Q1 Figures and Tables

Impact of FEMA designated risk zones on sale price

Setup stuff

Code

library(tidyverse)
library(fixest)
library(did)  
library(etwfe)
library(tidyverse)

library(modelsummary)
library(gt)

options(scipen = 999)

# bldg_sfha <- readRDS("data/processed/variants/df_prep__bldg_sfha.RDS")
# 
# parcels <- readRDS("data/processed/variants/df_prep__parcel_sfha.RDS")
# 
# bldg_1in500 <- readRDS("data/processed/variants/df_prep__bldg_1in500.RDS")

source("R/helper_pins_to_drop.R")

Code

lake100ft <- readxl::read_xlsx("data/processed/parcels_2025_lakeMI_within100ft_ExportTable_TableToExcel.xlsx")
lake500ft <- readxl::read_xlsx("data/processed/parcels_2025_lakeMI_within500ft_ExportTable_TableToExcel.xlsx")

bldg_sfha <- readRDS("data/processed/targets/sales/df_prep_bldg_v2026_03.RDS")

parcels <- readRDS("data/processed/targets/sales/df_prep_land_v2026_03.RDS")

bldg_1in500 <- readRDS("data/processed/targets/sales/df_prep_risk500_v2026_03.RDS")  |>
  filter(!pin10 %in% drop_parcels  |
               (pin10 == "0113301013" & sale_year > 2014))

# bldg_sfha <- readRDS("data/processed/variants/df_prep__bldg_sfhaarcgis_redo_2026sales.RDS")
# 
# parcels <- readRDS("data/processed/variants/df_prep__parcel_sfhaarcgis_redo_2026sales.RDS")
# 
# bldg_1in500 <- readRDS("data/processed/variants/df_prep__bldg_1in500arcgis_redo_2026sales.RDS") |> 
#   filter(!pin10 %in% drop_parcels  |          
#            (pin10 == "0113301013" & sale_year > 2014))


bldg_sfha <- bldg_sfha |> filter(!pin10 %in% drop_parcels  |
               (pin10 == "0113301013" & sale_year > 2014)
             )

parcels <- parcels |> filter(!pin10 %in% drop_parcels  |
               (pin10 == "0113301013" & sale_year > 2014)
             )

make_control_groups <- function(df, 
                                ever_added,
                                ever_removed,
                    pre_eff_change_type, 
                    type = c("pre", "eff")
                                ){
  
  # Treated Group A: Added to Risk Zone
  group_T_A <- df |>
    filter( {{ever_added}} == TRUE )
  
  # Treated Group B: Removed from Risk Zone
  group_T_B <- df |> 
      filter( {{ever_removed}}  == TRUE)

  # Control Group A:  Never in SFHA, FIRM panels never updated
    
   if(type == "pre"){
   group_C_A <- df |> 
      filter(
        {{pre_eff_change_type}} == "Never SFHA" &
             pre_date_chr %in% c("2005-01-01")
         )
  }else if(type == "eff"){  # for effective maps
  group_C_A <- df |> 
      filter(
        {{pre_eff_change_type}} == "Never SFHA" &
         eff_date_chr %in% c("2008-08-19") 
         ) 
  }  

    # Control Group B: Never in SFHA, FIRM Panels are updated
    group_C_B  <- df |> 
      filter( 
        {{pre_eff_change_type}} == "Never SFHA" &
         (eff_date_chr %in% c("2019-11-01", "2021-09-10") | 
             pre_date_chr %in% c("2015-02-12", "2019-07-01", "2021-09-22")
         )
         )

    # Control Group C: Always in SFHA, FIRM Panels are not updated 
    if(type == "pre"){
    group_C_C <- df |> 
      filter(
        {{pre_eff_change_type}} == "Always SFHA" &             
          pre_date_chr %in% c("2005-01-01")
      )}
    else if(type == "eff") {  # this probably is tainted by prelim map cohort in 2021 though....
    group_C_C <- df |> 
      filter(
        {{pre_eff_change_type}} == "Always SFHA" &
         (eff_date_chr %in% c("2008-08-19") )
      )
    
    group_C_C_alt <- df |> 
      filter(
        {{pre_eff_change_type}} == "Always SFHA" &
         (eff_date_chr %in% c("2008-08-19") | 
            pre_date_chr %in% c("2005-01-01")
          )
      )
    }
    
    # Always in SFHA but FIRM panels have been recently updated
   
    if(type == "eff") {
      group_C_D <- df |> 
        filter(
        {{pre_eff_change_type}} == "Always SFHA" &
         (eff_date_chr %in% c("2019-11-01", "2021-09-10"))
         )
      
       group_C_D_alt <- df |> 
        filter(
        {{pre_eff_change_type}} == "Always SFHA" &
         (eff_date_chr %in% c("2019-11-01", "2021-09-10") | 
             pre_date_chr %in% c("2015-02-12", "2019-07-01", "2021-09-22")
             )
         )
    } else if (type == "pre") {
      group_C_D <- df |> 
         filter(
        {{pre_eff_change_type}} == "Always SFHA" &
             pre_date_chr %in% c("2015-02-12", "2019-07-01", "2021-09-22")
         )
    }
      list(
    group_T_A = group_T_A,
    group_T_B = group_T_B,
    group_C_A = group_C_A,
    group_C_B = group_C_B,
    group_C_C = group_C_C,
    group_C_D = group_C_D

  )

}

groups <- make_control_groups(df = bldg_sfha, 
                    ever_added = ever_added_prelim, 
                    ever_removed = ever_removed_prelim, 
                    pre_eff_change_type = change_type_prelim, 
                    type = "pre")

groups_long <- imap_dfr(groups, ~ {
  .x |>
    mutate(group = .y)
})

Code

groups_long |>
  group_by(group) |>
  summarise(
    n = n(),
    n_properties = n_distinct(pin),
    mean_price = mean(sale_price, na.rm = TRUE),
    mean_log_price = mean(log_price, na.rm = TRUE)
  )

group	n	n_properties	mean_price	mean_log_price
group_C_A	490915	215004	303329.3	12.28778
group_C_B	162730	69781	429702.7	12.59933
group_C_C	14188	6206	268833.6	12.16733
group_C_D	595	257	505159.2	12.56855
group_T_A	13	6	274300.0	12.34933
group_T_B	155	67	124150.5	11.48264

Line Graphs

Code

group_trends <- groups_long |>
  group_by(group, sale_year) |>
  summarise(
    mean_log_price = mean(log_price, na.rm = TRUE),
    n = n(),
    .groups = "drop"
  )

ggplot(group_trends, aes(x = sale_year, y = mean_log_price, color = group)) +
  geom_line(linewidth = 1) +
  geom_point(size = 1.5) +
  labs(
    x = "Year",
    y = "Mean Log Price",
    color = "Group"
  ) +
  theme_minimal()

Code

groups_long <- groups_long |>
  mutate(group_type = case_when(
    group %in% c("group_T_A", "group_T_B") ~ "Treated",
    TRUE ~ "Control"
  ))

group_trends2 <- groups_long |>
  group_by(group_type, sale_year) |>
  summarise(
    mean_log_price = mean(log_price, na.rm = TRUE),
    .groups = "drop"
  )

ggplot(group_trends2, aes(x = sale_year, y = mean_log_price, color = group_type)) +
  geom_line(linewidth = 1.2)
ggplot(group_trends, aes(x = sale_year, y = mean_log_price)) +
  geom_line() +
  facet_wrap(~ group)

Question 1

Table 2: Sample Counts

Code

table(bldg_sfha$prelim_sfha_category)


    Added to prelim SFHA              Always SFHA               Never SFHA 
                      13                    14783                   653645 
Removed from prelim SFHA 
                     155

Code

table(bldg_sfha$change_type_prelim)


  Never SFHA  Always SFHA Changes SFHA 
      653645        14783          168

Code

df_es_added <- bldg_sfha |>
  filter(pin10 %in% groups$group_T_A$pin10 | pin10 %in% groups$group_C_B$pin10 )|>
 # filter(prelim_sfha_category %in% c("Added to prelim SFHA", "Never SFHA")) |>
  mutate(pin_num = as.numeric(pin)) |>
  dplyr::arrange(pin_num, sale_year, sale_date) |>
  dplyr::group_by(pin_num, sale_year, pre_date_chr) |>
  dplyr::slice_tail(n = 1) |>
  dplyr::ungroup() |>
  mutate(treated = prelim_sfha_category == "Added to prelim SFHA") |> 
  
   group_by(pin_num) |>
  mutate(
    g_add = if (any(addedto_prelim_sfha == 1, na.rm = TRUE)) {
      min(sale_year[addedto_prelim_sfha == 1], na.rm = TRUE)
    } else {
      0L
    }
  ) |>
  ungroup()


added_counts <- table(df_es_added$g_add, df_es_added$pre_date)



df_es_removed <- bldg_sfha |>
    filter(pin10 %in% groups$group_T_B$pin10 | pin10 %in% groups$group_C_D$pin10 )|>

  #filter(prelim_sfha_category %in% c("Removed from prelim SFHA", "Always SFHA")) |>
  mutate(pin_num = as.numeric(pin)) |>
  dplyr::arrange(pin_num, sale_year, sale_date) |>
  dplyr::group_by(pin_num, sale_year, pre_date_chr) |>
  dplyr::slice_tail(n = 1) |>
  dplyr::ungroup() |>
  mutate(treated = prelim_sfha_category == "Removed from prelim SFHA") |> 
  
   group_by(pin_num) |>
  mutate(
    g_remove = if (any(removedfrom_prelim_sfha == TRUE, na.rm = TRUE)) {
      min(sale_year[removedfrom_prelim_sfha == TRUE], na.rm = TRUE)
    } else {
      0L
    }
  ) |>
  ungroup()


removed_counts <- table(df_es_removed$g_remove, df_es_removed$pre_date)

