Re: Goodness of fit measures for a distribution
- From: Eric Bohlman <ebohlman@xxxxxxxxxx>
- Date: 7 Apr 2005 00:34:32 GMT
"Unknown" <noone@xxxxxxxxxxx> wrote in
news:d31i6v$ld5$1@xxxxxxxxxxxxxxxxxxx:
> Just a quick note to clear out any misconceptions about what I am
> trying to do.
>
> I know what the distribution of my data looks like and I have a good
> idea of what parametric model it follows.
>
> The problem is, that there are many distributions that VISUALY fit the
> data well.
> I want to find which of these distributions fits my data BEST.
>
> As you can understand I cannot rely only on visual results because
> they all look pretty much the same.
>
> That's why I need some NUMBERS!!!!
There's no single definition of "best" in this case; before you can even
start to consider metrics, *you* have to, based on subject-matter
knowledge and not statistical considerations, decide which forms of
departure between two distributions are of practical importance for your
work and which aren't. Different metrics will weight different aspects
of departure differently; you want one that will heavily weight the ones
that are important in your field and lightly weight those that aren't.
You'll need to consider the "costs" of a) assuming no difference in a
particular aspect when there really is one and b) assuming a difference
when there really isn't one.
.
- References:
- Goodness of fit measures for a distribution
- From: Unknown
- Re: Goodness of fit measures for a distribution
- From: Reef Fish
- Re: Goodness of fit measures for a distribution
- From: Unknown
- Goodness of fit measures for a distribution
- Prev by Date: Re: basic description of PCA terms
- Next by Date: Re: basic description of PCA terms
- Previous by thread: Re: Goodness of fit measures for a distribution
- Next by thread: Re: Goodness of fit measures for a distribution
- Index(es):
Relevant Pages
|
