skills.rest
xjtulyc
xjtulycCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

ml-econometrics

Estimate causal treatment effects with Post-LASSO and DML cross-fitting.

33|6|Updated Mar 17, 2026
Search Tags:#causal-inference#aipw#ml-econometrics#post-lasso#doubleml#dml-plr#causalforest
One-click install
npx skills add https://github.com/xjtulyc/awesome-rosetta-skills --skill ml-econometrics
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ml-econometrics
Source: https://github.com/xjtulyc/awesome-rosetta-skills/tree/main/skills/07-economics/ml-econometrics
Command: npx skills add https://github.com/xjtulyc/awesome-rosetta-skills --skill ml-econometrics

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires doubleml, econml, scikit-learn, numpy, pandas.

What problem does it solve?

This Skill helps you estimate causal treatment effects when the set of controls is high-dimensional, where naive regression or plain LASSO can produce biased estimates and invalid standard errors.

Core Features & Use Cases

  • Post-LASSO + valid inference: selects controls via Post-LASSO and then estimates treatment effects with OLS for more reliable inference in high-dimensional settings (e.g., Belloni-Chernozhukov-Hansen approach).
  • Double/Debiased Machine Learning (DML) for ATE/ATT: uses cross-fitting with nuisance models (E[Y|X] and E[D|X]) to reduce regularization bias and recover valid, asymptotically normal inference.
  • Heterogeneous effects via CausalForest (CATE): estimates conditional average treatment effects and surfaces feature importance for effect heterogeneity.
  • Doubly robust AIPW estimator: provides a doubly-robust ATE estimator with cross-fitting and propensity clipping to mitigate extreme weights.
  • Econometric use cases: program evaluation, policy impact analysis, and difference-in-differences–style workflows where rich covariates are required and causal identification depends on correct nuisance estimation.

Quick Start

Use the ml-econometrics skill to estimate the treatment effect (ATE) and uncertainty from your dataset by running Post-LASSO and then applying DML PLR with cross-fitting on the same outcome, treatment, and control variables.

Frequently Asked Questions about ml-econometrics

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I estimate causal treatment effects with high-dimensional controls without biased standard errors?▼

To estimate causal treatment effects with high-dimensional controls without biased standard errors, apply Post-LASSO for variable selection followed by OLS, or use Double/Debiased Machine Learning with cross-fitting. These approaches reduce regularization bias and recover valid, asymptotically normal inference for reliable effect estimates.

What is the doubly robust AIPW estimator and how does it handle extreme propensity scores?▼

The doubly robust AIPW estimator combines nuisance models for outcome and treatment to estimate average treatment effects, applying propensity score clipping to mitigate extreme weights. It uses cross-fitting to produce valid standard errors and confidence intervals even when one nuisance model is misspecified.

How do I estimate heterogeneous treatment effects using CausalForest?▼

To estimate heterogeneous treatment effects using CausalForest, train the model on your outcome, treatment, and covariates to estimate conditional average treatment effects (CATE). The method surfaces feature importance to identify drivers of effect heterogeneity across different subgroups in your data.

Can I use Double Machine Learning for program evaluation and policy impact analysis?▼

Yes, you can use Double Machine Learning for program evaluation and policy impact analysis. The DML PLR estimator with cross-fitting reduces regularization bias in high-dimensional settings, producing valid effect estimates and confidence intervals required for rigorous econometric policy evaluation.

Do I need doubleml and econml libraries to run DML and CausalForest estimations?▼

Yes, you need the doubleml and econml libraries along with scikit-learn, numpy, and pandas to run DML and CausalForest estimations. These dependencies provide the necessary cross-fitting, nuisance modeling, and causal forest implementations to produce effect estimates with standard errors.

Why does naive LASSO produce biased causal estimates in high-dimensional settings?▼

Naive LASSO produces biased causal estimates in high-dimensional settings because regularization introduces shrinkage bias that contaminates treatment effect estimation. Post-LASSO addresses this by selecting variables first and estimating effects via OLS, while DML uses cross-fitting to eliminate regularization bias and recover valid inference.

Related Skills

Explore More Education & Research Skills→
andikarachmanandikarachman
14

data-preprocessing

Preprocess raw data into clean, analysis-ready datasets with pandas and numpy.

Community
Advanced
sheehesheehe

ml-causal

Estimate ATE and CATE from observational data using DML and causal forests.

Community
Advanced
robotlearning123robotlearning123

scikit-learn

Build, evaluate, and refine machine learning models with scikit-learn.

Community
Advanced
tondevreltondevrel
19

dowhy

Estimate causal effects from observational data using the identify-estimate-refute framework.

Community
Advanced