Econometrics Skill

Causal inference for tabular data — 17 estimators from OLS to RDD, packaged as an AI coding-agent skill.

中文说明

What this is

A skill (structured prompt + bundled library) that gives AI coding agents the ability to run defensible causal-inference analyses on user-provided datasets. It covers the standard applied-econometrics toolkit:

The skill includes method selection guidance, automatic uploaded-data diagnostics, data preprocessing checklists, covariance/SE mapping tables, known library caveats, interpretation templates, an end-to-end project workflow, diagnostic checklists, result-table helpers, and an advanced applied workflow.

Supported agent frameworks

• Claude Code / OpenClaw — reads SKILL.md directly
• OpenAI/Codex skill UI metadata — see agents/openai.yaml; this is not a standalone OpenAI Agents SDK config
• Any framework that can read a Markdown skill definition

Repository

• GitHub: https://github.com/xiaomihu1992/econometrics-skill
• SSH: [email protected]:xiaomihu1992/econometrics-skill.git

Usage

# Clone the repository
git clone [email protected]:xiaomihu1992/econometrics-skill.git
cd econometrics-skill

# Install Python dependencies
pip install -r requirements.txt

# Run a smoke example
python examples/ols_example.py

For Claude Code or OpenClaw-style skill loading, install or copy the whole econometrics-skill/ folder into the environment’s skill directory, then invoke $econometrics in a causal-inference or uploaded-data diagnostic task. The main entrypoint is SKILL.md; detailed method signatures live in references/method_details.md.

Project structure

econometrics-skill/
├── SKILL.md                              # Main skill definition (English)
├── SKILL.zh-CN.md                        # 中文版 skill 定义
├── lib/
│   ├── econometric_algorithm.py          # 17 causal-inference functions
│   ├── data_preprocess.py                # load_table + uploaded-data diagnostics
│   └── result_tables.py                  # compact model comparison tables
├── references/
│   ├── method_selection.md               # Decision tree for picking the right estimator
│   ├── method_details.md                 # Exact signatures and minimal code per function
│   ├── interpretation.md                 # Reporting templates per method family
│   ├── applied_project_workflow.md       # End-to-end empirical project workflow
│   ├── data_preprocessing_advice.md      # Automatic dataset diagnostics and cleaning advice
│   ├── diagnostic_checklist.md           # Method-specific diagnostic checklist
│   ├── result_tables.md                  # Result table helper guide
│   └── advanced_applied_workflow.md      # Advanced applied workflow
├── examples/
│   ├── ols_example.py                    # OLS: schooling → wages
│   ├── psm_example.py                    # PSM + IPW with bootstrap
│   ├── iv_example.py                     # IV/2SLS: Card-style college → earnings
│   ├── did_example.py                    # Static DID: minimum wage panel
│   ├── did_event_study_example.py        # Staggered DID + event study
│   ├── rdd_example.py                    # Sharp RDD: merit scholarship
│   ├── rdd_fuzzy_example.py              # Fuzzy RDD: housing subsidy
│   ├── data_diagnostic_example.py        # Uploaded-data diagnostics and cleaning advice
│   └── skill_usage_example.zh-CN.md      # 中文使用示例：DID 政策评估
├── agents/
│   └── openai.yaml                       # OpenAI/Codex skill UI metadata
├── requirements.txt
├── LICENSE
└── README.md

Dependencies

• Python 3.10+
• numpy, pandas, matplotlib, statsmodels, linearmodels, scipy, openpyxl, xlrd

Practical helpers

• references/applied_project_workflow.md guides users from question to final report.
• lib/data_preprocess.py can inspect uploaded CSV/Excel data and produce cleaning advice.
• references/data_preprocessing_advice.md explains how agents should use the diagnostic output.
• references/diagnostic_checklist.md gives method-specific checks before causal interpretation.
• lib/result_tables.py builds compact model comparison tables for statsmodels and linearmodels results.

Known limitations

The library functions have several documented caveats (IV SE bias, non-standard AIPW, event-study lead/lag assumptions, etc.). These are fully documented in SKILL.md § “Known library limitations” and in references/method_details.md. The skill instructs the agent to disclose these to users when relevant.

License

MIT

---

中文说明

面向表格数据的因果推断 — 17 个估计器，从 OLS 到 RDD，打包为 AI 编程助手的 skill。

这是什么

一个结构化的 skill（提示词 + 算法库），让 AI 编程助手能够在用户提供的数据集上运行可辩护的因果推断分析。涵盖标准应用计量工具箱：

• OLS：带控制变量的回归
• 倾向得分方法：PS 构建、可视化、匹配（PSM）、逆概率加权（IPW）、PS 回归、AIPW、IPW-RA
• 工具变量：IV / 2SLS + 诊断检验
• 双重差分：静态 DID、交错 DID、事件研究 + 可视化
• 断点回归：Sharp RDD、Fuzzy RDD（两步法 + 全局多项式）

仓库地址与使用说明

• GitHub: https://github.com/xiaomihu1992/econometrics-skill
• SSH: [email protected]:xiaomihu1992/econometrics-skill.git

git clone [email protected]:xiaomihu1992/econometrics-skill.git
cd econometrics-skill
pip install -r requirements.txt
python examples/ols_example.py

在 Claude Code 或 OpenClaw 这类支持 skill 的环境中，把整个 econometrics-skill/ 目录安装或复制到 skill 目录，然后在因果推断、数据诊断或政策评估任务中调用 $econometrics。主入口是 SKILL.md，具体函数签名见 references/method_details.md。

快速开始

pip install -r requirements.txt
cd examples
python ols_example.py

深度模式

• 快速模式：核心三步工作流（识别策略 → 数据检查 → 结果解释）
• 项目工作流模式：研究问题 → 变量表 → 识别策略 → 诊断 → 稳健性 → 最终报告
• 数据诊断模式：自动分析上传数据并给出清洗、变量角色和建模准备建议
• 结果表模式：整理多个模型的系数、标准误、显著性标记和拟合统计量
• 诊断清单模式：在因果解释前检查各方法的关键风险
• 高级应用模式：高级研究流程（estimand → 识别备忘录 → 诊断 → 稳健性 → 异质性）

详细中文说明见 SKILL.zh-CN.md，使用示例见 examples/skill_usage_example.zh-CN.md。