Micro-Econometrics for Policy, Program, and Treatment Effects

by Lee, Myoung-jae

• ISBN13: 9780199267699
• ISBN10: 0199267693
• Format: Paperback
• Copyright: 2005-06-23
• Publisher: Oxford University Press

Summary

This title includes the following features: First book on treatmenteffects in econometrics that is suitable as a textbook; Provides a survey of theburgeoning econometric literature on treatment effects

Author Biography

Myoung-jae Lee is Associate Professor in the School of Economics and Social Sciences at Singapore Management University. He has held positions at Pennsylvania State University, Tilburg University in The Netherlands, London School of Economics, Tsukuba University in Japan and Sungkyunkwan University in Seoul.

Table of Contents

1 Tour of the book	1	(6)
2 Basics of treatment effect analysis	7	(36)
2.1 Treatment intervention, counter-factual, and causal relation	7	(4)
2.1.1 Potential outcomes and intervention	7	(2)
2.1.2 Causality and association	9	(1)
2.1.3 Partial equilibrium analysis and remarks	10	(1)
2.2 Various treatment effects and no effects	11	(5)
2.2.1 Various effects	11	(2)
2.2.2 Three no-effect concepts	13	(1)
2.2.3 Further remarks	14	(2)
2.3 Group-mean difference and randomization	16	(5)
2.3.1 Group-mean difference and mean effect	16	(2)
2.3.2 Consequences of randomization	18	(1)
2.3.3 Checking out covariate balance	19	(2)
2.4 Overt bias, hidden (covert) bias, and selection problems	21	(5)
2.4.1 Overt and hidden biases	21	(1)
2.4.2 Selection on observables and unobservables	22	(3)
2.4.3 Linear models and biases	25	(1)
2.5 Estimation with group mean difference and LSE	26	(6)
2.5.1 Group-mean difference and LSE	26	(2)
2.5.2 A job-training example	28	(2)
2.5.3 Linking counter-factuals to linear models	30	(2)
2.6 Structural form equations and treatment effect	32	(3)
2.7 On mean independence and independence	35	(3)
2.7.1 Independence and conditional independence	35	(1)
2.7.2 Symmetric and asymmetric mean-independence	36	(1)
2.7.3 Joint and marginal independence	37	(1)
2.8 Illustration of biases and Simpson's Paradox	38	(5)
2.8.1 Illustration of biases	38	(2)
2.8.2 Source of overt bias	40	(1)
2.8.3 Simpson's Paradox	41	(2)
3 Controlling for covariates	43	(36)
3.1 Variables to control for	43	(6)
3.1.1 Must cases	44	(1)
3.1.2 No-no cases	45	(1)
3.1.3 Yes/no cases	46	(1)
3.1.4 Option case	47	(1)
3.1.5 Proxy cases	48	(1)
3.2 Comparison group and controlling for observed variables	49	(7)
3.2.1 Comparison group bias	49	(2)
3.2.2 Dimension and support problems in conditioning	51	(2)
3.2.3 Parametric models to avoid dimension and support problems	53	(1)
3.2.4 Two-stage method for a semi-linear model	54	(2)
3.3 Regression discontinuity design (RDD) and before-after (BA)	56	(9)
3.3.1 Parametric regression discontinuity	56	(2)
3.3.2 Sharp nonparametric regression discontinuity	58	(3)
3.3.3 Fuzzy nonparametric regression discontinuity	61	(3)
3.3.4 Before-after (BA)	64	(1)
3.4 Treatment effect estimator with weighting	65	(7)
3.4.1 Effect on the untreated	67	(1)
3.4.2 Effects on the treated and on the population	68	(1)
3.4.3 Efficiency bounds and efficient estimators	69	(2)
3.4.4 An empirical example	71	(1)
3.5 Complete pairing with double sums	72	(7)
3.5.1 Discrete covariates	72	(2)
3.5.2 Continuous or mixed (continuous or discrete) covariates	74	(2)
3.5.3 An empirical example	76	(3)
4 Matching	79	(38)
4.1 Estimators with matching	80	(5)
4.1.1 Effects on the treated	80	(2)
4.1.2 Effects on the population	82	(2)
