power analysis - comparison of two occupancy estimates

[pcutter]

Hi,

I'm using equation 4 and the suggested optimization routine in:

MacKenzie, D. I., and J. A. Royle. 2005. Designing occupancy studies: general advice and allocating survey effort. Journal of Applied Ecology 42:1105-1114.

to calculate optimal survey effort allocation.

I would then like to reduce occupancy by a set proportion and compare the resulting means and variances in a power analysis framework.

Field et al (2007) did a nice job of this with a stochastic simulation. However, it seems that power could also be readily analyzed by capitalizing directly on the equation mentioned above.

Does anyone have any advice on whether this approach would be appropriate? Also, most two -sample power formulas rely on n samples but in this case it is really just one sample (the whole system) with an expected mean and variance. Are there better power fomulas under these conditions?

Thanks in advance for any insights.

[darryl]

You could use something along the lines of a paired-samples z-test as an approximation, although here you need to know the variance or standard error of the difference rather than of the 2 values. I don't know whether any formula has been published for SE(diff), but you could use a technique known as the analytic numeric approach (eg Burnham et al. 1987 Fisheries Monograph, Bailey et al. 2007 Ecological Applications) to determine it. This can be done relatively easily with a spreadsheet and PRESENCE or MARK.

Cheers
Darryl

[pcutter]

Hi Darryl,

Thanks for that. I had been exploring GenPres but went in a slightly different direction when I came across your suggested routine for finding optimal balance of sites and occasions (which is really my primary interest right now). I like being able to attach a cost function to the variance equation so that I can explore those patterns under different phi and p values and for different cost and variance constraints.

I have been able to use the numeric/analytic estimates from GenPres but I can't seem to put my finger on a way to run this so that large numbers of pairs of datasets (that would represent say a 10% reduction in occupancy under different sampling strategies, cost, or variance constraints) to be compared. Is there a way to do batch runs in GenPres. Or, for that matter, even outputting all the raw data sets in a simulation? I know I am probably missing something but i am stuck on figuring out a routine to optimize the cost/power tradeoffs under multiple survey configurations.

Any ideas for linking the optimization goal with the products of GenPres?

Thanks again for your thoughts on this.

Pete

[darryl]

I suspect there would be a way of of running GenPres in batch mode; Jim Hines would be the one to ask. But, GenPres is basically just a data-generation program which then links to either PRESENCE or MARK to analyse the data, then pulls the results back in.

However, if you're aim is to optimise design in some regard, then the first step would be write down a function expressing what exactly you are trying to optimise. After that, the best approach would be to find the optimum either by trial and error, or applying a bit of calculus (usually a much better approach in the long term). The latter is how we came up with those results in the paper you refer to.

Cheers
Darryl

[jhines]

Pete,

Under the 'options' menu in GENPRES, there is an option to save all output files. You could use this if you're simulating with the same parameter values.

I updated the GENPRES help file to show how to do batch simulations, and included a sample script.

Cheers,

Jim

[pcutter]

With regard to using GenPres, is there any way to produce simulations where Psi varies over say 2 or 3 seasons (years)?

This scenario seems to be at the core of many monitoring/power questions.

One could then compare summary statistics for Psi(.) with Psi(t) models for hypothesis-testing.

Thanks,
Pete

[darryl]

Pete,
In the 'model type' model you can select multi-season model to generate the data, then fit an appropriate model from the 'model' menu using either simulation or expected data. For more details, go to Help>Overview; I'm sure what you need to get started will be covered in there.
Cheers
Darryl

[pcutter]

Hmmm...

Hope I'm wrong but by all appearances, Psi seems to be constant (e.g. only one window for setting a single Psi over multiple years...)

Chalking it up to the wish list for now I guess.

Much appreciation for this great tool though.

Pete

[darryl]

When you select multi-season from model type menu you should get some boxes for specifying epsilon (extinction) and gamma (colonization) values. By changing these you can specify the future values of psi. See MacKenzie et al 2003 Estimating site occupancy ... Ecology, or MacKenzie et al 2006 Occupancy Estimation and Modeling if you're not sure of the relationship between occupancy, colonization and extinction.

Good Luck

[pcutter]

Jim and others,

I've got awk running but get the following when I try to run your example file:

Code: Select all

C:\Documents and Settings\cutt0016>awk -f C:/genpresBatchExample.awk
awk: C:/genpresBatchExample.awk:17:     a="cmd /c Q "c:\\progra~1\\presence\\genpres2" Q " 5 100
awk: C:/genpresBatchExample.awk:17:                   ^ syntax error
awk: C:/genpresBatchExample.awk:17:     a="cmd /c Q "c:\\progra~1\\presence\\genpres2" Q " 5 100
awk: C:/genpresBatchExample.awk:17:                    ^ backslash not last character on line

I've also tried with the following command:

Code: Select all

C:\Documents and Settings\cutt0016>awk C:/genpresBatchExample.awk
awk: C:/genpresBatchExample.awk
awk:  ^ syntax error
awk: C:/genpresBatchExample.awk
awk:    ^ unterminated regexp
errcount: 2

Am I leaving something out? Is the code possibly missing a "\" required by MARK somewhere?

Thanks for your help.

Pete