-------- Original Message --------
Subject: RE: Sampe size reduction testing bootstrap
Date: Thu, 28 May 2009 07:45:59 -0700 (PDT)
From: Louis Boell <[email protected]>
To: <[email protected]>
References: <[email protected]>
Dear Getulio,
the subsampling procedure you cited corresponds exactly to what I wish
to do. Since the subsampled datasets have to be analyzed in a CVA
afterwards, I looked on the net whether CVA has been implemented in R or
mathematica, with negative results. As far as I can see, this means:
either I implement the CVA myself, which I feel currently not able to
do, or I just implement the subsampling, create a limited number of
subsampled datasets (say, 100) and feed these manually, one by one, into
my morphometric software to create the desired distribution. The latter
procedure should be feasible (even if tedious).
Thanks again for the advice
Louis
Louis Boell
MPI für Evolutionsbiologie
August-Thienemannstr.2
24306 Plön
[email protected]
[email protected]
> Date: Wed, 27 May 2009 16:59:41 -0400
> From: [email protected]
> To: [email protected]
> Subject: RE: Sampe size reduction testing bootstrap
>
>
>
> -------- Original Message --------
> Subject: RE: Sampe size reduction testing bootstrap
> Date: Wed, 27 May 2009 13:07:48 -0700 (PDT)
> From: Getulio Amaral <[email protected]>
> To: [email protected]
> References: <[email protected]>
>
> Dear Loius,
>
> The procedure that you want is subsamplig. See:
>
> "Politis and Romano’s subsampling bootstrap takes samples without
> replacement of size b from the original sample of size n, generally with
> b < n (read “b much less than n”). Such samples are themselves
> samples of size b from the true unknown distribution F of the original
> sample." http://www.wepapers.com/Papers/8707/Subsampling_Bootstrap
>
> I can implement this procedure ir R but I do not use biological
programs.
> The procedure can be easyly implemented.
>
>
> Best wishes,
> Getulio
>
>
> >
> > -------- Original Message --------
> > Subject: RE: Sampe size reduction testing bootstrap
> > Date: Wed, 27 May 2009 00:56:38 -0700 (PDT)
> > From: Louis Boell <[email protected]>
> > To: <[email protected]>
> > References: <[email protected]>
> >
> >
> >
> > Dear Getulio,
> >
> > I have realized that "bootstrap" is indeed the inappropriate term for
> > what I wish to do. I would simply like to randomly draw 15 specimens
> > from my groups containing 30 specimens and to repeat this, say, 100
> > times in order to obtain a distribution of the Mahalanobis distances
> > between groups containing only 15 specimens when using 15 Landmarks.
> > Probably "permutation" fits this approach better.
> > Concerning R, I am decided to learn it as soon as possible...
> > Thanks for the advice,
> > Best wishes,
> >
> > Louis
> >
> >
> > Louis Boell
> > MPI für Evolutionsbiologie
> > August-Thienemannstr.2
> > 24306 Plön
> > [email protected]
> > [email protected]
> >
> >
> >
> >
> > > Date: Tue, 26 May 2009 15:23:03 -0400
> > > From: [email protected]
> > > To: [email protected]
> > > Subject: Re: Sampe size reduction testing bootstrap
> > >
> > >
> > >
> > > -------- Original Message --------
> > > Subject: Re: Sampe size reduction testing bootstrap
> > > Date: Tue, 26 May 2009 11:37:03 -0700 (PDT)
> > > From: Getulio Amaral <[email protected]>
> > > To: [email protected]
> > > References: <[email protected]>
> > >
> > > Dear Louis,
> > >
> > > The bootstrap uses a sample to a empirical distribution of the
> > variable.
> > > I do not think bootstrap is suitable in this case.
> > >
> > > Maybe you can use the Hotteling and Goodall tests on the tangent
space
> > > but the number of landmarks in each object must be the same.
