Model Averaging in unmarked

[Alejandro E]

Hi all,

I have a question, it may be very easy and basic, so, sorry about that.

I know that there are three diferents model averaging:

1. Model averaging suggested by Burnham and Anderson (2002) or

2. Natural averaging mencionated by Symmonds and Moussalli (2011) or 

3. Full-model averaging mencionated by Symmonds and Moussalli (2011)

So, my question would be: What kind of model averaging applies unmarked? 

Thanks for your answers, I really appreciate your help.

Burnham and Anderson. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach

Symmonds and Moussalli. 2011.A brief guide to model selection, multimodel inference and model averaging in behavioural  ecology using Akaike’s information criterion

[Mazerolle, Marc J.]

Hi Alejandro,

in the package AICcmodavg, modavg( ) implements natural averaging of beta estimates as discussed in Burnham and Anderson (2002), modavgShrink( ) implements the shrinkage estimator on beta estimates (also discussed in Burnham and Anderson 2002, p. 152; and Luckacs et al. 2010: Model selection bias and Freedman's paradox. Annals of the Institute of Statistical Mathematics 62:117-125). You can also model-average predictions or effect sizes (differences between specific groups) with modavgPred( ) and modavgEffect( ) - both implement the shrinkage estimator.

The natural average estimator is known to be biased, especially when the variable appears in models that do not have a lot of support (see Cade, B. S. 2015. Model averaging and muddled multimodel inferences. Ecology 96:2370-2382). 

Hope it helps,

Marc

--
____________________________________
Marc J. Mazerolle
Département des sciences du bois et de la forêt
2405 rue de la Terrasse
Université Laval
Québec, Québec G1V 0A6, Canada
Tel: (418) 656-2131 ext. 7120
Email: marc.ma...@uqat.ca

---

De : unma...@googlegroups.com [unma...@googlegroups.com] de la part de Alejandro E [biof...@gmail.com]
Envoyé : mercredi 31 août 2016 21:28
À : unmarked
Objet : [unmarked] Model Averaging in unmarked

--
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Richard Chandler]

Hi Alejandro,

I think the last two options you refer to involve averaging coefficients, which (intentionally) isn't possible in unmarked. In unmarked, you can model average predictions using the predict function applied to objects created by fitList(). Other options are possible with the AICcmodavg package (and I just noticed Marc's email with some details).

Richard

--
You received this message because you are subscribed to the Google Groups "unmarked" group.

To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Richard Chandler

Assistant Professor

Warnell School of Forestry and Natural Resources

University of Georgia

706-542-5815

Lab website

[Alejandro E]

Hi Marc and Richard,

Thanks for your responses, these were excellent to my question.

Greetings!

Alejandro

[James Shelley]

Hi,

I have another questions regarding model averaging predictions I was hoping you could help me with.

I would like to get average predictions for the whole model, but I can only seem to get them for each site...

The code I used is as follows...

##create fitlist 

ModelListTup <- fitList('psi(.)gam(.)eps(.)p(.)'    = fm0tup,
                  'model 1'    = fm1tup,
                  'model 2'    = fm2tup,
                  'model 3'    = fm3tup)

##create model list

Cands <- list(fm0tup, fm1tup, fm2tup, fm3tup)
Modnames <- c("1", "2", "3",
              "4")

##Do model averaging

Tup_occu_modavg_psi_predict <- modavgPred(Cands,
                                      parm.type = "psi", 
                                      newdata = Tupong_unmarkedMultFrame@siteCovs)[c("mod.avg.pred",
                                                                                     "lower.CL",
                                                                                     "upper.CL")]

## Put predictions, CI, and all site covariates into one data frame
Tup_occu_modavg_psi_predict_df <- data.frame(Predicted = Tup_occu_modavg_psi_predict$mod.avg.pred,
                                         lower = Tup_occu_modavg_psi_predict$lower.CL,
                                         upper = Tup_occu_modavg_psi_predict$upper.CL,
                                         site_cov)

The output looks like this (below), but I'm after a global prediction.

Can you please provide some help.

Thanks!

On Friday, September 2, 2016 at 11:26:28 AM UTC+10 biof...@gmail.com wrote:

Hi Marc and Richard,

Thanks for your responses, these were excellent to my question.

Greetings!

