[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: Equality of distributions
Nick & Stas
Thanks for your quick response..
The test statistics are requested in order to figure on the QQplot..
Nothing else than the KS? I'm encountering some differences over the tails
- between LogNormal and Gamma distributions for my continuous observed data
- between GCPP (Generalized Compound Poisson Pascal, see Brocket & all
JMR1996) and NBD distribution for ordinal data
De : firstname.lastname@example.org
[mailto:email@example.com]De la part de Nick Cox
Envoyé : mardi 2 novembre 2004 20:05
À : firstname.lastname@example.org
Objet : RE: st: Equality of distributions
Apart from the enormous caveat in Stas' statement, I'd
recommend a strongly graphical approach, especially
quantile-quantile plots, as giving various handles on
Single test statistics or attempts to identify a best
distribution often miss important details: where doesn't it
fit? In the tails? In the middle? etc. etc.
> As far as I can recall, Kolmogorov-Smirnov test (Stata: ksmirnov) is
> the uniformly most powerful test for comparison of an empirical to
> fully known theoretical distributions, if there is no need to estimate
> the parameters of the latter. It cannot answer your second question
> though; it just says if the difference between the two distributions
> is within the limits of sampling fluctuations.
> > I have two distributions (observed) and several theoretical
> (built upon
> > simulations), each is resumed into its own varaible (no expression)
> > I'm looking for tests in order to establish whether
> > - Reject/accept whether one theoretical distribution can
> describe the
> > observed data..
> > - Given degrees of freedom, which one describe "best" the
> observed data ?
> > References, commands in Stata are highly welcome..
> > Observed data are nominal, other are continuous
* For searches and help try: