MARK vs MSurge

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

MARK vs MSurge

Postby Mark Trinder » Wed Mar 08, 2006 1:14 pm

Hi

can anyone help me with what is either a MARK vs MSurge estimation difference, or (more likely) my inability to write GEMACO coding.

I would like to use MSurge for my current analysis as its coding flexibility is appealling since I may get into quite complicated models as I explore the data more fully, however I find the model definition language quite a challenge. I though I'd worked out how to code for a simple age structure, but when I compared the output to that from MARK, where I am more sure of the model structure, I had (noticeably) different parameter estimates. Here are the details:

Both sexes of birds have been first ringed in 3 age classes (juv, yearling, adult), giving 6 groups. I have been trying to fit a constant age-related survival model, where juveniles have 3 age classes (say parameters 1,2,3), yearlings have 2 (paras 2,3) and adults one (para 3). So as juvs and yearlings get older they contribute to the appropriate age survival estimates. In MARK this is straightforward to achieve using the PIMs and I am confident that part is right. I thought I had it right in MSurge too, using the following coding, adapted from the MSurge manual (pg15):

for males
a(1).[g(1)] + a(2).[g(1)&g(2)] + a(3).[g(1)&g(2)&g(3)]

and the same for females but substitute g(4-6) for g(1-3)

but I get different results:

MARK
0.914
1.0
0.89
1.0
0.68
0.82

MSurge
0.92
0.99
0.87
0.94
0.88
0.84

while most of these are similar, p4 and p5 are quite different. When I run simpler constant rate models I get more or less identical results, which suggests its my coding which is at fault, but I have tried all the combinations I can thnk of and have run out of ideas.

all help gratefully received,

Mark
Mark Trinder
 
Posts: 17
Joined: Tue Oct 28, 2003 7:42 am
Location: Slimbridge, UK

Re: MARK vs MSurge

Postby egc » Wed Mar 08, 2006 5:50 pm

Mark Trinder wrote:Hi

can anyone help me with what is either a MARK vs MSurge estimation difference, or (more likely) my inability to write GEMACO coding.
...
Mark


Since most of this is an M-SURGE question, I'm moving it to that part of the forum.
egc
Site Admin
 
Posts: 201
Joined: Thu May 15, 2003 3:25 pm

Model with real age,

Postby CHOQUET » Thu Mar 09, 2006 5:39 am

Hello,

I think that differences with estimates between M-SURGE and MARK
is not related to programs but to the sentence of the model with GEMACO. A correct way to defined a model with the real age is to defined each classes of real age. Let age0, age1, age2 be

age0 : individuals of age 0 to 1
age1 : individuals of age 1 to 2
age2 : individuals of age >2

In Gemaco with only 3 age classes,

age0 is [a(1).g(1)]
age1 is [a(2).g(1)&a(1).g(2)]
age2 is [a(3).g(1)&a(2).g(2)&g(3)]

So the model depending on three classes of the real age is simply:

age0+age1+age2

From this simple model, you can do more complex model easily as for example time dependence only on adult with the sentence :

age0+age1+age2.t

Sincerely,

Rémi
CHOQUET
 
Posts: 211
Joined: Thu Nov 24, 2005 4:58 am
Location: CEFE, Montpellier, FRANCE.

Postby Mark Trinder » Thu Mar 09, 2006 7:14 am

Thanks Remi

it wasn't quite the right answer, but it got me close enough to work out the rest. One thing I had to do was not reduce the age classes to 3, but leave it at the orginal 30 (31 occasions), and then the final coding to achieve the same answers as those from MARK is:

age0 [a(1).g(1)]
age1 [a(2).g(1)&a(1).g(2)]
age2 [a(3:30).g(1)&a(2:30).g(2)&a(1:30).g(3)]

Which leads me to ask why does changing the number of age classes have this effect?

cheers
Mark
Mark Trinder
 
Posts: 17
Joined: Tue Oct 28, 2003 7:42 am
Location: Slimbridge, UK

Influence of the number of age classes in M-SURGE

Postby CHOQUET » Fri Mar 10, 2006 5:08 am

Mark,

M-SURGE enables the user to chosse the number of age classes to any value from 1(time dependent) to the number of occasion minus one(full age and time dependent).

What happened in our case ?
reducing the number of age classes to 3, brings together age-classes 3 to 30. In this case a(3) is the same than a(3:30) in Gemaco. However when the number of age classes is kept to 30 then a(3) is not equal to a(3:30).


In our example, the model can be written by sentence age0+age1+age2

with 30 age classes, the shorcuts(age0,age1,age2) are defined by

age0 [a(1).g(1)]
age1 [a(2).g(1)&a(1).g(2)]
age2 [a(3:30).g(1)&a(2:30).g(2)&g(3)]

with 3 age classes, the above shorcuts are still valid but age2 can be rewritten

age2 [a(3).g(1)&a(2:3).g(2)&g(3)]

in a less general form.


So, why reducing the number of age classes to the minimum ?.
Reducing the number of age classes leads to faster convergence(with a new set of minimal statistics) and reduces the amount of memory(with a well adapted data structure). So, with this unique feature, M-SURGE
enable largest model to be fit.



Rémi
CHOQUET
 
Posts: 211
Joined: Thu Nov 24, 2005 4:58 am
Location: CEFE, Montpellier, FRANCE.


Return to analysis help

Who is online

Users browsing this forum: No registered users and 20 guests