Engression-Based Distributional Difference-in-Differences

Estimates distributional treatment effects by combining Lee & Wooldridge (2025) panel DiD transformations with engression distributional regression. Produces ATT, quantile treatment effects (QTE), and counterfactual distributions.

Usage

endid(
  data,
  y,
  ivar,
  tvar,
  gvar = NULL,
  post = NULL,
  dvar = NULL,
  rolling = "demean",
  control_group = "never_treated",
  aggregate = "overall",
  controls = NULL,
  season_var = NULL,
  quantiles = seq(0.1, 0.9, 0.1),
  nsample = 500,
  nboot = 200,
  noise_dim = 5,
  hidden_dim = 100,
  num_layer = 3,
  num_epochs = 1000,
  lr = 0.001,
  num_cores = 1,
  silent = TRUE
)

Arguments

data

A long-format panel data frame.

y

Character. Outcome column name.

ivar

Character. Unit identifier column name.

tvar

Character. Calendar time column name (numeric).

gvar

Character or NULL. First-treatment-year column for staggered designs. Units with value 0, NA, or Inf are never-treated.

post

Character or NULL. Binary post-treatment indicator column (0 = pre, 1 = post). Required when gvar = NULL.

dvar

Character or NULL. Binary treatment group indicator column (1 = treated unit, 0 = control). Required for common-timing designs when post is a calendar indicator (all units observed pre and post).

rolling

Character. Transformation method: "demean" (default), "detrend", "demeanq", "detrendq".

control_group

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

aggregate

Character. Aggregation for staggered designs: "overall" (default), "cohort", or "none".

controls

Character vector or NULL. Time-invariant control column names.

season_var

Character or NULL. Seasonal indicator column.

quantiles

Numeric vector. Quantiles for QTE (default: seq(0.1, 0.9, 0.1)).

nsample

Integer. Monte Carlo samples for engression predictions (default: 500).

nboot

Integer. Bootstrap replications for inference (default: 200).

noise_dim

Engression noise dimension (default: 5).

hidden_dim

Engression hidden layer width (default: 100).

num_layer

Engression number of layers (default: 3).

num_epochs

Engression training epochs (default: 1000).

lr

Engression learning rate (default: 1e-3).

num_cores

Integer. Number of cores for bootstrap parallelization (default: 1).

silent

Logical. Suppress engression training output (default: TRUE).

Value

An object of class "endid".

Examples

# \donttest{
  castle <- read.csv(system.file("extdata", "castle.csv", package = "endid"))
  castle$gvar <- castle$effyear
  castle$gvar[is.na(castle$gvar) | castle$gvar == 0] <- NA
  res <- endid(castle, "lhomicide", "sid", "year", gvar = "gvar",
               rolling = "demean", num_epochs = 500, nboot = 50)
  print(res)
# }