R/staggered_ife.R

In bcallaway11/ife: Treatment Effects in Interactive Fixed Effects Models

#' @title staggered_ife
#'
#' @description Compute treatment effects in interactive fixed effects models
#'  with a small number of time periods by exploiting staggered treatment
#'  adoption
#'
#' @inheritParams ife
#' @param nife the number of interactive fixed effects to include in the model
#'
#' @return \code{ptetools::pte_results} object
#'
#' @export
staggered_ife <- function(yname,
                          gname,
                          tname,
                          idname,
                          data,
                          nife,
                          xformla = ~1,
                          ret_ife_regs = TRUE,
                          anticipation = 0,
                          cband = TRUE,
                          alp = 0.05,
                          boot_type = "multiplier",
                          biters = 100,
                          cl = 1) {
  # set this in order to use `setup_pte` function provided by `pte` package
  required_pre_periods <- nife + 1

  res <- pte(
    yname = yname,
    gname = gname,
    tname = tname,
    idname = idname,
    data = data,
    setup_pte_fun = staggered_ife_setup_pte,
    subset_fun = ife_subset,
    attgt_fun = staggered_ife_attgt,
    nife = nife,
    xformla = xformla,
    ret_ife_regs = ret_ife_regs,
    required_pre_periods = required_pre_periods,
    anticipation = anticipation,
    cband = cband,
    alp = alp,
    boot_type = boot_type,
    biters = biters,
    cl = cl
  )

  res
}


#' @title staggered_ife2
#'
#' @description Compute treatment effects in interactive fixed effects models
#'  with a small number of time periods by exploiting staggered treatment
#'  adoption.  Unlike `staggered_ife`, this function uses all available
#'  pre-treatment periods for estimation.  This is the approach taken in
#'  Callaway and Tsyawo (2023).
#'
#' @inheritParams ife
#' @param nife the number of interactive fixed effects to include in the model
#' @param weighting_matrix which weighting matrix to use in the first step estimates.
#'  The default is "gmm" which delivers two-step gmm estimates.  Other options are
#'  "2sls" and "identity" which uses 2sls in the first stage or uses an identity
#'  weighting matrix in the first stage.
#'
#' @return \code{ptetools::pte_results} object
#'
#' @export
staggered_ife2 <- function(yname,
                           gname,
                           tname,
                           idname,
                           data,
                           nife,
                           weighting_matrix = "gmm",
                           xformla = ~1,
                           ret_ife_regs = TRUE,
                           anticipation = 0,
                           cband = TRUE,
                           alp = 0.05,
                           boot_type = "multiplier",
                           biters = 100,
                           cl = 1) {
  # set this in order to use `setup_pte` function provided by `pte` package
  required_pre_periods <- anticipation + (nife + 1)

  res <- pte(
    yname = yname,
    gname = gname,
    tname = tname,
    idname = idname,
    data = data,
    setup_pte_fun = staggered_ife_setup_data2,
    subset_fun = keep_all_pretreatment_subset,
    attgt_fun = staggered_ife_attgt2,
    nife = nife,
    weighting_matrix = weighting_matrix,
    xformla = xformla,
    ret_ife_regs = ret_ife_regs,
    required_pre_periods = required_pre_periods,
    anticipation = anticipation,
    cband = cband,
    alp = alp,
    boot_type = boot_type,
    biters = biters,
    cl = cl
  )

  res
}