|
Activity Number:
|
357
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #304016 |
|
Title:
|
An Adaptive Method for Collapsing Strata Using Regression Trees on Data from a Complex Design
|
|
Author(s):
|
Daniell Toth*+ and John L. Eltinge
|
|
Companies:
|
Bureau of Labor Statistics and Bureau of Labor Statistics
|
|
Address:
|
Office of Surey Methods Research, Washington, DC, 20212,
|
|
Keywords:
|
Recursive partition ; Degrees of freedom ; Sample surveys ; Design based inference ; Inferential power ; National Automotive Sampling System
|
|
Abstract:
|
When analyzing survey data it is often helpful to collapse strata in order to allow approximation of design-based variance estimators in cases that include strata with only one selected PSU and to improve the stability of these estimators. This work proposes an adaptive method for collapsing strata. For this we use recursively partition the data using strata indicator variables to build a regression tree. Important features of the proposed method include: a simple function representation of the tree; design based estimation of the variance of the tree coefficients; and use of backward elimination to collapse cells. We apply this method to data from the 2006 National Automotive Sampling System as an illustration.
|
