|
Abstract:
|
Loh, Eltinge, and Cho (2016) demonstrated that the Generalized, Unbiased, Interaction Detection, and Estimation (GUIDE) software package and other classification and regression tree algorithms can be used to impute missing data. Advantages of the tree algorithms for imputation are that they are less sensitive to model assumptions because they are non-parametric in nature, and that they can more easily handle a large number of variables and potentially numerous interaction terms in the imputation model. We want to expand current knowledge of the emerging tree-based imputation technique by comparing its performance under actual and artificially generated population datasets with several existing software packages, especially including AutoImpute, which was developed by Westat.
|
