. Author manuscript; available in PMC: 2018 Nov 30.

Published in final edited form as: Stat Med. 2018 Jul 18;37(27):4007–4021. doi: 10.1002/sim.7891

TABLE 1.

Estimates of the covariate-adjusted log-odds ratio for MCP-1 and spontaneous abortion, without (A) and with (B) the 30 replicate MCP-1 measurements incorporated. Values are estimated log-odds ratio (standard error), Akaike information criterion

		(A) Error Type
Method	Neither	PE only	ME only	PE and ME
LRF	0.012 (0.024), 2822.0	0.070 (0.115), 2697.6	−0.071 (−), 2762.4^b	Not identifiable
LRA	n/a^a	0.071 (0.115), 2697.5	−0.088 (−), 2762.4^b	Not identifiable
DFA	0.016 (0.025), 2277.6	0.090 (0.114), 2153.0	∞ (−), 2217.7^c	Not identifiable
		(B) Error Type
	Neither	PE only	ME only	PE and ME
LRF	n/a^d	n/a^d	0.026 (0.049), 2353.8	0.046 (0.082), 2340.8
LRA	n/a^d	n/a^d	0.026 (0.049), 2353.8	0.046 (0.082), 2340.8
DFA	n/a^d	n/a^d	0.030 (0.051), 1809.5	0.050 (0.081), 1796.5

There is no integral to approximate because assuming neither error type means MCP-1 is precisely measured.

Estimate of residual error variance in MCP-1 given covariates model hit lower bound of 0.001. Standard error omitted because variance-covariance matrix not positive definite.

Estimate of residual error variance in discriminant function model hit lower bound of 0.001, causing “blow-up” in log-OR estimate.

Cannot fit with replicates because no ME would imply that two distinct values are both the true MCP-1.

DFA, discriminant function approach; LRA, logistic regression with approximate maximum likelihood; LRF, logistic regression with full maximum likelihood; MCP-1, monocyte chemotactic protein-1; ME, measurement error; OR, odds ratio; PE, processing error.