Next: _DASTEPSTATE Up: Multivariate Macros Help File Previous: dasteplook()   Contents

dastepsetup()

Usage:
dastepsetup([Model] [,allin:T or in:logvec] [,silent:T]), Model a
  CHARACTER scalar glm model, usually of the form "y = groups"



Keywords: classification, discrimination, stepwise
You use macro dastepsetup() at the start of a forward or backward
stepwise selection of dependent variables in a discriminant analysis or
more generally in a multivariate linear model.

What it actually does is create and initialize invisible variable
_DASTEPSTATE which encapsulates information on which dependent
variables are "in" and which are "out" at any stage of the variable
selection process. See topic '_DASTEPSTATE'.

The most common use is in stepwise linear discriminant analysis where
you are trying to select a subset of reponse variables that effectively
discriminate among two or more groups.  It can also be used in any
linear model when you are trying to select a subset of reponse
variables that are responsible for any violation of the overall null
hypothesis H0: all model coefficients except constant term are 0.

dastepsetup(Model), where Model is a CHARACTER scalar specifying a GLM
model, initializes _DASTEPSTATE so that no variables are "in" and all
are "out".  This is appropriate at the start of forward stepwise
dependent variable selection.  In linear discrimination analysis, Model
has the form "y = groups", where groups is a factor defining the groups
to be discriminated.

A report of the current status is printed.  This includes all the
F-to-enter statistics and their P-values.

A copy of _DASTEPSTATE is returned as an "invisible" variable which can
be assigned but is not automatically printed.

dastepsetup(Model, silent:T) does the same, except the printed report
is suppressed.

dastepsetup([,silent:T]) does the same, except variable STRMODEL,
usually the most recent GLM model used, is taken as Model.

dastepsetup([Model], allin:T [,silent:T]) does the same, except that
all response variables are "in" and no variables are "out".  Component
'history' of _DASTEPSTATE is initialized to run(p), where p is the
number of variables.

dastepsetup([Model], in:ins [,silent:T]), where ins is a LOGICAL vector
of length p, does the same, except only variables j1, j2 , ... are "in"
where ins[j1], ins[j2] ... are T and the remaining elements are F.
Component 'history' is initialized to vector(j1,j2,...).

After dastepsetup(), your next step is to use daentervar() to enter a
new variable or daremovevar() to remove a variable.  The choice of which
variable to enter or remove is usually made on the basis of the
F-to-enter and/or F-to-remove statistics in the printed report.

See also topics daentervar(), daremovevar(), dastepstatus() and
dasteplook().


Gary Oehlert 2003-01-15