Table 2.

Simulated performance of different methods of estimation of a sieve effect using the match/mismatch metric for the mean number of infecting pathogens in a trial with passive surveillance. Pairs of rows report the average and variance of the estimated α₄ over 1000 simulated datasets. Data generated under a bivariate negative binomial model. X̄ is the mean X averaged over groups and f and then averaged over the simulations. Relative Efficiency is the ratio of squared Wald statistics or ${\mathcal{Z}}_A^2/{\mathcal{Z}}_B^2$, where 𝒵 is the sample average divided by the sample standard deviation.

						Relative efficiency
E{exp(bi)}	var{exp(bi)}	X̄	GEE	Single pathogen	EWCR	GEE/Single	GEE/ EWCR
Bivariate negative binomial
0.5	0.0	0.9	−1.302	−1.304	−1.304	2.68	1.18
			0.073	0.151	0.087
0.5	1.0	0.9	−1.314	−1.336	−1.325	2.51	1.51
			0.067	0.173	0.103
0.5	2.0	0.9	−1.289	−1.322	−1.292	3.49	1.82
			0.060	0.222	0.110
1.0	0.0	1.9	−1.283	−1.299	−1.283	3.33	1.31
			0.032	0.109	0.042
1.0	1.0	1.9	−1.295	−1.322	−1.299	4.89	1.96
			0.030	0.152	0.059
1.0	2.0	1.9	−1.279	−1.283	−1.278	5.00	2.19
			0.033	0.168	0.073
All-or-none
0.5	0.0	6.0	−0.997	−1.241	−1.237	1.21	0.78
			0.076	0.143	0.091
0.5	1.0	4.8	−0.863	−1.148	−1.137	1.02	0.72
			0.092	0.166	0.114
0.5	2.0	4.0	−0.799	−1.108	−1.075	2.75	0.68
			0.102	0.540	0.126
1.0	0.0	8.1	−0.742	−1.082	−1.065	1.34	0.62
			0.035	0.101	0.045
1.0	1.0	6.4	−0.660	−0.971	−0.965	1.24	0.65
			0.045	0.121	0.063
1.0	2.0	5.3	−0.610	−0.907	−0.899	1.20	0.63
			0.054	0.142	0.073