A Linear Representation of Regression Trees with Applications to Survey Data

• Activity Number: 338
• Type: Invited
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): John L. Eltinge* and Daniell Toth
• Companies: Bureau of Labor Statistics and Bureau of Labor Statistics
• Keywords: recursive partitioning ; sample design ; sample weights ; survey nonresponse ; incomplete data ; missing data

Abstract

We introduce a recursive algorithm to represent an arbitrary regression tree as a linear function of the splits. This representation is particularly useful for applications involving survey data. For instance, this representation makes the splits more interpretable when doing nonresponse analysis and aids checking for potential model misspecification and estimating the conditional sample variance. We demonstrate the added interpretability and usefulness with applications on Bureau of Labor Statistics survey data.