EventStudy: Estimates Equation (2) in Freyaldenhoven et al. (2021)

In eventstudyr: Estimation and Visualization of Linear Panel Event Studies

Estimates Equation (2) in Freyaldenhoven et al. (2021)

Description

EventStudy uses regression methods to estimate the effect of a policy on a given outcome.

Usage

EventStudy(
  estimator,
  data,
  outcomevar,
  policyvar,
  idvar,
  timevar,
  controls = NULL,
  proxy = NULL,
  proxyIV = NULL,
  FE = TRUE,
  TFE = TRUE,
  post,
  overidpost = 1,
  pre,
  overidpre = post + pre,
  normalize = -1 * (pre + 1),
  cluster = TRUE,
  anticipation_effects_normalization = TRUE,
  allow_duplicate_id = FALSE,
  avoid_internal_copy = FALSE,
  kernel = "estimatr"
)

Arguments

estimator	Accepts one of "OLS" or "FHS". If "OLS" is specified, implements Ordinary Least Squares. If "FHS" is specified, implements Instrumental Variables (IV) estimator proposed in Freyaldenhoven Hansen Shapiro (FHS, 2019).
data	Data frame containing the variables of interest.
outcomevar	Character indicating column of outcome variable y.
policyvar	Character indicating column of policy variable z.
idvar	Character indicating column of units.
timevar	Character indicating column of time periods.
controls	Optional character vector indicating a set of control variables q.
proxy	Character indicating column of variable that is thought to be affected by the confound but not by the policy. Should be specified if and only if estimator is specified as "FHS".
proxyIV	Character of column to be used as an instrument. Should be specified if and only if estimator is specified as "FHS". If NULL, defaults to the strongest lead of the policy variable based on the first stage.
FE	Logical indicating whether unit fixed-effects should be included. Defaults to TRUE.
TFE	Logical indicating whether time fixed-effects should be included. Defaults to TRUE.
post	Whole number indicating the number of periods in the past before which the past values of the policy are not supposed to affect the value of the outcome. Corresponds to M in equation (2) of Freyaldenhoven et al. (2021).
overidpost	Optional whole number indicating the number of event times after "post" to be included in estimation. Defaults to 1. Corresponds to L_M in equation (2) of Freyaldenhoven et al. (2021).
pre	Whole number indicating the number of periods in the future after which the future values of the policy are not supposed to affect the value of the outcome today. Corresponds to G in equation (2) of Freyaldenhoven et al. (2021).
overidpre	Optional whole number indicating the number of event times earlier than -"pre" to be included in estimation. Defaults to "post" + "pre". Corresponds to L_G in equation (2) of Freyaldenhoven et al. (2021).
normalize	Specifies the event-time coefficient to be normalized. Defaults to - pre - 1.
cluster	Logical indicating whether to use clustered errors by units. If FALSE, will use unclustered heteroskedasticity-robust standard errors. Defaults to TRUE. Must be TRUE if FE is TRUE.
anticipation_effects_normalization	If set to TRUE, runs the default process and switches coefficient to be normalized to 0 when there are anticipation effects. If set to FALSE, does not make the switch. Defaults to TRUE.
allow_duplicate_id	If TRUE, the function estimates a regression where duplicated ID-time rows are weighted by their duplication count. If FALSE, the function raises an error if duplicate unit-time keys exist in the input data. Default is FALSE.
avoid_internal_copy	If TRUE, the function avoids making an internal deep copy of the input data, and instead directly modifies the input data.table. Default is FALSE.
kernel	Accepts one of "estimatr" or "fixest". If "estimatr" is specified, uses the estimatr package for estimation. If "fixest" is specified, uses the fixest package for estimation. Defaults to "estimatr" (deprecated - will change to "fixest" in a future release).

Value

A list that contains, under "output", the estimation output as an lm_robust object, and under "arguments", the arguments passed to the function.

Examples

# A minimal example
eventstudy_model <-
  EventStudy(
    estimator = "OLS",
    data = example_data,
    outcomevar = "y_base",
    policyvar = "z",
    idvar = "id",
    timevar = "t",
    pre = 0, post = 3,
    normalize = -1,
    kernel = "fixest"
  )

### Access estimated model
eventstudy_model$output

summary(eventstudy_model$output)

### data.frame of estimates
fixest::coeftable(eventstudy_model$output) # for kernel='fixest'
# estimatr::tidy(eventstudy_model$output) # for kernel='estimatr'

### Access arguments
eventstudy_model$arguments

# A dynamic OLS model with anticipation effects and controls
eventstudy_model_dyn <-
  EventStudy(
    estimator = "OLS",
    data = example_data,
    outcomevar = "y_base",
    policyvar = "z",
    idvar = "id",
    timevar = "t",
    controls = "x_r",
    FE = TRUE, TFE = TRUE,
    post = 3, overidpost = 5,
    pre  = 2, overidpre  = 4,
    normalize = - 3,
    cluster = TRUE,
    kernel = "fixest",
    anticipation_effects_normalization = TRUE
  )

summary(eventstudy_model_dyn$output)

# A static model
eventstudy_model_static <-
  EventStudy(
    estimator = "OLS",
    data = example_data,
    outcomevar = "y_jump_m",
    policyvar = "z",
    idvar = "id",
    timevar = "t",
    FE = TRUE, TFE = TRUE,
    post = 0, overidpost = 0,
    pre  = 0, overidpre  = 0,
    cluster = TRUE,
    kernel = "fixest"
  )

summary(eventstudy_model_static$output)

# A dynamic model with an unbalanced panel
data_unbal <- example_data[1:(nrow(example_data)-1),]  # drop last row to make unbalanced

eventstudy_model_unbal <-
 EventStudy(
    estimator = "OLS",
    data = data_unbal,
    outcomevar = "y_base",
    policyvar = "z",
    idvar = "id",
    timevar = "t",
    pre = 0, post = 3,
    normalize = -1,
    kernel = "fixest"
  )

summary(eventstudy_model_unbal$output)

# A dynamic model estimated using IV
eventstudy_model_iv <-
  EventStudy(
    estimator = "FHS",
    data = example_data,
    outcomevar = "y_base",
    policyvar = "z",
    idvar = "id",
    timevar = "t",
    proxy = "x_r",
    FE = TRUE, TFE = TRUE,
    post = 2, overidpost = 1,
    pre  = 0, overidpre  = 3,
    normalize = -1,
    cluster = TRUE,
    kernel = "fixest"
  )

summary(eventstudy_model_iv$output)

eventstudyr documentation built on April 5, 2026, 5:06 p.m.