Multi-State Closed Robust Design - Assumption Violations

[oconnell]

I am working with a 6-year data set on a black bear population in Louisiana. We are using hair snaring to perform mark recapture estimation, and sample hair during each of the 8 weeks (secondary sampling) in summer and did so for all 6 years (primary sample) of the study and, thus are using Robust Design. In year 5 of this study, part of the study area was flooded (i.e. sites under 20 feet of water) by the ACOE to prevent flooding in New Orleans. Therefore, we could only sample outside of the flooded area that year. In year 6, we sampled the entire study area as before, thus providing us with before, during, and after data to analyze the flooding effects. We are interested in knowing if the bears died, if they remained and survived, if they left for the unflooded portions of the study area and stayed, or left for the unflooded portions and then returned after the flooding. We hope to address these questions using a RD multi-state model with capture in the flooded or non-flooded areas as the states. We would then estimate the transition probabilities to test the above hypotheses.

After looking at my raw data, I noticed that some bears (9 out of 109) moved from a flooded state to a non-flooded state during some of the secondary sessions, violating the assumption that an individual state remains constant within secondary sessions. I am considering using the RD multi-state model with state uncertainty to address this issue but it can be argued that, in our case, there is really no uncertainty because we know certain individuals are in 2 states.

Is this an appropriate use of this model? Does anyone have any other suggestions? Thanks in advance!

Kaitlin

[claudiapenaloza]

If we look at your data with a Multistate Robust Design (MSRD) with state uncertainty in mind… your certain vs. uncertain states would be a little different than you hinted at. Your only “certain” state would be a flooded/un-flooded or “migrant” bear, whereas for any bear found in either the flooded or un-flooded areas, there is uncertainty as to whether it was local to either of these areas or just a migrant that you haven’t detected in the other area yet.

So you could go down the state uncertainty path if you wish… but it seems you’d have fewer individuals in known states than you initially expected. I think it would work but I’m only mildly familiar with the state uncertainty models… so others may have different advice.

What I see as problematic is that you’ve defined the states “after the fact” and as a consequence of a human induced change, which may have never happened before, and may have no biological relevance to your study species, i.e., the “state” delineation may be artificial. Are you sure the flooded and un-flooded areas where those few bears were found within a primary period are not part of the same “state”? Do you have biological reasons to split up your sampling area into the “states” you’ve defined?

Have you thought about using a spatially explicit model instead of a MS model? (I’m sorry I can’t help you there, I’ve never used spatially explicit models… it just seems your data may have more connectivity between the different “states” than a MS model can handle).

There is a different way of resolving this issue, and I may risk public humiliation by saying it… but you could censor out those 9 bears.

[oconnell]

Thanks for the input!!

"If we look at your data with a Multistate Robust Design (MSRD) with state uncertainty in mind… your certain vs. uncertain states would be a little different than you hinted at. Your only “certain” state would be a flooded/un-flooded or “migrant” bear, whereas for any bear found in either the flooded or un-flooded areas, there is uncertainty as to whether it was local to either of these areas or just a migrant that you haven’t detected in the other area yet.

Yes. That is one of many issues I have with choosing state uncertainty for this data set.

"What I see as problematic is that you’ve defined the states “after the fact” and as a consequence of a human induced change, which may have never happened before, and may have no biological relevance to your study species, i.e., the “state” delineation may be artificial. Are you sure the flooded and un-flooded areas where those few bears were found within a primary period are not part of the same “state”? Do you have biological reasons to split up your sampling area into the “states” you’ve defined?"

Yes, the three states are separated by levees and highways, not impossible to travel from one state to another but not something a bear wants to do everyday either.

Have you thought about using a spatially explicit model instead of a MS model? (I’m sorry I can’t help you there, I’ve never used spatially explicit models… it just seems your data may have more connectivity between the different “states” than a MS model can handle).

Yes. I am in the beginning stages of getting density estimates in secr and am going to try and run models to get before, during, after flood density estimates.

There is a different way of resolving this issue, and I may risk public humiliation by saying it… but you could censor out those 9 bears.

This would be the simplest solution (and possibly the only). I would really like to hear others opinions on excluding those 9 bears ...

[claudiapenaloza]

Great! Its good to hear you been thinking about and dealing with these issues. I hope someone else pipes in about this.

Yes, the three states are separated by levees and highways, not impossible to travel from one state to another but not something a bear wants to do everyday either.

How many places are you sampling at? I had initially understood there were only two states, "flooded" and "unflooded". Can you change the spatial arrangement such that it makes sense with respect to the geographic barriers AND eliminates the within primary period state transitions (by having fewer states maybe/combining two)?

[oconnell]

Unfortunately no. My study are is a large block of land, the spillway/flooded state, with two smaller blocks on either side, Batchelor and Fordoche/non-flooded. All of the violations occur between the spillway and Fordoche or the spillway and Batchelor. For future management during a flood, we would like to know which non-flooded area the bears move to at a higher rate that would allow us to temporarily reduce the speed limit through those highways where bears might be moving, hence the 3 states.