Let $X_1,X_2,\ldots,X_n$ be i.i.d. random variables with cdf $F$ on $\mathbb{R}$ and $M_n$ be the sample median. The proof starts off with defining for any $a\in\mathbb{R}$, $S_n = \#\text{ of }X_i\text{s}>a/\sqrt{n}$. Then $S_n\sim\text{Bin}(n,p)$. We're interested in finding $P(M_n\le a/\sqrt{n})$. Why is the set $\{M_n\le a/\sqrt{n}\}$ equivalent to $\{S_n\le (n-1)/2\}$?


With a slight augmentation of your notation, define $$S_{n,a/\sqrt{n}}:=\sum\limits_{i=1}^{n}{\bf 1}_{X_i>a/\sqrt{n}}=\sum\limits_{i=1}^{n}(1-{\bf 1}_{X_i\leqslant a/\sqrt{n}})=n-\sum\limits_{i=1}^{n}{\bf 1}_{X_i\leqslant a/\sqrt{n}}\,,$$ where we have used the indicator functions $\ {\bf 1}_{X_i\leqslant a/\sqrt{n}}\ $.

Now, we simply observe the following:

$$ \begin{align} \omega\in\{M_n\leqslant a/\sqrt{n}\} &\iff \text{ at least }\frac{n+1}{2}\ \text{ of the }X_i\text{'}s\ \text{ are such that}\ X_i(\omega)\leqslant \frac{a}{\sqrt{n}}\\&\iff\ S_{n,a/\sqrt{n}}(\omega)=n-\sum\limits_{i=1}^{n}{\bf 1}_{X_i\leqslant a/\sqrt{n}}(\omega)\leqslant n-\frac{n+1}{2}=\frac{n-1}{2}\\ &\iff \omega \in \{S_{n,a/\sqrt{n}}\leqslant (n-1)/2\}\,. \end{align} $$

  • $\begingroup$ @stats134711, You are welcome. $\endgroup$
    – ki3i
    Apr 5 '15 at 3:54

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.