PLACE:  1104 Gilman

SPEAKER:
John Eltinge
U.S. Bureau of Labor Statistics & Texas A&M -Joint work with Sunyeong Heo and Amang Sukasih

TITLE:
On the Inferential Impact of Measurement Error Adjustments in the Analysis of Complex Survey Data

ABSTRACT:

Data collected through sample surveys can be affected by several sources of
nonsampling error, including frame error, nonresponse and measurement error.
Each of these sources can lead to bias in standard point estimators, and can
also lead to degradation of standard inferential tools, e.g., confidence
intervals or test statistics.  Consequently, one frequently considers
adjustments of point estimators and inference methods that are intended to
account for nonsampling error.  The properties of adjusted point estimators,
and associated inference methods, can be affected by several factors, including
the adequacy of explicit or implicit underlying models, and information
available for identification of these models.  This talk reviews these issues
with principal emphasis on the case of measurement error. Balanced
consideration is given to:  (1) power to detect population differences that are
of practical significance; (2) degradation in
(1) caused by deviations from an assumed measurement error model; and (3) power
to identify the deviations in (2).  Some of the proposed diagnostics are
applied to data from a health survey and from an economic survey.

Key words:  Chi-square test; confidence interval coverage rates and widths;
Consumer Expenditure Survey; misclassification error; National Health
Interview Survey; power curve; total survey error.
 

COFFEE: 3:45 p.m., 104 Snedecor