Imputation strategies for missing binary outcomes in cluster randomized trials

Table 1 Kappa statistics for different imputation strategies when missingness is completely at random

Imputation level	Imputation strategies	Percentage of missingness
		5%	10%	15%	20%	30%	50%
Within cluster	Logistic regression	0.954	0.913
	Propensity score	0.953	0.910	0.865	0.820	0.730	0.549
	MCMC¹	0.954	0.913	0.869	0.825	0.737	0.561
Across cluster	Propensity score	0.954	0.912	0.868	0.828	0.738	0.556
	Random-effects logistic regression	0.955	0.914	0.871	0.830	0.741	0.562
	Fixed-effects logistic regression	0.956	0.911	0.866	0.821	0.732	0.554
Ignore cluster	Logistic regression	0.954	0.907	0.861	0.814	0.722	0.537
	Propensity score	0.952	0.902	0.854	0.804	0.707	0.512
	MCMC¹	0.953	0.906	0.859	0.811	0.717	0.530

Note:
1. MCMC = Markov chain Monte Carlo. For the MCMC methods, we round the imputed values to 1 if it is equal or greater than 0.5 and to 0 otherwise.

ISSN: 1471-2288