1: Note that rotation does
not alter the basic aspects of the analysis, such
as the amount of variance extracted from the items. Indeed, although
eigenvalues might change as factor loadings are adjusted by rotation,
the overall percentage of variance accounted for will remain constant.
[return]
2: Alternatively, you could imagine
rotating each cluster of items toward the axis. It really works out
to be functionally the same.
[return]
3: Researchers
also tend to misinterpret the meaning of “orthogonal”
to mean that factor scores are also uncorrelated. Orthogonal factors
can (and often do) produce factor scores that are correlated (Nunnally
& Bernstein, 1994; Thompson, 2004). For more about factor scores,
see Chapter 9.
[return]
4: However, other authors have argued
that there are few substantive differences between the two oblique
rotations (Fabrigar, Wegener, MacCallum, & Strahan, 1999).
[return]
5: However,
some authors have argued that oblique rotations produce less replicable
results as they might overfit the data to a greater extent. We do
not think there is empirical evidence to support this argument, but
overfitting the data is a concern to all EFA analyses, as we will
discuss later in the book.
[return]
6: In this section we draw heavily
on Thompson (2004), which is always a good reference.
[return]
7: Except for nerds like us trying
to understand all this stuff.
[return]
8: Please note, the PRIORS = SMC
option is required to conduct a factor analysis as opposed to a PCA
when the PRINIT (iterated PAF) method is used. Also, the variables
included in the analysis are referred to as lists, such that the colon
after the EngProb prefix includes all variables in the analysis that
start with the specified prefix.
[return]
9: Please
note, the variables in the previous examples were put into their respective
models according to the theoretical factor each variable was associated
with; thus, the REORDER option would not have provided substantially
different factor loading matrices from what was presented.
[return]
10: This is calculated from the results
in the table entitled “Variance Explained by Each Factor.”
The explained variance is converted to a proportion by dividing it
by the total variance (equal to the number of items in the model;
in this case that is 30) and then multiplying by 100.
[return]