Evaluation of a weighting approach for performing sensitivity analysis after multiple imputation

Table 5 Estimates of the logistic regression coefficient (φ₁) under four analysis methods (n = 100, δ = 1); True value = 0.5

	Number of imputations (m)
	5	10	50	100	500	1000
Full dataset (before deletion)	0.524
Complete Case Analysis	0.329
Multiple Imputation under MAR	0.331	0.332	0.331	0.331	0.332	0.332
Sensitivity Analysis under MNAR	0.546	0.601	0.693	0.727	0.781	0.797

Note: The empirical Monte Carlo standard errors were all around 0.008 for MI for sensitivity analysis

ISSN: 1471-2288