I understand the classcial $\chi^2$ "goodness of fit" test used in Statistics, in which we compute $\sum_{i=1}^n \frac{(O_i - E_i)^2}{E_i}$ and, by comparing this quantity to a value found in a table of $\chi^2$ law (with a given risk $\alpha = 5\%$ for example), we decide if we should or not accept the hypothesis that the sample is likely to be an observation or not of a given distribution.

But I haven't found a good precise proof online yet, that shows that it's not only a good "recipe", but also has a strict proof, using probability theory (I know it exists, but I haven't found one yet).

Do you know a good detailed proof?

  • $\begingroup$ good find, @Basj $\endgroup$ – cmo Apr 9 '18 at 18:08

I think I finally found one in this document, page 2.

Theorem (Pearson): The random variable

$$\sum_{j=1}^n \frac{(\nu_j-n p_j)^2}{n p_j} \rightarrow \chi_{r-1}^2$$ converges in distribution to $\chi_{r-1}^2$-distribution with $r − 1$ degrees of freedom.

| cite | improve this answer | |
  • 1
    $\begingroup$ Thanks for the reference $\endgroup$ – Julien May 14 '18 at 20:17

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.