09-30-2014 04:44 PM
How do I interpret the fit statistics of proc LOESS to know if the form is a good fit? I ran PROC LOESS to explain price sensitivity (which appeared to me to be nonlinear, and not explained well with a curve). When I plot predicted values along with the actual behavior, the LOESS clearly explains it nicely, but, given the limited fit output (residual sum of squares, etc...) how can I assess the output? Thanks!
10-01-2014 01:42 PM
You can specify PLOT=DIAGNOSTICS on the PROC LOESS statement and look at the resulting diagnostics panel. Those are the same diagnostics plots that are produced by other regression routines.
If you want to be sure that the value of the smoothing parameter is the optimal value for the data, you can evaluate way that the smoothing parameter is seleted by requesting a CRITERIONPLOT. That shows the AICC criterion for various smoothing parameters.See the second example of this article: How to automatically select a smooth curve for a scatter plot in SAS - The DO Loop