Using E-Surge for individual covariates

[Guillaume Souchay]

Hi Rémi,

as I tell you 2 weeks ago, I've got a good model including a covariate. I'd like to show its result by a graphical figure.
In my model survival is computed :
For Initial State: IS - Step 1 - (0): to
For Transition: S - Step 1 - (20): f(1).a(1)+f(2,3,4).a(1)+f.a(3:6,7)+f(1,2,3,4).a(2).[i+xind]
For Transition: T - Step 2 - (49): f(1).to.ete.a(1)+f.to.a(2:7)+others
For Event: E - Step 1 - (97): firste+nexte(1:7).to.t

So I have an intercept and a slope for every location
As we use the logit link function, I need the slope and the intercept to show the impact on survival for every location using: logit(S)=µ+beta*BCI .
With beta do I have to use to do it?
Christophe Barbraud helped me and tell me I have to use beta from 42 to 45 for intercept and from 46 to 49 for slope. Is it good or not? How to find the good intercept and slope?

Thanks,

Guillaume

ps: if the output file of the model is needed, I can send it to you by mail

[CHOQUET]

>In my model survival is computed :
>For Initial State: IS - Step 1 - (0): to
>For Transition: S - Step 1 - (20): f(1).a(1)+f(2,3,4).a(1)+f.a(3:6,7)+f(1,2,3,4).a(2).[i+xind]
>For Transition: T - Step 2 - (49): f(1).to.ete.a(1)+f.to.a(2:7)+others
>For Event: E - Step 1 - (97): firste+nexte(1:7).to.t

>So I have an intercept and a slope for every location
>As we use the logit link function, I need the slope and the intercept to >show the impact on survival for every location using: logit(S)>=µ+beta*BCI .
>With beta do I have to use to do it?

In IVFV, you can get the coordinates of biological parameters to which
one beta is link. On additional think is that for interaction like a.b
then levels of a are described first.

> Christophe Barbraud helped me and tell me I have to use beta from 42 >to 45 for intercept and from 46 to 49 for slope. Is it good or not? How
> to find the good intercept and slope?
I think that intercepts are beta from 13 to 16 and slopes from 17 to 20.

Rémi

[Guillaume Souchay]

Hi Rémi,

thanks for your answer, with it, I totally understand why it was these parameters I had to use.
But, when I calculated each parameters for my birds, it shows that there is a difference between survival parameter between model with BCI and model without BCI.

survival estimation no BCI in model / BCI in model
france 0.6061786 / 0.9529166
esp 0.47240011 / 0.93123991
it 0.77565508 / 0.77635385
autre 0.86056931 / 0.98961168

For only one site, there is no difference.
I verified if the model with BCI was significantly better, it is, the AnoDev is also significant.

Could the difference be due to the way of Esurge to calculate estimators?
With Christophe, we are a little bit lost...
Thanks,
Guillaume

[CHOQUET]

In reply to:


> survival estimation no BCI in model / BCI in model
> france 0.6061786 / 0.9529166
> esp 0.47240011 / 0.93123991
> it 0.77565508 / 0.77635385
> autre 0.86056931 / 0.98961168

Are the values for BCI the mean among individuals ?

> Could the difference be due to the way of Esurge to calculate estimators?

No but sometimes, there is local minima.

Rémi

[Guillaume Souchay]

BCI are residuals from the regression tarsus*weight.
So for each individual, a BCI had been calculated.

Otherwise, the difference could be just the result of a local minima, but if there is local minima, is the model still the best or should I compute an other time the model to find if the parameters are the same?

Thanks a lot !!

Guillaume

[CHOQUET]

> BCI are residuals from the regression tarsus*weight.
> So for each individual, a BCI had been calculated.

My question was, are the BCI covariates standardized (mean=0, se=1) ?

> Otherwise, the difference could be just the result of a local minima, but if > there is local minima, is the model still the best or should I compute an
> other time the model to find if the parameters are the same?

There is an option to start from different values from E-SURGE.
It keep the parameters values link to the lower deviance.

Rémi

[Guillaume Souchay]

>My question was, are the BCI covariates standardized (mean=0, se=1) ?
We didn't standardized BCI covariates. We didn't transform the values resulting of the regression.


>There is an option to start from different values from E-SURGE.
It keep the parameters values link to the lower deviance.

To compute my model, I choose the option "multiple random" to start the model with different values and the model was computed 10 times before I had the final result.

[CHOQUET]

> We didn't standardized BCI covariates. We didn't transform the values
> resulting of the regression.

If the mean of the BCI covariates is not zero then the value for the survival
that you get by back-transforming b0 is not the mean value of phi with
logit(phi)=b0+b1*BCI

[Jocelyn]

Hi,
I loaded a file with individual covariates. ESURGE run correctly but it could not estimate any parameters. In the result, I have only the Beta!
Any suggestion?
Thanks a lot
jocelyn

Here is an extract of my data file in Mark format (the first histories)
1100000000000000 1 0 -1.60 ;
1111000000000000 0 1 -0.44 ;
1000000000000000 1 0 0.21 ;
1111000000000000 1 0 -0.88 ;
1110000000000000 1 0 -0.14 ;
1100000000000000 1 0 1.14 ;
1111000000000000 0 1 0.41 ;
1111101011110000 0 1 1.67 ;
1111000000000000 1 0 0.70 ;

Here are the matrices, the most simple:
%%%% Initial state %%%%%%
1
1 1 IS
*
%%%% Transition %%%%%%
1
2 2 T
y *
- *
%%%% Event %%%%%%
1
2 2 E
* b
* -

MDL:
T: i+xind
E: firste+nexte

[CHOQUET]

There is the manual and some complementary document from the menu of E-SURGE !!.
I don't really understand the question.
Rémi