Levy Elasticity Paper Tables

Code 
suppressPackageStartupMessages({
  library(tidyverse)
  library(haven)
  library(fixest)
  library(modelsummary)
  library(broom)
  library(glue)
  library(readr)
  library(gt)
})

options(scipen = 999)

table_dir <- fs::path("paper", "tables")

missing_output <- function(path) {
  cat(
    paste0(
      "<p><strong>Missing saved output:</strong> <code>",
      htmltools::htmlEscape(as.character(path)),
      "</code></p>",
      "<p>Render with <code>-P run_models:true</code> to create it.</p>"
    )
  )
}

display_model_table <- function(file_stub, title = NULL, notes = NULL, coef_map = NULL,
                                gof_omit = "IC|Log|Adj|RMSE|Std.Errors", gof_function = NULL, add_rows = NULL) {
  model_path <- fs::path(table_dir, paste0(file_stub, "_models.rds"))

  if (!fs::file_exists(model_path)) {
    missing_output(model_path)
    return(invisible(NULL))
  }

  models <- readRDS(model_path)

  modelsummary(
    models,
    title = title,
    notes = notes,
    coef_map = coef_map,
    stars = c("*" = .10, "**" = .05, "***" = .01),
    gof_omit = gof_omit,
    add_rows = add_rows
  )
}

display_df_table <- function(file_stub, title = NULL, digits = 3) {
  rds_path <- fs::path(table_dir, paste0(file_stub, ".rds"))

  if (!fs::file_exists(rds_path)) {
    missing_output(rds_path)
    return(invisible(NULL))
  }

  readRDS(rds_path) |>
    knitr::kable(
      format = "html",
      caption = title,
      digits = digits,
      escape = TRUE
    )
}

# ================================================================
# 4. HELPERS
# ================================================================

vcov_uid <- ~n_uniqueid

tidy_plus <- function(model, extra = NULL) {
  out <- broom::tidy(model)
  if (!is.null(extra)) {
    attr(out, "extra") <- extra
  }
  out
}

safe_wald_p <- function(model, hypothesis) {
  out <- tryCatch(fixest::wald(model, hypothesis), error = function(e) NULL)
  if (is.null(out)) return(NA_real_)
  out$p
}

safe_lincom <- function(model, combo) {
  out <- tryCatch(fixest::lincom(model, combo), error = function(e) NULL)
  if (is.null(out)) {
    return(tibble(estimate = NA_real_, ci_low = NA_real_, ci_high = NA_real_, p.value = NA_real_))
  }
  tibble(
    estimate = out$estimate,
    ci_low   = out$ci_low,
    ci_high  = out$ci_high,
    p.value  = out$p.value
  )
}

# ---- output folder ---------------------------------------------------------
# These HTML outputs are intentionally written to a stable, Quarto-friendly
# path for the replication site.
output_dir <- file.path("paper", "tables")
dir.create(output_dir, recursive = TRUE, showWarnings = FALSE)


save_table <- function(models, file_stub, title = NULL, notes = NULL,
                       coef_map = NULL, gof_omit = "IC|Log|Adj|RMSE|Std.Errors",
                       gof_function = NULL,
                       add_rows = NULL) {

  out_html <- file.path(output_dir, paste0(file_stub, ".html"))
  out_rds  <- file.path(output_dir, paste0(file_stub, "_models.rds"))

  modelsummary(
    models,
    output = out_html,
    title = title,
    notes = notes,
    coef_map = coef_map,
    stars = c("*" = .10, "**" = .05, "***" = .01),
    gof_omit = gof_omit
  )

  saveRDS(models, out_rds)
  invisible(models)
}

html_escape <- function(x) {
  x <- as.character(x)
  x <- gsub("&", "&amp;", x, fixed = TRUE)
  x <- gsub("<", "&lt;", x, fixed = TRUE)
  x <- gsub(">", "&gt;", x, fixed = TRUE)
  x
}

save_df_table <- function(df, file_stub, title = NULL, digits = 3) {
  out_html <- file.path(output_dir, paste0(file_stub, ".html"))
  out_rds  <- file.path(output_dir, paste0(file_stub, ".rds"))

  df_out <- df |>
    mutate(across(where(is.numeric), ~ round(.x, digits)))

  html <- knitr::kable(df_out, format = "html", caption = title, escape = TRUE)
  writeLines(html, out_html)
  saveRDS(df_out, out_rds)
  invisible(df_out)
}

safe_se <- function(model, term) {
  out <- tryCatch(fixest::se(model)[[term]], error = function(e) NA_real_)
  if (is.null(out)) NA_real_ else out
}

safe_coef <- function(model, term) {
  out <- tryCatch(stats::coef(model)[[term]], error = function(e) NA_real_)
  if (is.null(out)) NA_real_ else out
}

model_n <- function(model) {
  out <- tryCatch(fixest::nobs(model), error = function(e) NULL)
  if (is.null(out)) out <- tryCatch(stats::nobs(model), error = function(e) NULL)
  if (is.null(out)) out <- tryCatch(model$nobs, error = function(e) NULL)
  if (is.null(out) || length(out) == 0 || is.na(out[1])) return(NA_integer_)
  as.integer(out[1])
}

term_stats <- function(model, term = "d_eav", label = NULL) {
  est <- safe_coef(model, term)
  se  <- safe_se(model, term)
  tibble(
    model = label %||% deparse(substitute(model)),
    N = model_n(model),
    estimate = est,
    std_error = se,
    upper_bound = est + 1.645 * se,
    p_value = tryCatch(2 * pnorm(abs(est / se), lower.tail = FALSE), error = function(e) NA_real_)
  )
}

p_equal_terms <- function(model, lhs = "neg_d_eav", rhs = "pos_d_eav") {
  tryCatch(fixest::wald(model, paste0(lhs, " = ", rhs))$p, error = function(e) NA_real_)
}

lincom_sum <- function(model, terms, label = NULL) {
  b <- stats::coef(model)
  V <- tryCatch(stats::vcov(model), error = function(e) NULL)
  if (is.null(V) || !all(terms %in% names(b)) || !all(terms %in% rownames(V))) {
    return(tibble(
      model = label %||% deparse(substitute(model)),
      N = model_n(model),
      lagged_magnitude = NA_real_,
      std_error = NA_real_,
      A_upper_bound = NA_real_,
      lagged_p = NA_real_
    ))
  }
  est <- sum(b[terms])
  Vsub <- V[terms, terms, drop = FALSE]
  se <- sqrt(sum(Vsub))
  tibble(
    model = label %||% deparse(substitute(model)),
    N = model_n(model),
    lagged_magnitude = est,
    std_error = se,
    A_upper_bound = est + 1.645 * se,
    lagged_p = 2 * pnorm(abs(est / se), lower.tail = FALSE)
  )
}

sample_index <- function(model, data) {
  idx <- tryCatch(fixest::obs(model), error = function(e) NULL)
  if (!is.null(idx)) return(idx)
  # fallback: mark complete cases for variables used in the model is not perfect,
  # but prevents hard failure if fixest::obs changes.
  seq_len(nrow(data))
}

safe_first_stage_f <- function(model) {
  # fixest exposes several IV fit statistics, but names differ across versions.
  out <- tryCatch(fixest::fitstat(model, "ivf1"), error = function(e) NULL)
  if (is.null(out)) return(NA_real_)
  as.numeric(unlist(out))[1]
}

`%||%` <- function(x, y) if (is.null(x)) y else x



agency_label <- function(x) {
  recode(as.character(x),
    HR_muni = "Home-rule municipalities",
    NonHR_muni = "Non-home-rule municipalities",
    Muni = "All Munis",
    Other = "Other",
    Township = "Townships",
    School = "Schools",
    .default = as.character(x)
  )
}

Data preparation and shared helpers

This setup chunk reads the raw files, constructs the variables, defines the model helpers, and creates the results/tables/ folder. It is visible for transparency but folded by default.

Code 
# Setup ----------------------------------------------------------------------
# descriptive_stats_15_lags_cluster.R
# Rough R translation of: descriptive stats_15_lags_cluster.do
# Converted from Stata to R on 2026-03-06.
# Updated on May 11th 2026 using 22_lags_cluster.do
#
# Notes:
# - Stata's esttab output is translated to modelsummary output. You can
# - Some table formatting is approximate rather than identical.
Code 
# ================================================================
# 1. READ + MERGE DATA
# ================================================================

message(Sys.Date())
message(format(Sys.time(), "%H:%M:%S"))



# file DM used:  #"NTA_data_2024_10_14.csv"
# email shows that he had questions about this file and then michael sent one for 10_16
main_df <- read_csv("NTA_data_2024_10_14.csv")

# main_df |> distinct(agency_group)
#415 distinct agency_groups in the 10_14 file.

# File MVH sent him:
# main_df <- read_csv("NTA_data_2024_10_16.csv") |>
#   arrange(type)


# Most recent file AWM had on her computer:
# main_df <- read_csv("NTA_data_2024_11_08.csv") |>
# arrange(type)


agency_lookup <- read_dta("fips_all_agency_name.dta")

# anti_join(main_df, agency_lookup)

main_df <- main_df |>
  left_join(agency_lookup, by = "agency_group")
# 7470 observations before dropping NAs
# 7416 after dropping NAs
# Stata listed unmatched rows and dropped merge==1 rows.
# In dplyr terms: drop observations from master that did not match.
main_df <- main_df |>
  filter(!is.na(fipsid))

census_df <- read_dta("census_data.dta")

# 3,509 observations didn't have the census data and get dropped unless missing data is filled in between census years.
#anti_join(main_df, census_df)

main_df <- main_df |>
  left_join(census_df, by = c("fipsid", "year"))

# ================================================================
# 2. MERGE EQUALIZATION FACTORS + CLEAN NUMERIC FIELDS
# ================================================================

# Stata used a .dta called eq_factor after previously creating it from CSV.
# Adjust extension if your stored file is actually .csv.

eq_factor_path_dta <- file.path("eq_factor.dta")
eq_factor_path_csv <- file.path("Necessary_Files/eq_factor.csv")

eq_factor <- if (file.exists(eq_factor_path_dta)) {
  read_dta(eq_factor_path_dta)
} else if (file.exists(eq_factor_path_csv)) {
  read_csv(eq_factor_path_csv, show_col_types = FALSE)
} else {
  stop("Could not find eq_factor.dta or eq_factor.csv in post nta folder.")
}

main_df <- main_df |>
  left_join(eq_factor, by = "year")

string_num_vars <- c("assess_year_eav",  "assess_year_av", "av_true", "rate_smooth", "total_final_levy")

main_df <- main_df |>
  mutate(
    across(
      any_of(string_num_vars),
      ~ readr::parse_number(as.character(.x), na = c("NA", "", ".")),
      .names = "n_{.col}"
    )
  )

# ================================================================
# 3. PANEL SETUP + CONSTRUCTED VARIABLES
# ================================================================

main_df <- main_df |>
  mutate(
    n_agency_group = as.integer(factor(agency_group)),
    n_uniqueid     = as.integer(factor(uniqueid))
  ) |>
  arrange(n_agency_group, year) |>
  filter(!is.na(n_agency_group)) |>
  group_by(n_agency_group) |>
  arrange(year, .by_group = TRUE) |>
  mutate(
    lag_av = lag(av, 1),
    lag2_av = lag(av, 2),
    lag3_av = lag(av, 3),
    lead1_av = lead(av, 1),
    lead2_av = lead(av, 2),
    lead3_av = lead(av, 3),
    lag_eq_factor_final = lag(eq_factor_final, 1),
    lag2_eq_factor_final = lag(eq_factor_final, 2),
    lag3_eq_factor_final = lag(eq_factor_final, 3),
    lag_reassess_year = lag(reassess_year, 1),
    lag2_reassess_year = lag(reassess_year, 2)
  ) |>
  ungroup() |>
  filter(year >= 2008)

main_df <- main_df |>
  mutate(
    t = case_when(
      reassess_year == 1 ~ 1,
      lag_reassess_year == 1 ~ 2,
      lag2_reassess_year == 1 ~ 3,
    ),
    r = case_when(
      t == 1 ~ ((lead3_av / av)^(1 / 3)) - 1,
      t == 2 ~ ((lead2_av / lag_av)^(1 / 3)) - 1,
      t == 3 ~ ((lead1_av / lag2_av)^(1 / 3)) - 1,
      TRUE ~ 0
    ),
    EstV = case_when(
      t == 1 ~ av,
      t == 2 ~ lag_av * (1 + r),
      t == 3 ~ lag2_av * (1 + r)^2,
      TRUE ~ 0
    ),
    ln_EstV = log(EstV),
    d_av = 100 * log(av / lag_av)) |>
  group_by(n_agency_group) |>
  mutate(
    d_eav = 100 * log((av * eq_factor_final) / (lag_av * lag(eq_factor_final))),
    d_levy = 100 * log(total_final_levy / lag(total_final_levy)),
    d_total_ig_revenue = 100 * log(total_ig_revenue / lag(total_ig_revenue)),
    d_enrollment = 100 * log(enrollment / lag(enrollment)),
    has_ig_data = if_else(is.na(d_total_ig_revenue), 0, 1),
    type_2 = case_when(
      type == "Muni" & home_rule_ind == 1 ~ "HR_muni",

