lwdid

Lee-Wooldridge Difference-in-Differences Estimation for R

Overview

lwdid implements the Rolling Difference-in-Differences Estimator proposed by Lee and Wooldridge (2025, 2026) for R. Through unit-specific time-series transformations (demeaning or detrending), the package converts panel DiD estimation into standard cross-sectional treatment effect problems, enabling application of multiple estimators and exact small-sample inference.

The package provides fast and flexible estimation for both common timing and staggered adoption treatment settings in panel data, and covers standard large-N asymptotic inference as well as small-N settings where conventional inference may not be reliable.

Features:

• Two designs: Common Timing (simultaneous treatment) and Staggered Adoption (cohort-specific timing)
• Four estimators: RA (Regression Adjustment), IPW (Inverse Probability Weighting), IPWRA (Doubly Robust), PSM (Propensity Score Matching)
• Complete inference: Homoskedastic / HC0–HC4 / cluster-robust standard errors, Fisher Randomization Inference, Wild Cluster Bootstrap
• Rich diagnostics: Parallel trends tests, clustering diagnostics, selection mechanism diagnostics, sensitivity analysis
• Data preprocessing: Individual-level data aggregation to panel format via aggregate_to_panel()
• High performance: Vectorized data.table-based computation with optional parallel processing via future
• Export: LaTeX tables (to_latex(), to_latex_comparison()), CSV (to_csv()), event study plots

Contents

• Installation
• Quick Start
• Transformations
• Estimators
• Aggregation
• Variance-Covariance Estimation
• Diagnostics
• Inference
• Built-in Datasets
• Vignettes
• Citation
• Authors
• See Also

Installation

Requirements: R (>= 4.0.0)

From GitHub (development version)

# install.packages("devtools")
devtools::install_github("gorgeousfish/lwdid-r")

From CRAN (planned)

install.packages("lwdid")

Optional Dependencies

The core functionality requires only data.table and sandwich. Optional packages unlock additional features:

# Wild Cluster Bootstrap inference (archived from CRAN; install from R-universe)
install.packages("fwildclusterboot",
                 repos = "https://s3alfisc.r-universe.dev")

# Propensity Score Matching (PSM estimator)
install.packages(c("MatchIt", "cobalt"))

# Parallel computation (RI / Bootstrap / Staggered loops)
install.packages(c("future", "future.apply"))

# Visualization
install.packages("ggplot2")

Quick Start

Example 1: Common Timing (Smoking Data)

The smoking dataset contains the California Proposition 99 tobacco control study (1970–2000). California is the sole treated state.

library(lwdid)
data(smoking)

result <- lwdid(
  data = smoking,
  y = "lcigsale",
  ivar = "state",
  tvar = "year",
  d = "d",
  post = "post",
  rolling = "demean"
)

summary(result)

Example 2: Staggered Adoption (Castle Doctrine Data)

The castle dataset examines Castle Doctrine laws and homicide rates. States adopted the law at different times between 2005 and 2009.

data(castle)

result <- lwdid(
  data = castle,
  y = "lhomicide",
  ivar = "sid",
  tvar = "year",
  gvar = "gvar",
  rolling = "demean",
  aggregate = "overall"
)

summary(result)

Example 3: Event Study

result_es <- lwdid(
  data = castle,
  y = "lhomicide",
  ivar = "sid",
  tvar = "year",
  gvar = "gvar",
  rolling = "demean",
  aggregate = "event_time"
)

summary(result_es)
plot(result_es)

Example 4: Using Controls and IPW Estimator

result_ipw <- lwdid(
  data = castle,
  y = "lhomicide",
  ivar = "sid",
  tvar = "year",
  gvar = "gvar",
  rolling = "demean",
  estimator = "ipw",
  controls = c("police", "unemployrt", "income"),
  aggregate = "overall"
)

summary(result_ipw)

Transformations

rolling =	Description	Min. pre-treatment periods
"demean"	Subtract pre-treatment mean	1
"detrend"	Remove unit-specific linear trend	2
"demeanq"	Seasonal demeaning (quarterly data)	1 per season
"detrendq"	Seasonal detrending (quarterly data)	2 per season

Estimators

estimator =	Description	Controls required	Doubly robust
"ra"	Regression Adjustment	Optional	No
"ipw"	Inverse Probability Weighting	Yes	No
"ipwra"	Doubly Robust (IPW + RA)	Yes	Yes
"psm"	Propensity Score Matching	Yes	No

Aggregation

For staggered adoption designs, treatment effects are first estimated for each cohort-time pair and then aggregated:

aggregate =	Description
"cohort"	Group-time ATT by treatment cohort (default)
"overall"	Single weighted average ATT across all cohorts and periods
"event_time"	ATT by relative time to treatment (for event study plots)
"calendar_time"	ATT by calendar time period
"att_gt"	Return all individual group-time ATT(g,t) estimates

