> My question: what does it mean asymmetry distribution could
> affect PCA ? and also outliers could affect factors?
It means what it says. PCA will be affected by asymmetry and outliers will
affect the principal components (sometimes loosely called 'factors') In
particular an extreme outlying data point can cause at least one PC to be
essentially parallel to the vector between the outlier and the mean of the rest
of the data. If you want a picture of factors describing the bulk of the data
set, you need to chuck out the extreme points or use robust PCA.
Asymmetry I'd worry less about, at least for exploratory graphical
presentation; if I had a nice spherical data set I'd probably not be very
interested in the PCA because it'd not have much discriminatory power for
groups. But inference based on things like mahalanobis distance often relies
on some sense of multivariate normality or the like, and if the model used for
inference isn't built on a symmetric data set the inferences can be badly
wrong. Think Turkish flag; the star is 'obviously' not part of the crescent,
but in mahalanobis distance it's not much further from the (empty) centre of
the crescent than most of the crescent is.
*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
