parametrization of the Gamma distribution. $\endgroup$ – Alecos Papadopoulos Apr 21 '15 at 11:06
  • $\begingroup$ @AlecosPapadopoulos I wrote "different rate parameters (the second parameter)" so it should be clear. $\endgroup$ – Stéphane Laurent Apr 21 '15 at 12:16

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.