how to model the application of new marks

[SophieSmout]

Has anyone come across this problem?

The animals in my data set sometimes lose their marks (tags fall off).

New marks are also sometimes applied. Can I model this in an e-surge framework? One way I can think of is to have a covariate which is an indicator variable (it varies by year and by individual), which indicates whether or not a new mark has been applied.

However what I read in the manual suggests that covariates can be either time-dependent, or can vary by individual, but not both. Is that correct, and does anyone have another way to work around this issue? I'd be very grateful for any suggestions/ examples.

Thanks, Sophie

[sbonner]

Hi Sophie,

I think you are correct that E-surge can only handle covariates that are either constants for each individual or time-dependent but the same for all populations -- essentially all values of the covariate (for each individual at each time point after its first capture) have to be known values.

Because the covariate is a discrete variable you might be able to use a multi-state model. The states could be whether or not the individual has been re-marked or you could define states as marked one, marked twice,...

The difficulty that I forsee is in linking the different data from one individual that were collected under different marks. E-surge expects you to enter one capture history for each individual observed. If an individual loses its tag and is tagged again can you link back to its previous history, or does this get entered as an entirely new capture history.

Some work has been done on the issue of tag loss. My understanding is that you generally need some independent measure of the frequency with which tags are lost in order to estimate the proportion of extra capture histories in the data. This can be done by double tagging some individuals and recording how many come back with only one tag, or tagging some individuals in a closed environment and observing how many lose their tags. You might want to read through:

Cowen, L. and C. J. Schwarz. 2006. The Jolly-Seber model with tag loss. Biometric. 62, 699-705.

Hope this helps!

Cheers...