Re: optimal design for nonlinear model in R
- From: "hyena" <as@xxxxxx>
- Date: Mon, 29 May 2006 10:34:21 +0200
Thanks for all the answers!
"Old Mac User" <chendrixstats@xxxxxxxx> wrote in message
news:1148406442.709934.237590@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
The optimal placement of points (the "design") for a model which is
nonlinear in the parameters of the model will depend upon the values of
those parameters... alpha and beta. I know that sounds "strange"
because you don't know the values of those parameters. So the trick is
to get preliminary estimates of them and move forward from there.
Once we have early estimates of the parameters we can calculate the
optimal
(or near-optimal) placement of future experimental points (the
"design".)
In general, the optimal design will not be a factorial or fractional
factorial. For instance, if you have only two parameters to be
estimated (alpha and beta) then the optimal design will be two
experimental combinations. That begs the matter of having some
leftover degrees of freedom for "lack of fit" and/or to test the
validity of the model.
Designing experimental to estimate parameters in models which are
nonlinear in the parameters is a specialized subject... not easily
"canned" into software since it requires several operations, some of
which require judgement.
.
- References:
- optimal design for nonlinear model in R
- From: hyena
- Re: optimal design for nonlinear model in R
- From: Old Mac User
- optimal design for nonlinear model in R
- Prev by Date: Re: Minimum number of cases for a correlation?
- Next by Date: Comparing two variation coefficients
- Previous by thread: Re: optimal design for nonlinear model in R
- Next by thread: Tests for Seasonality
- Index(es):
Relevant Pages
|