cdid: The Chained Difference-in-Differences
The cdid package extends the popular did library by Brantly Callaway to improve efficiency and handle unbalanced panel data in staggered treatment designs. It implements the methodology introduced in:
Bellego, Benatia, and Dortet-Bernadet (2024), "The Chained Difference-in-Differences", Journal of Econometrics. https://doi.org/10.1016/j.jeconom.2024.105783
Features
- Enhanced Efficiency: Offers greater precision for balanced panels (smaller standard errors)
- Handles Missing Data: Offers greater precision for unbalanced panels, especially when errors are serially correlated.
- Reduced Bias: Less prone to bias when missingness is tied to unobservable heterogeneity.
- Flexible Control Groups: Supports control groups comprising either "nevertreated" or "notyettreated" units, like in the did library.
- Customizable Weighting: Implements aggregation with two weighting matrix options:
- Identity: Best for smaller datasets with less overidentifying restrictions (many missings like in rotating panel surveys)
- Two-step: Best for larger datasets with many overidentifying restrictions (few missings).
Future developments: generalized attrition model (MAR & sequential MAR), doubly-robust estimator (so far only ipw is implemented), and improved computational efficiency. Hopefully, it will be directly integrated within the did library.
Installation
The cdid package can be installed from CRAN using:
install.packages("cdid")Alternatively, for the development version:
remotes::install_github("joelcuerrier/cdid", ref = "main", build_vignettes = TRUE, force = TRUE)Getting Started
Example Usage with Simulated Data
library(did) #for comparison
library(cdid)
set.seed(123)
# Generate a balanced dataset with unit fixed-effects
# The true values of the coefficients are based on time-to-treatment. The treatment
# effect is zero before the treatment, 1.75 one period after, 1.5 two period after,
# 1.25 three period after, 1 four period after, 0.75 five period after, 0.5 six
# period after, etc.
data0 <- fonction_simu_attrition(
N = 500, T = 8,
theta2_alpha_Gg = 0.5, lambda1_alpha_St = 0,
sigma_alpha = 2, sigma_epsilon = 0.1, tprob = 0.5
)
# Ensure all observations are included for a balanced panel
data0$S <- 1
# Run the original `did` library estimation
did.results = did:: att_gt(
yname="Y",
tname="date",
idname = "id",
gname = "date_G",
xformla = ~X,
data = data0,
weightsname = NULL,
allow_unbalanced_panel = FALSE,
panel = TRUE,
control_group = "notyettreated",
alp = 0.05,
bstrap = TRUE,
cband = TRUE,
biters = 1000,
clustervars = NULL,
est_method = "ipw",
base_period = "varying",
print_details = FALSE,
pl = FALSE,
cores = 1
)
# Run `cdid` with 2-step weighting matrix
result_2step = att_gt_cdid(yname="Y", tname="date",
idname="id",
gname="date_G",
xformla=~X,
data=data0,
control_group="notyettreated",
alp=0.05,
bstrap=TRUE,
biters=1000,
clustervars=NULL,
cband=TRUE,
est_method="2-step",
base_period="varying",
print_details=FALSE,
pl=FALSE,
cores=1)
# Run `cdid` with identity weighting matrix
result_id = att_gt_cdid(yname="Y", tname="date",
idname="id",
gname="date_G",
xformla=~X,
data=data0,
control_group="notyettreated",
alp=0.05,
bstrap=TRUE,
biters=1000,
clustervars=NULL,
cband=TRUE,
est_method="Identity",
base_period="varying",
print_details=FALSE,
pl=FALSE,
cores=1)
# Print results
print(did.results)
print(result_2step)
print(result_id)Aggregating Results
After computing the group-time ATT estimates, aggregate results can be obtained using functions from did library
agg.es.did <- aggte(MP = did.results, type = 'dynamic')
agg.es.2step <- aggte(MP = result_2step, type = 'dynamic')
agg.es.id <- aggte(MP = result_id, type = 'dynamic')
# Print aggregate results
print(agg.es.did)
print(agg.es.2step)
print(agg.es.id)Working with Unbalanced Panels
The cdid library excels with unbalanced panels. Here's an example:
# Generate a dataset with missing observations based on sampling indicator S
data0 <- fonction_simu_attrition(
N = 500, T = 8,
theta2_alpha_Gg = 0.5, lambda1_alpha_St = 0,
sigma_alpha = 2, sigma_epsilon = 0.1, tprob = 0.5
)
# Keep only non-missing (S==1)
data0 <- data0[data0$S == 1, ]
# Run estimations as before, but specify panel = FALSE for did::att_gt()
did.results = did:: att_gt(
yname="Y",
tname="date",
idname = "id",
gname = "date_G",
xformla = ~X,
data = data0,
weightsname = NULL,
allow_unbalanced_panel = FALSE,
panel = FALSE,
control_group = "notyettreated",
alp = 0.05,
bstrap = TRUE,
cband = TRUE,
biters = 1000,
clustervars = NULL,
est_method = "ipw",
base_period = "varying",
print_details = FALSE,
pl = FALSE,
cores = 1
)
#For cdid, there is no difference
result_2step = att_gt_cdid(yname="Y", tname="date",
idname="id",
gname="date_G",
xformla=~X,
data=data0,
control_group="notyettreated",
alp=0.05,
bstrap=TRUE,
biters=1000,
clustervars=NULL,
cband=TRUE,
est_method="2-step",
base_period="varying",
print_details=FALSE,
pl=FALSE,
cores=1)
result_id = att_gt_cdid(yname="Y", tname="date",
idname="id",
gname="date_G",
xformla=~X,
data=data0,
control_group="notyettreated",
alp=0.05,
bstrap=TRUE,
biters=1000,
clustervars=NULL,
cband=TRUE,
est_method="Identity",
base_period="varying",
print_details=FALSE,
pl=FALSE,
cores=1)Documentation
Complete documentation and detailed examples are available through the package's help pages:
?cdid
browseVignettes("cdid")and a dedicated webpage: https://www.davidbenatia.com/projects/cdid-library/.
References
Bellego, C., Benatia, D., and Dortet-Bernadet, V. (2024). The Chained Difference-in-Differences. Journal of Econometrics. https://doi.org/10.1016/j.jeconom.2024.105783
Callaway, B., & Sant'Anna, P. H. C. (2021). Difference-in-Differences with Multiple Time Periods. Journal of Econometrics. https://doi.org/10.1016/j.jeconom.2020.12.001
License
This package is licensed under the GPL-2 license.