causal-inference-guide
安装方式
先用 Prompt 核对来源,也可切换到安装命令或下载可用的安装包。
用 AI 助手安装
复制提示词,让助手先核对固定版本和依赖。
查看 Prompt收起 Prompt
先检查当前 Agent 是否已安装 SkillPI Skill Finder。如果尚未安装,请按官方仓库的说明安装并启用:https://github.com/Findskills/skillpi。然后使用 Skill Finder 打开此 SkillPI 详情页并定位 "causal-inference-guide":https://skillpi.com/zh/skills/deepskill-555a11283de9f19c6781b1bdf24ead7d4e162bd8a64656b2dd6d97fcd86a047c。安装前核对上游来源和版本:https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/blob/bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba/skills/43-wentorai-research-plugins/skills/analysis/econometrics/causal-inference-guide/SKILL.md。来源不一致时停止并说明差异。核对通过后,用 Skill Finder 安装目标 Skill,并报告安装路径和来源版本。
此命令从 GitHub 来源仓库获取指定技能。运行前请先查看源文件。
npx skills add https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/tree/bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba/skills/43-wentorai-research-plugins/skills/analysis/econometrics/causal-inference-guide --skill causal-inference-guide来源与配置
查看 GitHub 源码brycewang-stanford/Auto-Empirical-Research-Skills固定提交 bfa0a6e5a843babbf45a1ef1663fd5dc4f07c0ba
配置方式待核实
请查看原始来源中的适用工具、依赖和权限要求。
安装实测待补充
本条目面向文献与学术工具场景,提供“causal-inference-guide”技能的目录入口。公开标签:learning、因果推断、双重差分、工具变量。请通过固定上游版本核对实际功能、依赖与权限;此页为 Skillπ 独立编写的用途介绍和原站链接,不代表运行安全保证。原文与附件依各自许可开放。
文件浏览
---
name: causal-inference-guide
description: "Causal inference methods including DiD, IV, RDD, and synthetic control"
metadata:
openclaw:
emoji: "🔗"
category: "analysis"
subcategory: "econometrics"
keywords: ["difference-in-differences", "DiD", "causal inference", "instrumental variables", "IV estimation", "regression discontinuity"]
source: "wentor"
---
# Causal Inference Guide
A skill for applying quasi-experimental causal inference methods in observational research. Covers difference-in-differences, instrumental variables, regression discontinuity designs, and synthetic control methods with implementation code and diagnostic checks.
## Difference-in-Differences (DiD)
### Classic Two-Period DiD
```python
import numpy as np
import pandas as pd
import statsmodels.formula.api as smf
def did_estimation(df: pd.DataFrame, outcome: str, treatment: str,
post: str, covariates: list[str] = None) -> dict:
"""
Estimate a difference-in-differences model.
Args:
df: Panel DataFrame
outcome: Name of outcome variable column
treatment: Name of treatment group indicator (0/1)
post: Name of post-treatment period indicator (0/1)
covariates: Optional list of control variable names
"""
# Create interaction term
df = df.copy()
df['did'] = df[treatment] * df[post]
# Build formula
formula = f"{outcome} ~ {treatment} + {post} + did"
if covariates:
formula += ' + ' + ' + '.join(covariates)
model = smf.ols(formula, data=df).fit(cov_type='cluster',
cov_kwds={'groups': df.get('unit_id', df.index)})
return {
'did_estimate': model.params['did'],
'se': model.bse['did'],
'p_value': model.pvalues['did'],
'ci_95': (model.conf_int().loc['did', 0], model.conf_int().loc['did', 1]),
'r_squared': model.rsquared,
'n_obs': model.nobs,
'interpretation': (
f"The treatment effect is {model.params['did']:.3f} "
f"(SE = {model.bse['did']:.3f}, p = {model.pvalues['did']:.4f}). "
f"{'Statistically significant' if model.pvalues['did'] < 0.05 else 'Not significant'} "
f"at the 5% level."
)
}
```
### Parallel Trends Test
The key identifying assumption. Test it with pre-treatment data:
```python
def test_parallel_trends(df: pd.DataFrame, outcome: str,
treatment: str, time: str,
treatment_period: int) -> dict:
"""
Test the parallel trends assumption using event study specification.
