Diagnostics for survey inference accounting for incomplete data and measurement error /
In the analysis of complex survey data, measurement errors and incomplete data can have a serious effect on the performance of large-sample inference methods. This work develops diagnostics to assess inferential performance, focusing primarily on the evaluation of hypothesis test power. Separate con...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1999.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=733671901&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | In the analysis of complex survey data, measurement errors and incomplete data can have a serious effect on the performance of large-sample inference methods. This work develops diagnostics to assess inferential performance, focusing primarily on the evaluation of hypothesis test power. Separate consideration is given to two cases, involving measurement errors in continuous and discrete data, respectively. For work with measurement errors in continuous data, we consider two related issues. First, we develop design-based estimators of the parameters of measurement error variance functions based on data from complex survey sampling. Second, these design based estimators are based on replicate measurements obtained from a limited subsample of the original sample persons. Consequently, it is important that the design based estimators account for the propensity of a given sample unit to be included in the re-measurement subsample. The proposed methods are motivated by, and illustrated with, an application to the U.S. Third National Health and Nutrition Examination Survey (NHANES III). For misclassification errors, the discrete observation version of measurement errors, we investigate the analysis of contingency tables based on data from a complex survey sample design. Specifically, this research develops methods to evaluate the power of chi-squared tests for homogeneity in the presence of misclassification errors. Results are developed for three separate cases: adjustment with a known misclassification matrix; adjustment with an estimated misclassification matrix; and no misclassification adjustment. All three cases are considered for both Wald tests and Rao-Scott adjusted tests. An extension to adjustments based on logistic regression modeling of misclassification is also considered. The proposed methods are motivated by, and illustrated with, an application to the Dual Frame National Health Interview Survey (NHIS)/ Random-Digit-Dialing (RDD) Methodology and Field Test Project. |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | xxviii, 250 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 229-232). |