Methods for significance testing of categorical covariates in logistic regression models after multiple imputation: power and applicability analysis

Table 1 Type I error for each pooling method for simulated data with beta equal to zero for 25% missing data in varying correlations between the variables where Factor1 is categorical and Covar1-Covar4 are continuous

Cor		Full data	RR	MR	CHI	VAR	MPR_in	MPR_out
0.2	Factor	0.057	^a	0.019	0.024	0.018	0.065	0.038
	covar1	0.056	0.048	0.052	0.057	0.048	0.104	0.028
	covar2	0.056	0.056	0.057	0.057	0.056	0.061	0.059
	covar3	0.043	0.051	0.057	0.055	0.051	0.063	0.052
	covar4	0.070	0.058	0.061	0.063	0.058	0.070	0.052
0.4	Factor	0.057	^a	0.020	0.026	0.025	0.065	0.035
	covar1	0.056	0.046	0.048	0.052	0.046	0.094	0.030
	covar2	0.056	0.051	0.054	0.054	0.051	0.060	0.056
	covar3	0.043	0.059	0.063	0.061	0.059	0.070	0.052
	covar4	0.070	0.056	0.057	0.056	0.056	0.057	0.055
0.6	Factor	0.061	^a	0.026	0.026	0.026	0.068	0.023
	covar1	0.058	0.048	0.049	0.051	0.048	0.088	0.031
	covar2	0.065	0.055	0.057	0.060	0.055	0.066	0.056
	covar3	0.059	0.051	0.054	0.053	0.051	0.058	0.051
	covar4	0.063	0.063	0.066	0.066	0.063	0.075	0.064
0.8	Factor	0.057	^a	0.026	0.026	0.025	0.077	0.033
	covar1	0.056	0.057	0.058	0.058	0.057	0.098	0.019
	covar2	0.056	0.058	0.060	0.061	0.058	0.063	0.055
	covar3	0.043	0.060	0.061	0.061	0.060	0.070	0.043
	covar4	0.070	0.053	0.052	0.054	0.053	0.062	0.059

^aFor the categorical variable the p-value could not be obtained by RR; Cor correlation between variables; Full data complete data; RR Rubin’s Rules, MR Meng and Rubin pooling, CHI chi-square test with multiple degrees of freedom, VAR pooled sampling variance method, MPR _in Median P Rule with the outcome included in model, MPR _out Median P Rule with the outcome excluded from model

ISSN: 1471-2288