Code

added_counts 
removed_counts

      
       2015-02-12 2019-07-01 2021-09-22
  0         11934     126483      19443
  2018          2          0          0
  2021          0          0          2
  2022          0          0          4
  2023          0          0          3
  2025          0          0          2

      
       2015-02-12 2019-07-01 2021-09-22
  0           287        314         72
  2015         19          0          0
  2016         12          0          0
  2017         15          0          0
  2018         11          0          0
  2019         13          0          0
  2020          8          0          0
  2021         14          0          5
  2022         19          0         11
  2023          2          0          4
  2024         10          0          2
  2025          2          0          2

Table 1: Table 1: Treated Sales by Cohort and Treatment Direction

Table 3

Code

sfha_category_stats <- bldg_sfha |>
  group_by(pin) |> mutate(n_sales = n()) |> ungroup() |>
  group_by(change_type_prelim) |>
 summarize(
    n_properties = n_distinct(pin),
    avg_sales_per_property = mean(n_sales),
    n_sales      = n(),
    avg_price = mean(sale_price),
    median_price = median(sale_price)  ) |>
  arrange(desc(n_properties))

sfha_category_stats
sfha_category_stats <- bldg_sfha |>
    mutate(ever_firm_update =  ifelse(pre_date =="2005-01-01", FALSE, TRUE) ) |>
  group_by(change_type_prelim) |>
  filter(ever_firm_update==TRUE) |>
  summarize(
    n_properties = n_distinct(pin),
    avg_sales_per_property = mean(n_sales),
    n_sales      = n(),
    avg_price = mean(sale_price),
    median_price = median(sale_price)  ) |>
  arrange(desc(n_properties))

sfha_category_stats
sfha_category_stats <- bldg_sfha |>
  group_by(prelim_sfha_category) |>
  summarize(
    n_properties = n_distinct(pin),
    avg_sales_per_property = mean(n_sales),
    n_sales      = n(),
    avg_price = mean(sale_price),
    median_price = median(sale_price)  ) |>
  arrange(desc(n_properties))

sfha_category_stats
sfha_category_stats <- bldg_sfha |>
    mutate(ever_firm_update =  ifelse(pre_date =="2005-01-01", FALSE, TRUE) ) |>
  group_by(prelim_sfha_category) |>
  filter(ever_firm_update==TRUE) |>
  summarize(
    n_properties = n_distinct(pin),
    avg_sales_per_property = mean(n_sales),
    n_sales      = n(),
    avg_price = mean(sale_price),
    median_price = median(sale_price)  ) |>
  arrange(desc(n_properties))

sfha_category_stats

change_type_prelim	n_properties	avg_sales_per_property	n_sales	avg_price	median_price
Never SFHA	284785	2.441165	653645	334790.9	250000
Always SFHA	6463	2.424474	14783	278345.5	197000
Changes SFHA	73	2.440476	168	135769.2	98250

change_type_prelim	n_properties	avg_sales_per_property	n_sales	avg_price	median_price
Never SFHA	69781	2.489990	162730	429702.7	300000
Always SFHA	257	2.448739	595	505159.2	250000
Changes SFHA	73	2.440476	168	135769.2	98250

prelim_sfha_category	n_properties	avg_sales_per_property	n_sales	avg_price	median_price
Never SFHA	284785	2.441815	653645	334790.9	250000
Always SFHA	6463	2.425015	14783	278345.5	197000
Removed from prelim SFHA	67	2.458064	155	124150.5	92000
Added to prelim SFHA	6	2.230769	13	274300.0	250000

prelim_sfha_category	n_properties	avg_sales_per_property	n_sales	avg_price	median_price
Never SFHA	69781	2.489990	162730	429702.7	300000
Always SFHA	257	2.448739	595	505159.2	250000
Removed from prelim SFHA	67	2.458064	155	124150.5	92000
Added to prelim SFHA	6	2.230769	13	274300.0	250000

Table 2: Table3: Descriptive Statistics for Sample

Code

allsales_bldg_sfha <- readRDS("data/processed/targets/sales/allsales_bldg_v2026_03.RDS")


allsales_bldg_sfha <- allsales_bldg_sfha |> 
  filter(!pin10 %in% c("0122400028", "0516106013"))|> 
  filter(!(Triad != "City" & log_price > 13.5))

Code

allsales_bldg_sfha |>
  filter(!pin10 %in% c("0122400028", "0516106013")) |> #filter(log_price > 13.5)|>
  filter(prelim_sfha_category %in% c("Always SFHA", "Never SFHA")) |> 
    group_by(sale_year, prelim_sfha_category, pre_date_chr) |> 
  summarise(avg_logprice = mean(log_price, na.rm = TRUE) )|> 

  ggplot(aes(x = sale_year , y = avg_logprice, shape = prelim_sfha_category)) +
  theme_classic() + 
  theme(legend.position = "bottom")+
  geom_smooth(aes(lty = prelim_sfha_category), method = "lm", se = FALSE, color = "darkgray", alpha= 0.5, lwd = 1) + 
  geom_line(aes(lty = prelim_sfha_category)) +
  geom_point(size = 2) + 
  
  scale_x_continuous(breaks = c(2010, 2015, 2020, 2026)) +
  labs(#title = "Cohort Trends by SFHA Group: Always In vs. Always Out",
       #subtitle = "Includes Chicago",
       y = "Avg. Logged Sale Price",
       x = NULL,
              shape = "SFHA Group",
       lty = "SFHA Group",
              caption = "Graphed with all sales, not just repeat sales. Includes Chicago") +
  facet_wrap(~pre_date_chr)

# EXCLUDE CHCICGO
allsales_bldg_sfha |>   
  filter(!pin10 %in% c("0122400028", "0516106013")) |> filter(Triad != "City")|>
  filter(prelim_sfha_category %in% c("Always SFHA", "Never SFHA")) |> 
    group_by(sale_year, prelim_sfha_category, pre_date_chr) |> 
  summarise(avg_logprice = mean(log_price, na.rm = TRUE) )|> 

  ggplot(aes(x = sale_year , y = avg_logprice, shape = prelim_sfha_category)) +
  theme_classic() + 
  theme(legend.position = "bottom")+
  geom_smooth(aes(lty = prelim_sfha_category), method = "lm", se = FALSE, color = "darkgray", alpha= 0.5, lwd = 1) + 
  geom_line(aes(lty = prelim_sfha_category)) +
  geom_point(size = 2) + 
  
  scale_x_continuous(breaks = c(2010, 2015, 2020, 2026)) + 
  labs(#title = "Cohort Trends by SFHA Group: Always In vs. Always Out",
       #subtitle = "Excludes Chicago",
       y = "Avg. Logged Sale Price",
       x = NULL,
              shape = "SFHA Group",
       lty = "SFHA Group",
       caption = "Graphed with all sales, not just repeat sales. Excludes Chicago") +
  facet_wrap(~pre_date_chr)

Code

allsales_bldg_sfha |>   
   filter(!pin10 %in% c("0122400028", "0516106013")) |> 
  #filter(!(Triad != "City" & log_price > 13.5))|>
  filter(prelim_sfha_category %in% c("Always SFHA", "Never SFHA")) |> 
    group_by(sale_year, prelim_sfha_category) |> 
  summarise(avg_logprice = mean(log_price, na.rm = TRUE) )|> 

  ggplot(aes(x = sale_year , y = avg_logprice, shape = prelim_sfha_category)) +
  theme_classic() + 
  theme(legend.position = "bottom")+
  geom_smooth(aes(lty = prelim_sfha_category), method = "lm", se = FALSE, color = "darkgray", alpha= 0.5, lwd = 1) + 
  geom_line(aes(lty = prelim_sfha_category)) +
  geom_point(size = 2.5) + 
  
  scale_x_continuous(breaks = c(2010, 2015, 2020, 2026)) + 
  labs(title = "County Trends by SFHA Group: Always In vs. Always Out",
       y = "Avg. Logged Sale Price",
       x = NULL,
       caption= "Uses All Sales",
              shape = "SFHA Group",
       lty = "SFHA Group")

Table 4: County Wide Effect

Code

bldg_sfha <- bldg_sfha |>
  mutate(nbhd_code = if_else(
    is.na(nbhd_code),
    "Missing",
    as.character(nbhd_code)
  ))

mods <- list (
 no_fe <- feols(
    log_price ~ in_prelim_sfha,
    vcov = ~pin10,
    data = bldg_sfha
  ),
  
 one_fe <- feols(
    log_price ~ in_prelim_sfha | sale_year,
    data = bldg_sfha,
     vcov = ~pin10
  ),
   pin_fe <- feols(
    log_price ~ in_prelim_sfha | pin ,
    data = bldg_sfha,
        vcov = ~pin10,

  ),
  
  two_fe <- feols(
    log_price ~ in_prelim_sfha | pin + sale_year,
    data = bldg_sfha,
     vcov = ~pin10
  ),
  
  three_fe <- feols(
    log_price ~ in_prelim_sfha | pin + sale_year + qtr,
    data = bldg_sfha,
     vcov = ~pin10
)
)
# mod_prelim_3fe <- feols(
#   log_price ~ in_prelim_sfha | pin + sale_year + qtr,
#   data = bldg_sfha
# )
# 
# mod_eff_3fe <- feols(
#   log_price ~ in_eff_sfha | pin + sale_year + qtr,
#   data = bldg_sfha
# )
# 
# mods_names <- list(
#   "Preliminary"      = mod_prelim,
#   "Effective"        = mod_eff
#   # , "Prelim w/ Qtr FE" = mod_prelim_3fe,
#   #"Eff w/ Qtr FE"    = mod_eff_3fe
# )
# 
# vcovs <- list(
#   "Parcel" = ~pin10,
#   "Municipality" = ~clean_name
#   # , "Neighborhood" = ~nbhd_code
# )
# 
# gof_labels <- c(
#   "nobs"      = "Observations",
#   "r.squared" = "R-squared"
# )

coef_labels = c(
  "in_prelim_sfhaTRUE" = "In SFHA (Preliminary Map)",
  "in_eff_sfhaTRUE"    = "In SFHA (Effective Map)"
)

