TABLE 4.

Simulation results for estimation of covariate-adjusted log-odds ratio relating MCP-1 and spontaneous abortion, with errors distributed lognormal (shifted to mean 0) with skewness of 1 and the same variance as in the normal-errors case (2500 trials, true value = 0.20)

	Mean Bias (Median Bias)	SD(IQR)	Mean SE	MSE	95% CI Coverage
PE only
Logistic regression ignoring MCP-1 errors	−0.097	0.050	0.048	0.012	0.458
LRF	0.012	0.102	0.099	0.011	0.956
LRA	0.012	0.101	0.099	0.010	0.956
DFA	0.004	0.094	0.094	0.009	0.960
ME only
Without replicates
Logistic regression ignoring MCP-1 errors	−0.024	0.064	0.063	0.005	0.921
LRF	0.028	0.109	0.104	0.013	0.969
LRA	0.027	0.107	0.103	0.012	0.969
DFA	0.004	0.085	0.094	0.007	0.964
With replicates
LRF	0.006	0.077	0.074	0.006	0.950
LRA	0.006	0.077	0.074	0.006	0.949
DFA	0.003	0.074	0.073	0.006	0.951
PE and ME
Without replicates
Logistic regression ignoring MCP-1 errors	(−0.106)	(0.062)	-	-	0.362
LRF	(0.067)	(0.366)	-	-	0.984b
LRA	(0.066)	(0.359)	-	-	0.985c
DFAa	(0.060)	(0.348)	-	-	0.984
With replicates
LRF	0.016	0.109	0.104	0.012	0.959
LRA	0.015	0.108	0.103	0.012	0.959
DFA	0.007	0.100	0.099	0.010	0.956

^anon-bias-adjusted version of estimator used because bias adjustment frequently flipped sign of log-OR estimate (76.0% of trials).

^bExcludes 88 trials in which variance-covariance matrix was not positive definite.

^cExcludes 99 trials in which variance-covariance matrix was not positive definite.

CI, confidence interval; DFA, discriminant function approach; IQR, interquartile range; LRA, logistic regression with approximate maximum likelihood; ME, measurement error; OR, odds ratio; PE, processing error; SD, standard deviation; SE, standard error; MSE, mean squared error.