The TRANSREG Procedure
- Overview
-
Getting Started
-
Syntax
-
Details
Model Statement Usage Box-Cox Transformations Using Splines and Knots Scoring Spline Variables Linear and Nonlinear Regression Functions Simultaneously Fitting Two Regression Functions Penalized B-Splines Smoothing Splines Smoothing Splines Changes and Enhancements Iteration History Changes and Enhancements ANOVA Codings Missing Values Missing Values, UNTIE, and Hypothesis Tests Controlling the Number of Iterations Using the REITERATE Algorithm Option Avoiding Constant Transformations Constant Variables Character OPSCORE Variables Convergence and Degeneracies Implicit and Explicit Intercepts Passive Observations Point Models Redundancy Analysis Optimal Scaling OPSCORE, MONOTONE, UNTIE, and LINEAR Transformations SPLINE and MSPLINE Transformations Specifying the Number of Knots SPLINE, BSPLINE, and PSPLINE Comparisons Hypothesis Tests Output Data Set OUTTEST= Output Data Set Computational Resources Unbalanced ANOVA without CLASS Variables Hypothesis Tests for Simple Univariate Models Hypothesis Tests with Monotonicity Constraints Hypothesis Tests with Dependent Variable Transformations Hypothesis Tests with One-Way ANOVA Using the DESIGN Output Option Discrete Choice Experiments: DESIGN, NORESTORE, NOZERO Centering Displayed Output ODS Table Names ODS Graphics
-
Examples
- References
| Centering |
You can use transformation options to center and standardize the variables in several ways. For example, the following MODEL statement creates three independent variables, x, 
, and 
:
model identity(y) = pspline(x);
The variables are not centered.
When the CENTER t-option is specified, as in the following statement, the independent variable is centered before squaring and cubing:
model identity(y) = pspline(x / center);
The three independent variables are 
, 
, and 
.
Since operations such as squaring occur after the centering, the resulting variables are not always centered. The CENTER t-option is particularly useful with polynomials since centering before squaring and cubing can help reduce collinearity and numerical problems. For example, if one of your variables is year, with values all greater than 1900, squaring and cubing without centering first will create variables that are all essentially perfectly correlated.
When the TSTANDARD=CENTER t-option is specified, as in the following model, the three independent variables are squared and cubed and then centered:
model identity(y) = pspline(x / tstandard=center);
The three independent variables are , 
, and 
.
You can specify both the CENTER and TSTANDARD=CENTER t-options to center the variables, then square and cube them, and then center the results, as in the following statement:
model identity(y) = pspline(x / center tstandard=center);
The three independent variables are , 
, and 
.