Effect of p on psi estimation

[Adrián Naveda-Rodríguez]

Hola,

I am trying to make raster of occupancy models. 

I ran the model in R and made the model spatially explicit in R using package Raster and using function predict(). I got a raster for the model occu(~tree~temp,umf) and this raster has values from 0.19 to 0.59.

 

I try to make the raster in ArcMap using occu(~1~temp,umf), and I got a raster with values from 0.38 to 0.78. I expected a raster with same values. But using raster calculator in arcmap I am not taking in account detection probability.

 

Do anybody have any idea on how detection probability influence psi estimation?

All the best,

Adrian

[Jim Baldwin]

It seems that your question is really "How does modeling the detection probability influence psi estimation?"

You are fitting one model with a constant detection probability (occu(~1~temp,umf)) and another with the detection probability being a function of the variable tree (occu(~tree~temp,umf)).  (The set of predictions using ArcMap and the set of predictions using the R package raster seem to have nothing to do with the question.  And the second model does take into account the detection probability: it's just that the detection probability is modeled as constant among visits.)

Have I totally misunderstood your question?

Jim

--
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.

[Adrián Naveda-Rodríguez]

Hola Jim,

Thanks for your input. My question is if psi estimation is afected by p.

If I ran a model using p covariate and psi covariate and I got a result.

After this, I ran a model without p covariate and the same psi covariate and I got a different result. I would expect the same result.

In distance sampling, if you have a low detection probability, you will get high abundance estimate and viceverse. I wonder if something similar occur in occupancy.

For me, making rasters of occupancy models in unmarked take five hours, whereas in ArcMap you can get the same product in a few seconds. For this reason I want to know how p affect psi in occu in order to include this in the equation in ArcMap.

All the best,

Adrian

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

[Kery Marc]

Dear Adrian,

I am not sure I find this confusing: in such hierarchical models, estimates of p will affect the estimates of abundance, or occupancy, or density. If you change something in the detection model, then p will (usually) be estimated at different values, so naturally, you will get different estimates of the biological state as well. In general, with any model, you can hardly ever expect exactly the same estimates from two models with different structure (e.g., different covariates).

Best wishes  ----  Marc

---

From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Adrián Naveda-Rodríguez [adrian...@gmail.com]
Sent: 12 May 2017 17:05
To: unmarked
Subject: Re: [unmarked] Effect of p on psi estimation

[Jim Baldwin]

Dear Marc:  As you may know from my responses in this forum, I'm easily confused.

Dear Adrian:  You say "But using raster calculator in arcmap I am not taking in account detection probability" but your second model does account for a constant detection probability.  In addition, for predictions of psi on a map only the function for psi is used.  Same for predictions using the raster package.  (Those statements and my general state of mind are the sources of my confusion.)

Jim

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.

--
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.

[Adrián Naveda-Rodríguez]

Dear Jim and Marc,

I apology for not explaining well my self.

My question is bacause I want to produce psi raster in ArcMap and not in R. 

If I run the model occu(~tree~temp,umf) I get a result =0.5

If I run the modelo occu(~1~temp,umf), I get a result = 0.4

It is clear that tree is having an effect on psi. I want to know what is that effect. I want to know what does tree is doing to psi estimate? maybe multiplying, dividing? 

There is a reason to modelate p before fitting a model to psi. what is that reason and how if works in the occu function.

Thanks,

Adrian

[Jim Baldwin]

I see two distinct issues here.

(1) Without knowing the meaning of "tree" (and why it should be related to detection) I would simply look at the AIC values.  If the model with "tree" has a much smaller AIC value, then you should probably drop the model that doesn't use tree.  (You still should look at some goodness-of-fit statistics to see how reasonable and consistent the model is with tree.)

(2) Whether you use ArcMap or the raster package in R, you're only using the psi part of the model for the predictions.  So either model can be used in both ArcMap and raster.

Jim

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

[Kery Marc]

Dear Adrian,

 

To be honest, I feel that you have a need to get your homework done before you ask further questions on this list. It may be that I don’t fully understand your problem, and if this is the case, I apologize, but if you are asking what I believe you do then clearly, you don’t grasp the meaning of the occupancy part of an occupancy model as a simple regression model. You have something there that looks like this:

 

logit(psi) = alpha + beta * tree.

 

unmarked yields estimates of the parameters alpha and beta. The magnitude and sign of beta give you a first indication of how the relationship between tree and psi looks like. And then,  you can combine alpha and beta with a range of values of the covariate tree to produce predictions of what one expects psi to be when one varies tree.

 

However, this is a basic problem about regression and not something that is particular about occupancy models. You can read up on this in any introductory stats book. For the specific case of a regression model as part of an occupancy or other hierarchical model, we have lots of examples of how to form predictions in our AHM1 book.

 

Why p should be modeled before psi ? Well, p is usually regarded as less interesting (a “nuisance parameter”) and in models such as occupancy models (but also Cormack-Jolly-Seber survival models), people usually feel that they want to get the nuisance done with before going to the (usually) biologically more interesting parameters, such as occupancy or survival.

 

Best regards    --- Marc

[Benedikt Schmidt]

Dear Adrian,

You wrote “I want to know what is that effect.” I think you have so far only compared AIC values of different models.

Now you need to have a look at the regression equation for occupancy (see Marc’s email). This is simply

summary(your.model.name)

Best,

Beni

 

 

 

 

Gesendet: Dienstag, 16. Mai 2017 um 09:06 Uhr
Von: "Kery Marc" <marc...@vogelwarte.ch>
An: "unma...@googlegroups.com" <unma...@googlegroups.com>
Betreff: AW: [unmarked] Effect of p on psi estimation