R Comparison#

This guide compares diff-diff with popular R packages for DiD analysis, helping users familiar with R transition to Python.

Overview#

Feature

diff-diff (Python)

did (R)

Other R

Basic DiD

✅ DifferenceInDifferences

✅ att_gt

✅ fixest::feols

Staggered DiD

✅ CallawaySantAnna

✅ att_gt

did2s, DRDID

Covariate adjustment

✅ DR, IPW, Reg

✅ DR, IPW, Reg

✅ Varies

Honest DiD

✅ HonestDiD

HonestDiD package

N/A

Synthetic DiD

✅ SyntheticDiD

synthdid package

N/A

Wild bootstrap

✅ wild_bootstrap_se

fwildclusterboot

N/A

Package Correspondence#

R did Package → diff-diff#

The R did package by Callaway and Sant’Anna is the gold standard for staggered DiD. Here’s how to translate common operations:

Basic estimation:

# R (did package)
library(did)
out <- att_gt(
  yname = "Y",
  tname = "period",
  idname = "id",
  gname = "G",
  data = data
)

# Python (diff-diff)
from diff_diff import CallawaySantAnna

cs = CallawaySantAnna()
results = cs.fit(
    data,
    outcome='Y',
    time='period',
    unit='id',
    first_treat='G'
)

With covariates (doubly robust):

# R
out <- att_gt(
  yname = "Y", tname = "period",
  idname = "id", gname = "G",
  xformla = ~ X1 + X2,
  est_method = "dr",
  data = data
)

# Python
cs = CallawaySantAnna(estimation_method='dr')
results = cs.fit(
    data,
    outcome='Y',
    time='period',
    unit='id',
    first_treat='G',
    covariates=['X1', 'X2']
)

Aggregations:

# R
agg_simple <- aggte(out, type = "simple")
agg_dynamic <- aggte(out, type = "dynamic")
agg_group <- aggte(out, type = "group")

# Python (unlike R's aggte(), aggregation is requested at fit time)
results = cs.fit(data, outcome='Y', time='period', unit='id',
                 first_treat='G', aggregate='all')
overall_att = results.overall_att  # Simple aggregation
event_study = results.event_study_effects  # Dynamic
by_group = results.group_effects  # By cohort

R HonestDiD Package → diff-diff#

The HonestDiD package implements Rambachan & Roth (2023) sensitivity analysis:

Relative magnitudes (ΔRM):

# R
library(HonestDiD)
delta_rm_results <- createSensitivityResults_relativeMagnitudes(
  betahat = beta_hat,
  sigma = sigma,
  numPrePeriods = 4,
  numPostPeriods = 3,
  Mbarvec = seq(0, 2, by = 0.5)
)

# Python
from diff_diff import HonestDiD

honest = HonestDiD(method='relative_magnitude', M=1.0)
results = honest.fit(event_study_results)

# Sensitivity analysis over M grid
sensitivity = honest.sensitivity_analysis(
    event_study_results,
    M_grid=[0, 0.5, 1.0, 1.5, 2.0]
)

Smoothness restrictions (ΔSD):

# R
delta_sd_results <- createSensitivityResults(
  betahat = beta_hat,
  sigma = sigma,
  numPrePeriods = 4,
  numPostPeriods = 3,
  Mvec = seq(0, 0.1, by = 0.02)
)

# Python
from diff_diff import HonestDiD

honest = HonestDiD(method='smoothness', M=0.05)
results = honest.fit(event_study_results)

R synthdid Package → diff-diff#

The synthdid package implements Arkhangelsky et al. (2021):

# R
library(synthdid)
setup <- panel.matrices(data, unit = "unit", time = "time",
                        outcome = "Y", treatment = "treatment")
tau.hat <- synthdid_estimate(setup$Y, setup$N0, setup$T0)

# Python
from diff_diff import SyntheticDiD

# SyntheticDiD requires a time-invariant ever-treated indicator
data['ever_treated'] = data.groupby('unit')['treatment'].transform('max')

# Derive post-treatment periods from treatment timing
post_periods = sorted(data.loc[data['treatment'] == 1, 'time'].unique())

sdid = SyntheticDiD()
results = sdid.fit(
    data,
    outcome='Y',
    unit='unit',
    time='time',
    treatment='ever_treated',
    post_periods=post_periods
)

Heterogeneous Adoption (HAD)#

When every unit is treated at the post period (universal-rollout policies, industry-wide regime changes) but treatment intensity varies across units, the standard R workhorses (did, fixest, synthdid, DIDmultiplegtDYN) assume an untreated comparison group exists and do not apply. The dedicated R package DIDHAD (de Chaisemartin et al., August 2025) covers the QUG case (Design 1’, d_lower = 0) from the same arXiv paper.

