Table 4.

Large-sample robustness of the assumed finite mixture (FM) model to the true dependence structure between tests given by the Gaussian random effects (GRE) model

r	σ1	Estimator (misspecified model)			Expected log-likelihood*
		$P_d^{\ast }$	SENS*	SPEC*	EGRE[log LGRE]	EGRE[log LFM]
0	1.5	.22	.84	.94	−1.74339	−1.74339
.2		.21	.82	.93	−1.78383	−1.81920
.5		.20	.78	.91	−1.86950	−1.91198
1		.20	.75	.90	−1.99560	−2.05440
0	3	.22	.86	.95	−1.61806	−1.61806
.2		.21	.82	.93	−1.67106	−1.70415
.5		.20	.76	.90	−1.75057	−1.79748
1		.20	.75	.90	−1.88307	−1.95351

NOTE: Verification is independent of Y_i and r denotes the proportion of random samples verified [i.e., P(V_i = 1) = r]. The true model is a GRE model with P_d = .2, SENS = .75, SPEC = .9, σ₀ = 1.5, and J = 5 for differing r.

^*Expected individual contribution to the log-likelihood.