The NLIN Procedure
- Overview
-
Getting Started
-
Syntax
-
Details
Automatic Derivatives Measures of Nonlinearity and Diagnostics Missing Values Special Variables Troubleshooting Computational Methods Output Data Sets Confidence Intervals Covariance Matrix of Parameter Estimates Convergence Measures Displayed Output Incompatibilities with SAS 6.11 and Earlier Versions of PROC NLIN ODS Table Names ODS Graphics
-
Examples
- References
| Displayed Output |
In addition to the output data sets, PROC NLIN also produces the following output objects:
the residual sums of squares associated with all or some of the combinations of possible starting values of the parameters
the estimates of the parameters and the residual sums of squares at each iteration
the estimation summary table, which displays information about the estimation method, the number of observations in the analysis, the objective function, and convergence measures
the analysis of variance table, including sums of squares for the "Model," "Residual," and "Total" sources of variation ("Corrected Total" or "Uncorrected Total"), and the model F test. Note that beginning in SAS® 9, only the uncorrected total SS is reported and the respective F test is based on the uncorrected total SS if PROC NLIN determines the model does not include an intercept. If PROC NLIN determines the model does include an intercept, only the corrected total SS is reported and the respective F test is based on the corrected total SS.
the table of parameter estimates, which contains for each parameter in the model its estimate, the approximate standard error of the estimate, and a 95% confidence interval based on the approximate standard error. The confidence level can be changed with the ALPHA= option in the PROC NLIN statement. The HOUGAARD option in the PROC NLIN statement requests that Hougaard’s skewness measure be added for each parameter. The standard errors and confidence limits are labeled approximate because they are valid asymptotically as the number of observations grows. If your model is linear in the parameters, the standard errors and confidence intervals are not approximate.
the approximate correlation matrix of the parameter estimates. This correlation matrix is labeled approximate because it is computed from the approximate covariance matrix of the parameter estimates. If your model is linear in the parameters, the correlation matrix is not approximate.