In jjchern/bartik.weight: Estimating Rotemberg Weights for a Bartik "Instrument"

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

About `bartik.weight`

The package bartik.weight is an implementation in R for estimating Rotemberg weights outlined in Goldsmith-Pinkham, Sorkin and Swift (2018).

The original Stata package by Paul Goldsmith-Pinkham can be found here: https://github.com/paulgp/bartik-weight.

Installation

You can install the current version of bartik.weight from Github with:

# install.packages("devtools")
devtools::install_github("jjchern/bartik.weight")

# To uninstall the package, use:
# remove.packages("bartik.weight")

Examples

The package includes example datasets:

library(tidyverse)

# index: czone, year
bartik.weight::ADH_master

# index: czone, year, ind
bartik.weight::ADH_local

# index: year, ind
bartik.weight::ADH_global

The main dataset bartik.weight::ADH_master contains the key variables at the level of czone-year.

The Bartik instrument is formed by interacting local industry shares (variable sh_ind_ in the ADH_local tibble) and the industry growth rates (variable trade_ in the ADH_global tibble).

To estimate the Rotemberg weights, it's necessary to transform the local tibble from long to wide format:

bartik.weight::ADH_local %>%
    mutate(ind = str_glue("t{year}_sh_ind_{ind}")) %>%
    spread(ind, sh_ind_, fill = 0) %>%
    print() -> ADH_local2

Once all the data is in proper format, the bw() function will return the weight, and the just-identified IV estimates:

library(bartik.weight)

# Prepare variables in the master tibble
y = "d_sh_empl_mfg"
x = "d_tradeusch_pw"
controls = c("reg_midatl", "reg_encen", "reg_wncen", "reg_satl",
  "reg_escen", "reg_wscen", "reg_mount", "reg_pacif", "l_sh_popedu_c",
  "l_sh_popfborn", "l_sh_empl_f", "l_sh_routine33", "l_task_outsource",
  "t2", "l_shind_manuf_cbp")
weight = "timepwt48"

# Prepare variables in the local tibble
Z = setdiff(names(ADH_local_wide), c("czone", "year"))

# Prepare variables in the global tibble
G = "trade_"

# Estimate the weight (alpha) and the IV estimates (beta)
bw = bw(ADH_master, y, x, controls, weight, ADH_local2, Z, ADH_global, G)
bw

The following table replicates Panel A of Table 5 in Goldsmith-Pinkham, Sorkin and Swift (2018):

bw %>% 
    mutate(Weight = if_else(alpha > 0, "Positive", "Negative")) %>% 
    group_by(Weight) %>% 
    summarise(Sum = sum(alpha), Mean = mean(alpha)) %>% 
    knitr::kable(digits = 3, caption = "Negative and positive weights")

Below is part of Panel D of Table 5:

bw %>% 
    top_n(5, alpha) %>% 
    arrange(desc(alpha)) %>% 
    mutate(ind = case_when(
        ind == "3571" ~ "Electronic Computers",
        ind == "3944" ~ "Games and Toys",
        ind == "3651" ~ "Household Audio and Video",
        ind == "3661" ~ "Telephone Apparatus",
        ind == "3577" ~ "Computer Equipment"
    )) %>% 
    rename(g = trade_) %>% 
    knitr::kable(digits = 3, caption = "Top five Rotemberg weight industries")