non-linearity of median - proof

261 Views Asked by At

I'd like to show counterexamples for:

a. $med(X+Y) = med(X) + med(Y)$

b. $med(aX+b) = a\cdot med(X) + b$.

  1. Showing that "a" is not correct:

Let $X, Y$ be independent random variables with exponential distribution and correspondeing: $F(t) = 1-e^{-t}$ (for $t \ge 0$).

To find the median I solve $F(t)=1-e^{-t}=1/2$ and then I see $med(X)+med(Y)=2log(2)$.

Now I'd like to take a look at $F_{X+Y}(t)$ - how can I get this function?

  1. Showing that "b" is not correct:

For showing this I need $F_{aX}(t)$ and $F_{aX+b}(t)$, correct? Well then I should await some answers to "1.", since I am stuck there.

1

There are 1 best solutions below

2
On BEST ANSWER

Part (a)

Your calculation of the sum of the medians of $X$ and $Y$ is correct and gives $2\ln2\approx1.3863$.

As you say the next step is to get the distribution for $X+Y$. The prob that $X+Y\le t$ is $\int_{r=0}^te^{-t}\int_{s=0}^{t-r}e^{-s}\ ds\ dr=\int_0^te^{-r}(1-e^{r-t}\ dr=1-e^{-t}-te^{-t}$.

You can either solve $1-e^{-t}-te^{-t}=\frac{1}{2}$ numerically to find $t=1.6784$, or you can note that the function is increasing and less than $\frac{1}{2}$ at $t=2\ln2$.

Part (b)

Hint: why would you expect this to be incorrect? You have $X\le m$ iff $aX+b\le am+b$.