mlogit specification problem in RMark

[shelly]

I have a query about the use of mlogit links in multi-state models in RMark.

I am using a 3 state multi-state model to compare disease infection rates at two sites. The states are: A = subadults (all healthy); B= healthy adults; C= diseased adults. State C is an absorbing state in that there is no recovery from disease (so transitions away from C are not allowed – fixed to zero). I want to compare infection rates (A->C and B->C) between the two sites (an unmanaged population with natural disease progression and a managed site at which a “test-and-cull” operation has been implemented).

My concern relates to how RMark implements the mlogit links that constrain all the transitions out of a particular state to =1. I want the mlogit links to be site dependent (so all transitions out of state A at site1 sum to 1 and all transitions out state A at site2 also sum to 1, etc). I’m not sure if this is how RMark implements mlogit links. When I run a model (with a >groups=(“site”) specified in the >process.data call) and look at the “links” output from a model (>model$links) each of the mlogit links has a different numeric value.

How can I be sure that the transitions away from any stratum are only being constrained to add to 1 within a site and not over all sites?

[jlaake]

I don't think the mlogits will have all different values as you said but it will appear that way because there will only be two of each. With 3 states, there are only 2 modelled transitions for each state-time pairing. The mlogits are linked by their index. If you were to group by site and you found there were more than 2 mlogis with the same index value then that would be a problem but I don't think you'll find that is the case. If it was incorrectly summing over all sites you would have k*2 mlogit links with the same value where k is number of sites. If it is summing within site as it should be then there will only be 2 similar indices for each mlogit.

You may also want to look at TransitionMatrix function which computes all real transition values including the one computed by subtraction. Let me know if you still have further concerns.

--jeff