diff-diff ships HeterogeneousAdoptionDiD, which implements de Chaisemartin, Ciccia, D’Haultfoeuille and Knau (2026, arXiv:2405.04465v6) and adds two surfaces beyond the QUG-focused R package: Design 1 (no QUG, d_lower > 0, targets WAS_{d_lower} under Assumption 6 or sign-only under Assumption 5), and survey-design integration via Binder (1983) Taylor-series linearization (sampling weights + optional strata / PSU / FPC). The diagnostic battery did_had_pretest_workflow() surfaces violations of the HAD identification assumptions (the design path is auto-detected separately by HeterogeneousAdoptionDiD.fit() from the dose support).

import numpy as np
import pandas as pd
from diff_diff import HeterogeneousAdoptionDiD

# Build a HAD-shape panel: D=0 in pre-periods (t < F), D > 0 only at F+.
rng = np.random.default_rng(42)
G, F, T = 200, 4, 5
doses = rng.beta(0.5, 1.0, size=G)
rows = []
for g in range(G):
    for t in range(1, T + 1):
        y = (rng.normal()
             + (doses[g] + doses[g] ** 2) * (t >= F)
             + rng.normal(0, 0.5))
        d = doses[g] if t >= F else 0.0
        rows.append({'unit': g, 'period': t, 'y': y, 'dose': d})
had_data = pd.DataFrame(rows)

est = HeterogeneousAdoptionDiD()
results = est.fit(had_data, outcome_col='y', unit_col='unit',
                  time_col='period', dose_col='dose',
                  aggregate='event_study')

Key Differences#

Design Philosophy#

  • diff-diff: sklearn-style API with fit() method, returning rich result objects

  • R packages: Function-based, returning lists or S3/S4 objects

Inference#

  • diff-diff: Analytical SEs by default, wild bootstrap available

  • R did: Multiplier bootstrap by default

Fixed Effects#

  • diff-diff: absorb parameter for high-dimensional FE (within transformation)

  • R fixest: feols with | notation for absorbed FE

Output Format#

diff-diff results have convenience methods:

results.summary()       # Print formatted table
results.to_dict()       # Dictionary representation
results.to_dataframe()  # pandas DataFrame

Feature Comparison Table#

Feature

diff-diff

R did

R HonestDiD

R synthdid

Basic 2x2 DiD

✅

✅

❌

❌

TWFE

✅

❌

❌

❌

Staggered DiD (CS)

✅

✅

❌

❌

Covariate adjustment

✅

✅

❌

❌

Doubly robust

✅

✅

❌

❌

Group-time effects

✅

✅

❌

❌

Event study

✅

✅

✅

❌

Synthetic DiD

✅

❌

❌

✅

Honest DiD (ΔRM)

✅

❌

✅

❌

Honest DiD (ΔSD)

✅

❌

✅

❌

Wild bootstrap

✅

❌

❌

❌

Cluster-robust SE

✅

✅

❌

✅

Placebo tests

✅

❌

❌

✅

Parallel trends tests

✅

✅

❌

❌

Bacon decomposition

✅

❌

❌

❌

Sun-Abraham

✅

❌

❌

❌

Imputation DiD

✅

❌

❌

❌

Two-Stage DiD (did2s)

✅

❌

❌

❌

Stacked DiD

✅

❌

❌

❌

Continuous DiD

✅

✅

❌

❌

Triple Difference (DDD)

✅

❌

❌

❌

TROP

✅

❌

❌

❌

Efficient DiD

✅

❌

❌

❌

Heterogeneous adoption (HAD)

✅

❌

❌

❌

Note

R equivalents for estimators not covered by the did, HonestDiD, or synthdid packages: Sun-Abraham is available via fixest::sunab(); Imputation DiD via the didimputation package; Two-Stage DiD via the did2s package; Bacon Decomposition via the bacondecomp package; Stacked DiD requires manual implementation or the stackedev package; Continuous DiD is available via the did package continuous extension; Triple Difference requires manual implementation in R. TROP and Efficient DiD have no direct R equivalents. HeterogeneousAdoptionDiD (dCDH 2026) overlaps with the dedicated R package DIDHAD (de Chaisemartin et al., 2025), which covers the QUG case (Design 1’); diff-diff additionally covers Design 1 (no QUG, WAS_{d_lower}) and survey-design integration via Binder TSL.

Migration Tips#

  1. Column names: diff-diff uses string column names, similar to R packages

  2. Formula interface: diff-diff supports R-style formulas for basic DiD: formula='y ~ treated * post'

  3. Results access: Use .att, .se, .ci instead of $att, $se

  4. Visualization: plot_event_study() produces matplotlib figures similar to ggdid() output

  5. Missing data: diff-diff requires complete data; use balance_panel() or dropna() first

  6. Heterogeneous Adoption (HAD): If you need surfaces the R DIDHAD package does not cover - Design 1 (no QUG, WAS_{d_lower}) or survey-design integration - reach for HeterogeneousAdoptionDiD. See the Heterogeneous Adoption (HAD) section above for the migration pattern.