Re: [Gretl-users] Principal component analysis: Missing values encountered
On Fri, 4 Nov 2016, Riccardo (Jack) Lucchetti wrote:
I guess that, for consistency with pther cpmmands (ols for example),
we ought to run PCA on an automatically trimmed dataset if there are
missing observations in the currently selected sample, and duly
indicate this in the output.
That is now the behaviour in git and snapshots. Illustrative script:
<hansl>
open data4-10
list L = CATHOL PUPIL WHITE ADMEXP
# full sample
pca L
WHITE[10] = NA
PUPIL[20] = NA
# 2 observations dropped
pca L
smpl ok(L) --restrict
# should give the same result as above
pca L
</hansl>
OTOH, I also think that our pca apparatus needs some tweaking: for
example, there is no GUI equivalent to the --save=n option, and that
we also should introduce a --varimax option for saved factors (and
possibly --quartimax). Perhaps we could get these done after
release.
Yes, I'd rather not get into those points right now.
Allin Cottrell