Hierarchical Elimination and Regionalization in Spatial Choice

Spatial choice models have been criticized for their restrictive assumptions. In particular, the independence or constant-ratio axiom of Luce has been questioned because of its failure to account for the effects of similarity between alternatives. Tversky's elimination by aspects model provides a natural explanation of similarity effects, but cannot be estimated from binary probabilities. The elimination by tree or, equivalently, hierarchical elimination model is less restrictive than the constant-ratio model and more parsimonious than elimination by aspects. To estimate this model in a spatial interaction context, a hierarchical regionalization based upon interareal flow data is first obtained. The lengths of the links in the resultant tree structure are reestimated on the basis of choice probability ratios.