4.1.3 Estimating asymptotic variance	84	(1)
4.2 Implementing matching	85	(7)
4.2.1 Decisions to make in matching	85	(3)
4.2.2 Evaluating matching success	88	(2)
4.2.3 Empirical examples	90	(2)
4.3 Propensity score matching	92	(5)
4.3.1 Balancing observables with propensity score	93	(1)
4.3.2 Removing overt bias with propensity-score	93	(2)
4.3.3 Empirical examples	95	(2)
4.4 Matching for hidden bias	97	(2)
4.5 Difference in differences (DD)	99	(12)
4.5.1 Mixture of before-after and matching	99	(1)
4.5.2 DD for post-treatment treated in no-mover panels	100	(3)
4.5.3 DD with repeated cross-sections or panels with movers	103	(2)
4.5.4 Linear models for DD	105	(3)
4.5.5 Estimation of DD	108	(3)
4.6 Triple differences (TD)	111	(6)
4.6.1 TD for qualified post-treatment treated	112	(1)
4.6.2 Linear models for TD	113	(2)
4.6.3 An empirical example	115	(2)
5 Design and instrument for hidden bias	117	(30)
5.1 Conditions for zero hidden bias	117	(2)
5.2 Multiple ordered treatment groups	119	(4)
5.2.1 Partial treatment	119	(3)
5.2.2 Reverse treatment	122	(1)
5.3 Multiple responses	123	(2)
5.4 Multiple control groups	125	(4)
5.5 Instrumental variable estimator (IVE)	129	(7)
5.5.1 Potential treatments	129	(2)
5.5.2 Sources for instruments	131	(3)
5.5.3 Relation to regression discontinuity design	134	(2)
5.6 Wald estimator, IVE, and compliers	136	(11)
5.6.1 Wald estimator under constant effects	136	(2)
5.6.2 IVE for heterogenous effects	138	(1)
5.6.3 Wald estimator as effect on compliers	139	(3)
5.6.4 Weighting estimators for complier effects	142	(5)
6 Other approaches for hidden bias	147	(24)
6.1 Sensitivity analysis	147	(13)
6.1.1 Unobserved confounder affecting treatment	148	(4)
6.1.2 Unobserved confounder affecting treatment and response	152	(5)
6.1.3 Average of ratios of biased to true effects	157	(3)
6.2 Selection correction methods	160	(3)
6.3 Nonparametric bounding approaches	163	(4)
6.4 Controlling for post-treatment variables to avoid confounder	167	(4)
7 Multiple and dynamic treatments	171	(20)
7.1 Multiple treatments	171	(6)
7.1.1 Parameters of interest	172	(2)
7.1.2 Balancing score and propensity score matching	174	(3)
7.2 Treatment duration effects with time-varying covariates	177	(4)
7.3 Dynamic treatment effects with interim outcomes	181	(10)
7.3.1 Motivation with two-period linear models	181	(5)
7.3.2 G algorithm under no unobserved confounder	186	(2)
7.3.3 G algorithm for three or more periods	188	(3)
Appendix	191	(42)
A.1 Kernel nonparametric regression	191	(5)
A.2 Appendix for Chapter 2	196	(5)
A.2.1 Comparison to a probabilistic causality	196	(2)
A.2.2 Learning about joint distribution from marginals	198	(3)
A.3 Appendix for Chapter 3	201	(3)
A.3.1 Derivation for a semi-linear model	201	(1)
A.3.2 Derivation for weighting estimators	202	(2)
A.4 Appendix for Chapter 4	204	(10)
A.4.1 Non-sequential matching with network flow algorithm	204	(2)
A.4.2 Greedy non-sequential multiple matching	206	(3)
A.4.3 Nonparametric matching and support discrepancy	209	(5)
A.5 Appendix for Chapter 5	214	(7)
A.5.1 Some remarks on LATE	214	(2)
A.5.2 Outcome distributions for compliers	216	(3)
A.5.3 Median treatment effect	219	(2)
A.6 Appendix for Chapter 6	221	(5)
A.6.1 Controlling for affected covariates in a linear model	221	(3)
A.6.2 Controlling for affected mean-surrogates	224	(2)
A.7 Appendix for Chapter 7	226	(7)
A.7.1 Regression models for discrete cardinal treatments	226	(2)
A.7.2 Complete pairing for censored responses	228	(5)
References	233	(12)
Index	245