CJS Modeling in MRA'

This vignette repeats and embellishes examples in help(F.cjs.estim).

Time-varying Models

The following demonstrates two methods for fitting a time-varying capture and survival model, the so-called "small t" model. First, we attach the mra library and obtain access to the example dipper data.


Method 1: Using factors

The following code constructs a factor variable containing one level for each capture occasion. The attribute of this factor tells mra the "other"" dimension of the problem. Internally, mra will use this attribute to replicate the factor into matrices that are the appropriate size.

ct <- as.factor( paste("T",1:ncol(dipper.histories), sep=""))

Next, call F.cjs.estim and specify that ct is a time-varying vector covariate using the tvar function. When the vector given to tvar is a factor, there are additional options which allow the user to drop certain levels of the factor. This is useful when coefficients for some levels are not estimable, as in the case of a completely time-varying CJS model. Here, there are 7 levels in factor ct, but only 6 capture and survival parameters are defined (recall, 1st capture parameter is not estimable, and 7th survival parameter between occasions 7 and 8 does not exist). Consequently, we tell tvar to drop the first two levels of ct from the capture model. We drop the level 1st because only 6 parameters exist. We drop the 2nd to break the colinearity of levels and define $p_2$ as the reference level. In the survival model, we drop the first, sixth, and seventh levels of ct. The first level is dropped to break the colinearity of levels and define $\phi_1$ as the reference level. We drop level 6 because the last survival and capture parameters are confounded in CJS models. We drop the 7th level because there are only 6 survival parameters. The call the F.cjs.estim is:

dipper1.cjs <- F.cjs.estim( ~tvar(ct,drop=c(1,2)), ~tvar(ct,drop=c(1,6,7)), dipper.histories )

Method 2: Using explicit 2-D matricies

While using factors (Method 1 above) produces the most economical code, it does not adequately illuminate the covariate matrices which are at the heart of CJS modeling. To illustrate covariates as explicit matricies, this method constructs one 2-D matrix for each paramter, then estimates the same model as Method 1.

First, we construct 6 matricies containing 1's in a single column only. In Method 1, this construction was performed behind-the-scenes by tvar. Note that only 6 matricies are required due to the number of parameters, breaking of colinearity, and confounding of CJS paramters mentioned above.

x2 <- matrix(c(0,1,0,0,0,0,0), nrow(dipper.histories), ncol(dipper.histories), byrow=TRUE)
x3 <- matrix(c(0,0,1,0,0,0,0), nrow(dipper.histories), ncol(dipper.histories), byrow=TRUE)
x4 <- matrix(c(0,0,0,1,0,0,0), nrow(dipper.histories), ncol(dipper.histories), byrow=TRUE)
x5 <- matrix(c(0,0,0,0,1,0,0), nrow(dipper.histories), ncol(dipper.histories), byrow=TRUE)
x6 <- matrix(c(0,0,0,0,0,1,0), nrow(dipper.histories), ncol(dipper.histories), byrow=TRUE)
x7 <- matrix(c(0,0,0,0,0,0,1), nrow(dipper.histories), ncol(dipper.histories), byrow=TRUE)

Each of the above matrices have a column of 1's corresponding to the effect they estimate. The first six rows of x3 and x4 are:


We now call F.cjs.extim without aid of tvar by explicitely specifying the matricies in each model. Note that x2 is not included in the capture model, and that x6 and x7 are not included in the survival model.

dipper2.cjs <- F.cjs.estim( ~x3+x4+x5+x6+x7, ~x2+x3+x4+x5, dipper.histories )

Note that parameter estimates produced by Method 1 and Method 2 are identical.

Plot: $\hat{N}_j$ estimates

Following is a plot of the Horvitz-Thomson population size estimates.


Plot: $\hat{\phi}_j$ estimates

Following is a plot of survival estimates containing one line per individual.


Try the mra package in your browser

Any scripts or data that you put into this service are public.

mra documentation built on May 1, 2019, 6:50 p.m.