Psi estimate less than naive estimate

questions concerning analysis/theory using program MARK

Psi estimate less than naive estimate

Postby heg90 » Wed May 22, 2019 8:32 pm

Hi,

I'm running a single-season single-species occupancy model and am struggling to understand why my top model estimate of occupancy (psi = 0.24) is less than my naive occupancy estimate (psi = 0.38). I detected individuals at 16/42 sites with no false positives, so I'm not sure how it's possible that my top model is reporting an occupancy of 0.24, when the null model has a much worse AIC score and reports an occupancy estimate of 0.42 in MARK.

I have been working in both PRESENCE and MARK. My models in both programs reflect the same AIC scores and beta estimates, so I know they're both running the same between programs. Oddly enough, when I calculate the overall psi estimate manually for my top model in PRESENCE by summing the psi-conditional and dividing it by the total number of sites I get psi = 0.42. Since I know the top model in MARK and PRESENCE are identical, why would I be getting a difference of psi = 0.24 in MARK and psi = 0.42 in PRESENCE? Am I looking for overall occupancy in the wrong place in MARK? I was using the estimate for psi that is found under 'real estimates.'

I did read a post in a different thread that said "You shouldn't be getting an occupancy estimate from the standard single season that is lower than the naive estimate. Or are you talking about a model that has covariates in it? If that case the individual estimate for some sites might be lower," but it was unclear what it meant. Does that mean this can sometimes occur when covariates are added? My top model has one covariate for psi and two for p.

Thanks for any insight!

~Holly
Last edited by heg90 on Wed May 22, 2019 8:39 pm, edited 1 time in total.
heg90
 
Posts: 20
Joined: Sun Nov 04, 2018 4:52 pm

Re: Psi estimate less than naive estimate

Postby darryl » Wed May 22, 2019 8:39 pm

Hi Holly,
I'm going to guess that your model has covariates in it, in which case the real estimate reported by MARK is the estimated for the first site only (unless you changed MARK's default options), not an overall estimate.
Cheers
Darryl
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Psi estimate less than naive estimate

Postby heg90 » Wed May 22, 2019 8:40 pm

Hi,

Thanks so much for your quick reply! You beat me to that, I was just editing my post to say I saw somewhere that having covariates will change the estimate. Is there a way in MARK (or PRESENCE for that matter) to get an overall estimate of psi with standard error?
heg90
 
Posts: 20
Joined: Sun Nov 04, 2018 4:52 pm

Re: Psi estimate less than naive estimate

Postby cooch » Wed May 22, 2019 9:00 pm

If you haven't already done so, you should read Chapter 21 in the MARK book -- specifically, section 21.1.3, which deals specifically with 'covariates'. The issue of an 'overall estimate' is vague, since modeling with covariates means the interest is in the value of a parameter estimated for a given value of the covariate. MARK allows a fair bit of flexibility with what actually gets reported (e.g., you can report for a mean value of the covariate, or a specific value, etc). Again, this is covered in the noted section of Chapter 21.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Psi estimate less than naive estimate

Postby cooch » Wed May 22, 2019 9:08 pm

darryl wrote:Hi Holly,
I'm going to guess that your model has covariates in it, in which case the real estimate reported by MARK is the estimated for the first site only (unless you changed MARK's default options), not an overall estimate.
Cheers
Darryl


Default in MARK is, IIRC, to estimate using the mean value of the covariate.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Psi estimate less than naive estimate

Postby heg90 » Wed May 22, 2019 9:41 pm

Thank you for your response.

I'm curious what you mean by "the issue of an 'overall estimate' is vague." It seems like in Ch. 21 it discusses the possibility of generating an overall estimate of occupancy from models with covariates:

"Another estimator that is often used, if you want to incorporate covariate effects, is to calculate the mean of the site-specific occupancy estimates for each site in the sample. For example, let’s suppose that our second model{ψBAOW pBAOW} was much better supported than the constant occupancy model. In this case,we might want to know the estimated proportion of sites occupied using this model because it accounts for variation in occupancy probabilities among sites."

It goes on to state that this is done by exporting the site specific estimates from the "individual covariate plot" and calculating the mean and standard deviation.

Do you mean that there is a fault to this method, or simply that MARK's way of reporting psi estimates with covariates is vague?
heg90
 
Posts: 20
Joined: Sun Nov 04, 2018 4:52 pm

Re: Psi estimate less than naive estimate

Postby cooch » Wed May 22, 2019 10:05 pm

heg90 wrote:Thank you for your response.

I'm curious what you mean by "the issue of an 'overall estimate' is vague."


Vague, in the sense that you didn't explicitly define what you meant by 'overall'. The word 'overall' could mean any number of different things. It could be the mean, or it could be something else entirely...
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Psi estimate less than naive estimate

Postby cooch » Wed May 22, 2019 10:07 pm

heg90 wrote:Do you mean that there is a fault to this method, or simply that MARK's way of reporting psi estimates with covariates is vague?


Nothing MARK (or PRESENCE) does is 'vague'. It reports what it reports. What was vague is a statement of what you were after. MARK or PRESENCE can provide whatever you want, generrally - provided you know what you want.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University


Return to analysis help

Who is online

Users browsing this forum: Google [Bot] and 9 guests