More parameters than expected according to GEMACO sentence

questions concerning analysis/theory using programs M-SURGE, E-SURGE and U-CARE

More parameters than expected according to GEMACO sentence

Postby simone77 » Fri Nov 18, 2011 1:32 pm

I am not sure it is a good idea to send a new topic before getting an answer to the first one. I apologize for that and hope these issues may be of some general interest.

This is my data specifications:
1 group,
0 individual covariates,
3 states,
4 events (not seen, no antibodies, antibodies, unknown)
1 age class
9 occasions

I am running an analysis in E-Surge whose pattern matrices are defined in GEPAT just like this (copied and pasted from another thread):
Image

I have defined in GEMACO these sentences (in capital letters the pattern matrices):
INITIAL t
SURVIVAL f.t
TRANSITION f.t
CAPTURE f.t
STATE ASSIGNMENT f.t

the problem arises from the results I get with state assignment (event, step 2), that are these:
Image
I can't understand why there is an age effect that (i) I have no specified and (ii) I have set just one age class in the data.
Anyway, in order to remove the age effect, I have tried to set:
STATE ASSIGNMENT [a(1:2)].f.t
but the result is just the same, why?

Thanks for any help,

Simone
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm

Re: More parameters than expected according to GEMACO senten

Postby ganghis » Fri Nov 18, 2011 1:55 pm

Hi Simone,
First, note that I believe ESURGE reports all combinations of real parameters for each value of from, to, age, etc. - if you look closely, you'll notice that a lot of them are repeats, i.e., there's no difference between an age 2, time 2 parameter and an age 1, time 2 parameter. So there isn't an age effect being estimated.

One trick with these models is that you DO always want an age effect on capture probability to account for first capture. In particular, since you're modeling the first capture occasion with these models, you want to set p=1 for age=1. So you'll want a model of the form "a(1)+a(2).[model]" where [model] is 'f' or whatever. Then make sure to fix the a(1) p parameter to 1.0 before running the model.

Paul
ganghis
 
Posts: 84
Joined: Tue Aug 10, 2004 2:05 pm

Re: More parameters than expected according to GEMACO senten

Postby simone77 » Fri Nov 18, 2011 2:16 pm

Uups...

You are right, I didn't realize that the values were equal. I was misled by the fact the first value for age 1 was different from age 2 but it is just because there is no time 1 for age 2!

Also, I have been misled because these all possible combinations (with repeated values as seen above) are shown just for the last pattern matrix (event, step 2). This seems to me something similar to the result of a DM in MARK where you have cut several columns to pass, for instance from a time effect to a constant and you keep the rows, and you get many identical values as rows you have for that parameter.

Thank you for your help

Simone
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm

Re: More parameters than expected according to GEMACO senten

Postby simone77 » Sun Nov 20, 2011 1:14 pm

Going on with this, I have done a trial to understand how the age effect works on the different kind of parameters (Initial State, Transition, Event).
The case study is the same specified above in the first post.
I have set one of the two sentences as the same for each step:
1) a(1).[f.t]+a(2)
2) a(1)+a(2).[f.t]

1) a(1).[f.t]+a(2)
Given I have specified just one age class I would think that the (1) should get as many parameters as resulting from the interaction of full time (t) and departure state (f) and nothing more, indeed I believed there is no a(2).
But, (1) gives the same results I would expect more one parameter of age 2 for event step one (Capture) and other 16 parameters (since time=2 to time=9) of event step two (in my case state assignment).
These last 16 parameters are of the same value (and CI) and this means that no time and departure state effect is shaping these parameters.

2) a(1)+a(2).[f.t]
For the same reasons I would expect that (2) should get just one parameter for each step, given age 2 does not exist.
In fact, (2) gives one parameter for Initial State, one for Transition step one (Survival) and one for Transition step two, but also it gets 17 parameters for Event step one (1 for age 1 and 16 for age 2 because are since time=2 to time=9) and 34 parameters for step two (18 for age one and 16 for age 2 because they are since time=2).
Actually, Event the 18 parameters for age 1 of Event step one are of the same value (and CI).

I would like if someone might explain this as it could be useful to be understood.

Simone
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm

Re: More parameters than expected according to GEMACO senten

Postby CHOQUET » Mon Nov 21, 2011 6:06 am

Selecting A classes of age in E-SURGE would mean :
For A=1, one class of age for initial states and transitions but two classes of age for event
For A=2, one class of age for initial states, 2 classes of age for transitions but three classes of age for event
etc ..
There is always A+1 classes of age for event, one more than for the transitions .
Sincerely,
CHOQUET
 
Posts: 211
Joined: Thu Nov 24, 2005 4:58 am
Location: CEFE, Montpellier, FRANCE.

Re: More parameters than expected according to GEMACO senten

Postby simone77 » Mon Nov 21, 2011 6:38 am

E-Surge 1.8 user's manual pg.6 wrote:In multievent formulations, the first encounter may be an event (...). Therefore, E-SURGE always considers at least two age classes for encounters, allowing the first event probability (first class of age) to be modelled or not. Thus if one chooses a maximum age A=1, which implies no age effect, E-SURGE creates 2 classes for first and next encounter. If one sets A=2, E-SURGE creates 3 classes (with age) for events.

I also underlined that sentence but forgot it, I have just remembered after this last answer.
Very sorry for the silly question.

Simone
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 10 guests