[R] Factor Analysis?

[Loren M. McCarter]

> Brian Ripley wrote:
> 
ripley> My feeling is that there are plenty of packages that do factor analysis,
ripley> and not much demand for it in R.  Anyone want to write a factor analysis
ripley> package for R?
ripley> 

I'm simultaneously working on my dissertation and also putting
together a collection of variations on principal components analysis
(pca) techniques into an R library. Unfortunately, I don't think that
I will have a complete R library until at least a few months as it has
to take a back seat to writing and graduating from school. Unless
someone else has taken the initiative by then, I will plan to begin
adding factor analysis to the library. And yes, the rotation (e.g.,
varimax) would be a necessary option for most factor analysts and yes,
it would take me some time to figure out how to write those rotation
functions.

I personnally use pca more than factor analysis because pca allows the
output of vectors containing "observed" component scores for each
individual, which is useful for my work. Factor analysis requires more
assumptions about the data and does not allow, theoretically, the
output of vectors containing the "observed" individual component
scores. Of course, factor analysis also has many advantages over pca
if assumptions are met and observed component scores are not
necessary.

Maria Wolters <wolters at ikp.uni-bonn.de> wrote:
wolters> Well, IM very HO, the problem is that factor analysis is frequently used in the social
wolters> sciences, and social science people tend to prefer more GUI-orineted
wolters> packages such as SPSS or Systat....

I'm a graduate student in behavioral neuroscience, which is officially
in psychology and thus I have some experience with this. Though there
is certainly variation among psychologists (many do use S-Plus and
many work with large complex datasets), you are correct that the
majority of psychologists do seem to use and prefer the GUI,
point-and-click interfaces (ala SPSS). The datasets of many
psychologists have a small n with many variables, and do not require
much computer programming to perform manipulations (e.g., little need
to use UNIX, perl, python etc. to manipulate matrices).

Among the statistically and computationally-sophisticated
psychologists, many are experts in techniques such as Structural
Equation Modeling (SEM), which can be described as a factor-analytic
approach to path analysis. Interestingly, the SEM software market has
been dominated by specialized statistical packages, not the
mainstream, one-size-fits-all, statistical packages. For example, most
psychologists do NOT use the SEM procedures bundled with SAS (i.e.,
PROC CALIS) but instead prefer LISREL, AMOS, EQS, MPlus and the like
(I believe that SPSS owns LISREL however). Many of these packages
allow the user to draw path diagrams in a GUI dialog box and then
apply SEM statistical tests to the drawing, which is quite useful for
testing path theories. There is a "freely downloadable" (though I do
not think that it is open-source) SEM package called Mx, by Michael
Neale (see http://views.vcu.edu/mx/; runs on Linux [also my FreeBSD
2.2.8 Linux emulation] and Windows).

I agree that an SEM library (including factor analysis functions)
would be a nice addition to R, but maybe it would be more reasonable
to create an R interface to Mx, which is already a full-blown SEM
package (of course this may be a stupid idea on my part, especially
since obtaining the Mx open-source may not be an option).

Loren

----------------------------------------
Loren McCarter
Graduate Student, Behavioral Neuroscience
UC Berkeley
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._