[R] starting values for uniquenesses in factanal()
Prof Brian Ripley
ripley at stats.ox.ac.uk
Fri Feb 9 17:30:34 CET 2001
On Fri, 9 Feb 2001, Barry Cooke wrote:
> Dear R-help,
> Using R 1.2.1 on Windows98 to run a factor analysis on a 64x150 matrix of
> data generated from a simulation model, factanal() reported that it failed
(I hope that's a typo. You have 64 observations on 150 variables?
Then you have a singular sample correlation matrix. Factor analysis cannot
be appropriate, since a Heywood case is needed to get a singular matrix.)
> to find a solution. Looking at the factanal code, I see the immediate
> condition that triggered the result:
> if (best == Inf)
> stop("Unable to optimize from these starting value(s)")
> So I am sure factanal() is giving the intended result. What I want to know
> is what sort of data properties lead to this result, and how does one go
> about choosing starting uniqueness values so that a (good) result can be
> computed? The factanal() help, in describing parameter "start", says it is:
> NULL or a matrix of starting values, each column giving an initial set of
> But (due to my own ignorance) this is a little sketchy. Any references or
> advice is appreciated.
> I realize this is not strictly an R-question ... so far. But because I did
> not encounter the same problem using Minitab's factor analysis,
Use Minitab's uniquenesses as the starting values: just give them as a
vector as the start argument.
> I am curious
> as to the reason for the different behaviour (and this is an R question).
> Maybe the response I got from R's factanal() is actually more appropriate?
> Is this another case of commercial software giving me an answer I want, even
> if it's not a good one? (I would prefer to use only R for the paper I am
> writing.) Any comments appreciated. To spare bandwidth I don't post the
> data here, but I can send it privately.
I don't know about Minitab, but I do know that most commercial statistical
packages will happily quote non-optimal answers, and indeed have default
methods that are statistically suspect.
If your first para is correct, this is a case in point.
Brian D. Ripley, ripley at stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch
More information about the R-help