Criteria for Variable Selection (REGRESSION command)
- The
ENTER,REMOVE, andTESTmethods use only theTOLERANCEcriterion. -
BACKWARDremoves variables according to the probability of F-to-remove (keywordPOUT). SpecifyFOUTto use F-to-remove instead. -
FORWARDenters variables according to the probability of F-to-enter (keywordPIN). SpecifyFINto use F-to-enter instead. -
STEPWISEuses bothPINandPOUT(orFINandFOUT) as criteria. If the criterion for entry (PINorFIN) is less stringent than the criterion for removal (POUTorFOUT), the same variable can cycle in and out until the maximum number of steps is reached. Therefore, ifPINis larger thanPOUTorFINis smaller thanFOUT,REGRESSIONadjustsPOUTorFOUTand issues a warning. - The values for these criteria are specified in parentheses. If a value is not specified, the default values are used.
DEFAULTS. PIN(0.05),
POUT(0.10), and TOLERANCE(0.0001). These are the defaults
if CRITERIA is omitted. If criteria
have been changed, DEFAULTS
restores these defaults.
PIN[(value)]. Probability
of F-to-enter. The
default value is 0.05. Either PIN or FIN can be specified.
If more than one is used, the last one specified is in effect.
FIN[(value)]. F-to-enter. The default value is 3.84. Either PIN or FIN can be specified. If
more than one is used, the last one specified is in effect.
POUT[(value)]. Probability
of F-to-remove. The
default value is 0.10. Either POUT or FOUT can be specified.
If more than one is used, the last one specified is in effect.
FOUT[(value)]. F-to-remove. The default value is 2.71. Either POUT or FOUT can be specified.
If more than one is used, the last one specified is in effect.
TOLERANCE[(value)]. Tolerance. The default value is 0.0001. If the specified tolerance is very
low, REGRESSION issues a warning.
MAXSTEPS[(n)]. Maximum number
of steps. The value of MAXSTEPS is the sum of the maximum number of steps for each method for the
equation. The default values are, for the BACKWARD or FORWARD methods, the number of variables meeting PIN/POUT or FIN/FOUT criteria, and for the STEPWISE method, twice the number of independent variables.