Table 4.

Performance of methods for regression of (dichotomous) distress at Wave II with missing baseline distress

% with complete data	Method	Diet (η1 = 0.754)				Distress (η2 = 6.260)				Health (η6 = 1.132)
		Bias	SE	StdBias	Coverage	Bias	SE	StdBias	Coverage	Bias	SE	StdBias	Coverage
90%	CCA	−0.115	0.257	−0.449	92.6%	0.021	0.197	0.106	95.0%	0.007	0.115	0.065	95.0%
	MVNI-skew0	0.002	0.226	0.008	93.9%	−0.003	0.199	−0.013	95.3%	0.005	0.106	0.048	95.1%
75%	CCA	−0.209	0.319	−0.653	91.9%	0.013	0.227	0.059	93.70%	0.006	0.133	0.047	95.4%
	MVNI-skew0	−0.007	0.233	−0.030	95.8%	−0.043	0.230	−0.187	94.6%	0.004	0.108	0.039	94.3%
50%	CCA	−0.243	0.485	−0.501	96.3%	0.046	0.306	0.149	94.5%	0.008	0.180	0.043	94.0%
	MVNI-skew0	0.014	0.251	0.057	95.9%	−0.082	0.304	−0.270	94.6%	0.008	0.115	0.066	95.6%
25%	CCA	−0.120	0.833	−0.144	82.4%	0.091	0.487	0.188	79.4%	0.021	0.286	0.073	80.0%
	MVNI-skew0	0.021	0.308	0.068	80.6%	−0.155	0.447	−0.348	78.5%	0.002	0.132	0.012	80.5%

Measures of performance are mean values in the estimation of the β parameters from Equation 1 across the 1000 simulated datasets of 1000 observations (compared to the true values from the synthetic population of 971,327). CCA = Complete Case Analysis; MVNI = Multivariate normal imputation; StdBias = standardised bias; SE = average (estimated) standard error across the 1000 datasets.

Note in this example it was not possible to include an analysis where only 10% of the data were complete due to a large number of datasets with zero counts in the cross-tabulation of diet and distress at Wave II.