# ms_models <- list()
# ms_vcov   <- list()
# 
# for (m in names(mods)) {
#   for (v in names(vcovs)) {
#     ms_models[[paste(m, v, sep = "\n")]] <- mods[[m]]
#     ms_vcov[[length(ms_vcov) + 1]] <- vcovs[[v]]
#   }
# }
# 
# modelsummary::modelsummary(
#   ms_models,
#   vcov = ms_vcov,
#   stars = TRUE,
# 
# )

modelsummary::modelsummary(
  mods,
  stars = TRUE,
 gof_omit = "IC|Log|Adj|Within"
#  ,
 # estimate = "{estimate}{stars}",
#  statistic = "\n({std.error})"#,
,  coef_map = coef_labels
)

	(1)	(2)	(3)	(4)	(5)
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001
In SFHA (Preliminary Map)	-0.187**	-0.184**	-0.445***	-0.028	-0.022
	(0.067)	(0.067)	(0.064)	(0.039)	(0.041)
Num.Obs.	668596	668596	667288	667288	667288
R2	0.001	0.076	0.796	0.859	0.860
RMSE	0.86	0.83	0.39	0.32	0.32
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10
FE: sale_year		X		X	X
FE: pin			X	X	X
FE: qtr					X

Table 3: Table 4

Code

modelsummary::modelsummary(
  mods,
  stars = TRUE,
 gof_omit = "IC|Log|Adj|Within",
vcov = ~clean_name
,  coef_map = coef_labels
)

	(1)	(2)	(3)	(4)	(5)
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001
In SFHA (Preliminary Map)	-0.187	-0.184	-0.445***	-0.028	-0.022
	(0.135)	(0.136)	(0.059)	(0.023)	(0.026)
Num.Obs.	668596	668596	667288	667288	667288
R2	0.001	0.076	0.796	0.859	0.860
RMSE	0.86	0.83	0.39	0.32	0.32
Std.Errors	by: clean_name	by: clean_name	by: clean_name	by: clean_name	by: clean_name
FE: sale_year		X		X	X
FE: pin			X	X	X
FE: qtr					X

Code

modelsummary::modelsummary(
  mods,
  stars = TRUE,
 gof_omit = "IC|Log|Adj|Within",
vcov = ~nbhd_code
,  coef_map = coef_labels
)

	(1)	(2)	(3)	(4)	(5)
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001
In SFHA (Preliminary Map)	-0.187	-0.184	-0.445***	-0.028	-0.022
	(0.122)	(0.123)	(0.082)	(0.046)	(0.049)
Num.Obs.	668596	668596	667288	667288	667288
R2	0.001	0.076	0.796	0.859	0.860
RMSE	0.86	0.83	0.39	0.32	0.32
Std.Errors	by: nbhd_code	by: nbhd_code	by: nbhd_code	by: nbhd_code	by: nbhd_code
FE: sale_year		X		X	X
FE: pin			X	X	X
FE: qtr					X

Table 4 robustness check

Table 5: Simple Directional Effects - Standard TWFE

Code

prelim_twfe <- list(

  
 "I" =  feols(
    log(sale_price) ~ addedto_prelim_sfha  | pin + sale_year,
    cluster = ~pin10,
        data = bldg_sfha
  ),

  "II" =  feols(
    log(sale_price) ~ removedfrom_prelim_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,

  ),
 
  # Asymmetric: added vs removed in one equation
  "0" = feols(
    log(sale_price) ~ addedto_prelim_sfha + removedfrom_prelim_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,
  ))
 


all_models_formatted <- modelsummary::modelsummary(prelim_twfe,
               output = "flextable",
               fmt = function(x) round(x, 2),

             stars = TRUE)

all_models_formatted

	I	II	0
addedto_prelim_sfhaTRUE	0.13		0.13
	(0.1)		(0.1)
removedfrom_prelim_sfhaTRUE		0.04	0.04
		(0.04)	(0.04)
Num.Obs.	667288	667288	667288
R2	0.859	0.859	0.859
R2 Adj.	0.751	0.751	0.751
R2 Within	0.000	0.000	0.000
R2 Within Adj.	-0.000	-0.000	-0.000
AIC	966832.3	966832.2	966833.7
BIC	4276346.6	4276346.5	4276359.4
RMSE	0.32	0.32	0.32
Std.Errors	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X
FE: sale_year	X	X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Table 4: Table 5: Effect of Being Mapped Into or Out of a FEMA Flood Plain

Table 5 & 6: Directional Effects & Control Group Subsamples

Code

##| eval: false

prelim_twfe <- list(

  
 "I" =  feols(
    log(sale_price) ~ addedto_prelim_sfha  | pin + sale_year,
    cluster = ~pin10,
        data = bldg_sfha
  ),

    "I Subset" = feols(
    log(sale_price) ~ addedto_prelim_sfha  | pin + sale_year,
    cluster = ~ pin10,
        data = (bldg_sfha |> filter(change_type_prelim != "Always SFHA" & treat_year_remove_prelim == 10000)),

  ),

  "II" =  feols(
    log(sale_price) ~ removedfrom_prelim_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,

  ),


  "II Subset" =  feols(
    log(sale_price) ~ removedfrom_prelim_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = (bldg_sfha |> filter(change_type_prelim != "Never SFHA" & treat_year_add_prelim == 10000)),

  ),
 
  # Symmetric effect of location in risk zone
    "Symmetric" = feols(
    log(sale_price) ~ in_prelim_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,
  ),
 
  # Asymmetric: added vs removed in one equation
  "0" = feols(
    log(sale_price) ~ addedto_prelim_sfha + removedfrom_prelim_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,
  )
)

all_models_formatted <- modelsummary::modelsummary(prelim_twfe,
               output = "flextable",
               fmt = function(x) round(x, 2),

             stars = TRUE)

all_models_formatted

	I	I Subset	II	II Subset	Symmetric	0
addedto_prelim_sfhaTRUE	0.13	0.13				0.13
	(0.1)	(0.1)				(0.1)
removedfrom_prelim_sfhaTRUE			0.04	0.08		0.04
			(0.04)	(0.05)		(0.04)
in_prelim_sfhaTRUE					-0.03
					(0.04)
Num.Obs.	667288	652380	667288	14908	667288	667288
R2	0.859	0.858	0.859	0.894	0.859	0.859
R2 Adj.	0.751	0.749	0.751	0.812	0.751	0.751
R2 Within	0.000	0.000	0.000	0.000	0.000	0.000
R2 Within Adj.	-0.000	-0.000	-0.000	0.000	-0.000	-0.000
AIC	966832.3	949742.5	966832.2	16063.4	966832.5	966833.7
BIC	4276346.6	4178679.2	4276346.5	65647.9	4276346.8	4276359.4
RMSE	0.32	0.32	0.32	0.27	0.32	0.32
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X	X	X	X
FE: sale_year	X	X	X	X	X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Table 5: Table 5 & 6 Effect of Being Mapped Into or Out of a FEMA Flood Plain. A lot of models all by each other.

Table 7: Restrict Samples - Updated FIRM Panels only

Code

# function with default options:
fn_valid_cohorts <- function(df){
  
  df |> 
    mutate(pre_date_chr = as.character(pre_date_chr)) |>
    group_by(pre_date_chr) |>
    summarise(
      n = n(),
      n_treated = sum(removedfrom_prelim_sfha == TRUE, na.rm = TRUE),
      n_untreated = sum(removedfrom_prelim_sfha == FALSE, na.rm = TRUE),
          .groups = "drop"

      ) |>
    filter(n_treated > 0,  n_untreated > 0) |>
    pull(pre_date_chr)
}



#table(bldg_sfha$Triad, bldg_sfha$prelim_sfha_category)

df_sub <- bldg_sfha |> 
      filter(pin10 %in% groups$group_T_B$pin10 | pin10 %in% groups$group_C_D$pin10 )
  # filter(change_type_prelim != "Never SFHA" & 
  #          # and properties that are never treated for being added to the SFHA
  #          treat_year_add_prelim == 10000 & treat_year_add_eff == 10000)

valid_cohorts_bldg <- fn_valid_cohorts(df_sub)
bldgs_fsplit <- df_sub |> filter(pre_date_chr %in% valid_cohorts_bldg)


#table(parcels$Triad, parcels$prelim_sfha_category)
#table(parcels$removed_pre_thisyear, parcels$year)

# df_sub_parcels <- parcels |> 
#       filter(pin10 %in% groups$group_T_B$pin10 | pin10 %in% groups$group_C_D$pin10 )
#   # filter(change_type_prelim != "Never SFHA" & 
#   #          # and properties that are never treated for being added to the SFHA
#   #          treat_year_add_prelim == 10000 & treat_year_add_eff == 10000)
# 
# valid_cohorts_parcels <- fn_valid_cohorts(df = df_sub_parcels)
# parcels_fsplit <- df_sub_parcels |> filter(pre_date_chr %in% valid_cohorts_parcels)
# 
# table(bldg_1in500$removed_pre_thisyear, bldg_1in500$year)
# table(bldg_1in500$removedfrom_prelim_sfha, bldg_1in500$year)
# table(bldg_1in500$Triad, bldg_1in500$prelim_sfha_category)
# df_sub_1in500 <- bldg_1in500 |> 
#   filter(change_type_prelim != "Never SFHA" & 
#            # and properties that are never treated for being added to the SFHA
#            treat_year_add_prelim == 10000 & treat_year_add_eff == 10000)
# valid_cohorts_1in500 <- fn_valid_cohorts(df = df_sub_1in500)
# bldg_1in500_fsplit <- df_sub_1in500 |> filter(pre_date_chr %in% valid_cohorts_1in500)
# 
# 




models <- list(
 "Bldg SFHA" = feols(
       log(sale_price) ~ removedfrom_prelim_sfha | pin + sale_year,

