The NLIN Procedure

PROFILE Statement

PROFILE parameter <…parameter> </ options> ;

A PROFILE statement selects parameters for profiling to assess their nonlinear characteristics. It can also gauge the influence of each observation on the selected parameters. To produce the plots that are controlled by the PROFILE statement, ODS Graphics must be enabled. If there are observations in the main data set that PROC NLIN deems unusable, the procedure issues a message stating that these observations are excluded from the parameter profiling. PROC NLIN ignores PROFILE statements for nonconvergent and singular models.

Table 67.8 summarizes the options available in the PROFILE statement.

Table 67.8: Summary of Options in PROFILE Statement

Option	Description
ALL	Produces all profiling related plots and statistics
CONFCURV	Produces confidence curves for the listed parameters
JACKKNIFE	Produces plots of the absolute relative percentage change by observation for the listed parameters
PROFDATA=	Specifies the profile output data set
RANGE=	Specifies a profiling range
TPLOT	Produces profile t plots for the listed parameters

ALL

produces all appropriate plots for the parameters that are listed in the PROFILE statement. This option overrides all other options. If the PROFDATA= option is specified, the values of the corresponding statistics are written to the data set.

CONFCURV

produces confidence curves for each parameter listed in the PROFILE statement. If the PROFDATA= option is specified, the values of the corresponding statistics are written to the data set. See the section Profiling Parameters and Assessing the Influence of Observations on Parameter Estimates for details.

JACKKNIFE

produces, for each parameter listed in the PROFILE statement, plots of absolute relative percentage change in the value of the parameter that results from a fit to a jackknife-resampled datum versus observation number. The jackknife resampling is performed with a leave-one-out method. If the PROFDATA= option is specified, the values of the corresponding statistics are written to the data set. See the section Profiling Parameters and Assessing the Influence of Observations on Parameter Estimates for details.

PROFDATA=SAS-data-set

specifies the SAS data set to be created by PROC NLIN when a PROFILE statement is included. See the section Output Data Sets for details about this data set.

RANGE= $\gamma _ f$ to $\gamma _ t$ by $\gamma _ b$

specifies a profiling range for each parameter listed in the PROFILE statement. $\gamma _ f$ , $\gamma _ t$ and $\gamma _ b$ represent the starting, final, and consecutive incremental values of the range in terms of the standard errors of the parameters to be profiled. If you do not specify the RANGE= option, PROC NLIN uses –2, 2, and 0.2 as default values for $\gamma _ f$ , $\gamma _ t$ , and $\gamma _ b$ , respectively. For example, if parameter $\beta$ is in the PROFILE statement, then during the profiling of $\beta$ , PROC NLIN performs approximately ( $\gamma _ t$ – $\gamma _ f$ )/ $\gamma _ b$ number of constrained optimizations. At the ith constrained optimization, the value of $\beta$ is constrained to $\beta _ i$ ,

$\beta _ i = \hat{\beta } + \mbox{stderr}_{\beta } \left(\gamma _ f + i\, \frac{\left(\gamma _ t - \gamma _ f \right)}{\gamma _ b} \right)$

and the rest of the parameters are freely optimized. In the preceding expression, $\hat{\beta }$ and $\mbox{stderr}_{\beta }$ are the estimated value and standard error, respectively, for $\beta$ . If parameter bounds are specified with BOUNDS statements, PROC NLIN enforces those bounds during the profiling optimizations.

Note: The profiling optimizations are performed using quasi-Newton algorithm. During these optimizations, any optimization options specified in the PROC NLIN statement and the _LOSS_ variable are ignored.

TPLOT

produces profile t plots for each parameter listed in the PROFILE statement. If the PROFDATA= option is specified, the values of the corresponding statistics are written to the data set. See the section Profiling Parameters and Assessing the Influence of Observations on Parameter Estimates for details.