The Prosecutor's Fallacy

Author: Carl Boettiger cboettig@gmail.com

Overview

This package is provided as supplementary material to fully document the research methods leading up to and used in my paper:

Boettiger, C. & Hastings, A. 2012 Early warning signals and the prosecutor’s fallacy. Proc. R. Soc. B Biol. Sci. (doi:10.1098/rspb.2012.2085)

The resulting data, along with the code as it appeared at the time of acceptance is permanently archived on Dryad.

This package was later used in deriving the additional results presented in the paper:

Boettiger, C. & Hastings, A. 2013 No early warning signals for stochastic transitions: insights from large deviation theory. Proc. R. Soc. B Biol. Sci. 280, 20131372–20131372. (doi:10.1098/rspb.2013.1372)

This package makes use of existing software libraries, and cannot guarentee future compatibility with later versions of those dependencies (For instance, the odesolve R package has since been removed from CRAN.) Intepretation and modification of the methods should still be facilitated by the availability of the code, however as time passes it will be increasingly unlikely that the package can simply be installed and the code executed without error. I welcome bug reports for any such difficulties and will try to address them as time allows.

Installation

Requirements

The devtools package (available on CRAN) should automatically handle most of the package dependencies, with one exception of a separate package I use to provide routines to simulate the individual-based population dynamics. See https://github.com/cboettig/populationdynamics for details.

If you have the GSL libraries installed on your machine, this might work for you:

• R (>= 2.15)
• R packages: devtools, ggplot2, deSolve, reshape2, psych, and plyr
• Install the package for generating the simulations

require(devtools)
install_github("populationdynamics", "cboettig")
install_github("prosecutors-fallacy", "cboettig")

Please file a bug report if you encounter problems.

Note: Several of the examples use code parallelized for multiple processors. If your machine has greater/fewer than the specified number of cpus available, modify the cpu settings in the code for optimal performance. Several of the examples may take several days to run on 16 processor machines specified in the code.

Reproducible code for the analysis

All code

• Code for analysis and plotting in Figure 1
• Code for the analysis and plotting in Figure 2
• Code for the analysis and plotting in Figure 3

Source code for example files

Examples are written in the dynamic document format knitr, which combines R code with rich formatting markup describing methods, results, and displaying figures. Source code for these files can be found in the .Rmd file of the same name in the same directory. A complete reproducible example can be generated by running the knit function on any of the corresponding .Rmd files.