st: appropriate analysis of tricky panel data

["Lachenbruch, Peter" <Peter.Lachenbruch@oregonstate.edu>]

A grad student came by with data on reimbursement that is collected at 6 month intervals with panels beginning every 18 months.  At about the midpoint of the overall interval, there is a change in the reimbursement scheme.  Thus, data may look like
Year 1	x 	x+6	x+12	x+18 	x+24	x+30
Year 2	y	y+6	y+12	y+18	y+24	y+30
Year 3	z	z+6	z+12	z+18	z+24	z+30
Year 4	w	w+6	w+12	w+18	w+24	w+30
Etc.
The value of x may be 0, y may be 18, z=36, w=54.  Thus the panels overlap only a bit.
The time of the change is about 48 months after the beginning of the study, so the year 1 and year 2 cohorts are all completed before the change occurs.  The third cohort will overlap the change it runs from 36 to 66, so part will be before the switch and part after.  The fourth and fifth cohort will be entirely after the switch.
It is believed that the switch will affect reimbursement substantially.  
I can analyze the data using the cohorts for years 1 and 2 and years 4 and 5 and ignore year 3 because it is partly before and partly after the switch.  This doesn't feel right.  One thing I could do is encode the switch variable as 0 0 0.5 1 1 and use it as a covariate or I could use the responses as occurring when the switch was in place or not.  It's unclear to me if the switch will affect the response immediately or not (probably not since this is related to behavior).
Any ideas?

Tony

Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001



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