Impact Evaluation: Treatment Effects and Causal Analysis
by Markus Frölich (Author), Stefan Sperlich (Author)

About the Author
Markus Frölich is Director of the Center for Evaluation and Development (C4ED), Professor of Econometrics at the Universität Mannheim, Germany and J-PAL Affiliate. He has twenty years of experience in impact evaluations, including the development of new econometric methods and numerous applied impact evaluations for organisations such as Green Climate Fund, International Fund for Agricultural Development (IFAD), International Labour Organization (ILO), UNICEF, World Food Programme (WFP), and the World Bank.

About this book
In recent years, interest in rigorous impact evaluation has grown tremendously in policy-making, economics, public health, social sciences and international relations. Evidence-based policy-making has become a recurring theme in public policy, alongside greater demands for accountability in public policies and public spending, and requests for independent and rigorous impact evaluations for policy evidence. Frölich and Sperlich offer a comprehensive and up-to-date approach to quantitative impact evaluation analysis, also known as causal inference or treatment effect analysis, illustrating the main approaches for identification and estimation: experimental studies, randomization inference and randomized control trials (RCTs), matching and propensity score matching and weighting, instrumental variable estimation, difference-in-differences, regression discontinuity designs, quantile treatment effects, and evaluation of dynamic treatments. The book is designed for economics graduate courses but can also serve as a manual for professionals in research institutes, governments, and international organizations, evaluating the impact of a wide range of public policies in health, environment, transport and economic development.

