252 – SPEED: Improving Survey Data Quality with Multiple Data Sources, Administrative Data, and Nonresponse Bias Control, Part 2
Hot-Deck Imputation Cells for the American Housing Survey
Richard Griffiths
Insight Policy Research
Chrystine Tadler
Insight Policy Research
One technique the American Housing Survey uses to address missing data is hot-deck imputation. This methodology has been largely unchanged since 1998, with limited documentation surrounding the formation of imputation cells. A review of the hot-deck imputation process and assessment of various methods of creating imputation cells has led to an alternate hot-deck imputation methodology, with an improvement in preserving associations between dependent and independent variables. Techniques including stepwise logistic regression and classification and regression trees identify variables highly correlated with those being imputed, forming the basis for improved imputation cells. Classification and regression trees are also used to determine improved sort order and grouping during production.