Package {didintrjl}

didintrjl: Intersection Difference-in-Differences

Description

An R wrapper for the Julia package DiDInt.jl, implementing intersection difference-in-differences (DID-INT).

Details

Implements intersection difference-in-differences (DID-INT), a method developed by Karim and Webb (2025) that allows for unbiased estimation of the average effect of treatment on the treated (ATT) in settings where the common causal covariates (CCC) assumption is violated. Supports common or staggered adoption and the inclusion of covariates whose effects on the outcome of interest may vary by state, time, or both. The package interfaces with the Julia package DiDInt.jl via JuliaConnectoR.

Estimation

• didint - Estimates the ATT, returning a DiDIntObj with print(), summary(), and coef() methods.

Plotting

• didint_plot - Produces parallel trends or event study plots, returning a DiDIntPlotObj with a plot() method.

Author(s)

Eric B. Jamieson

See Also

Useful links:

• https://ebjamieson97.github.io/didintrjl/

• https://github.com/ebjamieson97/undidR

• Report bugs at https://github.com/ebjamieson97/didintrjl/issues

Extract coefficients from DiDIntObj

Description

Extract coefficients from DiDIntObj

Usage

## S3 method for class 'DiDIntObj'
coef(object, level = c("agg", "sub"), ...)

Arguments

Value

A data frame of coefficient estimates

Estimate the ATT using DID-INT

Description

didint() estimates the average effect of treatment on the treated (ATT) using intersection difference-in-differences developped by Karim & Webb (2025). The method adjusts for covariates that may vary across states, over time, or jointly by state and time. This function is an R wrapper around the Julia implementation provided in the DiDInt.jl package. For more details on the didintrjl wrapper, visit the didintrjl documentation site: https://ebjamieson97.github.io/didintrjl/. For more details on the backend implementation, see: https://ebjamieson97.github.io/DiDInt.jl/stable/

Usage

didint(
  outcome,
  state,
  time,
  data,
  gvar = NULL,
  treated_states = NULL,
  treatment_times = NULL,
  date_format = NULL,
  covariates = NULL,
  ccc = "int",
  agg = "cohort",
  weighting = "both",
  ref = NULL,
  freq = NULL,
  freq_multiplier = 1,
  start_date = NULL,
  end_date = NULL,
  nperm = 999,
  verbose = TRUE,
  seed = sample.int(1e+06, 1),
  notyet = NULL,
  hc = "hc1",
  truejack = FALSE,
  edgecase = FALSE
)

Arguments

Details

The arguments treated_states and treatment_times must be supplied such that their ordering corresponds with one another. That is, the first element of treated_states refers to the state treated at the date given by the first element of treatment_times, and so on.

Dates can be entered as strings, numbers, or Date objects. When character strings are supplied, the input format must be specified via the date_format argument (e.g. "yyyy-mm-dd").

Period grids for staggered adoption are constructed automatically, based on the inputted data. Otherwise, the period grid can be created manually using the arguments freq, freq_multiplier, start_date, and end_date. More information on this process can be seen on the DiDInt.jl documentation site: https://ebjamieson97.github.io/DiDInt.jl/stable/.

Value

An object of class DiDIntObj, a list containing the aggregate results, sub-aggregate results, and model specifications. Has associated print.DiDIntObj, summary.DiDIntObj, and coef.DiDIntObj methods.

References

Karim & Webb (2025). Good Controls Gone Bad: Difference-in-Differences with Covariates. https://arxiv.org/abs/2412.14447

MacKinnon & Webb (2020). Randomization inference for difference-in-differences with few treated clusters. doi:10.1016/j.jeconom.2020.04.024

Examples

if (Sys.getenv("NOT_CRAN") == "true" && didintrjl_ready()) {
 file_path <- system.file("extdata", "merit.csv", package = "didintrjl")
 df <- utils::read.csv(file_path)
 res <- didint("coll", "state", "year", df, verbose = FALSE,
               treated_states = c(71, 58, 64, 59, 85, 57, 72, 61, 34, 88), nperm = 399,
               treatment_times = c(1991, 1993, 1996, 1997, 1997, 1998, 1998, 1999, 2000, 2000))
 summary(res)
 
}

Make event study or parallel trends plots

Description

didint_plot() produces either event study plots or parallel trends plots depending on what is specified via the event argument. The parallel trends plots, as well as the event study plots, are created using the means residualized by covariates under different model specifications that account for different violations of the common causal covaraites (CCC) assumptions.

