|
Activity Number:
|
344
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Consulting*
|
| Abstract - #301023 |
|
Title:
|
Semiparametric Estimation in Complex Surveys
|
|
Author(s):
|
Jean Opsomer*+ and Jay Breidt
|
|
Affiliation(s):
|
Iowa State University and Colorado State University
|
|
Address:
|
221 Snedecor Hall, Ames, Iowa, 50011, USA
|
|
Keywords:
|
local polynomial regression ; model-assisted estimation ; forestry survey
|
|
Abstract:
|
In large-scale surveys, design or practical considerations often result in sample data that are not available uniformly over space and time. Examples of such situations are longitudinal surveys with partially overlapping samples and multiphase surveys. Regression estimation provides a framework for combining the information available at the different time points or location scales. This approach has traditionally required parametric specification of the relationships between the variables. Breidt and Opsomer (2000) introduced local polynomial regression estimation, in which the relationship between the variables could be defined nonparametrically. In this presentation, the approach is extended to semiparametric additive models, in which some terms are modeled parametrically and others are maintained as nonparametric terms. We introduce semiparametric regression estimators for survey variables in complex surveys and describe their statistical properties, both asymptotically and through simulation experiments. We explain how to combine several models in situations with multiple phases and with partially overlapping samples. The approach is applied to data from the US Forest Service.
|
