Analyzing binary traits using live-recapture/dead-recovery

[BMcLoone]

I have live-recapture/dead-recovery data from a longitudinal study of a cohort of around 5,000 humans who were measured over the course of 60 years. The ages at the initial exam vary from late-20's to early 60's.

One of the traits that was measured in this study is blood pressure. An individual can receive a mark of "G" if his/her blood pressure is normal; a mark of "P" if his/her blood pressure is prehypertensive; a mark of "E" if his/her blood pressure is hypertensive; and a mark of "T" if his/her blood pressure is "stage two" hypertensive. I have given each individual a mark of "A" for all of the encounters before his/her first exam. "A" does not signify a measurement per se, but simply registers an individual in time as being alive.

e.g.:

/*XXXXXX*/ 0000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0G0G0P0P0G0P0P0P0G0G0G100000000000000000000000000000000 1 0;
/*XXXXXX*/ 0000000000000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0G0G0G0G0G0G0G0G0P0P000G0P0G000P0G0P0G0P0G00001000000 1 0;
/*XXXXXX*/ 000000000000000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0G0P0G0P0G0P0P0P0P000P0G0P0P0P0P0E0P0P0P0E0P0010000 1 0;
/*XXXXXX*/ 0000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0P0G0P0P0G0G0G0P0P0P0P0P0P0P0P0G0G0G0P0P0G0G00100000000 0 1;
/*XXXXXX*/ 000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0P0P0P0P0P0P0E0E0E0E0P0T0000000P00001000000000000000000 0 1;
/*XXXXXX*/ 000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0E0000000000000000000000000000000000000000000000000000000 0 1;
/*XXXXXX*/ 000000000000000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0G0G0G0P0G0P0P0P0T0G0T0P0P0P0E0E0P0E0P000P0G00001000000 0 1;
/*XXXXXX*/ 0000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0E0T0T0T0T0T0T0T0T0T000E0T0E0E001000000000000000000000000 0 1;

...and so on...

When I run this analysis with the five strata (A,G,P,E,T), it converges. However, when I run an analysis with just three strata (A,G, and P, where E and T are lumped with P), it does not.

e.g. (input file where E and T are lumped with P)

/*XXXXXX*/ 0000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0G0G0P0P0G0P0P0P0G0G0G100000000000000000000000000000000 1 0;
/*XXXXXX*/ 0000000000000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0G0G0G0G0G0G0G0G0P0P000G0P0G000P0G0P0G0P0G00001000000 1 0;
/*XXXXXX*/ 000000000000000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0G0P0G0P0G0P0P0P0P000P0G0P0P0P0P0P0P0P0P0P0P0010000 1 0;
/*XXXXXX*/ 0000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0P0G0P0P0G0G0G0P0P0P0P0P0P0P0P0G0G0G0P0P0G0G00100000000 0 1;
/*XXXXXX*/ 000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0P0P0P0P0P0P0P0P0P0P0P0P0000000P00001000000000000000000 0 1;
/*XXXXXX*/ 000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0000000000000000000000000000000000000000000000000000000 0 1;
/*XXXXXX*/ 000000000000000000000000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0G0G0G0P0G0P0P0P0P0G0P0P0P0P0P0P0P0P0P000P0G00001000000 0 1;
/*XXXXXX*/ 0000000000A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0A0P0P0P0P0P0P0P0P0P0P000P0P0P0P001000000000000000000000000 0 1;


I have tried to run analyses of other traits in which there are only three strata (A, plus two others), and those do not converge either. I believe the problem lies in having a binary representation of some trait, plus the stratum "A". I have tried a variety of models with this binary configuration, and none has worked.

Has anyone else had trouble running live-recapture/dead-recovery models in which the traits are binary?

Thank you in advance for any help!



Brian