Imputation
RTI statisticians have extensive expertise in imputation, which is the substitution of appropriately selected values when data are missing.
Methodologies/Techniques
- Chi-squared automatic interaction detection (CHAID) for development of imputation classes
- Hot-deck imputation -- classical and weighted sequential methods
- Regression imputation
- Predictive mean neighborhoods (PMN)
- Logical imputation
- Mean value imputation
Capabilities
- Our statisticians have comprehensive knowledge of and experience with commonly used software such as CHAID, classification and regression trees (CART), and SAS Enterprise Miner.
- Our statisticians are capable of adopting imputation methodologies developed by other researchers.
- Before any imputation task, we conduct a detailed analysis of item nonresponse bias to determine the most effective method of imputation.
Applications
- Consistent imputation of sets of related variables
- Highly effective multivariate imputation to preserve complex associations between responses when some are missing
- Weighted sequential hot-deck imputation software to guarantee the expectation over multiple imputations is the weighted respondent mean for each imputation class
- Predictive mean neighborhoods (PMN) proprietary software combines advantages of weighted hot deck and predictive mean matching
- Software for efficiently imputing data sets with high levels of nonresponse
rti.org