Suppose a random variable X follows the hypergeometric distribution with parameters $N$, $K$, $n$, where the pmf is given as

\begin{equation} Pr(X=k) = \frac{\binom{K}{k}\binom{N-K}{n-k}}{\binom{N}{n}}. \end{equation}

The question is: how closed is the probability that $X$ is even to $\frac12$, i.e. $\vert Pr(X~is~ even) - \frac12 \vert$?

As mentioned in wolfies' answer, it really depends on the parameters $N$, $K$, $n$. My conjecture is as follows:

Let $Pr_r(even)$ be the probability that the RV is even under the parameters $K_r$, $N_r$, $n_r$. If we assume $K_r= c_1 N_r$, $n_r = c_2 N_r$, for some universal constants $c_1$, $c_2$, then \begin{equation} Pr_r(even) \rightarrow 0.5,~as~N_r\rightarrow \infty. \end{equation} The reason I believe that this is true is that the hypergeometric distribution will converge to binomial distribution as $N \rightarrow \infty$, and in a binomial distribution $Pr(even)=0.5$. If this conjecture is true, the question then becomes: what is the rate of convergence?


The answer must be ---- it depends!

To make this clear, the Hypergeometric can have forms where there is a sizeable mass at 0 (which is even). And if you have say 70% of the density mass at $X = 0$, then it is clear that $P(X \text{ is even})$ is not going to be close to $\frac12$. To illustrate, here is a plot of the Hypergeometric pmf when $N = 200$, $n = 10$, and $r = 5$ (whatever notation one uses):

(source: tri.org.au)

If desired, one can formally derive, in say Mathematica, an expression for the sum of the pmf over even values: Sum[pmf, 0, n, 2] which returns a complicated mess involving HypergeometricPFQ functions ... but it comes back to the point made above: it depends on your parameter values.

  • $\begingroup$ Do you mean PDF instead of HypergeometricPFQ? $\endgroup$ – Sycorax Apr 20 '16 at 20:06
  • $\begingroup$ Thank you so much for your answer! I did not realize that the probability can be far away from 0.5. I modified the question to make it more concrete. $\endgroup$ – Martin Zhang Apr 20 '16 at 20:11
  • $\begingroup$ Mathematica yields a "complicated mess" because it doesn't know any better. The pmf for the even terms in any discrete distribution supported on the integers having pmf $p$ is simply $t\to (p(t)+p(-t))/2.$ In this case, $p$ can be expressed as a Riemann hypergeometric function $_2F_1.$ $\endgroup$ – whuber Jun 1 '19 at 16:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.