two species output doesn't match single species

[CJWM]

Greetings,

I'm trying to run a two-species analysis using the psiBa parameterization. We have previously run the two species separately using the single season, single species models, so I figured I should start with the covs that were important for each species from that analysis. But, when I run a model that assumes occupancy and detection of both species are independent, the values of the betas are completely different than those of the single species models. I have the design matrix arranged as:

psiA 1 0 cov1 cov2 0 0
psiBA 0 1 0 0 cov3 cov4
psiBa 0 1 0 0 cov3 cov4

and i have detection set up as pA=rA, and pB=rBA=rBa

Am I incorrect in assuming that estimates from this model should mirror those of the single species models?

Thanks,

Chris

[jhines]

Chris,

Yes, the 2 species output should match the single-species output. I'd be glad to check into it if you send me the 2 project backup files (presence_backup5.zip).

Cheers,

Jim

[dgtendale]

Dears,

I would like to know the final answer of this question, please. I am having almost the same problem.

I have two species. I analysed them using "Two species" single-season using PRESENCE 3.0.

I want to know the independent as well as interaction occupancy of the species. I am in a situation that could not differentiate which species represented as "A" and "B". I thought the program assign the names (A and B) according to their sequence in the Input Data Form. To make sure I am talking about the same species, I entered the data in reverse order and saved it in other name and analysed it in the same way. I got the same real estimates for the same A and B (in the same order). So I do not know which species is represented as "A" and "B".

It is maybe stupid but to get some clue I analysed each species in “single season” separately. But their result is far different from the results I got from "two species" analysis. Probably, this has something to do with the interaction effect. But would you give me a clue to fix this, please?

Totally, I have 1000 rows (sites) for both. I have 11 columns for each species so totally I have 22 columns. I entered the first species in the first 11 columns, the next 11 for the second species.

Regards,
degitu

[jhines]

The data for the two-species, single-season model should contain all sites for the first species (species A), followed by all sites for the 2nd species (species B). It sounds like you joined the data horizontally, instead of vertically. If you have 1000 sites for each species and 11 surveys, you should have 2000 rows and 11 columns of data. PRESENCE will assume the first 1000 sites are for species A, and the rest are for species B.

Alternatively, you could create a file with 1000 rows and 11 columns, where you use codes 0,1,2 and 3 for each site and survey. In this format, 0 means neither species detected, 1 means only species A detected, 2 means only species B detected, and 3 means both species detected. PRESENCE will detect which format is used by looking at the data.

Jim

[dgtendale]

Hello Jim,

Wow,I definitely messed it up then.

Thank you so so much.

degitu

[tajensen]

Hello,

I am also running a two-species analysis and am getting different estimates than than those obtained from the single species analysis. For example, pA = 0.605 (0.085) from the single species analysis, but pA = 0.444 (0.172) from the two-species analysis. I have double checked my data file and it is formatted correctly, with species A stacked above species B. My design matrix looks like this:

psiA 1 0 0
psiB 0 1 0
phi 0 0 1

pA[1] 1 0 0 0 0
pA[2] 1 0 0 0 0
pA[3] 1 0 0 0 0
pB[1] 0 1 0 0 0
pB[2] 0 1 0 0 0
pB[3] 0 1 0 0 0
rA[1] 0 0 1 0 0
rA[2] 0 0 1 0 0
rA[3] 0 0 1 0 0
rB[1] 0 0 0 1 0
rB[2] 0 0 0 1 0
rB[3] 0 0 0 1 0
delta[1] 0 0 0 0 1
delta[2] 0 0 0 0 1
delta[3] 0 0 0 0 1

There are 20 sites and 3 surveys - which I realize is less than desirable - so I'm wondering if the difference in estimates is due to the small number of sites? I have also run the analysis using the psiBa/rBa parameterization and am getting similar discrepancies with the single species analysis.

Tim

[darryl]

Hi Tim,
Note that pA has different interpretations in single vs 2 species models. In order for the two analyses to match up you need to set up the equivalent 2-species model such that everything is independent, as per the first post in this thread (but note the first poster is using a different parameterisation to yourself). To make everything independent you need use the following design matrix (with your parameterisation) to get the equivalent of a psi(.)p(.) model fit to each species seperately.

Cheers
Darryl


psiA 1 0
psiB 0 1
phi 0 0

pA[1] 1 0
pA[2] 1 0
pA[3] 1 0
pB[1] 0 1
pB[2] 0 1
pB[3] 0 1
rA[1] 1 0
rA[2] 1 0
rA[3] 1 0
rB[1] 0 1
rB[2] 0 1
rB[3] 0 1
delta[1] 0 0
delta[2] 0 0
delta[3] 0 0

[tajensen]

Yes, that worked. Thanks Darryl.

-Tim

[jhines]

Another way to think about this is that the single-species model has 2 parameters (for not survey-specific detection) for each species, so the 2-species model should have 4 estimated parameters. It might be a little easier to use the 2nd parameterization for the comparison so you can constrain the parameters as follows:

Code: Select all

-     a1  a2
psiA   1  0
psiBA  0  1
psiBa  0  1

-   b1  b2
pA   1  0
pB   0  1
rA   1  0
rBA  0  1
rBa  0  1


So, occupancy for species B is the same, regardless of whether species A is present or not (psiBA=psiBa). Detection probabilities don't depend on whether or not both species are present (pA=rA, pB=rBA=rBa). (I didn't list each survey in the rows of the detection design matrix, but each row would be repeated for the number of surveys.)

Note: I noticed a bug in my output for this parameterization where it used the wrong link function for the last parameter (rBa). I've corrected this and uploaded the new version.

Jim