تقييم الأثر والاستدلال السببي للبرامج والسياسات العامة

import numpy as np
import pandas as pd
import statsmodels.formula.api as smf

# Synthetic panel: 100 districts x 8 years, half treated
rng = np.random.default_rng(1)
df = pd.DataFrame(
    [(i, t) for i in range(100) for t in range(8)],
    columns=["unit", "year"])
df["treated"] = (df.unit < 50).astype(int)
df["post"] = (df.year >= 4).astype(int)
u = rng.normal(0, 1, 100)[df.unit]  # unit heterogeneity
df["y"] = (u + 0.3 * df.year + 2.0 * df.treated * df.post
           + rng.normal(0, 1, len(df)))

# Two-way fixed effects DiD, SEs clustered by district
m = smf.ols("y ~ treated:post + C(unit) + C(year)",
            data=df).fit(cov_type="cluster",
                         cov_kwds={"groups": df["unit"]})
print("ATT:", round(m.params["treated:post"], 3),
      "SE:", round(m.bse["treated:post"], 3))
print(m.conf_int().loc["treated:post"].round(3).tolist())

library(MatchIt)
library(marginaleffects)
data("lalonde", package = "MatchIt")

# 1:1 nearest-neighbour matching on a logit propensity score
m <- matchit(treat ~ age + educ + race + married +
               nodegree + re74 + re75,
             data = lalonde, method = "nearest",
             distance = "glm", caliper = 0.2)

# Covariate balance after matching (|SMD| < 0.1 target)
summary(m)$sum.matched[, "Std. Mean Diff."]

# ATT with cluster-robust SEs by matched pair
md <- match.data(m)
fit <- lm(re78 ~ treat * (age + educ + re74 + re75),
          data = md, weights = weights)
avg_comparisons(fit, variables = "treat",
                vcov = ~subclass,
                newdata = subset(md, treat == 1))