Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

How do you prove that the sample median of $X_i, \ldots, X_n$ where $i$ is odd converges to the median of U$(0,1)$? Do you have to find the cdf of the median, find the p.m.f. of median, and then show that as $n$ increases, the variance of the p.m.f. gets smaller, eventually becoming a point mass around the true median?

share|cite|improve this question

Since your post is incomplete, I will answer the question in my interpretation. Assume $X_i$ are independent identically distributed standard uniform random variables. We seek to show that the median of a sample of size $n = 2m+1$ converges in probability to $\frac{1}{2}$ as $m \to \infty$.

See wikipedia article on order statistics, and for the expression for the probability density of the median $X_{(m+1)}$, which is $m+1$-st smallest element in the sample. $$ f_{X_{(m+1)}}(u)\mathrm{d} u = \frac{(2m+1)!}{(m)!^2} u^{m} (1-u)^{m} \mathbf{1}_{0 < u < 1} \mathrm{d} u = (2m+1)\binom{2m}{m} (u(1-u))^{m} \mathbf{1}_{0 < u < 1} \mathrm{d} u $$

By definition of convergence in probability, we seek to evaluate: $$ \begin{eqnarray} p_{m,\epsilon} &=&\mathbb{P}\left( \vert X_{(m+1)} - \frac{1}{2} \vert \ge \epsilon \right) = 2 \int_0^{\frac{1}{2}-\epsilon} \frac{(2m+1)!}{(m)!^2} u^{m} (1-u)^{m} \mathbf{1}_{0 < u < 1} \mathrm{d} u \\ &=& 2 \int_0^{\frac{1}{2}-\epsilon} (2m+1)\binom{2m}{m} (u(1-u))^{m} \mathrm{d} u \\ & \le & (2m+1)\binom{2m}{m} \left( \frac{1}{4} - \epsilon^2 \right)^m \int_0^{\frac{1}{2} -\epsilon} 2 \mathrm{d} u \end{eqnarray} $$ The last inequality follows because $u(1-u)$ is increasing for $0<u<\frac{1}{2}- \epsilon$.

We should prove that for any fixed $\epsilon >0$, $\lim_{m \to \infty} p_{m, \epsilon} = 0$.

For large $m$, $\binom{2m}{m} \sim \frac{4^m}{\sqrt{m \pi}} $, therefore $$ \begin{eqnarray} 0 \le \lim_{m \to \infty} p_{m, \epsilon} &\le& \lim_{m \to \infty} (2m+1)\binom{2m}{m} \left( \frac{1}{4} - \epsilon^2 \right)^m (1-2 \epsilon) \\ &=& \lim_{m \to \infty} (2m+1) \frac{4^m}{\sqrt{m \pi}} \left( \frac{1}{4} - \epsilon^2 \right)^m (1-2 \epsilon) \\ &=& \lim_{m \to \infty} \frac{2m+1}{\sqrt{m \pi}} \left( 1 - 4 \epsilon^2 \right)^m (1-2 \epsilon)= 0 \end{eqnarray} $$

share|cite|improve this answer
If you were to shift the distribution to U(theta+ 0.5,theta-0.5) how would the distribution change? – lord12 Nov 17 '11 at 16:45
@lord12 You know the density of $X$, now you should find the density of $X + \theta - \frac{1}{2}$. Do you know how ? – Sasha Nov 17 '11 at 16:47
Do you just change the lower bound for the integration to theta - 1/2? You want P(|X-theta} > e)= 0 since theta is median of this distribution. – lord12 Nov 17 '11 at 17:29
In the integration, I don't get how you just substituted in (1/4-e^2) without integrating u(1-u)^m – lord12 Nov 17 '11 at 17:59
@lord12 Since $u(1-u)$ is increasing on the domain $0<u< v = \frac{1}{2} -\epsilon$, it follows that $\int_0^v (u(1-u))^m du \le \int_0^v (v(1-v))^m d u = (v(1-v))^m v$. – Sasha Nov 17 '11 at 19:57

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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