Inclusion Levels (DISCRIMINANT command)
When you specify a stepwise method on the METHOD
subcommand (any method other than
the default direct-entry method), you can control the order in which
variables are considered for entry or removal by specifying inclusion
levels on the ANALYSIS
subcommand.
By default, all variables in the analysis are entered according to
the criterion requested on the METHOD
subcommand.
- An inclusion level is an integer between 0 and 99, specified in parentheses after a
variable or list of variables on the
ANALYSIS
subcommand. - The default inclusion level is 1.
- Variables with higher inclusion levels are considered for entry before variables with lower inclusion levels.
- Variables with even inclusion levels are entered as a group.
- Variables with odd inclusion levels are entered individually,
according to the stepwise method specified on the
METHOD
subcommand. - Only variables with an inclusion level of 1 are considered
for removal. To make a variable with a higher inclusion level eligible
for removal, name it twice on the
ANALYSIS
subcommand, first specifying the desired inclusion level and then an inclusion level of 1. - Variables with an inclusion level of 0 are never entered. However, the statistical criterion for entry is computed and displayed.
- Variables that fail the tolerance criterion are not entered regardless of their inclusion level.
The following are some common methods of entering
variables and the inclusion levels that could be used to achieve them.
These examples assume that one of the stepwise methods is specified
on the METHOD
subcommand (otherwise,
inclusion levels have no effect).
Direct. ANALYSIS=ALL(2)
forces all
variables into the equation. (This is the default and can be requested
with METHOD
=DIRECT
or simply by omitting the METHOD
subcommand.)
Stepwise. ANALYSIS=ALL(1)
yields a stepwise
solution in which variables are entered and removed in stepwise fashion.
(This is the default when anything other than DIRECT
is specified on the METHOD
subcommand.)
Forward. ANALYSIS=ALL(3)
enters variables
into the equation stepwise but does not remove variables.
Backward. ANALYSIS=ALL(2) ALL(1)
forces
all variables into the equation and then allows them to be removed
stepwise if they satisfy the criterion for removal.
Inclusion Levels Used With a Stepwise Method
DISCRIMINANT GROUPS=SUCCESS(0,1)
/VARIABLES=A, B, C, D, E
/ANALYSIS=A TO C (2) D, E (1)
/METHOD=WILKS.
- A, B, and C are entered into the analysis first, assuming that they pass the tolerance criterion. Since their inclusion level is even, they are entered together.
- D and E are then entered stepwise. The one that minimizes the overall value of Wilks’ lambda is entered first.
- After entering D and E, the program checks whether
the partial F for either one
justifies removal from the equation (see the
FOUT
andPOUT
subcommands).
Inclusion Levels Without a Stepwise Method
DISCRIMINANT GROUPS=SUCCESS(0,1)
/VARIABLES=A, B, C, D, E
/ANALYSIS=A TO C (2) D, E (1).
- Since no stepwise method is specified, inclusion levels have no effect and all variables are entered into the model at once.