      ## Added this row below!!
      type == "Muni" & home_rule_ind == 0 ~ "NonHR_muni",
      TRUE ~ as.character(type)
    )
  ) |> ungroup()



reg_df <- main_df |>
  filter(year > 2008)


reg_df <- reg_df |>
  mutate(
    # Keep municipal types separate in by-type models.
    type_2 = case_when(
      type == "Muni" & home_rule_ind == 1 ~ "HR_muni",
      type == "Muni" & home_rule_ind == 0 ~ "NonHR_muni",
      TRUE ~ as.character(type)
    ),
    # Use a non-missing home-rule flag for year-by-home-rule fixed effects.
    home_rule_for_fe = replace_na(as.integer(home_rule_ind), 0L)
  )


gov_types <- c("Other", "Township", "HR_muni", "NonHR_muni", "School")
gov_types_B <- c("HR_muni", "NonHR_muni", "School")
minor_types <- c("ELEMENTARY", "SECONDARY")

gof_omit = "IC|Log|Adj|RMSE|Std.Errors"

Main paper tables

Table 1. Distribution of Percentage Changes in Levies and EAV

Code 
# summarize d_levy and d_eav by type
## Table 1. Distribution of Percentage Changes in Levies and EAV

table1 <- reg_df |>
  filter(year >= 2009) |>
  mutate(
    agency_type = case_when(
      type == "Muni" ~ "All Munis",
      type == "School" ~ "Schools",
      type == "Township" ~ "Townships",
      type == "Other" ~ "Other Districts"
    )
  ) |>
  select(agency_type, d_levy, d_eav) |>
  pivot_longer(
    cols = c(d_levy, d_eav),
    names_to = "stat",
    values_to = "value"
  ) |>
  group_by(agency_type, stat) |>
  summarize(
    N = sum(!is.na(value)),
    p5 = quantile(value, .05, na.rm = TRUE),
    p25 = quantile(value, .25, na.rm = TRUE),
    p50 = quantile(value, .50, na.rm = TRUE),
    p75 = quantile(value, .75, na.rm = TRUE),
    p95 = quantile(value, .95, na.rm = TRUE),
    .groups = "drop"
  )

table1 |>
  gt() |>
  fmt_number(
    columns = c(p5, p25, p50, p75, p95),
    decimals = 1
  ) |>
  tab_header(
    title = md("**Table 1. Distribution of Percentage Changes in Levies and EAV**")
  )
Table 1. Distribution of Percentage Changes in Levies and EAV
agency_type stat N p5 p25 p50 p75 p95
All Munis d_eav 1605 −15.3 −6.9 −1.5 4.2 23.8
All Munis d_levy 1601 −1.4 0.8 2.6 4.6 11.7
Other Districts d_eav 2160 −16.1 −7.2 −1.6 4.2 24.0
Other Districts d_levy 2160 −4.5 1.1 2.3 3.4 8.3
Schools d_eav 1995 −14.8 −7.1 −1.6 4.1 23.3
Schools d_levy 1995 −2.7 1.4 2.3 3.5 8.6
Townships d_eav 420 −14.3 −7.2 −1.4 3.8 20.8
Townships d_levy 420 −2.0 1.2 2.7 4.1 6.8

Table 2. All-agency OLS

“At the bottom of table 2, we also include our estimated “upper bound” for ε_b—this tells us the maximum value of the elasticity that we fail to reject with 95 percent confidence—i.e. we can reject the hypothesis that ε_b is greater than the upper bound with 95 percent confidence. As shown in the table, even these upper bounds are quite small and always much less than 1.”

Code 
all_ols_1 <- feols(d_levy ~ d_eav,
  data = reg_df, cluster = vcov_uid)

all_ols_2 <- feols(d_levy ~ d_eav | year^home_rule_for_fe,
  data = reg_df, cluster = vcov_uid)

all_ols_3 <- feols(d_levy ~ d_eav | year^home_rule_for_fe + n_uniqueid,
  data = reg_df, cluster = vcov_uid)

all_ols_models_v22 <- list(M1 = all_ols_1, M2 = all_ols_2, M3 = all_ols_3)

save_table(
  all_ols_models_v22,
  file_stub = "v22_table_02_OLS_all_agencies",
  title = "Table 2: All agencies OLS Predict levy using d_eav",
  notes = c(
    "Cook County, Illinois data from 2008 through 2023.",
    "Columns 2 and 3 include year-by-home-rule fixed effects; column 3 includes unit fixed effects."
  ),
  coef_map = c("d_eav" = "Change in EAV")
)

save_df_table(
  bind_rows(
    term_stats(all_ols_1, "d_eav", "M1"),
    term_stats(all_ols_2, "d_eav", "M2"),
    term_stats(all_ols_3, "d_eav", "M3")
  ),
  "v22_table_02_OLS_all_agencies_upper_bounds",
  "Table 2: Upper bounds for all-agency OLS"
)

# ================================================================
Code 
display_model_table(
  "v22_table_02_OLS_all_agencies",
  title = "Table 2. All-agency OLS estimates",
  notes = c(
    "Cook County, Illinois data from 2008 through 2023.",
    "Columns 2 and 3 include year-by-home-rule fixed effects; column 3 includes unit fixed effects."
  ),
  coef_map = c("d_eav" = "Change in EAV")
)
Table 2. All-agency OLS estimates
M1 M2 M3
* p < 0.1, ** p < 0.05, *** p < 0.01
Cook County, Illinois data from 2008 through 2023.
Columns 2 and 3 include year-by-home-rule fixed effects; column 3 includes unit fixed effects.
Change in EAV 0.072*** 0.089*** 0.074***
(0.010) (0.025) (0.021)
Num.Obs. 6176 6176 6174
R2 0.004 0.016 0.046
R2 Within 0.002 0.002
FE: year^home_rule_for_fe X X
FE: n_uniqueid X
Code 
display_df_table(
  "v22_table_02_OLS_all_agencies_upper_bounds",
  title = "Upper-bound summary for Table 2"
)
Upper-bound summary for Table 2
model N estimate std_error upper_bound p_value
M1 NA 0.072 0.010 0.089 0.000
M2 NA 0.089 0.025 0.130 0.000
M3 NA 0.074 0.021 0.109 0.001

Table 3. All-agency IV

“In table 3 we treat the change in the tax base (d_eav) as endogenous and instrument for it using the timing of reassessments. For the timing of reassessments to be a valid instrument for the change in the tax base requires that (i) reassessments are a strong predictor of changes in EAV and (ii) reassessments do not influence the change in the levy except to the extent that they affect the change in the tax base.”

Code 
library(fixest)
library(tibble)

# -----------------------------
# 1. Run IV models
# -----------------------------
all_iv_1 <- feols(
  d_levy ~ 1 | 0 | d_eav ~ reassess_year,
  data = reg_df,
  cluster = vcov_uid
)

all_iv_2 <- feols(
  d_levy ~ 1 | year^home_rule_for_fe | d_eav ~ reassess_year,
  data = reg_df,
  cluster = vcov_uid
)

all_iv_3 <- feols(
  d_levy ~ 1 | year^home_rule_for_fe + n_uniqueid | d_eav ~ reassess_year,
  data = reg_df,
  cluster = vcov_uid
)

all_iv_models_v22 <- list(
  all_iv_1,
  all_iv_2,
  all_iv_3
)


modelsummary(list(M1 = all_iv_1, M2 = all_iv_2, M3 = all_iv_3
  ),
    gof_omit = gof_omit,
 coef_map = c("fit_d_eav" = "Change in EAV"),
  file_stub = "IV_all_agencies",
  title = "Table X: IV Predict levy using d_eav",
  notes = c(
    "Cook County, Illinois data from 2008 through 2023.",
    "Columns 2, 3, and 4 include year fixed effects; column 4 includes unit fixed effects.",
    "d_eav is treated as endogenous and instrumented by reassessment year."
  )
)
Table X: IV Predict levy using d_eav
M1 M2 M3
Cook County, Illinois data from 2008 through 2023.
Columns 2, 3, and 4 include year fixed effects; column 4 includes unit fixed effects.
d_eav is treated as endogenous and instrumented by reassessment year.
Change in EAV -0.010 -0.085 -0.086
(0.031) (0.062) (0.062)
Num.Obs. 6176 6176 6174
R2 0.000 0.014 0.045
R2 Within 0.000 0.000
FE: year^home_rule_for_fe X X
FE: n_uniqueid X

upper bound for ε_b allows us to rule out all but tiny values for this elasticity

  • upper_bound=_b[d_eav]+(1.645 * _se[d_eav])

The p-exog statistic is the p value for a test of the hypothesis that d_eav is exogenous

We also report the F statistics from the first stage regressions to predict d_eav.

Code 
rows_to_join <-  bind_rows(
    term_stats(all_iv_1, "fit_d_eav", "M1"),
    term_stats(all_iv_2, "fit_d_eav", "M2"),
    term_stats(all_iv_3, "fit_d_eav", "M3")
  ) |>
    mutate(first_stage_F = c(safe_first_stage_f(all_iv_1),
      safe_first_stage_f(all_iv_2),
      safe_first_stage_f(all_iv_3)))

rows_to_join |> select(-N)
# A tibble: 3 × 6
  model estimate std_error upper_bound p_value first_stage_F
  <chr>    <dbl>     <dbl>       <dbl>   <dbl>         <dbl>
1 M1    -0.00979    0.0309      0.0410   0.751          921.
2 M2    -0.0854     0.0615      0.0159   0.165          988.
3 M3    -0.0857     0.0616      0.0156   0.164         1021.
Code 
# save_table(
#   all_iv_models_v22,
#   file_stub = "v22_table_03_IV_all_agencies",
#   title = "Table 3: All agencies IV Predict levy using d_eav",
#   notes = c(
#     "Cook County, Illinois data from 2008 through 2023.",
#     "Columns 2 and 3 include year-by-home-rule fixed effects; column 3 includes unit fixed effects.",
#     "d_eav is treated as endogenous and instrumented by reassessment year."
#   ),
#   coef_map = c("fit_d_eav" = "Change in EAV")
# )
# 
# save_df_table(
#   bind_rows(
#     term_stats(all_iv_1, "fit_d_eav", "M1"),
#     term_stats(all_iv_2, "fit_d_eav", "M2"),
#     term_stats(all_iv_3, "fit_d_eav", "M3")
#   ) |>
#     mutate(first_stage_F = c(
#       safe_first_stage_f(all_iv_1), 
#       safe_first_stage_f(all_iv_2), 
#       safe_first_stage_f(all_iv_3))),
#   "v22_table_03_IV_all_agencies_upper_bounds",
#   "Table 3: Upper bounds for all-agency IV"
# )

# display_model_table(
#   "v22_table_03_IV_all_agencies",
#   title = "Table 3. All-agency IV estimates",
#   notes = c(
#     "Cook County, Illinois data from 2008 through 2023.",
#     "Columns 2 and 3 include year-by-home-rule fixed effects; column 3 includes unit fixed effects.",
#     "Change in EAV is treated as endogenous and instrumented by reassessment year."
#   ),
#   coef_map = c("fit_d_eav" = "Change in EAV")
# )
# 
# display_df_table(
#   "v22_table_03_IV_all_agencies_upper_bounds",
#   title = "Upper-bound summary for Table 3"
# )

