12-07-2015 03:21 PM
I'm unsure whether this is the appropriate board to post this question in, but I'm looking to find a way to interpolate or predict/estimate values from a nonlinear regression that already had it's unknown parameter estimated.
Currently, I've successfully used the PROC NLIN to estimate unknown parameters within my nonlinear regression (yay!). However, I'm looking to use SAS to estimate new functions that would use the parameters predicted in my nonlinear regression. In other words, I want to interpolate information from my nonlinear regression.
In SYSTAT, the equivalent command would be FUNPAR name=function (e.g. FUNPAR LC50=alpha/beta). Using the FUNPAR command in SYSTAT would provide estimates and related statistics (e.g. Standard error, 95%CI, etc.) for each function I write.
Below is the SAS syntax for my nonlinear regression where I'm trying to use my nonlinear regression (R4M2WK_M=(Con+LOG(2)*(MeaConc*(ILC50+KI)/(1+MeaConc*KI))*100)**0.25) to also predict LC50, LC25, and TermK (in italics).
PROC NLIN Data=Clade1Cu METHOD=MARQUARDT MAXITER=1000;
PARMS Con=0.02 ILC50=0.001 KI=0.01
LC50=500 LC25=350 TermK=100;
LC50 = (1/ILC50);
LC25 = 1/((ILC50+KI)/(LOG(4/3)/LOG(2))**(1100)-KI));
TermK = (1/KI);
output out=Clade1Cugraph predicted=R4M2WK_MPre;
I realize that the above syntax doesn't quite work with the italicized commands in SAS.
Is there any syntax or command where I can use SAS to predict a new function using the parameters estimated in the nonlinear regression? Would PROC NLIN still be appropriate for this use? Or is there a better command or syntax for this purpose?
I hope the above makes sense and thank you in advance for any help anyone can offer!
07-20-2016 02:26 AM
As I understand it, you want to calculate predicted values for cases that are not in your sample dataset. To do this, create a dummy dataset with values for the RHS variables, but a missing value for the dependent variable. Append this to your main dataset and run the estimation. SAS will not use the cases with missing dependent variable in the analysis, but will produce predicted values for these cases via the Output statement.