The LOGSELECT Procedure
SELECTION Statement
SELECTION <METHOD=method<(method-options)>><options>;
The SELECTION statement performs model selection by examining whether effects should be added to or removed from the model according to rules that are defined by model selection methods. The statement is fully documented in the section SELECTION Statement in Chapter 3: Shared Concepts.
The LOGSELECT procedure supports the following effect-selection methods in the SELECTION statement:
- BACKWARD
performs backward elimination. This method starts with all effects in the model and deletes effects.
- BACKWARD(FAST)
performs fast backward elimination. This method starts with all effects in the model and deletes effects without refitting the model.
- FORWARD
performs forward selection. This method starts with no effects in the model and adds effects.
- LASSO
performs model selection by the group LASSO method. This method adds and removes effects by using a sequence of LASSO steps. For more information, see the section Group LASSO Selection in Chapter 3: Shared Concepts.
- NONE
results in no model selection. This method fits the full model.
- STEPWISE
performs stepwise selection. This method is similar to the FORWARD method except that effects already in the model do not necessarily stay there.
By default, METHOD=STEPWISE.
The SELECT=, CHOOSE=, and STOP= method-options default to SBC. These defaults differ from their corresponding defaults in the HPLOGISTIC procedure in SAS/STAT software.
You can specify the following criteria in the SELECT=, CHOOSE=, and STOP= method-options:
- AIC
uses Akaike’s information criterion (Akaike 1974) computed on the training data.
- AICC
uses a small-sample bias-corrected version of Akaike’s information criterion, as promoted in Hurvich and Tsai (1989) and Burnham and Anderson (1998), computed on the training data.
- SBC | BIC
uses the Schwarz Bayesian criterion (Schwarz 1978) computed on the training data.
- SL
uses the significance level of the score test computed on the training data as the criterion (not available for the CHOOSE= option).
- VALIDATE
uses the average square error (ASE) computed on the validation data as the criterion (not available for the SELECT= option).
For more information, see the section Information Criteria.
If you specify METHOD=LASSO and you do not specify either the CHOOSE= or STOP= option, then the model in the last LASSO step is chosen as the selected model.
If you specify METHOD=LASSO, then the STOPHORIZON= option has no effect.
Note: If you use the fast backward elimination method, then the –2 log likelihood, AIC, AICC, and SBC statistics are approximated at each step where the model is not refit, and hence they do not match the values that are computed when that model is fit outside the selection routine. Similarly, if you specify SELECT=AIC, AICC, or SBC, the selection criteria are estimated (Lawless and Singhal 1978), and hence they do not match the values that are computed when that model is fit outside the selection routine.
Note: The default model hierarchy method is HIERARCHY=NONE for the stepwise, forward, and fast backward selection methods. The backward elimination method always uses the HIERARCHY=SINGLE method-option.
The LASSO method produces a summary table that displays the effects that are added or removed at each step; the LASSO regularization parameter; and the AIC, AICC, and SBC fit statistics. For the other methods, you can specify the following values for the DETAILS= option:
- SUMMARY
produces a summary table that shows the effect that is added or removed at each step along with the SELECT=, CHOOSE=, and STOP= criteria. The summary table is produced by default if the DETAILS= option is not specified.
- STEPS
produces the preceding summary table and displays the results from fitting each model at each step.
- ALL
produces the preceding tables and a detailed listing of all candidates at each step along with their ranking in terms of the selection criterion for addition to or removal from the model.
If you specify the PLOTS=CRITERIA or PLOTS=ALL option, then a plot of the fit criterion by the selection step is created for the AIC, AICC, and SBC statistics. If you also specify a PARTITION statement or the PARTFIT option, then the same type of plot is created for the ASE (for each role), max-rescaled R-square, McFadden’s R-square, misclassification rate, and difference of means statistics.
If you specify the PLOTS=FITBYROLE or PLOTS=ALL option and a PARTITION statement, then a plot of the ASE by the selection step for each role is created.
The PLOTS= option is not available for the LASSO method.