Table 4 & 5?. By agency type: OLS

Code 
m_ols_3 <- feols(d_levy ~ d_eav | year + n_uniqueid, data = reg_df, cluster = vcov_uid)
m_iv_3 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = reg_df, cluster = vcov_uid)


modelsummary(list(m_ols_3, m_iv_3), stars = TRUE, gof_omit= gof_omit)
Table 1: Compares output for the tax base elasticity using taxing agency and tax year fixed effects with OLS and IV models
(1) (2)
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
d_eav 0.073***
(0.021)
fit_d_eav -0.085
(0.063)
Num.Obs. 6174 6174
R2 0.042 0.040
R2 Within 0.002 0.000
FE: year X X
FE: n_uniqueid X X

Appendix A2 & A3

Code 
m_ols_3 <- feols(d_levy ~ d_eav | year + n_uniqueid, data = reg_df, cluster = vcov_uid, fsplit = ~type_2)
m_iv_3 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = reg_df, cluster = vcov_uid, fsplit = ~type_2)


modelsummary(list(m_ols_3), stars = TRUE, gof_omit= gof_omit)
modelsummary(list(m_iv_3), stars = TRUE, gof_omit= gof_omit)
Table 2: Compares output for the tax base elasticity for each type of taxing agency using taxing agency and tax year fixed effects with OLS and IV models. Matches Values in A2 and A3 for the 3rd row that uses TWFE.
sample: Full sample sample: HR_muni sample: NonHR_muni sample: Other sample: School sample: Township
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
d_eav 0.073*** 0.073 -0.052 0.085* 0.102* 0.010
(0.021) (0.046) (0.048) (0.036) (0.040) (0.029)
Num.Obs. 6174 1030 571 2158 1995 420
R2 0.042 0.188 0.078 0.036 0.151 0.154
R2 Within 0.002 0.010 0.001 0.001 0.026 0.000
FE: year X X X X X X
FE: n_uniqueid X X X X X X
sample: Full sample sample: HR_muni sample: NonHR_muni sample: Other sample: School sample: Township
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
fit_d_eav -0.085 -0.007 -0.060 -0.201 0.011 -0.065
(0.063) (0.070) (0.135) (0.174) (0.033) (0.089)
Num.Obs. 6174 1030 571 2158 1995 420
R2 0.040 0.180 0.077 0.036 0.128 0.155
R2 Within 0.000 0.000 0.000 0.001 0.000 0.001
FE: year X X X X X X
FE: n_uniqueid X X X X X X

Upper bound estimates, all model specifications

Code 
# ================================================================

run_type_v22 <- function(df, gov, iv = FALSE) {
  d <- df |> filter(type_2 == gov)
  if (!iv) {
    m1 <- feols(d_levy ~ d_eav, data = d, cluster = vcov_uid)
    m2 <- feols(d_levy ~ d_eav | year, data = d, cluster = vcov_uid)
    m3 <- feols(d_levy ~ d_eav | year + n_uniqueid, data = d, cluster = vcov_uid)
    return(list(m1 = m1, m2 = m2, m3 = m3))
  }
  m1 <- feols(d_levy ~ 1 | 0 | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  m2 <- feols(d_levy ~ 1 | year | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  m3 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  list(m1 = m1, m2 = m2, m3 = m3)
}

run_type_B_v22 <- function(df, gov, iv = FALSE) {
  d <- df |> filter(type_2 == gov)
  if (!iv) {
    m4 <- feols(d_levy ~ d_eav + d_total_ig_revenue | year + n_uniqueid, data = d, cluster = vcov_uid)
    idx <- sample_index(m4, d)
    m5 <- feols(d_levy ~ d_eav | year + n_uniqueid, data = d[idx, ], cluster = vcov_uid)
    return(list(m4 = m4, m5 = m5))
  }
  m4 <- feols(d_levy ~ d_total_ig_revenue | year + n_uniqueid | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  idx <- sample_index(m4, d)
  m5 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = d[idx, ], cluster = vcov_uid)
  list(m4 = m4, m5 = m5)
}

school_extra_v22 <- function(df, iv = FALSE) {
  d <- df |> filter(type_2 == "School")
  if (!iv) {
    return(feols(d_levy ~ d_eav + d_enrollment | year + n_uniqueid, data = d, cluster = vcov_uid))
  }
  feols(d_levy ~ d_enrollment | year + n_uniqueid | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
}

make_by_type_summary <- function(models_A, models_B, school_extra, iv = FALSE) {
  term <- if (iv) "fit_d_eav" else "d_eav"
  rows <- list()
  for (spec in paste0("m", 1:3)) {
    rows[[spec]] <- map_dfr(gov_types, function(gov) {
      term_stats(models_A[[gov]][[spec]], term, gov) |>
        transmute(row = spec, agency = model, N, upper_bound, estimate, std_error, p_value)
    })
  }
  for (spec in paste0("m", 4:5)) {
    rows[[spec]] <- map_dfr(gov_types_B, function(gov) {
      term_stats(models_B[[gov]][[spec]], term, gov) |>
        transmute(row = spec, agency = model, N, upper_bound, estimate, std_error, p_value)
    })
  }
  rows[["m6"]] <- term_stats(school_extra, term, "School") |>
    transmute(row = "m6", agency = model, N, upper_bound, estimate, std_error, p_value)
  bind_rows(rows) |>
    mutate(
      agency = agency_label(agency),
      row = recode(row,
      m1 = "No controls",
      m2 = "Year FE",
      m3 = "Year + unit FE",
      m4 = "IG revenue + year + unit FE",
      m5 = "Same sample as IG row, no IG control",
      m6 = "Enrollment + year + unit FE"
    ))
}

ols_type_A <- setNames(map(gov_types, ~ run_type_v22(reg_df, .x, iv = FALSE)), gov_types)
ols_type_B <- setNames(map(gov_types_B, ~ run_type_B_v22(reg_df, .x, iv = FALSE)), gov_types_B)
ols_school_extra <- school_extra_v22(reg_df, iv = FALSE)

iv_type_A <- setNames(map(gov_types, ~ run_type_v22(reg_df, .x, iv = TRUE)), gov_types)
iv_type_B <- setNames(map(gov_types_B, ~ run_type_B_v22(reg_df, .x, iv = TRUE)), gov_types_B)
iv_school_extra <- school_extra_v22(reg_df, iv = TRUE)

ols_by_type_summary <- make_by_type_summary(ols_type_A, ols_type_B, ols_school_extra, iv = FALSE)

iv_by_type_summary <- make_by_type_summary(iv_type_A, iv_type_B, iv_school_extra, iv = TRUE)
# 
# save_df_table(ols_by_type_summary, "v22_table_04_OLS_by_type_upper_bounds", "Table 4: By agency type OLS upper-bound estimates")
# save_df_table(ols_by_type_summary |> select(row, agency, N, estimate, std_error, p_value), "v22_table_A2_OLS_by_type_point_estimates", "Table A2: By agency type OLS point estimates")
# 
# 
# save_df_table(iv_by_type_summary, "v22_table_05_IV_by_type_upper_bounds", "Table 5: By agency type IV upper-bound estimates")
# save_df_table(iv_by_type_summary |> select(row, agency, N, estimate, std_error, p_value), "v22_table_A3_IV_by_type_point_estimates", "Table A3: By agency type IV point estimates")
Code 
display_df_table(
  "v22_table_04_OLS_by_type_upper_bounds",
  title = "Table 4. By agency type OLS upper-bound estimates"
)
Table 4. By agency type OLS upper-bound estimates
row agency N upper_bound estimate std_error p_value
No controls Other NA 0.123 0.093 0.018 0.000
No controls Townships NA 0.052 0.006 0.028 0.835
No controls Home-rule municipalities NA 0.062 0.018 0.026 0.485
No controls Non-home-rule municipalities NA 0.075 0.039 0.022 0.079
No controls Schools NA 0.126 0.096 0.018 0.000
Year FE Other NA 0.180 0.109 0.043 0.012
Year FE Townships NA 0.057 0.010 0.029 0.737
Year FE Home-rule municipalities NA 0.163 0.079 0.051 0.117
Year FE Non-home-rule municipalities NA 0.044 -0.036 0.048 0.460
Year FE Schools NA 0.187 0.114 0.044 0.009
Year + unit FE Other NA 0.144 0.085 0.036 0.018
Year + unit FE Townships NA 0.058 0.010 0.029 0.741
Year + unit FE Home-rule municipalities NA 0.148 0.073 0.046 0.115
Year + unit FE Non-home-rule municipalities NA 0.026 -0.052 0.048 0.272
Year + unit FE Schools NA 0.167 0.102 0.040 0.010
IG revenue + year + unit FE Home-rule municipalities NA 0.412 0.228 0.112 0.041
IG revenue + year + unit FE Non-home-rule municipalities NA 0.379 0.139 0.146 0.342
IG revenue + year + unit FE Schools NA 0.265 0.157 0.066 0.018
Same sample as IG row, no IG control Home-rule municipalities NA 0.410 0.228 0.111 0.040
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.348 0.127 0.135 0.346
Same sample as IG row, no IG control Schools NA 0.264 0.156 0.066 0.018
Enrollment + year + unit FE Schools NA 0.264 0.155 0.066 0.019
Code 
display_df_table(
  "v22_table_05_IV_by_type_upper_bounds",
  title = "Table 5. By agency type IV upper-bound estimates"
)
Table 5. By agency type IV upper-bound estimates
row agency N upper_bound estimate std_error p_value
No controls Other NA 0.079 -0.052 0.080 0.512
No controls Townships NA 0.042 -0.058 0.061 0.341
No controls Home-rule municipalities NA 0.036 -0.037 0.044 0.406
No controls Non-home-rule municipalities NA 0.061 -0.029 0.055 0.594
No controls Schools NA 0.108 0.065 0.026 0.013
Year FE Other NA 0.085 -0.201 0.174 0.248
Year FE Townships NA 0.082 -0.065 0.089 0.465
Year FE Home-rule municipalities NA 0.108 -0.007 0.070 0.919
Year FE Non-home-rule municipalities NA 0.165 -0.052 0.132 0.691
Year FE Schools NA 0.065 0.011 0.033 0.733
Year + unit FE Other NA 0.085 -0.201 0.174 0.248
Year + unit FE Townships NA 0.082 -0.065 0.089 0.465
Year + unit FE Home-rule municipalities NA 0.107 -0.007 0.070 0.914
Year + unit FE Non-home-rule municipalities NA 0.163 -0.060 0.135 0.658
Year + unit FE Schools NA 0.065 0.011 0.033 0.741
IG revenue + year + unit FE Home-rule municipalities NA 0.302 -0.240 0.330 0.466
IG revenue + year + unit FE Non-home-rule municipalities NA 0.339 -0.141 0.292 0.629
IG revenue + year + unit FE Schools NA 0.128 -0.037 0.100 0.714
Same sample as IG row, no IG control Home-rule municipalities NA 0.300 -0.243 0.330 0.462
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.323 -0.142 0.283 0.616
Same sample as IG row, no IG control Schools NA 0.130 -0.035 0.101 0.728
Enrollment + year + unit FE Schools NA 0.121 -0.045 0.101 0.658

Table 6 & 7. All-agency asymmetric models

TWFE specifications

Code 
# my way of coding it
reg_df <- reg_df |>
  mutate(
    eav_growth = case_when(
      d_eav > 0 ~ "Increase",
      d_eav < 0 ~ "Decrease",
      TRUE ~ "No Change"))


reg_B <- feols(d_levy ~ d_eav:eav_growth | year + n_uniqueid, data = reg_df, cluster = vcov_uid)
reg_B |> summary()
OLS estimation, Dep. Var.: d_levy
Observations: 6,174
Fixed-effects: year: 15,  n_uniqueid: 419
Standard-errors: Clustered (n_uniqueid) 
                         Estimate Std. Error  t value Pr(>|t|)    
d_eav:eav_growthDecrease 0.041102   0.088167 0.466180 0.641330    
d_eav:eav_growthIncrease 0.081360   0.034054 2.389137 0.017331 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 12.7     Adj. R2: -0.030748
             Within R2:  0.001709
Code 
reg_B <- feols(d_levy ~ d_eav * eav_growth | year^home_rule_ind + n_uniqueid, data = reg_df, cluster = vcov_uid)
reg_B |> summary()
OLS estimation, Dep. Var.: d_levy
Observations: 6,174
Fixed-effects: year^home_rule_ind: 30,  n_uniqueid: 419
Standard-errors: Clustered (n_uniqueid) 
                         Estimate Std. Error  t value Pr(>|t|) 
d_eav                    0.031442   0.082986 0.378889  0.70496 
eav_growthIncrease       0.303799   0.418813 0.725380  0.46862 
d_eav:eav_growthIncrease 0.045254   0.110278 0.410360  0.68175 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 12.6     Adj. R2: -0.028899
             Within R2:  0.00179

d_eav:eav_growth is the difference between his neg_d_eav = 0.43) and pos_d_eav = 0.082*. just another way to estimate it.