   # log(sale_price) ~ event_remapped*ever_removed_prelim | pin + sale_year,
    data = bldgs_fsplit,
  fsplit = ~pre_date_chr,
    cluster = ~pin10
  )
  # 
  # "Bldg 1in500" = feols(
  #          log(sale_price) ~ removedfrom_prelim_sfha | pin + sale_year,
  # 
  # #  log(sale_price) ~ event_remapped*ever_removed_prelim | pin + sale_year,
  #   data = bldg_1in500_fsplit,
  # fsplit = ~pre_date_chr,
  #   cluster = ~pin10
  # 
  # ),
  # 
  # "Parcels" = feols(
  #          log(sale_price) ~ removedfrom_prelim_sfha | pin + sale_year,
  # 
  # #  log(sale_price) ~ event_remapped*ever_removed_prelim | pin + sale_year,
  #   data = parcels_fsplit,
  # fsplit = ~pre_date_chr,
  #   cluster = ~pin10
  # 
  # )
)

all_models_formatted <- modelsummary::modelsummary(models,
               output = "flextable",
               fmt = function(x) round(x, 2),

             stars = TRUE)

all_models_formatted

	Bldg SFHA sample: Full sample	Bldg SFHA sample: 2015-02-12	Bldg SFHA sample: 2021-09-22
removedfrom_prelim_sfhaTRUE	-0.08	-0.06	0.09
	(0.08)	(0.1)	(0.21)
Num.Obs.	533	431	100
R2	0.940	0.931	0.959
R2 Adj.	0.888	0.870	0.900
R2 Within	0.006	0.003	0.005
R2 Within Adj.	0.002	-0.001	-0.019
AIC	187.7	94.5	84.2
BIC	1244.5	915.8	237.9
RMSE	0.18	0.17	0.20
Std.Errors	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X
FE: sale_year	X	X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Table 6: Table 7: Mapped out of Preliminary SFHA - Restricted Samples

Table 8

Code

#table(bldg_sfha$added_pre_thisyear, bldg_sfha$year)

# function with default options:
fn_valid_cohorts <- function(df){
  
  df |>  
    mutate(pre_date_chr = as.character(pre_date_chr)) |>
    group_by(pre_date_chr) |>
    summarise(
      n = n(),
      n_treated = sum(addedto_prelim_sfha == TRUE, na.rm = TRUE),
      n_untreated = sum(addedto_prelim_sfha == FALSE, na.rm = TRUE),
          .groups = "drop"

      ) |>
    filter(n_treated > 0,  n_untreated > 0) |>
    pull(pre_date_chr)
}




table(bldg_sfha$Triad, bldg_sfha$prelim_sfha_category)

       
        Added to prelim SFHA Always SFHA Never SFHA Removed from prelim SFHA
  South                    2        5244     157264                      129
  North                   11        5228     145900                       26
  City                     0        4311     350481                        0

Code

df_sub <- bldg_sfha |> 
    filter(pin10 %in% groups$group_T_A$pin10 | pin10 %in% groups$group_C_B$pin10 )
  # filter(change_type_prelim != "Always SFHA" & 
  #          # and properties that are never treated for being removed from the SFHA
  #          treat_year_remove_prelim == 10000 & treat_year_remove_eff == 10000)

valid_cohorts_bldg <- fn_valid_cohorts(df_sub)
bldgs_fsplit <- df_sub |> filter(pre_date_chr %in% valid_cohorts_bldg)


# table(parcels$Triad, parcels$prelim_sfha_category)
# table(parcels$removed_pre_thisyear, parcels$year)

df_sub_parcels <- parcels |> 
    filter(pin10 %in% groups$group_T_A$pin10 | pin10 %in% groups$group_C_B$pin10 )
  # filter(change_type_prelim != "Always SFHA" & 
  #          # and properties that are never treated for being removed from the SFHA
  #          treat_year_remove_prelim == 10000 & treat_year_remove_eff == 10000)
valid_cohorts_parcels <- fn_valid_cohorts(df = df_sub_parcels)
parcels_fsplit <- df_sub_parcels |> filter(pre_date_chr %in% valid_cohorts_parcels)


# 
# table(bldg_1in500$removed_pre_thisyear, bldg_1in500$year)
# table(bldg_1in500$removedfrom_prelim_sfha, bldg_1in500$year)
# table(bldg_1in500$Triad, bldg_1in500$prelim_sfha_category)

# df_sub_1in500 <- bldg_1in500 |> 
#     filter(pin10 %in% groups$group_T_A$pin10 | pin10 %in% groups$group_C_B$pin10 )
  # filter(change_type_prelim != "Always SFHA" & 
  #          # and properties that are never treated for being removed from the SFHA
  #          treat_year_remove_prelim == 10000 & treat_year_remove_eff == 10000)
#valid_cohorts_1in500 <- fn_valid_cohorts(df = df_sub_1in500)
#bldg_1in500_fsplit <- df_sub_1in500 |> filter(pre_date_chr %in% valid_cohorts_1in500)




models_pre_add <- list(
  # Heterogeneity by Triad: added/removed interacted with Triad
  # 
  # # Heterogeneity by Triad: added/removed interacted with Triad
  "Buildings in SFHA" =  feols(
    log(sale_price) ~ addedto_prelim_sfha | pin + sale_year,
    fsplit = ~pre_date_chr,
    cluster = ~pin10,
    data = bldgs_fsplit
  ),
    "Buildings in SFHA" =  feols(
    log(sale_price) ~ addedto_prelim_sfha*condo | pin + sale_year,
    fsplit = ~pre_date_chr,
    cluster = ~pin10,
    data = bldgs_fsplit
  ),

  "Buildings in SFHA" =  feols(
    log(sale_price) ~ addedto_prelim_sfha | pin + sale_year,
    cluster = ~pin10,
    data = bldg_sfha
  )
  # 
  # "Buildings - 1-in-500" =  feols(
  #   log(sale_price) ~ addedto_prelim_sfha | pin + year,
  #   fsplit = ~pre_date_chr,
  #   cluster = ~pin10,
  #   data = bldg_1in500_fsplit
  # ),
  # 
  # "Buildings - 1-in-500" =  feols(
  #   log(sale_price) ~ addedto_prelim_sfha | pin + sale_year,
  #   cluster = ~pin10,
  #   data = bldg_1in500
  # )
)

all_models_formatted <- modelsummary::modelsummary(models_pre_add,
                                                   output = "flextable",
                                                   fmt = function(x) round(x, 2),
                                                   
                                                   stars = TRUE)

all_models_formatted

	Buildings in SFHA sample: Full sample	Buildings in SFHA sample: 2015-02-12	Buildings in SFHA sample: 2021-09-22	Buildings in SFHA sample: Full sample	Buildings in SFHA sample: 2015-02-12	Buildings in SFHA sample: 2021-09-22	Buildings in SFHA
addedto_prelim_sfhaTRUE	0.12	0.11***	0.13	0.12	0.11***	0.13	0.13
	(0.11)	(0.02)	(0.13)	(0.11)	(0.02)	(0.13)	(0.1)
condoNot Condo				0.25***		0.29***
				(0.01)		(0.01)
Num.Obs.	33067	12509	20558	33067	12509	20558	667288
R2	0.906	0.893	0.913	0.906	0.893	0.913	0.859
R2 Adj.	0.832	0.807	0.844	0.832	0.807	0.844	0.751
R2 Within	0.000	0.000	0.000	0.001	0.000	0.002	0.000
R2 Within Adj.	-0.000	-0.000	-0.000	0.001	-0.000	0.002	-0.000
AIC	14266.7	4679.3	9133.6	14236.9	4679.3	9095.2	966832.3
BIC	137309.6	46006.0	81258.2	137288.2	46006.0	81227.7	4276346.6
RMSE	0.19	0.19	0.19	0.19	0.19	0.19	0.32
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X	X	X	X	X
FE: sale_year	X	X	X	X	X	X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Table 7: Table 8: Mapped into Preliminary SFHA - Restricted Sample

Table 9: County-wide and Cohort Directional Models using Extended TWFE

Added To Risk Zone

Code

mod1c = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_add_prelim, 
  vcov = ~pin10, 
  data = (bldg_sfha |>
            filter(
             (pin %in% groups$group_T_A$pin |
              pin %in% groups$group_C_A$pin |
              pin %in% groups$group_C_B$pin | 
              pin %in% groups$group_C_C$pin |
              pin %in% groups$group_C_D$pin )
                     )
          )
  )
# mod1c
emfx(mod1c)

Code

emfx(mod1c, type = "group")


 .Dtreat Estimate Std. Error     z Pr(>|z|)     S   2.5 %  97.5 %
    TRUE  -0.0173   0.000737 -23.5   <0.001 401.6 -0.0187 -0.0158

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE


 treat_year_add_prelim Estimate Std. Error     z Pr(>|z|)     S   2.5 % 97.5 %
                  2015   0.0000         NA    NA       NA    NA      NA     NA
                  2021  -0.0207   0.000884 -23.5   <0.001 401.6 -0.0225 -0.019

