principal factor vs. principal-component factor option for factor analysis

[Johann von Roetel]

Hello everyone!

Can someone explain me the difference between the "principal-factor" and "principal-component factor" options for factor analysis in stata 15? On my data, they lead to fundamentally different results. I know that this question has already been answered multiple times, but in my opinion not to a satisfactory extend. I have seen two posts about this, here and on research gate that refer to

Harman (1976). It is, however, not clear where Harman describes the "principal-component factor" method. Maybe in Chapter 8.2 "Component Analysis"?

Thank you for your reply.

Similar Discussion: https://www.stata.com/statalist/arch.../msg00321.html

Harman, Harry H. (1976): Modern Factor Analysis. Third Edition Revised. The University of Chicago Press.

[Johann von Roetel]

Sorry for the push, but i just wanted to add this here:
https://www.statalist.org/forums/for...-pcf-on-factor

As you can see, Boris Ivanov demonstrated that -pca- and -factor, pca- lead to different loadings. I have, however, observed that the structure of factors produced by -factor, pca- resembles those produced by -pca- more closely than those produced by -factor-.

Any ideas?

[Marcos Almeida]

The book "A Gentle Introduction to Stata (Alan Acock, StataCorp) has a nice chapter on this topic truly worth the reading.

A caveat is the use of different terms (under different software/books) to convey the same meaning. On the hand, the use of the same term (in different software/books) to convey a different meaning.

[Johann von Roetel]

Thanks for the tip Marcos! I just got the book and decided to share some of the information obtained:

According to Acock (2016) principle factor analysis (PF) corresponds to what is more generally called exploratory factor analysis. Principal-component factor analysis (PCF) corresponds to what SPSS calls principal component analysis. While PF tries to explain the variance shared among a set of items, PCF does not distinguish between shared and error variance, but rather tries to explain all variance. Acock recommends to use PCF when you're trying to develop a measure of a concept.

Acock, Alan C. (2016): A Gentle Introduction to Stata. Fifth Edition. TX: Stata Press.