I need to find an appropriate statistical test (likelihood ratio test, t-test, etc.) on the following: Let $\{X_i;Y_i\}^n_{i=1}$ be an i.i.d. sample of a random vector $(X;Y)$ and assume that $\bigl( \begin{smallmatrix} Y\\ X \end{smallmatrix} \bigr)$~$N$ $\left[\bigl( \begin{smallmatrix} \mu_1\\ \mu_2 \end{smallmatrix} \bigr), \bigl( \begin{smallmatrix} 1 & .5\\ .5 & 1 \end{smallmatrix} \bigr) \right]$. The hypotheses are: $H_0=\mu_1+\mu_2\le 1$; $H_1=\mu_1+\mu_2\gt 1$

By looking at this information, how do I know which test is the most appropriate? Is it because the data is i.i.d. I can simply take a likelihood ratio test? A good explanation on what test is more appropriate than another one would be greatly appreciated. This would definitely clear my mind.

  • 14
    $\begingroup$ Have you noticed that $X+Y\sim N(\mu_1+\mu_2, 3)$ and $X-Y\sim N(\mu_1-\mu_2, 1)$ are uncorrelated and jointly normal, whence they are independent? Thus you can digest your dataset into $\{(X_i+Y_i)\}$, view it as a set of iid realizations of a Normal distribution with known variance and unknown mean, and ask how to compare its mean to zero. This is an elementary textbook problem with a well-known answer (a Z test). $\endgroup$ – whuber Oct 9 '12 at 4:54
  • $\begingroup$ @whuber thanks! I will look into this more carefully. Thanks for the insight. $\endgroup$ – CharlesM Oct 9 '12 at 4:58
  • $\begingroup$ @whuber what I find difficult though is that I face a composite hypothesis testing and I don't know how to set this is up. any suggestion would be welcomed $\endgroup$ – CharlesM Oct 9 '12 at 19:11
  • 1
    $\begingroup$ @whuber it is a previous year practice exam question - so yes not the test itself $\endgroup$ – CharlesM Oct 10 '12 at 17:10
  • 1
    $\begingroup$ @whuber Shouldn't the $X-Y$ distribution have $\mu_1-\mu_2$ as its mean? I realize it doesn't matter for this problem, but it just worries me to see the typo sitting there. $\endgroup$ – Glen_b May 16 '13 at 8:26

Let's investigate the distribution of $Z=X+Y$.

$E[X+Y] = \mu_1 + \mu_2$


$var(Z) = var(X+Y) = var(X) + var(Y) + 2Cov(X,Y)$ which equals to 3 in your case.

What remains is testing $H_0: Z < 1$ which can be done with the usual t-test.

Hope this helps.


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.