It is known (cf. Wikipedia, Noncentral_chi_distribution) that the non-central chi-square distribution with k degrees of freedom is a Poisson weighted mixture of central chi-squared distributions).

There are many proofs but all are basically the same (see same ref section "Derivation of the pdf" for a sketch and here libre stats for a freely accessible complete proof).

I'm looking for the intuition behind this result, that is, a way to interpret and explain it. Otherwise such the result seems kind of accident. Any idea ?

best, G.


Patnaik writes on p. 203:

The [non-central $\chi^2$ distribution] has been [sic] obtained by [Fisher, R. A. (1928). Proc. Roy. Soc. A, 121, 654] as a particular case of the distribution of the multiple correlation coefficient. A purely analytical proof was given by [Tang, P. C. (1938). Statist. Res. Mem. 2, 126]. [...] We give a direct geometrical derivation of the [non-central $\chi^2$ distribution].

The geometric derivation by Patnaik uses a series expansion of the density, which is equivalent to the series expansion of the characteristic or moment generating function in the usual quick derivation of the non-central $\chi^2$ distribution as a Poisson mixture of central $\chi^2$ distributions.

So, I think there has never been any "intuition behind this result" -- just a lucky and rather simple analytical observation.

  • $\begingroup$ Yes, I saw the paper and some others (for instance Kerridge Aust J Stat 7(1965)37-39 ) and there is no intuition behind just series expansion. But I still find weird that all constants line up so remarkably, My idea was to look for an intuition by decomposing somehow chi-square as a sum, the same for Poisson, both have additive properties... $\endgroup$
    – Gabriel
    Feb 22 at 19:48
  • $\begingroup$ @Gabriel : Intuition is defined as the "faculty of knowing or understanding something without reasoning or proof" (thefreedictionary.com/intuition). It seems very likely that there has never been such intuition here. As for "all constants lin[ing] up so remarkably", I guess you mean the Poisson mixture. But the Poisson probabilities are just the terms of the normalized Maclaurin series for the exponential function, which are quite ubiquitous. $\endgroup$ Feb 23 at 0:33
  • $\begingroup$ I guess you are right... $\endgroup$
    – Gabriel
    Feb 24 at 16:25

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.