Measurement error and misclassification in statistics and epidemiology : impacts and Bayesian adjustments / Paul Gustafson.

Format:

Book

Author/Creator:

Gustafson, Paul, 1968-

Series:

Interdisciplinary statistics

Interdisciplinary statistics series

Language:

English

Subjects (All):

Error analysis (Mathematics).

Sequential analysis.

Bayesian statistical decision theory.

Bias.

Medical Subjects:

Bias.

Physical Description:

x, 188 pages : illustrations ; 25 cm.

Place of Publication:

Boca Raton, Fla. : Chapman & Hall/CRC, [2004]

Summary:

Mismeasurement of explanatory variables is a common hazard when using statistical modeling techniques, and particularly so in fields such as biostatistics and epidemiology where perceived risk factors cannot always be measured accurately. With this perspective and a focus on both continuous and categorical variables, Measurement Error and Misclassification in Statistics and Epidemiology: Impacts and Bayesian Adjustments examines the consequences and Bayesian remedies in those cases where the explanatory variable cannot be measured with precision. The author explores both measurement error in continuous variables and misclassification in discrete variables, and shows how Bayesian methods might be used to allow for mismeasurement. A broad range of topics, from basic research to more complex concepts such as "wrong-model" fitting, make this a useful research work for practitioners, students and researchers in biostatistics and epidemiology."

Contents:

1.1 Examples of Mismeasurement 1

1.2 The Mismeasurement Phenomenon 3

2 The Impact of Mismeasured Continuous Variables 9

2.1 The Archetypical Scenario 9

2.3 Multiplicative Measurement Error 15

2.4 Multiple Mismeasured Predictors 18

2.5 What about Variability and Small Samples? 21

2.6 Logistic Regression 24

2.7 Beyond Nondifferential and Unbiased Measurement Error 26

3 The Impact of Mismeasured Categorical Variables 35

3.1 The Linear Model Case 35

3.3 Inferences on Odds-Ratios 39

3.4 Logistic Regression 44

3.5 Differential Misclassification 46

3.6 Polychotomous Variables 47

4 Adjusting for Mismeasured Continuous Variables 51

4.1 Posterior Distributions 51

4.2 A Simple Scenario 56

4.3 Nonlinear Mixed Effects Model: Viral Dynamics 62

4.4 Logistic Regression I: Smoking and Bladder Cancer 69

4.5 Logistic Regression II: Framingham Heart Study 72

4.6 Issues in Specifying the Exposure Model 74

4.7 More Flexible Exposure Models 79

4.8 Retrospective Analysis 83

4.9 Comparison with Non-Bayesian Approaches 88

5 Adjusting for Mismeasured Categorical Variables 99

5.1 A Simple Scenario 99

5.2 Partial Knowledge of Misclassification Probabilities 104

5.3 Dual Exposure Assessment 111

5.4 Models with Additional Explanatory Variables 119

6 Further Topics 139

6.1 Dichotomization of Mismeasured Continuous Variables 139

6.2 Mismeasurement Bias and Model Misspecification Bias 147

6.3 Identifiability in Mismeasurement Models 152

Appendix Bayes-MCMC Inference 165

A.1 Bayes Theorem 165

A.2 Point and Interval Estimates 168

A.3 Markov Chain Monte Carlo 169

A.4 Prior Selection 176

A.5 MCMC and Unobserved Structure 177.

Notes:

Includes bibliographical references (pages [179]-186) and index.

ISBN:

1584883359

OCLC:

52587710