|
Activity Number:
|
24
|
|
Type:
|
Other
|
|
Date/Time:
|
Sunday, August 11, 2002 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
ASA
|
| Abstract - #302033 |
|
Title:
|
To Model or Not to Model: Competing Modes of Inference for Finite Population Sampling
|
|
Author(s):
|
Roderick Little*+
|
|
Affiliation(s):
|
University of Michigan
|
|
Address:
|
1420 Washington Heights, Ann Arbor, Michigan, 48109-2029, USA
|
|
Keywords:
|
|
|
Abstract:
|
Frequentist and Bayesian paradigms of statistics have always been fiercely debated. The debate is particularly sharply drawn in the field of finite population sampling. Descriptive inferences about finite population quantities from probability samples are in large part "design-based," meaning that they are based on the randomization distribution induced by the survey design. Models for survey outcomes may be used to "assist" the choice of estimator, but not as the basis for inference. This is in sharp contrast to a Bayesian approach to surveys, where the model is the basis of inference, and random sampling plays the assisting role by conveying robustness to model misspecification. Bayes has failed to attract much of a following amongst survey practitioners. The talk will consider why and will discuss pros and cons of the design and model-based approaches to survey inference. It is suggested that a modern Bayesian approach to survey analysis that pays careful attention to the survey design can provide a unified framework for all inferential problems involving probability sampling.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2002 program |