Alejandro

2016-09-01 7:54 GMT-05:00 Richard Chandler <rcha...@warnell.uga.edu>:

Hi Alejandro,

I think the last two options you refer to involve averaging coefficients, which (intentionally) isn't possible in unmarked. In unmarked, you can model average predictions using the predict function applied to objects created by fitList(). Other options are possible with the AICcmodavg package (and I just noticed Marc's email with some details).

Richard

On Wed, Aug 31, 2016 at 9:28 PM, Alejandro E <biof...@gmail.com> wrote:

Hi all,

I have a question, it may be very easy and basic, so, sorry about that.

I know that there are three diferents model averaging:

1. Model averaging suggested by Burnham and Anderson (2002) or

2. Natural averaging mencionated by Symmonds and Moussalli (2011) or 

3. Full-model averaging mencionated by Symmonds and Moussalli (2011)

So, my question would be: What kind of model averaging applies unmarked? 

Thanks for your answers, I really appreciate your help.

Burnham and Anderson. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach

Symmonds and Moussalli. 2011.A brief guide to model selection, multimodel inference and model averaging in behavioural  ecology using Akaike’s information criterion

--
You received this message because you are subscribed to the Google Groups "unmarked" group.

To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Richard Chandler

Assistant Professor

Warnell School of Forestry and Natural Resources

University of Georgia

706-542-5815

Lab website

--
You received this message because you are subscribed to the Google Groups "unmarked" group.

To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.

[James Shelley]

Sorry, this is the correct image of the output for psi predictions.

[Marc J. Mazerolle]

Hi James,

Not sure what you mean by getting "average predictions for the whole
model". If you want to get model-averaged predictions for an average
site, one way to go about it would be to supply values in the newdata
data frame in modavgPred( ) that correspond to average values for a
site.

If you want to average the model-averaged predictions across all sites,
you could potentially use a bootstrap approach to get confidence
intervals around the average, using a few loops.

Best,

Marc
--
____________________________________
Marc J. Mazerolle

Professeur agrégé et directeur du bac. en environnements naturels et
aménagés

Département des sciences du bois et de la forêt
2405 rue de la Terrasse
Université Laval
Québec, Québec G1V 0A6, Canada

Tel: (418) 656-2131 ext. 407120
Email: marc.ma...@sbf.ulaval.ca

-------- Message initial --------
De: James Shelley <jamessh...@gmail.com>
Répondre à: unma...@googlegroups.com
À: unmarked <unma...@googlegroups.com>
Objet: Re: [unmarked] Model Averaging in unmarked
Date: Thu, 01 Feb 2024 18:25:29 -0800

[James Shelley]

Thanks for getting back to me Marc.

I'd like to produce a table like this one (below)  Dynamic occupancy modelling to determine the status of a Critically Endangered lizard | Oryx | Cambridge Core. 

They don't provide details on how it was done, but it looks like they averaged the predictions across the three best models. 

Does that help?

[Marc J. Mazerolle]

Hi James,

The type of presentation in the table you included is unusual and could
be misleading, as it is unclear what the authors did to obtain a single
value of psi, gamma, epsilon, and p. If certain variables are important
predictors of some of these parameters, then I'm not convinced that
producing an average value for psi, gamma, epsilon or p is a good
idea. 

My suggestion is to present the results (model-averaged predictions) in
terms of the variables for which the parameters varied and illustrate
these in plots vs the variables of interest. This is a more
straightforward way to present and interpret the results in my opinion.

Best,

Marc
--
____________________________________
Marc J. Mazerolle
Professeur agrégé et directeur du bac. en environnements naturels et
aménagés
Département des sciences du bois et de la forêt
2405 rue de la Terrasse
Université Laval
Québec, Québec G1V 0A6, Canada
Tel: (418) 656-2131 ext. 407120
Email: marc.ma...@sbf.ulaval.ca

-------- Message initial --------
De: James Shelley <jamessh...@gmail.com>
Répondre à: unma...@googlegroups.com
À: unmarked <unma...@googlegroups.com>
Objet: Re: [unmarked] Model Averaging in unmarked

Date: Thu, 01 Feb 2024 21:48:04 -0800

Thanks for getting back to me Marc.
I'd like to produce a table like this one (below) Dynamic occupancy
modelling to determine the status of a Critically Endangered lizard |
Oryx | Cambridge Core. 
They don't provide details on how it was done, but it looks like they
averaged the predictions across the three best models. 

Does that help?