Hello,
I'm working with a dynamic system that I've started to
analyse using msm(I've emailed to chris, orignator of
the program, separately, but maybe he's on holiday).
The data is obtained from a large cohort of students
and consists of a model of learning states that
students pass through over a period of one school
year. As analyzed with the msm program, the data shows
'no lack of fit' with some covariates that have been
included.
Using the msm program gives nice Q and P matrices
for the system. Furthermore, the p matrix can be
adjusted from t=0 to practically any time range.
However, I'm especially interested in the question of
the significance of the covariates in the
system:(Pardon me if this question is obvious from the
literature; I haven't seen it discussed as yet)can we
determine if a dynamic Markov system is homogenous or
nonhomogenous on the basis of whether covariates
included in the model are significant(as indicated by
msm's p-values)? Is this the only way?
One reason I'd like to go further than msm is taking
me right now is that this kind of data analysis seems
very pertinent in lots of situations, but there's a
certain narrowness in msm(look away from it's
tremendous precision and other good qualities for a
moment): how about situations where people are moving
through states that are not consecutive(1 to 5, as
opposed to only from one state to the next), as msm
seems constrained to limit itself to? I guess I'm
also wondering if anyone
has analyzed such empirical data using the state space
system in MATLAB or another analysis/simulation
program?
thanks,
s
