Stable Isotope Bayesian Ellipses in R
SIBER is a series of tools for comparing variation within and among groups of isotope-space data as described in Jackson, A.L., Parnell, A.C., Inger R., & Bearhop, S. 2011. Comparing isotopic niche widths among and within communities: SIBER – Stable Isotope Bayesian Ellipses in R. Journal of Animal Ecology, 80, 595-602. SIBER was previously part of the siar package which also included a mixing model for determining diet contributions from sources to a consumer (or any sources that form a mixture). Siar will become defucnt at some point in the future (possibly by end of 2015) with the mixing model component being moved to a new standalone package simmr maintained by Andrew Parnell, and SIBER taking on a dedicated analysis for isotopic niche width.
The move away from SIAR has arisen primarily as we have moved to fitting the Bayesian models using jags, specifically through the R package rjags. In the case of SIBER, this move to jags makes it easier to control the parameters of the MCMC sampling process such as through burnin length, number of chains and the amount of thinning. It also makes the output more amenable to convergence checking through the rjags package. Other than this additional functionality, the move to jags makes no discernible difference to model fitting.
However, there is a difference in the fitting procedure that does lead to different, and more accurate results. In the implementation of SIBER, the data are z-score transformed prior to the esimation of uncertainty in their means and covariance matrices. These z-score fitted posterior distibutions are then back-transformed internally in SIBER so that they are on the same scale and location as the original data. This fixes an issue that occasionally arose in data where the x and y variables were on very different scales, and/or the data showed strong correlation in the x-y data. This issue meant that there was an obvious mis-match between the Bayesian and Maximium Likelihood estimates of the Standard Ellipse Area.
The learning resources for SIBER are currently in development with the launch of this standalone package. The package vingette includes code and examples of the main analyses. For now, until I get a chance to update them, below is a series of screencast tutorials in which Andrew Jackson talks through various analyses in the SIBER components. THe concepts and overall look of the analyses are identical between SIBER and SIAR, but the code implementation in the new format is slightly different (but more user friendly and helpful I hope!). The scripts and data used in SIAR are available at SIAR-examples-and-queries
SIBER was developed by Andrew Jackson, Andrew Parnell, Richard Inger and Stuart Bearhop.
Having trouble with SIBER? Check out the learning resources pages for SIBER. Bugs and suggestions should be left as issues on the main SIAR development page.