Modeling Sex Differences

questions concerning anlysis/theory using program DENSITY and R package secr. Focus on spatially-explicit analysis.

Modeling Sex Differences

Postby adamdillon » Thu Jan 14, 2016 7:06 pm

Hi Murray,

I have been modeling sex differences a few different ways and the resulting estimated densities vary quite a bit depending on which method I’m using, despite the data being the exact same. I’m not exactly sure why this is and I’m hoping you could clarify it for me.
My data consists of a single fox trapping season with 6 occasions and I have modeled sex differences in 3 different ways.

1: Sex as different sessions
Data & Model Example:
#Session FoxID Occasion Trap-ID
F F118 4 MG13-6

Capthist <- read.capthist(captfile="Capture_File.txt", trapfile="Detection_File.txt", detector = "multi", fmt="trapID", noccasions=6)
Mask <- make.mask(traps(Capthist), spacing = 125, session=c(1,1), buffer = 400, type = "trapbuffer")
D.1.g.1.s.sex <- secr.fit(Capthist, model = list(D ~ 1, g0 ~ 1, sigma ~ session), trace = TRUE, mask=Mask, detectfn=0)

2. Sex as individual covariate
Data & Model Example:
#Session FoxID Occasion Trap-ID Sex
1 F118 4 MG13-6 F

Capthist <- read.capthist(captfile="Capture_File.txt", trapfile="Detection_File.txt", detector = "multi", fmt="trapID", noccasions=6, covnames = "sex")
Mask <- make.mask(traps(Capthist), spacing = 125, session=c(1), buffer = 400, type = "trapbuffer")
D.1.g.1.s.sex <- secr.fit(Capthist, model = list(D ~ 1, g0 ~ 1, sigma ~ sex), trace = TRUE, mask=Mask, detectfn=0, CL=TRUE)

3. Sex as hybrid mixture model
Data & Model Example:
#Session FoxID Occasion Trap-ID Sex
1 F118 4 MG13-6 F

Capthist <- read.capthist(captfile="Capture_File.txt", trapfile="Detection_File.txt", detector = "multi", fmt="trapID", noccasions=6)
Mask <- make.mask(traps(Capthist), spacing = 125, session=c(1), buffer = 400, type = "trapbuffer")
D.1.g.1.s.sex <- secr.fit(Capthist, model = list(D ~ 1, g0 ~ 1, sigma ~ h2), trace = TRUE, mask=Mask, detectfn=0)

I have the exact same model set for all 3 methods and these are the results I get:

Method 1: Sex As Session
Model AIC cwt Density
D.1.g.sex.s.B Model 10 .20 .02415907 (*2) = .04831814
D.1.g.1.s.B Model 4 .14 .02377003 (*2) = .04754006
D.1.g.sex.B.s.B Model 27 .14 .0239219 (*2) = .0478438
D.1.g.B.s.B Model 19 .11 .02351786 (*2) = .04703572

Method 2: Sex As Ind Cov
Model AIC cwt Density
D.1.g.1.s.B Model 4 .20 .0475402
D.1.g.B.s.B Model 19 .16 . .04703572
D.1.g.sex.s.B Model 10 .16 .04780902
D.1.g.sex.B.s.B Model 27 .11 .0473202

Method 3: Sex As H Cov
Model AIC cwt Density
D.1.g.sex.B.s.sex Model 26 .18 .05065689
D.1.g.B.s.sex.B Model 20 .18 .09050037
D.1.g.sex.s.B Model 10 .17 .05538701
D.1.g.1.s.sex.B Model 6 .17 .09478066

First of all when I model sex as a session it estimates density for each session correct? Since I have no variation (sex) on density the estimated density of males and females is equal. But in order to actually estimate the density of foxes (both male and females) I need to multiple the estimate times 2 correct?
When comparing the estimates from Method 1 and Method 2, the different models had different AICc weight and therefore different ranks. Do you know why this is? I would expect the models to have the same weights given they are identical model sets and the data is the exact same. Although the estimates for a single model (i.e. Model 10) was similar across these two methods.

When I model sex as a mixture models my AIC table is vastly different. As in methods 1 and 2, 4 models had AICc weight > .1, yet there was only 1 model (Model 10) in method 3 that was present in the first 2 methods. Instead there were 3 other models that showed substantial weight in method 3 than had very little support in method 1. Additionally when looking at the actual estimated densities, they vary by almost 2 fold. What is it about these three methods of modeling sex that would account for different AICc weights and different estimates of density, given the same model sets and same data?

Also when modeling sex as an individual covariate, I must use the Conditional Likelihood, but is there a way to model average with the CL? I can determine a model average for both method 1 and method 3 but if I try and save the derived estimates of each model separately and model average it gives me an error and tells me that the objects must be of class ‘secr’ or’secrlist’. If I model averaging across all models and then trying to derive the density estimates it gives me an error and tells me that derived requires a fitted secr model.

And lastly as I mentioned before in method 1 the print out of the real estimates shows a density for session 1 (females) and a density for session 2 (male) and these two densities are equal because I don’t have any variation on density (D.1). But in order to estimate the density of foxes (males and females) I need to add these two estimates together, right? But then in method 3 the print out of the real estimates from the mixture model show a density for h2=1 and h2=2. Again the density estimates are equal because I have no variation on density (D.1) but in this case the estimated density for each mixture is actually the density of total foxes? Or do I need to add these two densities together in order to determine total fox density (males and females). If I need to add them together than these estimates are way higher than those determined in method 1 and 2.

I’m just looking for some clarification and guidance before I move forward on this. Thanks a lot.
Adam
adamdillon
 
Posts: 21
Joined: Tue Mar 03, 2015 2:11 pm

Re: Modeling Sex Differences

Postby murray.efford » Fri Jan 15, 2016 12:01 am

Your code does not specify a hybrid mixture model - for that specify hcov.
CL models cannot be 'identical' to full likelihood models. 'secr' does not offer model averaging for CL models - you're on your own there, but it shouldn't be hard. Why D~1 in your CL model? Why a session argument in make.mask? The rest I think you can work out for yourself - the answer to most of your questions is Yes. I'm not about to write an essay.
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand


Return to analysis help

Who is online

Users browsing this forum: No registered users and 12 guests

cron