C Bo 9781139025751 A 006

Contents
Preface
PA RT I
page xvii
I N T RO D U C T I O N
Causality: The Basic Framework

1.1
Introduction
1.2
Potential Outcomes
1.3
Definition of Causal Effects
1.4
Causal Effects in Common Usage
1.5
Learning about Causal Effects: Multiple Units
1.6
The Stable Unit Treatment Value Assumption
1.7
The Assignment Mechanism: An Introduction
1.8
Attributes, Pre-Treatment Variables, or Covariates
1.9
Potential Outcomes and Lords Paradox
1.10 Causal Estimands
1.11 Structure of the Book
1.12 Samples, Populations, and Super-Populations
1.13 Conclusion
Notes
A Brief History of the Potential Outcomes Approach to

Causal Inference
2.1
Introduction
2.2
Potential Outcomes and the Assignment Mechanism
before Neyman
2.3
Neymans (1923) Potential Outcome Notation in Randomized
Experiments
2.4
Earlier Hints for Physical Randomizing
2.5
Fishers (1925) Proposal to Randomize Treatments to Units
2.6
The Observed Outcome Notation in Observational Studies for
Causal Effects
2.7
Early Uses of Potential Outcomes in Observational Studies in
Social Sciences
3
3
3
5
7
8
9
13
15
16
18
20
20
21
21
23
23
24
25
26
26
27
28
vii
Downloaded from Cambridge Books Online by IP 193.205.23.67 on Mon Feb 15 08:45:09 GMT 2016.
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139025751
Cambridge Books Online Cambridge University Press, 2016
viii
Contents
2.8
Potential Outcomes and the Assignment Mechanism in

Observational Studies: Rubin (1974)
Notes
A Classification of Assignment Mechanisms

3.1
Introduction
3.2
Notation
3.3
Assignment Probabilities
3.4
Restrictions on the Assignment Mechanism
3.5
Assignment Mechanisms and Super-Populations
3.6
Randomized Experiments
3.7
Observational Studies: Regular Assignment Mechanisms
3.8
Observational Studies: Irregular Assignment Mechanisms
3.9
Conclusion
Notes
29
30
31
31
33
34
37
39
40
41
42
43
43
PA RT I I C L A S S I C A L R A N D O M I Z E D E X P E R I M E N T S
A Taxonomy of Classical Randomized Experiments

4.1
Introduction
4.2
Notation
4.3
Bernoulli Trials
4.4
Completely Randomized Experiments
4.5
Stratified Randomized Experiments
4.6
Paired Randomized Experiments
4.7
Discussion
4.8
Conclusion
Notes
47
47
48
48
50
51
52
53
55
56
Fishers Exact P-Values for Completely Randomized Experiments

5.1
Introduction
5.2
The Paul et al. Honey Experiment Data
5.3
A Simple Example with Six Units
5.4
The Choice of Null Hypothesis
5.5
The Choice of Statistic
5.6
A Small Simulation Study
5.7
Interval Estimates Based on Fisher P-Value Calculations
5.8
Computation of P-Values
5.9
Fisher Exact P-Values with Covariates
5.10 Fisher Exact P-Values for the Honey Data
5.11 Conclusion
Notes
57
57
59
59
63
64
72
74
75
78
80
81
81
Neymans Repeated Sampling Approach to Completely

Randomized Experiments
6.1
Introduction
6.2
The Duflo-Hanna-Ryan Teacher-Incentive Experiment Data
6.3
Unbiased Estimation of the Average Treatment Effect
83
83
84
85
Contents
6.4
6.5
6.6
6.7
6.8
6.9
6.10
ix
The Sampling Variance of the Neyman Estimator

Estimating the Sampling Variance
Confidence Intervals and Testing
Inference for Population Average Treatment Effects
Neymans Approach with Covariates
Results for the Duflo-Hanna-Ryan Teacher-Incentive Data
Conclusion
Notes
Appendix A Sampling Variance Calculations
Appendix B Random Sampling from a Super-Population
87
92
95
98
101
102
104
104
105
109
Regression Methods for Completely Randomized Experiments

