Imputing the Missing Y’s: Implications for Survey Producers and Survey Users
Imputing the Missing Y’s: Implications for Survey Producers and Survey Users
Key takeaways
This is such an obnoxiously odd article. Why did we need an article published to tell us the obvious that imputation should include the dependent variable of interest...
Bibliography: Young, R., Johnson, D.R., 2011. Imputing the Missing Y’s: Implications for Survey Producers and Survey Users. Proceedings of the AAPOR Conference Abstracts.
Authors:: Rebekah Young, David R Johnson
Collections:: Methods
First-page:
Multiple imputation is a popular technique used to handle item-level missing data. Recent studies, however, have generated serious concerns about the best practices for statistical analysis with an imputed dependent variable. We use an example from observed data to examine three multiple imputation strategies: (1) excluding the dependent variable from the imputation model, (2) multiple imputation then deletion, and (3) including the dependent variable in the imputation model and retaining the imputed values in the subsequent analysis. Consistent with previous research, our results suggest that the dependent variable should be included in the imputation model. Under conditions where it is most practical to do so, survey users may be able to retain the imputed values in their analysis, provided that a sufficient number of datasets was generated.
content: "@youngImputingMissingImplications2011" -file:@youngImputingMissingImplications2011
Reading notes
Annotations
(08/05/2024, 21:46:01)
“Multiple imputation is a popular technique used to handle item-level missing data. Recent studies, however, have generated serious concerns about the best practices for statistical analysis with an imputed dependent variable” (Young and Johnson, 2011, p. 6242)
“Consistent with previous research, our results suggest that the dependent variable should be included in the imputation model. Under conditions where it is most practical to do so, survey users may be able to retain the imputed values in their analysis, provided that a sufficient number of datasets was generated.” (Young and Johnson, 2011, p. 6242)