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Table 8: Control Groups A, B, C, & D (so all PINs that never changed SFHA status; both always in, and always out)

Mapped Out of SFHA, Column 3

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_remove_prelim, 
  vcov = ~pin10, 
  data = (bldg_sfha|> 
                filter(pin10 %in% groups$group_T_B$pin10 | 
                         pin10 %in% groups$group_C_D$pin10 | 
                         pin10 %in% groups$group_C_C$pin10 |
                         pin10 %in% groups$group_C_A$pin10 | 
                         pin10 %in% groups$group_C_B$pin10 )

          )
  )
# mod2
emfx(mod2)


 .Dtreat Estimate Std. Error    z Pr(>|z|)   S  2.5 % 97.5 %
    TRUE    0.119     0.0549 2.17   0.0301 5.1 0.0115  0.227

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

emfx(mod2, type = "group")


 treat_year_remove_prelim Estimate Std. Error    z Pr(>|z|)    S   2.5 % 97.5 %
                     2015    0.109     0.0649 1.68   0.0931  3.4 -0.0182  0.236
                     2021    0.174     0.0255 6.82   <0.001 36.7  0.1242  0.224

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

plot(emfx(mod2, type = "group"))

Removed - Subset (Column 4)

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_remove_prelim, 
  vcov = ~pin10, 
  data = (bldg_sfha|> 
                filter(pin10 %in% groups$group_T_B$pin10 | 
                         pin10 %in% groups$group_C_D$pin10 | 
                         pin10 %in% groups$group_C_C$pin10 )

          )
  )
# mod2
emfx(mod2)


 .Dtreat Estimate Std. Error    z Pr(>|z|)   S  2.5 % 97.5 %
    TRUE    0.138     0.0612 2.26   0.0239 5.4 0.0183  0.258

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

emfx(mod2, type = "group")


 treat_year_remove_prelim Estimate Std. Error    z Pr(>|z|)    S   2.5 % 97.5 %
                     2015    0.131     0.0715 1.83   0.0673  3.9 -0.0093  0.271
                     2021    0.178     0.0316 5.62   <0.001 25.6  0.1157  0.240

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

#plot(emfx(mod2, type = "group"))

Table 10: Regional Slopes

Mapped into SFHA

Code

library(etwfe)

mod1 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  xvar = Triad, 
  cgroup = "never",
  gvar = treat_year_add_prelim,
  vcov = ~pin10, 
  data = (bldg_sfha |> 
            filter(treat_year_remove_prelim==10000
                   
                   & change_type_prelim != "Always SFHA")
          )
  
  )
#mod1
emfx(mod1)


 .Dtreat Triad Estimate Std. Error     z Pr(>|z|)     S   2.5 %  97.5 %
    TRUE South   0.0000         NA    NA       NA    NA      NA      NA
    TRUE North  -0.0388    0.00154 -25.2   <0.001 461.9 -0.0419 -0.0358

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

emfx(mod1, type = "group")


 treat_year_add_prelim Triad Estimate Std. Error     z Pr(>|z|)     S   2.5 %
                  2015 South   0.0000         NA    NA       NA    NA      NA
                  2021 North  -0.0388    0.00154 -25.2   <0.001 461.9 -0.0419
  97.5 %
      NA
 -0.0358

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

plot(emfx(mod1))

Added with regional slopes

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_add_prelim, 
    xvar = Triad, 

  vcov = ~pin10, 
  data = (bldg_sfha|> 
                filter(pin10 %in% groups$group_T_A$pin10 | 
                         pin10 %in% groups$group_C_A$pin10 |
                         pin10 %in% groups$group_C_B$pin10)
          )
  )
# mod2
emfx(mod2)


 .Dtreat Triad Estimate Std. Error     z Pr(>|z|)     S   2.5 %  97.5 %
    TRUE South   0.0000         NA    NA       NA    NA      NA      NA
    TRUE North  -0.0388    0.00154 -25.2   <0.001 462.1 -0.0419 -0.0358

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

emfx(mod2, type = "group")


 treat_year_add_prelim Triad Estimate Std. Error     z Pr(>|z|)     S   2.5 %
                  2015 South   0.0000         NA    NA       NA    NA      NA
                  2021 North  -0.0388    0.00154 -25.2   <0.001 462.1 -0.0419
  97.5 %
      NA
 -0.0358

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_add_prelim, 
  xvar = group_name_prelim, 
  vcov = ~pin10, 
  data = (bldg_sfha|> 
                filter(pin10 %in% groups$group_T_A$pin10 | 
                         pin10 %in% groups$group_C_A$pin10 |
                         pin10 %in% groups$group_C_B$pin10 
                         )
          )
  )
# mod2
emfx(mod2)


 group_name_prelim .Dtreat Estimate Std. Error     z Pr(>|z|)     S   2.5 %
              2015    TRUE   0.0000         NA    NA       NA    NA      NA
              2021    TRUE  -0.0251    0.00106 -23.8   <0.001 411.9 -0.0272
 97.5 %
     NA
 -0.023

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

emfx(mod2, type = "group")


 group_name_prelim treat_year_add_prelim Estimate Std. Error     z Pr(>|z|)
              2015                  2015   0.0000         NA    NA       NA
              2021                  2021  -0.0251    0.00106 -23.8   <0.001
     S   2.5 % 97.5 %
    NA      NA     NA
 411.9 -0.0272 -0.023

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_remove_prelim, 
    xvar = Triad, 

  vcov = ~pin10, 
  data = (bldg_sfha|> 
                filter(pin10 %in% groups$group_T_B$pin10 | 
                         pin10 %in% groups$group_C_D$pin10 |
                         pin10 %in% groups$group_C_C$pin10 
                         )
          )
  )
# mod2
#emfx(mod2)
emfx(mod2, type = "group")


 treat_year_remove_prelim Triad Estimate Std. Error      z Pr(>|z|)   S   2.5 %
                     2015 South  -0.0436     0.0718 -0.607  0.54369 0.9 -0.1842
                     2021 North   0.1072     0.0366  2.926  0.00343 8.2  0.0354
 97.5 %
 0.0971
 0.1790

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  cgroup = "never",
  gvar = treat_year_remove_prelim, 
  xvar = group_name_prelim, 

  vcov = ~pin10, 
  data = (bldg_sfha|> 
                filter(pin10 %in% groups$group_T_B$pin10 | 
                         pin10 %in% groups$group_C_D$pin10 |
                         pin10 %in% groups$group_C_C$pin10 
                         )
          )
  )
# mod2
emfx(mod2)


 group_name_prelim .Dtreat Estimate Std. Error    z Pr(>|z|)    S    2.5 %
              2015    TRUE    0.132     0.0721 1.83   0.0669  3.9 -0.00919
              2021    TRUE    0.181     0.0318 5.68   <0.001 26.2  0.11855
 97.5 %
  0.273
  0.243

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

emfx(mod2, type = "group")


 group_name_prelim treat_year_remove_prelim Estimate Std. Error    z Pr(>|z|)
              2015                     2015    0.132     0.0721 1.83   0.0669
              2021                     2021    0.181     0.0318 5.68   <0.001
    S    2.5 % 97.5 %
  3.9 -0.00919  0.273
 26.2  0.11855  0.243

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

#table(bldg_sfha$added_pre_thisyear, bldg_sfha$year)

# function with default options:
fn_valid_cohorts <- function(df){
  
  df |> 
    mutate(eff_date_chr = as.character(eff_date_chr)) |>
    group_by(eff_date_chr) |>
    summarise(
      n = n(),
      n_treated = sum(removedfrom_eff_sfha == TRUE, na.rm = TRUE),
      n_untreated = sum(removedfrom_eff_sfha == FALSE, na.rm = TRUE),
          .groups = "drop"

      ) |>
    filter(n_treated > 0,  n_untreated > 0) |>
    pull(eff_date_chr)
}





df_sub <- bldg_sfha |> 
      filter(pin10 %in% groups$group_T_B$pin10 | pin10 %in% groups$group_C_D$pin10 )
  # filter(change_type_prelim != "Never SFHA" & 
  #          # and properties that are never treated for being addedto from the SFHA
  #          treat_year_add_prelim == 10000 & treat_year_add_eff == 10000)

valid_cohorts_bldg <- fn_valid_cohorts(df_sub)
bldgs_fsplit <- df_sub |> filter(eff_date_chr %in% valid_cohorts_bldg)


# table(parcels$Triad, parcels$prelim_sfha_category)
# table(parcels$removed_pre_thisyear, parcels$year)

df_sub_parcels <- parcels |> 
      filter(pin10 %in% groups$group_T_B$pin10 | pin10 %in% groups$group_C_D$pin10 )
  # filter(change_type_prelim != "Never SFHA" & 
  #          # and properties that are never treated for being added to the SFHA
  #          treat_year_add_prelim == 10000 & treat_year_add_eff == 10000)
valid_cohorts_parcels <- fn_valid_cohorts(df = df_sub_parcels)
parcels_fsplit <- df_sub_parcels |> filter(eff_date_chr %in% valid_cohorts_parcels)


# 
# table(bldg_1in500$removed_pre_thisyear, bldg_1in500$year)
# table(bldg_1in500$removedfrom_prelim_sfha, bldg_1in500$year)
# table(bldg_1in500$Triad, bldg_1in500$prelim_sfha_category)

df_sub_1in500 <- bldg_1in500 |> 
      filter(pin10 %in% groups$group_T_B$pin10 | pin10 %in% groups$group_C_D$pin10 ) 
  # filter(change_type_prelim != "Never SFHA" & 
  #          # and properties that are never treated for being addedto the SFHA
  #          treat_year_add_prelim == 10000 & treat_year_add_eff == 10000)
valid_cohorts_1in500 <- fn_valid_cohorts(df = df_sub_1in500)
bldg_1in500_fsplit <- df_sub_1in500 |> filter(eff_date_chr %in% valid_cohorts_1in500)




models <- list(
  
  # "Buildings in SFHA" =  feols(
  #   log(sale_price) ~ addedto_eff_sfha | pin + sale_year,
  #   fsplit = ~eff_date_chr,
  #   cluster = ~pin10,
  #   data = bldgs_fsplit
  # ),

