SAVE Subcommand (NLR command)

SAVE is used to save the temporary variables for the predicted values, residuals, and derivatives that are created by the model and the derivatives programs.

  • The minimum specification is a single keyword.
  • The variables to be saved must have unique names on the active dataset. If a naming conflict exists, the variables are not saved.
  • Temporary variables—for example, variables that are created after a TEMPORARY command and parameters that are specified by the model program—are not saved in the active dataset. They will not cause naming conflicts.

The following keywords are available and can be used in any combination and in any order. The new variables are always appended to the active dataset in the order in which these keywords are presented here:

PRED. Save the predicted values. The variable's name, label, and formats are those specified for it (or assigned by default) in the model program.

RESID [(varname)]. Save the residuals variable. You can specify a variable name in parentheses following the keyword. If no variable name is specified, the name of this variable is the same as the specification that you use for this keyword. For example, if you use the three-character abbreviation RES, the default variable name will be RES. The variable has the same print and write format as the predicted values variable that is created by the model program. It has no variable label and no user-defined missing values. It is system-missing for any case in which either the dependent variable is missing or the predicted value cannot be computed.

DERIVATIVES. Save the derivative variables. The derivative variables are named with the prefix D. to the first six characters of the parameter names. Derivative variables use the print and write formats of the predicted values variable and have no value labels or user-missing values. Derivative variables are saved in the same order as the parameters named on MODEL PROGRAM. Derivatives are saved for all parameters, whether or not the derivative was supplied in the derivatives program.

LOSS. Save the user-specified loss function variable. This specification is available only with CNLR and only if the LOSS subcommand has been specified.

Asymptotic standard errors of predicted values and residuals, and special residuals used for outlier detection and influential case analysis are not provided by the [C]NLR procedure. However, for a squared loss function, the asymptotically correct values for all these statistics can be calculated by using the SAVE subcommand with [C]NLR and then using the REGRESSION procedure. In REGRESSION, the dependent variable is still the same, and derivatives of the model parameters are used as independent variables. Casewise plots, standard errors of prediction, partial regression plots, and other diagnostics of the regression are valid for the nonlinear model.

Example

MODEL PROGRAM A=.5 B=1.6.
COMPUTE PSTOP=A*SPEED**B.
NLR STOP /PRED=PSTOP
  /SAVE=RESID(RSTOP) DERIVATIVES PRED.
REGRESSION VARIABLES=STOP D.A D.B /ORIGIN
  /DEPENDENT=STOP /ENTER D.A D.B /RESIDUALS.
  • The SAVE subcommand creates the residuals variable RSTOP and the derivative variables D.A and D.B.
  • Because the PRED subcommand identifies PSTOP as the variable for predicted values in the nonlinear model, keyword PRED on SAVE adds the variable PSTOP to the active dataset.
  • The new variables are added to the active dataset in the following order: PSTOP, RSTOP, D.A, and D.B.
  • The subcommand RESIDUALS for REGRESSION produces the default analysis of residuals.