ABSTRACT
Iachine (2001) discussed the estimation method for bivariate frailty models. Maximum likelihood is not the only possible estimation method for survival models with random eects (Korsgaard et al., 1998; Yau and McGilchrist, 1998; Hougaard 2000, p. 274). However, it has several appealing features which is why it is popular. First, the presence of asymptotic consistency and asymptotic normality properties for nite-dimensional parametric models is a good starting point for estimation and inference. For semiparametric models the results of Murphy (1995), Parner (1998), and Korsholm (1999) justify the semiparametric estimation and inference procedures in gamma-frailty models. For more general frailty models the work of Murphy and Vaart (2000) on prole likelihood may provide a foundation for semiparametric maximum likelihood estimation.
