Relationship to MARK - For the most part, these routines perform a subset of the analyses available in program MARK or via the MARK front-end package, RMark. However, there are differences. The most significant difference between this package and MARK is parameterization. The parameterization used here does not utilize triangular "parameter information matrices" (PIMs) as MARK (and RMark) does. Because of this, the "design" matrix utilized by this package is not parallel to the "design" matrix of program MARK. For those new to mark-recapture analysis, this parameterization difference will be inconsequential. The approach taken here provides equivalent modeling flexibility, yet is easier to grasp and visualize, in our opinion. For those already familiar with the PIMs used by program MARK, it is helpful to view the "PIMs" of this package as rectangular matrices of the real parameters. I.e., the "PIMs" of this package are rectangular matrices where cell (i,j) contains the real parameter (capture or survival) for individual i at capture occasion j.
Analyses available here that are not included in program MARK include:
History - These routines grew from consulting work on multiple mark-recapture projects. The original Fortran code, upon which the package is based, was written by Dr. Bryan Manly for a northern spotted owl similation project in 1991. Dr. Manly is the one who originally envisioned and programed the non-PIM (or rectangular PIM, if you prefer) approach. However, Dr. Manly did not realize what he had done. In 1997, Dr. Trent McDonald almost completely revised the original Fortran routines for use on a polar bear mark-recapture project. At that time, the routines were stand-alone Fortran executables. Subsequent revisions required by other projects included addition of closed-form variance estimates (originally, variances were estimated by the bootstrap), the Horvitz-Thompson size estimates, and goodness of fit testing. In 2003, it finally dawned on Dr. McDonald how to call a Fortran DLL from S-Plus and R, thus eliminating the executable version and allowing S-Plus or R to do front-end data manipulation and plotting. S-Plus was abandoned in favor of R in 2004. After publication of Amstrup et al. (2005), Dr. McDonald realized that an official R package with documentation was needed, and learned how to make a package (not an easy process for him). Version 1.X of MRA contained routines for open population CJS models only. Version 2.X of MRA added closed population Huggins model estimation routines. In future, addition of a major model type will increment the primary version number, revisions to routines for existing models will increment the secondary version number. Thus, the number of major model types available in MRA will be the primary version number. Throughout the process, several statisticians, including Dr. Manly, Dr. Jeff Laake and Dr. Gary White, have provided comments that helped shape the approach.
Ways You Can Help - It is a no-brainer that others have R routines that perform various mark-recapture analyzes. The author of MRA views this effort as collaborative and welcomes routines, comments, and assistance in developing MRA. The overall goal is to develop MRA into a comprehensive repository for mark-recapture analyzes in R. Indeed, collaboration is the strength of R and open source code.
Along these lines, the author of MRA would especially welcome routines that perform the following analyzes:
The above is only a partial list. All routines are welcome and will be considered. Original authors will of course be acknowledged in the routine's documentation. Assistance in formating documentation files is available. I.e., either the author of MRA will write the documentation or the routine's original author will be sent a text file to edit that contains all the necessary mark-up and sectioning.
If you have a routine that would be useful, email MRA's maintainer at the address below. Supporting papers and data sets can be distributed with MRA, and are encouraged.
| Package: |
| mra |
| Type: |
| Package |
| License: |
| GNU General Public License |
List of routines:
F.cjs.covars Returns matricies that can be used to fit a CJS model
F.cjs.estim Estimation of Cormack-Jolly-Seber (CJS) open population model
F.cjs.gof Goodness-of-fit tests for CJS models
F.cr.model.matrix Function that returns two 3-D design matricies.
F.3d.model.matrix Function that expands a formula into the type of 3-D array
needed for estimation.
F.huggins.estim Estimation of Huggin's closed population model
F.sat.lik Returns the saturated likelihood for a CJS model that does
not contain individual covariates
F.fit.table Prints table of fit statistics from a list of MRA objects
F.cjs.simulate Simulates capture data from MRA CJS model
F.step.cjs Performs step-wise variables selection for CJS models
F.cr.model.avg Performs model averaging from a list of MRA objects
mra.control Controls aspects of model fitting (interations,
tolerance, tracing, etc.)
dipper.data European Dipper data set
lines.cjs Lines method for cjs objects
plot.cjs Plot method for cjs objects
predict.cjs Predicted values for active cells of a CJS model.
print.cjs Print method for cjs objects
print.hug Print method for hug (Huggin's model) objects
print.nhat Pring method for size estimates from a CJS model
residuals.cjs Residuals for CJS models
ivar Function for specifying individual varying effects
tvar Function for specifying time varying effects