Abstract: | This paper discusses the relevance of hybrid log-linear models for applied research based on categorial data. Such models may lead to statistically significant results based on a reduced number of parameters. The use of long-linear modeling is illustrated by means of spatial interaction analysis with categorial data. Special attention is given to the problem of so-called structural zeros and of stability of spatial interaction patterns. Various constraints on the set of interaction parameters emerging from the off-diagonal elements of a spatial interaction matrix are taken into account by means of an entropy function. The analysis is illustrated by means of an empirical application to migration flows in Dutch housing market areas. |