EVALUATION OF SOCIAL PROGRAMS IN
THE PRESENCE OF SELECTION BIAS
Motakis
Efthimios
Supervisor: I. Panaretos
CHAPTER
1
INTRODUCTION TO SOCIAL PROGRAM EVALUATION
1.1 Introduction
1.2 The Research Framework
1.3 Evaluating Social Programs with Nonrandom Samples1.4 Evaluating Social Programs
MATHEMATICAL AND THEORETICAL BACKGROUND
2.1 Introduction
2.2 A Mathematical Description of the Evaluation Process
2.3 Useful Notation2.4 Evaluation Process
2.5 Other Evaluation Questions
2.6 Solutions to Evaluation Problem
2.7 Discriminating Sample Selection Bias
STATISTICAL APPROACHES TO THE EVALUATION PROBLEM:
RANDOMIZED SOCIAL EXPERIMENTS
3.1 Introduction
3.2 Historical Review
3.3 Experimental Designs
3.4 A Description of the Experimental Procedure in Evaluation Studies
3.5 The Evaluation Problem in Randomized Experiments
3.5.1 Solution of the Evaluation Problem
3.5.2 Elimination of Selection Bias
3.5.3 Identification Assumptions
3.6 Stages for Randomization
3.7 Identification of the Impact Distribution
3.7.1 Distribution Identification Assumptions - Response Patterns
3.8 Ethical Issues
3.9 Evaluation Under Dropouts
3.10 The Case For and Against Social Experimentation
3.10.1 Advantages of Experiments
3.10.2 Limitations of Experiments
STATISTICAL APPROACHES TO THE EVALUATION PROBLEM:
MATCHING METHODS
4.1 Introduction
4.2 The Idea of Matching Methods
4.3 Conditions of Matching
4.4 Matching Measures
4.5 Matching Estimators
4.6 Other Methods
4.7 Identification of the Impact Distribution
4.8 Discussion on Matching
ECONOMETRIC APPROACHES TO THE EVALUATION PROBLEM:
STRUCTURAL MODELS
EXTENSION TO LONGITUDINAL DATA5.1 Introduction
5.2 Historical Review
5.3 Useful Notation
5.3.1 The Evaluation Problem
5.3.2 Description of the Structural Approach
5.3.3 Definition of Mean Impacts
5.4 The Conventional Selection Bias Model (Tobit Type II Censoring)
5.5 Parametric Methods
5.5.1 Maximum Likelihood Estimation
5.5.2 2-step Estimation
5.6 Dummy Endogenous Variable Model
5.7 Properties of the Estimators
5.8 Testing the Normality Assumption
5.9 Relaxation of Normality
5.10 Instrumental Variable Estimation
5.10.1 Definition of the IV Estimator
5.10.2 Criticisms
5.10.3 The Local Instrumental Variable Effect
5.11 Calculation of Mean Parameters Using Heckman's 2-step Procedure
5.12 Calculation of ATE Using the Method of Bounding
5.13 Semi-Parametric Methods
5.13.1 Maximum Likelihood Estimation
5.13.2 Heckman's 2-step Estimation
5.13.2.1 Estimation of the Selection Model
5.13.2.2 Estimation of the Primary Equation
5.13.2.3 Identification of the Primary's Equation Intercept
5.14 Sample Selection Models With Alternative Censoring Rules
5.14.1 Other Tobit Type Censoring Rules
5.14.2 Multiple Alternatives - Ordered Censoring Rules
5.14.3 Multiple Alternatives - Unordered Censoring Rules (Polychotomous Model)
5.14.4 Censoring Rules Based on Multiple Indices
5.14.5 Other Types of Censoring
5.15 The EM Algorithm
5.16 Identification of the Distribution of Impacts
5.17 Discussion on the Non-Experimental Methods
INTRODUCTION TO DISCRETE CHOICE MODELS6.1 Introduction
6.2 Panel Data - The Experimental Case
6.2.1 The Before-After Estimator
6.2.2 The Difference-in-Differences Estimator
6.3 Panel Data - The Non-Experimental Case
6.3.1 Maximum Likelihood Estimators
6.3.2 2-step Estimators
6.3.2.1 Some Early Approaches
6.3.2.2 Specification Tests
6.3.2.3 Modern Approaches
6.4 Evaluation Under Dropouts
6.5 Discussion on the Panel Estimators
CONCLUSIONS AND TOPICS FOR FURTHER RESEARCH7.1 Discussion
7.2 Multinomial Logit Model
7.3 Multinomial Probit Model
8.1 Concluding Remarks
8.2 Topics for Further Research