Hi, It's not quite so simple, at what stage do you recombine your samples. Recombining the covariances into an "average" covariance from the imputed samples has different implications than recombining the projected data. I think that in terms of pca uncertainty the projection directions is the issue. Nicholas
Message: 1
Date: Mon, 6 Apr 2009 12:21:51 -0400
From: Farrar.David at epamail.epa.gov
Subject: Re: [R-sig-eco] missing data in PCA
To: St?phane Dray <dray at biomserv.univ-lyon1.fr>
Cc: r-sig-ecology-bounces at r-project.org, Jason S <jas2339 at yahoo.com>,
r-sig-ecology at r-project.org
Message-ID:
<OF3DFCFEFF.E4C894BF-ON85257590.00568507-85257590.0059E106 at epamail.epa.gov>
Content-Type: text/plain
For combining imputation with PCA, transcan{Design} could be interest.
For imputation, I have been hearing about the mice program.
The documentation for princomp() suggests that a covariance matrix can be
entered via the covmat argument.
David F
r-sig-ecology-bounces at r-project.org wrote on 04/03/2009 10:34:51 AM:
[image removed] Re: [R-sig-eco] missing data in PCA St?phane Dray to: Jason S 04/03/2009 10:48 AM Sent by: r-sig-ecology-bounces at r-project.org Cc: r-sig-ecology ade4 has nipals() which deals with row data. Cheers. Jason S wrote:
Dear Jari, Thanks for the hints. A package such as mvnmle would be ideal, but
it provides only the estimate of the covariance matrix, not the raw data. However, comments like princomp and pca require the raw data.
Do you (or anyone else) know a command to do PCA straight from a
covariance matrix?
best, Jason
________________________________ From: Jari Oksanen <jari.oksanen at oulu.fi> Cc: Jari Oksanen <jari.oksanen at oulu.fi>; r-sig-ecology at r-project.org Sent: Friday, April 3, 2009 2:55:20 AM Subject: Re: [R-sig-eco] missing data in PCA On 03/04/2009, at 00:32 AM, Jason S wrote: Dear all, I was wondering if you have good options to deal with missing
data on a PCA in R. I guess I could simply delete those cases, but I think there should be better options in terms of predicting them. Any
hints?
Howdy, There may be better ways, but they are not easy... Check paragraph
on missing data in multivariate task view. This lists several alternatives of multiple imputation data. About a year ago I tried some of them for multivariate analysis, but that was not quite straightforward. The problem was summarizing multivariate results. Things may have been changed since then, and there may be some canned routines.
Here is one link to multivariate task view (but you can use a
mirror close to you):
http://cran.r-project.org/web/views/Multivariate.html One thing you should remember: do not replace missing values with
means.
Cheers, Jari Oksanen [[alternative HTML version deleted]]
_______________________________________________ R-sig-ecology mailing list R-sig-ecology at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
-- St?phane DRAY (dray at biomserv.univ-lyon1.fr ) Laboratoire BBE-CNRS-UMR-5558, Univ. C. Bernard - Lyon I 43, Bd du 11 Novembre 1918, 69622 Villeurbanne Cedex, France Tel: 33 4 72 43 27 57 Fax: 33 4 72 43 13 88 http://pbil.univ-lyon1.fr/members/dray/
_______________________________________________ R-sig-ecology mailing list R-sig-ecology at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[[alternative HTML version deleted]] ------------------------------
_______________________________________________ R-sig-ecology mailing list R-sig-ecology at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology End of R-sig-ecology Digest, Vol 13, Issue 3 ********************************************