Conditional independence relations and log-linear models for random matchings (Q1046845)

The paper examines the probability distributions on all possible complete matchings in a complete bipartite graph admitting linear ordering for the vertices. To do so the author defines a distribution providing the implicit and explicit parametric description. The derived results are non-trivial for \(n \geq 4\). The author introduces the idea of \(L\)-decomposability and then characterizes random matchings for which the distribution of permutations is \(L\)-decomposable. Characterizations of only strictly positive distributions are given. Later, the author focuses on the log-linear model. He proves that the family of strictly positive bi-decomposable distributions can be written as a log-linear model. The number of free parameters in this family is also examined. Maximum likelihood estimation is discussed from a theoretical perspective. Ordering is also treated as an unknown parameter to be estimated. The results obtained in the paper are mainly theoretical. No real data application is provided, although the author claims for such practical aspects of the derived results.