Testing for the mixture hypothesis of geometric distributions

Use of the likelihood ratio (LR) statistic is examined to test for the mixture assumption of geometric distributions. As the asymptotic null distribution of the LR statistic is not a standard chi-square due to the fact that there are a boundary parameter problem and a nuisance parameter not identified under the null, we derive it separately and also provide a method to obtain the asymptotic critical values. Further, the finite sample properties of the LR test are evaluated by Monte Carlo simulations by examining the levels and powers of the LR test. Finally, using Kennan's (1985) strike data in labor economics, we conclude that unobserved heterogeneity is present in the data, which cannot be captured by specifying a geometric distribution.