PCA and hierarchical and non-hierarchical CA were
performed on data sets of each site using Statistica 6 software.
PCA analysis was performed on the concentration
matrix following the data scaling procedure [31]. The elements
that were not quantified (under MDL or comparable
with signal of blank filter) in more than 20% of the
samples have been excluded from the analysis similarly
to the selection reported in literature [32]. For data not
quantified values equal to the maximum between the
detection limit and the standard deviation of blank filters
have been used in the analysis [4]. The cluster analysis is
an effective multivariate statistical method that could be
used to achieve a greater confidence in the classification
obtained with PCA. It has been shown that CA can assist
PCA in the identification of clusters in atmospheric aerosol
particles [33]. Therefore in this work the cluster analysis
was essentially used for comparison purposes.
Dendrogram has been obtained using Ward’s method.