|
Hey Krishna:
The model may be indicating "non-convergence", but in actuallity is just fine. Since you have 3-parameters and 3-observations, the model reduces to a system of 3-equations and 3-unknowns. Due to the properties of least squares, your "predicted line" should pass through these 3 points (x,y). However, you will not be able to estimate a "residual variance", since you have no residual degrees of freedom. The MSE should = ~ 0. One way to check this would be to look at the residuals (predicted-observed), or (observed-predicted , but the former makes for much easier interpretation!!!). If these residuals are 0, then you should be in the ball park. Note that SAS NLIN will not give exact values of "0", more likely some really small #, i.e. 1.0E-8.
lewis
*****************************
Lewis Jordan
Weyerhaeuser:
Southern Timberlands Group
Cell (Primary): 662-889-4514
Office: 662-245-5227
lewis.jordan@weyerhaeuser.com
*****************************
________________________________________
From: SAS(r) Discussion [SAS-L@LISTSERV.UGA.EDU] On Behalf Of SUBSCRIBE SAS-L Anonymous [rkbattula@DRREDDYS.COM]
Sent: Wednesday, November 30, 2011 3:45 AM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Unable to fit the model with PROC NLIN
Hi every body,
I tried to fit the model with three observations. I have three different
doses and their PD responses. I used following code:
proc nlin data = it method=marquardt hougaard;
parms E0 =1 ED50 =1 Emax = 1 ; parameters;
model FFEP=E0 + ((Emax * Dose) / (ED50 + Dose));
Der.E0 = 1;
Der.Emax = Dose / ( ED50 + Dose);
Der.ED50 = - (( Emax * Dose)/ ((ED50 + Dose)**2));
output out=pred p=yhat;
run;
Model is not able to fit since I have only three observations. Is there
any way to work on this? Your help is greatly appriciated.
Thanks,
krishna.
|
