A class of smooth models satisfying marginal and context specific conditional independencies
We study a class of conditional independence models for discrete data with the property that one or more log-linear interactions are defined within two different