Table 1.

Missing data definitions

Term	Definition
Complete Records Analysis (CRA)	Analysis is restricted to subjects who have complete data for all variables in the analysis model
Missing Completely At Random (MCAR)	The probability that data are missing is independent of the observed and missing values of variables in the analysis model, and of any related variables. Data can be MCAR if missingness is caused by a variable independent of those in the analysis model e.g. if missingness is for administrative reasons
Missing At Random (MAR)	Given the observed data, the probability that data are missing is independent of the true values of the partially observed variable. Any systematic differences between the observed and missing values can be explained by associations with the observed data
Missing Not At Random (MNAR)	If data are not MCAR nor MAR, data are said to be MNAR. The probability that data are missing depends on the (unobserved) values of the partially observed variable, even after conditioning on the observed data
Multiple Imputation (MI)	MI is a method for handling missing data. It consists of three steps: 1. An imputation model is fitted to the observed data (this is usually some form of regression model). The missing values are replaced with draws (“imputed”) from its predictive distribution (after first perturbing the model parameters). This imputation stage is carried out multiple (M) times, to give M completed datasets 2. The analysis model is fitted to each of the M completed datasets 3. The M sets of results are combined using Rubin’s rules, [5] to correctly account for the uncertainty about the missing values
Predictive Mean Matching (PMM)	PMM is an MI approach that uses an alternative method in Step 1 of the MI process: instead of imputing missing values directly from the conditional predictive distribution of the missing data given the observed data, each missing value is replaced with an observed value randomly chosen from a donor pool anchored on the conditional predicted mean
Auxiliary variable	A variable that is not in the analysis model but that is included as a predictor in the imputation model