Lee-Wooldridge DiD via unit-specific pre-treatment transformations

Main entry point implementing the Lee & Wooldridge (2026) panel difference-in-differences estimator. Supports both common-timing and staggered adoption designs. Each unit's outcome is residualised using only its own pre-treatment observations before running a pooled cross-sectional OLS.

Usage

lwdid(
  data,
  y,
  ivar,
  tvar,
  gvar = NULL,
  post = NULL,
  dvar = NULL,
  rolling = "demean",
  method = NULL,
  control_group = "never_treated",
  aggregate = "overall",
  vce = NULL,
  cluster_var = NULL,
  controls = NULL,
  season_var = NULL,
  pre = -1L,
  never = FALSE,
  attgt = FALSE,
  ydot = FALSE,
  reps = 999L,
  seed = NULL,
  level = 95,
  nose = FALSE,
  nboot = 999,
  nperm = 999,
  vce_inner = "hc3"
)

Arguments

data

A long-format panel data frame (one row per unit-period).

y

Character. Name of the outcome column.

ivar

Character. Name of the unit identifier column.

tvar

Character. Name of the calendar time column (numeric or integer).

gvar

Character or NULL. Name of the first-treatment-year column for staggered designs. Units with value 0 or NA are treated as never-treated. Set to NULL (default) for common-timing designs (supply post instead).

post

Character or NULL. Name of a binary post-treatment indicator column (0 = pre, 1 = post). Required when gvar = NULL; ignored otherwise.

dvar

Character or NULL. Name of a unit-level treatment-group indicator for common-timing designs. If omitted, lwdid() keeps backwards-compatible behavior when post is the treatment-on indicator; if post is a calendar post indicator, it looks for an unambiguous unit-invariant treatment column named treat, treated, D, or d.

rolling

Character. Transformation method applied to each unit's pre-treatment observations:

"demean": Subtract the unit's pre-period mean (default).
"detrend": Remove a linear trend fitted on pre-periods.
"demeanq": Seasonal demeaning; requires season_var.
"detrendq": Seasonal detrending; requires season_var.

method

Character or NULL. If one of "ra", "ipw", "ipwra", the large-N path (Lee & Wooldridge 2026a) is used: per-cohort/period ATT(g,t) via regression adjustment, inverse-probability weighting, or doubly-robust IPWRA, aggregated to the event-study WATT(r) path with wild-cluster-bootstrap inference. ipw/ipwra require controls. rolling must be "demean" or "detrend". If NULL (default), the small-N path is used.

control_group

Character. Control group for staggered designs: "never_treated" (default) or "not_yet_treated".

aggregate

Character. Aggregation level for staggered designs: "overall" (default), "cohort", or "none" (returns all (g,r) pairs).

vce

Character or NULL. Variance-covariance estimator: NULL (homoskedastic OLS), "hc1", "hc3", "cluster", "wildboot" (wild cluster bootstrap), or "permutation" (randomisation inference). The last two are distribution-free and recommended at small N.

cluster_var

Character or NULL. Column name for clustering; required when vce = "cluster".

controls

Character vector or NULL. Names of time-invariant control variables to include in the cross-sectional regression.

season_var

Character or NULL. Column name of the seasonal indicator (required for rolling = "demeanq" or "detrendq").

pre

Integer. Large-N only. Number of most-recent pre-periods used in the transformation; -1 (default) uses all pre-periods.

never

Logical. Large-N only. If TRUE, the comparison group is never-treated units only (default FALSE uses not-yet-treated units too).

attgt

Logical. Large-N only. If TRUE, the ATT(g,t) cell estimates are returned (and printed).

ydot

Logical. Large-N only. If TRUE, the per-cohort transformed outcomes are returned.

reps

Integer. Large-N only. Wild cluster bootstrap replications (default 999).

seed

Integer or NULL. Large-N only. Seed for the wild bootstrap.

level

Numeric. Confidence level for the large-N path (default 95).

nose

Logical. Large-N only. If TRUE, skip standard errors (faster).

nboot

Integer. Number of bootstrap replications for vce = "wildboot" (default 999).

nperm

Integer. Number of permutations for vce = "permutation" (default 999).

vce_inner

Character. Inner variance estimator used when computing the observed t-statistic inside the wild bootstrap (default "hc3").

Value

An object of class "lwdid", a list containing:

design: "staggered" or "common_timing".
att_overall: Estimated overall ATT.
se_overall: Standard error of overall ATT.
tstat: t-statistic.
pvalue: Two-sided p-value.
att_by_cohort: Data frame of cohort-specific ATTs (staggered only).
att_by_cohort_time: Data frame of (g,r)-specific ATTs (staggered only).
att_by_period: Data frame of period-specific ATTs (common timing only).
ci_lower, ci_upper: 95% confidence interval bounds (common timing only).
N: Sample size at first post-treatment period (common timing only).

References

Lee, S. J., & Wooldridge, J. M. (2026a). Simple Transformation Approach to Difference-in-Differences Estimation for Panel Data. Journal of Business & Economic Statistics, 1-27. doi:10.1080/07350015.2026.2683047

Lee, S. J., & Wooldridge, J. M. (2026b). Simple Approaches to Inference with Difference-in-Differences Estimators with Small Cross-Sectional Sample Sizes. Working paper, SSRN 5325686. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5325686

lwdidR is an R port of the original Stata lwdid command by Soo Jeong Lee and Jeffrey M. Wooldridge: https://github.com/Soo-econ/lwdid. The Castle Doctrine example (Section 7.2) and the small-sample inference options are from (2026b).

Examples

# Load bundled Castle Doctrine dataset
castle <- read.csv(system.file("extdata", "castle.csv", package = "lwdidR"))
castle$gvar <- castle$effyear
castle$gvar[is.na(castle$gvar) | castle$gvar == 0] <- NA

# Staggered design with demeaning and HC3 standard errors
res <- lwdid(castle, "lhomicide", "sid", "year",
             gvar = "gvar", rolling = "demean", vce = "hc3")
print(res)
#> 
#> Lee-Wooldridge DiD (lwdidR)
#> Design:      staggered
#> Transf.:     demean
#> VCE:         HC3
#> --------------------------------------------------
#> Overall ATT:   0.0917
#> SE:            0.0612
#> t-stat:        1.4997
#> p-value:       0.1402
#> --------------------------------------------------
#> 
#> Cohort-specific effects:
#>  cohort     att      se tstat    pvalue
#>    2005 0.08017 0.03215 2.493 1.884e-02
#>    2006 0.06824 0.08920 0.765 4.488e-01
#>    2007 0.11406 0.09838 1.159 2.552e-01
#>    2008 0.14605 0.08203 1.780 8.548e-02
#>    2009 0.21108 0.03550 5.946 2.115e-06
#>