Date:         Mon, 19 Jul 2004 07:29:36 -0700
Reply-To:     jeremie guedj <jeremie.guedj@ISPED.U-BORDEAUX2.FR>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         jeremie guedj <jeremie.guedj@ISPED.U-BORDEAUX2.FR>
Organization: http://groups.google.com
Subject:      fit differential equation proc model
Content-Type: text/plain; charset=ISO-8859-1

Hi,

I have the following problem with the "proc model"

I try to estimate parameters of a system of ODE and I have 2 problems
:

- In a first time, I want to simulate my data with known parameters,
add a noise, take a sample of these data and see if I can recover the
initial parameters.
Moreover, the variance of the noise is supposed to be known.
How can I specifiate that this variance is known ?
I tried the option "h.parameters=sigma" but the results are not well.

- I would like to use the maximum likelihood.
There is the option "fiml".
But the data that I have to fit are not directly related to the system
of ODE but are a combination of them. For example, if my system is :

df=h1(t,f,g)*dt
dg=h2(t,f,g)*dt

data to fit are ln(f) and f+g.
So I have an error message ; it seams that you can use fiml only if
data to fit are directly related to the system of ODE.

Is it right ?

Thank you in advance