Faktoranalyse—hold deg unna varimax rotasjon !
Av Ronny Klæboe Dato: 4.5.2009
Utsnitt fra innlegg på SEMNET (nyhetsliste) om faktoranalyse fra en av ekspertene på området:
...... Lengre utredning om hvordan Thurstone oppfattet “simple structure” og hvor faktorer tillates å korrellere.
Don’t do Kaiser’s varimax rotation! Orthogonal common factors are a researcher artifact.
You cannot derive second-order common factors from uncorrelated factors either. Kaiser’s method was perhaps the easiest way mathematically challenged researchers could understand rotated factors, but it was based on a missleading concept of simple structure.
Det er svært vanlig at folk bruker varimax-rotering. Det er vel den som også er default i SPSS.
If you are using direct oblimin (min utheving/rk), you are getting a near optimal simple structure solution.
There may be no such thing as a unique
simple structure solution. Rozeboom first reported local minima (many) for some rotational schemes to get simple structure.
There are some new rotational methods that yield very good simple structure solutions, but you have to run the programs with different starting solutions to find the one that seems better than all the others. I have a revision of my factor analysis book coming out this fall that has a new chapter describing these new rotational methods as well as Jennrich’s gradient projection algorithm that allows us to obtain all of the rotational solutions by simple changes in the program code.