www.phidot.org

Hi all. I am trying to compare tag loss rates (it's slightly similar to Laake et al (2014) but there are enough differences that I would love some help on how best to proceed). Instead of comparing 2 tags, I have 3 tags: ear tag, pit tag 1 and pit tag 2 and no permanent marks. Essentially I want to know if using 2 PIT tags is better than using 1 ear tag for marking species? Additionally, I am comparing tag loss rates among species.

Question 1: I assume I cannot use the Model = “hmmcjs2tl” because I don’t have a permanent mark and I have 3 different tags? Instead, should I use “hmmMSCJS” or something else? If I use the “hmmMSCJS”, I assume I would set the following stratum with the assumption that losing the front PIT tag is no less likely than losing the back PIT tag:

1=3 tags
2=2 pit tags, 0 ear tag
3=1 pit tags, 0 ear tag
4=1 pit tags, 1 ear tag
5=0 pit tags, 1 ear tag

Whereby transitions that are possible include (and then I would fix the others to 0 as we did not re-tag any individual):

1 to 2 or 3 or 4 or 5
2 to 3
4 to 3 or 5
With the transitions between 1 to 2 and 1 to 5 being of most interest?

But I can see how the probability of of losing 1 PIT tag depends on losing the other PIT tag so if I want to be able to model that relationship, I assume I would need to include the following:

1=FPIT, RPIT, ET
2=FPIT, RPIT
3=FPIT
4=RPIT
5=FPIT, ET
6=RPIT, ET
7=ET

Is this how I would go about doing this? Or is there a better way? I’m still most interested in the transition from 1 to 2/3/4 and transition from 1 to 7.
I’m not 100% how best to handle the triple loss individuals as they will be unobservable which is different from dead. I assume triple tag loss is minimal for our study period so is it ok to just assume it happens so rarely it is negligible to the analysis?

Question 2: Just to complicate things a bit further, I also have some missed captures periods for some species. Below is a traditional Mark CJS input dataset to show what I am trying to say:

CH Freq Species
101..11 1 spec1
1111011 1 spec2
100..11 1 spec1
1000010 1 spec2
100..10 1 spec1

So is there a way to incorporate this into the models in marked or do I just set the missed periods to 0 and then fix the probability of capture (p) to 0 for those time periods/species since there is no possibility of capturing that individual then?

Thank you for your help,
Amanda

You are correct that you cannot use 2tl unless you assume that one is permanent. If you go with a multi-state modelling approach I would suggest using RMark and MARK. But I'm contemplating some different approaches where you condition on survival and possibly status of a tag to more simply estimate tag loss to evaluate differences. I'll reply back here later. On vacation xurrently.

Jeff, can't she just use the "mvmscjs" model? The situation is just about the same as the sea lion analysis in the paper (Amanda see https://projecteuclid.org/journals/statistical-science/volume-31/issue-2/Multivariate-State-Hidden-Markov-Models-for-Mark-Recapture-Data/10.1214/15-STS542.full). The only difference is her 3rd state transition is a tag instead of a location. Granted, she doesn't have a permanent mark but she can at least model 3 independent marks. There might be a small amount of information to get the pit tag dependence because the ear tag is independent. i.e., you see the 00 for the pit tags sometimes. But, I'm not sure on that one. A far as the missed sampling occasions, yes, just fix p=0 for those times and capture histories.

Thank you both for your quick replies. Jeff, thank you so much for replying even on vacation. Whenever you are back, if you have any further thoughts, suggestions, ideas, I look forward to hearing about them. I will look over the Sea Lion paper in more detail but it does sound like it might be a good option for this dataset.

Cheers,
Amanda

I guess I'm technically always on vacation since I have been retired for 7 years but I was away and busy with family.

Great idea Devin. I hadn't considered that. Been awhile since we did that paper and it wasn't on the top of my mind. It could possibly simplify the modelling using pit tags as one variable and the ear tag as the second variable. Will have to set p to 0 for 00 pit tag observation when ear tag is 0.

I'm wondering why you are even considering a single ear tag as a possibility because mortality and tag loss would be completely confounded. One option to evaluate consider is an ear tag and single pit tag. They are likely to be independent. The ear tag or torn ear would signal you to look for pit tag exhaustively.

Jeff

Jeff,
Thank you very much for your response and I very much appreciate you continuing to be active while "always on vacation". Sorry for my slow response, it was my turn to be on vacation and I did not have my laptop with me. I don't think I am fully following your response. A torn ear could signal a lost ear tag but their ears also are just sometimes naturally torn. Furthermore, I don't have data on whether an ear was torn or not. We scan every capture for both PIT tags and the ear tag so we always know (with the assumption that the PIT tags do not fail) whether the individual still has their tags or not. I'm not really interested in survival other than the individual has to survive to know if they are still tagged or not. I'm most interested in probability of losing both PIT tags vs a single PIT tag vs a single ear tag and whether that differs among species (for instance is it better to ear tag species A but better to PIT tag species B). Basically to help inform what the best method would be when you can't necessarily always use 3 tags (whether it be cost, animal size restrictions, etc).

Cheers,
Amanda

So what you are trying to decide is between the following:

1) 2 pit tags and 1 ear tag
2) 1 pit tag and 1 ear tag
3) 2 pit tags

What I was saying (maybe not well) was that using just a single ear tag should not be considered because tag loss and mortality are confounded. I was also suggesting that with a lost ear tag that it would flag to look closely for pit tags. I wasn't suggesting that torn ears be used other than that. I believe that 1 pit tag and 1 ear tag will end up being the preferred alternative to 3 tags because it is likely that pit tag and ear tag losses will be independent. What critter are you studying? Do ear tags become unreadable? With sea lions flipper tags became unreadable and losses were dependent. If you'll send me a snippet of your data I'll use it to set up a template for you.

--jeff