creating cohort for square PIMS

posts related to the RMark library, which may not be of general interest to users of 'classic' MARK

creating cohort for square PIMS

Postby j.harv3y » Thu Feb 06, 2020 12:24 pm

Hi,
I'm trying to create a cohort variable for a square PIMS. I have added the year first seen in the input ch (as Year) for each animal and am using the following;

Code: Select all
capt.pr <- process.data(data,begin.time = 2014, model = "POPAN",
                       groups = c("sex", "Year"))


I'm not sure if I've done this correctly because despite being the 'top model' (with lowest AICc) the standard errors are absolutely huge. I would guess that the data isn't sufficient for year covariets however the AIC is far 'lower' and any dot or otherwise models.

Code: Select all
                                                   model npar     AICc  DeltaAICc       weight Deviance
52                     Phi(~Year)p(~time)pent(~Year)N(~1)   19 505.6249   0.000000 5.761502e-01 185.3868

Output summary for POPAN model
Name : Phi(~Year)p(~time)pent(~Year)N(~1)

Npar :  19  (unadjusted=10)
-2lnL:  463.8811
AICc :  505.6249  (unadjusted=484.91881)

Beta
                    estimate           se           lcl           ucl
Phi:(Intercept)    1.0088768 1.965698e-01  6.236000e-01  1.394154e+00
Phi:Year2015      -0.0482977 4.064769e-01 -8.449924e-01  7.483971e-01
Phi:Year2016      -1.0978776 5.134463e-01 -2.104233e+00 -9.152280e-02
Phi:Year2017      -0.5171342 5.622330e-01 -1.619111e+00  5.848425e-01
Phi:Year2018     -41.8050200 4.801592e+04 -9.415301e+04  9.406940e+04
Phi:Year2019      63.8141650 7.270434e+04 -1.424367e+05  1.425643e+05
p:(Intercept)      2.9985974 8.850260e-01  1.263946e+00  4.733249e+00
p:time2015        -0.5741594 1.010561e+00 -2.554858e+00  1.406540e+00
p:time2016         0.4431593 1.335746e+00 -2.174902e+00  3.061221e+00
p:time2017        -1.3503223 1.021704e+00 -3.352861e+00  6.522169e-01
p:time2018        -1.2904555 1.024393e+00 -3.298266e+00  7.173554e-01
p:time2019        12.7330420 2.247753e+03 -4.392863e+03  4.418329e+03
pent:(Intercept)  -5.5745870 1.761071e+00 -9.026287e+00 -2.122887e+00
pent:Year2015     78.2985190 8.895199e+04 -1.742676e+05  1.744242e+05
pent:Year2016     72.5578700 8.251250e+04 -1.616519e+05  1.617970e+05
pent:Year2017     42.7271000 4.905025e+04 -9.609577e+04  9.618123e+04
pent:Year2018     57.8648340 6.603079e+04 -1.293625e+05  1.294782e+05
pent:Year2019     55.8441220 6.376409e+04 -1.249218e+05  1.250335e+05
N:(Intercept)    -19.9187590 3.402224e+03 -6.688277e+03  6.648440e+03


I have two further questions about including the cohort covariate which I'm a little confused about (and have read the workshop notes) and I'm wondering if this could be the problem partly

-how can I include first capture event in the likelihood?

- There is a bias towards when surveying started (2014) so would it be best to set '2014' to NA in the input, that way the model will just look at animals definitely seen in 2015+?

Looking forward to any suggestions!

Many thanks in advance
j.harv3y
 
Posts: 45
Joined: Mon Oct 08, 2018 4:45 am

Re: creating cohort for square PIMS

Postby jlaake » Thu Feb 06, 2020 7:49 pm

You should not be doing this. What is your reason for including a cohort variable? Essentially you have used the observed entries (year) to estimate the probability of entry. Do you see the problem with that? That would be like trying to estimate p using a covariate that is 1 for those seen and 0 for those not seen. Data should not be used as a covariate to explain the same data. That is what you are doing. Sure the aic will be better because you are using the data to explain the data. Not very useful.
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: creating cohort for square PIMS

Postby j.harv3y » Fri Feb 07, 2020 7:09 am

jlaake wrote:probability
not sure why I can't quote your reply!

Ah my bad, I just used the cml to run all possible combinations and hadn't thought about not including it in pent :oops:, completely makes sense! Thanks for explaining.

I want to see if year effects survival of the animals, so would you recommend doing this a different way for phi?

Thanks so much!
j.harv3y
 
Posts: 45
Joined: Mon Oct 08, 2018 4:45 am

Re: creating cohort for square PIMS

Postby jlaake » Fri Feb 07, 2020 9:52 am

Jolly-Seber type models use the first capture event in the data. This means that a 0 before the first 1 is relevant. It can be 0 because the animal hasn't entered the population yet or it entered but wasn't seen. If it had entered the population then it has to survive until the first time it was seen. Now consider what survival rate would you apply for that animal prior to it being first seen. You don't know when it first entered so you can't assign it's cohort value (your year variable). You can make survival vary by time (not cohort) because that is known for all animals. In essence you are assuming survival is the same for those first seen and those in the population but not yet seen. Use time and not year.
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: creating cohort for square PIMS

Postby j.harv3y » Fri Feb 07, 2020 5:46 pm

jlaake wrote:Jolly-Seber type models use the first capture event in the data. This means that a 0 before the first 1 is relevant. It can be 0 because the animal hasn't entered the population yet or it entered but wasn't seen. If it had entered the population then it has to survive until the first time it was seen. Now consider what survival rate would you apply for that animal prior to it being first seen. You don't know when it first entered so you can't assign it's cohort value (your year variable). You can make survival vary by time (not cohort) because that is known for all animals. In essence you are assuming survival is the same for those first seen and those in the population but not yet seen. Use time and not year.


Ah ok makes sense, thanks for explaining! I was getting confused between cohort and TSM/age, after rereading the chapter in the handbook and the workshop notes its actually TSM I think I need.
Can I just double check then, TSM can be done (depending on the question) for JS models, using viewtopic.php?f=21&t=2591 as a guide on how to create the TSM in .ddl?

Thanks again really appreciate the guidance,

Jess
j.harv3y
 
Posts: 45
Joined: Mon Oct 08, 2018 4:45 am

Re: creating cohort for square PIMS

Postby jlaake » Fri Feb 07, 2020 7:45 pm

I have never tried that but it should work since newly caught and not previously caught will both have TSM=0. So that seems valid.
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA


Return to RMark

Who is online

Users browsing this forum: Google [Bot] and 8 guests