Skip to content

brycewang-stanford/Auto-Empirical-Research-Skills / skills/11-James-Traina-compound-science/skills/causal-inference

causal-inference

Verified

This skill covers causal inference methods in observational and quasi-experimental settings. Use when the user is implementing, choosing between, or debugging causal identification strategies — including instrumental variables, difference-in-differences, regression discontinuity, synthetic control, or matching estimators. Triggers on "causal effect", "identification strategy", "instrumental variable", "2SLS", "GMM", "difference-in-differences", "DiD", "staggered treatment", "regression discontinuity", "RDD", "synthetic control", "matching", "propensity score", "IPW", "AIPW", "doubly robust", "LATE", "ATT", "ATE", "parallel trends", "exclusion restriction", "first stage", "weak instruments", or "endogeneity".

FileKindSize
SKILL.mdskill md12.3 KB
references/method-implementations.mdmarkdown10.9 KB
references/staggered-did.mdmarkdown3.3 KB
references/synthetic-control.mdmarkdown3.5 KB

references/synthetic-control.md

View on GitHub
# Synthetic Control — Implementation Reference

## Standard Synthetic Control

```python
# Python: SparseSC, pensynth, or manual implementation
# The R package Synth is the standard reference implementation

import numpy as np
from scipy.optimize import minimize

def synthetic_control(Y, treated_idx, pre_periods, post_periods):
    """
    Basic synthetic control: find weights W such that
    Y_treated(pre) ≈ Y_controls(pre) @ W

    Y: (T x N) matrix of outcomes
    treated_idx: index of treated unit
    pre_periods: indices of pre-treatment periods
    post_periods: indices of post-treatment periods
    """
    Y_pre = Y[pre_periods, :]
    control_idx = [i for i in range(Y.shape[1]) if i != treated_idx]

    Y1_pre = Y_pre[:, treated_idx]           # treated unit, pre-treatment
    Y0_pre = Y_pre[:, control_idx]            # control units, pre-treatment

    n_controls = len(control_idx)

    # Minimize || Y1_pre - Y0_pre @ w ||^2
    # subject to: w >= 0, sum(w) = 1
    def objective(w):
        return np.sum((Y1_pre - Y0_pre @ w) ** 2)

    constraints = [
        {'type': 'eq', 'fun': lambda w: np.sum(w) - 1}
    ]
    bounds = [(0, 1)] * n_controls

    w0 = np.ones(n_controls) / n_controls
    result = minimize(objective, w0, method='SLSQP',
                      bounds=bounds, constraints=constraints)

    weights = result.x

    # Synthetic control outcome for all periods
    Y_synth = Y[:, control_idx] @ weights

    # Treatment effect = treated - synthetic
    effect = Y[:, treated_idx] - Y_synth

    return weights, Y_synth, effect

# Inference: permutation (placebo) test
def placebo_test(Y, treated_idx, pre_periods, post_periods):
    """
    Run synthetic control for every unit as if it were treated.
    Compare treated unit's gap to placebo distribution.
    """
    effects = {}
    for i in range(Y.shape[1]):
        w, y_synth, eff = synthetic_control(Y, i, pre_periods, post_periods)

        # Pre-treatment RMSPE (for quality filter)
        pre_rmspe = np.sqrt(np.mean(eff[pre_periods] ** 2))
        post_rmspe = np.sqrt(np.mean(eff[post_periods] ** 2))

        effects[i] = {
            'effect': eff,
            'pre_rmspe': pre_rmspe,
            'post_rmspe': post_rmspe,
            'ratio': post_rmspe / pre_rmspe if pre_rmspe > 0 else np.inf
        }

    # p-value: fraction of placebos with ratio >= treated unit's ratio
    treated_ratio = effects[treated_idx]['ratio']
    ratios = [v['ratio'] for v in effects.values()]
    p_value = np.mean([r >= treated_ratio for r in ratios])

    return effects, p_value
```

## Augmented Synthetic Control (Ben-Michael et al. 2021)

Combines synthetic control with outcome modeling to reduce bias:

```python
# R: augsynth package
# library(augsynth)
# result <- augsynth(
#     outcome ~ treatment,
#     unit = unit_id, time = year,
#     data = df,
#     progfunc = "ridge",  # or "none" for standard SC
#     scm = TRUE
# )
# summary(result)
```

## Synthetic Control Diagnostics

- [ ] **Pre-treatment fit**: Synthetic control closely tracks treated unit pre-treatment (plot and report RMSPE)
- [ ] **Weight sparsity**: Examine weights — too many near-zero weights suggest poor donor pool
- [ ] **Placebo tests**: Permutation inference over all donor units
- [ ] **Leave-one-out**: Remove each control unit with positive weight; results should be stable
- [ ] **Time placebo**: Assign treatment to an earlier date; should find no effect
- [ ] **Predictor balance**: Match on pre-treatment covariates, not just outcome
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 323kUpdated Claude CodeCodexVerifiedInstall