# What does it mean to say that one random variable is “greater” than another?

This might be a very basic question, but I've just started with order statistics so it is tripping me up a bit.

What does it mean to say that a random variable, say $Q_1$, is greater than another, say $Q_2$?

The way I've heard order statistics explained is that we have some observations, e.g., $x_1 = 1$, $x_2 = 3$, $x_3 = 2$. We can then say that $x_1 < x_3 < x_2$, i.e. $x_1 = x_{(1)}, x_2 = x_{(3)}, x_3 = x_{(2)}$. So far, this makes sense to me, because we are ordering numbers, not random variables.

Now however, people start to say that (and we denote random variables with upper case) $X_{(1)} < X_{(2)} < X_{(3)}$ and that we can find the probability distributions of each of these: which I do not understand since, if we already have the samples and values, why is there a probability distribution since we know for certain what they are? If random variables are meant to represent an unknown quantity, how can they be ordered - and if they represent a known quantity, why is there any uncertainty/probability involved?

• I would say $X<Y$ iff $P(X<Y)=1$. – Lord Shark the Unknown Jan 19 '18 at 6:18
• I would say $Y=X+Z$ where $P(Z \geq 0)=1$ (which is in fact equivalent of what @Lord Shark the Unknown just mentionned by setting $Z:=Y-X$ :). – Jean Marie Jan 19 '18 at 6:21

One straightforward way to order random variables is to treat them as functions on your probability space $(\Omega,P,\Sigma)$: $Q_1:\Omega\rightarrow\mathbb{R}$ and $Q_2:\Omega\rightarrow\mathbb{R}.$ Then you can define $$Q_1 > Q_2 \quad \text{ iff } \quad \forall \omega\in\Omega: Q_1(\omega)>Q_2(\omega).$$ Notice that this definition makes perfect sense even if you do not know anything about specific values of $Q_1,Q_2.$ If you do have a probability defined, it makes sense to weaken this requirement and exclude sets of zero probability. This is what Lord Shark the Unknown suggested above.
This way of ordering works also for order statistics. In this case your random variable $X$ is not real valued but $X:\Omega\rightarrow\mathbb{R}^n$ an $n$-vector, where $n$ is the size of your sample. In the case $n=2$ for example, the brief statement $X_{(1)}\leq X_{(2)}$ now means in "natural language":