CANON <data>,L
computes canonical correlations and variables for two sets of variables.
The first set (X variables) should be denoted by 'X' masks and the
second set (Y variables) by 'Y' masks.
IND and CASES specifications are allowed for selecting observations.
CANON gives the results from line L (optional) onwards. Furthermore
the following matrix files are produced by CANON:
LCAN.M vector of canonical correlations
XCAN.M correlations of canonical variables with X variables
YCAN.M correlations of canonical variables with Y variables
XCOEFF.M coefficients for the first set of canonical variables
YCOEFF.M coefficients for the second set of canonical variables
Values of canonical variables can be computed afterwards by using
XCOEFF.M and YCOEFF.M in the LINCO operation.
For example, LINCO <data>,XCOEFF.M(C1,C2).
The calculations needed are based on orthogonalized data matrices (not
on correlations).
For educational purposes (showing usage of MAT operations) also a sucro
/CANCORR is available. To see its usage, activate /CANCORR without
parameters.
Optionally confounding variables, whose effects on X and Y variables
are removed by linear regression, may be included. Confounding variables
are to be activated by 'Z'. All the results obtained then from CANON
are related to the transformed X and Y data sets.
CANON creates several work matrices (names initiated by &). These are
automatically deleted, unless RESULTS=100 is given. All the work
matrices are on the current data disk.
The MATRUN operations CANON2 and CANON3 work on the same basis as CANON.
See e.g.
MATRUN CANON2?
M = More information on multivariate analysis
Help System (web edition)