Loss on capture for multistate models?

[Fishhead]

Hi there. Are losses on capture handled by the multistate models implemented in marked?

I've been working through some of the examples that use the built-in mstrata dataset and modifying the data to add a capture history indicating losses on capture with a negative frequency. As far as I can tell, the accumulate option simply sums over common capture history frequencies and reduces freq accordingly due to the negative value. This seems to simply remove the individuals entirely from the data set. Setting accumulate = F returns an error for an invalid "times" argument.

I tried to dig deeper in the ddl, tracking through the call of create.dmdf, where losses on capture are explicitly mentioned in this function's help. But I still couldn't confirm if and how losses on capture are handled by multistate models.

Any insights are much appreciated!

[jlaake]

This is something I should have documented. Sorry. None of the models handle loss on capture by default. Do not use negative frequency. Because marked creates a parameter for each occasion for each animal all you have to do is set fix variable for p to be 0 for each occasion after the loss occasion. For a multistate model it will have to be assigned to 0 for each state. For any p that should be estimated fix should be NA.

Let me know if you have any problems doing this.

Jeff

[Fishhead]

Thanks for the quick response. Your suggestion makes sense. I assume I'll have to use individual capture histories and set accumulate = F. One potential issue is that I have >300 k individuals (these are PIT tags). So I may encounter run time limitations at the individual level. Collapsing the individual capture histories into frequencies would likely reduce run time considerably but this doesn't appear to be possible because p needs to be fixed for specific individuals. I haven't run this yet, so don't know for sure if 300k individuals will be an issue. Any thoughts or suggestions?

[jlaake]

You might want to switch to RMark/MARK. But if you want to stay with marked I would accumulate the frequencies except those that are LOC. Make that your input data and set accumulate to FALSE on the run.