In our mind, the primary issue to be decided prior
to a higher-order factor analysis is whether the
initial factor structure is appropriate or not. If we extract five
factors, and then decide there is a single second-order factor (or
even two second-order factors), the first question we would ask is
whether the original solution was correct, or whether the correct
first-order structure should have been one (or two) factors rather
than five.
In the seven or eight
decades since this discussion began in earnest, many things have changed
in quantitative methods. One is the easy access to confirmatory factor
analysis techniques. Although we have not discussed confirmatory techniques,
they are methods for directly testing hypotheses such as whether a
particular data set is best characterized as one, two, or five factors.
Thus, before launching into higher-order factor analysis, we would
evaluate (and replicate) whether the initial solution was correct.
We suspect that, in many cases, the initial solution was indefensible
or suboptimal, and that the “higher-order factors” are
really just the more parsimonious version of what should have been
extracted initially. We would recommend exploring higher-order factors
only after the initial factor structure has been thoroughly vetted
through CFA as the most parsimonious and desirable. Of course, once
in CFA, higher-order factors can be modeled and tested in that confirmatory
framework!
If you want to explore
this aspect of your data, we can, in the spirit of intrepid exploration,
briefly cover some of the mechanics of the process.