> > >
> > > If you use R, there is the function shapes, where at the link
> > >
> > > http://cran.r-project.org/web/packages/shapes/index.html
> > >
> > > You can download the pdf file with all the instruction.
> > >
> > > Best regards,
> > > Getulio J. A. Amaral
> > > Departamento de Estatística
> > > Centro de Ciências Exatas e da Natureza
> > > Universidade Federal de Pernambuco
> > > http://www.de.ufpe.br/~gjaa
> > >
> > > > -------- Original Message --------
> > > > Subject: Sampe size reduction testing bootstrap
> > > > Date: Tue, 26 May 2009 07:40:35 -0700 (PDT)
> > > > From: Louis Boell <[email protected]>
> > > > To: <[email protected]>
> > > >
> > > >
> > > >
> > > > Dear colleagues,
> > > >
> > > > I would like to ask for your advice about the following
problem: In
> > a
> > > > landmark-based GM context I am calculating Mahalanobis distances
> > beween
> > > > several groups. Now my material consists of museum material,
thus I
> > do
> > > > have sample size problems. My samples consist of 15 to 30
> > Landmarks per
> > > > group. I want to use 15 2D landmarks, which means that I need 30
> > > > specimens per group, which is not possible for some groups. In
> > order to
> > > > estimate the amount of deviation from the "true" Mahalanobis
> > distances
> > > > caused by having less than 30 specimens in a group, I randomly
> > discarded
> > > > 15 specimens from some groups for which I have 30 specimens and
> > > > calculated the correlation between the Mahalanobis distances from
> > the
> > > > unreduced sample set and those from the reduced sample set, which
> > gave
> > > > me an rsq of about .8.
> > > > Now I think that this calculation is in principle ok, but the
> > choice of
> > > > specimens to be discarded for the calculation of this correlation
> > > > should be bootstrapped. I have been using MorphoJ so far, but this
> > does
> > > > not allow for implementig a self-written bootstrapping procedure
> > (which
> > > > I do also not yet have). What would you do in such a situation?
> > > > Thanks in advance for any help with this,
> > > > Best wishes,
> > > >
> > > > Louis Boell
> > > >
> > > >
> > > > Louis Boell
> > > > MPI für Evolutionsbiologie
> > > > August-Thienemannstr.2
> > > > 24306 Plön
> > > > [email protected]
> > > > [email protected]
> > > >
> > > >
> > > >
> > > >
> >
------------------------------------------------------------------------
> > > > Die virtuelle Festplatte im Netz - 25 GB für deine Bilder,
> > Musikfiles
> > > > und Dateien! Jetzt kostenlos sichern!
> > > > <http://redirect.gimas.net/?n=M0905WLPhotos>
> > > >
> > > > --
> > > > Replies will be sent to the list.
> > > > For more information visit http://www.morphometrics.org
> > > >
> > >
> > >
> > > Atenciosamente,
> > >
> > >
> > >
> > >
> > > --
> > > Replies will be sent to the list.
> > > For more information visit http://www.morphometrics.org
> > >
> >
> >
------------------------------------------------------------------------
> > Kein Werbe-Blabla - hier klicken!
> > <http://redirect.gimas.net/?n=M0905FTP_SpringCampaign1>
> >
> > --
> > Replies will be sent to the list.
> > For more information visit http://www.morphometrics.org
> >
>
>
> Atenciosamente,
> Getulio J. A. Amaral
> Departamento de Estatística
> Centro de Ciências Exatas e da Natureza
> Universidade Federal de Pernambuco
> http://www.de.ufpe.br/~gjaa
>
>
>
> --
> Replies will be sent to the list.
> For more information visit http://www.morphometrics.org
>
------------------------------------------------------------------------
Surfen – optimiert für MSN. Hol Dir die neue Version des Internet
Explorers - kostenlos! <http://redirect.gimas.net/?n=M0905IE8_MSN1>
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
