Prev: fopen funniness
Next: how to make dynamic matrix(urgent)
From: Peter Perkins on 8 Aug 2010 18:18
On 8/8/2010 4:31 AM, L N wrote:
> Thanks Peter. One more question: I wanted to use a non-parametric
> distribution because it doesn't assume any models - is it generally
> considered ok to use a parametric function based only on closeness of
> fit? What if the process doesn't correspond to the distribution model at
> all?

You need to extrapolate to a region where you have no data at all, so
you're going to have to make some strong assumptions.

If you have a theuretical basis for such assumptions, that's good. If
not, about the best you can do is experiment with different
distributions that all fit your data about equally well. See how
sensitive your conclusions about the region where you don't have data
are to the particular distribution you've chosen. And even then, there
will be other distributions that you have not considered that might have
fit your data and led you to completely different conclusions about what
happens below the truncation point.

You can't get something from nothing.

Hope this helps.
From: L N on 10 Aug 2010 04:28
I don't think the differences between reasonably well fitted distributions will alter the conclusion greatly because I am just trying to find the total number of samples, and the distribution is fairly flat and "featureless" left of the truncation point. Also, the graph was exaggerated a bit - the actual truncation point is <50% of the mean. Thanks for your advice again.

Peter Perkins <Peter.Perkins(a)MathRemoveThisWorks.com> wrote in message <i3naf6$s1g$1(a)fred.mathworks.com>...
> On 8/8/2010 4:31 AM, L N wrote:
> > Thanks Peter. One more question: I wanted to use a non-parametric
> > distribution because it doesn't assume any models - is it generally
> > considered ok to use a parametric function based only on closeness of
> > fit? What if the process doesn't correspond to the distribution model at
> > all?
>
> You need to extrapolate to a region where you have no data at all, so
> you're going to have to make some strong assumptions.
>
> If you have a theuretical basis for such assumptions, that's good. If
> not, about the best you can do is experiment with different
> distributions that all fit your data about equally well. See how
> sensitive your conclusions about the region where you don't have data
> are to the particular distribution you've chosen. And even then, there
> will be other distributions that you have not considered that might have
> fit your data and led you to completely different conclusions about what
> happens below the truncation point.
>
> You can't get something from nothing.
>
> Hope this helps.
First  |  Prev  | 
Pages: 1 2
Prev: fopen funniness
Next: how to make dynamic matrix(urgent)