The control_group parameter controls which untreated units serve as comparisons:

control_group =	Description
"not_yet_treated"	Units not yet treated at timet (default; larger comparison group)
"never_treated"	Only units that are never treated during the sample period

Variance-Covariance Estimation

vce =	Description
NULL	OLS (homoskedastic)
"robust" / "hc1"	HC1 heteroskedasticity-robust
"hc0" – "hc4"	HC-family (HC3 recommended for smallN)
"cluster"	Cluster-robust (requires cluster_var)

Diagnostics

# Parallel trends test (staggered)
pt <- lwdid_test_parallel_trends(
  data = castle, y = "lhomicide",
  ivar = "sid", tvar = "year", gvar = "gvar"
)
summary(pt)

# Sensitivity analysis (common timing)
sens <- lwdid_sensitivity(
  data = smoking, y = "lcigsale",
  ivar = "state", tvar = "year",
  d = "d", post = "post",
  type = "all"
)
summary(sens)

# Full diagnostic suite (staggered)
diag <- lwdid_diagnose(
  data = castle, y = "lhomicide",
  ivar = "sid", tvar = "year", gvar = "gvar"
)
summary(diag)

Inference

Randomization Inference

Randomization inference (RI) provides exact p-values without relying on asymptotic normality, which is particularly useful in small-N settings:

result_ri <- lwdid(
  data = smoking,
  y = "lcigsale",
  ivar = "state",
  tvar = "year",
  d = "d",
  post = "post",
  rolling = "demean",
  ri = TRUE,
  rireps = 1000,
  seed = 42
)

summary(result_ri)

Wild Cluster Bootstrap

Wild Cluster Bootstrap (WCB) is automatically applied when vce = "cluster" is specified, providing asymptotic refinement for cluster-robust inference:

result_wcb <- lwdid(
  data = castle,
  y = "lhomicide",
  ivar = "sid",
  tvar = "year",
  gvar = "gvar",
  rolling = "demean",
  vce = "cluster",
  cluster_var = "sid",
  aggregate = "overall"
)

summary(result_wcb)

Built-in Datasets

Dataset	Design	Description	Obs	Units
smoking	Common Timing	California Proposition 99 tobacco control (1970–2000)	1,209	39 states
castle	Staggered	Castle Doctrine / Stand Your Ground laws (2000–2010)	550	50 states
walmart	Staggered	Walmart store openings and local labor markets (1977–1999)	29,440	1,280 counties

Vignettes

• vignette("getting-started"): Data format, transformations, common timing and staggered examples
• vignette("advanced-usage"): Multiple estimators, sensitivity, diagnostics, WCB, RI, parallel computation, export
• vignette("methodology"): Mathematical framework, identification, and comparison with Callaway & Sant'Anna (2021)

Citation

If you use lwdid in your research, please cite the following papers:

APA Format:

Cai, X., & Xu, W. (2026). lwdid: Lee-Wooldridge Difference-in-Differences Estimation for R (Version 0.1.0) [Computer software]. GitHub. https://github.com/gorgeousfish/lwdid-r

Lee, S. J., & Wooldridge, J. M. (2025). A simple transformation approach to difference-in-differences estimation for panel data. Available at SSRN 4516518.

Lee, S. J., & Wooldridge, J. M. (2026). Simple Approaches to Inference with Difference-in-Differences Estimators with Small Cross-Sectional Sample Sizes. Available at SSRN 5325686.

BibTeX:

@software{lwdid2026r,
  title={lwdid: Lee-Wooldridge Difference-in-Differences Estimation for R},
  author={Xuanyu Cai and Wenli Xu},
  year={2026},
  version={0.1.0},
  url={https://github.com/gorgeousfish/lwdid-r}
}

@article{lee2025simple,
  title={A Simple Transformation Approach to Difference-in-Differences
         Estimation for Panel Data},
  author={Lee, Soo Jeong and Wooldridge, Jeffrey M.},
  year={2025},
  note={Available at SSRN 4516518}
}

@article{lee2026simple,
  title={Simple Approaches to Inference with Difference-in-Differences
         Estimators with Small Cross-Sectional Sample Sizes},
  author={Lee, Soo Jeong and Wooldridge, Jeffrey M.},
  year={2026},
  note={Available at SSRN 5325686}
}

Authors

R Implementation:

• Xuanyu Cai, City University of Macau Email: xuanyuCAI@outlook.com
• Wenli Xu, City University of Macau Email: wlxu@cityu.edu.mo

Methodology:

• Soo Jeong Lee, Michigan State University
• Jeffrey M. Wooldridge, Michigan State University

License

AGPL-3.0. See LICENSE.md for details.

See Also

• Original Stata package by Lee and Wooldridge: https://github.com/Soo-econ/lwdid
• Python implementation: https://github.com/gorgeousfish/lwdid-py