7.1
Introduction
7.2
The LRC-CPPT Cholesterol Data
7.3
The Super-Population Average Treatment Effects
7.4
Linear Regression with No Covariates
7.5
Linear Regression with Additional Covariates
7.6
Linear Regression with Covariates and Interactions
7.7
Transformations of the Outcome Variable
7.8
The Limits on Increases in Precision Due to Covariates
7.9
Testing for the Presence of Treatment Effects
7.10 Estimates for LRC-CPPT Cholesterol Data
7.11 Conclusion
Notes
Appendix
113
113
115
116
118
122
125
127
128
129
131
133
134
135
Model-Based Inference for Completely Randomized Experiments

8.1
Introduction
8.2
The Lalonde NSW Experimental Job-Training Data
8.3
A Simple Example: Naive and More Sophisticated Approaches
to Imputation
8.4
Bayesian Model-Based Imputation in the Absence of Covariates
8.5
Simulation Methods in the Model-Based Approach
8.6
Dependence between Potential Outcomes
8.7
Model-Based Imputation with Covariates
8.8
Super-Population Average Treatment Effects
8.9
A Frequentist Perspective
8.10 Model-Based Estimates of the Effect of the NSW Program
8.11 Conclusion
Notes
Appendix A Posterior Distributions for Normal Models
Appendix B Analytic Derivations with Known Covariance
Matrix
141
141
144
Stratified Randomized Experiments

9.1
Introduction
9.2
The Tennesee Project Star Data
146
150
163
165
169
171
172
174
177
177
178
181
187
187
188
Contents
9.3
9.4
9.5
9.6
9.7
9.8
9.9
10
11
The Structure of Stratified Randomized Experiments

Fishers Exact P-Values in Stratified Randomized Experiments
The Analysis of Stratified Randomized Experiments from
Neymans Repeated Sampling Perspective
Regression Analysis of Stratified Randomized Experiments
Model-Based Analysis of Stratified Randomized Experiments
Design Issues: Stratified versus Completely Randomized
Experiments
Conclusion
Notes
Appendix A: Student-Level Analyses
Appendix B: Proofs of Theorems 9.1 and 9.2
Pairwise Randomized Experiments

10.1 Introduction
10.2 The Childrens Television Workshop Experiment Data
10.3 Pairwise Randomized Experiments
10.4 Fishers Exact P-Values in Pairwise Randomized Experiments
10.5 The Analysis of Pairwise Randomized Experiments from
Neymans Repeated Sampling Perspective
10.6 Regression-Based Analysis of Pairwise Randomized
Experiments
10.7 Model-Based Analysis of Pairwise Randomized Experiments
10.8 Conclusion
Notes
Appendix: Proofs
Case Study: An Experimental Evaluation of a Labor Market
Program
11.1 Introduction
11.2 The San Diego SWIM Program Data
11.3 Fishers Exact P-Values
11.4 Neymans Repeated Sampling-Based Point Estimates and
Large-Sample Confidence Intervals
11.5 Regression-Based Estimates
11.6 Model-Based Point Estimates
11.7 Conclusion
Notes
189
192
201
205
207
211
212
212
213
214
219
219
220
220
222
224
229
231
233
234
234
240
240
240
242
245
247
250
253
253
PA RT I I I R E G U L A R A S S I G N M E N T M E C H A N I S M S : D E S I G N
12
Unconfounded Treatment Assignment

12.1 Introduction
12.2 Regular Assignment Mechanisms
12.3 Balancing Scores and the Propensity Score
12.4 Estimation and Inference
12.5 Design Phase
257
257
258
266
268
276
Contents
12.6
12.7
13
14
15
16
xi
Assessing Unconfoundedness
Conclusion
Notes
278
279
279
Estimating the Propensity Score

13.1 Introduction
13.2 The Reinisch et al. Barbituate Exposure Data
13.3 Selecting the Covariates and Interactions
13.4 Choosing the Specification of the Propensity Score for
the Barbituate Data
13.5 Constructing Propensity-Score Strata
13.6 Choosing Strata for the Barbituate Data
13.7 Assessing Balance Conditional on the Estimated
Propensity Score
13.8 Assessing Covariate Balance for the Barbituate Data
13.9 Conclusion
Notes
Appendix: Logistic Regression
281
281
284
285
Assessing Overlap in Covariate Distributions

