Single site analysis?

[amdurso]

This may be a rudimentary question, but I've been asked to analyze some data (estimate detection probability) that spans several years (multiple visits per year) of surveying only a single site. I tried running a set of models with covariates for p that used only one site and it seemed to work, although I noticed that many of the models received identical support (i.e. the AIC was exactly the same for several models of different structure). Is this data structure going to cause problems/violate assumptions/limit the application of the output? I'd love to know about any pitfalls or prophylactic remedies that I could apply before I do the full analysis.

Thoughts?

-Andrew Durso
Graduate Student, Biological Sciences
Eastern Illinois University
amdurso@gmail.com

[bacollier]

Andrew:

Define "site"? I assume you mean you have multiple sample points where P/A surveys are conducted at those points with a single study site (like point samples for birds a woodlot) multiple times each year? Thus, you have a capture history with presence/absence data which looks like

101 110 001 111 (4 years with 3 surveys per year-robust design)?

I guess am unclear on what you mean "running a set of models with covariates for p that used only one site"?

Can you clarify?

Bret (Go Panthers! EIU 97')

[amdurso]

Sorry, I should have been more specific. The data are for snakes caught along a road route - so there are no discrete points at which resampling took place - all the captures are opportunistically taken from the road at the point where the animal happened to be - I know the duration of each sampling period, but nothing about where specifically the animals were or even how I could divide up the route into sections. It's a single elongate sample point.

You're right on with the capture history - although there are different numbers of samples each year.

The model set I ran allowed p to vary with each of four covariates individually, plus 2- and 3-way combos, plus a null model and a global model. The data had only a single site with many samples.

[bacollier]

Sorry, I should have been more specific. The data are for snakes caught along a road route - so there are no discrete points at which resampling took place - all the captures are opportunistically taken from the road at the point where the animal happened to be - I know the duration of each sampling period, but nothing about where specifically the animals were or even how I could divide up the route into sections. It's a single elongate sample point.

You're right on with the capture history - although there are different numbers of samples each year.

The model set I ran allowed p to vary with each of four covariates individually, plus 2- and 3-way combos, plus a null model and a global model. The data had only a single site with many samples.

Andrew:
OK, so your saying that you had 1 sampling location (the road), and resamples were taken opportunistically all along that road? How did you create the capture history? Based on what you said, you have 1 row of data (where a row represents the sampling location), and either you found, or did not find a snake on the road (sample transect, not point) during each survey period. If so, then you may be out of luck as there is not much you can do with one row of data in a CMR framework.

Am I missing something, did you have multiple roads which were surveyed so you have >1 row of data?

Also, what is the logic in your posited set of candidate models? Seems like you are running a pretty big subset if you class all the combos you describe.

Bret

[amdurso]

That's right - a single row of data. I didn't do the data collection - I was just asked to help analyze it. There are many species of snake that we're interested in - so interspecies comparisons are the goal of the study - but I think the data collection began way before the analysis framework was considered. I suppose the road could be subdivided into sections based on the surrounding habitat (the road is >20 km long), but I'd need a lot more information (which I might be able to get) - even then, each "site" would be a section along a single transect and they wouldn't really be independent.

[amdurso]

Regarding the logic for the model set - I was more interested in seeing how the weights would vary between models in a large set than picking the most biologically meaningful analysis - but if the data can be analyzed at all, I'd parse that set down to only the ones that made sense.