2024.8.19 详细议程
会议主持：李俊青-教授 南开大学
2024.8.19 上午安排如何解决能源消费所带来的环境污染对公众健康的负面影响，是实现健康中国目标的重要挑战，而清洁能源发展则为此提供了一个可行的治理路径。本文以西气东输二线工程的投产运营作为准自然实验，利用2006~2015年的中国健康与营养调查数据（CHNS），实证考察了清洁能源发展如何影响公众健康。研究发现：西气东输工程产生了健康效应，在通过多种稳健性检验后，仍能显著提升沿线地区公众健康水平。但这一效应主要体现在城市居民和老年人群体，且家庭用能结构改善、企业污染减排、城市环境质量提升是主要作用渠道。进一步分析表明，“煤改气”政策有助于增强工程的健康效应。福利分析显示，该工程降低了个人及地区医疗支出，并促进了当地就业。本文的发现对如何深化能源供给侧结构性改革，助力实施健康中国战略，提供了鲜明的政策启示。
关键词：西气东输工程 清洁能源发展 健康效应
This talk will demonstrate how to produce informative, robust, and complex graphs using reproducible official and community-contributed routines in Stata. We will also discuss commonly used programming tools and tips for creating more engaging graphs.
To make the conventional synthetic control method more flexible to estimate the average treatment effect (ATE), this article proposes a quasi- synthetic control method for nonlinear models under the index model framework with possible high-dimensional covariates, together with a suggestion of using the minimum average variance estimation (MAVE) method to estimate parameters and the LASSO type procedure to choose high-dimensional covariates. We derive the asymptotic distribution of the proposed ATE estimators for both finite and diverging dimensions of covariates. A properly designed Bootstrap method is proposed to obtain confidence intervals and its theoretical justification is provided. When the dimension of covariates is greater than the sample size, we suggest using the robust version of sure independence screening procedure based on the distance correlation to first reduce the dimensionality and then apply the MAVE approach to estimate parameters. Finally, Monte Carlo simulation studies are conducted to examine the finite sample performance of our proposed estimators and Bootstrap procedure. In addition, an empirical application to reanalyzing data from the National Supported Work Demonstration demonstrates the practical usefulness of our proposed method.
会议主持：王群勇-教授 南开大学
2024.8.19 下午安排本文重点讨论新质生产力研究现状以及测度、评估过程中应该注意的问题。并没有进行新质生产力发展指数的实质性计算。建议国家统计局把“新质生产力发展指数”作为产品定期生产、发布。而不是由民间自行发布。
Average causal response function (ACRF) is a useful tool to assess treatment effect with dose functions, especially when the treatment is endogenous. This paper presents the identification and estimation of an ACRF with sample selection and a high dimensional controls. We derive the Ney- man orthogonal moments with multiple nuisance parameters and utilize double machine learning method and typical nonparametric techniques to estimate the proposed estimators. Asymptotics for proposed estimators are derived and Monte Carlo simulations demonstrate their good finite sample properties. Our identification and estimation results could be readily extended to the case with more complex sample selection mechanisms. We apply the proposed method to US Job Corps data to evaluate the heterogeneous effect of residential components, which yields new insights for policy makers.
作为最重要的准实验因果推断方法之一，断点回归设计有两大分析框架，二者无论在前提假定、带宽选择还是推断方法上均有相当差异。其中，基于连续性的框架假定潜在结果的条件期望连续，在实证研究中广泛应用。局部随机化的框架则为后起之秀，该框架假定在断点附近的小窗口，驱动变量可视为随机分配。本讲座将介绍这两大框架的原理与技术，包括识别、估计、推断，并通过蒙特卡罗模拟与Stata案例比较二者的差异，以及应用前景。
The interaction effect in endogenous probit model with an interaction term is consistently estimated in Zhou and Li (2021). However, the estimation and test are time-consuming when the sample size is large. In this article, a new Stata command, eivprobit, is developed to implement Zhou-Li’s method with much less time. Besides, the marginal effects of the two interacted regressors and the quadratic effect of a regressor with a squared term can also be estimated by the command. The eivprobit estimation is based on the control function approach and the standard errors of the estimated effects are obtained by nonparametric bootstrapping. Moreover, the finite sample Monto Carlo simulation shows that the estimator of the interaction effect behaves well and better than the usual methods such as Ai and Norton (2003)’s estimator ignoring endogeneity or the coefficient estimator of the interaction term in IV-probit estimation.
Mediation models with censored outcomes play a crucial role in social and medical sciences. However, the inherent censoring characteristics of the data often lead existing models to rely on assumptions of linearity, homogeneity, and normality for estimation. Unfortunately, these assumptions may not align with the complexities of real-world problems, limiting the persuasiveness of causal analyses. In this study, we investigate causal mediation analysis within a counterfactual framework by framing it as a neural style transfer problem commonly encountered in image processing. Acknowledging the impressive capabilities of generative adversarial networks (GANs) in handling neural style transfer, we propose a novel GAN-based model named generative adversarial censored mediation network to address mediation issues under our concern. Our model employs rectified linear unit (ReLU) activation function and designs a particular multi-channel network structure to implement the censored outcome mechanism while accommodating multiple mediators. To guide our model in accurately learning the underlying data patterns, we also develop a novel min-max optimization problem. Leveraging the strengths of GANs, our model fundamentally relaxes the stringent assumptions present in traditional models, resulting in more precise estimations of mediation effects and promising inference outcomes, especially in the context of intricate data patterns. Through unique insights and techniques, this study illustrates how generative learning methods can serve as an effective and robust approach for diverse causal mediation problems. We substantiate our claims with numerical results obtained from synthetic and realistic datasets, showcasing the superior performance of our method.