  "Buildings in SFHA" =  feols(
    log(sale_price) ~ removedfrom_eff_sfha | pin + sale_year,
    fsplit = ~eff_date_chr,

    cluster = ~pin10,
    data = parcels_fsplit
  ),

  "Buildings - 1-in-500" =  feols(
    log(sale_price) ~ removedfrom_eff_sfha | pin + year,
    fsplit = ~eff_date_chr,
    cluster = ~pin10,
    data = bldg_1in500_fsplit
  )
)

all_models_formatted <- modelsummary::modelsummary(models,
                                                   output = "flextable",
                                                   fmt = function(x) round(x, 2),
                                                   
                                                   stars = TRUE)

all_models_formatted

	Buildings in SFHA sample: Full sample	Buildings in SFHA sample: 2019-11-01	Buildings - 1-in-500 sample: Full sample	Buildings - 1-in-500 sample: 2019-11-01
removedfrom_eff_sfhaTRUE	0.07	0.07	-0.13	-0.13
	(0.07)	(0.07)	(0.09)	(0.09)
Num.Obs.	431	431	431	431
R2	0.931	0.931	0.931	0.931
R2 Adj.	0.870	0.870	0.871	0.871
R2 Within	0.006	0.006	0.009	0.009
R2 Within Adj.	0.001	0.001	0.004	0.004
AIC	93.4	93.4	92.1	92.1
BIC	914.8	914.8	913.5	913.5
RMSE	0.17	0.17	0.17	0.17
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X	X
FE: year			X	X
FE: sale_year	X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Table 9: Table 11: Mapped out of EFFECTIVE SFHA - Restricted Sample

Code

#table(bldg_sfha$added_pre_thisyear, bldg_sfha$year)

# function with default options:
fn_valid_cohorts <- function(df){
  
  df |> 
    mutate(eff_date_chr = as.character(eff_date_chr)) |>
    group_by(eff_date_chr) |>
    summarise(
      n = n(),
      n_treated = sum(addedto_eff_sfha == TRUE, na.rm = TRUE),
      n_untreated = sum(addedto_eff_sfha == FALSE, na.rm = TRUE),
          .groups = "drop"

      ) |>
    filter(n_treated > 0,  n_untreated > 0) |>
    pull(eff_date_chr)
}




table(bldg_sfha$Triad, bldg_sfha$prelim_sfha_category)

       
        Added to prelim SFHA Always SFHA Never SFHA Removed from prelim SFHA
  South                    2        5244     157264                      129
  North                   11        5228     145900                       26
  City                     0        4311     350481                        0

Code

df_sub <- bldg_sfha |> 
  filter(change_type_prelim != "Always SFHA" & 
           # and properties that are never treated for being removed from the SFHA
           treat_year_remove_prelim == 10000 & treat_year_remove_eff == 10000)

valid_cohorts_bldg <- fn_valid_cohorts(df_sub)
bldgs_fsplit <- df_sub |> filter(eff_date_chr %in% valid_cohorts_bldg)


# table(parcels$Triad, parcels$prelim_sfha_category)
# table(parcels$removed_pre_thisyear, parcels$year)

df_sub_parcels <- parcels |> 
  filter(change_type_prelim != "Always SFHA" & 
           # and properties that are never treated for being removed from the SFHA
           treat_year_remove_prelim == 10000 & treat_year_remove_eff == 10000)
valid_cohorts_parcels <- fn_valid_cohorts(df = df_sub_parcels)
parcels_fsplit <- df_sub_parcels |> filter(eff_date_chr %in% valid_cohorts_parcels)


# 
# table(bldg_1in500$removed_pre_thisyear, bldg_1in500$year)
# table(bldg_1in500$removedfrom_prelim_sfha, bldg_1in500$year)
# table(bldg_1in500$Triad, bldg_1in500$prelim_sfha_category)

df_sub_1in500 <- bldg_1in500 |> 
  filter(change_type != "Always SFHA" & 
           # and properties that are never treated for being removed from the SFHA
           treat_year_remove_prelim == 10000 & treat_year_remove_eff == 10000)
valid_cohorts_1in500 <- fn_valid_cohorts(df = df_sub_1in500)
bldg_1in500_fsplit <- df_sub_1in500 |> filter(eff_date_chr %in% valid_cohorts_1in500)




models <- list(

  # "Buildings in SFHA" =  feols(
  #   log(sale_price) ~ addedto_eff_sfha | pin + sale_year,
  #   fsplit = ~eff_date_chr,
  #   cluster = ~pin10,
  #   data = bldgs_fsplit
  # ),

  "Buildings in SFHA" =  feols(
    log(sale_price) ~ addedto_eff_sfha | pin + sale_year,
    cluster = ~pin10,
    data = bldg_sfha
  ),

  "Buildings - 1-in-500" =  feols(
    log(sale_price) ~ addedto_eff_sfha | pin + year,
    fsplit = ~eff_date_chr,
    cluster = ~pin10,
    data = bldg_1in500_fsplit
  ),

  "Buildings - 1-in-500" =  feols(
    log(sale_price) ~ addedto_eff_sfha | pin + sale_year,
    cluster = ~pin10,
    data = bldg_1in500
  )
)

all_models_formatted <- modelsummary::modelsummary(models,
                                                   output = "flextable",
                                                   fmt = function(x) round(x, 2),
                                                   
                                                   stars = TRUE)

all_models_formatted

	Buildings in SFHA	Buildings - 1-in-500 sample: Full sample	Buildings - 1-in-500 sample: 2019-11-01	Buildings - 1-in-500
addedto_eff_sfhaTRUE	0.07	0.05**	0.05**	0.05**
	(0.09)	(0.02)	(0.02)	(0.02)
Num.Obs.	667288	11614	11614	667288
R2	0.859	0.884	0.884	0.859
R2 Adj.	0.751	0.792	0.792	0.751
R2 Within	0.000	0.000	0.000	0.000
R2 Within Adj.	-0.000	0.000	0.000	-0.000
AIC	966832.7	4509.7	4509.7	966831.6
BIC	4276347.0	42546.0	42546.0	4276345.8
RMSE	0.32	0.19	0.19	0.32
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X	X
FE: sale_year	X			X
FE: year		X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Table 11: Mapped into EFFECTIVE SFHA - Restricted Sample

Table 11: Countywide Effective Map Dates

Code

eff_twfe = list(
 "I-Eff" =  feols(
    log(sale_price) ~ addedto_eff_sfha  | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha
  ),

    "I-Eff Subset" = feols(
    log(sale_price) ~ addedto_eff_sfha  | pin + sale_year,
    cluster = ~ pin10,
        data = (bldg_sfha |> filter(change_type != "Always SFHA" & treat_year_remove_eff == 10000)),

  ),

  "II-Eff" =  feols(
    log(sale_price) ~ removedfrom_eff_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,

  ),


  "II-Eff Subset" =  feols(
    log(sale_price) ~ removedfrom_eff_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha |> filter(change_type != "Never SFHA" & treat_year_add_prelim == 10000),

  ),
 
  # Symmetric effect of location in risk zone
    "III-Eff" = feols(
    log(sale_price) ~ in_eff_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,

  ),
  # Asymmetric: added vs removed in one equation
  "0-Eff" = feols(
    log(sale_price) ~ addedto_eff_sfha + removedfrom_eff_sfha | pin + sale_year,
    cluster = ~ pin10,
        data = bldg_sfha,

  )
 )
all_models_formatted <- modelsummary::modelsummary(eff_twfe,
               output = "flextable",
               fmt = function(x) round(x, 2),

             stars = TRUE)

all_models_formatted

	I-Eff	I-Eff Subset	II-Eff	II-Eff Subset	III-Eff	0-Eff
addedto_eff_sfhaTRUE	0.07	0.07				0.07
	(0.09)	(0.09)				(0.09)
removedfrom_eff_sfhaTRUE			0.18***	0.22***		0.18***
			(0.03)	(0.05)		(0.03)
in_eff_sfhaTRUE					-0.17***
					(0.03)
Num.Obs.	667288	652241	667288	15053	667288	667288
R2	0.859	0.858	0.859	0.895	0.859	0.859
R2 Adj.	0.751	0.749	0.751	0.813	0.751	0.751
R2 Within	0.000	0.000	0.000	0.002	0.000	0.000
R2 Within Adj.	-0.000	-0.000	0.000	0.002	0.000	0.000
AIC	966832.7	949625.6	966819.2	16137.7	966820.0	966821.1
BIC	4276347.0	4177830.0	4276333.4	66257.7	4276334.2	4276346.8
RMSE	0.32	0.32	0.32	0.27	0.32	0.32
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10
FE: pin	X	X	X	X	X	X
FE: sale_year	X	X	X	X	X	X
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

