Abstract:
Exploratory Factor Analysis (EFA) is a technique to explore the underlying factors of a large
set of observed variables which cannot be measured directly. In general there are seven types
of factor extraction methods. For a meaningful interpretation of occurred factor model, the
extraction method usually followed by either Orthogonal rotation or Oblique rotation
method. However it has not been recommended a particular method of EFA for a given set
of data. Further most of the researchers are misusing Principal Component Analysis (PCA)
with Exploratory Factor Analysis. Therefore, this study was carried out to investigate a
possibility of recommending a particular method for a given set of data using a data set
comprising seven variables on crimes. Data were analysed using the statistical software
SPSS.
To illustrate the contrast of PCA and EFA, analysis was begun with Principle Component.
For the comparison of different types of extraction methods under EFA, variables were
extracted using Maximum Likelihood Factoring, Principle Axis Factoring and General Least
Squares followed by all the Orthogonal rotation methods separately. The steps of the
analysis in EFA were quite same with all the extraction methods, however the final result
and the effect of the prior assumptions make difference. It is very important to confirm that
KMO statistic to be greater than 0.6, prior to carry out EFA for the adequacy of sample size
in order to derive valid statistical inferences. If the variables having multivariate normal
distribution it is recommended to conduct Maximum Likelihood or General Least Squares.
For the non normal distributions, Principle axis factoring is recommended. However it is
recommended to compare the results from each method irrespective of the distribution of
data set.
Among Orthogonal rotations Varimax rotation is recommended as it provides simple factor
loadings to interpret. Quartimax generally does not provide simple factor loadings as in
Varimax. It is not recommended to carry out ail possible combinations of factor extraction
methods and rotation methods to any set of data, as same results will not be produced by
each combination. The recommendation given for the particular data set was confirmed
using Jackknife validation method.
