Combining MS trapping and telemetry data

[harding]

I posted this in the RMark forum but was encouraged to repost here as a better place for the question.

When I was first learning MARK, an experienced user told me I could use a dot notation (“.”) to merge radio telemetry and trapping data for multistate models. Specifically, there were times when a collared animal was not trapped during a trapping session, but was known to be alive because it was tracked by telemetry. However, because the animal was not captured, its state could not be determined. I used dots in the capture histories to denote this sort of occasion where an animal was known to be alive, but the state was unknown.(Telemetry was primarily used to evaluate space use and was only incorporated into the capture histories because it seemed like a waste to not use all the information available).

Based on a recent posting on this forum, I am now uncertain if this usage was correct. As I understand it now, what I really did was to fix the capture probability to 0 for those times that I used the dot notation. However, I only used the dot notation for collared animals, and I had also set capture probability to 1 for all collared animals. Does anyone know which of these wins out as for how capture probabillity for those animals was actually handled? When running models using these histories, I did not get any error messages and the number of histories that MARK read was the same as the number in my input file.

If my usage was incorrect, can anyone suggest:
1) a better means of handling this sort of situation
2) how/if this coding might have influenced my results?

Fewer than 2% of the captures had a dot notation in their histories.


Thanks in advance for any suggestions or wisdom anyone can offer!

[claudiapenaloza]

To my understanding, when you use a "dot" notation in a capture history you are telling MARK you didn't "look" for that animal on that occassion... that would be consistent with setting p=0.
Whether or not your fixing p=1 wins over the "dot" setting it =0, I don't know...
but...
Have you thought of using a misclassification model instead?
You would have how ever many states you are already using, plus a unknown state (u) for those times when you would have put a "dot" in the encounter histories, because you don't know in which state your telemetered animals are (right?).

Also... I have a feeling you can make better use of your joint telemetry/mark-recapture data... wish I knew how.

Hope that helps,
cp

[Bill Kendall]

I agree with Claudia. I believe using a model with state uncertainty is best. It sounds like you have included two groups in MARK (telemetered and conventionally tagged). Your p is 1 for telemetered and <1 for conventionally tagged. However, you know state for each of the captured individuals, but none of the telemetry detections? MS models with state uncertainty involve a probability of identifying the state of the detected individual (delta). In your case you would set delta to 1.0 for conventionally marked and 0 for telemetered (or estimate it if you know state sometimes). If you know program E-SURGE and you don't have robust design data, then that is your best bet. You can also do it in MARK, but for now you'll have to email me directly in order to explain it. First, there is not yet documentation for this feature. Second, you'll need to trick a robust design module into doing a non-robust design analysis (easy to do).

Having said all that, it's not clear how your telemetry data is contributing to survival (its primary purpose in this analysis although you said its primary purpose in the study is information on space use). If you have mortality information then you will need to included a dead state in the model.

Bill Kendall