I'd add a perhaps unfashionable view that the principal components or factors to work with are likely to be those that can be interpreted scientifically or practically. But clearly much depends on the problem and on approved tribal behaviour.
Nick
[email protected]
Philip Ender
My understanding is that the Kaiser criterion is, in fact, based on
the unreduced correlation matrix. So yes, you should use the number
of eigenvalues greater than or equal to one from the -pca-. Selecting
the number of factors using this approach is not recommended much
these days. Goodness of fit, RMSEA and the Tucker-Lewis Index, which
are not available in Stata -ipf-, are more up-to-date methods.
Andrés Cardona Jaramillo wrote:
I'm having some doubts in choosing the number of factors to retain
after an iterated principal axis analysis (factor, ipf). I normally
use the Kaiser Criterion (eigenvalues > 1) to solve this issue with
the command line "factor, ipf mineigen(1)". However I've seen that
other statistical packages like SPSS figure out the number of factor
to be extracted in a principal axis analysis based on the eigenvalues
of a principal component analysis an not the eigenvalues of the former
(which seems to be more intuitive). The latter would be obtainer in
Stata through the following 3 command lines: factor, pc // write down
the number of extracted factor "x" // factor, ipf factors(x)
My question is: are there any theoretical (or maybe practical) reason
to first estimate a principal component analysis (factor, pc) and to
use these eigenvalues in choosing the number of factors to retain in
an iterated principal axis analysis (factor, ipf)
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
