The HPQUANTSELECT Procedure

PROC HPQUANTSELECT Contrasted with Other SAS Procedures

For general contrasts between SAS High-Performance Analytics procedures and other SAS procedures, see the section Common Features of SAS High-Performance Statistical Procedures in SAS/STAT 14.1 User's Guide: High-Performance Procedures. The following remarks contrast the HPQUANTSELECT procedure with the QUANTSELECT and QUANTREG procedures in SAS/STAT.

The major functional differences between the HPQUANTSELECT and QUANTSELECT procedures are as follows:

  • The HPQUANTSELECT procedure uses an interior point algorithm for model fitting. The QUANTSELECT procedure uses a flexible simplex algorithm for model fitting.

  • The HPQUANTSELECT procedure can output confidence limits for parameter estimates.

  • The HPQUANTSELECT procedure does not support the LASSO and adaptive LASSO effect-selection methods.

  • The HPQUANTSELECT procedure does not support the TESTDATA= and VALDATA= options in its PROC statement.

  • The HPQUANTSELECT procedure does not support graphical summaries for the effect selection processes.

Both the HPQUANTSELECT and QUANTSELECT procedures support the forward, backward, and stepwise effect-selection methods and the ability to use separate validation and test data via the PARTITION statement. For more information about the QUANTSELECT procedure, see Chapter 96: The QUANTSELECT Procedure.

The major functional differences between the HPQUANTSELECT and QUANTREG procedures are as follows:

  • The QUANTREG procedure does not support any effect-selection methods. It does not output the following fit statistics: AIC, AICC, SBS, VALIDATE, TEST, R1, and adjusted R1. And it does not support the PARTITION statement.

  • The QUANTREG procedure provides three algorithms for fitting quantile regression models: simplex algorithm, interior point algorithm, and smoothing algorithm. The HPQUANTSELECT procedure supports only the interior point algorithm.

  • The QUANTREG procedure supports the rank test, which is not available for the HPQUANTSELECT procedure. Both the QUANTREG and HPQUANTSELECT procedures support the Wald test and the likelihood test.

  • The QUANTREG procedure supports two methods of estimating the covariance matrix of the parameter estimates: an asymptotic method and a bootstrap method. The HPQUANTSELECT procedure supports only the asymptotic method.

For more information about the QUANTREG procedure, see Chapter 95: The QUANTREG Procedure.

The HPQUANTSELECT procedure is also different from the QUANTSELECT and QUANTREG procedures in the following respects:

  • The HPQUANTSELECT procedure supports the CODE statement, which is not available in the QUANTSELECT and QUANTREG procedures.

  • The HPQUANTSELECT procedure does not support quantile process regression, whereas the QUANTREG procedure does support quantile process regression. The QUANTSELECT procedure supports effect selection for quantile process regression.

  • The HPQUANTSELECT procedure does not support the EFFECT statement, which provides constructed effects such as polynomial effects, spline effects, collection effects, and multimember classification effects.

  • The HPQUANTSELECT procedure can output confidence limits for mean predicted quantiles; this functionality is not available in the QUANTSELECT and QUANTREG procedures.

In addition to having many similarities to the QUANTSELECT and QUANTREG procedures, the HPQUANTSELECT procedure also compares closely to the HPREG procedure. PROC HPREG is a high-performance procedure that performs effect selection in the framework of general linear models. The HPQUANTSELECT procedure inherits most of its syntax from the HPREG and QUANTREG procedures. The HPQUANTSELECT procedure provides results that are similar to those of the HPREG and QUANTSELECT procedures.