2024.8.20 详细议程
会议主持：李宝伟-副教授 南开大学
2024.8.20 上午安排You can use treatment-effects estimators to draw causal inferences from observational data. You can use lasso when you want to control for many potential covariates. With standard treatment-effects models, there is an intrinsic conflict between two required assumptions. The conditional independence assumption is likely to be satisfied with many variables in the model, while the overlap assumption is likely to be satisfied with fewer variables in the model. This presentation shows how to overcome this conflict by using Stata's telasso command. telasso estimates the average treatment effects with high-dimensional controls while using lasso for model selection. This estimator is robust to model-selection mistakes. Moreover, it is doubly robust, so only one of the outcome or treatment model needs to be correctly specified.
过去三十多年间，项目评估计量经济学经历了长足的发展，同时得益于计量分析软件技术的不断进步和数据资料的日益丰富，经济学实证研究范式乃至整个经济学的研究范式发生了巨大转变，深刻影响了经济学的教学和研究。过去五年间，围绕DID, IV, RD等主流项目评估计量经济学方法，又出现了不少新的计量理论进展，一方面对原有的理论方法和应用实践做了修补和完善，另一方面也推动了项目评估计量经济学方法进一步向前发展。我将对这些进展做一个概括性的介绍。
同行效应（或邻居效应）模型是研究个体之间相互影响的重要模型，其设定与空间计量模型相类似。但在空间计量模型中，邻接矩阵往往被视作外生的。如果邻接矩阵不是地理网络而是社会或经济网络，那么外生性假定是不合理的。本文提出了异质性同行效应模型和内生性同行效应模型的Stata估计指令，snreghnet和snregenet。snreghnet可以考察模型的行异质性和列异质性。snregenet计算模型的两阶段工具变量估计，并采用野蛮自举计算标准误差。
会议主持：颜冠鹏-青年讲师 山东财经大学
2024.8.20 下午安排Li et al. (2024) 扩展了用于估计和推断交互固定效应面板模型处理效应的因子化方法 (the factor-based approach)。本演讲介绍了Stata新命令——xtteifeci，该命令可逐期生成处理效应的置信区间和p值，且支持多种模型设定，包括模型中包含协变量和/或非平稳趋势等。最后，以经典案例详细介绍该命令的具体操作。
随着信息技术的飞速发展和全球经济一体化的深入，数字经济已经成为推动全球经济增长的重要引擎。然而，由于不同地区的资源禀赋、经济基础、政策支持等因素存在差异，数字经济的发展水平在不同地区呈现出明显的空间异质性。因此，对数字经济发展水平的空间异质性进行深入分析，具有重要的必要性和现实意义。这不仅有助于我们全面了解数字经济在全国乃至全球范围内的分布状况和发展趋势，还有助于揭示数字经济与区域经济发展之间的互动关系、识别数字经济在区域间的发展差距和潜在风险，以及推动经济的可持续发展。
随着时代的发展与技术的进步，一般性的统计数据得到了广泛的应用。与此同时，以文本形式存在的非结构化数据也正在逐渐成为经管实证领域的中坚力量。值得注意的是，广大研究者在进行文本分析时通常会优先考虑使用Python等工具。但从Stata到Python的工具迁移往往伴随不小的学习成本。在这种情况下我们不禁会想，是否可以使用Stata做文本分析内容？本主题旨在介绍经管领域主流的文本分析方法，并探讨使用Stata进行这些文本分析的可能性与局限性。
当一个单位的处置也会影响其他单位的结果时的干扰情况下，传统的因果推断的SUTVA假定被违反，当干扰起作用时，政策评估主要依赖于在集群干扰和二元处理下的随机实验的假设。相反，我们考虑在连续治疗和网络干扰下的非实验处置。具体来说，我们通过将网络处置的暴露程度定义为通过物理、社会或经济互动连接的单位所接受的处理的加权平均值来定义溢出效应。在Forastiere等人(2021)的基础上，我们提供了一个基于广义倾向得分的估计量来估计连续处置的直接和溢出效应。我们的估计量还允许考虑以不同强度为特征的非对称网络连接。本演讲介绍了一个Stata新命令，该命令结合Mathematica的优点，采用线性回归和机器学习的方法估计个体倾向得分和邻域倾向得分，支持多种模型设定，采用广义线性模型估计结果模型和剂量反应函数（ADRF），并采用自举法计算标准误差。最后，以一个实际的案例详细介绍该命令的具体操作。
Stata官方命令merge和joinby可以进行1:1、m:1、1:m和m:m的数据横向合并。但这两个命令需要将使用数据集保存到硬盘中，不仅增加了时间成本并可能产生大量中间文件，影响效率。Stata的数据框（frame）功能允许用户在内存中同时操作多个数据集，无需保存数据到硬盘，但其frlink和frget命令仅支持1:1和m:1的合并类型，不支持1:m和m:m。通常的解决方法是将使用数据框的数据保存到硬盘，再借助merge和joinby命令将数据合并到主数据框中。显然，当需要合并的数据量较大或涉及的数据集较多时，这种方式的效率很低。本演讲介绍一个新命令——framerge，以解决上述问题。framerge命令不仅支持多种数据合并关系（包括1:1, m:1, 1:m和m:m），还能够在内存中直接操作数据，无需读写硬盘，从而提高了数据处理的效率。最后以案例详细介绍该命令的具体操作。倾向得分，支持多种模型设定，采用广义线性模型估计结果模型和剂量反应函数（ADRF），并采用自举法计算标准误差。最后，以一个实际的案例详细介绍该命令的具体操作。
会议官网链接
