nlmefit compound symmetry covariance pattern?

조회 수: 1 (최근 30일)
Sanjay Manohar
Sanjay Manohar 2016년 2월 15일
답변: the cyclist 2016년 2월 15일
Hi, I am using nlmefit to fit a model of a "repeated measures" kind. The function provides very powerful control over the covariance matrix, but I could not find how to set compound symmetry. I have four measures, indicated in the design matrix as separate columns:
X = [ measure==1 measure==2 measure==3 measure==4 ]
I would like the form
[ x y y y
y x y y
y y x y
y y y x ]
Such that there is a fixed correlation between the four measures. But if I use
'covpattern', eye(4)+1
It doesn't constrain the `y` variances to be constant. When I examine `psi`, they are all free/estimated.

답변 (1개)

the cyclist
the cyclist 2016년 2월 15일
My understanding of the "CovPattern" input is that zero/nonzero entries merely indicate which terms can have covariance with each. For example, one of the documentation examples is
P =
1 1 0
1 1 0
0 0 1
nlmefit(...,'CovPattern',P)
and the output covariance is
PSI =
0.5180 0.1069 0
0.1069 0.0221 0
0 0 0.0454
In your case, you put all nonzero entries -- the magnitudes are irrelevant, I think -- and you got covariance entries for all terms.
I don't know of a way to specify the covariance structure you want.

카테고리

Help CenterFile Exchange에서 Design of Experiments (DOE)에 대해 자세히 알아보기

태그

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by