Bayesian Communication Software

The software provided at this site enables you to accomplish many of the goals of Bayesian communication in a user-friendly way. The software uses the output of any posterior simulator to accomplish many tasks:

Calculate posterior means and posterior standard deviations, and assess the numerical accuracy of the posterior mean (moment)
Evaluate the marginal likelihood of a model (mlike)
Combine the runs of several posterior simulator runs for the same model, checking that the runs are mutually consistant (apm)
Change the prior distribution by reweighting the posterior simulator output, and/or evaluate alternative posterior moments (reweight)
Prepare input for standard plotting software, to approximate posterior probability density functions (graph)
Calculate upper and lower bounds on posterior moments, as the prior distribution is varied from its original specification.(robust)
Prepare a summary of a posterior simulator file that is useful in designating the transient phase of a Markov chain Monte Carlo posterior simulator (converge)

To use any of the seven routines, it is necessary that posterior simulator output be organized in a standard way. The reweight software generates new posterior simulator files in the same, standard way.

Posterior simulator file structure

The following is a quick technical summary. For a more comprehensive introduction, see the files indicated below.

The initial record of a posterior simulator file consists of two integers: the first is the number of iterations, and the second is the number of entries in the vector written in each iteration.

For each iteration, two records are written. The first record is an integer followed by three real constants. The integer is the iteration number; it reflects the number of skips, if any, between iterations. (This integer is only for your convenience in e xamining the posterior simulator file should you wish to do so. It is not used in any way by the software. The first real constant is the logarithm of the weight function, i.e., the log ratio of posterior density kernel to importance sampling kernel; for many methods htis value is zero. The second real constant is the logarithm of the prior density (not merely the kernel) at the parameter vector for the iteration, and the third real constant is the logarithm of the data density (not merely the kernel) at the parameter vector for the iteration.

The second record for each iteration is a parameter vector, written five entries per line and in general occupying multiple lines. The organization of this vector is specific to the particular application. If you have not generated the posterior simulator file yourself, it will be necessary to find documentation indicating how this file has been organized.

The software is available in six languages: Fortran 77; c+; Gauss; Mathematica; Matlab; and S-Plus.
The above introduction and explanation is available in pdf and postscript format.

Index

Workstations

The following executables are compiled to run on Sun-Ultra workstations such as Gibbs and Daedalus. It may happen that these file will not be executable after downloading. Type chmod 711 [filename] to make them executable.

DOS

The following executables are 32-bit DOS executables. They should be used in any PC that has a 32-bit environment. See models for more details.
There are not any executables for the routines MLIKE and REWEIGHT as these require user defined subroutines for each problem. Refer to Models for the DOS versions of MLIKE for each of the models defined there.