Multivariate data imputation using trees.
ISSN: 1134-8984
Année de publication: 2002
Número: 5
Type: Working Paper
D'autres publications dans: Documentos de Trabajo BILTOKI
Résumé
We address the problem of completing two files with records containing a fully observed common subset of variables. The tecnique investigated involves the use of regression and/or classification trees. An extension of current methodology (the intersection-seeking or ``forest-climbing'' algorithm) is proposed to deal with multivariate response variables. The method is demonstrated and shown to be feasible and have some desirable properties.