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

If a continuous random variable $X$ has a symmetric distribution around $0$, what is the conditional distribution of $X$ given $|X|$?

share|cite|improve this question
Which definition of the conditional distribution do you know? – Did Oct 3 '12 at 13:46
disintegration. I guess the answer is $\frac{F_x(x)}{F_X(x)+ F_X(-x)}, but I need a formal proof. – ie86 Oct 3 '12 at 14:01
disintegration... Is this a definition? You might try to be more explicit. (And your guess is wrong.) – Did Oct 3 '12 at 14:55
@Martin : Don't confuse $F$ with $f$. In conventional notation, the former is the c.d.f. and the latter is the density. Your guess would be right if you had $f$ rather than $F$. – Michael Hardy Oct 3 '12 at 16:26
It was just a typo. Thanks – ie86 Oct 3 '12 at 17:09
up vote 3 down vote accepted

For a rigorous answer one needs to rely on a precise definition of conditional distribution. Recall that, for any random variables $X$ and $Y$, one calls conditional distribution of $X$ conditionally on $Y$, any family $(\mu_y)_y$ of probability measures on the state space of $X$, indexed by the state space of $Y$, such that, for every bounded measurable function $u$, $$ \mathbb E(u(X)\mid Y)=L_u(Y)\ \text{almost surely},\quad\text{where}\quad L_u(y)=\int u(z)\mathrm d\mu_y(z). $$ In other words, one asks that, for every bounded measurable function $v$, $$ \mathbb E(u(X)v(Y))=\mathbb E(L_u(Y)v(Y)). $$

In the case at hand, $Y=|X|$ has density $g(y)=f_X(y)+f_X(-y)$ on $y\gt0$, hence $$ \mathbb E(L_u(Y)v(Y))=\int_{y\gt0}L_u(y)v(y)g(y)\mathrm dy. $$ On the other hand, $$ \mathbb E(u(X)v(Y))=\int u(x)v(|x|)f_X(x)\mathrm dx=\int_{y\gt0} (u(y)f_X(y)+u(-y)f_X(-y))v(y)\mathrm dy. $$ The only way these can coincide for every function $v$ is that, for Lebesgue almost every $y$, $$ L_u(y)g(y)=u(y)f_X(y)+u(-y)f_X(-y), $$ that is, $$ L_u(y)=\frac{u(y)f_X(y)+u(-y)f_X(-y)}{f_X(y)+f_X(-y)}. $$ Finally, the only way $L_u$ may be as above is when the family $(\mu_y)_y$ is such that, $\mathbb P_Y(\mathrm dy)$ almost surely, $$ \mu_y(\mathrm dx)=\frac{f_X(y)\delta_y(\mathrm dx)+f_X(-y)\delta_{-y}(\mathrm dx)}{f_X(y)+f_X(-y)}. $$ In particular, if the distribution of $X$ is symmetric, $f_X(y)=f_X(-y)$ hence, as you guessed, $$ \mu_y(\mathrm dx)=\frac12(\delta_y(\mathrm dx)+\delta_{-y}(\mathrm dx)). $$
A shortcut in the symmetric case is to note that $(X,|X|)$ and $(-X,|X|)$ coincide in distribution, hence $$ \mathbb E(u(X)\mid |X|)=\mathbb E(u(-X)\mid |X|)=\mathbb E(L_u(|X|)\mid |X|)=L_u(|X|), $$ where $L_u(y)=\frac12(u(y)+u(-y))$ for $y\gt0$. This yields directly that the measure $\mu_y$ is uniform on $\{-y,y\}$.

share|cite|improve this answer
Thanks. Good, liked both your answers. – ie86 Oct 4 '12 at 0:25

Let $f_X(x)$ be the density function. The event that $|X|=x$ is the same as $X=\text{either }x\text{ or }-x$. So you have $$ \Pr(X=x\mid |X|=x)=\frac{f_X(x)}{f_X(x)+f_X(-x)}, $$ and $\Pr(X=-x\mid |X|=x)$ is complementary to that.

But you say it's symmetric about $0$, so that probability is $1/2$.

share|cite|improve this answer
Thanks. That's what I guess but typed wrongly above. How do you formally prove it? – ie86 Oct 3 '12 at 15:35
$\Pr(X=x)=0$. $ $ – Did Oct 3 '12 at 15:41
@did : Typo fixed. – Michael Hardy Oct 3 '12 at 15:45
I think I may return to write a proof later...... – Michael Hardy Oct 3 '12 at 18:46
@Did : Yes, it does say the distribution is continuous, so my argument is perhaps incomplete. The condional probability $\Pr(X=x\mid |X|)$ should be a random variable that is a function of $|X|$. – Michael Hardy Feb 4 '13 at 17:05

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.