Amount of missing data | Rule |
---|---|
Less than 5% | Complete case analysis will be performed, that is excluding cases with missing data. |
Between 5% and 15% | Marginal mean imputation will be performed, that is imputing the overall median or mean. |
Between 15% and 25% | Conditional mean imputation methods will be used. This involves predicting the outcome from a regression model from (linearly related) covariates. |
Above 25% | Multiple imputation will be considered. A general imputation model that uses an iterative procedure to generate imputed values will be used to generate multiple complete data sets. The model of interest will be fitted to each of the complete data sets and effect estimates combined using Rubin’s rules. |