A Monte Carlo study of tests for the independence of irrelevant alternatives property

A plethora of tests for the Independence of Irrelevant Alternatives (IIA) property of Logit models of discrete choice behavior has been proposed in the literature. These tests are based upon asymptotic arguments and little is known about their size and power properties in finite samples. This paper uses a Monte Carlo simulation study to investigate the size and power properties of six tests for IIA in the multinomial Logit model. Our results show that the majority of tests based upon partitioning the choice set appear to have very poor size and power properties in small samples. Tests for IIA based upon the DOGIT model, similarly have poor size properties, but in some circumstances do have reasonable power properties.