Table 6 trial

OLS

Code 
library(fixest)
library(ivreg)
library(modelsummary)

# ==============================================================================
# 1. NEW EXTRACTOR FOR FIXEST MODELS (feols)
# ==============================================================================
glance_custom.fixest <- function(x, ...) {
  # fixest objects naturally carry clustered/robust SEs in standard extraction functions
  beta_vector <- coef(x)
  se_vector   <- se(x)  # fixest specific function to pull adjusted SEs
  
  # Calculate upper bounds dynamically for whatever variables exist in the column
  ub_neg <- if ("neg_d_eav" %in% names(beta_vector)) as.numeric(beta_vector["neg_d_eav"] + (1.645 * se_vector["neg_d_eav"])) else NA
  ub_pos <- if ("pos_d_eav" %in% names(beta_vector)) as.numeric(beta_vector["pos_d_eav"] + (1.645 * se_vector["pos_d_eav"])) else NA
  ub_all <- if ("d_eav"     %in% names(beta_vector)) as.numeric(beta_vector["d_eav"]     + (1.645 * se_vector["d_eav"]))     else NA

  return(data.frame(
    "regF"        = NA, 
    "p_wu"        = NA, 
    "ub_d_eav"    = ub_all,
    "ub_neg_eav"  = ub_neg, 
    "ub_pos_eav"  = ub_pos
  ))
}

# ==============================================================================
# 2. UPDATED EXTRACTOR FOR IV MODELS (ivreg)
# ==============================================================================
glance_custom.ivreg <- function(x, ...) {
  diag_tests  <- summary(x, diagnostics = TRUE)$diagnostics
  regF_val    <- diag_tests["Weak instruments", "statistic"]
  p_wu_val    <- diag_tests["Wu-Hausman", "p-value"]
  
  beta_vector <- coef(x)
  se_vector   <- sqrt(diag(vcov(x)))
  
  ub_neg <- if ("neg_d_eav" %in% names(beta_vector)) as.numeric(beta_vector["neg_d_eav"] + (1.645 * se_vector["neg_d_eav"])) else NA
  ub_pos <- if ("pos_d_eav" %in% names(beta_vector)) as.numeric(beta_vector["pos_d_eav"] + (1.645 * se_vector["pos_d_eav"])) else NA
  ub_all <- if ("d_eav"     %in% names(beta_vector)) as.numeric(beta_vector["d_eav"]     + (1.645 * se_vector["d_eav"]))     else NA

  return(data.frame(
    "regF"        = regF_val, 
    "p_wu"        = p_wu_val, 
    "ub_d_eav"    = ub_all,
    "ub_neg_eav"  = ub_neg, 
    "ub_pos_eav"  = ub_pos
  ))
}

# ==============================================================================
# 3. EXPANDED TABLE MAP
# ==============================================================================
# 
# custom_gof <- modelsummary::gof_map
# custom_gof <- rbind(custom_gof,
#                     data.frame(raw = "regF",        clean = "First-Stage F",             fmt = 2, omit = FALSE),
#                     data.frame(raw = "p_wu",        clean = "Wu-Hausman (p-value)",      fmt = 3, omit = FALSE),
#                     data.frame(raw = "ub_d_eav", clean = "Upper Bound",   fmt = 3, omit = FALSE),
#                     data.frame(raw = "ub_neg_eav", clean = "Upper Bound",   fmt = 3, omit = FALSE),
#                     data.frame(raw = "ub_pos_eav", clean = "Upper Bound",   fmt = 3, omit = FALSE)
#                     
# )
# 
# 
# custom_gof <- data.frame(
#   raw   = c("nobs", "r.squared", "regF", "p_wu", "ub_d_eav", "ub_neg_eav", "ub_pos_eav"),
#   clean = c("Num. Obs.", "R²", "First-Stage F", "Wu-Hausman (p-value)", 
#             "90% UB (d_eav)", "90% UB (neg_d_eav)", "90% UB (pos_d_eav)"),
#   fmt   = c(0, 3, 2, 3, 3, 3, 3),
#   omit  = c(FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE)
# )
Code 
# DFM's way of coding it
reg_df <- reg_df |>
  mutate(
    eav_growth = if_else(d_eav > 0, 1, 0),
    pos_d_eav = d_eav * eav_growth,
    neg_d_eav = d_eav * (1 - eav_growth),
  )

first_stage_asym <- feols(d_eav ~ reassess_year | year, data = reg_df)

reg_df <- reg_df |>
  mutate(
    d_eav_hat = fitted(first_stage_asym),
    pos_d_eav_hat = pos_d_eav * d_eav_hat,
    neg_d_eav_hat = neg_d_eav * d_eav_hat
  )


asym_ols_1 <- feols(d_levy ~ neg_d_eav + pos_d_eav, data = reg_df, cluster = vcov_uid)
asym_ols_2 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year^home_rule_for_fe, data = reg_df, cluster = vcov_uid)
asym_ols_3 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year^home_rule_for_fe + n_uniqueid, data = reg_df, cluster = vcov_uid)

asym_ols_models_v22 <- list(M1 = asym_ols_1, M2 = asym_ols_2, M3 = asym_ols_3)


asym_iv_1 <- feols(d_levy ~ 1 | 0 | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat,
  data = reg_df, cluster = vcov_uid)
asym_iv_2 <- feols(d_levy ~ 1 | year^home_rule_for_fe | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat,
  data = reg_df, cluster = vcov_uid)
asym_iv_3 <- feols(d_levy ~ 1 | year^home_rule_for_fe + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat,
  data = reg_df, cluster = vcov_uid)

asym_iv_models_v22 <- list(M1 = asym_iv_1, M2 = asym_iv_2, M3 = asym_iv_3)
# 
# row3 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year^home_rule_for_fe + n_uniqueid, data = reg_df, cluster = vcov_uid)


# 2. Build a fresh, clean gof_map from scratch (Highly Recommended)
# This prevents default hidden rules from overriding or hiding your custom row
custom_gof <- data.frame(
  raw   = c("nobs", "r.squared", "regF", "p_wu", "upper_bound"),
  clean = c("Num. Obs.", "R²", "First-Stage F", "Wu-Hausman (p-value)", "90% Upper Bound (d_eav)"),
  fmt   = c(0, 3, 2, 3, 3),
  omit  = c(FALSE, FALSE, FALSE, FALSE, FALSE)
)

modelsummary(asym_ols_models_v22, stars = TRUE, title = "Table 6 in Paper", gof_omit = gof_omit,
             gof_map = custom_gof)
Table 6 in Paper
M1 M2 M3
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
(Intercept) 2.620***
(0.189)
neg_d_eav 0.078** 0.079 0.043
(0.025) (0.086) (0.088)
pos_d_eav 0.068*** 0.092* 0.082*
(0.017) (0.036) (0.035)
Num. Obs. 6176 6176 6174
R² 0.004 0.016 0.046

Using neg_d_eav_hat

Code 
asym_ols_3 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year^home_rule_for_fe + n_uniqueid, data = reg_df, cluster = vcov_uid)

asym_iv_3 <- feols(d_levy ~ 1 | year^home_rule_for_fe + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + neg_d_eav_hat,
  data = reg_df, cluster = vcov_uid)

modelsummary(list(asym_ols_3, asym_iv_3), stars = TRUE, title = "Table 6 in Paper, TWFE specifications", gof_omit = gof_omit)
Table 6 in Paper, TWFE specifications
(1) (2)
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
neg_d_eav 0.043
(0.088)
pos_d_eav 0.082*
(0.035)
fit_neg_d_eav 0.086
(0.083)
fit_pos_d_eav -0.040
(0.056)
Num.Obs. 6174 6174
R2 0.046 0.045
R2 Within 0.002 0.000
FE: year^home_rule_for_fe X X
FE: n_uniqueid X X
ub_pos_eav 0.139522350054269
ub_neg_eav 0.188324814015241

IVs

Code 
modelsummary(list(asym_iv_1, asym_iv_2, asym_iv_3),
             gof_map = "nobs|r.squared|regF|p.value|p.value.Weak.instrument|p.value.Wu.Hausman|ub_d_eav|ub_neg_eav|ub_pos_eav"

            #   c("nobs", "r.squared", "regF", "p.value", "p.value.Weak.instrument","p.value.Wu.Hausman", "ub_d_eav", "ub_neg_eav", "ub_pos_eav")
             )
(1) (2) (3)
(Intercept) 3.282
(0.363)
fit_neg_d_eav 0.187 0.372 0.086
(0.087) (0.308) (0.083)
fit_pos_d_eav 0.014 0.037 -0.040
(0.018) (0.044) (0.056)
Code 
#"nobs|r.squared|regF|p.value|p.value.Weak.instrument|p.value.Wu.Hausman|ub_d_eav|ub_neg_eav|ub_pos_eav"

all models

And then by Type.

Code 
# ================================================================
# Tables 6/A4/A5 and 7/A6/A7. Asymmetric OLS + IV
# ================================================================

reg_df <- reg_df |>
  mutate(
    eav_growth = if_else(d_eav > 0, 1, 0),
    pos_d_eav = d_eav * eav_growth,
    neg_d_eav = d_eav * (1 - eav_growth)
  )

first_stage_asym <- feols(d_eav ~ reassess_year | year, data = reg_df)
reg_df <- reg_df |>
  mutate(
    d_eav_hat = fitted(first_stage_asym),
    pos_d_eav_hat = pos_d_eav * d_eav_hat
  )

asym_ols_1 <- feols(d_levy ~ neg_d_eav + pos_d_eav, data = reg_df, cluster = vcov_uid)
asym_ols_2 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year^home_rule_for_fe, data = reg_df, cluster = vcov_uid)
asym_ols_3 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year^home_rule_for_fe + n_uniqueid, data = reg_df, cluster = vcov_uid)

asym_ols_models_v22 <- list(M1 = asym_ols_1, M2 = asym_ols_2, M3 = asym_ols_3)

save_table(
  asym_ols_models_v22,
  file_stub = "v22_table_06_OLS_all_agencies_asym",
  title = "Table 6: All agencies OLS allowing for asymmetry",
  coef_map = c("pos_d_eav" = "Positive EAV change", "neg_d_eav" = "Negative EAV change")
)

save_df_table(
  bind_rows(
    term_stats(asym_ols_1, "pos_d_eav", "M1"),
    term_stats(asym_ols_2, "pos_d_eav", "M2"),
    term_stats(asym_ols_3, "pos_d_eav", "M3")
  ) |>
    mutate(p_equal = c(p_equal_terms(asym_ols_1), p_equal_terms(asym_ols_2), p_equal_terms(asym_ols_3))),
  "v22_table_06_OLS_all_agencies_asym_upper_bounds",
  "Table 6: Upper bounds for positive EAV changes"
)

asym_iv_1 <- feols(d_levy ~ 1 | 0 | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat,
  data = reg_df, cluster = vcov_uid)
asym_iv_2 <- feols(d_levy ~ 1 | year^home_rule_for_fe | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat,
  data = reg_df, cluster = vcov_uid)
asym_iv_3 <- feols(d_levy ~ 1 | year^home_rule_for_fe + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat,
  data = reg_df, cluster = vcov_uid)

asym_iv_models_v22 <- list(M1 = asym_iv_1, M2 = asym_iv_2, M3 = asym_iv_3)

save_table(
  asym_iv_models_v22,
  file_stub = "v22_table_07_IV_all_agencies_asym",
  title = "Table 7: All agencies IV allowing for asymmetry",
  coef_map = c("fit_pos_d_eav" = "Positive EAV change", "fit_neg_d_eav" = "Negative EAV change")
)

