Table 1

Scenario 1a: Correctly Specified Model for Intercepts
	Complete-case MLE				MLE				Pseudo-conditional Likelihood Method
	θ⁽⁰⁾ =0.35				θ⁽⁰⁾=0.35				θ⁽⁰⁾=0.35
Mean	0.357	0.149	0	0.497	0.354	0.147	0.001	0.498	0.351	0.148	0.002	0.5
Median	0.353	0.149	0.003	0.498	0.349	0.146	0.004	0.496	0.348	0.146	0.005	0.498
MAD	0.17	0.157	0.161	0.154	0.127	0.13	0.123	0.128	0.129	0.133	0.129	0.127
Emp. SE	0.161	0.152	0.153	0.158	0.126	0.128	0.125	0.126	0.129	0.132	0.128	0.128
Est. SE	0.163	0.153	0.152	0.159	0.129	0.125	0.124	0.129	0.131	0.128	0.128	0.133
Bias	0.007	−0.001	0	−0.003	0.004	−0.003	0.001	−0.002	0.001	−0.002	0.002	−0.000
B. Score	1.985	−0.189	0.046	−0.720	1.406	−0.904	0.455	−0.791	0.448	−0.550	0.693	−0.064
RMSE	0.161	0.152	0.153	0.158	0.127	0.128	0.125	0.126	0.129	0.132	0.128	0.128
CP	0.948	0.955	0.952	0.949	0.958	0.949	0.948	0.956	0.95	0.948	0.951	0.963
Scenario 1a: Correctly Specified Model for Intercepts
	Complete-case MLE				MLE				Pseudo-conditional Likelihood Method
	θ⁽⁰⁾ =0.35				θ⁽⁰⁾=0.35				θ⁽⁰⁾=0.35
Mean	0.475	0.027	−0.077	0.456	0.476	0.023	−0.076	0.459	0.383	0.122	−0.017	0.482
Median	0.478	0.031	−0.077	0.451	0.476	0.022	−0.075	0.457	0.383	0.12	−0.016	0.479
MAD	0.147	0.146	0.155	0.147	0.117	0.125	0.124	0.122	0.133	0.13	0.134	0.124
Emp. SE	0.148	0.15	0.152	0.152	0.117	0.122	0.124	0.122	0.133	0.128	0.13	0.127
Est. SE	0.159	0.149	0.149	0.155	0.126	0.122	0.122	0.125	0.136	0.128	0.127	0.132
Bias	0.125	−0.123	−0.077	−0.044	0.126	−0.127	−0.076	−0.041	0.033	−0.028	−0.017	−0.018
B. Score	37.776	−36.571	−22.603	−12.880	48.513	−46.309	−27.410	−15.099	11.16	−9.918	−5.717	−6.248
RMSE	0.194	0.194	0.171	0.158	0.172	0.176	0.145	0.129	0.137	0.131	0.131	0.129
CP	0.898	0.86	0.916	0.942	0.858	0.812	0.898	0.942	0.954	0.946	0.946	0.954

Table 1: Simulation results for the complete-case MLE, the MLE, and the pseudo-conditional likelihood method. Here MAD, Emp. SE, Est. SE, Bias, B. Score, RMSE, and CP denote median absolute deviation, empirical standard error, estimated standard error, bias, bias score, root mean squared error, and 95% coverage probability based on the Wald-type confidence intervals, respectively. The results were based on 2,000 replications. There were 2×2×2=8 disease subtypes. The missingness probabilities depended on the covariate.