by jhines » Thu Dec 13, 2007 3:34 pm
To model 10 species in PRESENCE, construct the data such that each site/species combination is a separate record. Then construct site covariates with one column for each species, where the column would contain '0' if the particular record did not corresponded to that species, and '1' if the record did correspond to that species.
So, if the first 5 histories were observations for species1, then the first species covariate column would contain '1' and the other 9 species covariates would contain '0' for the first 5 records. If histories 6-10 were observations (at the same or different sites) for species2, then the first species covariate column would contain '0', the 2nd covariate would contain '1', and the rest would contain '0' for records 6-10.
To build a model where all 10 species have the same occupancy(psi) and detection(p), run a single-season model as you would for only one species, ignoring the 10 new covariates.
To build a model where each of the 10 species has a different psi, add a 9 columns in the 'psi' page of the design matrix, and enter the name of each of the species covariates (except the first) in the newly created cells (click blank cell, goto 'Init' menu, select covariate name). This causes PRESENCE to compute occupancy as a function of an 'intercept' term plus a term for each species (except the first). Since the covariates are either zero or one, occupancy will be computed as:
psi(species1) = f(1*a1) { since all covariates for spec2-spec10 are zero}
psi(species2) = f(1*a1+1*a2) { since covariate for spec2=1, others=0}
psi(species3) = f(1*a1+1*a3) { since covariate for spec3=1, others=0}
psi(species4) = f(1*a1+1*a4) { since covariate for spec3=1, others=0}
PRESENCE will estimate a1,a2,...a10, and compute a different psi for each species. Detection probabilities would be done the same way, except that you have the possibility of different estimates for each survey (in combination with each species). For a survey-specific p and different p for each species, you would need 2 columns (one intercept and 9 columns for each species covariate) for each survey.
More documentation on building models with a design matrix is available in the Cooch's online MARK book.
Jim