A MISSPECIFICATION TEST FOR MODELS

OF MULTINOMIAL DISTRIBUTIONS

by

Thomas Laitila

Department of Forest Economics, SLU, 901 83 Umeå

ABSTRACT

A general test of model misspecification is derived for models of multinomially distributed data. It is suggested for models where category probabilities depend on continuous explanatory variables. The test makes an alternative to usual chi-2 and likelihood ratio tests for contingency tables with low cell frequencies. Size and power properties of the test and earlier proposed tests are studied by means of simulation.