Skip to content

brycewang-stanford/Auto-Empirical-Research-Skills / skills/10-Jill0099-causal-inference-mixtape

causal-inference-mixtape

Verified

This skill should be used when the user asks to "implement a DiD regression", "write a causal inference pipeline", "set up an event study", "implement instrumental variables", "run a regression discontinuity design", "build a synthetic control model", "implement propensity score matching", "write parallel trends test", "implement Bacon decomposition", or needs code templates for causal inference methods in Python, R, or Stata. Based on Scott Cunningham's Causal Inference: The Mixtape.

references/r-stata-comparison.md

View on GitHub
# R & Stata Comparison — Methods Not Available in Python

Cross-language coverage gaps and package recommendations.

---

## Method Coverage Matrix

| Method | Python | R | Stata |
|--------|--------|---|-------|
| OLS / Robust SE | statsmodels | estimatr | reg, robust |
| Cluster SE | statsmodels | estimatr/lfe | cluster() |
| Two-way FE | statsmodels (slow) | fixest (fast) | reghdfe (fast) |
| DiD (2x2) | statsmodels | did/fixest | reghdfe |
| Event Study | manual | fixest/did | reghdfe + coefplot |
| Bacon Decomposition | **None** | bacondecomp | bacondecomp |
| Callaway-Sant'Anna | **None** | did | csdid |
| Sun & Abraham | **None** | fixest (sunab) | eventstudyinteract |
| Sharp RDD | statsmodels (manual) | rdrobust | rdrobust |
| Fuzzy RDD | linearmodels | rdrobust | rdrobust |
| McCrary Test | **None** | rdd | rddensity |
| IV / 2SLS | linearmodels | AER/ivreg | ivregress |
| JIVE | **None** | **None** | jive |
| Synthetic Control | **rpy2 only** | Synth + SCtools | synth |
| Augmented SC | **None** | augsynth | sdid |
| PSM / Matching | manual | MatchIt | teffects psmatch |
| CEM | **None** | MatchIt (method="cem") | cem |
| IPW | manual | ipw | teffects ipw |
| Randomization Inference | manual loop | ri2 | ritest |
| DAGs | **None** | dagitty + ggdag | **None** |

---

## Python Gaps — Recommended Workarounds

### Synthetic Control

No mature Python package. Two options:

1. **rpy2 bridge** (recommended for production):
```python
import rpy2.robjects as ro
from rpy2.robjects import pandas2ri
pandas2ri.activate()
ro.globalenv['df'] = df
ro.r('library(Synth); ...')
```

2. **SparseSC** (experimental): `pip install SparseSC` — limited functionality

### Bacon Decomposition

No Python implementation. Use R:
```r
library(bacondecomp)
bacon(y ~ treatment, data = df, id_var = "id", time_var = "year")
```

### Coarsened Exact Matching (CEM)

Stata's `cem` command is the gold standard. R alternative:
```r
library(MatchIt)
m.out <- matchit(treated ~ x1 + x2, data = df, method = "cem")
```

### McCrary Density Test

R implementation:
```r
library(rdd)
DCdensity(running_var, cutpoint = cutoff, plot = TRUE)
```

---

## Package Quick Reference

### R Packages

| Package | Purpose | Install |
|---------|---------|---------|
| estimatr | Robust/cluster SE OLS | `install.packages("estimatr")` |
| lfe | High-dimensional FE | `install.packages("lfe")` |
| fixest | Fast FE estimation | `install.packages("fixest")` |
| AER | IV / 2SLS | `install.packages("AER")` |
| rdrobust | RDD estimation | `install.packages("rdrobust")` |
| Synth | Synthetic control | `install.packages("Synth")` |
| SCtools | SC placebo tests | `install.packages("SCtools")` |
| MatchIt | Matching (PSM/CEM) | `install.packages("MatchIt")` |
| did | Callaway-Sant'Anna | `install.packages("did")` |
| bacondecomp | Bacon decomposition | `install.packages("bacondecomp")` |
| dagitty | DAG analysis | `install.packages("dagitty")` |
| ri2 | Randomization inference | `install.packages("ri2")` |
| ipw | Inverse probability weighting | `install.packages("ipw")` |

### Stata Packages

| Package | Purpose | Install |
|---------|---------|---------|
| reghdfe | High-dimensional FE | `ssc install reghdfe` |
| rdrobust | RDD estimation | `ssc install rdrobust` |
| synth | Synthetic control | `ssc install synth` |
| cem | Coarsened exact matching | `ssc install cem` |
| bacondecomp | Bacon decomposition | `ssc install bacondecomp` |
| ritest | Randomization inference | `ssc install ritest` |
| did_multiplegt | Staggered DiD | `ssc install did_multiplegt` |
| eventstudyinteract | Sun & Abraham | `ssc install eventstudyinteract` |
| csdid | Callaway-Sant'Anna | `ssc install csdid` |

### Python Packages

| Package | Purpose | Install |
|---------|---------|---------|
| statsmodels | OLS / WLS / GLM / logit | `pip install statsmodels` |
| linearmodels | IV2SLS / panel models | `pip install linearmodels` |
| plotnine | ggplot2-style plotting | `pip install plotnine` |
| rpy2 | Call R from Python | `pip install rpy2` |

---

## When to Switch Languages

| Situation | Recommendation |
|-----------|---------------|
| Already in a Python ML pipeline | Stay in Python, use rpy2 for gaps |
| Need Bacon decomposition | Switch to R or Stata |
| Synthetic control analysis | Use R (Synth) or Stata (synth) |
| Publication-ready regression tables | Stata (esttab) or R (modelsummary) |
| Exploratory / quick prototyping | Python statsmodels |
| Teaching / reproducibility | R (tidyverse ecosystem) |
| Referee asks for specific robustness | Match the language to the available package |
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 323kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 323kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 323kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall