Huggins Closed Capture- Removal Model for Pt Ct data

[jlaake]

db-

If you are interested contact me (jeff.laake@noaa.gov) and I can send you some relevant pdfs from the double observer line transect literature that you may find useful if you don't already have them. Both outline the idea I described earlier and the first may be similiar to what you are doing with point data.

regards --jeff

LAAKE, J., M. J. DAWSON and J. HONE. 2008. Visibility bias in aerial survey: mark-recapture, line-transect or both? Wildlife Research 35: 299-309.

BORCHERS, D. L., J. L. LAAKE, C. SOUTHWELL and C. G. M. PAXTON. 2006. Accommodating unmodeled heterogeneity in double-observer distance sampling surveys. Biometrics 62: 372-378.

[marshbirder]

APOLOGIES! I already screwed up my first post and posted this in a Huggins thread in the RMark forum. Please see below.

I'm trying to use Huggins as well, based on Thompson and La Sorte 2008. I have looked through the forum here for advice before posting and have found some previous responses helpful. I understand the points made in this thread regarding removal models for point count data, but that is what is expected of me by my advisor, our project partners, and it seems like many editors these days...

Right now I have one year of data, but I will eventually have three. So, for now, I just want to be able to estimate abundance without looking at changes over time. I see that you are saying here I shouldn't expect to get abundance at each point count station with Huggins, which is fine. My covariates should tease out the differences I'm looking for (which would be based on point count locations). My research is looking at songbird abundance relative to silviculture, with focal species I can use, rather than the entire community so that I can use the more abundant birds for analysis. I have two sites that are already harvested and two that will have one year of pre- and at least two years of post-harvest sampling.

*123 point count stations
*2013 data are two sites harvested and two sites pre-harvest
*3 visits per year
*10-min point counts with five 2-min intervals
*We will use only detections within 50m
*I want to relate abundance to slope, aspect, and basal area gradient (measured in 4 prism plots at each point count station) with more variables to be added later

It was suggested that I run Huggins to get p and c and then use a GLMM to model abundance relative to my linear variables. The MARK portion of this seems much simpler than what I was initially thinking I would have to do, although I'm not sure how to code for more than 3 species (0 1, 1 0, 0 0, ?). Thompson and La Sorte used 6 species in their analyses. Does this seem like a good approach or is there a model I can use that would incorporate my continuous variables?

I have only ~6 weeks of a 1-credit graduate level MARK course under my belt.

Thanks!
Gretchen

[bacollier]

APOLOGIES! I already screwed up my first post and posted this in a Huggins thread in the RMark forum. Please see below.

I'm trying to use Huggins as well, based on Thompson and La Sorte 2008. I have looked through the forum here for advice before posting and have found some previous responses helpful. I understand the points made in this thread regarding removal models for point count data, but that is what is expected of me by my advisor, our project partners, and it seems like many editors these days...

I would bet that what is correct and accurate would probably be acceptable as well

Right now I have one year of data, but I will eventually have three. So, for now, I just want to be able to estimate abundance without looking at changes over time. I see that you are saying here I shouldn't expect to get abundance at each point count station with Huggins, which is fine.

Recently there was a thread on here about estimating point-specific abundance that you were on if I remember right, so you can get abundance at the point level if you wanted. Do you mean 'expect' in that MARK won't output it for you and you have to derive them yourself?

My covariates should tease out the differences I'm looking for (which would be based on point count locations). My research is looking at songbird abundance relative to silviculture, with focal species I can use, rather than the entire community so that I can use the more abundant birds for analysis. I have two sites that are already harvested and two that will have one year of pre- and at least two years of post-harvest sampling.

*123 point count stations
*2013 data are two sites harvested and two sites pre-harvest
*3 visits per year
*10-min point counts with five 2-min intervals
*We will use only detections within 50m
*I want to relate abundance to slope, aspect, and basal area gradient (measured in 4 prism plots at each point count station) with more variables to be added later

It was suggested that I run Huggins to get p and c and then use a GLMM to model abundance relative to my linear variables. The MARK portion of this seems much simpler than what I was initially thinking I would have to do, although I'm not sure how to code for more than 3 species (0 1, 1 0, 0 0, ?). Thompson and La Sorte used 6 species in their analyses. Does this seem like a good approach or is there a model I can use that would incorporate my continuous variables?

There is no limit on the number of groups you can set up in MARK in the DM, there are quite a few examples of various ways to do that that in the help files.

With regards to the rest and your question on the GLMM and "does this seem like a good approach", the short answer is no, what you are describing is doing statistics on statistics and a number of folks on this list would advise you against doing for any number of reasons, yet it seems to happen pretty regularly in the wildlife literature. As for options, you can look at ways to integrate your covariates into your analyses in MARK (which will thus primarily focus on the impact those metrics have on p and c), or you may have to move out of MARK into a hierarchical approach such that you can model the impact of those covariates on N (I think Royle and Dorazio has some closed capture examples in it that you could adjust appropriately to your situation, also might look at Kery and Schaub, which I just started reading but I suspect there are examples there as well).

\bret

[jlaufenb]

As Bret mentioned, there are hierarchical modeling examples similar to your sampling design presented in the Kery and Schaub book (Bayesian Population Analysis Using WinBUGS, 2012). In particular, have a look at chapter 12 and the open-population binomial mixture models. I also agree with Bret's comment on "doing statistics on statistics" which, unfortunately, is still commonly finding its way into being used based on what I call the "legacy effect" of doing things a certain way because it is the way is has always been done. Soap box aside, I encourage you to also check into the R package 'unmarked' which fits hierarchical models to count data within a MLE framework. I'm fairly certain it offers model options that would fit your situation. A final comment is my concern about how many of your sites have zero detections. If you have a large proportion of sites with zero detections and you go the binomial mixture model route, you may need to consider accounting for those zeros by incorporating a zero-inflated component into your models (see Kery and Schaub 12.3.2 for an example). Unfortunately, I do not think 'unmarked' currently offers this model.

Good luck!
Jared

[marshbirder]

Recently there was a thread on here about estimating point-specific abundance that you were on if I remember right, so you can get abundance at the point level if you wanted. Do you mean 'expect' in that MARK won't output it for you and you have to derive them yourself?

I was referring to your response in the other thread. I guess I meant from MARK, but whether I have to derive them myself or get it from MARK, that's fine. I have to do some more hands-on work with MARK to understand what I'm doing

There is no limit on the number of groups you can set up in MARK in the DM, there are quite a few examples of various ways to do that that in the help files.

With regards to the rest and your question on the GLMM and "does this seem like a good approach", the short answer is no, what you are describing is doing statistics on statistics and a number of folks on this list would advise you against doing for any number of reasons, yet it seems to happen pretty regularly in the wildlife literature. As for options, you can look at ways to integrate your covariates into your analyses in MARK (which will thus primarily focus on the impact those metrics have on p and c), or you may have to move out of MARK into a hierarchical approach such that you can model the impact of those covariates on N (I think Royle and Dorazio has some closed capture examples in it that you could adjust appropriately to your situation, also might look at Kery and Schaub, which I just started reading but I suspect there are examples there as well).

\bret

Thanks for that input. I have to submit something from MARK for this class I'm in, but I'm not tied to it for my dissertation analyses. I will look into your suggestions. I'm sure there is a way I can integrate my covariates for now.

And, thank you, Jared.