Package 'fdid'

Factorial Difference-in-Differences Estimation

Description

Performs factorial difference-in-differences (FDID) estimation using various methods and variance estimation techniques.

Usage

fdid(
  s,
  tr_period,
  ref_period,
  entire_period = NULL,
  method = "ols1",
  vartype = "robust",
  missing_data = c("listwise", "available"),
  nsims = 1000,
  parallel = FALSE,
  cores = 2,
  target.pop = c("all", "1", "0")
)

Arguments

s	A data frame prepared using fdid_prepare.
tr_period	A numeric vector specifying the treatment periods.
ref_period	A numeric scalar specifying the reference period.
entire_period	A numeric vector specifying the total range of time periods. If NULL, estimation is performed on all available time periods. Example: c(1958, 1959, 1960, 1961).
method	A string specifying the estimation method. Options: "ols1", "ols2", "did", "ebal", "ipw", "aipw". Default is "ols1".
vartype	A string specifying the variance estimation type. Options: "robust", "bootstrap", "jackknife". Default is "robust".
missing_data	How to handle missing data. Two options: "listwise": Drop any row missing any relevant column (including outcomes in the periods used). "available": Drop rows only if they are missing in group/covariates/cluster columns, but allow partial usage of outcomes. Default is "listwise".
nsims	Number of simulations for bootstrap variance estimation. Default is 1000.
parallel	Logical; whether to perform parallel computations. Default is FALSE.
cores	Number of cores for parallel computations. Default is 2.
target.pop	Character; the target population for averaging: "all", "1", or "0". "all" corresponds to the full sample. "1" targets the G=1 population. "0" targets the G=0 population. Default is "all".

Value

A list with the following components:

est	A list with three elements: $pre, $event, and $post containing aggregated pre-treatment, overall event, and post-treatment FDID estimates, respectively.
dynamic	Dynamic FDID estimates for each time in entire_period.
raw_means	Raw mean outcomes by group for each time in entire_period.
tr_period	Treatment periods used.
ref_period	Reference period used.
entire_period	All time periods for dynamic estimation.
method	Method used.
vartype	Variance type used.
times	All numeric time columns found.
G	Group indicator (0/1).
ps	Propensity scores (if ipw or aipw method used).
call	The matched call.
target.pop	Character indicating the target population used.

Author(s)

Rivka Lipkovitz, Enhan Liu

Examples

data(fdid)
mortality$uniqueid <- paste(mortality$provid, mortality$countyid, sep = "-")
mortality$G <- ifelse(mortality$pczupu >= median(mortality$pczupu, na.rm = TRUE), 1, 0)
s <- fdid_prepare(
  data = mortality, Y_label = "mortality",
  X_labels = c("avggrain", "lnpop"),
  G_label = "G", unit_label = "uniqueid", time_label = "year"
)
result <- fdid(s, tr_period = 1958:1961, ref_period = 1957)
summary(result)

---

Create an 'fdid_list' Object

Description

Bundles multiple 'fdid' objects into a single list with class '"fdid_list"' for convenient collective handling.

Usage

fdid_list(..., validate = TRUE)

Arguments

...	One or more objects of class '"fdid"', or a single list of them.
validate	Logical; if 'TRUE' (default) verify each element inherits from '"fdid"'.

Value

A list with classes 'c("fdid_list", "list")'.

Author(s)

Rivka Lipkovitz

---

Prepare Data for Factorial Difference-in-Differences Analysis

Description

Prepares a dataset for factorial difference-in-differences (FDID) analysis by reshaping the data into a wide format, averaging time-varying covariates, and renaming columns for consistency in subsequent analysis.

Usage

fdid_prepare(
  data,
  Y_label,
  X_labels = NULL,
  G_label,
  unit_label,
  time_label,
  cluster_label = NULL
)

Arguments

data	A data frame containing the dataset to be processed.
Y_label	A string specifying the column name of the outcome variable.
X_labels	A character vector specifying the column names of the time-varying covariates.
G_label	A string specifying the column name of the group variable (e.g., treatment vs. control).
unit_label	A string specifying the column name of the unit identifier (e.g., individual or entity).
time_label	A string specifying the column name of the time variable.
cluster_label	An optional string specifying the column name of the clustering variable. Default is 'NULL'.

Value

A data frame in wide format with the following: - Outcome variable pivoted to wide format with time columns. - Time-varying covariates averaged across time. - Columns renamed: - Unit identifier -> 'unit' - Covariates -> 'x1', 'x2', ... - Group variable -> 'G' - Clustering variable (if provided) -> 'c'

Author(s)

Rivka Lipkovitz

Examples

data <- data.frame(
  id = rep(1:3, each = 4),
  time = rep(1:4, times = 3),
  outcome = rnorm(12),
  covar1 = runif(12),
  covar2 = runif(12),
  group = c(0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1)
)
fdid_data <- fdid_prepare(
  data = data,
  Y_label = "outcome",
  X_labels = c("covar1", "covar2"),
  G_label = "group",
  unit_label = "id",
  time_label = "time"
)
head(fdid_data)

---

FDID example dataset

Description

A long-format panel dataset for demonstrating the fdid package.

