Occupancy Estimation for 8 Individual Sites

[bluegrassjunkie]

Hi there.

I am studying raptor visitation to fields with or without artificial perches. There are 4 control and 4 treatment sites; the experimental units are the 8 fields. I'd like to analyze each field independently using occupancy estimation in MARK, then compare the psi and p estimates between the control and treatment fields. This is a bit different than I've seen in the literature, so I'm not sure how to format the data to create each site's input file.

There were 21 surveys in each field (except field NSU2 which was surveyed 23 times) over 4 months. At this point, I do not plan to include covariates and am including all raptor species detected in one encounter history for each field. Here's what my 8 input files currently look like:

Control:

/*NSU2*/ 11111111111111111111111 1;

/*NSU3*/ 101111111001101111111 1;

/*NSU4*/ 100110111101110101111 1;

/*KEES*/ 111111101001101111111 1;

Artificial Perches:

/*NSU1*/ 111111111111111111111 1;

/*NSU5*/ 111111111011011011111 1;

/*MART*/ 011111111111111110111 1;

/*DANO*/ 101010010110011011011 1;

In the "Enter Specifications for MARK Analysis" box I enter 21 or 23 encounter occasions, 1 group, and 0 covariates, and select Occupancy Estimation for the data type. When I hit "OK" in that box, I get a box saying "The encounter histories file ... only contains 1 semicolon. Is this file a proper MARK encounter histories file?" I click "OK" and a results database is created alright.

Please let me know if I am doing this correctly, or if there is any more information I can provide.

Thanks!

[darryl]

A few things.
1. you want to have the same number of characters in each encounter history. You need to include missing values '.', so there's 23 in each
2. Do you have the labels "Control:' and 'Artificial perches:' in the input file? If so you need to comment them out.
3. what's you're actually research question and why are you using occupancy models? You already have 100% observed occupancy assuming the whoel 4-month period is a single 'season'.
Cheers
Darryl

[bluegrassjunkie]

Hi Darryl, thanks for your response.

1. Do I need 23 occasions for each history, even though each history is its own input file?

2. No, these labels are not included in the input file- that was just to clarify my data for my post (perhaps unnecessarily).

3. My objectives are to determine if artificial perches affect raptor visitation and/or affect small mammal populations (a whole other MARK analysis, live recaptures). Yes, I am considering the 4-mo period a single season. I'm using occupancy modeling to estimate site 'use' by raptors in each field. I hope to look at species-specific differences later on. Again, there were 4 treatment (with artificial perches) and 4 control (no artificial perches) fields. I'd like to create models for each field, then compare the estimates for the control fields to the estimates for the treatment fields. Is a 1-line input file acceptable for occupancy estimation in MARK?

By the way and FYI, this is my master's research.

Thanks,
Jordan

[darryl]

1. own input file? You want to have all your data in a single file as you need to have data from more than 1 'site' in order for any of this to work.

3. Occupancy modelling is based on presence/absence data so you've got to have a sampling unit defined for which you're trying to establish either species presence or absence. In your case are you interested in presence/absence of raptors at the scale of the field, or is presence/absence meaningful at some smaller scale? Your comment about visitation rates makes me think you're more interested in how often the raptors are in a field, which in terms of the single season occupancy model parameters means perhaps your inference is more directed at p (detection) rather than psi (occupancy)? Is that what you were thinking? If so an issue is that I imagine you could have false zeros in your visitation data in that a raptor visited the field, but you didn't see it? Hence 'p' would be a combination of the true visitation rate * probability you see the raptors (which presumably is <1), ie p is going to underestimate the true visitation rate.

Actually, based on what you've said you want to achieve, I'm not sure occupancy models are the best approach to take, at least not given the information you've supplied here. I suggest you go back and think things through a bit more and perhaps do a bit more reading to figure out how occupancy modelling is going to provide you with the answers you're looking for, given the data that you have collected. Prior to collecting the data, how did you plan to analyse it?

