“autocorrelation” of nominal variable

Question

I have data on patients attendance of a series of sessions of counseling. At each session, a patient could be A, S or D. Each patient has data for 12 sessions, and there are ~100 patients.

I am interested in how the variables are 'autocorrelate'. I could just make 3x3 tables of (e.g.) time 1 vs. time 2, time 1 vs. 3, time 2 vs. 3, etc. but I am curious if there is something more sophisticated.

Googling this revealed some measures for spatial autocorrelation, e.g join-counts, but nothng specifically for temporal autocorrelation. I could also try to adapt the join counts to temporal data, but, again, am curious if someone has done this already.

I have access to R and SAS packages

Answer 1

One other thing to consider, besides markov-chain is sequence analysis / mining. There are several really good packages in R:

TraMineR which has the following very good docs:

• here
• here

There is also that package arulesSequence.