14.1 Introduction
14.2 Assessing Balance in Univariate Distributions
14.3 Direct Assessment of Balance in Multivariate Distributions
14.4 Assessing Balance in Multivariate Distributions Using the
Propensity Score
14.5 Assessing the Ability to Adjust for Differences in Covariates
by Treatment Status
14.6 Assessing Balance: Four Illustrations
14.7 Sensitivity of Regression Estimates to Lack of Overlap
14.8 Conclusion
Notes
309
309
310
313
Matching to Improve Balance in Covariate Distributions

15.1 Introduction
15.2 The Reinisch et al. Barbituate Exposure Data
15.3 Selecting a Subsample of Controls through Matching to
Improve Balance
15.4 An Illustration of Propensity Score Matching with Six
Observations
15.5 Theoretical Properties of Matching Procedures
15.6 Creating Matched Samples for the Barbituate Data
15.7 Conclusion
Notes
337
337
339
Trimming to Improve Balance in Covariate Distributions

16.1 Introduction
16.2 The Right Heart Catheterization Data
16.3 An Example with a Single Binary Covariate
359
359
360
362
288
290
294
296
300
306
306
307
314
317
318
332
336
336
339
344
345
349
358
358
xii
Contents
16.4
16.5
16.6
Selecting a Subsample Based on the Propensity Score

The Optimal Subsample for the Right Heart
Catheterization Data
Conclusion
Notes
366
368
373
374
PA RT I V R E G U L A R A S S I G N M E N T M E C H A N I S M S : A N A LY S I S
17
18
19
Subclassification on the Propensity Score

17.1 Introduction
17.2 The Imbens-Rubin-Sacerdote Lottery Data
17.3 Subclassification on the Propensity Score and Bias Reduction
17.4 Subclassification and the Lottery Data
17.5 Estimation Based on Subclassification with Additional
Bias Reduction
17.6 Neymanian Inference
17.7 Average Treatment Effects for the Lottery Data
17.8 Weighting Estimators and Subclassification
17.9 Conclusion
Notes
377
377
378
380
385
Matching Estimators
18.1 Introduction
18.2 The Card-Krueger New Jersey and Pennsylvania Minimum
Wage Data
18.3 Exact Matching without Replacement
18.4 Inexact Matching without Replacement
18.5 Distance Measures
18.6 Matching and the Card-Krueger Data
18.7 The Bias of Matching Estimators
18.8 Bias-Corrected Matching Estimators
18.9 Matching with Replacement
18.10 The Number of Matches
18.11 Matching Estimators for the Average Treatment Effect for the
Controls and for the Full Sample
18.12 Matching Estimates of the Effect of the Minimum Wage
Increase
18.13 Conclusion
Notes
401
401
A General Method for Estimating Sampling Variances for

Standard Estimators for Average Causal Effects
19.1 Introduction
19.3 Estimands
386
388
390
392
399
399
404
405
407
410
412
415
416
424
425
427
428
430
431
433
433
435
436
Contents
xiii
19.4
The Common Structure of Standard Estimators for Average

Treatment Effects
19.5 A General Formula for the Conditional Sampling Variance
19.6 A Simple Estimator for the Unit-Level Conditional Sampling
Variance
19.7 An Estimator for the Sampling Variance of Conditional on
Covariates
19.8 An Estimator for the Sampling Variance for the Estimator for the
Average Effect for the Treated
19.9 An Estimator for the Sampling Variance for the Population
Average Treatment Effect
19.10 Alternative Estimators for the Sampling Variance
19.11 Conclusion
Notes
20
Inference for General Causal Estimands

20.1 Introduction
20.2 The Lalonde NSW Observational Job-Training Data
20.3 Causal Estimands
20.4 A Model for the Conditional Potential Outcome
Distributions
20.5 Implementation
20.6 Results for the Lalonde Data
20.7 Conclusion
Notes
PA RT V
441
445
446
452
452
454
456
460
460
461
461
462
465
468
472
473
474
474
R E G U L A R A S S I G N M E N T M E C H A N I S M S : S U P P L E M E N TA RY
A N A LY S E S
21
Assessing Unconfoundedness
21.1 Introduction
21.2 Setup
21.3 Estimating Effects on Pseudo-Outcomes
21.4 Estimating Effects of Pseudo-Treatments
21.5 Robustness to the Set of Pre-Treatment Variables
21.7 Conclusion
Notes
479
479
482
482
485
487
490
495
495
22
Sensitivity Analysis and Bounds

