nlmefit compound symmetry covariance pattern?
    조회 수: 7 (최근 30일)
  
       이전 댓글 표시
    
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.
댓글 수: 0
답변 (1개)
  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.
댓글 수: 0
참고 항목
카테고리
				Help Center 및 File Exchange에서 BPSK에 대해 자세히 알아보기
			
	Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!

