處理效果的三個議題

Three Essays on Treatment Effects

賴宗志(Tsung-Chih Lai)

指導教授 ： 許育進

共同指導教授 ： 管中閔(Chung-Ming Kuan)

國立臺灣大學/社會科學院/經濟學研究所/博士(2017年)

https://doi.org/10.6342/NTU201700187

摘要

無資料

關鍵字

治療效果 ； 分位數 ； 乘數拔靴法 ； 外部有效性 ； 結構函數 ； 工具變數

並列摘要

This dissertation contains three essays on estimation and inference for the quantile treatment effects in different models and settings. The methods I propose are useful for practitioners to answer interesting questions in applied microeconomics, such as examining possible explanations for the ineffectiveness of some job training program, estimating ceteris paribus causal effect of smoking on wage distribution, and testing fertility effect on family labor supply over a continuum of quantiles. This abstract serves as a brief introduction to the proposed methods as well as the empirical studies in which the methods can be applied.

In the first essay entitled “Estimating Counterfactual Treatment Effects to Assess External Validity,” I examine the quantile treatment effect in the counterfactual environment based on a similar status quo treatment whose impact has been credibly estimated. Under the unconfoundedness and invariance of conditional distributions assumptions, the parameter of interest can be identified and nonparametrically estimated based on the kernel method. I then show the estimator converges weakly to a zero mean Gaussian process at the parametric rate, and I also propose a multiplier bootstrap to construct the corresponding uniform confidence bands. In the application, the methods are utilized to examine two possible explanations given by Eren and Ozbeklik (2014) for the ineffectiveness of Job Corps at the lower quantiles. However, the empirical findings are only in favor of the strong economic performance explanation while no evidence can be found to support the skill hypothesis.

The second essay is titled as “Distribution and Quantile Structural Functions in Treatment Effect Models: Application to Smoking Effects on Wages.” This paper aims to answer the question raised by Fortin, Lemieux, and Firpo (2011): “By how much the treatment effect would change if the covariate was switched from its original value to its average value?” Dealing with this question with the help of the structural functions enables us to distinguish the observed causal effect from different levels of characteristics. I then provide a thorough treatment to the identification, estimation and inference for the parameters of interest under unconfoundedness. I also apply the methods to investigate various smoking effects on wage distribution in the empirical study. Nevertheless, the key finding suggests that given observed characteristics fixed, the smoking wage penalty does not occur for male workers at the bottom of the individual heterogeneity distribution. This result is new to the literature and is robust regardless of which fixed value is selected.

In the third essay with the title “Estimation and Inference for Distribution Functions and Quantile Functions in Endogenous Treatment Effect Models,” I consider the setting with instrumental variable which allows us to credibly estimate the causal effects among the group of compliers, that is, individuals whose treatment status is affected by the instrument. I follow Frölich and Melly (2013) to explore the local quantile treatment effect and I derive the uniform asymptotic properties which can be seen as an extension to their pointwise result. In addition, I also propose an estimator and a simulation method for the treated compliers which is not considered in their paper. In the application I revisit the effect of fertility on family income using twin birth as a legitimate instrument. However, the result based on the uniform confidence bands cannot support the hypothesis that fertility effect is heterogeneous across all quantiles, which is in sharp contrast to the finding of Frölich and Melly (2013) based on the pointwise asymptotics.

並列關鍵字

Treatment Effect ； Quantile ； Multiplier Bootstrap ； External Validity ； Structural Function ； Instrumental Variable

參考文獻

Abadie, A. (2003): “Semiparametric Instrument Variable Estimation of Treatment Response Models,” Journal of Econometrics, 113, 231–263.

Abrevaya, J., Y.-C. Hsu and R. P. Lieli (2015): “Estimating Conditional Average Treatment Effects,” Journal of Business and Economic Statistics, 33, 485-505.

Allcott, H. (2015) “Site Selection Bias in Program Evaluation,” The Quarterly Journal of Economics, 130, 1117-1165.

Andrews, D. W. (1994): “Empirical Process Methods in Econometrics,” in Handbook of Econometrics, 4, 2247-2294.

Angrist, J. D. and W. N. Evans (1998): “Children and Their Parents’ Labor Supply: Evidence from Exogenous Variation in Family Size,” American Economic Review, 88, 450–477.