|
1
|
Greeting to all SAS users, Quick Stat question: I have the opportunity to use the ORTHPOLY command which apparently orthogonalize some correlated polynomials. In essence is the idea to avoid a full rank matrix and other floating point error? What about the parameters - can they be interpreted the usual way? For instance if i have y=ax-b(x*x) then provided that b< Thanks to all |
|||
|
|
|
1
|
Your question is twolfold. About the purpose of the method, I think you are right. This method seems however to have stability problems (see http://en.wikipedia.org/wiki/Gram%E2%80%93Schmidt_process). With respect to interpretation, the meaning of corresponding parameters is not so easy as a positive value of one parameter refers to several categories (same for negative values). I would suggest you to use the default parametrisation (GLM) which gives good results (even with a full rank matrix) and much more easier to interpret. Kind regards |
|||
|
