 |
 |
Dear BUGS users, I am fairly new to BUGS, but excited about its capabilities. I am working on complex problem that I hope BUGS can offer some improvement. If you are interested in sharing your insights on how this problem should be approached and offer some reference, I would greatly appreciate it. I am running a semi-Markov process (SMP) model on health events observed over a 4-year period with annual follow-up interviews. The survey has a rotating panel design so that each person contributes up to 4 observations and then rotated out. The goal of the study is to apply the SMP model on different types of events and derive the duration-dependent age-specific transition probabilities so that I can map out the health trajectories (by using a life-table frame work), from age 65, for example, to death. By using a Monte Carlo simulation, I can get a large collection of these independent health histories for further summary analysis. The problem arises when I try to measure the duration, or elapsed time, for each individual event. Everyone at the beginning of the survey (baseline) occupies a certain state. (Here I am only using 5 finite health states and death.) He may or may not experience an event during the 4-year interview. Either way I do not observe the entry into these baseline states which is prior to survey beginning. Obviously to measure the true duration effect I need to impute these unobserved portions R of the total durations. Everyone's data needs imputation. Starting at baseline one either has an event later during the survey so that the event is left-censored, or has no event till the end of survey which makes the 4-year observation interval-censored. Subsequent events don't need imputation because their beginnings are observed. What I am doing right now is a) apply a SMP model (based on accelerate failure time regressions in current study) to all events/spells, get transition parameters from estimated coefficients, run a Monte Carlo simulation and get an estimate of the duration distributions for each origin-destination-specific events; b) estimate R for each origin-destination pair of left-censored (i.e., without observed origin) and interval-censored by randomly draw a single r* from the estimated matching duration distributions; form a complete data set from r* and original durations for subsequent events; c) repeat step a) & b) until estimates of some summary statistics like life expectancy at age 65 stabilizes. My question is, would Gibbs sampling be able to improve the above models by offering better control for convergence and provide standard error estimates? Has anyone used BUGS for a similar problem and can provide some reference to help me get started? I appreciate your help. Liming ------------------------------------------------------------------- This list is for discussion of modelling issues and the BUGS software. For help with crashes and error messages, first mail [log in to unmask] To mail the BUGS list, mail to [log in to unmask] Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html Please do not mail attachments to the list. To leave the BUGS list, send LEAVE BUGS to [log in to unmask] If this fails, mail [log in to unmask], NOT the whole list