If $X$ is a discrete random variable that takes values on the positive integers, it is true that

$$E(X) = \sum_{k=1}^{\infty} P(X \ge k)\;.$$

I know how to prove this (by expressing the summand as a sum of all probabilities going from $n=k$ to $n=\infty$ and then playing around with the double sums) but I'm having trouble developing an intuition for why the above result holds.

I am thinking of a number line where the equation says that if I'm stepping through $k$ from $1$ to $\infty$ and take the sum of probabilities of everything to the right of $k$ at each step, that gives the expectation of $X$, which does not make sense to me.

This can be flipped to be $1 - \sum_{k=1}^{\infty} P(x \le k)$, whose interpretation would then be stepping though $k$ but summing the probabilities of everything to the left of $k$ at each step, but this does not help me understand either.

The result is from exercise 4.7.8 in Rice's statistics book.


1 Answer 1


One way to see this is as follows:

For a discrete random variable $X$, we obviously have that

$$E[X] = \sum_{k=1}^\infty kP(X=k)$$

so our goal is to see how $\sum_{k=1}^{\infty} P(X\geq k)$ relates to that. Since $X$ is a discrete rv, we also have that

$$P(X>n) = \sum_{k=n}^{\infty} P(X=k)$$

which in words is saying that $P(X>n)$ is the same as just summing every individual $P(X=k)$ from $n$ onwards. Then if we were to sum all of the $P(X>n)$ from $n=1$ to $\infty$, we have

$$\sum_{n=1}^\infty P(X>n) = \sum_{n=1}^\infty \sum_{k=n}^{\infty} P(X=k)$$

Things get confusing with this double sum, so let's take a moment to think about it intuitively. When $n=1$, we have one $P(X=k)$ for every single $k=1,\dots,\infty$. When $n=2$, we have one $P(X=k)$ for every single $k=2,\dots,\infty$. So every $P(X=k)$ gets another count, except the first one! And it's clear to see that after $n=1$, we will never count another $P(X=1)$ since the sums are $k=n,\dots,\infty$ for any $n$. So we only get one $P(X=1)$. To make things easier on our brain, let's just store them in an infinite dimensional matrix $A$ where the $(i,k)$-th element is whether $P(X=k)$ gets a count for when $n=i$ ($1$ if gets a count, $0$ otherwise). So when $n=1$ (the first row of the matrix), every $P(X=j)$ gets a count. So the first row is just an infinite vector of $1$s. For $n=2$, every $P(X=j)$ for $j\geq 2$ gets a count, so the first entry is $0$, and the rest are all $1$. For $n=3$, every $P(X=j)$ for $j \geq 3$ gets a count, so the first two entries is $0$, and the rest are all $1$.

The pattern is easy to see: for $n=i$, every $P(X=j)$ for $j \geq i$ will get a count. So let's do this till infinity, and let's count how many times we count each $P(X=j)$. For $j=1$, we only count it once. For $j=2$, we count it twice. For $j=i$, we count it $i$ times. So it must be that

$$\sum_{n=1}^\infty P(X>n) = \sum_{n=1}^\infty \sum_{k=n}^{\infty} P(X=k) = 1\times P(X=1) + 2\times P(X=2) + 3\times P(X=3) + \dots$$

But wait! That looks awfully like that first equation we wrote! And indeed, it is equal to $\sum_{k=1}^\infty P(X=k)$, which is equal to $E[X]$.

The intuition is that by counting sums of $P(X=k)$ in the way described, we are counting each one $k$ times, which is the natural way to describe the 'average' value of a discrete random variable, since this is another way to write that we are weighting each value $k$ by $P(X=k)$.


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.