Here - Tilburg University
Here - Tilburg University
Here - Tilburg University
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Author and presenter<br />
Forcina, Antonio; Dept. of Economics, Finance and Statistics, <strong>University</strong> of<br />
Perugia, Italy<br />
Title<br />
Smoothness of Conditional Independence Models for Discrete Data<br />
Abstract<br />
The paper is about a family of conditional independence models which require<br />
constraints on complete but non hierarchical marginal log-linear parameters. For<br />
such models, whose dependence structure cannot be represented by any of the<br />
known graphical separation criteria, it is not known whether the model is<br />
smooth, so that the usual asymptotics can be applied. A model is called non<br />
smooth when the variety which it defines in the parameter space contains points<br />
which do not admit a local approximation by a linear space.<br />
By exploiting results on the mixed parameterization within the exponential<br />
family, we determine a condition which has to be satisfied for the model to be<br />
smooth. The condition has to do with the possibility to reconstruct the joint<br />
distribution from the set of marginal log-linear parameters in a unique way. In<br />
technical terms, the condition require that a certain jacobian matrix has spectral<br />
radius strictly less than 1. In the simple context when only two marginals are<br />
involved, we show that this condition is always satisfied. In the general case, we<br />
describe an efficient numerical test for checking whether the condition is<br />
satisfied with high probability. This approach is illustrated with several examples<br />
of non hierarchical conditional independence models and by a directed cyclic<br />
graph model; we establish that all these models smooth.