Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

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

Can anyone shed some light on the below:

  1. Consider a set with $N$ distinct members, and a function $f$ defined on $\mathbb Q$ that takes the values $0$, $1$ such that $\frac1N\sum_{x\in\mathbb Q} f(x) = p$. For a subset $S$ of $\mathbb Q$ of size $n$, define the sample proportion $$p = p(S) = \frac1N\sum_{x\in S} f(x)$$ If each subset of size $n$ is chosen with equal probability, calculate the expectation and standard deviation of the random variable $p$.

    1. Let $X\sim \mathcal N(0, 1)$ be a normally distributed random variable with mean 0 and variance 1. Suppose that $x \in \mathbb R, x > 0$. Find upper and lower bounds for the conditional expectation $E(X \mid X >x)$
    2. Now suppose that $X$ has a power law distribution, $P(X >x) = ax^{-b}$, for $x>x_0>0$, and some $a> 0, b> 1$. Calculate the conditional expectation $E(X\mid X>x), x >x_0$

Many thanks in advance.

share|cite|improve this question
Two different exercises (nearly three). Please ask them as two separate questions, add to each your own thoughts, what you tried, which related results you know. – Did Jan 12 '13 at 15:54
...and use of LaTeX would greatly increase legibility – Hagen von Eitzen Jan 12 '13 at 16:05
The only thing I can think of for the first one is expectation of \hat{p} = p and the limit of the variance of \hat{p} as n -> N is 0. The variance of p is 0. Don't think this is correct though. – SilvaSquarred Jan 13 '13 at 0:38

Taking on the second question.. $$ E(X|X>x) = \int_{x}^{\infty}tP(t,X>x)dt = \int_{x}^{\infty}t\frac{P(t)}{P(X>x)}dt\tag{1} $$ where $$ P(x) = \mathcal N(0, 1) = \frac{1}{\sqrt{2\pi}}\mathrm{e}^{-\frac{x^2}{2}} $$ In Eq.1 I use dummy variables to lessen the confusion. $$ E(X|X>x) = \frac{1}{\sqrt{2\pi}}\frac{1}{P(X>x)}\int_{x}^{\infty}t\mathrm{e}^{-\frac{t^2}{2}}dt $$ the integral is trivial results in $$ E(X|X>x) =\frac{1}{\sqrt{2\pi}}\frac{1}{P(X>x)}\mathrm{e}^{-x^2/2} $$ remembering that $$ P(X>x) = \frac{1}{\sqrt{2\pi}}\int_x^{\infty}\mathrm{e}^{-t^2/2}dt, $$ which is also equal to $\frac{1}{2}\mathrm{erfc}(x/\sqrt{2})$

From Abramowitz and Stegun you can obtain an upper and lower bound to the erfc function as $$ \sqrt{\frac{2}{\pi}}\frac{\mathrm{e}^{-x^2/2}}{x+\sqrt{x^2+4}}< \frac{1}{2}\mathrm{erfc}(x/\sqrt{2}) < \sqrt{\frac{2}{\pi}}\frac{\mathrm{e}^{-x^2/2}}{x+\sqrt{x^2+\frac{8}{\pi}}} $$ and further more $$ \sqrt{\frac{\pi}{2}}\frac{ x+\sqrt{x^2+\frac{8}{\pi}}}{\mathrm{e}^{-x^2/2}}< \frac{1}{\frac{1}{2}\mathrm{erfc}}< \sqrt{\frac{\pi}{2}}\frac{ x+\sqrt{x^2+4}}{\mathrm{e}^{-x^2/2}} $$ putting it all together we find $$ \frac{x+\sqrt{x^2+\frac{8}{\pi}}}{2} < E(X|X>x) < \frac{x+\sqrt{x^2+4}}{2} $$