save_df_table(
  bind_rows(
    term_stats(asym_iv_1, "fit_pos_d_eav", "M1"),
    term_stats(asym_iv_2, "fit_pos_d_eav", "M2"),
    term_stats(asym_iv_3, "fit_pos_d_eav", "M3")
  ) |>
    mutate(p_equal = c(p_equal_terms(asym_iv_1, "fit_neg_d_eav", "fit_pos_d_eav"), p_equal_terms(asym_iv_2, "fit_neg_d_eav", "fit_pos_d_eav"), p_equal_terms(asym_iv_3, "fit_neg_d_eav", "fit_pos_d_eav"))),
  "v22_table_07_IV_all_agencies_asym_upper_bounds",
  "Table 7: Upper bounds for positive EAV changes, IV"
)
Code 
display_model_table(
  "v22_table_06_OLS_all_agencies_asym",
  title = "Table 6. All-agency OLS estimates allowing for asymmetry",
  coef_map = c("pos_d_eav" = "Positive EAV change", "neg_d_eav" = "Negative EAV change")
)
Table 6. All-agency OLS estimates allowing for asymmetry
M1 M2 M3
* p < 0.1, ** p < 0.05, *** p < 0.01
Positive EAV change 0.068*** 0.092** 0.082**
(0.017) (0.036) (0.035)
Negative EAV change 0.078*** 0.079 0.043
(0.025) (0.086) (0.088)
Num.Obs. 6176 6176 6174
R2 0.004 0.016 0.046
R2 Within 0.002 0.002
ub_neg_eav 0.118922469269611 0.219797129834847 0.188324814015241
ub_pos_eav 0.0965482515794449 0.151865951699359 0.139522350054269
FE: year^home_rule_for_fe X X
FE: n_uniqueid X
Code 
display_df_table(
  "v22_table_06_OLS_all_agencies_asym_upper_bounds",
  title = "Upper-bound summary for Table 6"
)
Upper-bound summary for Table 6
model N estimate std_error upper_bound p_value p_equal
M1 NA 0.068 0.017 0.097 0.000 NA
M2 NA 0.092 0.036 0.152 0.011 NA
M3 NA 0.082 0.035 0.140 0.019 NA
Code 
display_model_table(
  "v22_table_07_IV_all_agencies_asym",
  title = "Table 7. All-agency IV estimates allowing for asymmetry",
  coef_map = c("fit_pos_d_eav" = "Positive EAV change", "fit_neg_d_eav" = "Negative EAV change")
)
Table 7. All-agency IV estimates allowing for asymmetry
M1 M2 M3
* p < 0.1, ** p < 0.05, *** p < 0.01
Positive EAV change 0.014 0.037 0.027
(0.018) (0.044) (0.039)
Negative EAV change 0.187** 0.372 0.336
(0.087) (0.308) (0.290)
Num.Obs. 6176 6176 6174
R2 0.001 0.015 0.046
R2 Within 0.001 0.001
FE: year^home_rule_for_fe X X
FE: n_uniqueid X
Code 
display_df_table(
  "v22_table_07_IV_all_agencies_asym_upper_bounds",
  title = "Upper-bound summary for Table 7"
)
Upper-bound summary for Table 7
model N estimate std_error upper_bound p_value p_equal
M1 NA 0.014 0.018 0.043 0.442 NA
M2 NA 0.037 0.044 0.109 0.405 NA
M3 NA 0.027 0.039 0.091 0.498 NA

Table 8 & 9. Asymmetric models by agency type

main row row3 TWFE

Code 
row3_tab8_ols <- feols(d_levy ~ neg_d_eav + pos_d_eav | year + n_uniqueid, data = reg_df, cluster = vcov_uid, fsplit = ~type_2)


modelsummary(list(row3_tab8_ols), stars = TRUE)
sample: Full sample sample: HR_muni sample: NonHR_muni sample: Other sample: School sample: Township
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
neg_d_eav 0.041 0.076 -0.273 0.002 0.131** 0.211*
(0.088) (0.062) (0.167) (0.197) (0.044) (0.085)
pos_d_eav 0.081* 0.072 0.016 0.110 0.096+ -0.030
(0.034) (0.058) (0.054) (0.068) (0.049) (0.039)
Num.Obs. 6174 1030 571 2158 1995 420
R2 0.042 0.188 0.082 0.036 0.151 0.161
R2 Adj. -0.031 0.115 -0.014 -0.041 0.081 0.065
R2 Within 0.002 0.010 0.006 0.001 0.027 0.009
R2 Within Adj. 0.001 0.007 0.002 0.000 0.026 0.003
AIC 49730.4 6529.5 4151.4 19372.6 11794.0 2660.7
BIC 52657.1 6954.1 4390.5 20286.6 12650.6 2838.5
RMSE 12.65 5.30 8.33 19.99 4.31 5.17
Std.Errors by: n_uniqueid by: n_uniqueid by: n_uniqueid by: n_uniqueid by: n_uniqueid by: n_uniqueid
FE: year X X X X X X
FE: n_uniqueid X X X X X X
ub_neg_eav 0.186136338077469 0.1779068768088 0.00176302661826211 0.326032198627545 0.204052753458301 0.351131422043423
ub_pos_eav 0.137379580830717 0.167544301206615 0.105872014890694 0.222206051897779 0.177179176545785 0.0333288831571422
Code 
row3_tab9_iv <- feols(d_levy ~ 1 | year + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = reg_df, cluster = vcov_uid)
  
modelsummary(list(row3_tab9_iv), stars = TRUE)
(1)
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
fit_neg_d_eav 0.339
(0.289)
fit_pos_d_eav 0.028
(0.038)
Num.Obs. 6174
R2 0.041
R2 Adj. -0.031
R2 Within 0.001
R2 Within Adj. 0.001
AIC 49759.4
BIC 52686.1
RMSE 12.68
Std.Errors by: n_uniqueid
FE: year X
FE: n_uniqueid X

combined tables

Code 
run_asym_type_v22 <- function(df, gov, iv = FALSE) {
  d <- reg_df |> filter(type_2 %in% gov_types)
  if (!iv) {
    m1 <- feols(d_levy ~ neg_d_eav + pos_d_eav, data = d, cluster = vcov_uid)
    m2 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year, data = d, cluster = vcov_uid)
    m3 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year + n_uniqueid, data = d, cluster = vcov_uid)
    return(list(m1 = m1, m2 = m2, 
                m3 = m3))
  }
  m1 <- feols(d_levy ~ 1 | 0 | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = d, cluster = vcov_uid)
  m2 <- feols(d_levy ~ 1 | year | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = d, cluster = vcov_uid)
  m3 <- feols(d_levy ~ 1 | year + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = d, cluster = vcov_uid)
  list(m1 = m1, m2 = m2, 
       m3 = m3)

}



run_asym_type_B_v22 <- function(df, gov, iv = FALSE) {
  d <- df |> filter(type_2 == gov)
  if (!iv) {
    m4 <- feols(d_levy ~ neg_d_eav + pos_d_eav + d_total_ig_revenue | year + n_uniqueid, data = d, cluster = vcov_uid)
    idx <- sample_index(m4, d)
    m5 <- feols(d_levy ~ neg_d_eav + pos_d_eav | year + n_uniqueid, data = d[idx, ], cluster = vcov_uid)
    return(list(m4 = m4, m5 = m5))
  }
  m4 <- feols(d_levy ~ d_total_ig_revenue | year + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = d, cluster = vcov_uid)
  idx <- sample_index(m4, d)
  m5 <- feols(d_levy ~ 1 | year + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = d[idx, ], cluster = vcov_uid)
  list(m4 = m4, m5 = m5)
}

asym_school_extra_v22 <- function(df, iv = FALSE) {
  d <- df |> filter(type_2 == "School")
  if (!iv) {
    return(feols(d_levy ~ neg_d_eav + pos_d_eav + d_enrollment | year + n_uniqueid, data = d, cluster = vcov_uid))
  }
  feols(d_levy ~ d_enrollment | year + n_uniqueid | neg_d_eav + pos_d_eav ~ reassess_year + pos_d_eav_hat, data = d, cluster = vcov_uid)
}
Code 
make_asym_by_type_summary <- function(models_A, models_B, school_extra, iv = FALSE) {
  term_pos <- if (iv) "fit_pos_d_eav" else "pos_d_eav"
  term_neg <- if (iv) "fit_neg_d_eav" else "neg_d_eav"
  rows <- list()
  for (spec in paste0("m", 1:3)) {
    rows[[spec]] <- map_dfr(gov_types, function(gov) {
      m <- models_A[[gov]][[spec]]
      term_stats(m, term_pos, gov) |>
        transmute(row = spec, agency = model, N, upper_bound, estimate, std_error, p_value, p_equal = p_equal_terms(m, term_neg, term_pos))
    })
  }
  for (spec in paste0("m", 4:5)) {
    rows[[spec]] <- map_dfr(gov_types_B, function(gov) {
      m <- models_B[[gov]][[spec]]
      term_stats(m, term_pos, gov) |>
        transmute(row = spec, agency = model, N, upper_bound, estimate, std_error, p_value, p_equal = p_equal_terms(m, term_neg, term_pos))
    })
  }
  rows[["m6"]] <- term_stats(school_extra, term_pos, "School") |>
    transmute(row = "m6", agency = model, N, upper_bound, estimate, std_error, p_value, p_equal = p_equal_terms(school_extra, term_neg, term_pos))
  bind_rows(rows) |>
    mutate(
      agency = agency_label(agency),
      row = recode(row,
      m1 = "No controls",
      m2 = "Year FE",
      m3 = "Year + unit FE",
      m4 = "IG revenue + year + unit FE",
      m5 = "Same sample as IG row, no IG control",
      m6 = "Enrollment + year + unit FE"
    ))
}

asym_ols_type_A <- setNames(map(gov_types, ~ run_asym_type_v22(reg_df, .x, iv = FALSE)), gov_types)
asym_ols_type_B <- setNames(map(gov_types_B, ~ run_asym_type_B_v22(reg_df, .x, iv = FALSE)), gov_types_B)
asym_ols_school_extra <- asym_school_extra_v22(reg_df, iv = FALSE)

asym_iv_type_A <- setNames(map(gov_types, ~ run_asym_type_v22(reg_df, .x, iv = TRUE)), gov_types)
asym_iv_type_B <- setNames(map(gov_types_B, ~ run_asym_type_B_v22(reg_df, .x, iv = TRUE)), gov_types_B)
asym_iv_school_extra <- asym_school_extra_v22(reg_df, iv = TRUE)

asym_ols_by_type_summary <- make_asym_by_type_summary(asym_ols_type_A, asym_ols_type_B, asym_ols_school_extra, iv = FALSE)
asym_iv_by_type_summary <- make_asym_by_type_summary(asym_iv_type_A, asym_iv_type_B, asym_iv_school_extra, iv = TRUE)

save_df_table(asym_ols_by_type_summary, "v22_table_08_OLS_by_type_asym_upper_bounds", "Table 8: OLS upper-bound estimates allowing for asymmetry")
save_df_table(asym_ols_by_type_summary |> select(row, agency, p_equal), "v22_table_A4_OLS_by_type_asym_p_equal", "Table A4: P-values for OLS symmetry tests")
save_df_table(asym_ols_by_type_summary |> select(row, agency, N, estimate, std_error, p_value), "v22_table_A5_OLS_by_type_asym_point_estimates", "Table A5: OLS point estimates for positive EAV changes")



save_df_table(asym_iv_by_type_summary, "v22_table_09_IV_by_type_asym_upper_bounds", "Table 9: IV upper-bound estimates allowing for asymmetry")
save_df_table(asym_iv_by_type_summary |> select(row, agency, p_equal), "v22_table_A6_IV_by_type_asym_p_equal", "Table A6: P-values for IV symmetry tests")
save_df_table(asym_iv_by_type_summary |> select(row, agency, N, estimate, std_error, p_value), "v22_table_A7_IV_by_type_asym_point_estimates", "Table A7: IV point estimates for positive EAV changes")