Code

mods <- list (
 no_fe <- feols(
    log_price ~ in_eff_sfha,
    vcov = ~pin10,
    data = bldg_sfha
  ),
  
 one_fe <- feols(
    log_price ~ in_eff_sfha | sale_year,
    data = bldg_sfha,
     vcov = ~pin10
  ),
pin_fe <- feols(
    log_price ~ in_eff_sfha | pin ,
    data = bldg_sfha,
        vcov = ~pin10,
  ),
  
  two_fe <- feols(
    log_price ~ in_eff_sfha | pin + sale_year,
    data = bldg_sfha,
     vcov = ~pin10
  ),
  
  three_fe <- feols(
    log_price ~ in_eff_sfha | pin + sale_year + qtr,
    data = bldg_sfha,
     vcov = ~pin10
)
)


coef_labels = c(
  "in_prelim_sfhaTRUE" = "In SFHA (Preliminary Map)",
  "in_eff_sfhaTRUE"    = "In SFHA (Effective Map)"
)

modelsummary::modelsummary(
  mods,
  stars = TRUE,
 gof_omit = "IC|Log|Adj|Within"
,  coef_map = coef_labels
)

	(1)	(2)	(3)	(4)	(5)
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001
In SFHA (Effective Map)	-0.194**	-0.190**	-0.615***	-0.169***	-0.164***
	(0.067)	(0.067)	(0.040)	(0.034)	(0.033)
Num.Obs.	668596	668596	667288	667288	667288
R2	0.001	0.076	0.796	0.859	0.860
RMSE	0.86	0.83	0.39	0.32	0.32
Std.Errors	by: pin10	by: pin10	by: pin10	by: pin10	by: pin10
FE: sale_year		X		X	X
FE: pin			X	X	X
FE: qtr					X

Table 10: Table 11

Table 12 & 13

Code

tbl <- readRDS("outputs/tables/grid_tbl_twfe.rds")
tbl

Model	Term	Bldg - SFHA		Parcel - SFHA		Bldg - 1-in-500		Parcel - 1-in-500
Model	Term	Preliminary	Effective	Preliminary	Effective	Preliminary	Effective	Preliminary	Effective
County - Symmetric	In SFHA	-0.03 (0.04)	-0.17*** (0.03)	0.04 (0.04)	-0.14*** (0.03)	-0.01 (0.02)	0.02 (0.02)	0.05 (0.04)	0.02 (0.04)
County - R	Mapped Out	0.04 (0.04)	0.18*** (0.03)	-0.05 (0.04)	0.15*** (0.03)	0.05 (0.08)	0.09 (0.06)	-0.10+ (0.06)	0.02 (0.05)
County - Add	Mapped In	0.13 (0.10)	0.07 (0.09)	0.03 (0.09)	0.00 (0.04)	0.00 (0.02)	0.05** (0.02)	0.03 (0.05)	0.06 (0.05)
Triads - Symmetric	In SFHA	-0.01 (0.03)	-0.17*** (0.03)	0.03 (0.04)	-0.14*** (0.03)	0.00 (0.02)	0.02 (0.02)	0.07 (0.05)	0.02 (0.04)
Triads - Symmetric	In SFHA × North	-0.07 (0.12)	NA	0.06 (0.09)	NA	-0.05 (0.09)	NA	-0.05 (0.07)	NA
Triads - R	Mapped Out	0.01 (0.03)	0.18*** (0.03)	-0.03 (0.04)	0.15*** (0.03)	-0.05 (0.06)	0.09 (0.06)	-0.07 (0.06)	0.02 (0.05)
Triads - R	Mapped Out × North	0.17 (0.13)	NA	-0.12 (0.12)	NA	0.23+ (0.14)	NA	-0.12 (0.17)	NA
Triads - A	Mapped In	0.03*** (0.00)	0.07 (0.09)	-0.13** (0.04)	0.00 (0.04)	-0.01 (0.02)	0.05** (0.02)	0.07 (0.08)	0.06 (0.05)
Triads - A	Mapped In × North	0.11 (0.12)	NA	0.17 (0.11)	NA	0.07 (0.06)	NA	-0.07 (0.10)	NA
* p < 0.1, p < 0.05, * p < 0.01

Table 14

Code

tbl <- readRDS("outputs/tables/grid_tbl_etwfe.rds")
tbl

Model	Term	Buildings - SFHA		Parcels - SFHA		Buildings - 1 in 500		Parcels - 1 in 500
Model	Term	Preliminary	Effective	Preliminary	Effective	Preliminary	Effective	Preliminary	Effective
County - Add	Overall ATT	-0.017*** (0.001)	0.111*** (0.033)	0.265* (0.140)	0.043*** (0.001)	0.214*** (0.039)	0.104*** (0.014)	0.110 (0.073)	0.057 (0.054)
County - R	Overall ATT	0.138** (0.061)	0.216*** (0.066)	0.102 (1333.565)	0.131** (0.060)	-0.044 (0.070)	0.036 (0.137)	-0.076 (1706.536)	0.014 (0.161)
Triads - Add	North	-0.039*** (0.002)	NA	0.214 (0.159)	NA	-0.016 (0.088)	NA	0.068 (0.098)	NA
Triads - Add	South	0.000 ( NA)	0.030 (0.033)	-0.177*** (0.007)	0.048*** (0.003)	0.112*** (0.043)	-0.014 (0.014)	-0.074 (0.110)	-0.077 (0.054)
Triads - R	North	0.107*** (0.037)	NA	0.136 (10565.428)	NA	0.061* (0.032)	NA	-0.817 (6959.621)	NA
Triads - R	South	-0.043 (0.072)	0.017 (0.068)	-0.028 (0.062)	0.022 (0.061)	-0.369*** (0.116)	-0.143 (0.137)	-0.094 (0.094)	-0.113 (0.162)
Cohorts - Add	2015	0.000 ( NA)	NA	-0.143*** (0.005)	NA	0.193*** (0.043)	NA	0.003 (0.110)	NA
Cohorts - Add	2019	NA	0.088*** (0.033)	NA	0.046*** (0.002)	NA	0.070*** (0.014)	NA	0.018 (0.054)
Cohorts - Add	2021	-0.025*** (0.001)	NA	0.294* (0.158)	NA	0.057 (0.088)	NA	0.146 (0.098)	NA
Cohorts - R	2015	0.132* (0.072)	NA	0.084 (0.062)	NA	-0.168 (0.117)	NA	0.058 (0.095)	NA
Cohorts - R	2019	NA	0.215*** (0.066)	NA	0.130** (0.060)	NA	0.040 (0.137)	NA	0.014 (0.161)
Cohorts - R	2021	0.181*** (0.032)	NA	0.309 (11627.637)	NA	0.137*** (0.030)	NA	-0.465 (7696.784)	NA
p < 0.1, p < 0.05, * p < 0.01 p < 0.1, p < 0.05, * p < 0.01

Appendix A

Table 16 & 17, and Figure 11

Code

out_dir <- "outputs/q2_marginaleffects"
prep_path <- file.path(out_dir, "bldg_sfha_q2_dissertation_prepped.rds")
if (!file.exists(prep_path)) {
  source("scripts/q2_dissertation_00_prep_data.R")
}

bldg_sfha <- readRDS(prep_path)

df_prep <- bldg_sfha |>
  mutate(
    pre_year = if ("pre_year" %in% names(df)) pre_year else as.integer(format(pre_date, "%Y")),
    eff_year = if ("eff_year" %in% names(df)) eff_year else as.integer(format(eff_date, "%Y")),
    event_time = sale_year - pre_year,
    event_time = ifelse(event_time == 0 & sale_date < pre_date, -1, event_time)
  )

# ----  bin leads/lags ----
df_prep <- df_prep |>
  mutate(
    event_time_binned = case_when(
      event_time <= -5 ~ -5L,
      event_time >=  5 ~  5L,
      TRUE ~ as.integer(event_time)
    )
  )

# ---- define sample + treatment (Added vs Never) ----
# Added to SFHA comparison!
df_es_add <- df_prep |>
  filter(prelim_sfha_category %in% c("Added to prelim SFHA", "Never SFHA")) |>
  mutate(treated = prelim_sfha_category == "Added to prelim SFHA")

# ---- (A) event-study regression ----
# base model, no flexible fixed effects
mod_es_add_none <- feols(
  log(sale_price) ~ i(event_time_binned, treated, ref = -1) | pin + sale_year,
  cluster = ~ pin10,
  data = df_es_add
)


# ----  plot ----
iplot(mod_es_add_none, xlab = "Years relative to map year", main = "Event Study: Added")

Code

# ---- pre-trend check (leads only) ----
pretrends <- broom::tidy(mod_es_add_none) |>
  filter(str_detect(term, "^event_time_binned::")) |>
  mutate(rel_time = as.integer(str_extract(term, "-?\\d+"))) |>
  filter(rel_time < 0) |>
  arrange(rel_time)

pretrends

term	estimate	std.error	statistic	p.value	rel_time
event_time_binned::-5:treated	0.1104162	0.0395717	2.790280	0.0052668	-5
event_time_binned::-4:treated	-0.5796238	0.0063581	-91.162598	0.0000000	-4
event_time_binned::-3:treated	-0.2672640	0.0038471	-69.470805	0.0000000	-3
event_time_binned::-2:treated	-0.0340017	0.0048248	-7.047211	0.0000000	-2

Code

# ----  joint test of pre-trends ----
wald(mod_es_add_none, keep = "event_time_binned::-(5|4|3|2)")

Wald test, H0: joint nullity of event_time_binned::-5:treated, event_time_binned::-4:treated, event_time_binned::-3:treated and event_time_binned::-2:treated
 stat = 2,874.2, p-value < 2.2e-16, on 4 and 368,846 DoF, VCOV: Clustered (pin10).