$\textbf{Part 2}$ We have to determine the probability of $P(x)$ first to be able to compute the conditional expectation. $$ P(x) = \frac{\alpha-1}{x_0}\left(\frac{x}{x_0}\right)^{-\alpha} $$ here $x_0$ is the min value as outlined in the OP. $$ \int_x^{\infty}\frac{\alpha-1}{x_0}\left(\frac{t}{x_0}\right)^{-\alpha}dt =ax^{-b} $$ we find that $$ \alpha = b+1,\\ \left(\frac{1}{x_0}\right)^{1-\alpha} = a. $$ this yields that the original distribution should be $$ P(x) = abx^{-(b+1)}. $$

Then to compute $$ \begin{eqnarray} E(X|X>x) &=& \frac{1}{P(X>x)}\int_{x}^{\infty}t\left(abt^{-(b+1)}\right)dt \\ &=& \frac{ab}{ax^{-b}}\int_x^{\infty}t^{-b}dt \\ &=& \frac{b}{x^{-b}}\frac{x^{1-b}}{b-1}\\ &=& \frac{b}{b-1}x \end{eqnarray} $$

share|cite|improve this answer
not sure I agree with you. Isn't eq (1) the folloing: $$ E(X|X>x) = \int_{x}^{\infty} t P(t|t>x)dt = \int_{x}^{\infty}t\frac{P(t)}{P(t>x)}dt\tag{1} $$ ? Thanks! – IcannotFixThis Oct 1 '14 at 13:42
@icannotfixthis it is confusing notation on my part. The denominator is already an integrated quantity so does not factor into the expectation I.e can be treated as a constant. Does that make it clearer? – Chinny84 Oct 1 '14 at 18:33

Regarding the upper limit of $E(x|x>X)$ when x is Normal distributed, a simple and more stringent value is

$E(x|x>X)\leq X+\sqrt{2/\pi}$.

Just looking at the Gaussian function one can infer that the difference

$E(x|x>X)-X$ decreases with $X$, so that it will have its maximum value ($\sqrt{2/\pi}$) when $X=0$. Thus for all $X\geq0$ one has $E(x|x>X)-X \leq E(x|x>0)$. It follows that

$E(x|x>X) \leq X + (E(x|x>0)) \rightarrow E(x|x>X) \leq X + \sqrt{2/\pi}$.

I just plotted the result and it works better than the one derived from the Abramowitz and Stegun limit, for all $X\geq 0$.

A trivial lower limit is of course $E(x|x>X)\geq X$

share|cite|improve this answer

The way I see it, $f \left( x \right)$ follows a Bernoulli distribution with success probability $p$. Assuming this is correct, then the distribution of $n \hat{p}$ given $S$ would be a binomial random variable with success probability $p$ and $n$ trials. \begin{equation} \left. n \hat{p} \middle| S \right. \sim \mathcal{B} \left( p , n \right) \text{.} \end{equation} The conditional mean and variance of $n \hat{p}$ are \begin{align*} \mathrm{E} \left[ n \hat{p} \middle| S \right] = & n p \text{,} \\ \mathrm{Var} \left[ n \hat{p} \middle| S \right] = & n p \left( 1 - p \right) \text{,} \end{align*} and from the properties of the expectation it follows that \begin{align*} \mathrm{E} \left[ \hat{p} \middle| S \right] = & p \text{,} \\ \mathrm{Var} \left[ \hat{p} \middle| S \right] = & \frac{p \left( 1 - p \right)}{n} \text{.} \end{align*} Now it's time to marginalize out $S$, \begin{align*} \mathrm{E} \left[ \hat{p} \right] = & p \text{,} \\ \mathrm{Var} \left[ \hat{p} \right] = & p \left( 1 - p \right) \sum_{n = 1}^N \frac{P \left( n \right)}{n} \text{,} \end{align*} where $P \left( n \right)$ is $n$'s probability mass function. Since $S$ is chosen with equal probability, and the number of distinct $n$-element subsets in an $N$-element set such as $\Omega$ is \begin{equation*} \left( \begin{array}{c} n \\ N \end{array} \right) = \frac{N!}{n! \left( N - n \right)!} \text{,} \end{equation*} I'd be tempted to say that $P \left( n \right)$ is the reciprocal of this coefficient. But this doesn't lead anywhere.

share|cite|improve this answer

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.