Generic Machine Learning Inference on Heterogeneous Treatment Effects in Randomized Experiments, with an Application to Immunization in India DOI
Victor Chernozhukov,

Mert Demirer,

Esther Duflo

et al.

Published: June 1, 2018

We propose strategies to estimate and make inference on key features of heterogeneous effects in randomized experiments.These include best linear predictors the using machine learning proxies, average sorted by impact groups, characteristics most least impacted units.The approach is valid high dimensional settings, where are proxied (but not necessarily consistently estimated) predictive causal methods.We post-process these proxies into estimates features.Our generic, it can be used conjunction with penalized methods, neural networks, random forests, boosted trees, ensemble both causal.Estimation based repeated data splitting avoid overfitting achieve validity.We use quantile aggregation results across many potential splits, particular taking medians p-values other quantiles confidence intervals.We show that lowers estimation risks over a single split procedure, establish its principal inferential properties.Finally, our analysis reveals ways build provably better through learning: we objective functions develop construct effects, obtain initial step.We illustrate tools learners field experiment evaluates combination nudges stimulate demand for immunization India.

Language: Английский

Local green finance policies and corporate ESG performance DOI
Qihang Xue, Huimin Wang, Caiquan Bai

et al.

International Review of Finance, Journal Year: 2023, Volume and Issue: 23(4), P. 721 - 749

Published: May 22, 2023

Abstract Based on China's government‐business relations theory, we use difference‐in‐differences and causal forest to find that local green finance policies can significantly enhance corporate ESG performance especially for nonstate‐owned companies, companies with high levels of executive social capital, non‐heavily polluting in developed regions. We also the financing constraint mitigation effect regional environmental regulation are important mechanisms promoting performance. Additionally, strengthen positive role enhancing value, which is conducive sustainability.

Language: Английский

Citations

49
The value added of machine learning to causal inference: evidence from revisited studies DOI Creative Commons

Anna Baiardi,

Andrea A. Naghi

Econometrics Journal, Journal Year: 2024, Volume and Issue: 27(2), P. 213 - 234

Published: Feb. 6, 2024

Summary A new and rapidly growing econometric literature is making advances in the problem of using machine learning methods for causal inference questions. Yet, empirical economics has not started to fully exploit strengths these modern methods. We revisit influential studies with aiming connect theory on economics. focus double learning, forest, generic methods, context both average heterogeneous treatment effects. illustrate implementation a variety settings highlight relevance value added relative traditional used original studies.

Language: Английский

Citations

26
The “Double-Edged Sword” effect of air quality information disclosure policy—Empirical evidence based on the digital transformation of Chinese listed companies DOI
Deheng Xiao,

Xu Jinlong,

Qiyuan Li

et al.

Energy Economics, Journal Year: 2024, Volume and Issue: 133, P. 107513 - 107513

Published: March 27, 2024

Language: Английский

Citations

19
Reducing Mortality from Air Pollution in the United States by Targeting Specific Emission Sources DOI Creative Commons
Sumil K Thakrar, Srinidhi Balasubramanian, P. J. Adams

et al.

Environmental Science & Technology Letters, Journal Year: 2020, Volume and Issue: 7(9), P. 639 - 645

Published: July 15, 2020

Air quality in the United States has dramatically improved, yet exposure to air pollution is still associated with 100000–200000 deaths annually. Reducing number of effectively, efficiently, and equitably relies on attributing them specific emission sources, but so far, this been done for only highly aggregated groups or a select few sources interest. Here, we estimate mortality attributable all domestic, human-caused emissions primary PM2.5 secondary precursors. We present detailed source-specific attributions four alternate groupings relevant identifying promising ways reduce mortality. find that nearly half can be attributed just five activities, different sectors. Around fossil fuel combustion, remainder combustion nonfossil fuels, agricultural processes, other noncombustion processes. Both are important, including from currently unregulated precursor pollutants such as ammonia. suggest improvements realized by continued reductions traditionally important novel strategies reducing emerging relative importance research focus. Such changes contribute improved health outcomes environmental goals.

Language: Английский

Citations

134
Generic Machine Learning Inference on Heterogeneous Treatment Effects in Randomized Experiments, with an Application to Immunization in India DOI
Victor Chernozhukov,

Mert Demirer,

Esther Duflo

et al.

Published: June 1, 2018

We propose strategies to estimate and make inference on key features of heterogeneous effects in randomized experiments.These include best linear predictors the using machine learning proxies, average sorted by impact groups, characteristics most least impacted units.The approach is valid high dimensional settings, where are proxied (but not necessarily consistently estimated) predictive causal methods.We post-process these proxies into estimates features.Our generic, it can be used conjunction with penalized methods, neural networks, random forests, boosted trees, ensemble both causal.Estimation based repeated data splitting avoid overfitting achieve validity.We use quantile aggregation results across many potential splits, particular taking medians p-values other quantiles confidence intervals.We show that lowers estimation risks over a single split procedure, establish its principal inferential properties.Finally, our analysis reveals ways build provably better through learning: we objective functions develop construct effects, obtain initial step.We illustrate tools learners field experiment evaluates combination nudges stimulate demand for immunization India.

Language: Английский

Citations

133