-------- Original Message --------
Subject: Re: bootstrap
Date: Tue, 24 Mar 2009 13:41:43 -0700 (PDT)
From: andrea cardini <[email protected]>
To: [email protected]
Dear Yu,
I guess that one of the first question is whether you're interested to
estimate the support from your variables (as one would do in phlylogenetic
analyses) or the effect of sampling error on the phenogram topology (e.g.,
if you're doing a cluster analysis of mean shapes from several samples).
If your aim is the first one, I am not sure that bootstrapping is a good
idea for shape variables from Procrustes based landmark analyses.
If you want to estimate repeatability of clusters in relation to sampling
error, this is described in a couple of papers of mine.
CARDINI A., Elton S., 2008 - Variation in guenon skulls I: species
divergence, ecological and genetic differences. Journal of Human Evolution,
54: 615-637.
CARDINI A., Elton S., 2008 - Does the skull carry a phylogenetic signal?
Evolution and modularity and in the guenons. Biological Journal of the
Linnean Society, 93: 813-834.
In the second one there's also a small section on phylogenetic trees based
on shape variables that I strongly suggest you to skip. It was a mistake to
add it (has little relevance for the main aim of the study) and the
reconstruction based on maximum likelyhood (not used but mentioned) is
flawed.
Below, you can find excerpts from one of these paper where we say something
about bootstraps. Everything was done in NTSYSpc 2.20.
Good luck.
Cheers
Andrea
... bootstrap samples were built for each species (separate sexes),
and bootstrapped species mean shapes were computed. For instance,
100 bootstrap samples of n = 37 were built using the
37 females of C. ascanius and the mean was computed for
each of them (hence, producing 100 bootstrap means for this
species). Thus, for each species (with separate analyses for females
and males) 100 means are generated by bootstrapping
the original sample and computing a new mean for each bootstrap
sample.
... Cluster analyses were performed on the matrix of shape distances
of observed means and on bootstrapped species mean
shapes. Thus, the mean of the first bootstrap sample for A. nigroviridis
was analysed together with that of the first bootstrap sample
of C. aethiops plus that of the first bootstrap sample of C.
ascanius and so on for all 23 species. This was done 100 times.
Then, pairwise shape distances between the 23 species means of
the first bootstrap were computed and an UPGMA phenogram
calculated. This was done for all 100 bootstrap samples, producing
100 phenograms. Eventually, a 50%majority rule consensus
tree of the 100 phenograms (plus the observed) was built and
percentages of phenograms supporting the observed topology
were shown. Bootstrap support of phenogram clusters provides
information on how consistently a cluster is found in phenograms
when species means are computed after removing some
specimens and replacing them with others in the same sample.
Thus, it indicates how inaccuracies of estimates of mean shapes
(based on available samples) might influence the reconstruction
of species phenetic relationships.
... The reproducibility of results can also depend on
the choice of landmarks used to capture the shape of
a FDM. Lockwood et al. (2004) and Couette, Escarguel
& Montuire (2005) addressed this issue by bootstrapping
shape variables. This is the most widely
used test for the effect of character sampling on tree
topologies in cladistic analyses of meristic characters.
However, Caumul & Polly (2006: 2464) observed that,
when estimating the support for the nodes in a tree
using ordinary phylogenetic data, the characters are
sampled with replacement to generate a new dataset
[but] for morphometric data, the characters are
scores whose values depend on the ordination of the
mean shapes. The reproducibility of results can also depend on
the choice of landmarks used to capture the shape of
a FDM. Lockwood et al. (2004) and Couette, Escarguel
& Montuire (2005) addressed this issue by bootstrapping
shape variables. This is the most widely
used test for the effect of character sampling on tree
topologies in cladistic analyses of meristic characters.
However, Caumul & Polly (2006: 2464) observed that,
when estimating the support for the nodes in a tree
using ordinary phylogenetic data, the characters are
sampled with replacement to generate a new dataset
[but] for morphometric data, the characters are
scores whose values depend on the ordination of the
mean shapes. Without discussing if and to what
extent bootstrapping shape variables might be (in)appropriate,
we suggest an alternative procedure, also
borrowed from the cladistic literature (Felsenstein,
2004a) and easily applicable to geometric morphometric
data. A jack-knife procedure can be applied to
the matrix of landmark coordinates in order to
produce replicate landmark configurations where different
bits of information are excluded from the
analysis. Thus, one at a time, each landmark (more
precisely its x, y, and z coordinates) is excluded, the
species mean shapes are Procrustes superimposed,
and a matrix of pairwise Procrustes distances is computed.
This is repeated as many times as the total
number of landmarks in the original configuration.
The corresponding matrices of variables or distances
are then used for tree building. The proportions of
clusters found in the jack-knife replicate trees are
the estimates of the character (i.e. landmark configuration)
support of tree nodes. Had this been
applied to the guenon chondrocranium, the character
support would have been, on average, approximately
20% higher than our bootstrap proportions of
node repeatability. Random replacement of shape
coordinates, as in the bootstrap procedure used in
other studies, is very likely to produce results similar
to our jack-knife procedure. Indeed, as was the case
in our jack-knife example, both Lockwood et al.
(2004) and Couette et al. (2005) found very high percentages
of tree node support. Although we do not
argue against the utility of estimating uncertainties
in tree nodes due to character sampling (which we
would do, however, using a jack-knife when applied
to shape coordinates), we do suggest that the effects
of sampling error should also be estimated. Since this
could be large (Cole et al., 2002; Caumul & Polly,
2006; Cardini & Elton, 2007), its consequences could
potentially be more profound than those due to
uncertainties in character sampling.
At 07:21 24/03/2009 -0400, you wrote:
-------- Original Message --------
Subject: bootstrap
Date: Sat, 21 Mar 2009 07:49:14 -0700 (PDT)
From: ÓáÇà <[email protected]>
To: <[email protected]>
Dear Morphometricians
I conducted a geometric morphometric analysis of tooth .I want to do a
cluster analysis(ntsyspc2.1) using UPGMA with the Euclidean distance as
distance measure. Should I do a Bootstrap analysis and what software can
be applied to do it generally?
Thanks
Yu qing
dentist
Department of Prosthodontics,
Stomatolgy hospital affiliated to medical school,
Nanjing University, Nanjing, 210008, China.
Phone:+ 86 83620261
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
Dr. Andrea Cardini
Lecturer in Animal Biology
Museo di Paleobiologia e dell'Orto Botanico, Universitá di Modena e Reggio
Emilia
via Università 4, 41100, Modena, Italy
tel: 0039 059 2056532; fax: 0039 059 2056535
Honorary Fellow
Functional Morphology and Evolution Unit, Hull York Medical School
University of Hull, Cottingham Road, Hull, HU6 7RX, UK
University of York, Heslington, York YO10 5DD, UK
E-mail address: [email protected], [email protected],
[email protected]
http://hyms.fme.googlepages.com/drandreacardini
http://ads.ahds.ac.uk/catalogue/archive/cerco_lt_2007/overview.cfm#metadata
More on publications at:
http://www.cons-dev.org/marm/MARM/EMARM/framarm/framarm.html
CLICK ON THE LETTER C AND LOOK FOR "CARDINI" (p. 8-9 until March 2009)
http://hyms.fme.googlepages.com/dr.sarahelton-publications
LOOK FOR "CARDINI"
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
