[R] PCA with NA

[Hartmut Oldenbürger]

Hi Birgit, and All

Possibly you should not consider the case completed ;-)
There is an important alternative to imputing means, or estimates from
first or second order
regression (Frane, Psychometrica, BMDP):  partial-least-squares (Wold),
which uses as much
information from the data as possible to estimate the principal
components, or the missing data.

Stephane Dray, also from Lyon, provides 'nipals' here:
    http://biomserv.univ-lyon1.fr/~dray/software.php
There is also an interesting paper.  - In case, you use PLS to estimate
and impute, set the
number of factors as high as reasonably possible, e.g. m-1, when m is
the number of variables.
                                                best - Hartmut Oldenbürger
---
holdenb1 at uni-goettingen.de
http://www.wipaed.wiso.uni-goettingen.de/~holdenb1