Thanks for your advice Glen.

CCC

"Glen Barnett" <[EMAIL PROTECTED]> wrote in message
acs3bv$fsp$[EMAIL PROTECTED]">news:acs3bv$fsp$[EMAIL PROTECTED]...
>
> Chia C Chong <[EMAIL PROTECTED]> wrote in message
> acro4s$72c$[EMAIL PROTECTED]">news:acro4s$72c$[EMAIL PROTECTED]...
> > I have a random variable,X with 1000 samples (all discrete values). I
want
> > to test whether X can sastify a Poisson process or not. How should I
test
> > it?
>
> I assume you mean that you have one sample with 1000 observations.
>
> Firstly, I want to caution you against using hypothesis tests for goodness
of
> fit
> of distributions in general. Usually it turns out that hypothesis tests
aren't
> actually
> what is needed for the problem at hand (which you don't specify), so be
sure
> that it's really a hypothesis test that is what's required to answer your
> underlying
> question of interest!
>
> (To begin with, we already know the answer - the distribution from which
the
> data comes is not precisely Poisson, so the real question must be somewhat
> different. Often, for example, the question of interest is more along the
lines
> of "How badly will my results be affected by the fact that the data
doesn't
> exactly come from a Poisson distribution?". That question /isn't/ answered
by
> hypothesis tests at all. There are some questions for which hypothesis
tests
> are a reasonable answer, it's just that people are not usually trying to
get
> answers to those questions.)
>
> As for how to test it, it depends on what your alternatives are.
>
> For example, if you want to test it against some overdispersed model that
the
> Poisson might be a special case of, you do something different than if
you're
> interested in testing against a general discrete alternative on the
> non-negative
> integers.
>
> Glen
>
>


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to