Greetings! I am having some difficulty modelling a longitudinal data set and
was wondering if anyone had some useful tips.
I am trying to model a continuous outcome (transformed to normality) as a
function of two other variables. My dataset is longitudinal with a number of
individuals and I thought that the most appropriate model would be a GEE
(gaussian with identity link). My only problem has been specifiying the
correlation structure in STATA. Observing the data has shown that
individuals have very different autocorrelation patterns and since there was
no reason why (physiologically) the data would possess a particular
structure, I have opted for the "unstructured" correlation but my model does
not converge. In fact, the only models that convereged were "independent"
and "ar 1".. I was wondering if anyone would have any ideas about other
structures or modelling approaches that would be appropriate in this
instance.
Thank you very much for your help.
Kind regards.
Stephanie MacNeill
|