METHOD Subcommand (REGRESSION command)
METHOD specifies a variable selection method and names a block of variables to be evaluated using that method. METHOD is required.
- The minimum specification is a method keyword and, for some methods, a list of variables. The actual keyword METHOD can be omitted.
- When more than one METHOD subcommand is specified, each METHOD subcommand is applied to the equation that resulted from the previous METHOD subcommands.
- The default variable list for methods FORWARD, BACKWARD, STEPWISE, and ENTER consists of all variables named on VARIABLES that are not named on the DEPENDENT subcommand. If VARIABLES=(COLLECT), the variables must be specified for these methods.
- There is no default variable list for the REMOVE and TEST methods.
- Keyword TO in a variable list on METHOD refers to the order in which variables are specified on the VARIABLES subcommand. If VARIABLES=(COLLECT), TO refers to the order of variables in the active dataset.
The available stepwise methods are as follows:
BACKWARD [varlist]. Backward elimination. Variables in the block are considered for removal. At each step, the variable with the largest probability-of-F value is removed, provided that the value is larger than POUT. See the topic CRITERIA Subcommand (REGRESSION command) for more information. If no variables are in the equation when BACKWARD is specified, all independent variables in the block are first entered.
FORWARD [varlist]. Forward entry. Variables in the block are added to the equation one at a time. At each step, the variable not in the equation with the smallest probability of F is entered if the value is smaller than PIN. See the topic CRITERIA Subcommand (REGRESSION command) for more information.
STEPWISE [varlist]. Stepwise selection. If there are independent variables already in the equation, the variable with the largest probability of F is removed if the value is larger than POUT. The equation is recomputed without the variable and the process is repeated until no more independent variables can be removed. Then, the independent variable not in the equation with the smallest probability of F is entered if the value is smaller than PIN. All variables in the equation are again examined for removal. This process continues until no variables in the equation can be removed and no variables not in the equation are eligible for entry, or until the maximum number of steps has been reached. See the topic CRITERIA Subcommand (REGRESSION command) for more information.
The methods that enter or remove the entire variable block in a single step are as follows:
ENTER [varlist]. Forced entry. All variables specified are entered in a single step in order of decreasing tolerance. You can control the order in which variables are entered by specifying the variables on multiple METHOD=ENTER subcommands.
REMOVE varlist. Forced removal. All variables specified are removed in a single step. REMOVE requires a variable list.
TEST (varlist) (varlist). R ^{2} change and its significance for sets of independent variables. This method first adds all variables specified on TEST to the current equation. It then removes in turn each subset from the equation and displays requested statistics. Specify test subsets in parentheses. A variable can be used in more than one subset, and each subset can include any number of variables. Variables named on TEST remain in the equation when the method is completed.
Example
REGRESSION VARIABLES=POP15 TO GROWTH, SAVINGS
/DEPENDENT=SAVINGS
/METHOD=STEPWISE
/METHOD=ENTER.
- STEPWISE applies the stepwise procedure to variables POP15 to GROWTH.
- All variables not in the equation when the STEPWISE method is completed will be forced into the equation with ENTER.
Example
REGRESSION VARIABLES=(COLLECT)
/DEPENDENT=SAVINGS
/METHOD=TEST(MEASURE3 TO MEASURE9)(MEASURE3,INCOME)
/METHOD=ENTER GROWTH.
- The VARIABLES=(COLLECT) specification assembles a correlation matrix that includes all variables named on the DEPENDENT and METHOD subcommands.
- REGRESSION first builds the full equation of all the variables named on the first METHOD subcommand: SAVINGS regressed on MEASURE3 to MEASURE9 and INCOME. For each set of test variables (MEASURE3 to MEASURE9, and MEASURE3 and INCOME), the R ^{2} change, F, probability, sums of squares, and degrees of freedom are displayed.
- GROWTH is added to the equation by the second METHOD subcommand. Variables MEASURE3 to MEASURE9 and INCOME are still in the equation when this subcommand is executed.