Usage

mortality

Format

A data frame with 11973 rows and 17 columns:

provid

Province ID

countyid

County ID

zupu

Genealogy book count

pczupu

Genealogy book density (per capita); 45% of counties have zero

lnpczupu

Log-transformed genealogy density: log(pczupu + 1); used as a continuous treatment in Xu, Zhao, and Ding (2026)

anyzupu

Indicator: any genealogy book present

avggrain

Average grain output

nograin

Indicator: no grain data

urban

Urban population share

dis_bj

Distance to Beijing

dis_pc

Distance to provincial capital

rice

Rice cultivation indicator

minority

Minority population share

edu

Education level

lnpop

Log population

year

Year (1954–1966)

mortality

Mortality rate

---

Plot Results from FDID Analysis

Description

Provides visualisations for FDID results, including raw means, dynamic effects, and propensity-score overlap. The comparison plot of multiple methods has been removed; use plot.fdid_list() for that.

Usage

## S3 method for class 'fdid'
plot(
  x,
  type = c("raw", "dynamic", "overlap"),
  connected = FALSE,
  ci = TRUE,
  shade_periods = x$tr_period,
  alpha_shade = 0.2,
  palette = "Set2",
  group_labels = c("Group 0", "Group 1"),
  xlab = NULL,
  ylab = NULL,
  main = NULL,
  ylim = NULL,
  ...
)

Arguments

x	An fdid object.
type	One of "raw", "dynamic", or "overlap".
connected	Logical; if TRUE, connects points with lines in the "raw" and "dynamic" plots. Default is FALSE.
ci	Logical; if TRUE, draw 95% CIs when available. Default is TRUE.
shade_periods	Shaded intervals on the time axis. Default uses x$tr_period, i.e. event periods. Set to NULL to remove shaded area.
alpha_shade	Transparency for shading the treatment period.
palette	A palette name from RColorBrewer. Default "Set2".
group_labels	Labels for the two groups.
xlab, ylab, main	Axis labels and main title.
ylim	Y-axis limits. Default NULL (computed automatically).
...	Additional graphics parameters.

Value

Produces a plot; invisibly returns NULL.

Author(s)

Rivka Lipkovitz, Enhan Liu

Examples

data(fdid)
mortality$uniqueid <- paste(mortality$provid, mortality$countyid, sep = "-")
mortality$G <- ifelse(mortality$pczupu >= median(mortality$pczupu, na.rm = TRUE), 1, 0)
s <- fdid_prepare(
  data = mortality, Y_label = "mortality",
  X_labels = c("avggrain", "lnpop"),
  G_label = "G", unit_label = "uniqueid", time_label = "year"
)
result <- fdid(s, tr_period = 1958:1961, ref_period = 1957)
plot(result, type = "raw")
plot(result, type = "dynamic")

---

Plot Multiple FDID Estimates

Description

Creates a comparison plot of point estimates and confidence intervals for every element of an 'fdid_list'.

Usage

## S3 method for class 'fdid_list'
plot(
  x,
  xlab = NULL,
  ylab = NULL,
  main = NULL,
  ylim = NULL,
  vertical = TRUE,
  show_vartype = TRUE,
  ...
)

Arguments

x	An object of class '"fdid_list"'.
xlab, ylab, main	Axis labels and title. If 'NULL', sensible defaults are used.
ylim	Optional numeric vector of length two giving the *estimate-axis* limits. (Backward compatible: for horizontal plots this is the x-limit; for vertical plots this is the y-limit.)
vertical	Logical; default is TRUE.
show_vartype	Logical; include vartype in labels. Default is TRUE.
...	Additional graphics parameters passed to plot().

Value

Invisibly returns 'x'; called for its side-effect of drawing a plot.

Author(s)

Rivka Lipkovitz, Enhan Liu

---

Print Method for FDID Objects

Description

Print Method for FDID Objects

Usage

## S3 method for class 'fdid'
print(x, ...)

Arguments

x	An object of class 'fdid'.
...	Additional arguments (not used).

Value

Prints a brief overview of the 'fdid' object

Author(s)

Rivka Lipkovitz.

---

Summary Method for FDID Objects

Description

Summary Method for FDID Objects

Usage

## S3 method for class 'fdid'
summary(object, ...)

Arguments

object	An object of class fdid.
...	Additional arguments (not used).

Value

Prints a summary of the fdid object.

Author(s)

Rivka Lipkovitz, Enhan Liu

Examples

data(fdid)
mortality$uniqueid <- paste(mortality$provid, mortality$countyid, sep = "-")
mortality$G <- ifelse(mortality$pczupu >= median(mortality$pczupu, na.rm = TRUE), 1, 0)
s <- fdid_prepare(
  data = mortality, Y_label = "mortality",
  X_labels = c("avggrain", "lnpop"),
  G_label = "G", unit_label = "uniqueid", time_label = "year"
)
result <- fdid(s, tr_period = 1958:1961, ref_period = 1957)
summary(result)

---