WRONG: FIX SOON

Code 
display_df_table(
  "v22_table_08_OLS_by_type_asym_upper_bounds",
  title = "Table 8. OLS upper-bound estimates allowing for asymmetry"
)
Table 8. OLS upper-bound estimates allowing for asymmetry
row agency N upper_bound estimate std_error p_value p_equal
No controls Other NA 0.097 0.068 0.017 0.000 NA
No controls Townships NA 0.097 0.068 0.017 0.000 NA
No controls Home-rule municipalities NA 0.097 0.068 0.017 0.000 NA
No controls Non-home-rule municipalities NA 0.097 0.068 0.017 0.000 NA
No controls Schools NA 0.097 0.068 0.017 0.000 NA
Year FE Other NA 0.149 0.091 0.035 0.010 NA
Year FE Townships NA 0.149 0.091 0.035 0.010 NA
Year FE Home-rule municipalities NA 0.149 0.091 0.035 0.010 NA
Year FE Non-home-rule municipalities NA 0.149 0.091 0.035 0.010 NA
Year FE Schools NA 0.149 0.091 0.035 0.010 NA
Year + unit FE Other NA 0.137 0.081 0.034 0.017 NA
Year + unit FE Townships NA 0.137 0.081 0.034 0.017 NA
Year + unit FE Home-rule municipalities NA 0.137 0.081 0.034 0.017 NA
Year + unit FE Non-home-rule municipalities NA 0.137 0.081 0.034 0.017 NA
Year + unit FE Schools NA 0.137 0.081 0.034 0.017 NA
IG revenue + year + unit FE Home-rule municipalities NA 0.565 0.295 0.164 0.072 NA
IG revenue + year + unit FE Non-home-rule municipalities NA 0.528 0.211 0.193 0.275 NA
IG revenue + year + unit FE Schools NA 0.331 0.175 0.095 0.065 NA
Same sample as IG row, no IG control Home-rule municipalities NA 0.562 0.295 0.163 0.070 NA
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.508 0.197 0.189 0.297 NA
Same sample as IG row, no IG control Schools NA 0.330 0.175 0.095 0.065 NA
Enrollment + year + unit FE Schools NA 0.330 0.174 0.095 0.068 NA
Code 
display_df_table(
  "v22_table_09_IV_by_type_asym_upper_bounds",
  title = "Table 9. IV upper-bound estimates allowing for asymmetry"
)
Table 9. IV upper-bound estimates allowing for asymmetry
row agency N upper_bound estimate std_error p_value p_equal
No controls Other NA 0.043 0.014 0.018 0.442 NA
No controls Townships NA 0.043 0.014 0.018 0.442 NA
No controls Home-rule municipalities NA 0.043 0.014 0.018 0.442 NA
No controls Non-home-rule municipalities NA 0.043 0.014 0.018 0.442 NA
No controls Schools NA 0.043 0.014 0.018 0.442 NA
Year FE Other NA 0.107 0.037 0.043 0.386 NA
Year FE Townships NA 0.107 0.037 0.043 0.386 NA
Year FE Home-rule municipalities NA 0.107 0.037 0.043 0.386 NA
Year FE Non-home-rule municipalities NA 0.107 0.037 0.043 0.386 NA
Year FE Schools NA 0.107 0.037 0.043 0.386 NA
Year + unit FE Other NA 0.090 0.028 0.038 0.460 NA
Year + unit FE Townships NA 0.090 0.028 0.038 0.460 NA
Year + unit FE Home-rule municipalities NA 0.090 0.028 0.038 0.460 NA
Year + unit FE Non-home-rule municipalities NA 0.090 0.028 0.038 0.460 NA
Year + unit FE Schools NA 0.090 0.028 0.038 0.460 NA
IG revenue + year + unit FE Home-rule municipalities NA 0.038 -0.112 0.091 0.220 NA
IG revenue + year + unit FE Non-home-rule municipalities NA 0.574 0.154 0.255 0.547 NA
IG revenue + year + unit FE Schools NA 0.046 -0.009 0.034 0.780 NA
Same sample as IG row, no IG control Home-rule municipalities NA 0.039 -0.110 0.091 0.223 NA
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.532 0.138 0.239 0.563 NA
Same sample as IG row, no IG control Schools NA 0.045 -0.010 0.034 0.772 NA
Enrollment + year + unit FE Schools NA 0.044 -0.011 0.033 0.746 NA

Table 10 & 11. Lagged EAV models

Code 
# ================================================================
# Tables 10/11/A8/A9. Lagged OLS models
# ================================================================

reg_df <- reg_df |>
  group_by(n_agency_group) |>
  arrange(year, .by_group = TRUE) |>
  mutate(
    d_2_eav = 100 * log((lag_av * lag_eq_factor_final) / (lag2_av * lag2_eq_factor_final)),
    d_3_eav = 100 * log((lag2_av * lag2_eq_factor_final) / (lag3_av * lag3_eq_factor_final))
  ) |>
  ungroup()

lag_all_ols_1 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav, data = reg_df, cluster = vcov_uid)
lag_all_ols_2 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav | year^home_rule_for_fe, data = reg_df, cluster = vcov_uid)
lag_all_ols_3 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav | year^home_rule_for_fe + n_uniqueid, data = reg_df, cluster = vcov_uid)

save_table(
  list(M1 = lag_all_ols_1, M2 = lag_all_ols_2, M3 = lag_all_ols_3),
  file_stub = "v22_table_10_OLS_all_agencies_lagged",
  title = "Table 10: All agencies OLS allowing for two lags of d_eav",
  coef_map = c("d_eav" = "Current EAV change", "d_2_eav" = "Lag 1 EAV change", "d_3_eav" = "Lag 2 EAV change")
)

save_df_table(
  bind_rows(
    lincom_sum(lag_all_ols_1, c("d_eav", "d_2_eav", "d_3_eav"), "M1"),
    lincom_sum(lag_all_ols_2, c("d_eav", "d_2_eav", "d_3_eav"), "M2"),
    lincom_sum(lag_all_ols_3, c("d_eav", "d_2_eav", "d_3_eav"), "M3")
  ),
  "v22_table_10_OLS_all_agencies_lagged_sums",
  "Table 10: Sum of current and lagged EAV effects"
)

run_lag_type_v22 <- function(df, gov) {
  d <- df |> filter(type_2 == gov)
  m1 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav, data = d, cluster = vcov_uid)
  m2 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav | year, data = d, cluster = vcov_uid)
  m3 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav | year + n_uniqueid, data = d, cluster = vcov_uid)
  list(m1 = m1, m2 = m2, m3 = m3)
}

run_lag_type_B_v22 <- function(df, gov) {
  d <- df |> filter(type_2 == gov)
  m4 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav + d_total_ig_revenue | year + n_uniqueid, data = d, cluster = vcov_uid)
  idx <- sample_index(m4, d)
  m5 <- feols(d_levy ~ d_eav + d_2_eav + d_3_eav | year + n_uniqueid, data = d[idx, ], cluster = vcov_uid)
  list(m4 = m4, m5 = m5)
}

lag_school_extra_v22 <- function(df) {
  d <- df |> filter(type_2 == "School")
  feols(d_levy ~ d_eav + d_2_eav + d_3_eav + d_enrollment | year + n_uniqueid, data = d, cluster = vcov_uid)
}

make_lag_by_type_summary <- function(models_A, models_B, school_extra) {
  rows <- list()
  for (spec in paste0("m", 1:3)) {
    rows[[spec]] <- map_dfr(gov_types, function(gov) {
      lincom_sum(models_A[[gov]][[spec]], c("d_eav", "d_2_eav", "d_3_eav"), gov) |>
        transmute(row = spec, agency = model, N, A_upper_bound, lagged_magnitude, std_error, lagged_p)
    })
  }
  for (spec in paste0("m", 4:5)) {
    rows[[spec]] <- map_dfr(gov_types_B, function(gov) {
      lincom_sum(models_B[[gov]][[spec]], c("d_eav", "d_2_eav", "d_3_eav"), gov) |>
        transmute(row = spec, agency = model, N, A_upper_bound, lagged_magnitude, std_error, lagged_p)
    })
  }
  rows[["m6"]] <- lincom_sum(school_extra, c("d_eav", "d_2_eav", "d_3_eav"), "School") |>
    transmute(row = "m6", agency = model, N, A_upper_bound, lagged_magnitude, std_error, lagged_p)
  bind_rows(rows) |>
    mutate(
      agency = agency_label(agency),
      row = recode(row,
      m1 = "No controls",
      m2 = "Year FE",
      m3 = "Year + unit FE",
      m4 = "IG revenue + year + unit FE",
      m5 = "Same sample as IG row, no IG control",
      m6 = "Enrollment + year + unit FE"
    ))
}

lag_type_A <- setNames(map(gov_types, ~ run_lag_type_v22(reg_df, .x)), gov_types)
lag_type_B <- setNames(map(gov_types_B, ~ run_lag_type_B_v22(reg_df, .x)), gov_types_B)
lag_school_extra <- lag_school_extra_v22(reg_df)

lag_by_type_summary <- make_lag_by_type_summary(lag_type_A, lag_type_B, lag_school_extra)

save_df_table(lag_by_type_summary, "v22_table_11_OLS_by_type_lagged_upper_bounds", "Table 11: OLS upper-bound estimates allowing for two lags of d_eav")
save_df_table(lag_by_type_summary |> select(row, agency, N, lagged_p), "v22_table_A8_OLS_by_type_lagged_p_values", "Table A8: P-values for lagged effects")
save_df_table(lag_by_type_summary |> select(row, agency, N, lagged_magnitude, std_error, lagged_p), "v22_table_A9_OLS_by_type_lagged_point_estimates", "Table A9: Sum of current and lagged EAV coefficients")
Code 
display_model_table(
  "v22_table_10_OLS_all_agencies_lagged",
  title = "Table 10. All-agency OLS estimates with lagged EAV changes",
  coef_map = c("d_eav" = "Change in EAV", "d_2_eav" = "One-year lag", "d_3_eav" = "Two-year lag")
)
Table 10. All-agency OLS estimates with lagged EAV changes
M1 M2 M3
* p < 0.1, ** p < 0.05, *** p < 0.01
Change in EAV 0.077*** 0.121*** 0.100***
(0.011) (0.034) (0.032)
One-year lag -0.037 0.050 0.026
(0.031) (0.031) (0.033)
Two-year lag 0.079** 0.136** 0.116
(0.037) (0.066) (0.071)
Num.Obs. 6176 6176 6174
R2 0.008 0.019 0.048
R2 Within 0.006 0.004
ub_d_eav 0.0954438446626904 0.176443184560096 0.153626531647265
FE: year^home_rule_for_fe X X
FE: n_uniqueid X
Code 
display_df_table(
  "v22_table_10_OLS_all_agencies_lagged_sums",
  title = "Lagged-effect summary for Table 10"
)
Lagged-effect summary for Table 10
model N lagged_magnitude std_error A_upper_bound lagged_p
M1 NA 0.120 0.019 0.152 0.000
M2 NA 0.306 0.112 0.490 0.006
M3 NA 0.242 0.120 0.440 0.044
Code 
display_df_table(
  "v22_table_11_OLS_by_type_lagged_upper_bounds",
  title = "Table 11. By agency type lagged EAV estimates"
)
Table 11. By agency type lagged EAV estimates
row agency N A_upper_bound lagged_magnitude std_error lagged_p
No controls Other NA 0.240 0.170 0.043 0.000
No controls Townships NA 0.093 0.051 0.026 0.047
No controls Home-rule municipalities NA 0.142 0.078 0.039 0.043
No controls Non-home-rule municipalities NA 0.136 0.065 0.043 0.135
No controls Schools NA 0.149 0.114 0.021 0.000
Year FE Other NA 0.787 0.388 0.242 0.109
Year FE Townships NA 0.185 0.014 0.104 0.896
Year FE Home-rule municipalities NA 0.430 0.142 0.175 0.418
Year FE Non-home-rule municipalities NA 0.449 0.236 0.130 0.069
Year FE Schools NA 0.478 0.343 0.082 0.000
Year + unit FE Other NA 0.736 0.300 0.265 0.257
Year + unit FE Townships NA 0.272 0.049 0.135 0.715
Year + unit FE Home-rule municipalities NA 0.437 0.175 0.159 0.270
Year + unit FE Non-home-rule municipalities NA 0.397 0.166 0.141 0.239
Year + unit FE Schools NA 0.385 0.257 0.078 0.001
IG revenue + year + unit FE Home-rule municipalities NA 0.837 0.413 0.257 0.108
IG revenue + year + unit FE Non-home-rule municipalities NA 0.424 0.066 0.218 0.760
IG revenue + year + unit FE Schools NA 0.438 0.280 0.096 0.004
Same sample as IG row, no IG control Home-rule municipalities NA 0.837 0.413 0.258 0.109
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.363 0.042 0.195 0.828
Same sample as IG row, no IG control Schools NA 0.436 0.278 0.096 0.004
Enrollment + year + unit FE Schools NA 0.433 0.278 0.095 0.003

Appendix tables

Appendix 1

Source: Civic Federation Property Tax Primer

Property Tax Process

Appendix 2: Instrument validity

With two-way FE, your identifying variation becomes: deviations from unit-specific mean and year-specific shocks

So the question becomes:conditional on unit and year shocks, does reassessment timing still move EAV, and only through EAV?

