|
Activity Number:
|
44
- Tukey (1962) and the Subsequent Sixty Years of Data Analysis: Perspectives from Government and Social Science Applications
|
|
Type:
|
Invited
|
|
Date/Time:
|
Sunday, August 7, 2022 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Government Statistics Section
|
|
Abstract #320388
|
|
|
Title:
|
Risk of Model Misspecification in Survey Estimation
|
|
Author(s):
|
Daniell Toth*
|
|
Companies:
|
U.S. Bureau of Labor Statistics
|
|
Keywords:
|
survey data;
sample design;
estimation;
official statistics
|
|
Abstract:
|
Trying to satisfy the demands for more accurate estimates at finer levels of detail is a constant occupation of the survey statistician. These efforts most always require some type of modeling of the data in order to produce estimates of areas or domains where no (or little) data was collected by borrowing strength from data in other areas or by incorporating data from outside sources that did not come from a probability sample. There has been much work done recently on choosing the best models for these purposes. This choice involves a trade-off between efficiency and bias and accepting the risk of model misspecification. Accepting the risk of bias for a big gain in efficiency is routinely done and is an easy choice for certain applications; but how appropriate is this trade-off for official statistics? We look at different model types and the risks they impose on the estimates and discuss ways of possibly identifying model misspecification and mitigating these risks.
|
Authors who are presenting talks have a * after their name.