"""
df = df.copy()
pre_periods = sorted(df[df[time] < treatment_period][time].unique())
# Create period dummies interacted with treatment
for t in pre_periods:
df[f'pre_{t}'] = ((df[time] == t) & (df[treatment] == 1)).astype(int)
period_vars = [f'pre_{t}' for t in pre_periods[:-1]] # omit last pre-period (reference)
formula = f"{outcome} ~ {' + '.join(period_vars)} + C({time}) + C(unit_id)"
model = smf.ols(formula, data=df).fit()
# Joint F-test: all pre-treatment interactions = 0
f_test = model.f_test(' = '.join([f'{v} = 0' for v in period_vars]))
return {
'pre_period_coefficients': {v: model.params[v] for v in period_vars},
'f_statistic': f_test.fvalue[0][0],
'f_pvalue': f_test.pvalue,
'parallel_trends_hold': f_test.pvalue > 0.05,
'interpretation': (
'Parallel trends assumption supported (cannot reject joint null)'
if f_test.pvalue > 0.05
else 'WARNING: Parallel trends assumption may be violated'
)
}
```
## Instrumental Variables (IV)
### Two-Stage Least Squares
```python
from linearmodels.iv import IV2SLS
def iv_estimation(df: pd.DataFrame, outcome: str, endogenous: str,
instrument: str, exogenous: list[str] = None) -> dict:
"""
Estimate an IV model using 2SLS.
Args:
outcome: Dependent variable
endogenous: Endogenous regressor
instrument: Instrumental variable
exogenous: List of exogenous control variables
"""
exog_formula = '1'
if exogenous:
exog_formula += ' + ' + ' + '.join(exogenous)
model = IV2SLS(
dependent=df[outcome],
exog=df[exogenous] if exogenous else None,
endog=df[[endogenous]],
instruments=df[[instrument]]
).fit(cov_type='robust')
# First-stage F-statistic
first_stage = smf.ols(f"{endogenous} ~ {instrument}", data=df).fit()
f_stat = first_stage.fvalue
return {
'iv_estimate': model.params[endogenous],
'se': model.std_errors[endogenous],
'p_value': model.pvalues[endogenous],
'first_stage_F': f_stat,
'weak_instrument': f_stat < 10, # Stock-Yogo rule of thumb
'interpretation': (
f"IV estimate: {model.params[endogenous]:.3f}. "
f"First-stage F = {f_stat:.1f} "
f"({'Strong' if f_stat >= 10 else 'WEAK'} instrument)."
)
}
```
### IV Diagnostic Checklist
1. **Relevance**: First-stage F > 10 (Stock & Yogo, 2005)
2. **Exclusion restriction**: Instrument affects outcome only through the endogenous variable (untestable, argue conceptually)
3. **Overidentification test**: Sargan/Hansen J-test when you have more instruments than endogenous variables
## Regression Discontinuity Design (RDD)
```python
def rdd_estimation(df: pd.DataFrame, outcome: str, running_var: str,
cutoff: float, bandwidth: float = None) -> dict:
"""
Sharp regression discontinuity design estimation.
"""
df = df.copy()
df['centered'] = df[running_var] - cutoff
df['treated'] = (df[running_var] >= cutoff).astype(int)
if bandwidth is None:
bandwidth = df['centered'].std() # simple default
# Restrict to bandwidth
local = df[df['centered'].abs() <= bandwidth]
# Local linear regression
formula = f"{outcome} ~ treated * centered"
model = smf.ols(formula, data=local).fit(cov_type='HC1')
return {
'rdd_estimate': model.params['treated'],
'se': model.bse['treated'],
'p_value': model.pvalues['treated'],
'bandwidth': bandwidth,
'n_obs': len(local),
'n_treated': local['treated'].sum(),
'n_control': len(local) - local['treated'].sum()
}
```
## Best Practices
- Always visualize your data: plot outcome trends over time (DiD), first-stage relationships (IV), or running variable distributions (RDD)
- Report robustness checks: varying bandwidths, alternative specifications, placebo tests
- Use cluster-robust standard errors at the appropriate level (usually the treatment unit level)
- Be transparent about identifying assumptions and potential violations
- Pre-register your analysis plan when possible to avoid p-hacking concerns
版本历史
