|
Activity Number:
|
502
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 10, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #306625 |
|
Title:
|
Nonparametric Regression with Complex Survey Data
|
|
Author(s):
|
Torsten Harms*+ and Pierre Duchesne
|
|
Companies:
|
Freie Universität Berlin and Université de Montréal
|
|
Address:
|
Ebersstrasse 69, Berlin, 10827, Germany
|
|
Keywords:
|
nonparametric regression ; survey sampling ; design weights ; bandwidth
|
|
Abstract:
|
Nonparametric regression has found widespread application for data that is generated from independent and identically distributed variables. The resulting methods - particularly for bandwidth selection - do however not apply in the context of survey sampling. We will present a modification of the Nadaraya-Watson estimator that does correct for complex survey designs. Using a general superpopulation model, we can show consistency as well as derive analytical expressions for the AMSE which also depends on the sampling design. This allows for the construction of adjustment factors for the optimal bandwidth in order to correct for different inclusion probabilities of elements as well as varying sample size. An empirical study will evaluate this new estimator under different sampling designs and for different populations.
|
