Hello!
I'm
using the NLPCA to reduce the
dimensionality of nine variables
(4 nominal /
3 ordinal /2 numeric)
to obtain the object-scores to be used as dependent variable in a
regression model.
I'm using the package homals (http://www.jstatsoft.org/v31/i04/paper).
The output is:
Call: homals (date = date, Ndim = 1, rank = 1, level = c ("numerical", rep
("ordinal", 3), "numerical",
rep ("nominal", 4), active = TRUE)
Loss: 0.0002050824
Eigenvalues??: D1 0.0212
I'm having
the following questions:
1)
Is it best to consider Ndim = rank = 1 or Ndim = rank = max (rank) to
reduce the dimensionality of data?
2) Is there a command to automatically calculate
the proportion of variance explained
by the first component? Otherwise, how can I calculate it by hand?3) Is it
necessary to standardize numeric variables before perfoming "homals"?
If anyone
has any thoughts for this, responses would be greatly appreciated.
Thanks.
Lucia
[[alternative HTML version deleted]]