timevaryingcov: Time-varying Covariates

Description

Extract or replace time varying covariates

Usage

timevaryingcov(object, ...)
timevaryingcov(object) <- value

Value

timevaryingcov(object) returns the timevaryingcov attribute of

object (may be NULL).

Arguments

object: an object of class traps or capthist
value: a list of named vectors
...: other arguments (not used)

Details

The timevaryingcov attribute is a list of one or more named vectors. Each vector identifies a subset of columns of covariates(object), one for each occasion. If character values are used they should correspond to covariate names.

In secr models, time-varying covariates are restricted to traps objects. Time-varying (session-specific) individual covariates may be used in openCR. The following remarks apply to time-varying traps covariates.

The name of the vector may be used in a model formula; when the model is fitted, the value of the trap covariate on a particular occasion is retrieved from the column indexed by the vector.

For replacement, if object already has a usage attribute, the length of each vector in value must match exactly the number of columns in usage(object).

When converting a multi-session capthist object into a robust-design ``single-session'' object with function join the argument `timevaryingcov' is used to collate covariate values across sessions in a form suitable for inclusion in openCR models (see join).

Examples

Run this code


# make a trapping grid with simple covariates
temptrap <- make.grid(nx = 6, ny = 8, detector = "multi") 
covariates (temptrap) <- data.frame(matrix(
    c(rep(1,48*3),rep(2,48*2)), ncol = 5))
head(covariates (temptrap))

# identify columns 1-5 as daily covariates
timevaryingcov(temptrap) <- list(blockt = 1:5)
timevaryingcov(temptrap)

if (FALSE) {

# default density = 5/ha, noccasions = 5
CH <- sim.capthist(temptrap, detectpar = list(g0 = c(0.15, 0.15,
    0.15, 0.3, 0.3), sigma = 25))

fit.1 <- secr.fit(CH, trace = FALSE) 
fit.tvc2 <- secr.fit(CH, model = g0 ~ blockt, trace = FALSE) 

# because variation aligns with occasions, we get the same with:
fit.t2 <- secr.fit(CH, model = g0 ~ tcov, timecov = c(1,1,1,2,2),
    trace = FALSE) 

predict(fit.t2, newdata = data.frame(tcov = 1:2))
predict(fit.tvc2, newdata = data.frame(blockt = 1:2))

# now model some more messy variation
covariates (traps(CH))[1:10,] <- 3
fit.tvc3 <- secr.fit(CH, model = g0 ~ blockt, trace = FALSE) 

AIC(fit.tvc2, fit.t2, fit.tvc3)
# fit.tvc3 is the 'wrong' model

}