www.phidot.org

How can I use Program MARK or Rmark to pool mark-recapture data from three isolated sites with different and unequal survey intervals?

We used mark-recapture to survey for active infections of a disease in small mammals at three sites over eight years. We conducted 3-6 trapping sessions per year per site weather permitting, and it was impossible for our small crew to survey multiple sites in a day, so the trap dates are different for all three sites. There is no movement between sites, but site size and catchability are constant.

For the species that we have enough data for, I want to compare survival (phi) between animals that were infected at capture at least once and animals that were never infected at capture to test whether or not the disease reduces survival. We are not interested in differences between sites and we want to pool the data for a more robust estimate.

Is there a way I can enter three different sets of encounter histories into the same analysis?

Thank you in advance for your help!

I am not sure it would be easy to do that, according to what you have explained. It depends, among other things, on how much the dates of the sessions differ in the three sites. Anyway, I'm not sure you would have much to gain from doing this. By doing that you would be the possibility of testing, within the analyses, if there are differences between the three sites but you are not interested in that.

However, something that seems more important to me is what you have mentioned about how you want to test the effect of the disease on the survival of individuals. Being infected, or not, is a state and therefore, I do not suggest that you treat it as a group. Infection is usually a dynamic trait. It is something that changes throughout an individual's life. That is what in capture-recapture is called a state. Sex, or belonging to a population (closed to exchanges with others) is usually a fixed feature of the individual, something that in capture-recapture is usually modeled as a group. Making two groups, infected at least once vs. never infected, it doesn't seem like the right way to answer your question. First, because you are assuming that an individual never detected as infected has, in effect, never been infected. This is contrary to the logic of capture-recapture modeling where detectability is something you want to account for. Second, in the "never infected" group there will be individuals who have been infected but at a time when you did not capture them (by the way: did you find out the infected / uninfected status of all the individuals each time you captured them?). Third, it may be the case that the probability of capture varies according to the infectious state. Fourth, and perhaps most importantly, that classification criterion runs the risk of biasing your results. On average, the more times an individual is captured, the higher their survival will be, so that those you classify as "infected at least once" will have their survival inflated by the classification criteria itself.

Solution? Try to use a multi-state approach.

Thank you simone77 for your thoughtful response!

You brought up some excellent points that I somehow failed to consider. I will play with the multi-state models a bit, but ultimately I now think my approach is pretty flawed here because of the issues you mentioned.

And just to answer your question: yes, we were able to obtain qPCR results for all 2300 of our captures.

Looks like it’s back to the drawing board for me.

Cheers!