0

I stumbled upon an expression in an article of statistics for an $n$-th moment with $X$ being a random variable over $[0, \infty)$.

$$\mathbb{E} X^{n} = \int^{\infty}_{0} nz^{n-1}\; \text{Pr}(X > z) \; \text{dz}$$

Could someone enlighten me on why the above is true? It indeed works for the exponential distribution.

Dilip Sarwate
  • 24,512
  • 3
  • 46
  • 114
johnny
  • 2,223
  • 1
  • 22
  • 26
  • 5
    If you are familiar with the formula $$E[Y] = \int_0^{\infty} P\{Y > y\}\,\mathrm dy$$ for a nonnegative random variable, set $Y = X^n$ and do a change of variable $z = y^{1/n}$ in $$E[Y] = \int_0^{\infty} P\{Y > y\}\,\mathrm dy = \int_0^{\infty} P\{X^n > y\}\,\mathrm dy = \int_0^{\infty} P\{X > y^{1/n}\}\,\mathrm dy.$$ – Dilip Sarwate Jun 14 '12 at 11:45
  • Thank you! Knowing the formula surely helps! – johnny Jun 14 '12 at 12:14
  • 1
    You are welcome. For the intuition behind the formula as well as a formal proof via Tonelli/Fubini as you asked of @StefanHansen, see the answers to [Intuition behind using complementary CDF to compute expectation for nonnegative random variables](http://math.stackexchange.com/q/64186/15941) – Dilip Sarwate Jun 14 '12 at 12:30
  • The article I'm reading is treating a continuously distributed random variable, so that's probably why they won't make a difference. But your point is of course very valid. – johnny Jun 14 '12 at 12:42
  • 1
    The functions $P(X>z)$ and $P(X\geq z)$ can only differ at a countable set of $z$ values, and so it doesn't matter which of these you integrate. –  Jun 14 '12 at 13:12

1 Answers1

2

First, use Tonelli's theorem to conclude that (write the probability as an integral and interchange the two integrals) $$ E[X^n]=\int_0^\infty P\left(X^n> z\right)\, \mathrm{d} z. $$ Now write $P\left(X^n> z\right)=P\big(X> z^{1/n}\big)$ and use change of variables with $t=z^{1/n}$.

Stefan Hansen
  • 24,846
  • 7
  • 56
  • 86
  • Would you add a line or two to show how you apply Tonelli's theorem to get the formula for expectation? I can't see it immediately. I'll accept the answer then. Thanks! – johnny Jun 14 '12 at 12:13
  • It's alright. See @DilipSarwate comment for a link to the proof of that relation. – johnny Jun 14 '12 at 12:32
  • @DilipSarwate: Of course, I edited it now. Thanks for the heads up. – Stefan Hansen Jun 14 '12 at 13:02