Re: Distribution of a Percetile
- From: "\"Luis A. Afonso\"" <licas_@xxxxxxxxxxx>
- Date: Thu, 22 Sep 2005 05:31:31 EDT
Jack said
***Well, you can always use a nonparametric approach. If, however, you can use a correct parametric model, the methods would be more powerful. Jack***
My response:
The simple fact that nonparametric methods are less powerful than parametric is not a sufficient raison that these ones must always be preferred.
Why?
By two ponderous things.
_______Because you seldom know by theoretical bases what is the Distribution and therefore we must *guess* it using the sample (a risky task).
_______Even you had the good luck to fid out the Distribution the parameter value (or parameters) evaluated from the sample is never exact.
Jack could
Firstly *read thoroughly what is asked for* , then answer adequately in the context and NEVER to blunder something out. Inappropriate generalities are of scarce help in concrete problems.
________________________________
Richard Ulrich said
Jack
***Well, you can always use a nonparametric approach. If, however, you can use a correct parametric model, the methods would be more powerful- Personally, I don't think there is much loss in power, but that's not the reason that I rather like the use of the order statistics for CI's on percentiles. Gain in generality? I suppose that I assume that someone is interested in the percentiles *because* there is doubt about the distributional forms I would check Conover for a test, rather than CI's or in addition to CI's
Rich Ulrich, ***
My response
MY GOD. The method I posted (and still I think the more appropriate) based in ORDER Statistics?
(mais, faîtes comme vous voulez. Je m'enfus)
_________________licas
.
- References:
- Distribution of a Percetile
- From: Stephen Clark
- Distribution of a Percetile
- Prev by Date: Re: Distribution of a Percetile
- Next by Date: K-L Distance
- Previous by thread: Re: Distribution of a Percetile
- Next by thread: Re: Distribution of a Percetile
- Index(es):
Relevant Pages
|
Loading