-
v2026.07
GitHub ↗
原版更新说明
## Highlights - **First tagged release.** Every gate is green: `make check` (validate + 183 unit tests + eval-harness + numeric benchmarks), validate-catalog, quality-evals, and OpenSSF Scorecard on `main`. - **Rigor double loop at 13 method families.** Each family closes taxonomy tag → eval scenario (does the agent avoid the method's known traps?) → deterministic numeric benchmark (does it recover the true effect?). New this cycle: heterogeneous treatment effects (CATE) and quantile treatment effects (QTE). - **Paper-WorkFlow flagship at 29/29 executable rigor gates**, now merged to its `main` with a drift-gated README badge — the credibility artifact is checked by CI, so it can never silently go stale ([RIGOR.md](https://github.com/brycewang-stanford/Paper-WorkFlow/blob/main/RIGOR.md)). - **1,150 vendored skills / 69 collections**, all license-audited, security-scanned, tracked in a machine-readable catalog with CI-enforced freshness, and importable as a single whole-repo router skill. - **Hardened weekly upstream syncs**: synced trees must pass the full gate before merge; this cycle both sync PRs were verified, patched (StatsPAI's SkillOpt execution-gate card restored), and merged. - **First-party plugin marketplace** (Python / R / Stata empirical-analysis plugins + guide) shipped in the previous cycle, now under a tagged version. ## Known Follow-Ups - Ecosystem benchmark comparison (AERS vs econometrics agents on the shared benchmark suite) — [PLAN-2026-07 week 3](https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills/blob/main/docs/PLAN-2026-07.md). - Decomposition (Oaxaca–Blinder) and shift-share (Bartik) method families (weeks 3–4). - GitHub Pages catalog site, README demo screenshots, five-language README consistency script (week 4). ## Catalog - Top-level collections: **69** - `SKILL.md` files: **1150** - Cataloged agent tools: **335** ([`tools/CATALOG.md`](../tools/CATALOG.md)) ## Methodological rigor - Method families in the coverage map: **13** ([`RIGOR_COVERAGE.md`](RIGOR_COVERAGE.md)) - Eval scenarios: **26** (138 rubric items, 122 auto-checkable) ([`../eval-harness/`](../eval-harness/README.md)) - Benchmark tasks: **13** (70 gold checks, 58 required) ([`../benchmark/`](../benchmark/README.md)) ### Method families (skills tagged) - `iv`: 27 - `rdd`: 25 - `panel-fe`: 19 - `staggered-did`: 17 - `bayesian`: 13 - `did`: 10 - `event-study`: 10 - `matching`: 10 - `synthetic-control`: 10 - `cate`: 8 - `dml`: 8 - `survival`: 3 - `quantile`: 2 ### Eval scenarios by category - `causal-identification`: 14 - `citation-hygiene`: 1 - `reproducibility`: 2 - `research-integrity`: 1 - `runtime-safety`: 1 - `statistical-validity`: 2 - `writing-compliance`: 3 - `writing-style`: 2 ## License buckets - AGPL-3.0: 1 - Apache-2.0: 1 - CC-BY-NC-4.0 (non-commercial): 1 - CC-BY-SA-4.0: 1 - CC-BY-SA-4.0 (repository default): 4 - GPL-3.0: 1 - MIT: 35 - MIT Non-Commercial: 1 - Mixed (per-skill; see folder): 2 - UNKNOWN - check upstream: 22 ## Quality gates - `make check` — catalog/provenance/audit/eval freshness, repo hygiene, Paper-WorkFlow gates, stdlib unit tests, eval-harness lint + graded smoke, and the strict numeric benchmark (reference-gold drift fails the build). - External links and OpenSSF Scorecard run on their scheduled workflows.
当前收录版本
0.0.1
质量评测
满分 100。前四项来自 DeepSkill 收录的质量报告。
评分依据
提供 DiD、IV、RDD、合成控制法等主流因果推断方法的完整实现
评分依据
触发词精准覆盖 DiD、IV、RDD 等核心术语
评分依据
模块化结构清晰,每方法独立成章
评分依据
工具库来源明确(statsmodels、linearmodels)
计算方式
取前四项原始分值的算术平均,四舍五入取整。可通过收录报告核对各项评分与说明,运行记录需单独查证。
权限与使用前检查
根据当前主文档自动整理,命中内容可能是示例或防护说明;不是实际权限声明,也未覆盖附件、外部依赖或运行行为。
未从主文档识别到上述常见操作,不能据此判断无需权限或没有风险。
安全评级与访问范围
risk · 静态证据核验,不代表运行安全保证。
使用前请由宿主核对并授权实际文件读写、外部服务和联网目的地;本站的核验不会授予任何运行权限。
用户讨论
还没有已核实的点评。