We use these (exogenously determined) reassessment years as an “instrument” which purges observed changes in the tax base of the component that is predictable because of the re-assessment year.

If you test the instrument in a different specification (say no fixed effects), you are asking a different question:

  • Without FE: “Is reassessment year correlated with raw variation in EAV?”
  • With FE: “Is reassessment year correlated with within-unit deviations from trend in EAV?”

The code below creates the instrument-validity table. The combined municipality column is labeled All Munis.

Code 
instrument_df <- main_df |>
  filter(year > 2008)

table(instrument_df$year) # 2009 through 2023

2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 
 412  412  412  412  412  412  412  412  412  412  412  412  412  412  412
Code 
# Appendix A2 

# all munis
feols(d_eav ~ reassess_year | year + uniqueid,
  data = instrument_df,
  cluster = vcov_uid,
  fsplit = ~type)
                              x.1               x.2               x.3
Sample (type)         Full sample              Muni             Other
Dependent Var.:             d_eav             d_eav             d_eav
                                                                     
reassess_year   6.722*** (0.1341) 6.591*** (0.2649) 6.736*** (0.2504)
Fixed-Effects:  ----------------- ----------------- -----------------
year                          Yes               Yes               Yes
uniqueid                      Yes               Yes               Yes
_______________ _________________ _________________ _________________
S.E.: Clustered    by: n_uniqueid    by: n_uniqueid    by: n_uniqueid
Observations                6,178             1,605             2,158
R2                        0.68560           0.70518           0.65647
Within R2                 0.14292           0.13905           0.13240

                              x.4               x.5
Sample (type)              School          Township
Dependent Var.:             d_eav             d_eav
                                                   
reassess_year   6.824*** (0.2388) 6.785*** (0.1754)
Fixed-Effects:  ----------------- -----------------
year                          Yes               Yes
uniqueid                      Yes               Yes
_______________ _________________ _________________
S.E.: Clustered    by: n_uniqueid    by: n_uniqueid
Observations                1,995               420
R2                        0.69738           0.73580
Within R2                 0.15071           0.19791
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Code 
# homerule and nonhomerule munis and drop other types.
feols(d_eav ~ reassess_year | year + uniqueid,
  data = instrument_df |> filter(type == "Muni"),
  cluster = vcov_uid,
  fsplit = ~home_rule_ind)
                                     x.1              x.2               x.3
Sample (home_rule_ind)       Full sample                0                 1
Dependent Var.:                    d_eav            d_eav             d_eav
                                                                           
reassess_year          6.591*** (0.2649) 5.951*** (1.076) 6.631*** (0.3341)
Fixed-Effects:         ----------------- ---------------- -----------------
year                                 Yes              Yes               Yes
uniqueid                             Yes              Yes               Yes
______________________ _________________ ________________ _________________
S.E.: Clustered           by: n_uniqueid   by: n_uniqueid    by: n_uniqueid
Observations                       1,605              573             1,032
R2                               0.70518          0.81859           0.66724
Within R2                        0.13905          0.08609           0.14429
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Code 
all_instrument      <- feols(d_eav ~ reassess_year| year + uniqueid,
  data = instrument_df, cluster = vcov_uid)

muni_instrument    <- feols(d_eav ~ reassess_year| year + uniqueid,
  data = filter(instrument_df, type == "Muni"),
  cluster = vcov_uid)

other_instrument    <- feols(d_eav ~ reassess_year | year + uniqueid,
  data = filter(instrument_df, type == "Other"),
  cluster = vcov_uid)

school_instrument   <- feols(d_eav ~ reassess_year| year + uniqueid,
  data = filter(instrument_df, type == "School"),
  cluster = vcov_uid)

township_instrument <- feols(d_eav ~ reassess_year| year + uniqueid,
  data = filter(instrument_df, type == "Township"),
  cluster = vcov_uid)
# 
# save_table(
#   list(
#     "All agencies" = all_instrument,
#     "All Munis" = muni_instrument,
#     "Other" = other_instrument,
#     "Schools" = school_instrument,
#     "Townships" = township_instrument
#   ),
#   file_stub = "instrument_validity_checks",
#   title = "Table X: Predict assessments by reassessment year by type",
# #   notes = "Cook County, Illinois data from 2008 through 2023"
# # )
# 
# #| label: instrument-validity-display
# 
# 
# display_model_table(
#   "instrument_validity_checks",
#   title = "Table 1. Reassessment year as predictor of EAV change",
#   notes = "Cook County, Illinois data from 2008 through 2023"
# )
Code 
modelsummary(  list(
    "All agencies" = all_instrument,
    "All Munis" = muni_instrument,
    "Other" = other_instrument,
    "Schools" = school_instrument,
    "Townships" = township_instrument), 
    stars = TRUE, gof_omit = gof_omit
)

Table X: Predict assessments by reassessment year by type. Cook County, Illinois data from 2008 through 2023
All agencies All Munis Other Schools Townships
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
reassess_year 6.722*** 6.591*** 6.736*** 6.824*** 6.785***
(0.134) (0.265) (0.250) (0.239) (0.175)
Num.Obs. 6178 1605 2158 1995 420
R2 0.686 0.705 0.656 0.697 0.736
R2 Within 0.143 0.139 0.132 0.151 0.198
FE: year X X X X X
FE: uniqueid X X X X X

These appendix outputs are created by the table chunks above and displayed here so readers do not have to search through the output folder.

Appendix 3

OLS by agency type point estimates

Code 
# ================================================================
# Tables 4/A2 and 5/A3. By agency type: OLS + IV summary rows
# ================================================================

run_type_v22 <- function(df, gov, iv = FALSE) {
  d <- df |> filter(type_2 == gov)
  if (!iv) {
    m1 <- feols(d_levy ~ d_eav, data = d, cluster = vcov_uid)
    m2 <- feols(d_levy ~ d_eav | year, data = d, cluster = vcov_uid)
    m3 <- feols(d_levy ~ d_eav | year + n_uniqueid, data = d, cluster = vcov_uid)
    return(list(m1 = m1, m2 = m2, m3 = m3))
  }
  m1 <- feols(d_levy ~ 1 | 0 | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  m2 <- feols(d_levy ~ 1 | year | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  m3 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  list(m1 = m1, m2 = m2, m3 = m3)
}

run_type_B_v22 <- function(df, gov, iv = FALSE) {
  d <- df |> filter(type_2 == gov)
  if (!iv) {
    m4 <- feols(d_levy ~ d_eav + d_total_ig_revenue | year + n_uniqueid, data = d, cluster = vcov_uid)
    idx <- sample_index(m4, d)
    m5 <- feols(d_levy ~ d_eav | year + n_uniqueid, data = d[idx, ], cluster = vcov_uid)
    return(list(m4 = m4, m5 = m5))
  }
  m4 <- feols(d_levy ~ d_total_ig_revenue | year + n_uniqueid | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
  idx <- sample_index(m4, d)
  m5 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = d[idx, ], cluster = vcov_uid)
  list(m4 = m4, m5 = m5)
}

school_extra_v22 <- function(df, iv = FALSE) {
  d <- df |> filter(type_2 == "School")
  if (!iv) {
    return(feols(d_levy ~ d_eav + d_enrollment | year + n_uniqueid, data = d, cluster = vcov_uid))
  }
  feols(d_levy ~ d_enrollment | year + n_uniqueid | d_eav ~ reassess_year, data = d, cluster = vcov_uid)
}

make_by_type_summary <- function(models_A, models_B, school_extra, iv = FALSE) {
  term <- if (iv) "fit_d_eav" else "d_eav"
  rows <- list()
  for (spec in paste0("m", 1:3)) {
    rows[[spec]] <- map_dfr(gov_types, function(gov) {
      term_stats(models_A[[gov]][[spec]], term, gov) |>
        transmute(row = spec, agency = model, N, upper_bound, estimate, p_value)
    })
  }
  for (spec in paste0("m", 4:5)) {
    rows[[spec]] <- map_dfr(gov_types_B, function(gov) {
      term_stats(models_B[[gov]][[spec]], term, gov) |>
        transmute(row = spec, agency = model, N, upper_bound, estimate, p_value)
    })
  }
  rows[["m6"]] <- term_stats(school_extra, term, "School") |>
    transmute(row = "m6", agency = model, N, upper_bound, estimate, p_value)
  bind_rows(rows) |>
    mutate(row = recode(row,
      m1 = "No controls",
      m2 = "Year FE",
      m3 = "Year + unit FE",
      m4 = "IG revenue + year + unit FE",
      m5 = "Same sample as IG row, no IG control",
      m6 = "Enrollment + year + unit FE"
    ))
}

ols_type_A <- setNames(map(gov_types, ~ run_type_v22(reg_df, .x, iv = FALSE)), gov_types)
ols_type_B <- setNames(map(gov_types_B, ~ run_type_B_v22(reg_df, .x, iv = FALSE)), gov_types_B)
ols_school_extra <- school_extra_v22(reg_df, iv = FALSE)

iv_type_A <- setNames(map(gov_types, ~ run_type_v22(reg_df, .x, iv = TRUE)), gov_types)
iv_type_B <- setNames(map(gov_types_B, ~ run_type_B_v22(reg_df, .x, iv = TRUE)), gov_types_B)
iv_school_extra <- school_extra_v22(reg_df, iv = TRUE)

ols_by_type_summary <- make_by_type_summary(ols_type_A, ols_type_B, ols_school_extra, iv = FALSE)
iv_by_type_summary <- make_by_type_summary(iv_type_A, iv_type_B, iv_school_extra, iv = TRUE)
Code 
display_df_table("v22_table_A2_OLS_by_type_point_estimates", title = "Table A2. By agency type OLS point estimates")
Table A2. By agency type OLS point estimates
row agency N estimate std_error p_value
No controls Other NA 0.093 0.018 0.000
No controls Townships NA 0.006 0.028 0.835
No controls Home-rule municipalities NA 0.018 0.026 0.485
No controls Non-home-rule municipalities NA 0.039 0.022 0.079
No controls Schools NA 0.096 0.018 0.000
Year FE Other NA 0.109 0.043 0.012
Year FE Townships NA 0.010 0.029 0.737
Year FE Home-rule municipalities NA 0.079 0.051 0.117
Year FE Non-home-rule municipalities NA -0.036 0.048 0.460
Year FE Schools NA 0.114 0.044 0.009
Year + unit FE Other NA 0.085 0.036 0.018
Year + unit FE Townships NA 0.010 0.029 0.741
Year + unit FE Home-rule municipalities NA 0.073 0.046 0.115
Year + unit FE Non-home-rule municipalities NA -0.052 0.048 0.272
Year + unit FE Schools NA 0.102 0.040 0.010
IG revenue + year + unit FE Home-rule municipalities NA 0.228 0.112 0.041
IG revenue + year + unit FE Non-home-rule municipalities NA 0.139 0.146 0.342
IG revenue + year + unit FE Schools NA 0.157 0.066 0.018
Same sample as IG row, no IG control Home-rule municipalities NA 0.228 0.111 0.040
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.127 0.135 0.346
Same sample as IG row, no IG control Schools NA 0.156 0.066 0.018
Enrollment + year + unit FE Schools NA 0.155 0.066 0.019