[bluegrassjunkie]

It was decided that occupancy modeling would be used in my project prior to data collection. I’ve never considered mine to be an occupancy study, but based on my reading of occupancy modeling I thought that such analysis could still be useful for my objectives by allowing me to take into account detectability and estimate raptor ‘use’ of control vs. exp. sites.

My sampling units are the 8 fields; yes, I'm interested in the presence/absence of raptors at the scale of the field. I’d like to incorporate occupancy modeling to estimate psi (site use) for control fields and for experimental fields, then use ANOVA to determine differences between treatments. During each survey, I recorded raptor visitation data (e.g. time spent in the field and activities) so that I can also compare numbers of visits and visitation rates, etc.

I understand about including detection histories from several sites into one file. In addition to the ‘raptors-all species’ detection histories, I have compiled detection histories by species. I would first like to look for differences between treatments for raptors in general, then for each species. What I’m not clear on is how to create my input files so that I can compare parameter estimates between control and experimental sites. Does it make sense to analyze the 4 control sites in one file, and the 4 experimental sites in another file, then statistically compare the estimates?

If I am misunderstanding the applicability of occupancy modeling for determining my objectives, can you suggest a more appropriate approach?

Thanks so much.

[bacollier]

My sampling units are the 8 fields; yes, I'm interested in the presence/absence of raptors at the scale of the field. I’d like to incorporate occupancy modeling to estimate psi (site use) for control fields and for experimental fields, then use ANOVA to determine differences between treatments. During each survey, I recorded raptor visitation data (e.g. time spent in the field and activities) so that I can also compare numbers of visits and visitation rates, etc.

Revisiting what Darryl had said in his earlier response, if you surveyed 8 fields, and your data are as you show earlier in this thread, then estimating 'occupancy' is done, the answer is 1 as all your sites were occupied by raptors at least once (e.g. you have no all zero sites). If you had the data to compare occupancy probability for treatment vs. control sites (which you don't imho), I would not recommend using AOV, rather you should create a couple of models, one with a effect of group (trt or control), one without that effect, and then see which model was ranked (based on AIC) best (this approach was actually an example in several of the B&A articles and there is an example in the MARKBOOK as well).

I understand about including detection histories from several sites into one file. In addition to the ‘raptors-all species’ detection histories, I have compiled detection histories by species.

It is plausible that you can do species-specific occurrence estimation at the field level (but see Darryl's previous comments on what you are estimating).

I would first like to look for differences between treatments for raptors in general, then for each species. What I’m not clear on is how to create my input files so that I can compare parameter estimates between control and experimental sites. Does it make sense to analyze the 4 control sites in one file, and the 4 experimental sites in another file, then statistically compare the estimates?

Why separate them and then compare statistically? Create one file for each species with a 'group' effect of treatment/not and compare as I outlined above using AIC. FYI, all this is detailed in the MARKBOOK, RTM.

If I am misunderstanding the applicability of occupancy modeling for determining my objectives, can you suggest a more appropriate approach?

My $0.02 (which basically mirrors what Darryl said earlier)-At the 'sample unit' level (field), your data already indicate that occupancy is 1 for all sites (e.g., raptors were located at least 1 time), so estimating occupancy is moot, its 1 for both treatment and control. You might tease out some detection/non-detection stuff as related to use as Darryl indicated, and you might be able to estimate treatment/control at the individual level depending on your data structure for individual species (but with a sample of 8, I doubt seriously you will be able to extract any useful data on the relative effects of the treatment vs. control unless one group is all occupied and one is never occupied, and then is may just be a sampling artifact as I assume that you are likely violating the closure assumption which is needed for the occupancy modeling that you are attempting).

Bret

[bluegrassjunkie]

I appreciate your help. I ran the models (by species) looking at group effect (control vs. treatment) as you suggested. Sure enough, as far as I can tell the results are junk, probably due (as you predicted) to the small sample size.

At this point I am searching for another possible way to model my visitation data. Any suggestions?

Thanks!