Table 1.

Percent bias in conditional maximum likelihood estimates based on the outcome-dependent sample (ODS) and maximum likelihood estimates based on the original cohort (OC), and the RVs and RRMSEs from the 2 approaches. Statistics are reported as percentages, and empirical variances (as opposed to average estimated variances) are used in calculations. RV and RRMSE are defined as 100 times the estimate based on the OC analysis divided by the estimate based on the ODS analysis. For N = 400 (N = 4000), we used 1000 (500) replicates

							β0				β1
							Percent bias		Relative efficiency		Percent bias		Relative efficiency
N	ni	γ	ρx	β0	β1	Ave Nr	OC	ODS	RV	RRMSE	OC	ODS	RV	RRMSE
As a function of ρx
4000	14	1.5	1	− 2.75	0.15	2034	0	0	32	55	− 1	0	37	60
			0.5			2050	0	0	33	57	− 1	− 1	74	85
			0			2046	0	0	32	56	1	1	100	99
400	14	1.5	1	− 2.75	0.15	203	0	0	31	54	− 1	− 2	37	62
			0.5			205	0	0	32	56	1	1	74	87
			0			204	0	0	32	59	0	1	99	100
As a function of ni, γ, β1
4000	14	1.5	0	− 2.75	0.15	2046	0	0	32	56	1	1	100	99
					0.5	2180	0	0	38	61	0	0	98	100
4000	7	1.5	0	− 2	0.15	1894	0	0	30	56	1	1	99	101
					0.5	2015	0	0	34	61	0	0	95	97
4000	14	3	0	− 2.5	0.15	1782	0	0	30	54	0	0	99	100
					0.5	1976	0	0	37	60	0	0	96	97
4000	7	3	0	− 1.75	0.15	1477	0	− 1	26	47	0	− 1	98	98
					0.5	1685	0	− 1	33	55	0	0	89	94
400	14	1.5	0	− 2.75	0.15	204	0	0	32	59	0	1	99	100
					0.5	218	0	0	38	60	0	0	98	99
400	7	1.5	0	− 2	0.15	189	0	1	29	54	0	0	99	100
					0.5	202	0	1	34	57	0	0	94	96
400	14	3	0	− 2.5	0.15	178	0	0	30	53	1	1	99	100
					0.5	198	0	1	37	63	1	1	96	98
400	7	3	0	− 1.75	0.15	148	0	0	24	46	2	2	98	99
					0.5	169	0	0	32	54	0	0	89	93
As a function of ni
4000	14	3	0	− 2.5	0.5	1976	0	0	37	60	0	0	96	97
	U(5, 23)					1896	0	0	42	66	0	0	97	98
400	14	3	0	− 2.5	0.5	198	0	1	37	63	1	1	96	98
	U(5, 23)					189	0	0	42	63	0	0	97	98