Let $X_i$ be iid exponential random variables. I want to calculate $Var\left(\sqrt{\bar{X}/6}\right).$

The idea I had to simplify this is expressing it as $$Var\left(\sqrt{\frac{\bar{X}}{6}}\right) = E\left(\frac{\bar{X}}{6}\right) - \left[E\left(\sqrt{\frac{\bar{X}}{6}}\right)\right]^2.$$

This still leaves me with the problem of solving $E(\sqrt{\bar{X}/6})$. How do I deal with the square root inside the expectation?

  • 3
    $\begingroup$ It is not going to be a real number because a normal random variable takes on negative values. $\endgroup$ – user3903581 Apr 2 '17 at 1:08
  • $\begingroup$ @user3903581 Thank you, let me amend the question to make it more practical $\endgroup$ – kingledion Apr 2 '17 at 1:09
  • $\begingroup$ Is the sample size $n$ or 6? Anyway, you may be able to use the fact that the sum of Exponentials is a Gamma distribution ... $\endgroup$ – wolfies Apr 2 '17 at 6:30

The sample mean from an exponential($\lambda$) sample is gamma($n$, $\lambda/n$) distributed.

It's a fact that the square root of $\overline{X}$ therefore has a generalized gamma distribution with parameters $p=2$, $d=2n$, and $a = \sqrt{\lambda/n}$, so it's mean is

$$ a \frac{ \Gamma( (d+1)/p )}{\Gamma(d/p)} = \sqrt{\lambda/n} \cdot \frac{ \Gamma( (2n+1)/2)}{ \Gamma(n)}$$

Divide this by $\sqrt{6}$ and there's your answer.

  • 1
    $\begingroup$ Is $\lambda$ standing for a scale parameter in your formulation? $\endgroup$ – Alecos Papadopoulos Apr 2 '17 at 14:03
  • $\begingroup$ Yes. If $\lambda$ is supposed to be the rate, replace $\lambda$ with $1/\lambda$ in the final answer. $\endgroup$ – gammer Apr 2 '17 at 14:44

Say the parameter to each distribution is $\lambda$. First, $$ E\left[\frac{\bar{X}}{6}\right] = [6\lambda]^{-1} $$ by linearity. Then, because $Y = \sum_i X_i \sim \text{Gamma}(n,1/\lambda)$ \begin{align*} E\left(\sqrt{\frac{\bar{X}}{6}}\right)&= \int \sqrt{\frac{y}{6n}} \frac{\lambda^n }{\Gamma(n)}y^{n-1}e^{-y\lambda} dy \\ &= (6n)^{-1/2}\frac{\lambda^n}{(n-1)!} \int y^{n + 1/2 - 1}e^{-y\lambda}dy \\ &= (6n)^{-1/2}\frac{\lambda^n}{(n-1)!} \left[\frac{\Gamma(n+1/2) \lambda^{-(n+1/2)} }{\Gamma(n+1/2)\lambda^{-(n+1/2)}}\right] \int y^{n + 1/2 }e^{-y\lambda}dy \\ &= (6n)^{-1/2}\frac{\lambda^n}{(n-1)!} \Gamma(n+1/2) \lambda^{-(n+1/2-1)} \\ &= (6n)^{-1/2} \lambda^{1/2} \frac{\Gamma(n+1/2) }{\Gamma(n) } \end{align*}


The Exponential distribution is a special case of a Gamma ($G$) distribution. If $X \sim \text{Exp}(\lambda)$ where $\lambda$ is the rate parameter (so ($E(X) = 1/\lambda)$, then (shape -scale parametrization)

$$ \sum_{i=1}^n X_i \sim G(n, 1/\lambda) \implies \bar X = \frac 1n \sum_{i=1}^n X_i \sim G(n, 1/n\lambda) $$

Now, $$Z = \sqrt {\bar X} \implies \bar X = Z^2 \implies \frac {\partial \bar X}{\partial Z} = 2Z$$

Then by the change-of-variable formula we have that the density of $Z$ is

$$f_Z(z) = \left |\frac {\partial \bar X}{\partial Z}\right| \cdot f_{\bar X}(z)$$

$$\implies f_Z(z) = \frac {2z}{\Gamma(n)\cdot (1/n\lambda)^n}\cdot (z^2)^{n-1} \cdot\exp \{- z^2/(1/n\lambda)\} $$

$$\implies f_Z(z) = \frac {2}{\Gamma(n)\cdot (1/n\lambda)^n}\cdot z^{2n-1} \cdot\exp \{- z^2/(1/n\lambda)\}$$

Set $p\equiv 2,\;\; n \equiv d/p \equiv d/2,\;\; a\equiv (1/n\lambda)^{1/2}$. Then we can write

$$f_Z(z) = \frac {p}{\Gamma(d/p)\cdot a^d}\cdot z^{d-1} \cdot\exp \{- z^2/a^2\}$$

This is the density of the Generalized Gamma distribution, and so we have

$$E(Z) = E\left[\sqrt{\bar X}\right] = a\frac {\Gamma[(d+1)/p]}{\Gamma(d/p)}$$

and reverting back to the original parameters we obtain

$$E\left[\sqrt{\bar X}\right] = \frac{1}{\sqrt{n\lambda}}\cdot \frac {\Gamma[n+(1/2)]}{\Gamma(n)}$$


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.