Skip to main content


Fig. 4 | BMC Medical Research Methodology

Fig. 4

From: Multiple imputation methods for handling missing values in a longitudinal categorical variable with restrictions on transitions over time: a simulation study

Fig. 4

Absolute bias, Relative bias (%), Coverage (%), and Mean square error for complete case analysis (CCA), available case analysis (ACA), indicator based imputation using multivariate normal imputation with projected distance-based rounding (indicator-PDBR), imputation as a continuous variable using multivariate normal imputation with calibration (continuous-calibration), and predictive mean matching (PMM) for handling increasing proportions of missing data (0.45, 0.65), for the parameter estimates for ex-smokers relative to never-smokers for the simulation study, when data are a) missing completely at random; b) missing at random (weak); c) missing at random (strong). Results are not shown for fully conditional specification with multinomial and ordinal logistic imputation and two-fold fully conditional specification methods because the imputation models failed to converge in some or all of the simulations. Minimal differences were observed between the results of predictive mean matching with 5 and 10 nearest observations. Therefore, only the results for this method with 5 nearest observations are presented. Complete case analysis and available case analysis are presented under without restrictions for comparison purposes only

Back to article page