Table 3 in Fafchamps et al. (2014), “Microenterprise growth and the flypaper effect: Evidence from a randomized experiment in Ghana”

To read the created .tex file, the following \(\LaTeX\) packages are required:

• float
• pdflscape
• booktabs
• multicol

Following the checklist for tables by DIME Analytics, the Constant coefficient is dropped and means of outcome variables are added instead.

# Install and load packages ---------------
packages <- c(
  "tidyverse",
  "haven",
  "lfe",
  "stargazer",
  "broom",
  "kableExtra",
  "XML",
  "magrittr",
  "pdftools"
)

pacman::p_load(packages, character.only = TRUE, install = TRUE)

# Load an example dataset ---------------
data <- read_dta("data/ReplicationDataGhanaJDE.dta")

outcome <- "realfinalprofit"

rhs_var_1 <- c("atreatcash", "atreatequip")
rhs_var_2 <- c("atreatcashfemale", "atreatequipfemale", "atreatcashmale", "atreatequipmale")

control_var_1 <- colnames(data)[grepl("^wave\\d$", colnames(data))]
control_var_2 <- colnames(data)[grepl("^wave\\d(_female)?$", colnames(data))]
control_var_3 <- colnames(data)[grepl("^wave6(_female)?$", colnames(data))]

fe_1 <- "groupnum"
fe_2 <- "sheno"
iv <- "0"
cluster <- "sheno"

# function to create formula which will be passed to lfe::felm
create_formula <- function(outcome, rhs_var, control_var, fe, iv, cluster) {
  as.formula(
    paste(
      paste(outcome, paste(c(rhs_var, control_var), collapse = " + "), sep = " ~ "),
      fe, iv, cluster, sep = " | "
    )
  )
}

# data.frame of formulas
formulas <- c(
  replicate(2, create_formula(outcome, rhs_var_1, control_var_1, fe_1, iv, cluster)),
  replicate(2, create_formula(outcome, rhs_var_1, control_var_1, fe_2, iv, cluster)),
  replicate(2, create_formula(outcome, rhs_var_2, control_var_2, fe_1, iv, cluster)),
  replicate(2, create_formula(outcome, rhs_var_2, control_var_2, fe_2, iv, cluster)),
  replicate(2, create_formula(outcome, rhs_var_2, control_var_3, fe_1, iv, cluster))
  ) %>%
  enframe("model_no", "formula")

# list of functions to filter data, which will be used to create subsamples
# in each regression
data_filter_list <- vector(mode = "list", length = 10)
for (i in seq(10)) {
  if (i %in% c(1, 3, 5, 7)) {
    data_filter_list[[i]] <- filter
  } else if (i %in% c(2, 4, 6, 8)) {
    data_filter_list[[i]] <- function (x) {filter(x, is.na(trimgroup))}
  } else if (i == 9) {
    data_filter_list[[i]] <- function (x) {filter(x, wave >= 5)}
  } else if (i == 10) {
    data_filter_list[[i]] <- function (x) {filter(x, is.na(trimgroup), wave >= 5)}
  }
}

# list of functions to conduct F-test
test_fun_list <- replicate(5, vector(mode = "list", length = 10), FALSE)
for (i in seq(10)) {
  if (i <= 4) {
    test_fun_list[[1]][[i]] <- function(x) {
      formatC(
        waldtest(x, ~ atreatcash - atreatequip)["p.F"],
        3, format = "f"
        )
      }
    test_fun_list[[2]][[i]] <- function (x) {return("")}
    test_fun_list[[3]][[i]] <- function (x) {return("")}
    test_fun_list[[4]][[i]] <- function (x) {return("")}
    test_fun_list[[5]][[i]] <- function (x) {return("")}
  } else if (i >= 5) {
    test_fun_list[[1]][[i]] <- function (x) {return("")}
    test_fun_list[[2]][[i]] <- function(x) {
      formatC(
        waldtest(x, ~ atreatcashfemale - atreatequipfemale)["p.F"],
        3, format = "f"
        )
      }
    test_fun_list[[3]][[i]] <- function(x) {
      formatC(
        waldtest(x, ~ atreatcashmale - atreatequipmale)["p.F"],
        3, format = "f"
        )
      }
    test_fun_list[[4]][[i]] <- function(x) {
      formatC(
        waldtest(x, ~ atreatcashmale - atreatcashfemale)["p.F"],
        3, format = "f"
        )
      }
    test_fun_list[[5]][[i]] <- function(x) {
      formatC(
        waldtest(x, ~ atreatequipmale - atreatequipfemale)["p.F"],
        3, format = "f"
        )
      }
  }
}

# regression results and p-values from F-test
reg_res <- formulas %>%
  mutate(
    model = map2(formula, data_filter_list, function(x, y) felm(x, y(data))),
    test_1 = map2_chr(model, test_fun_list[[1]], function(x, y) y(x)),
    test_2 = map2_chr(model, test_fun_list[[2]], function(x, y) y(x)),
    test_3 = map2_chr(model, test_fun_list[[3]], function(x, y) y(x)),
    test_4 = map2_chr(model, test_fun_list[[4]], function(x, y) y(x)),
    test_5 = map2_chr(model, test_fun_list[[5]], function(x, y) y(x)),
    n_firm = map_int(model, function(x) n_distinct(model.frame(x)$sheno)),
    mean_outcome = map_chr(model, function(x) formatC(colMeans(model.frame(x)[outcome]), 2, format = "f"))
    )

# result output as .tex file, using stargazer package
reg_res %>%
  pull(model) %>%
  stargazer(
    dep.var.labels.include = FALSE,
    column.labels = c(rep(c("OLS", "OLS", "FE", "FE"), 2), rep("OLS", 2)),
    covariate.labels = c(
      "Cash treatment",
      "In-kind treatment",
      "Cash treatment * female",
      "In-kind treatment * female",
      "Cash treatment * male",
      "In-kind treatment * male"
    ),
    keep = c(rhs_var_1, rhs_var_2),
    title = "",
    add.lines = list(
      c("Number of firms", reg_res$n_firm),
      c("Waves", rep("All", 8), rep("5 and 6", 2)),
      c("Baseline trimming", rep(c("No", "Yes"), 5)),
      c("$p$-values for testing:", rep("", 10)),
      c("\\  Cash = In-kind", reg_res$test_1),
      c("\\  Cash = In-kind for females", reg_res$test_2),
      c("\\  Cash = In-kind for males", reg_res$test_3),
      c("\\  Cash male = cash Female", reg_res$test_4),
      c("\\  In-kind male = in-kind female", reg_res$test_5),
      c("Mean of outcomes", reg_res$mean_outcome)
    ),
    type = "latex",
    out = "tex/FMQW2014_table3_replicate.tex",
    omit.stat = c("adj.rsq", "ser"),
    table.layout = "=#c-t-sa-n",
    digits = 2
  )