Removed from SFHA - Event Study

Code

# --- define sample + treatment (Removed vs Always) ----
df_es_removed <- df_prep |>
  filter(prelim_sfha_category %in% c("Removed from prelim SFHA", "Always SFHA")) |>
  mutate(treated = prelim_sfha_category == "Removed from prelim SFHA")

range(df_es_removed$sale_year, na.rm = TRUE)

[1] 2010 2025

Code

# ---- event-study regression ----
mod_es_removed_none <- feols(
  log(sale_price) ~ i(event_time_binned, treated, ref = -1) | pin + sale_year,
  cluster = ~ pin10,
  data = df_es_removed
)

# ----  plot ----
iplot(mod_es_removed_none, xlab = "Years relative to map year", main = "Event Study: Removed")

This spec is implicitly assuming North Cook, South Cook, and Chicago share a single price trend. That assumption is false. Regional growth differences get misattributed to the flood-risk information shock.

Code

# ---- pre-trend check (leads only) ----
pretrends <- broom::tidy(mod_es_removed_none) |>
  filter(str_detect(term, "^event_time_binned::")) |>
  mutate(rel_time = as.integer(str_extract(term, "-?\\d+"))) |>
  filter(rel_time < 0) |>
  arrange(rel_time)

pretrends

term	estimate	std.error	statistic	p.value	rel_time
event_time_binned::-5:treated	-0.0272036	0.1429671	-0.1902788	0.8491055	-5
event_time_binned::-4:treated	0.1548045	0.0941084	1.6449593	0.1000987	-4
event_time_binned::-3:treated	0.0670327	0.0770443	0.8700546	0.3843506	-3
event_time_binned::-2:treated	-0.0432119	0.1293224	-0.3341406	0.7383003	-2

Code

# ----  joint test of pre-trends ----
wald(mod_es_removed_none, keep = "event_time_binned::-(5|4|3|2)")

Wald test, H0: joint nullity of event_time_binned::-5:treated, event_time_binned::-4:treated, event_time_binned::-3:treated and event_time_binned::-2:treated
 stat = 1.16631, p-value = 0.323484, on 4 and 8,383 DoF, VCOV: Clustered (pin10).

Passed pre-trend test: Coefficients were not statistically significant from each other.

Figure 12: Local Average Treatment Effect

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  xvar = group_name_prelim, 
    cgroup = "never",

  gvar = treat_year_remove_prelim, 
  vcov = ~pin10, 
  data = (df_prep|> 
            # do not want treated properties that move in opposite direction. 
            filter(treat_year_add_prelim==10000 &
                     # and only want to compare to properties that were Always in the SFHA (also keeps properties that changed SFHA status), 
                     change_type_prelim != "Never SFHA") 
          )
  )
#mod2
emfx(mod2, type = "group")


 group_name_prelim treat_year_remove_prelim Estimate Std. Error    z Pr(>|z|)
              2015                     2015    0.132     0.0721 1.83   0.0669
              2021                     2021    0.181     0.0318 5.68   <0.001
    S   2.5 % 97.5 %
  3.9 -0.0092  0.273
 26.2  0.1186  0.243

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

plot(emfx(mod2), pch = c(16, 15, 18, 19))

Code

plot(emfx(mod2, type = "event"), pch = c(16, 15, 18, 19))

Code

mod2 = etwfe(
  fml = log_price ~ 1, 
  tvar = sale_year, 
  ivar = pin, 
  xvar = group_name_prelim, 
    cgroup = "never",

  gvar = treat_year_add_prelim, 
  vcov = ~pin10, 
  data = (df_prep|> 
            # do not want treated properties that move in opposite direction. 
            filter(treat_year_remove_prelim==10000 &
                     # and only want to compare to properties that were Never in the SFHA (also keeps properties that changed SFHA status), 
                     change_type_prelim != "Always SFHA") 
          )
  )
#mod2
emfx(mod2, type = "group")


 group_name_prelim treat_year_add_prelim Estimate Std. Error     z Pr(>|z|)
              2015                  2015   0.0000         NA    NA       NA
              2021                  2021  -0.0251    0.00106 -23.8   <0.001
     S   2.5 % 97.5 %
    NA      NA     NA
 411.9 -0.0272 -0.023

Term: .Dtreat
Type: response
Comparison: TRUE - FALSE

Code

plot(emfx(mod2), pch = c(16, 15, 18, 19))

Code

plot(emfx(mod2, type = "event"), pch = c(16, 15, 18, 19))

Code

# mod2 = etwfe(
#   fml = log_price ~ 1, 
#   tvar = sale_year, 
#   ivar = pin, 
#   xvar = group_name_prelim, 
#     cgroup = "notyet",
# 
#   gvar = treat_year_add_prelim, 
#   vcov = ~pin10, 
#   data = (df_prep|> 
#             # do not want treated properties that move in opposite direction. 
#             filter(treat_year_remove_prelim==10000 &
#                      # and only want to compare to properties that were Never in the SFHA (also keeps properties that changed SFHA status), 
#                      change_type_prelim != "Always SFHA") 
#           )
#   )
# #mod2
# emfx(mod2, type = "group")
# plot(emfx(mod2), pch = c(16, 15, 18, 19))
# plot(emfx(mod2, type = "event"), pch = c(16, 15, 18, 19))

Table 17: Q1 Placebo Test

Purpose: Estimate treatment effects in periods before any treatment occurs.

Why: If “effects” show up before treatment, you’ve got either model misspecification or anticipation dynamics. Either way, your core assumptions are shaky.

Code

# df_prep must have: pin, pin10, sale_date, sale_year, pre_date, prelim_sfha_category, log_price

run_placebo_did <- function(df,
                            fake_shift_years = 2,
                            treated_label = "Added to prelim SFHA",
                            unit_fe = "pin",
                            time_fe = "sale_year",
                            cluster_var = "pin10") {
  if(treated_label == "Added to prelim SFHA"){
    df <- df |> filter(prelim_sfha_category %in% c("Added to prelim SFHA", "Never SFHA"))
  }
  if(treated_label == "Removed from prelim SFHA"){
    df <- df |> filter(prelim_sfha_category %in% c("Removed from prelim SFHA", "Always SFHA"))
  }
  df_placebo <- df |>
    mutate(
      treated = prelim_sfha_category == treated_label,
      placebo_date = pre_date %m-% years(fake_shift_years),
      placebo_post = sale_date >= placebo_date
    ) |>
    # keep only observations strictly before the *real* event (so placebo is truly placebo)
    filter(sale_date < pre_date) |>
    # optional: keep only observations with a defined placebo_date
    filter(!is.na(placebo_date))

  fml <- as.formula(paste0(
    "log_price ~ treated * placebo_post | ",
    unit_fe, " + ", time_fe
  ))

  feols(
    fml,
    cluster = as.formula(paste0("~", cluster_var)),
    data = df_placebo
  )
}


m_placebo <- run_placebo_did(bldg_sfha, treated_label = "Removed from prelim SFHA",
                             fake_shift_years = 2)
etable(m_placebo)

	m_placebo
Dependent Var.:	log_price

placebo_postTRUE	-0.1085 (0.1330)
treatedTRUE x placebo_postTRUE	-0.1975 (0.1934)
Fixed-Effects:	—————-
pin	Yes
sale_year	Yes
______________________________	________________
S.E.: Clustered	by: pin10
Observations	224
R2	0.96363
Within R2	0.01679

Code

m_placebo <- run_placebo_did(bldg_sfha, treated_label = "Added to prelim SFHA",
                             fake_shift_years = 2)
etable(m_placebo)

	m_placebo
Dependent Var.:	log_price

placebo_postTRUE	0.0126 (0.0105)
treatedTRUE x placebo_postTRUE	0.3056*** (0.0127)
Fixed-Effects:	——————
pin	Yes
sale_year	Yes
______________________________	__________________
S.E.: Clustered	by: pin10
Observations	64,308
R2	0.89959
Within R2	6.12e-5

Code

run_normal_did <- function(df,
                            treated_label = "Added to prelim SFHA",
                            unit_fe = "pin",
                            time_fe = "sale_year",
                            cluster_var = "pin10") {
  if(treated_label == "Added to prelim SFHA"){
    df <- df |> filter(prelim_sfha_category %in% c("Added to prelim SFHA", "Never SFHA"))
  }
  if(treated_label == "Removed from prelim SFHA"){
    df <- df |> filter(prelim_sfha_category %in% c("Removed from prelim SFHA", "Always SFHA"))
  }

  df_did <- df |>
    mutate(
      treated = prelim_sfha_category == treated_label, #== treated_label,
      treated_date = pre_date,
      event_post = (sale_date >= treated_date  & pre_date_chr != "2005-01-01")
    ) 

  fml <- as.formula(paste0(
    "log_price ~ treated * event_post | ",
    unit_fe, " + ", time_fe
  ))

  feols(
    fml,
    cluster = as.formula(paste0("~", cluster_var)),
    data = df_did
  )
}


normal_did_add <- run_normal_did(bldg_sfha, treated_label = "Added to prelim SFHA")
etable(normal_did_add)
normal_did_remove <- run_normal_did(bldg_sfha, treated_label = "Removed from prelim SFHA")
etable(normal_did_remove)

	normal_did_add
Dependent Var.:	log_price

event_postTRUE	-0.1968*** (0.0074)
treatedTRUE x event_postTRUE	0.2844** (0.1003)
Fixed-Effects:	——————-
pin	Yes
sale_year	Yes
____________________________	___________________
S.E.: Clustered	by: pin10
Observations	652,380
R2	0.85960
Within R2	0.01033

	normal_did_rem..
Dependent Var.:	log_price

event_postTRUE	-0.0080 (0.0483)
treatedTRUE x event_postTRUE	0.0897 (0.0568)
Fixed-Effects:	—————-
pin	Yes
sale_year	Yes
____________________________	________________
S.E.: Clustered	by: pin10
Observations	14,908
R2	0.89396
Within R2	0.00023

Table 11: Same set up as placebo test, but without the placebo. Includes all years of data.