Stacked Difference-in-Differences

Builds the stacked dataset described in Baker et al. (2022) and estimates TWFE regressions to recover the ATT and dynamic effects free of contamination from heterogeneous treatment timing.

Usage

stacked_did(
  data,
  unit_var,
  time_var,
  outcome_var,
  treat_time_var,
  anticipation = 0,
  clean_controls = TRUE,
  plot = TRUE
)

Arguments

data

A data frame in balanced long panel format.

unit_var

Character. Name of the unit/panel ID column.

time_var

Character. Name of the time period column (integer/numeric).

outcome_var

Character. Name of the outcome variable column.

treat_time_var

Character. Name of the column recording each unit's first treatment period (NA or a large number for never-treated).

anticipation

Integer. Number of pre-treatment periods to exclude from the "clean" window. Default 0.

clean_controls

Logical. If TRUE (default), exclude already-treated units from the control group within each sub-experiment.

plot

Logical. Whether to return an event-study plot. Default TRUE.

Value

A named list:

estimate

Scalar ATT (weighted average across cohorts).

event_study

Data frame with columns rel_period, estimate, ci_lo, ci_hi.

stacked_data

The stacked data frame used for estimation.

plot

A ggplot2 event-study plot, or NULL if plot = FALSE.

Details

Stacked Difference-in-Differences Estimator

Implements the stacked DiD estimator following Cengiz et al. (2019) and Baker et al. (2022). For each treatment cohort, a "sub-experiment" dataset is created by stacking the cohort's treated units with clean control units (never-treated or not-yet-treated). TWFE is then estimated on this stacked dataset with cohort-by-period interactions, returning an average treatment effect on the treated (ATT) and event-study dynamic effects.

References

Baker, A. C., Larcker, D. F., & Wang, C. C. Y. (2022). How much should we trust staggered difference-in-differences estimates? Journal of Financial Economics, 144(2), 370-395.

Cengiz, D., Dube, A., Lindner, A., & Zipperer, B. (2019). The effect of minimum wages on low-wage jobs. Quarterly Journal of Economics, 134(3), 1405-1454.

Examples

if (requireNamespace("fixest", quietly = TRUE)) {
  data("base_stagg", package = "fixest")
  # base_stagg has: id, year, treated, treatment_time, y, ...
  res <- stacked_did(
    data           = base_stagg,
    unit_var       = "id",
    time_var       = "year",
    outcome_var    = "y",
    treat_time_var = "year_treated",
    anticipation   = 0,
    clean_controls = TRUE,
    plot           = TRUE
  )
  print(res$estimate)
  res$plot
}
#> Error in loadNamespace(x) : there is no package called ‘..cohort..’
#> ATT estimation failed: Error in res[[2]] : subscript out of bounds
#> This error was unforeseen by the author of the function feols. If you think your call to the function is legitimate, could you report?
#> [1] NA