• :

    There are many prior distributions for random partition models in literature.  These prior distributions often have the exchangeability property, but a growing number of works contain priors that are not exchangeable.  These non-exchangeable priors are useful when, a priori, some similarity or clustering information between items is known.  When this occurs, it useful to incorporate as much of that information into the prior as possible.  For example, if it is known that two items are likely in the same partition with a specific prior probability an effective way of quantifying should produce a more accurate posterior distribution.  Our approach differs methodologically from the previously proposed "Centered Partition Processes" and has advantages of improved posterior computation times and additional flexibility in incorporating prior information.  The proposed "Random Focal Partition Prior" is demonstrated on a quality control example.

  • : Dahl, David; Warr, Richard; Jensen, Tom
  • : Brigham Young University
  • : Richard Warr
  • : statistics
  • : intermediate
  • : warr@stat.byu.edu
Random Focal Partition Priors