Al possible subsets regression is a recommended supplementary analysis to stepwise procedures.
It performs a multiple regression using all the possible combinations of predictor variables and compares these with respect to various statistical criteria.
For example, if we have three predictors, x1, x2 and x3 then there are seven possible models featuring the seven different combinations of the predictors: x1, x2, x3, x1 x2, x1 x3, x2 x3 and x1 x2 x3.
The interest is in total model prediction (R-squared) rather than the performance of individual predictors.
By comparing a range of models simultaneously we can see just how much better the optimal combination of predictors accounts for the most variation in outcome compared to the other models.
It may be the best combination of predictors accounts for far more variation than any other combination, or alternatively, several predictor combinations may do almost as good a job of predicting the outcome. This is information which you don't get out of stepwise methods.
Unlike stepwise methods you need special syntax to run all possible subsets regressions in SPSS.
The programs all_possible.sps (outputs statistical model criteria for all regression models) and all_possi2.sps (plots statistical criteria) perform all possible subsets analysis.
Run the all_possible.sps first in a syntax window in SPSS - please read the instructions at the top of the file first. Then upon successful completion run all_possi2.sps to plot the statistical criteria for all the models fitted. The outputted criteria are Mallows Cp, adjusted R-squared and mean squared error.