Simulations for robust occupancy models

[Linda Allison]

I would like to use the simulation module to compare effectiveness of different study designs for robust occupancy. The study designs involve different permutations of running 80 site-visits per year. In a given year, some sites will be visited multiple times; others only once. To accommodate other work we want to do, we are considering whether we have to focus on a limited number of sites over the whole study, or whether we can afford to keep many more sites in the target group, without visiting all sites each year.

No matter what we do, and in common with published studies using robust occupancy designs, all designs we are considering include lots of missing data, representing site-date combinations that were not surveyed. Is there a way to have MARK simulate data with missing values? Since this is a property of the data/design, not of the occupancy/detection parameters, I tried generating the ‘true’ data assuming all sites were visited on all dates. Then for my estimation models, I treated each site as a group, and set p(detection)=0 for dates when I wanted missing data. Assuming this would compromise parameter estimates, I thought it would still give a reasonable ranking of different study designs. However, in my test runs, MARK usually shuts down.

Any suggestions?

Linda

[cooch]

While the simulation capabilities in MARK are pretty good, they are not sufficiently general to handle al the sorts of situations you're likely to run into in 'real world' application (at least, in my opinion).

My general recommendation then is to write your own simulation code. This does 2 things: (1) gives you a general, and flexible code-base to allow you to easily simulate problems that either are not implemented in MARK, or are too cumbersome to execute in MARK, and (2) writing the code yourself forces you to really understand how the data are generated (this is also one of the advantages to using SURVIV for analyses) - by writing the logic sequences to generate the probability structures of your model, you will invariably learn a lot more about the models, and the underlying theory, than you will if you simply rely on the simulation capacity in MARK.

In general, writing your own code involves 2 steps: generating the basic probability structures and assigning 'fates' to each simulated individual, and then figuring out how to get your program to generate encounter histories. Most people find the former easier than the latter, in fact.

I do all of my simulation work in SAS - all of the 'made up' examples in 'the book' were generated that way.

[Linda Allison]

I did use SPSS to simulate an initial file of encounter histories in order to build models for different designs. So I was able to rank models for one set of data. However, I do not plan to simulate 100 different history files, respecifying the same models in Program MARK for each file!

Can get to a more general result by adding simulations through Program MARK? Is there a way to specify missing site-time information that is inherent in robust occupancy/proportion-area-occupied designs?

On the other hand, my experience with these designs and empirical datasets so far is that delta-AICs for different models are relatively large/huge. Perhaps model rankings are pretty stable and many simulations aren't needed to get a general result. I wonder if the large delta AICs are due to the large number of missing values in these datasets...

Thanks

[tomo_eguchi]

Linda,

There is a way to run Mark.exe without going through the GUI. If you have 100 simulated data files, you can process those files by writing input files to mark.exe (which can be done with a programming language of your choice), then running mark.exe with a batch file.

Look at intermediate files that Mark creates when you run a specific model of your one dataset. I think these files have extensions, .x, .y, and .z. I forget the details but one of them is the input to mark.exe. Imitate the file, creating necessary input files for all simulated datasets. Then create a batch file, which looks like this:

mark i=inputFilename1.inp o=outfilename1
mark i=inputFilename2.inp o=outfilename2
...

save this file filename.bat, where filename is anything you'd like to call this file. Double click the filename in your windows explorer and you should see mark goes through all these data files. Note if you place this batch file in a directory that doesn't have mark.exe, specify the exact path where mark.exe is in the batch file. For example, the previous three lines look like:

c:\programs\mark\mark i=inputFilename1.inp o=outfilename1
c:\programs\mark\mark i=inputFilename2.inp o=outfilename2
...

Good luck,

Tomo