Bayesian Capabilities in SAS/STAT Procedures
Bayesian analysis techniques are now available in both SAS/STAT 9.1.3 and SAS/STAT 9.2, Phase 1.
Bayesian Capabilities in SAS 9.1.3
SAS/STAT software now provides Bayesian analysis in downloadable, experimental versions of three procedures for SAS 9.1.3 on Windows: GENMOD, LIFEREG, and PHREG. The new BAYES statement in these procedures produces Bayesian modeling and inference capability in generalized linear models, accelerated life failure models, Cox regression models, and piecewise constant baseline hazard models (also known as piecewise exponential models). These versions are named BGENMOD, BLIFEREG, and BPHREG, respectively, and they otherwise contain the full functionality of the original procedures.
The following are some highlights:
- Computations use the Gibbs sampler to obtain posterior distributions.
- Convergence diagnostics such as the Gelman-Rubin, Geweke, Heidelberger-Welch, and Raftery-Lewis tests are produced as well as trace plots.
- All procedures offer the normal and uniform prior, and the BGENMOD procedure also provides Jeffreys’ prior.
- You can output the posterior distribution to a SAS data set for use in additional analysis.
The experimental BGENMOD, BLIFEREG and BPHREG procedures are being made available through the SAS web site so that users can provide feedback on this new software. The documentation provides an introduction to Bayesian analyses as well as a comprehensive reading list for further information. Additional chapters contain syntax, details, and examples for the individual procedures BGENMOD, BLIFEREG, and BPHREG. These chapters do not repeat information that is included in the SAS/STAT documentation for SAS 9.1.3.
Bayesian Capabilities in SAS 9.2, Phase 1
These same capabilities described above are production in SAS/STAT 9.2, First Phase, now available for all platforms. These capabilities have been rolled into the GENMOD, LIFEREG, and PHREG procedures in this release.
The experimental MCMC procedure is a flexible simulation-based procedure that is suitable for fitting a wide range of Bayesian models. You specify a likelihood function for the data and a prior distribution for the parameters and PROC MCMC obtains samples from the corresponding posterior distributions. It produces summary and diagnostic statistics.- Use programming statements similiar to to those used in PROC NLMIXED.
- Specify hyperprior distributions to fit hierarchical model.
- Save posterior samples to an output data set.
- Parameters can enter the model linearly or in any nonlinear functional form.
For more information, please see the PROC MCMC documentation, as well as
Focus on Bayesian Methods
.