I shouldn’t do this, because we’ve all made mistakes, especially me. But I took a peek at a website with the intriguing title “100+ commonly asked data science interview questions” with the thought “Maybe I could be a data scientist”. But the author of this list choked on the very first question. It’s interesting to examining why the question is bad.
The question reads: “Prove that a random variable with a distribution on [0,1] (that is, the density function is equal to 0 outside [0, 1]) has an expectation always between 0 and 1. Prove that its variance is maximum and equal to 1/12 if and only if the distribution is uniform on [0, 1].”
The first half is okay. The integral of xf(x) over the interval 0 to 1 is bounded above by the integral of f(x) over the interval 0 to 1 and you know that the latter has to equal 1 for it to be a density function.
The second half, though, is wrong. A Bernoulli random variable with probability 1/2 is bounded between 0 and 1 and has variance 1/4. That’s a whole lot bigger than 1/12.
You can confirm this with a quick simulation in R. The single line
var(rbinom(1000, 1, 0.5))
will give you an answer that is pretty close to 0.25.
No fair, you claim. The Bernoulli distribution does not have a density function because it is not a continuous random variable.
That’s true. Let’s consider a different case then. Let’s consider a beta random variable with parameters alpha=0.5 and beta=0.5. I had to peek at Wikipedia, but the variance of a beta distribution is
alpha*beta / ((alpha+beta)^2 * (alpha+beta+1).
Plug in alpha=1 and beta=1 as a quick check and you do indeed get 1/12. When you plug in alpha=0.5 and beta=0.5, you get 1/8. Check this in R with
var(rbeta(1000, 0.5, 0.5))
and you’ll get a value close to 0.125.
Maybe, you’ll say something like, no fair because the density function of this particular beta distribution is unbounded at both 0 and 1.
Fair enough. How about a distribution that is uniform on the interval 0 to 1/3 and 2/3 to 1? That has a variance of roughly 2/9.
You say that you were talking about unimodal distributions only. I’m not sure, but I think you might be right about 1/12 being the largest variance possible…
…except that a uniform distribution is multi-modal.
You can find an earlier version of this page on my blog.