# Card (1993) Table 5 – interaction IV (parental education × college proximity).
library(tidyverse)
library(here)
library(gt)
library(AER)
source(here("04-topics/rep-card1993/Rcode/card1993-data-prep.R"))
source(here("04-topics/rep-card1993/Rcode/card1993-gt-quarto.R"))

data <- load_card93_data()
wage <- card93_wage_sample(data)

x <- c(card93_x_base, card93_fb_full)
nearc_x_f <- paste0("nearc4_f", 1:8)
for (i in 1:8) {
  wage[[nearc_x_f[i]]] <- wage$nearc4 * wage[[paste0("f", i)]]
}

rf_rhs <- paste(c("nearc4", "nearc4_low", card93_exp, x), collapse = " + ")

rf_ed <- card93_coef_se(lm(as.formula(paste("ed76 ~", rf_rhs)), data = wage), "nearc4")
rf_ed_low <- card93_coef_se(lm(as.formula(paste("ed76 ~", rf_rhs)), data = wage), "nearc4_low")
rf_w <- card93_coef_se(lm(as.formula(paste("lwage76 ~", rf_rhs)), data = wage), "nearc4")
rf_w_low <- card93_coef_se(lm(as.formula(paste("lwage76 ~", rf_rhs)), data = wage), "nearc4_low")

# Structural col 3: IV return = Wald ratio RF(nearc4_low); direct nearc4 from 2SLS.
wald_iv <- function(num, den) {
  est <- num$estimate / den$estimate
  se <- abs(est) * sqrt((num$std.error / num$estimate)^2 +
                          (den$std.error / den$estimate)^2)
  list(estimate = est, std.error = se)
}
iv_ed_3 <- wald_iv(rf_w_low, rf_ed_low)

iv_3 <- card93_iv_panel_b(
  wage, "lwage76", "ed76", "nearc4_low",
  c(x, "nearc4", "nearc4_low")
)
nearc_3 <- card93_coef_se(iv_3, "nearc4")

# Structural col 4: instruments = nearc4 × f1–f8.
inst4 <- c(nearc_x_f, "age76", "age2", x, "nearc4")
fml4 <- as.formula(
  paste(
    "lwage76 ~ ed76 + exp76 + exp2 + nearc4 +", paste(x, collapse = " + "),
    "|", paste(inst4, collapse = " + ")
  )
)
iv_4 <- ivreg(fml4, data = wage)
iv_ed_4 <- card93_coef_se(iv_4, "ed76")
nearc_4 <- card93_coef_se(iv_4, "nearc4")

paper_targets <- tribble(
  ~col, ~var, ~value, ~se,
  1L, "nearc4_ed", 0.154, 0.135,
  1L, "nearc4_low_ed", 0.462, 0.186,
  2L, "nearc4_w", 0.029, 0.024,
  2L, "nearc4_low_w", 0.043, 0.032,
  3L, "iv_ed", 0.093, 0.065,
  3L, "nearc4", 0.015, 0.029,
  4L, "iv_ed", 0.097, 0.048,
  4L, "nearc4", 0.013, 0.024
)

replication <- tibble(
  col = c(1, 1, 2, 2, 3, 3, 4, 4),
  var = c("nearc4_ed", "nearc4_low_ed", "nearc4_w", "nearc4_low_w",
          "iv_ed", "nearc4_direct", "iv_ed", "nearc4_direct"),
  estimate = c(
    rf_ed$estimate, rf_ed_low$estimate,
    rf_w$estimate, rf_w_low$estimate,
    iv_ed_3$estimate, nearc_3$estimate,
    iv_ed_4$estimate, nearc_4$estimate
  ),
  std.error = c(
    rf_ed$std.error, rf_ed_low$std.error,
    rf_w$std.error, rf_w_low$std.error,
    iv_ed_3$std.error, nearc_3$std.error,
    iv_ed_4$std.error, nearc_4$std.error
  )
)

