ABSTRACT
The discussion to this point has made heavy use of the concepts of independence and unrelatedness. Most models have been put in these terms. Of course, any model for a multifactor table other than complete independence is also a description of dependence, in the sense that it allows dependencies among the combinations of factors that are not specifically asserted to be unrelated. Nevertheless, the models are built around the supposition that certain parameters of the saturated model are zero. When these models are rejected, one does not have very sophisticated ways to describe the association. One can list the values of the relevant λ parameters or can examine the residuals from a simpler model, but these are only general tabulations. One needs a better way to characterize dependence.
