In article <3184062B.3FC0@vtt.fi> timo alanko <timo.alanko@vtt.fi> writes: >From: timo alanko <timo.alanko@vtt.fi> >Subject: Re: Maximum Likelihood Estimation >Date: Sun, 28 Apr 1996 16:58:35 -0700

>jargo@utdallas.edu wrote: >> >> Is there a PROC that allows a user to specify a likelihood function and >> gives the MLE parameter estimates? I thought maybe I could do this with >> PROC NLIN, but after reading the manual, I got the impression that it is not >> really set up to do this, although perhaps there is a workaround. >> >> -- >> -- Paul A. Jargowsky (jargo @ utdallas.edu)

>PROC NLIN can be used for this exactly the way described in BMDP manuals (I don't >have recent ones but in the 1983 manual section 14.3 after BMDP3R and BMDPAR describes >this in detail). In fact, PROC NLIN is much more flexible than the old BMDP because NLIN >allows the full power of the SAS language for model specification. The technical reference is:

>If you can't get the manuals I will be glad to provide technical details. >I have used PROC NLIN for fairly complicated ML-estimation tasks successfully. >If you cannot use partial derivatives, the derivative-free DUD algorithm is very >fast and produces, again in my experience, 'accurate' numeric standard errors.

>Cheers,

>Timo Alanko

>(Timo.Alanko@VTT.FI)

If it is not too much trouble, there is at least one other person out here who wouldn't mind seeing the technical details posted.

I've always found the manuals (at least the recent ones) to be confusing wrt NLIN. I know that it can do this based on the sample code but not so much from the manual. Does NLIN treat only one equation or can it handle systems?

Ken Schaudt