overall psi and p estimates

[sushma]

Hi, I'm a new user of the Presence software and I'm using version 5.1. I have a basic query. I would like to know how to obtain overall estimates of psi and p for each model when covariates are used since Presence output gives estimates for each site and not one estimate for the model.

[jhines]

You could just take the average of each site, which gives a single estimate weighted equally by each site. Or, just use the estimate from a model without covariates, which takes into account the variance of each site. The 2nd method seems preferable as imprecise site estimates will have less influence on the result than precise estimates.

Jim

[darryl]

Depends on what you mean in terms of 'overall'. Are you only interested in the places you've surveyed or do you want to extrapolate to other places too? Strictly speaking, psi and p are probabilities and if you've identified important covariates that suggests there is no 'overall' value as the probabilities are different under different conditions.
Darryl

[sushma]

Hi, what i mean is, when you're modelling occupancy and detection probabilities as a function of certain covariates, Presence returns estimates for each of the sites. How do you obtain one single value and the associated SE for a particular model?

[darryl]

I understand that. My question is what is that single number supposed to represent? Is it an average probability for all surveyed places, an estimate of what proportion of the surveyed places were occupied (not these are not the same thing), or you wanting to use that number of apply it to all places of interest including that those were not surveyed? Once we understand what you want that single number to mean, we can make some suggestions on how to use your results most appropriately to get that number.
Darryl

[sushma]

Hi, I want to obtain an estimate of what proportion of surveyed sites were occupied and average detection probability as a function of covariates. Additionally, I'm also interested in obtaining occupany estimate for unsampled sites given a set of site-specific covariates. My site covariates are all GIS based covariates.

sushma

[jhines]

To get the expected number of surveyed sites which are occupied, you could sum all of the conditional psi's. Then convert that to a proportion by dividing by the number of sites.

To get the prob of occupancy for unsurveyed sites, you can compute psi for them by plugging the beta's into the inverse-logit function. For example, if psi is a function of a single site covariate, X(i), where X(i) is the covariate value for site, i, then psi(i)=1/(1+exp(-Beta0 - Beta1*X(i)). Or, an easy way to get PRESENCE to do this for you is to add the additional sites to your data, entering the covariate data as well and entering missing values for each site. (Just don't add in these sites to the sum mentioned above.)

I'd compute average detection probability by copying all of the detection probabilities for each site and survey which did not have missing data to a spreadsheet and compute the average.

[stephanietodd]

Hi,
What about the the standard error for the expected number of surveyed sites that were occupied? I would like to compare the naive occupancy estimate with the estimated occupancy when detection is taken into account for my top model/s
Cheers,
Steph