Many of us, even those with years of experience using
EFA, remain unclear on some of the nuances and details of what exactly
is happening “under the hood” when we perform this analysis.
Sticking with the default settings in most modern statistical software
will generally not lead to using best practices. In PROC
FACTOR
, the default method of extraction is actually
not even a method of factor analysis — it’s PCA! This
is a solid choice if you were a psychologist in the 1960s, but in
the 21st century, we can do better. So, pay attention: the extraction
method must be specified in order for a factor analysis to be conducted.
As for which extraction
method to use—our examples in this chapter demonstrated that
most extraction techniques can be used when the data has a clear factor
structure and meets basic assumptions of normality. When the data
does not meet assumptions of normality, the iterated PAF or ULS extraction
techniques can provide the best estimates. This led us to conclude
that extraction method can matter more when assumptions are violated
and less when assumptions are met.
However, there is general
consensus in the literature that ML is the preferred choice for when
data exhibits multivariate normality and iterated PAF or ULS for when
that assumption is violated (Fabrigar, Wegener, MacCallum, & Strahan,
1999; Nunnally & Bernstein, 1994). Other extraction techniques
seem to be vulnerable to violations of this assumption, and do not
seem to provide any substantial benefit. Thus, the general recommendation
to use either ML, iterated PAF, or ULS seems sensible.