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