With SAS 9.2, PROC TRANSREG no longer always prints an iteration history table by default, and in some cases, the table it prints is not the same as it was previously. This change is due to the increasing use of PROC TRANSREG with transformations that are not based on alternating least squares. Here is some background for the change. PROC TRANSREG’s processing can be divided into three steps. In the first step, the data are read and certain transformations, such as SMOOTH, PBSPLINE, BOXCOX, RANK, LOG and the other nonoptimal transformations, are performed. These transformations are not based on OLS. In the second step, the alternating least squares iterations are performed according to METHOD=UNIVARIATE, MORALS, REDUNDANCY, or CANALS. It is in the second step that the alternating least squares transformations (SPLINE, MSPLINE, MONOTONE, OPSCORE, LINEAR, and UNTIE) are iteratively found. In the third step, the results are displayed. In some cases, the results are appropriately based on using the method of OLS applied to the optimally transformed variables. In other cases, such as with smoothing splines and penalized B-splines, OLS-based results are not appropriate. Furthermore, for many of these types of models, nothing changes in the iterations, so the computations needed to realize that nothing changes are not needed, nor is the iteration history table.
With SAS 9.2, the iteration history is not printed for models where it is known that nothing will change in the iterations. Suppose the NOMISS option is specified or there are no missing data. If METHOD=UNIVARIATE, if there are no iterative transformations (SPLINE, MSPLINE, MONOTONE, OPSCORE, LINEAR, and UNTIE), and if the MAXITER= option is not specified, then by default, an iteration history table is not produced. If you want to see an iteration history, there are many things you can do, such as specifying MAXITER=, changing the method to MORALS, or changing IDENTITY to LINEAR.
With models with smoothing splines or penalized B-splines, the iteration history will not contain an R square. This is because the iterations are based on the method of alternating least squares, but the smoothing splines and penalized B-splines are not based on a least squares model. Hence, an ordinary R square in the iterations, based on a computed intercept, which is typically not exactly zero, and a computed slope, which is typically not exactly one, will not be exactly the same as the correct R square, which is based on an intercept and slope of zero and one. The final reported results include the correct R square in the fit statistics table after the ANOVA table. If you want to see only the correct R square from the results, without the iteration history, you can specify the new RSQUARE option.