Fused Extended Two-Way Fixed Effects

Calculates the fused extended two-way fixed effects (FETWFE) estimator for unbiased and efficient estimation of difference-in-differences in panel data with staggered treatment adoption. This estimator eliminates bias inherent in conventional two-way fixed effects estimators, while also employing a novel bridge regression regularization approach to improve efficiency and yield valid standard errors. Also implements extended TWFE (etwfe) and bridge-penalized ETWFE (betwfe). Provides S3 classes for streamlined workflow and supports flexible tuning (ridge and rank-condition guarantees), automatic covariate centering/scaling, and detailed overall and cohort-specific effect estimates with valid standard errors. Includes simulation and formatting utilities, extensive diagnostic tools, vignettes, and examples. See Faletto (2025) ().

Fused Extended Two-Way Fixed Effects

The {fetwfe} package implements fused extended two-way fixed effects (FETWFE), a methodology for estimating treatment effects in difference-in-differences with staggered adoptions.

For a brief introduction to the methodology, as well as background on difference-in-differences with staggered adoptions and motivation for FETWFE, see this blog post.
For more detailed slides on the methodology (but less detailed than the paper), see here.
Check out the most recent draft of the full paper here.
This blog post explains a little more about what the package does under the hood, if you're interested.

fetwfePackage

To install the {fetwfe} package, simply use

install.packages("fetwfe")

You can also install the latest development version by using

# install.packages("remotes")  # if needed
remotes::install_github("gregfaletto/fetwfePackage")

The primary function in the {fetwfe} is fetwfe(), which implements fused extended two-way fixed effects. Here's some example code applying fetwfe() to the castle data set from the bacondecomp package:

library(fetwfe)
library(bacondecomp)

data(castle)

# Response: the log homicide rate. Treatment: `cdl` records the share of
# the year the castle-doctrine law was in effect, so `cdl > 0` gives the
# absorbing 0/1 treatment indicator `fetwfe()` requires.
castle$l_homicide <- log(castle$homicide)
castle$treated <- as.integer(castle$cdl > 0)

res <- fetwfe(
    pdata = castle,
    time_var = "year",
    unit_var = "state",
    treatment = "treated",
    response = "l_homicide",
    verbose = TRUE)

summary(res)

For vignettes and full documentation, check out the page for the {fetwfe} package on CRAN.