Psi output in MARK

[campbelb]

A quick question regarding the output from MARK for real parameters of Psi. I've pasted an example below.

Real Function Parameters of { p(~SURVEY + effort)Psi(~adPF500) }

I constructed a model where detectability varies as logit function of survey (factor) and effort (no. of point counts) and occupancy varies as logit function of forest cover. Am I correct interpreting this part of the output as the occupancy (Psi) held at means of the covariates which accounts for both sites occupied and detected (Psi * p) and sites which are occupied but species are undetected (Psi * (1-p))?

Brian

[jlaake]

Brian-

I'll address your question but first be careful of using something like Survey and Effort. If Survey is a factor variable and Effort is a measure of effort for each Survey then these covariates are redundant. The only way that is a valid model is if you have variation of Effort within levels of Survey.

Now to your question. Psi is getting measured for each level of the covariate adpf500 (whatever that is). I'm going to assume it is a factor (category) variable. You can imagine computing a naive occupancy estimate for each level of that covariate as the number of sites observed occupied divided by the number of sites with that level of adpf500. For each set of sites with all 0 capture histories, some of those are truly unoccupied and others are occupied but you simply didn't see them during the visits. The model for p is used to estimate the proportion of the 0 capture histories that were actually occupied. The estimate of the probability that the species would be observed on at least one occasion (p*) depends on the survey and effort variables specific to each site/occasion. You can think of the estimate of false 0's (truly occupied with nothing observed) as being "added" to those observed to be occupied to get an estimate of Psi at each covariate level. Read ch 4 of McKenzie et al for a more thorough explanation.

regards --jeff

[darryl]

Hi Brian
Not sure I fully understand your question, but here's my response to what I think you're asking.

Basically the covariates are averaged across all sites and surveys (where appropriate), and those values are plugged in as the predictor variable values in the set of logistic regression equations defined by you design matrix. The results are reported as the Real Function Parameters so you should have 1 psi value and the number of p values will depend on how you set up the PIMs and design matrix, but at most will be equal to the number of surveys per site.

Cheers
Darryl