The (mis)specification of discrete time duration models with unobserved heterogenity: a Monte Carlo study