RMark (version 3.0.0)

deer: White-tailed deer double observer spotlight capture-recapture analysis

Description

This data represents a set of independent double observer road-transect survey data of white-tailed deer on Brosnan Forest, South Carolina surveyed in August, 2005-2009. The primary reason for this package is to provide a completely reproducible example of the analysis from Collier et al. (2012). We used a Huggins closed capture model implemented in MARK http://www.phidot.org/software/mark/ via RMark both of which will need to be installed on the system to use this package. The data have 2 time periods (primary observer (t1) was a thermal imager, secondary observer (t2) was a spotlight observer in the same vehicle on the same side) with the primary objective of the study being to evaluate the detection (recapture) rates of white-tailed deer using spotlights as a survey method.

Format

The format is a data frame with 4508 observations on the following 7 variables.

SL (spotlight)

0/1 whether deer was missed/seen by the spotlight observer

TI (thermal imager)

0/1 whether deer was missed/seen by the thermal imager observer

Group

Factor with 79 levels representing each unique paired (TI-SL) survey conducted

Year

Factor with 5 levels for year of survey

MaxCount

Count of maximum number of deer seen for each survey, only needed for bootstrapp analysis in MARK, not used in bfdeeR package

Cluster

Value assigning each deer to a specific observation cluster, only needed for bootstrapp analysis in MARK, not used in bfdeeR package

MgmtUnit

Management unit identification

Author

Bret Collier

Details

In addition to detailing the analysis used by Collier et al. (2012), this example documents the use of the share argument in the RMark parameter specification because there is presently very little documentation on the use of share. Parameters in MARK models rarely share columns of the design matrix. For example while you might want to use the same covariate for survival and capture probability, you would never use the same beta (same column of the design matrix) for each parameter. However, there are exceptions when the parameters represent similar quantities and that is when the share argument is useful. For example, in the closed capture models p is initial capture probability and c is recapture probability. In this case, it would make perfect sense to use the same column of the design matrix for both parameters. The most obvious case is to fit a model in which p=c.

In RMark, certain pairs of parameters have been identified as similar and shareable. These can be found in the file parameters.txt which is in the RMark directory in your R library. With each pair that is shareable, the first one listed is the primary parameter. When you want to share columns in the design matrix, share=TRUE is added to the specification of the primary parameter. A parameter specification is not given for the other secondary parameter when they are shared. When RMark, sees that the parameters are to be shared it creates a pooled set of design data and adds a column with the name of the secondary parameter and its value is 0 for the rows for the primary parameter and 1 for the rows for the secondary parameter. For example, with the closed capture model if share=TRUE is added to the parameter specification for p, a model is not specified for c, and the pooled design data set contains a field called c. The added field allows construction of models where there are restricted differences between the parameters. For example, p=list(formula=~time+c,share=TRUE) will fit a model in which capture probability varies by time and recapture probability includes an additive difference on the link scale. Because the design data are pooled when you share parameters, if you modify design data for one of the parameters, the other most be modified as as well, so the columns of the design data for both parameters are the same or RMark will give an error.

The argument share is used in all the candidate models in the below example analysis. As a simplified example of how share works, look at the candidate models in the bfrun{} function call named mod.2 and mod.2a (note that mod.2a was not included in the supplemental file available from the Journal of Wildlife Management and is only included in this package). Both of these models are conducting the exact same analysis, with the first mod.2, we used the formula ~time (if you don't know what this means go read the MARKBOOK at http://www.phidot.org/software/mark/. Notice, however, we used the argument share in mod.2, which tells RMark to share columns of the MARK design matrix. For comparison, so you can evaluate how share works for yourself, mod.2a recreates the same analysis as mod.2, but uses the approach more typical to MARK analyses where each parameter is specified independently and uniquely.

References

Collier, B. A., S. S. Ditchkoff, J. B. Raglin, and C. R. Ruth. 2012. Spotlight surveys for white-tailed deer: monitoring panacea or exercise in futility? Journal of Wildlife Management, In Press.

Examples

# \donttest{
# This example is excluded from testing to reduce package check time
data(deer)
x=data.frame(ch=paste(deer$TI, deer$SL, sep=""), Survey=factor(deer$Group), 
     Year=factor(deer$Year), Cluster=deer$Cluster, MgtUnit=factor(deer$MgmtUnit))
x$ch=as.character(x$ch)
bfrun=function(){
x.proc=process.data(x, model="Huggins", groups=c("Survey", "Year", "MgtUnit"))
x.ddl=make.design.data(x.proc)

#Silly Null model, constant p & c sharing 1 parameter (one detection estimate)
p.shared=list(formula=~1,share=TRUE)
mod.1=mark(x.proc, x.ddl, model.parameters=list(p=p.shared), invisible=FALSE,delete=TRUE)
 
#2 Parameter Null Model, constant p, constant c, different p and c (one estimate for each; p ne c)
#p(time), c(-), share=TRUE, detection is time dependent, with recapture parameter shared
p.sharetime=list(formula=~time, share=TRUE)
mod.2=mark(x.proc, x.ddl, model.parameters=list(p=p.sharetime), invisible=FALSE,delete=TRUE)

#2a Parameter Null Model, constant p, constant c,
# different p and c (one estimate for each; p ne c) not using share
mod.2a=mark(x.proc, x.ddl, model.parameters=list(p=list(formula=~1), c=list(formula=~1)),
            delete=TRUE)

#Fully parameterized model, different p and c for each survey transect replicate, 
# management unit, method (TI or SL) and any observers
p.survey=list(formula=~Survey*time, share=TRUE)
mod.3=mark(x.proc, x.ddl, model.parameters=list(p=p.survey), invisible=FALSE,delete=TRUE)

#p(MU), c(MU), initial detection and recapture differ and are management unit dependent
p.mu=list(formula=~MgtUnit*time, share=TRUE)
mod.4=mark(x.proc, x.ddl, model.parameters=list(p=p.mu), invisible=FALSE,delete=TRUE)

#p(MU) detection is management unit dependent
p.mu=list(formula=~MgtUnit, share=TRUE)
mod.5=mark(x.proc, x.ddl, model.parameters=list(p=p.mu), invisible=FALSE,delete=TRUE)

#p(Yr + MgtUnit),  detection is year + MgtUnit
p.yearMgtUnit=list(formula=~Year*time+MgtUnit, share=TRUE)
mod.6=mark(x.proc, x.ddl, model.parameters=list(p=p.yearMgtUnit), invisible=FALSE,delete=TRUE)

#p(Year), initial detection and recapture are year dependent
p.year=list(formula=~Year*time, share=TRUE)
mod.7=mark(x.proc, x.ddl, model.parameters=list(p=p.year), invisible=FALSE,delete=TRUE)

return(collect.models())
}
bf.out=bfrun()
bf.out

#export function to send dataset and covariates data to MARK for bootstrap analysis 
#(not run but here for completeness)
#export.MARK(x.proc, "BFdeer", mod.3, replace=TRUE, ind.covariates="all")
# }