|Title:||Model-based Multiple Imputation for Multilevel Data: Methodological Extensions and Software Enhancements|
|Principal Investigator:||Enders, Craig||Awardee:||University of California, Los Angeles|
|Program:||Statistical and Research Methodology in Education [Program Details]|
|Award Period:||3 years (07/01/19 – 06/30/22)||Award Amount:||$868,046|
|Type:||Methodological Innovation||Award Number:||R305D190002|
Co-Principal Investigators: Du, Han; Keller, Brian
The purpose of this grant was to develop, implement, and evaluate new methodologies that will substantially increase the range of substantive applications of the Blimp multiple e imputation software package. Blimp was initially developed by the research team through a previous IES grant, Multiple Imputation Procedures for Multilevel Data, and the software in its current form is freely available. Through this grant, the research team worked to expand Blimp's capabilities in two ways. First, they rendered Blimp able to accommodate a comprehensive set of not missing at random (NMAR) missing data mechanisms. Second, they further developed Blimp so that it is a general-use Bayesian analysis package.
The expansion of Blimp to account for data that are NMAR is important for education research. For example, it is quite likely that low-achieving students are more likely to opt out of state and district standardized tests, thus creating a situation where the reason for the missing value is related to a student's unobserved achievement score. Adding Bayesian analysis capabilities to Blimp allows researchers to impute missing data and run statistical analyses in the same software package, rather than having to export the imputed datasets to a different software package. The specific tasks for incorporating Bayesian statistical analysis into the software were: 1) redesigning the graphical interface to include an output window that displays basic summaries of all model parameters; 2) graphical tools such a trace plots and kernel density plots; 3) a comprehensive set of prior distributions; 4) additional convergence diagnostic measures; and 5) the deviance information criterion (DIC) for evaluating model fit. The research team provided the updated software as a free package here. In addition, the researchers disseminated research findings and software enhancements through methodological and applied peer-reviewed journal manuscripts and presentations at conferences.
Project Website: http://www.appliedmissingdata.com/multilevel-imputation
Related IES Projects: Multiple Imputation Procedures for Multilevel Data (R305D150056); Dealing with Missing Data in Educational Research: Methodological Innovations and Contemporary Recommendations (R305D220001)