Hello everyone
I would like to carry out PCA on a group of variables; some are of a
continuous nature, some ordinal and some binary. My question is: can
correlation-based PCA be performed using such a group of variables (i.e.
based on the covariance matrix of the standardised variables)? This would
avoid certain variables having undue influence on the PCs. I am concerned
as some of the variables making up the group are binary and I wondered if
correlation-based PCA is appropriate for these data? Perhaps I should
instead perform covariance based PCA where the variables are not
standardised?
Many thanks for your views,
Kim.
|