linear/quadratic effect of age

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

linear/quadratic effect of age

Postby Sébastien » Mon May 01, 2006 10:32 am

Hi,
I am using M-Surge for testing hypotheses about costs of reproduction (with 2 states: breeder-non breeder). I am trying to test for a linear or quadratic effect of age on transition and survival probabilities. I am using a file of external covariates defined as:

2
14 14
1 2 3 ...14
1 4 9 ... 196

With such a file, my results are pretty surprising: for example, the deviance of model Phi(t.from) Psi(t.from.a*x(1)) p(i) is higher than deviance of model Phi(t.from) Psi(t.from) p(i), and the number of identifiable parameters is the same?

When I use an additive effect of age, it seems OK: the model Phi(t.from) Psi(t.from+a*x(1)) p(i) has a lower deviance (and 2 more parameters) than model Phi(t.from) Psi(t.from) p(i).

So, can anyone explain me these surprising results and, if I am wrong (which is probably the case!) tell me how to code for a linear (and quadratic) effect of age with GEMACO?

Thanks a lot,
Sebastien
Sébastien
 
Posts: 3
Joined: Fri Apr 28, 2006 2:17 pm
Location: UQAR

Postby CHOQUET » Wed May 10, 2006 8:13 am

Hello,

Several thinks must be said for the use of covariates:

First, covariates must be normalized:
a) to be of mean 0 to test for positif of negatif slope,
b) to be of variance 1 for numerical reasons.

Secondly, starting from the model where covariates
are removed is a good starting point to avoid local minima.

Thirdly, user must take care about the use of covariates inside
transition with more than 3 states because multinomial logit
might perturb constraints defined by the covariates.
In this case, the identity link is recommended.

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

Re: linear/quadratic effect of age

Postby lucette » Tue Feb 07, 2012 4:38 am

Hi,
I'm a new user of M-SURGE.

I'm doing analyses on Breeder/ not Breeder, and I would like to input an external covariate with the number of Breeder according to the Age. This is closer from a guassian than a quadratic evolution.
So first i would like to know if it"s possible to work with a gaussian external covariate on M-SURGE.

Secondly I would like more precision about how write the matrix of the covariate. I have the Age (from2 to 20) and the proportion of breed associated. So does my matrix must look like this:

2
2 3 4 .... 20
0.01 0.05 O.15...0.01


????

and do I must put this in MARK format to (.INP)

Thank you very much for your help and attention.

Cheers
lucette
 
Posts: 5
Joined: Mon May 23, 2011 6:04 am

Re: linear/quadratic effect of age

Postby CHOQUET » Tue Feb 07, 2012 10:55 am

Hello,

In M-Surge (the same in E-SURGE), create a text file with the number of Breeder (Ni) according to the age as:

1
19
N2 .... N20

In gemaco, define this text-file as the external covariates file and give the following sentence for survival for example:

i+a(2_20)*x

It will create the correct design matrix.
sincerely,
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 15 guests

cron