message("Table V replication vs paper:")
print(replication)

table_data <- tibble(
  stub = c(
    "Live Near College in 1966",
    "Live College * Low Parental Education",
    "Education",
    "Family Background Variables"
  ),
  `(1)` = c(
    card93_coef_cell(rf_ed$estimate, rf_ed$std.error),
    card93_coef_cell(rf_ed_low$estimate, rf_ed_low$std.error),
    "--", "yes"
  ),
  `(2)` = c(
    card93_coef_cell(rf_w$estimate, rf_w$std.error),
    card93_coef_cell(rf_w_low$estimate, rf_w_low$std.error),
    "--", "yes"
  ),
  `(3)` = c(
    card93_coef_cell(nearc_3$estimate, nearc_3$std.error),
    "--",
    card93_coef_cell(iv_ed_3$estimate, iv_ed_3$std.error),
    "yes"
  ),
  `(4)` = c(
    card93_coef_cell(nearc_4$estimate, nearc_4$std.error),
    "--",
    card93_coef_cell(iv_ed_4$estimate, iv_ed_4$std.error),
    "yes"
  )
)

table_data <- ak91_quarto_blank_df(table_data)

gt_tbl <- table_data |>
  gt() |>
  tab_spanner(label = "Reduced Form Models", columns = c(`(1)`, `(2)`)) |>
  tab_spanner(label = "Structural Models of Earnings", columns = c(`(3)`, `(4)`)) |>
  cols_label(
    `(1)` = html("Education"),
    `(2)` = html("Earnings"),
    `(3)` = html(" "),
    `(4)` = html(" ")
  )

gt_tbl <- card93_gt_standard(
  gt_tbl,
  stub_col = "stub",
  data_cols = c("(1)", "(2)", "(3)", "(4)"),
  header_rows = NULL
)

gt_tbl <- gt_tbl |>
  card93_gt_source_notes(
    c(
      paste(
        "**Notes:** Standard errors in parentheses. Sample size is 3010.",
        "Dependent variable: log hourly wages in 1976 (mean 6.262, SD 0.444).",
        "All models include black, 1976 South/SMSA, 1966 region/SMSA, experience",
        "and experience-squared, and full family background. Experience is endogenous",
        "in structural columns; age and age-squared are instruments."
      ),
      paste(
        "**Replication:** Reduced-form columns (1)–2) use OLS; with endogenous",
        "experience, Panel B RF via `ivreg` gives slightly different point estimates.",
        "Direct effect of `nearc4` in columns (3)–4) is small and imprecise."
      )
    )
  ) |>
  card93_gt_stub_footnote(
    table_data,
    "Live College * Low Parental Education",
    paste(
      "Interaction of an indicator for living near a 4-year college in 1966 with an",
      "indicator for both parents having less than high-school education (`lowfam = 1`",
      "if `famed = 9`; `nearc4_low = nearc4` $\\times$ `lowfam`)."
    )
  ) |>
  card93_gt_stub_footnote(
    table_data,
    "Education",
    paste(
      "Column (3): instrument for schooling is `nearc4_low`; coefficient is the Wald",
      "ratio of reduced-form earnings to schooling effects (paper: 0.093). Column (4):",
      "instruments are `nearc4` $\\times$ `f1`–`nearc4` $\\times$ `f8` (paper IV $\\approx$ 0.097)."
    )
  ) |>
  card93_gt_stub_footnote(
    table_data,
    "Family Background Variables",
    paste(
      "Fourteen controls (`card93_fb_full`): parental education (years and missing",
      "indicators), eight `famed` interaction classes, and two family-structure",
      "indicators at age 14."
    )
  )

gt_tbl <- card93_gt_finalize(gt_tbl)

save(table_data, gt_tbl, replication, paper_targets, file = here("04-topics/rep-card1993/Rcode/Table_V.RData"))