Estimates with linearComb

266 views
Skip to first unread message

Aude Corbani

unread,
Jun 11, 2012, 2:53:17 PM6/11/12
to unmarked
Hi all,
Imagine a model (p~covariate1, psi~covariate1) using the occu
function.
Now, you want an estimate of psi and p for the model fitted.
What estimates the linearComb function? Is it correct to use the model
coefficients to make estimations for the model fitted?
Thanks,
Aude

Aude CORBANI
PhD Student
Faculté de foresterie, de géographie et de géomatique
Pavillon Abitibi-Price
2405, rue de la Terrasse
Université Laval
Québec (Québec) G1V 0A6
CANADA

Kery Marc

unread,
Jun 11, 2012, 4:06:36 PM6/11/12
to unma...@googlegroups.com
Dear Aude,

you can get estimates from the model by hand or use the functionality of unmarked.

(a) You can type plogis(X) for a logit-scale parameter or prediction X (for occupancy or detection) to backtransform it from the logit to the probability scale. This will however not give you a SE or confidence interval
(b) For a "Null model", with just an intercept in the state and the detection model (i.e., for the model you imagine), you can use backTransform(), which will also yield a SE.
(c) For models with covariates, use predict, which will also yield SEs and 95% CIs.

Check the vignettes that come with unmarked. There is information on this in "Overview of unmarked" and also "Dynamic occupancy models in unmarked". Both are relevant for your occupancy model.

Regards -- Marc

______________________________________________________________

Marc Kéry
Tel. ++41 41 462 97 93
marc...@vogelwarte.ch
www.vogelwarte.ch

Swiss Ornithological Institute | Seerose 1 | CH-6204 Sempach | Switzerland
______________________________________________________________

*** Introductory book on Bayesian statistical modeling: Kéry (2010), Introduction to WinBUGS for Ecologists, Academic Press; see www.mbr-pwrc.usgs.gov/pubanalysis/kerybook
*** New book on Bayesian statistical modeling: Kéry & Schaub (2012), Bayesian Population Analysis using WinBUGS, Academic Press; see www.vogelwarte.ch/bpa


________________________________________
From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Aude Corbani [audec...@yahoo.fr]
Sent: 11 June 2012 20:53
To: unmarked
Subject: [unmarked] Estimates with linearComb

Aude Corbani

unread,
Jun 12, 2012, 9:51:31 AM6/12/12
to unmarked
Hi Marc,

Thank you for your answer. I think I'll use your "c" option because my
best model contains one covariate for each parameter (psi and p). I
already use the function predict for my dataset and I calculate the
mean value for psi and p as estimates (is this correct?), but I want
to be sure it is the right way to obtain my estimates.

Aude

Aude Corbani
PhD Student
> Faculté de foresterie, de géographie et de géomatique
> Pavillon Abitibi-Price
> 2405, rue de la Terrasse
> Université Laval
> Québec (Québec) G1V 0A6
> CANADA

On 11 juin, 16:06, Kery Marc <marc.k...@vogelwarte.ch> wrote:
> Dear Aude,
>
> you can get estimates from the model by hand or use the functionality of unmarked.
>
> (a) You can type plogis(X) for a logit-scale parameter or prediction X (for occupancy or detection) to backtransform it from the logit to the probability scale. This will however not give you a SE or confidence interval
> (b) For a "Null model", with just an intercept in the state and the detection model (i.e., for the model you imagine), you can use backTransform(), which will also yield a SE.
> (c) For models with covariates, use predict, which will also yield SEs and 95% CIs.
>
> Check the vignettes that come with unmarked. There is information on this in "Overview of unmarked" and also "Dynamic occupancy models in unmarked". Both are relevant for your occupancy model.
>
> Regards  --  Marc
>
> ______________________________________________________________
>
> Marc Kéry
> Tel. ++41 41 462 97 93
> marc.k...@vogelwarte.chwww.vogelwarte.ch
>
> Swiss Ornithological Institute | Seerose 1 | CH-6204 Sempach | Switzerland
> ______________________________________________________________
>
> *** Introductory book on Bayesian statistical modeling: Kéry (2010), Introduction to WinBUGS for Ecologists, Academic Press; seewww.mbr-pwrc.usgs.gov/pubanalysis/kerybook
> *** New book on Bayesian statistical modeling: Kéry & Schaub (2012), Bayesian Population Analysis using WinBUGS, Academic Press; seewww.vogelwarte.ch/bpa
>
> ________________________________________
> From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Aude Corbani [audecorb...@yahoo.fr]

Kery Marc

unread,
Jun 12, 2012, 10:04:14 AM6/12/12
to unma...@googlegroups.com
Dear Aude,

I am not sure I understand what you mean by "calculate the mean value for psi and p as estimates" ?

Regards --- Marc


-----Ursprüngliche Nachricht-----
Von: unma...@googlegroups.com [mailto:unma...@googlegroups.com] Im Auftrag von Aude Corbani
Gesendet: Dienstag, 12. Juni 2012 15:52
An: unmarked
Betreff: [unmarked] Re: Estimates with linearComb

Aude Corbani

unread,
Jun 12, 2012, 10:31:05 AM6/12/12
to unmarked
I use the function predict to obtain an estimate of psi and p for each
"line" of my data set. Then I obtain as many values as lines. Finally,
I calculate a mean of all values and I obtain one value for psi and
one value for p.

Regards,

Aude

Aude Corbani

unread,
Jun 12, 2012, 10:36:01 AM6/12/12
to unmarked
Just an example to be clear:

psi se
Site1 0.35 0.02
Site2 0.27 0.05
Site3 0.33 0.03

General estimate: psi = 0.32 se = 0.03

Aude

Kery Marc

unread,
Jun 12, 2012, 10:56:32 AM6/12/12
to unma...@googlegroups.com
Dear Aude,

if you want to characterise the average occupancy probability and the average detection probability, then this approach is fine, what concerns the mean. However, for the standard error you would have to use the delta rule: you can't simply average the standard errors !

An alternative would be to fit the mean-centered covariates, in which case the intercept would represent the logit-scale occupancy and detection values for a typical covariate value, namely the mean of the observed values. You could then directly backtransform that to get both the estimate at the probability scale and its uncertainty (SE).

Centering covariates is very useful for interpretation of regression relationships, because it makes the intercept directly interpretable, something it is usually not, when it means the value of the response at a value of x=0.

Regards --- Marc



-----Ursprüngliche Nachricht-----
Von: unma...@googlegroups.com [mailto:unma...@googlegroups.com] Im Auftrag von Aude Corbani
Gesendet: Dienstag, 12. Juni 2012 16:36

Aude Corbani

unread,
Jun 12, 2012, 11:16:54 AM6/12/12
to unmarked
I already centered my covariate so I just have to use the function
backtransform with the intercept to obtain my estimate.

Thank you very much for your advices. It helps me a lot.

Regards,

Aude

Rylee

unread,
Nov 27, 2012, 3:30:26 PM11/27/12
to unma...@googlegroups.com
Would the delta rule in this case just involve calculating variance from the values of psi from each site, then, subsequently, the standard deviation and standard error?

Rylee
Reply all
Reply to author
Forward
0 new messages