Let $X_1, X_2, \cdots, X_n$ be an identical and independently distributed sample from $N(\mu, \sigma^2)$, define:

$$D = \frac{1}{t}\left[\overline{X} + \frac{1-\rho}{2} S^2\right]$$


$t$ and $p$ are fixed constants

$\overline{X} = \frac{1}{n}\sum_{i=1}^n X_i$, i.e, the sample mean

$S^2 = \frac{1}{n} \sum_{i=1}^n \left(X_i-\overline{X}\right)^2$, i.e, the biased sample variance

Then what is the sampling distribution of $D$?

I know that $\overline{X} \sim N\left(\mu, \frac{\sigma^2}{n}\right)$ and $n\frac{S^2}{\sigma^2} \sim \chi^2(n-1)$, but how do I derive the sampling distribution of $D$?

  • 1
    $\begingroup$ The trick to solving this is a special known result that $\overline{X}$ and $S^2$ are independent (even though both are based on the same $X_i$~$N(\mu, \sigma^2)$), so your problem essentially reduces to finding the sum of a Normal random variable and a Chisquared random variable. That is the upside. The downside is that finding the sum of a Normal and a Chisquared may be much more tricky than it sounds ... $\endgroup$
    – wolfies
    Commented Nov 22, 2013 at 17:16
  • $\begingroup$ What do you want to use the result for/what kind of an answer do you need? $\endgroup$
    – Glen_b
    Commented Nov 22, 2013 at 22:01
  • $\begingroup$ @wolfies haha yes, I tried using mgf's to no avail... it really looks much more difficult than it seems. Also since there is a constant term in front of the chi squared random variable, wouldn't that make the entire term a Gamma distributed random variable? $\endgroup$
    – TeTs
    Commented Nov 23, 2013 at 5:34
  • 1
    $\begingroup$ @Glen_b This is part of a bigger problem which (ideally) requires a formal proof of the sampling distribution of $D$. This is because I would like to create test statistics for $D$ and conduct hypothesis testing. $\endgroup$
    – TeTs
    Commented Nov 23, 2013 at 5:57

1 Answer 1


I would go with Monte Carlo simulation. Pretty straightforward and less chance of mistakes. There may be some slick math-stats solution too, but Monte Carlo is probably faster.

  • $\begingroup$ Thanks for the suggestion. I might actually try that first just to see the results. To clarify, do I just generate random values according to the distribution of $\overline{X} \sim N\left(\mu, \frac{\sigma^2}{n}\right)$ and $n\frac{S^2}{\sigma^2} \sim \chi^2(n-1)$, then substitute back into the expression to calculate the corresponding value for $D$, then apply a kernel density estimation technique to estimate the distribution for $D$? $\endgroup$
    – TeTs
    Commented Nov 23, 2013 at 5:41
  • $\begingroup$ Yea that would work fine $\endgroup$
    – user31668
    Commented Nov 23, 2013 at 15:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.