. Author manuscript; available in PMC: 2009 Jun 25.

Published in final edited form as: J Am Stat Assoc. 2009 Jun 1;104(486):512–523. doi: 10.1198/jasa.2009.0017

Table 6.

The estimation and coverage performance of each model on MSI sensitivity (true value = 0.90) based on simulation studies with 2,000 replicates

True models		Random effects models using NLMIXED							Bayesian hierarchical models using WinBUGS
True models		None	ε_i	μ_iA	μ_iB	ε_i, μ_iA	ε_i, μ_iB	μ_iA, μ_iB	None	ε_i	μ_iA	μ_iB	ε_i, μ_iA	ε_i, μ_iB	μ_iA, μ_iB
None	Mean	902	900	903	900	901	900	902	897	900	919	902	918	902	922
	95% CI length^*	83	89	98	89	96	89	97	79	84	105	83	101	81	101
	95% CICP^*	961	968	975	964	977	969	976	938	949	944	951	942	946	920

ε_i	Mean	902	900	897	896	902	900	890	897	900	902	897	916	902	896
	95% CI length^*	85	86	123	92	96	87	129	80	81	126	86	103	79	130
	95% CICP^*	961	966	972	956	977	968	956	941	958	981	946	950	945	977

μ_iA	Mean	893	891	900	891	898	891	899	888	892	911	894	911	895	915
	95% CI length^*	85	91	111	91	111	91	111	81	86	112	85	108	83	108
	95% CICP^*	864	879	949	873	942	876	950	829	879	959	890	951	875	948

ε_i, μ_iA	Mean	893	892	890	887	899	892	882	889	893	894	888	909	895	887
	95% CI length^*	85	86	132	92	106	87	139	82	84	134	88	110	81	138
	95% CICP^*	860	885	944	852	952	886	898	831	873	977	855	957	866	963

NOTE: The numbers have been multiplied by 1,000 for presentation. The bolded cells represent the correctly chosen model.

95% CICp = 95% confience interval coverage probability 95% CICP, and 95% CI length are based on logit-normal assumption for the random effects models using NLMIXED and equal tail credible intervals for the Bayesian hierarchical models using WinBUGS.