Table 5.

Bias, 95% Confidence Interval Coverage, and Mean Squared Error for Simulation Studies^a Under 6 Scenarios for Differential Misclassification

Sensitivityb	Specificity	Sample Size	Validation Percentc	Naive			Validation			Direct Maximum Likelihood			Multiple Imputation
				Biasd	Covere	MSEf	Bias	Cover	MSE	Bias	Cover	MSE	Bias	Cover	MSE
(0.95, 0.85)	0.9	1,000	10	34	36	14	−5	96	47	−5	96	41	4	100	12
		1,000	30	34	36	14	−1	95	10	−1	95	6	1	97	6
		300	30	34	75	19	−5	96	47	−4	96	42	3	99	17
(0.70, 0.50)	0.9	1,000	10	60	4	39	−5	96	47	−5	96	46	3	98	17
		1,000	30	60	4	39	−1	95	10	−1	95	8	0	96	7
		300	30	60	47	45	−5	96	47	−5	96	49	2	98	23

Abbreviation: MSE, mean squared error.

^a Results are summarized over 10,000 simulations.

^b Sensitivity differs by exposure group; presented as (sensitivity for X = 1, sensitivity for X = 0).

^c Percent of all participants included in the validation subgroup.

^d Bias was defined as 100 times the difference between the average estimated log odds ratio and the true log odds ratio.

^e Confidence interval coverage was calculated as the percentage of simulations in which the estimated 95% Wald-type confidence limits included the true value.

^f MSE was calculated as the sum of the bias squared and the variance.