If $$X \sim \mathcal{N}(\mu,\sigma)$$

then $$X^2 \sim \frac{e^{-\frac{\left(\mu +\sqrt{x}\right)^2}{2 \sigma ^2}} \left(e^{\frac{2 \mu \sqrt{x}}{\sigma ^2}}+1\right)}{2 \sqrt{2 \pi } \sigma \sqrt{x}} \hspace{3 mm}, \hspace{3 mm} x>0$$

If $X^2$ has been known as non-central chi square distribution ($\mathcal{X^2(1,\lambda)}$) then how to calculate the non-centrality parameter in context of above distribution of $X^2$, so that both the distributions become equal? Any help please.

  • 1
    $\begingroup$ You are mistaken in your premises: $X^2$ does not have a $\chi^2(1,\lambda)$ distribution. It is $\sigma^2$ times such a distribution. By definition, the noncentrality parameter of that distribution is $\lambda=(\mu/\sigma)^2$. $\endgroup$ – whuber Oct 13 '14 at 4:31
  • $\begingroup$ I mean a non-central chi square distribution, which has two parameters: degree of freedom and non-central parameter. $\endgroup$ – kaka Oct 13 '14 at 7:50
  • 2
    $\begingroup$ Yes, it is clear that's what you mean--but you are wrong to suppose that the $X^2$ you exhibit here has such a non-central $\chi^2$ distribution. It does not. After you rescale it by $1/\sigma$, it will have that distribution, whose PDF is given by a modified Bessel function $I_{-1/2}$. This fact is readily checked after you recognize that $I_{-1/2}(z) = \sqrt{1/(2\pi z)}\left(\exp(z)+\exp(-z)\right)$. $\endgroup$ – whuber Oct 13 '14 at 16:23
  • $\begingroup$ @whuber Nice description. So $\mathcal{N}(\mu,\sigma)$ doesn't have a non central Chi-Square distribution. so does the distribution of $X^2$ where $X \sim \mathcal{N}(\mu,\sigma)$ has any particular name? $\endgroup$ – kaka Oct 13 '14 at 21:28
  • 1
    $\begingroup$ A multiple of a distribution occurs so often and is so simply related to the distribution itself that it need not have any special name. For that reason, for example, the square of a Normal$(0,\sigma)$ distribution has no special name, either: it's just $\sigma^2$ times a $\chi^2(1)$ distribution. $\endgroup$ – whuber Oct 13 '14 at 21:33

As already answered by whuber in above comments:

If $$X \sim \mathcal{N}(\mu,\sigma),$$ and $$Y \sim \chi^2 \Big(k=1,\lambda=\Big(\frac{\mu}{\sigma}\Big)^2 \Big),$$

then $$X^2\stackrel{d}{=}\sigma^2 Y.$$

In words, if X is normal random variable with non-zero mean and variance and Y is non-central chi squared random variable with one degree of freedom and non-central parameter $\lambda=(\frac{\mu}{\sigma})^2$, then $X^2$ is equal in distribution to $\sigma^2$Y.

Moreover, a scaled non-central chi square variable doesn't have the non-central chi squared distribution.

It should be $X^2$ in the conclusion.


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.