st: xtmixed Hessian error

[Sarah Vinette <sarah.a.vinette@gmail.com>]

Hi,

I have a question regarding an error I get when I try to run a mixed
model linear regression. My data has a fair amount of variation so I
felt it was best to use a model that allowed both the slopes and
intercepts to be different for each participant. For my study, I have
a few subjects where Oxygen levels are measured repeatedly over time.
I am hoping to examine the slopes of these groups as well as compare
the slopes of two different groups. I am using the following commands:

xtmixed Oxy Time Timeb Groupb, || ID: Time, covariance(unstructured) reml

xtmixed Oxy Time, || ID: Time, covariance(unstructured) reml


With some of my groups - but not others, I get the following error:

"Hessian is not negative semidefinite conformability error”


This generally goes away if I use ml instead of reml or if I take out
the covariance part of the command or if I only allow the intercepts
to vary by participant (and not the slopes).


I do not have a ton of experience with mixed models.  Would someone be
able to tell me what this error means and why it might be occurring?
I am planning to proceed using ml instead to prevent this error
(although I had been using reml as, from what I understand, it is
unbiased with smaller sample sizes and some of my groups have few
participants), is there a better way to deal with this?


Thank you very much in advance for your advice and help!


Sarah

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/