Weisberg Bias and Causation

Models and Judgment for Valid Comparisons
1. Auflage 2010
ISBN: 978-0-470-63109-6
Verlag: John Wiley & Sons
Format: PDF
Kopierschutz: Adobe DRM (»Systemvoraussetzungen)

E-Book, Englisch, 376 Seiten, E-Book

Reihe: Wiley Series in Probability and Statistics

A one-of-a-kind resource on identifying and dealing with bias instatistical research on causal effects
Do cell phones cause cancer? Can a new curriculum increasestudent achievement? Determining what the real causes of suchproblems are, and how powerful their effects may be, are centralissues in research across various fields of study. Some researchersare highly skeptical of drawing causal conclusions except intightly controlled randomized experiments, while others discountthe threats posed by different sources of bias, even in lessrigorous observational studies. Bias and Causation presents acomplete treatment of the subject, organizing and clarifying thediverse types of biases into a conceptual framework. The booktreats various sources of bias in comparative studies--bothrandomized and observational--and offers guidance on how theyshould be addressed by researchers.
Utilizing a relatively simple mathematical approach, the authordevelops a theory of bias that outlines the essential nature of theproblem and identifies the various sources of bias that areencountered in modern research. The book begins with anintroduction to the study of causal inference and the relatedconcepts and terminology. Next, an overview is provided of themethodological issues at the core of the difficulties posed bybias. Subsequent chapters explain the concepts of selection bias,confounding, intermediate causal factors, and information biasalong with the distortion of a causal effect that can result whenthe exposure and/or the outcome is measured with error. The bookconcludes with a new classification of twenty general sources ofbias and practical advice on how mathematical modeling and expertjudgment can be combined to achieve the most credible causalconclusions.
Throughout the book, examples from the fields of medicine,public policy, and education are incorporated into the presentationof various topics. In addition, six detailed case studiesillustrate concrete examples of the significance of biases ineveryday research.
Requiring only a basic understanding of statistics andprobability theory, Bias and Causation is an excellent supplementfor courses on research methods and applied statistics at theupper-undergraduate and graduate level. It is also a valuablereference for practicing researchers and methodologists in variousfields of study who work with statistical data.
This book was selected as the 2011 Ziegel PrizeWinner in Technometrics for the best book reviewed by thejournal.
It is also the winner of the 2010 PROSE Award forMathematics from The American Publishers Awards forProfessional and Scholarly Excellence

Autoren/Hrsg.

Weisberg, Herbert I.

Inhaltsverzeichnis

Preface xi
1. What Is Bias? 1
1.1 Apples and Oranges, 2
1.2 Statistics vs. Causation, 3
1.3 Bias in the Real World, 6
Guidepost 1, 23
2. Causality and Comparative Studies 24
2.1 Bias and Causation, 24
2.2 Causality and Counterfactuals, 26
2.3 Why Counterfactuals? 32
2.4 Causal Effects, 33
2.5 Empirical Effects, 38
Guidepost 2, 46
3. Estimating Causal Effects 47
3.1 External Validity, 48
3.2 Measures of Empirical Effects, 50
3.3 Difference of Means, 52
3.4 Risk Difference and Risk Ratio, 55
3.5 Potential Outcomes, 57
3.6 Time-Dependent Outcomes, 60
3.7 Intermediate Variables, 63
3.8 Measurement of Exposure, 64
3.9 Measurement of the Outcome Value, 68
3.10 Confounding Bias, 70
Guidepost 3, 71
4. Varieties of Bias 72
4.1 Research Designs and Bias, 73
4.2 Bias in Biomedical Research, 81
4.3 Bias in Social Science Research, 85
4.4 Sources of Bias: A Proposed Taxonomy, 90
Guidepost 4, 92
5. Selection Bias 93
5.1 Selection Processes and Bias, 93
5.2 Traditional Selection Model: Dichotomous Outcome, 100
5.3 Causal Selection Model: Dichotomous Outcome, 102
5.4 Randomized Experiments, 104
5.5 Observational Cohort Studies, 108
5.6 Traditional Selection Model: Numerical Outcome, 111
5.7 Causal Selection Model: Numerical Outcome, 114
Guidepost 5, 121
Appendix, 122
6. Confounding: An Enigma? 126
6.1 What is the Real Problem? 127
6.2 Confounding and Extraneous Causes, 127
6.3 Confounding and Statistical Control, 131
6.4 Confounding and Comparability, 137
6.5 Confounding and the Assignment Mechanism, 139
6.6 Confounding and Model Specifi cation, 141
Guidepost 6, 144
7. Confounding: Essence, Correction, and Detection 145
7.1 Essence: The Nature of Confounding, 146
7.2 Correction: Statistical Control for Confounding, 172
7.3 Detection: Adequacy of Statistical Adjustment, 180
Guidepost 7, 191
Appendix, 192
8. Intermediate Causal Factors 195
8.1 Direct and Indirect Effects, 195
8.2 Principal Stratifi cation, 200
8.3 Noncompliance, 209
8.4 Attrition, 214
Guidepost 8, 215
9. Information Bias 217
9.1 Basic Concepts, 218
9.2 Classical Measurement Model: Dichotomous Outcome, 223
9.3 Causal Measurement Model: Dichotomous Outcome, 230
9.4 Classical Measurement Model: Numerical Outcome, 239
9.5 Causal Measurement Model: Numerical Outcome, 242
9.6 Covariates Measured with Error, 246
Guidepost 9, 250
10. Sources of Bias 252
10.1 Sampling, 254
10.2 Assignment, 260
10.3 Adherence, 266
10.4 Exposure Ascertainment, 269
10.5 Outcome Measurement, 273
Guidepost 10, 277
11. Contending with Bias 279
11.1 Conventional Solutions, 280
11.2 Standard Statistical Paradigm, 286
11.3 Toward a Broader Perspective, 288
11.4 Real-World Bias Revisited, 293
11.5 Statistics and Causation, 303
Glossary 309
Bibliography 321
Index 340

Über Autor(innen)

HERBERT I. WEISBERG, PhD, is founder and President of Correlation Research Inc., a consulting firm that specializes in the application of statistics to various business and legal issues. Dr. Weisberg has over forty years of statistical consulting experience and has published numerous articles related to bias assessment and reduction.