Package 'unitdid'

Title: Unit-level Difference-in-Difference Estimator
Description: The package provides a function to estimate the unit-level difference-in-difference estimator proposed by Arkhangelsky, Yanagimoto, and Zohar (2024).
Authors: Kazuharu Yanagimoto [aut, cre]
Maintainer: Kazuharu Yanagimoto <[email protected]>
License: MIT + file LICENSE
Version: 0.0.6.2
Built: 2025-02-23 06:18:27 UTC
Source: https://github.com/kazuyanagimoto/unitdid

Help Index

Aggregate the mean and variance of the estimated unit-level DiD effects

Description

Aggregate the mean and variance of the estimated unit-level DiD effects

Usage

aggregate_unitdid(
  object,
  agg = "full",
  na.rm = TRUE,
  by = NULL,
  normalized = NULL,
  allow_negative_var = FALSE,
  only_full_horizon = FALSE
)

Arguments

object

unitdid object
agg

Aggregation method. One of c("full", "event", "event_age") and the default is full. If by is provided in the model, all the options will separately aggregate by its group. The event option aggregates by the group of the event timing. The event_age option aggregates by the group of the age at the event time. event_age requires the bname to be provided in the model.
na.rm

Logical. If TRUE, remove NA values for the aggregation. The default is TRUE.
by

A character vector of variables to aggregate separately by. Default is inherited from the unitdid object but you can override it here. You can estimate the unit-level DiD effects separately by by in unitdid but you can also aggregate the estimates by (higher-level) by here. You can use "rel_time" as the highest level of aggregation.
normalized

Logical. If TRUE, the function will normalize the aggregated mean and variance by the mean of the imputed outcome variable. Default is inherited from the unitdid object.
allow_negative_var

Logical. If FALSE, the function will return the estimated variance trimmed at zero. Default is FALSE.
only_full_horizon

Logical. If TRUE, when you aggregate the unit-level treatment effect, only the event year (ename) with full horizon (k_min:k_max) will be included. This is recommended in the case that you do not want to change the composition of the event year (or age for the child penalties) for each estimated point in k_min:k_max. Default is FALSE.

Value

A tibble with the aggregated mean and variance of the estimated unit-level DiD effects

Simulated Individual Child Panalty Data

Description

Simulated Individual Child Panalty Data

Usage

base_heterocp

Format

base_heterocp

A dataframe with 1000 individuals for each birth year from 1965 to 1984:

id

Individual identifier

year

Year of observation

byear

Birth year

cage

Age at first birth

rel_time

Relative time to first birth

y

Outcome variable

Source

Generated by gen_heterocp() with seed 1234

Generate Sample Heterogenous Child Penalty Data

Description

Generate Sample Heterogenous Child Penalty Data

Usage

gen_heterocp(size_cohort = 300)

Arguments

size_cohort

n_obsumber of individuals per birth year

Value

A sample dataframe with heterogenous child penalty over the age at first birth

Examples

set.seed(1234)
base_heterocp <- gen_heterocp()

Get unit-level Difference-in-Differences estimates

Description

Get unit-level Difference-in-Differences estimates

Usage

get_unitdid(
  object,
  normalized = NULL,
  export = TRUE,
  only_full_horizon = FALSE
)

Arguments

object

unitdid object
normalized

Logical. If TRUE, the function will normalize them by the mean of the imputed outcome variable. Default is inherited from the unitdid object.
export

Logical. If TRUE, the function will not export the columns with the zz000 prefix, which are used in the internal computation.
only_full_horizon

Logical. If TRUE, only the event year (ename) with full horizon (k_min:k_max) will be exported. This is recommended in the case that you do not want to change the composition of the event year (or age for the child penalties) for each estimated point in k_min:k_max for aggregation. Default is FALSE.

Value

A dataframe with a new column of the unit-level DiD estimates

Aggregate the mean and variance of the estimated unit-level DiD effects

Description

Aggregate the mean and variance of the estimated unit-level DiD effects

Usage

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

Arguments

object

unitdid object
...

aggregate_unitdid arguments

Value

A tibble with the summary statistics

Examples

library(unitdid)
mdl_base <- base_heterocp |>
  unitdid(yname = "y",
          iname = "id",
          tname = "year",
          ename = "cyear",
          bname = "byear")
summary(mdl_base, agg = "event_age")

A function estimates unit-level difference-in-differences

Description

A function estimates unit-level difference-in-differences

Usage

unitdid(
  data,
  yname,
  iname,
  tname,
  ename,
  first_stage = NULL,
  wname = NULL,
  k_min = 0,
  k_max = 5,
  compute_varcov = "none",
  by = NULL,
  bname = NULL,
  normalized = FALSE,
  newnames = NULL
)

Arguments

data

The dataframe containing all the variables
yname

Outcome variable
iname

Unit identifier
tname

Time variable
ename

Event timing variable
first_stage

Formula for Y(0). Formula follows fixest::feols. If not specified, unit (iname) and time (tname) fixed effects will be used.
wname

Optional. The name of the weight variable.
k_min

Relative time to treatment at which treatment starts. Default is 0.
k_max

Relative time to treatment at which treatment ends. Default is 5.
compute_varcov

One of c("none", "var", "cov") and Default is "none". If "var", the function will estimate the unit-level variance of the outcome variable. If "cov", the function will estimate the unit-level covariance of the outcome variable for each pair within k_min:k_max.
by

A character vector of variables to estimate separately by. Default is NULL.
bname

Birth year variable. Default is NULL. Necessary to aggregate the estimates by age at event.
normalized

Logical. If TRUE, the function will normalize the outcome variable scale. Default is FALSE.
newnames

Optional. A list of new names for the output variables. ytildename is the name of the imputed outcome variable. Default is paste0(yname, "_tilde"). yvarname is the name of the unit-level variance of the outcome variable. Default is paste0(yname, "_var"). yvarrawname is the name of the raw unit-level variance of the outcome variable, which is the variance before subtracting the variance of the measurement error. Default is paste0(yname, "_varraw"). yvarerrname is the name of the unit-level variance of the measurement error. Default is paste0(yname, "_varerr"). ycovname is the name of the unit-level covariance of the outcome variable. Default is paste0(yname, "_cov"). ycovrawname is the name of the raw unit-level covariance of the outcome variable, which is the covariance before subtracting the covariance of the measurement error. Default is paste0(yname, "_covraw"). ycoverrname is the name of the unit-level covariance of the measurement error. Default is paste0(yname, "_coverr"). kprimename is the name of the relative time to treatment. This is used for the second column name of the relative time of the unit-level covariance estimation. Default is "kprime".

Value

A unitdid class object.