We are given a random sample of size $n$ and let the corresponding order statistics are as follows: $$Y_1 \leq Y_2 \leq Y_3.......Y_{n-1}\leq Y_n$$

The distribution of the sample median for $n$ even is required, so if $n=2k$ the sample median is given as : $\dfrac{Y_k + Y_{k+1}}{2}$

Also, the joint distribution of $Y_k,Y_{k+1}$ is given as follows (only defined for $x<y)$: $$f_{Y_k , Y_{k+1}}(x,y)=\frac{(2k)!}{((k-1)!)^2}[F(x)(1-F(x))]^{k-1}f(x)f(y)$$

Let $U=\dfrac{Y_k + Y_{k+1}}{2}$, starting off as follows:

$P(U\leq u)=P(\dfrac{Y_k + Y_{k+1}}{2}\leq u)=P(Y_k + Y_{k+1}\leq 2u)$

Thus, we are required to integrate the function $f(x,y)$ over the given shaded region:

enter image description here

Thus, the integral looks like this:$\int_{- \infty}^{u} \int_{x}^{2u-x}f(x,y)dydx$.

Is that expression okay ?

Because if I proceed further I get the result as $0$, whereas if I don't consider the condition $x<y$ the result is non-zero, but from the definition of the joint distribution of order statistics that condition is necessary, isn't it ?

Kindly help !


1 Answer 1


mistake: I think the joint distribution of $(Y_{(k)},Y_{(k+1)})$ is $$ f_{Y_{(k)},Y_{(k+1)}}(x,y)=\frac{(2k)!}{((k-1)!)^2}[F(x)(1-F(y))]^{k-1}f(x)f(y)\mathbb{I}_{x\le y} $$ With this correction, you should get the proper cdf of the median from the double integral $$ \int_{- \infty}^{u} \int_{x}^{2u-x}f(x,y)\,\text{d}y\text{d}x $$ that indeed incorporates the constraint $x<y$ (or $x\le y$).


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.