TABLE 2.

Comparison of multilevel approach with the marginal approach and the available case approach from simulation study for continuous outcome where working models were misspecified by omitting an important covariate

				β 0				β 1
				Emp var ratiob		MSE ratiob		Emp var ratiob		MSE ratiob
Effect strength	% Miss.	Ra	Ya	Available case approach	Marginal approach	Available case approach	Marginal approach	Available case approach	Marginal approach	Available case approach	Marginal approach
Weak	20	T	T	0.994	0.844	0.992	0.845	1.103	0.858	1.104	0.858
		T	F	0.993	0.847	0.994	0.850	1.104	0.861	1.106	0.862
		F	T	0.993	0.846	0.991	0.847	1.101	0.861	1.102	0.861
		F	F	0.993	0.848	1.002	0.851	1.102	0.863	1.105	0.864
	35	T	T	1.002	0.671	0.971	0.672	1.042	0.667	1.040	0.667
		T	F	1.009	0.679	0.983	0.684	1.054	0.676	1.053	0.677
		F	T	0.997	0.671	0.966	0.671	1.037	0.668	1.036	0.669
		F	F	1.003	0.679	0.991	0.684	1.050	0.676	1.049	0.677
Strong	20	T	T	0.901	0.891	0.626	0.891	0.965	0.907	0.905	0.907
		T	F	0.964	0.935	0.707	0.985	1.023	0.954	0.968	0.963
		F	T	0.892	0.913	0.620	0.912	0.951	0.928	0.892	0.928
		F	F	0.996	0.937	1.385	0.973	1.021	0.953	1.136	0.965
	35	T	T	0.877	0.739	0.398	0.738	0.915	0.762	0.824	0.762
		T	F	1.123	0.857	0.602	1.010	1.151	0.866	1.061	0.886
		F	T	0.811	0.795	0.368	0.795	0.848	0.819	0.764	0.819
		F	F	1.044	0.890	2.016	1.007	1.058	0.896	1.338	0.946

^aT = Working model specified correctly. F = Working model misspecified by excluding the covariate Z_3,ij.

^bRatio comparing multilevel approach to corresponding comparison method.