Programme Evaluation and Treatment Choice

1 Introduction.- 2 Programme Evaluation and Treatment Choice — An Overview.- 2.1 Introduction in Programme Evaluation.- 2.1.1 Potential Outcomes.- 2.1.2 Stable-Unit-Treatment-Value Assumption.- 2.1.3 Average Treatment Effects and Selection Bias.- 2.1.4 Identification of Average Treatment Effects.- 2.1.5 Estimation of Mean Counterfactual Outcomes.- 2.2 Optimal Treatment Choice.- 2.2.1 Definition of Optimal Treatment.- 2.2.2 Profiling and Targeting of Programmes in Practice.- 2.2.3 Estimating the Optimal Treatment.- 2.3 Nonparametric Regression.- 2.3.1 Nearest Neighbours and Local Polynomial Regression.- 2.3.2 Properties of Local Polynomial Regression.- 3 Nonparametric Covariate Adjustment in Finite Samples.- 3.1 Potential Efficiency Gains of Local Polynomial Matching.- 3.1.1 Simulation Results at the Optimal Bandwidth Value.- 3.1.2 Sensitivity to the Bandwidth Value.- 3.2 Approximation to the MSE and Bandwidth Choice.- 3.2.1 Bandwidth Choice.- 3.2.2 MSE Approximation of Local Polynomial Matching.- 3.2.3 Approximation Accuracy in Finite Samples.- 3.3 Data-driven Bandwidth Choice by Cross-Validation.- 3.4 Matching with Unknown Propensity Score.- 4 Semiparametric Estimation of Optimal Treatment Choices.- 4.1 Estimation of Conditional Expected Potential Outcomes.- 4.1.1 Semiparametric GMM Estimator.- 4.1.2 Monte Carlo Simulation.- 4.2 Optimal Choice and Swedish Rehabilitation Programmes.- 5 Conclusions.- A Appendix.- B Appendix.- C Appendix.- MSE-Approximation for Local Polynomial Matching.- Additional Tables to Chapter 3.- D Appendix.- D.1 Simulated Mean Squared Error for Sample Size 40.- D.2 Simulated Mean Squared Error for Sample Size 200.- D.3 Simulated Mean Squared Error for Sample Size 1000.- D.4 MSE Approximation: Kernel Matching, Sample Size 200.- D.5 MSE Approximation: Kernel Matching, Sample Size 1000.- D.6 MSE Approximation: Local Linear, Sample Size 200.- D.7 MSE Approximation: Local Linear, Sample Size 1000.- E Appendix.- E.1 Asymptotic Properties of the GMM Estimator.- E.2 Power of the J-tests — Additional Monte Carlo Results.- E.3 Additional Tables to Swedish Rehabilitation Programmes.- References.