A Bayesian nonparametric model for combining data from different experiments

A number of models have been recently proposed in the Bayesian non-parametric literature for dealing with data arising from different related studies. In this paper we consider a modeling approach that relies on canonically correlated Poisson random measures. These lead to define vectors of dependent random prob-ability measures, which are useful in the contexts of density estimation and survival analysis. With reference to the former we point out results useful for devising a Gibbs sampling algorithm. This is then used to emphasize some remarkable features, especially in terms of the clustering behavior and the borrowing information across datasets, of a class of dependent nonparametric priors based on the normalized sigma-stable process.