IV by agency type point estimates

Code 
display_df_table("v22_table_A3_IV_by_type_point_estimates", title = "Table A3. By agency type IV point estimates")
Table A3. By agency type IV point estimates
row agency N estimate std_error p_value
No controls Other NA -0.052 0.080 0.512
No controls Townships NA -0.058 0.061 0.341
No controls Home-rule municipalities NA -0.037 0.044 0.406
No controls Non-home-rule municipalities NA -0.029 0.055 0.594
No controls Schools NA 0.065 0.026 0.013
Year FE Other NA -0.201 0.174 0.248
Year FE Townships NA -0.065 0.089 0.465
Year FE Home-rule municipalities NA -0.007 0.070 0.919
Year FE Non-home-rule municipalities NA -0.052 0.132 0.691
Year FE Schools NA 0.011 0.033 0.733
Year + unit FE Other NA -0.201 0.174 0.248
Year + unit FE Townships NA -0.065 0.089 0.465
Year + unit FE Home-rule municipalities NA -0.007 0.070 0.914
Year + unit FE Non-home-rule municipalities NA -0.060 0.135 0.658
Year + unit FE Schools NA 0.011 0.033 0.741
IG revenue + year + unit FE Home-rule municipalities NA -0.240 0.330 0.466
IG revenue + year + unit FE Non-home-rule municipalities NA -0.141 0.292 0.629
IG revenue + year + unit FE Schools NA -0.037 0.100 0.714
Same sample as IG row, no IG control Home-rule municipalities NA -0.243 0.330 0.462
Same sample as IG row, no IG control Non-home-rule municipalities NA -0.142 0.283 0.616
Same sample as IG row, no IG control Schools NA -0.035 0.101 0.728
Enrollment + year + unit FE Schools NA -0.045 0.101 0.658

Appendix 4

OLS symmetry tests

Code 
display_df_table("v22_table_A4_OLS_by_type_asym_p_equal", title = "Table A4. P-values for OLS symmetry tests")
Table A4. P-values for OLS symmetry tests
row agency p_equal
No controls Other NA
No controls Townships NA
No controls Home-rule municipalities NA
No controls Non-home-rule municipalities NA
No controls Schools NA
Year FE Other NA
Year FE Townships NA
Year FE Home-rule municipalities NA
Year FE Non-home-rule municipalities NA
Year FE Schools NA
Year + unit FE Other NA
Year + unit FE Townships NA
Year + unit FE Home-rule municipalities NA
Year + unit FE Non-home-rule municipalities NA
Year + unit FE Schools NA
IG revenue + year + unit FE Home-rule municipalities NA
IG revenue + year + unit FE Non-home-rule municipalities NA
IG revenue + year + unit FE Schools NA
Same sample as IG row, no IG control Home-rule municipalities NA
Same sample as IG row, no IG control Non-home-rule municipalities NA
Same sample as IG row, no IG control Schools NA
Enrollment + year + unit FE Schools NA

Appendix 5

OLS asymmetric point estimates

Code 
display_df_table("v22_table_A5_OLS_by_type_asym_point_estimates", title = "Table A5. OLS point estimates for positive EAV changes")
Table A5. OLS point estimates for positive EAV changes
row agency N estimate std_error p_value
No controls Other NA 0.068 0.017 0.000
No controls Townships NA 0.068 0.017 0.000
No controls Home-rule municipalities NA 0.068 0.017 0.000
No controls Non-home-rule municipalities NA 0.068 0.017 0.000
No controls Schools NA 0.068 0.017 0.000
Year FE Other NA 0.091 0.035 0.010
Year FE Townships NA 0.091 0.035 0.010
Year FE Home-rule municipalities NA 0.091 0.035 0.010
Year FE Non-home-rule municipalities NA 0.091 0.035 0.010
Year FE Schools NA 0.091 0.035 0.010
Year + unit FE Other NA 0.081 0.034 0.017
Year + unit FE Townships NA 0.081 0.034 0.017
Year + unit FE Home-rule municipalities NA 0.081 0.034 0.017
Year + unit FE Non-home-rule municipalities NA 0.081 0.034 0.017
Year + unit FE Schools NA 0.081 0.034 0.017
IG revenue + year + unit FE Home-rule municipalities NA 0.295 0.164 0.072
IG revenue + year + unit FE Non-home-rule municipalities NA 0.211 0.193 0.275
IG revenue + year + unit FE Schools NA 0.175 0.095 0.065
Same sample as IG row, no IG control Home-rule municipalities NA 0.295 0.163 0.070
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.197 0.189 0.297
Same sample as IG row, no IG control Schools NA 0.175 0.095 0.065
Enrollment + year + unit FE Schools NA 0.174 0.095 0.068

IV symmetry tests

gen pos_inst = reassess_year * eav_growth gen neg_inst = reassess_year * (1 - eav_growth) ivregress 2sls d_levy (neg_d_eav pos_d_eav = reassess_year pos_inst), vce(cluster n_uniqueid)

reassess_year is your instrument eav_growth is exogenous (constructed from outcomes, but deterministic split)

Code 
display_df_table("v22_table_A6_IV_by_type_asym_p_equal", title = "Table A6. P-values for IV symmetry tests")
Table A6. P-values for IV symmetry tests
row agency p_equal
No controls Other NA
No controls Townships NA
No controls Home-rule municipalities NA
No controls Non-home-rule municipalities NA
No controls Schools NA
Year FE Other NA
Year FE Townships NA
Year FE Home-rule municipalities NA
Year FE Non-home-rule municipalities NA
Year FE Schools NA
Year + unit FE Other NA
Year + unit FE Townships NA
Year + unit FE Home-rule municipalities NA
Year + unit FE Non-home-rule municipalities NA
Year + unit FE Schools NA
IG revenue + year + unit FE Home-rule municipalities NA
IG revenue + year + unit FE Non-home-rule municipalities NA
IG revenue + year + unit FE Schools NA
Same sample as IG row, no IG control Home-rule municipalities NA
Same sample as IG row, no IG control Non-home-rule municipalities NA
Same sample as IG row, no IG control Schools NA
Enrollment + year + unit FE Schools NA

IV asymmetric point estimates

Code 
reg_df <- reg_df %>%
  mutate(
    eav_growth = ifelse(d_eav > 0, 1, 0),
    pos_d_eav = d_eav * eav_growth,
    neg_d_eav = d_eav * (1 - eav_growth),

    # correct instruments
    pos_inst = reassess_year * eav_growth
  )

m3 <- feols(d_levy ~ 1 | year + n_uniqueid | d_eav ~ reassess_year, data = reg_df, cluster = vcov_uid)

m3
TSLS estimation - Dep. Var.: d_levy
                  Endo.    : d_eav
                  Instr.   : reassess_year
Second stage: Dep. Var.: d_levy
Observations: 6,174
Fixed-effects: year: 15,  n_uniqueid: 419
Standard-errors: Clustered (n_uniqueid) 
          Estimate Std. Error t value Pr(>|t|) 
fit_d_eav -0.08472   0.062765 -1.3498  0.17781 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 12.7     Adj. R2: -0.032007
             Within R2:  3.156e-4
F-test (1st stage), d_eav: stat = 1,027.3158, p < 2.2e-16 , on 1 and 6,158 DoF.
               Wu-Hausman: stat =     7.3673, p = 0.006662, on 1 and 5,739 DoF.
Code 
iv_model <- feols(
  d_levy ~ 1 | n_uniqueid + year | pos_d_eav ~ reassess_year + pos_inst,
  data = reg_df,
  cluster = ~n_uniqueid
)

summary(iv_model)
TSLS estimation - Dep. Var.: d_levy
                  Endo.    : pos_d_eav
                  Instr.   : reassess_year, pos_inst
Second stage: Dep. Var.: d_levy
Observations: 6,174
Fixed-effects: n_uniqueid: 419,  year: 15
Standard-errors: Clustered (n_uniqueid) 
              Estimate Std. Error  t value Pr(>|t|) 
fit_pos_d_eav -0.04726   0.042289 -1.11754  0.26441 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 12.7     Adj. R2: -0.032015
             Within R2:  3.081e-4
F-test (1st stage), pos_d_eav: stat = 4,462.88199, p < 2.2e-16 , on 2 and 6,157 DoF.
                   Wu-Hausman: stat =    33.42098, p = 7.813e-9, on 1 and 5,739 DoF.
                       Sargan: stat =     0.27445, p = 0.600363, on 1 DoF.
Code 
display_df_table("v22_table_A7_IV_by_type_asym_point_estimates", title = "Table A7. IV point estimates for positive EAV changes")
Table A7. IV point estimates for positive EAV changes
row agency N estimate std_error p_value
No controls Other NA 0.014 0.018 0.442
No controls Townships NA 0.014 0.018 0.442
No controls Home-rule municipalities NA 0.014 0.018 0.442
No controls Non-home-rule municipalities NA 0.014 0.018 0.442
No controls Schools NA 0.014 0.018 0.442
Year FE Other NA 0.037 0.043 0.386
Year FE Townships NA 0.037 0.043 0.386
Year FE Home-rule municipalities NA 0.037 0.043 0.386
Year FE Non-home-rule municipalities NA 0.037 0.043 0.386
Year FE Schools NA 0.037 0.043 0.386
Year + unit FE Other NA 0.028 0.038 0.460
Year + unit FE Townships NA 0.028 0.038 0.460
Year + unit FE Home-rule municipalities NA 0.028 0.038 0.460
Year + unit FE Non-home-rule municipalities NA 0.028 0.038 0.460
Year + unit FE Schools NA 0.028 0.038 0.460
IG revenue + year + unit FE Home-rule municipalities NA -0.112 0.091 0.220
IG revenue + year + unit FE Non-home-rule municipalities NA 0.154 0.255 0.547
IG revenue + year + unit FE Schools NA -0.009 0.034 0.780
Same sample as IG row, no IG control Home-rule municipalities NA -0.110 0.091 0.223
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.138 0.239 0.563
Same sample as IG row, no IG control Schools NA -0.010 0.034 0.772
Enrollment + year + unit FE Schools NA -0.011 0.033 0.746

Lagged model p-values

Code 
display_df_table("v22_table_A8_OLS_by_type_lagged_p_values", title = "Table A8. P-values for lagged effects")
Table A8. P-values for lagged effects
row agency N lagged_p
No controls Other NA 0.000
No controls Townships NA 0.047
No controls Home-rule municipalities NA 0.043
No controls Non-home-rule municipalities NA 0.135
No controls Schools NA 0.000
Year FE Other NA 0.109
Year FE Townships NA 0.896
Year FE Home-rule municipalities NA 0.418
Year FE Non-home-rule municipalities NA 0.069
Year FE Schools NA 0.000
Year + unit FE Other NA 0.257
Year + unit FE Townships NA 0.715
Year + unit FE Home-rule municipalities NA 0.270
Year + unit FE Non-home-rule municipalities NA 0.239
Year + unit FE Schools NA 0.001
IG revenue + year + unit FE Home-rule municipalities NA 0.108
IG revenue + year + unit FE Non-home-rule municipalities NA 0.760
IG revenue + year + unit FE Schools NA 0.004
Same sample as IG row, no IG control Home-rule municipalities NA 0.109
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.828
Same sample as IG row, no IG control Schools NA 0.004
Enrollment + year + unit FE Schools NA 0.003

Lagged model point estimates

Code 
display_df_table("v22_table_A9_OLS_by_type_lagged_point_estimates", title = "Table A9. Sum of current and lagged EAV coefficients")
Table A9. Sum of current and lagged EAV coefficients
row agency N lagged_magnitude std_error lagged_p
No controls Other NA 0.170 0.043 0.000
No controls Townships NA 0.051 0.026 0.047
No controls Home-rule municipalities NA 0.078 0.039 0.043
No controls Non-home-rule municipalities NA 0.065 0.043 0.135
No controls Schools NA 0.114 0.021 0.000
Year FE Other NA 0.388 0.242 0.109
Year FE Townships NA 0.014 0.104 0.896
Year FE Home-rule municipalities NA 0.142 0.175 0.418
Year FE Non-home-rule municipalities NA 0.236 0.130 0.069
Year FE Schools NA 0.343 0.082 0.000
Year + unit FE Other NA 0.300 0.265 0.257
Year + unit FE Townships NA 0.049 0.135 0.715
Year + unit FE Home-rule municipalities NA 0.175 0.159 0.270
Year + unit FE Non-home-rule municipalities NA 0.166 0.141 0.239
Year + unit FE Schools NA 0.257 0.078 0.001
IG revenue + year + unit FE Home-rule municipalities NA 0.413 0.257 0.108
IG revenue + year + unit FE Non-home-rule municipalities NA 0.066 0.218 0.760
IG revenue + year + unit FE Schools NA 0.280 0.096 0.004
Same sample as IG row, no IG control Home-rule municipalities NA 0.413 0.258 0.109
Same sample as IG row, no IG control Non-home-rule municipalities NA 0.042 0.195 0.828
Same sample as IG row, no IG control Schools NA 0.278 0.096 0.004
Enrollment + year + unit FE Schools NA 0.278 0.095 0.003