MCMC Procedure

The MCMC procedure provides the following new capabilities:

  • The MODEL statement augments missing values in the response variable by default. PROC MCMC treats missing values as unknown parameters and incorporates the sampling of the missing data as part of the Markov chain.

  • The RANDOM statement supports multilevel hierarchical modeling to an arbitrary depth; a random effect can appear in the distributional hierarchy of other random effects.

  • More distributions, such as multivariate normal distribution with autoregressive structure, Poisson distribution, and general distribution (for the construction of nonstandard distributions), are made available for the RANDOM statement.

  • Direct sampling and more conjugate sampling algorithms are available for all parameters in the model (including model parameters, random-effects parameters, and missing data variables) when appropriate.

  • A slice sampler is an alternative sampling algorithm for both the model parameters and random-effects parameters.