Contributed Discussion on: Centered Partition Processes: Informative Priors for Clustering by Paganin, S., Herring, A.H., Olshan A.F., Dunson, D.B

We congratulate the authors on the development of a broad new class of Bayesian clus- tering models that allow for inclusion of prior information on the clustering structure, and result in improvement of model performance in practice when such prior informa- tion is available. The authors applied this method in an interesting and novel context where there are N = 26 different birth defects, and for each defect i ∈ {1,...,N}, there is a highly variable number of observations. The prior knowledge of the initial partition c0 of the birth defects provided by experts is merged with information in the data through a grouped logistic regression model to produce a posterior distribution on the partition, which also characterizes uncertainty.