Goodness-of-fit tests and related diagnostics for response probability models in the analysis of complex survey data /
This dissertation presents a method of goodness-of-fit test for response probability models from complex survey data that is subject to nonresponse. In survey work, nonresponse adjustment may be carried out by weighting observations according to the inverse of estimated response probabilities. Estim...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
2002.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=726458671&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | This dissertation presents a method of goodness-of-fit test for response probability models from complex survey data that is subject to nonresponse. In survey work, nonresponse adjustment may be carried out by weighting observations according to the inverse of estimated response probabilities. Estimation of these probabilities always depends to some degree on underlying models, which may use varying degrees of parametric structure. For example, one may use probability estimates based on a relatively simple logistic regression model; or probability estimates computed through weighting cells. Comparison of the probability estimates computed through these two methods leads to a quadratic form Wald type test statistic. This test can also be used to check whether a logistic regression model used to compute estimated response probabilities fits the data. An alternative by using Bonferroni test is also discussed and compared to this Wald type test. Limiting distributions of the test statistic are first derived under the basic assumption of independent and identically distributed observations, then are extended to observations obtained through a complex sample design. Special attention will be given to the variance estimation used in the test, the distribution of the test statistic, the type I error and power of the test, and the sensitivity of the test from model misspecification. Related issues involving the power of the test were studied under conditions of bias of point estimator and variance of the point estimator. Finally, some of the proposed methods are applied to data from the Third National Health and Nutrition Examination Survey (NHANES III). |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | xvi, 174 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 169-171). |