22.1 Introduction
22.3 Bounds
22.4 Binary Outcomes: The Rosenbaum-Rubin
Sensitivity Analysis
496
496
497
497
500
xiv
Contents
22.5
22.6
PA RT V I
Binary Outcomes: The Rosenbaum Sensitivity Analysis

for P-Values
Conclusion
Notes
506
509
509
REGULAR ASSIGNMENT MECHANISMS WITH

N O N C O M P L I A N C E : A N A LY S I S
23
24
25
Instrumental Variables Analysis of Randomized Experiments with

One-Sided Noncompliance
23.1 Introduction
23.2 The Sommer-Zeger Vitamin A Supplement Data
23.3 Setup
23.4 Intention-to-Treat Effects
23.5 Compliance Status
23.6 Instrumental Variables
23.7 Moment-Based Instrumental Variables Estimators
23.8 Linear Models and Instrumental Variables
23.9 Naive Analyses: As-Treated, Per Protocol,
and Unconfoundedness
23.10 Conclusion
Notes
Appendix
Instrumental Variables Analysis of Randomized Experiments
with Two-Sided Noncompliance
24.1 Introduction
24.2 The Angrist Draft Lottery Data
24.3 Compliance Status
24.4 Intention-to-Treat Effects
24.5 Instrumental Variables
24.6 Traditional Econometric Methods for Instrumental
Variables
24.7 Conclusion
Notes
Model-Based Analysis in Instrumental Variable Settings:
Randomized Experiments with Two-Sided Noncompliance
25.1 Introduction
25.2 The McDonald-Hiu-Tierney Influenza Vaccination Data
25.3 Covariates
25.4 Model-Based Instrumental Variables Analyses for Randomized
Experiments with Two-Sided Noncompliance
25.5 Simulation Methods for Obtaining Draws from the Posterior
Distribution of the Estimand Given the Data
25.6 Models for the Influenza Vaccination Data
513
513
516
517
519
522
526
530
531
535
539
539
541
542
542
543
544
546
548
556
559
559
560
560
561
567
568
574
578
Contents
xv
25.7
25.8
PA RT V I I
26
Results for the Influenza Vaccination Data

Conclusion
Notes
581
584
584
CONCLUSION
Conclusions and Extensions

Notes
References
Author Index
Subject Index
589
590
591
605
609

C Bo 9781139025751 A 006

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

C Bo 9781139025751 A 006

Hochgeladen von

Copyright:

Verfügbare Formate

Contents

Causality: The Basic Framework

A Brief History of the Potential Outcomes Approach to

Potential Outcomes and the Assignment Mechanism in

A Classification of Assignment Mechanisms

A Taxonomy of Classical Randomized Experiments

Fishers Exact P-Values for Completely Randomized Experiments

Neymans Repeated Sampling Approach to Completely

The Sampling Variance of the Neyman Estimator

Regression Methods for Completely Randomized Experiments

Model-Based Inference for Completely Randomized Experiments

Stratified Randomized Experiments

The Structure of Stratified Randomized Experiments

Pairwise Randomized Experiments

Unconfounded Treatment Assignment

Estimating the Propensity Score

Assessing Overlap in Covariate Distributions

Matching to Improve Balance in Covariate Distributions

Trimming to Improve Balance in Covariate Distributions

Selecting a Subsample Based on the Propensity Score

Subclassification on the Propensity Score

A General Method for Estimating Sampling Variances for

The Common Structure of Standard Estimators for Average

Inference for General Causal Estimands

Sensitivity Analysis and Bounds

Binary Outcomes: The Rosenbaum Sensitivity Analysis

REGULAR ASSIGNMENT MECHANISMS WITH

Instrumental Variables Analysis of Randomized Experiments with

Results for the Influenza Vaccination Data

Conclusions and Extensions

Das könnte Ihnen auch gefallen