As we learned in Chapter
5, sample size can greatly affect the results of an EFA. Small samples
can lead to the identification of incorrect factor structures and
item loadings (both direction and magnitude). They also limit our
ability to check the replicability of our results through internal
replication. (See Chapter 6.) Thus, small samples can affect your
results and limit your ability to discern how they might affect them!
Now the best solution
to the small sample dilemma is to get more data. More subjects or
observations can be added to a sample to improve the potential replicability
of the results. However, if this is not a possibility (we all have
limited resources and funds!), an alternative solution is to explore
the replicability of the results through bootstrap resampling techniques.
Bootstrap resampling is a subsampling and averaging procedure that
can be used to produce confidence intervals for any and all of the
statistics that we choose. We can get confidence intervals (CIs) around
eigenvalues, communalities, factor loadings, and more. These CIs can
then inform us about the precision of each estimate and the range
in the estimates that we might see among other samples. This information
provides another means through which to examine the replicability
of our results.
The benefits of bootstrap
resampling are not limited to small samples. If we consider what bootstrap
resampling can tell us —the potential error in our estimates
and generalizability to other samples —we can quickly see
the benefit of this practice to any analysis. Although it is true
that large samples are more likely to yield replicable results, this
hypothesis can be tested through bootstrap analysis. These methods
help inform us about the replicability and generalizability of our
results.