Brief contents
Introduction 1
1 Basic Definitions, Assumptions and Randomised Experiments 3
    1.1 Treatment Effects: Definitions, Assumptions and Problems 3
        1.1.1 What Is a Treatment Effect? 5
        1.1.2 Formal Definitions: ATE, ATET, ATEN 10
        1.1.3 Stable-Unit-Treatment-Value Assumption 13
        1.1.4 Conditional Independence Assumption and Selection Bias 15
    1.2 Randomised Controlled Trials 17
        1.2.1 Definition and Examples for Controlled Trials 18
        1.2.2 Randomisation Methods and Statistical Properties 21
        1.2.3 Difficulties and Remedies in Practice 29
    1.3 Respecting Heterogeneity: Non-Experimental Data and Distributional Effects 33
        1.3.1 Non-Separability and Consequences 34
        1.3.2 Distributional Effects 36
    1.4 Bibliographic and Computational Notes 38
        1.4.1 Further Reading and Bibliographic Notes 38
        1.4.2 Computational Notes 40
    1.5 Exercises 41
2 An Introduction to Non-Parametric Identification and Estimation 42
    2.1 An Illustrative Approach to the Identification of Causality 43
        2.1.1 Introduction to Causal Graphs and Conditional Independence 44
        2.1.2 Back Door Identification 48
        2.1.3 Front Door Identification 51
        2.1.4 Total versus Partial Effects, Post-Treatment Covariates and Instruments 54
    2.2 Non- and Semi-Parametric Estimation 60
        2.2.1 A Brief Introduction to Non-Parametric Regression 61
        2.2.2 Extensions: Bandwidth Choice, Bias Reduction, Discrete Covariates and Estimating Conditional Distribution Functions 82
        2.2.3 A Brief Introduction to Semi-Parametric Regression 89
        2.2.4 A Note on Sieves: Series Estimators and Splines 104
    2.3 Bibliographic and Computational Notes 111
        2.3.1 Further Reading and Bibliographic Notes 111
        2.3.2 Computational Notes 112
    2.4 Exercises 113
3 Selection on Observables: Matching, Regression and Propensity Score Estimators 116
    3.1 Preliminaries: General Ideas 116
        3.1.1 How to Apply the CIA for Identification and Estimation 117
        3.1.2 Selection Bias and Common Support 121
        3.1.3 Using Linear Regression Models? 125
    3.2 ATE and ATET Estimation Based on CIA 126
        3.2.1 Definition of Matching and Regression Estimators 127
        3.2.2 Statistical Properties of Matching 129
        3.2.3 Statistical Properties of Regression Estimators 133
    3.3 Propensity Score-Based Estimator 140
        3.3.1 Propensity Score Matching 140
        3.3.2 Propensity Score Weighting 145
        3.3.3 Combination of Weighting and Regression: Double Robust Estimators 149
    3.4 Practical Issues on Matching and Propensity Score Estimation 153
        3.4.1 Summary of Estimators, Finite Sample Performance and Inference 153
        3.4.2 When Using Propensity Scores 157
        3.4.3 Testing the Validity of the Conditional Independence Assumption 162
        3.4.4 Multiple Treatment Evaluation 165
    3.5 Bibliographic and Computational Notes 168
        3.5.1 Further Reading and Bibliographic Notes 168
        3.5.2 Computational Notes 171
    3.6 Exercises 173
4 Selection on Unobservables: Non-Parametric IV and Structural Equation Approaches 175
    4.1 Preliminaries: General Ideas and LATE 175
        4.1.1 General Ideas 176
        4.1.2 Local Average Treatment Effect: LATE 180
        4.1.3 Special Cases and First Extensions 187
    4.2 LATE with Covariates 190
        4.2.1 Identification of the LATE by Conditioning 190
        4.2.2 A Feasible LATE Estimator with Confounders: The unconditional LATE 195
        4.2.3 LATE Estimation with Propensity Scores 198
        4.2.4 IV with Non-Binary Instruments 201
    4.3 Marginal Treatment Effects 204
        4.3.1 The Concept of Marginal Treatment Effects 204
        4.3.2 Relating Other Treatment Effects to MTEs 208
        4.3.3 Extensions: Identification of Distributions of Potential Outcomes and Increasing the Identification Region 211
    4.4 Non-Binary Models with Monotonicity in Choice Equation 213
        4.4.1 Continuous Treatment with Triangularity 213
        4.4.2 Ordered Discrete Treatment with Triangularity 218
    4.5 Bibliographic and Computational Notes 221
        4.5.1 Further Reading and Bibliographic Notes 221
        4.5.2 Computational Notes 223
    4.6 Exercises 224
5 Difference-in-Differences Estimation: Selection on Observables and Unobservables 227
    5.1 The Difference-in-Differences Estimator with Two Time Periods 228
        5.1.1 Diff-in-Diff, the Simple Case 229
        5.1.2 Diff-in-Diff, Conditional on Confounders 234
        5.1.3 Relationship to Linear Panel Models 237
    5.2 Multiple Groups and Multiple Time Periods 239
        5.2.1 Triple Differences and Higher Differences 242
    5.3 The Changes-in-Changes Concept 244
        5.3.1 Changes-in-Changes with Continuous Outcome Y 245
        5.3.2 Changes-in-Changes with Discrete Outcome Y and Interval Identification 251
        5.3.3 Changes-in-Changes with Discrete Outcome Y but Point Identification 255
        5.3.4 Relationship to Panel Data Analysis and Selection on Observables 257
    5.4 Bibliographic and Computational Notes 260
        5.4.1 Further Reading and Bibliographic Notes 260
        5.4.2 Computational Notes 261
    5.5 Exercises 262
6 Regression Discontinuity Design 264
    6.1 Regression Discontinuity Design without Covariates 267
        6.1.1 Identification in the Regression Discontinuity Designs 267
        6.1.2 Estimation of RDD-Based Treatment Effects 276
        6.1.3 RDD with Multiple Thresholds 281
    6.2 Regression Discontinuity Design with Covariates 285
        6.2.1 Motivations for Including Covariates 285
        6.2.2 Identification of Treatment Effect with RDD and Covariates 293
    6.3 Plausibility Checks and Extensions 299
        6.3.1 Manipulation of the Assignment Variable 299
        6.3.2 Further Diagnostic Checks for RDD 302
        6.3.3 DiD-RDD and Pseudo Treatment Tests 306
    6.4 Bibliographic and Computational Notes 309
        6.4.1 Further Reading and Bibliographic Notes 309
        6.4.2 Computational Notes 312
    6.5 Exercises 313
7 Distributional Policy Analysis and Quantile Treatment Effects 315
    7.1 A Brief Introduction to (Conditional) Quantile Analysis 317
        7.1.1 What Is Quantile Regression and Where Is It Good For? 318
        7.1.2 The Linear Regression Quantile Estimator 324
    7.2 Quantile Treatment Effects 329
        7.2.1 Quantile Treatment Effects under Selection on Observables 332
        7.2.2 Quantile Treatment Effects under Endogeneity: Instrumental Variables 335
    7.3 Quantile Treatment Effects under Endogeneity: RDD 342
    7.4 Bibliographic and Computational Notes 350
        7.4.1 Further Reading and Bibliographic Notes 350
        7.4.2 Computational Notes 352
    7.5 Exercises 353
8 Dynamic Treatment Evaluation 355
    8.1 Motivation and Introduction 355
    8.2 Dynamic Potential Outcomes Model 360
        8.2.1 Equivalence to Static Model 366
        8.2.2 Sequential Conditional Independence Assumptions 368
        8.2.3 Sequential Matching or Propensity Score Weighting 374
    8.3 Duration Models and the Timing of Treatments 377
        8.3.1 A Brief Introduction to Duration Analysis in Econometrics 378
        8.3.2 From Competing Risks Models to Treatment Effects in Duration Analysis 388
    8.4 Bibliographic and Computational Notes 396
        8.4.1 Further Reading and Bibliographic Notes 396
        8.4.2 Computational Notes 397
    8.5 Exercises 398
Bibliography 399
Index 415

Pages: 428 pages
Publisher: Cambridge University Press (May 9, 2019)
Language: English
ISBN-10: 1107616069
ISBN-13: 978-1107616066