A central hallmark of
science is replication (Killeen, 2008; Thompson, 2002; Yu, 2003).
Unfortunately, replication in EFA can be quite tricky! Even with very
large samples, the factor structure and parameter estimates are often
unstable or inaccurate reflections of the population. In addition,
there is no single effect or test of statistical significance that
can be compared across samples—instead there are communalities,
eigenvalues, factor loadings, and much more.
Despite the difficulties,
replication procedures have a lot to offer EFA. In Chapter 6, we explored
a more classic replication methodology of splitting an existing sample
(internal replication) or gathering two independent samples (external
replication). These classic methods are not as useful with small samples,
might require more resources (to recruit larger or duplicate samples),
and they can inform us only of the replicability of our estimates
in an adjacent sample. However, they do indeed address the question
of replicability and provide us with valuable information. In the
current chapter we explored an alternative replication method that
complements and, in some ways, improves upon the previous methodology.
This method of bootstrap resampling seeks to gain insight into the
population parameters through the estimation of CIs. While not perfect,
resampling methods can inform the researcher as to the relative stability
or instability (i.e., replicability or non-replicability) of an effect
or result. To our knowledge, this technique has not been widely applied
to EFA.
Although these procedures
can be somewhat time-consuming and programmatically complex, we hope
you see the myriad of benefits that they can offer. We rarely have
the ability to know the “true” factor structure in the
population (leaving aside the fact that factor structures can vary
across subpopulations). Most of the time we have only a sample, a
hope that the factor structure will match our theoretical model or
models, and optimism that our results will generalize to new samples
and studies. The past century or so of exploratory factor analysis
has been almost entirely devoted to exploring data in a theoretical
way, or seeking to confirm that an instrument matches a theoretical
model. It has been rare to see any attention given to what should
be at least as important—whether the findings will generalize.
In this chapter, we described and explored a methodology to move in
that direction. It is not a perfect methodology, as a poor sample
will lead to bootstrap analyses that are poor — but, at the
least, we can show how precise our results are and how confident we
can be about those estimates. As Bruce Thompson has suggested, bootstrap
methods can give us valuable information about our results. It can
help researchers gain information about their results that they might
otherwise not have.