|
Abstract:
|
A stacked multiple imputation approach stacks the M imputed data sets into a single data set in order to obtain parameter estimates. Due to challenges in standard error estimation with the stacked approach, a pooled multiple imputation approach is instead usually used, where estimates are obtained for each imputed data set and combined using Rubin’s rules. We explore missing data situations where a stacked approach may have potential advantages, such as small sample sizes, categorical outcomes, or when bootstrapping is of interest. We consider several different simulation scenarios and multiple methods for obtaining standard errors for the stacked data estimates.
|
