Advised missing data method

No longitudinal analysis > MNAR > <50% missing data > <75% item missings

The methods that are presented and discussed on this website (e.g., multiple imputation), assume a Missing at Random (MAR) mechanism. The MNAR mechanism is more strict and multiple imputation might not always work for missing data that is MNAR. However, when additional auxiliary variables are included in the missing data handling, to make the MAR assumption more plausible, multiple imputation can give valid analysis results. The missing data method should be applied to the the item scores in order to use the available item information. In this situation multiple imputation of the item scores can be performed and was shown to work well even in this MNAR missing data mechanism (Eekhout et al., 2014). However, the percentage of subjects with missing values cannot be too large and auxiliary variables that might partly explain the missing data or correlate to the variables with missing data should be included in the imputation model.