1
$\begingroup$

If we chop a standard normal distribution in half and use only the positive side (scaled up by a factor of $2$ to maintain a proper density), then we get the so-called ‘half normal’ density:

$$f_X(x)=\sqrt{\frac{2}{\pi}}\exp(-\frac{1}{2}x^2),x>0$$

If $Z ∼ N(0, 1)$ then $|Z|$ has a half normal density.

How if $X$ is half normal and $S = ±1$ with probability half each, independently of $X$, then $Z = SX ∼ N(0, 1).$ $S ∼ U{(−1, +1)}$ to indicate that the distribution of $S$ is uniformly distributed over the finite set $\{−1, +1\}$?

$\endgroup$
1
$\begingroup$

$S$ is either $+1$ or $-1$ with equal probabilities. So $SX$ gives you back the negative side of the original normal you had. And you would scale it down (in your words) by a factor of two, regaining the original standard normal random variable.

$\endgroup$
  • $\begingroup$ Can't $S$ accept value between $-1$ and $1$ ? When we say $X\sim U(0,1)$, we mean $X$ can assume any value between $0$ and $1$ $\endgroup$ – ABC Sep 24 '13 at 1:11
  • $\begingroup$ From the problem description, $S$ is discrete and cannot take on any values except $-1$ and $+1.$ I agree the notation is confusing. Whoever wrote it should have been more careful. $\endgroup$ – soakley Sep 24 '13 at 3:10
1
$\begingroup$

I will derive the mean and variance, and leave the distribution to you. Define $I\{S=1\}\equiv I_1$ for simplicity, to be the indicator function taking the value $1$ when $S=1$ and the value $0$ when $S=-1$. Note also that when $S=1 \Rightarrow Z = X$ and when $S=-1 \Rightarrow Z = -X$. Combining we can re-write the functional form of $Z$ as

$$Z= XI_1 - X(1-I_1) = X(2I_1-1)$$

Then, given also independence between $X$ and $S$ $$E(Z) = E\left(X(2I_1-1)\right) = E(X)E(2I_1-1) = \sqrt{\frac{2}{\pi}}\cdot(2E(I_1)-1)$$

and by the properties of indicator function, $E(I_1) = P(S=1) = 1/2$ we get

$$E(Z) = \sqrt{\frac{2}{\pi}}\cdot(2\frac 12-1) = 0$$

For the variance we have, since the mean is zero,

$$ \text {Var}(Z) = E(Z^2) = E\left(X^2(2I_1-1)^2\right) = E\left(X^2\right)E\left(4I_1^2 - 4I_1+1\right)$$

Now $$\text {Var}(X) = E(X^2) - \left[E(X)\right]^2 \Rightarrow E(X^2) = \text {Var}(X) + \left[E(X)\right]^2 = 1-\frac {2}{\pi} + \frac {2}{\pi} = 1$$

So $$ \text {Var}(Z) = 1\cdot E\left(4I_1^2 - 4I_1+1\right)= 4E\left(I_1^2\right) - 4P(S=1)+1 = 4E\left(I_1^2\right) - 1$$

By the properties of the product of the indicator function, $I_1^2= I_1\cdot I_1 = \min \{I_1,I_1\} = I_1$

So

$$\text {Var}(Z) = 4E\left(I_1\right) - 1 = 4P(S=1) - 1 = 2 - 1 = 1$$

$\endgroup$

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.