fixing transitions and recapture rates in MS models

[shelly]

Hi

I am working with a cmr dataset from a population of diseased tasmanian devils and want to examine the impact of disease on survival and transition (infection) rates.

I have set up a 3 strata MS model as such:
S1= Healthy subadults (devils contract the disease as adults) = must transition to either Healthy adults or Diseased adults (yearly survival estimates = must age to adults).
S2= Healthy adults = transition to D adults or stay as H adults.
S3= D adults = can only stay D adults (there is no recovery - yet!).
The following transitions are fixed to 0
S1 -> S1 = O; S2-> S1 = 0; S3->S1 = 0; S3 -> S2 = 0

For this to work I have also fixed recapture rate for S1 (subadults) to 0 (because they must move) and recapture rate for S3 (D adults) to 1 (because they can't move). Doing this improves the model support significantly but I'm not sure if this is valid since diseased adults are only ever captured once (survival of S3 strata is virtually zero).

Any advice is most welcome

Shelly

[ganghis]

Hi Shelley,

Fixing transition and other parameters is a completely valid thing to do given your problem. However, make sure to set S3 -> S3 = 1 or you may have parameter ID issues, and make sure to change PIM definitions so that you get the right transition parameter by subtraction so that transitions will add to one (e.g. from transitions away from strata S1, change definitions to get either S1 _> S2 or S1 -> S3 by subtraction).

I'm not sure what to say about your last issue with capture probability for S3 = 1. If you don't have any survivors for the S3 class my temptation would be to set survival for S3=0 and to keep capture probability for S3 =1. If you do have some survivors, your current solution should work to keep parameter ID issues at bay.

Finally, keep an eye out for estimated standard errors of 0 or really high values, together with results from the MARK's singular value decomposition to try to diagnose situations where you are trying to estimate too many parameters.

Cheers, Paul Conn

[shelly]

Thanks Paul,

Its good to know I'm on the right track and fixing S3 -> S3=1 seemed to work better (smaller confidence intervals on my parameters).
cheers
Shelly