Usage

didint_plot(
  outcome,
  state,
  time,
  data,
  gvar = NULL,
  treated_states = NULL,
  treatment_times = NULL,
  date_format = NULL,
  covariates = NULL,
  ref = NULL,
  ccc = "all",
  event = FALSE,
  weights = TRUE,
  ci = 0.95,
  freq = NULL,
  freq_multiplier = 1,
  start_date = NULL,
  end_date = NULL,
  hc = "hc1"
)

Arguments

Details

The arguments treated_states and treatment_times must be supplied such that their ordering corresponds with one another. That is, the first element of treated_states refers to the state treated at the date given by the first element of treatment_times, and so on.

Dates can be entered as strings, numbers, or Date objects. When character strings are supplied, the input format must be specified via the date_format argument (e.g. "yyyy-mm-dd").

Period grids are constructed automatically, based on the inputted data Otherwise, the period grid can be created manually using the arguments freq, freq_multiplier, start_date, end_date. More information on this process can be seen on the didintrjl documentation site: https://ebjamieson97.github.io/didintrjl/.

Value

An object of class DiDIntPlotObj, a list containing the parallel trends data or event study data (if event is set to TRUE) and the name of the outcome variable. Has an associated plot.DiDIntPlotObj method for producing event study or parallel trends plots.

References

Karim & Webb (2025). Good Controls Gone Bad: Difference-in-Differences with Covariates. https://arxiv.org/abs/2412.14447

Examples

if (Sys.getenv("NOT_CRAN") == "true" && didintrjl_ready()) {
 file_path <- system.file("extdata", "merit.csv", package = "didintrjl")
 df <- utils::read.csv(file_path)
 res_event <- didint_plot(
   "coll", "state", "year", df, event = TRUE,
   treated_states = c(71, 58, 64, 59, 85, 57, 72, 61, 34, 88),
   treatment_times = c(1991, 1993, 1996, 1997, 1997, 1998, 1998, 1999,
                       2000, 2000),
   covariates = c("asian", "black", "male")
 )
 plot(res_event)
 
}

Check if didintrjl is ready to go

Description

Checks whether Julia is set up correctly via JuliaConnectoR and the DiDInt.jl package (version >= 0.9.6) is available. Used to guard examples and tests that require a live Julia session.

Usage

didintrjl_ready()

Value

A single logical value: TRUE if Julia, JuliaConnectoR, and DiDInt.jl (>= 0.9.6) are all available; FALSE otherwise.

Plot method for DiDIntPlotObj

Description

Plot method for DiDIntPlotObj

Usage

## S3 method for class 'DiDIntPlotObj'
plot(x, y = NULL, ccc = "all", groupmin = 3, window = NULL, ...)

Arguments

Value

A ggplot object showing either the event study plot or the parallel trends plot.

Print method for DiDIntObj

Description

Print method for DiDIntObj

Usage

## S3 method for class 'DiDIntObj'
print(x, level = c("agg", "sub"), ...)

Arguments

Value

The DiDIntObj object, returned invisibly. Called for its side effect of printing results to the console.

Summary method for DiDIntObj

Description

Summary method for DiDIntObj

Usage

## S3 method for class 'DiDIntObj'
summary(object, level = c("all", "agg", "sub"), ...)

Arguments

Value

The DiDIntObj object, returned invisibly. Called for its side effect of printing summary results to the console.