Data Analysis with R, Second Edition - Second Edition

Multiple imputation

The big idea behind multiple imputation is that instead of generating one set of imputed data with our best estimation of the missing data, we generate multiple versions of the imputed data where the imputed values are drawn from a distribution. The uncertainty about what the imputed values should be is reflected in the variation between the multiple imputed datasets.

We perform our intended analysis separately with each of these m amounts of completed datasets. These analyses will then yield m different parameter estimates (like regression coefficients, and so on). The critical point is that these parameter estimates are different solely due to the variability in the imputed missing values, and hence, our uncertainty ...