2

Suppose that ${y_1}, ..., {y_n}$ is a random sample from an ${N(\mu,\sigma^2)}$ distribution. Then $$ {\sum_{i=1}^{n}{\frac {(y_{i}-\bar{y})^{2}}{\sigma^{2}}}}$$

has a $\chi^{2}_{n-1}$ distribution. Why is this the sum of ${\chi^{2}}$ distributions that sum to a chi-square distribution with ${n-1}$ degrees of freedom instead ${n}$ degree? Can anyone prove it for me?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Sirius Hou
  • 73
  • 1
  • 5
  • There's some discussion here; [Why is the sampling distribution of variance a chi-squared distribution?](http://stats.stackexchange.com/questions/121662/why-is-the-sampling-distribution-of-variance-a-chi-squared-distribution). One way to prove it is via Cochran's theorem ; the wikipedia page for it ((see the link in the answer) contains a proof of Cochran's theorem and has a section showing how the distribution of the variance follows as a consequence. Two other methods of proof are mentioned by Henry L in comments under the answer. – Glen_b Mar 05 '17 at 01:05
  • 1
    I expect a proof (rather than a discussion and a link to one) has been given directly in an answer in CV before (if I locate one of them this should close as a duplicate). – Glen_b Mar 05 '17 at 01:09

1 Answers1

-1

It is well-known that this is because the sample mean is used in place of $\mu$. Note that this is the sum of n chi-square random variables but they are dependent due to the use of the sample mean in place of $\mu$.

Michael R. Chernick
  • 39,640
  • 28
  • 74
  • 143