Table 1.

Simulation results for the general measurement error model with Gaussian subject-specific and random error. For 1000 simulated data sets, the mean bias, empirical standard deviation (SD), bootstrap standard deviation (BSD), root mean squared error (RSME) and estimated 95% coverage probability (CP) are given for β = log 1.5, log 3.

	Moderate Subject-Specific Bias					Strong Subject-Specific Bias
β = log 1.5	Bias	SD	BSD	RMSE	CP	Bias	SD	BSD	RMSE	CP
True	0.000	0.041	0.041	0.041	96.5	0.000	0.041	0.041	0.041	96.5
Naive	−0.217	0.032	0.032	0.219	0.0	−0.265	0.037	0.037	0.268	0.0
RC	−0.014	0.063	0.062	0.065	92.9	−0.016	0.069	0.069	0.071	93.3
RRC	−0.012	0.068	0.073	0.069	95.4	−0.013	0.073	0.073	0.074	93.4
CS B	0.010	0.099	0.102	0.099	95.8	0.010	0.099	0.102	0.099	95.8
CS W	0.011	0.091	0.096	0.092	96.5	0.010	0.102	0.105	0.102	95.6
NP B	0.014	0.123	0.128	0.124	96.9	0.014	0.123	0.128	0.124	96.9
NP W	0.012	0.119	0.123	0.120	96.7	0.007	0.122	0.126	0.122	95.7

β = log 3	Bias	SD	BSD	RMSE	CP	Bias	SD	BSD	RMSE	CP
True	0.001	0.050	0.051	0.050	95.4	0.001	0.050	0.051	0.050	95.4
Naive	−0.678	0.035	0.034	0.679	0.0	−0.806	0.039	0.038	0.806	0.0
RC	−0.190	0.095	0.094	0.213	46.2	−0.219	0.098	0.097	0.240	38.5
RRC	−0.120	0.107	0.119	0.161	79.7	−0.146	0.107	0.108	0.181	69.9
CS B	0.035	0.182	0.198	0.186	97.4	0.035	0.182	0.198	0.186	97.4
CS W	0.023	0.171	0.183	0.172	96.5	0.025	0.180	0.195	0.181	97.6
NP B	0.070	0.255	0.289	0.265	96.6	0.070	0.255	0.289	0.265	96.6
NP W	0.056	0.246	0.275	0.252	96.3	0.041	0.262	0.276	0.265	95.2

True: Cox regression with true X; Naive: Cox regression with unadjusted Q; RC: ordinary regression calibration; RRC: risk set regression calibration; CS B: CS from Tsiatis and Davidian (2001) using biomarker W data only; CS W: weighted combination of CS using Q only and W only; NP B: NP equation from Huang and Wang (2000) using only W; NP W: weighted combination of NP using Q only and W only.