repeated measures + SAS + covariance structure

#1
Hello,

My first time posting here, and really hoping that someone can help me out!
I have data from a repeated measures experiment in which the time points were not evenly spaced + some missing data from individual animals, so I am using proc mixed and attempting to find the appropriate covariance structure. I am doing this in two different models - one that includes 3 different time points (days 4, 6, and 7) and one that includes 5 different time points (days 8, 9, 10, 11 and 13).
I know that the unequal spacing in time isn't hugely "unequal" but in any case it isn't completely consistent, either...
Initially I tried sp(pow) and it gives me the message that the model converges but is not Hessian positive definite. Then I tried ante(1) and got the same message. My statistics consultant on campus said that the problem may be that I have relatively few time points and it is trying to estimate the appropriate parameters based on just a few time points and that I shouldn't necessarily be concerned with it not being Hessian positive definite.
I wonder if anyone can help me determine whether there are other options or whether I should just "turn a blind eye" to this problem and continue along in my analysis as if I wasn't getting this message.
I'm new to all of this so I have no idea what would definitively tell me if it's "ok" or not, and I am anxious to proceed with my analysis.

Thank you,

Rachel
 

jrai

New Member
#2
The hessian being singular suggests that the model has been specified wrong. For example where there is perfect colinearity... Check the data and model.

Hessian is a second order matrix & the warning is saying that the matrix isn't invertible & that is mostly when it is not full rank matrix.