Q2 Figures and Tables

Impact of visible flood factor scores on sale price

Combined Indicator & Preliminary SFHA Status

Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index
Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index
me_highff_post_by_change_type <- avg_comparisons(
  highff_post_models[["Change Type × HighFF Post"]],
  variables = "highff_post",
  by = "change_type"
)
term contrast change_type estimate std.error conf.low conf.high p.value
highff_post TRUE - FALSE Never SFHA -0.10 0.01 -0.12 -0.08 0.00
highff_post TRUE - FALSE Always SFHA -0.12 0.04 -0.21 -0.04 0.00
highff_post TRUE - FALSE Changes SFHA 0.07 0.09 -0.10 0.23 0.44
hyp_highff_post_by_change_type <- avg_comparisons(
  highff_post_models[["Change Type × HighFF Post"]],
  variables = "highff_post",
  by = "change_type",
  hypothesis = "b1=b2"
)
estimate std.error conf.low conf.high p.value
0.03 0.04 -0.06 0.11 0.57
me_highff_post_triad_by_change_type <- avg_comparisons(
  highff_post_models[["Change Type × HighFF Post × Triad"]],
  variables = "highff_post",
  by = "change_type"
)
term contrast change_type estimate std.error conf.low conf.high p.value
highff_post TRUE - FALSE Never SFHA -0.06 0.01 -0.07 -0.04 0.00
highff_post TRUE - FALSE Always SFHA -0.04 0.01 -0.07 -0.02 0.00
highff_post TRUE - FALSE Changes SFHA 0.07 0.09 -0.10 0.23 0.44
hyp_highff_post_triad_by_change_type <- avg_comparisons(
  highff_post_models[["Change Type × HighFF Post × Triad"]],
  variables = "highff_post",
  by = "change_type",
  hypothesis = "b1=b2"
)
estimate std.error conf.low conf.high p.value
-0.01 0.01 -0.04 0.02 0.39
me_highff_post_partial_by_change_type <- avg_comparisons(
  highff_post_models[["Partially interacted"]],
  variables = "highff_post",
  by = "change_type"
)
term contrast change_type estimate std.error conf.low conf.high p.value
highff_post TRUE - FALSE Never SFHA -0.05 0.01 -0.07 -0.03 0.00
highff_post TRUE - FALSE Always SFHA -0.09 0.03 -0.14 -0.04 0.00
highff_post TRUE - FALSE Changes SFHA 0.07 0.09 -0.10 0.23 0.44
me_highff_post_grid_by_change_type_triad <- comparisons(
  highff_post_models[["Change Type × HighFF Post × Triad"]],
  variables = "highff_post",
  newdata = datagrid(
    change_type = c("Never SFHA", "Always SFHA", "Changes SFHA"),
    Triad = c("South", "North", "City")
  ),
  vcov = ~pin10
)
term contrast change_type Triad highff_post estimate std.error conf.low conf.high p.value
highff_post TRUE - FALSE Never SFHA City FALSE -0.15 0.01 -0.18 -0.13 0.00
highff_post TRUE - FALSE Never SFHA North FALSE -0.05 0.01 -0.08 -0.02 0.00
highff_post TRUE - FALSE Never SFHA South FALSE 0.15 0.01 0.12 0.17 0.00
highff_post TRUE - FALSE Always SFHA City FALSE -0.39 0.02 -0.44 -0.35 0.00
highff_post TRUE - FALSE Always SFHA North FALSE 0.02 0.02 -0.02 0.06 0.39
highff_post TRUE - FALSE Always SFHA South FALSE 0.17 0.02 0.13 0.22 0.00
highff_post TRUE - FALSE Changes SFHA City FALSE -0.23 0.09 -0.41 -0.06 0.01
highff_post TRUE - FALSE Changes SFHA North FALSE -0.13 0.09 -0.30 0.04 0.14
highff_post TRUE - FALSE Changes SFHA South FALSE 0.07 0.09 -0.10 0.23 0.44

Separated Timing and Underlying Risk: Base Models

This file now loads saved model objects, tables, and marginal effects from scripts rather than estimating the models while rendering the Quarto document. The heavy computations are created by the scripts in the scripts/ folder and saved under outputs/q2_marginaleffects/.

event == TRUE for all sales that occurred after First Street released its flood risk data. high_ff_score == TRUE for properties with a Flood Factor score of 5 or higher.

Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index
Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index

Allows for year-region specific slopes by using year-region fixed effects.

(1) (2) (3) (4)
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
eventTRUE 0.061*** 0.060*** 0.058*** 0.049***
(0.006) (0.006) (0.006) (0.005)
high_ff_scoreTRUE × eventTRUE -0.089*** -0.085*** -0.088*** -0.003
(0.011) (0.011) (0.010) (0.005)
Num.Obs. 666531 665049 666566 666488
R2 0.865 0.863 0.861 0.879
R2 Adj. 0.760 0.758 0.754 0.784
R2 Within 0.002 0.001 0.001 0.000
R2 Within Adj. 0.002 0.001 0.001 0.000
AIC 939731.9 944417.5 955808.2 870894.4
BIC 4267355.2 4246831.0 4261482.8 4224125.3
RMSE 0.32 0.32 0.32 0.30
FE: pin X X X X
FE: year^clean_name X
FE: year^Township X
FE: year^Triad X
FE: year^FIRM_PAN X

Alternate FEs - PIN + Year + Region

(1) (2) (3) (4)
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
high_ff_scoreTRUE -0.620 -0.633 -0.695 -0.623
(7050.379) (7441.052) (3761.178) (7061.393)
eventTRUE 0.058*** 0.059*** 0.058*** 0.058***
(0.006) (0.006) (0.006) (0.006)
high_ff_scoreTRUE × eventTRUE -0.106*** -0.106*** -0.106*** -0.106***
(0.011) (0.011) (0.011) (0.011)
Num.Obs. 666566 665049 666566 666566
R2 0.859 0.859 0.859 0.859
R2 Adj. 0.751 0.751 0.751 0.751
R2 Within 0.002 0.002 0.002 0.002
R2 Within Adj. 0.002 0.002 0.002 0.002
AIC 964829.9 962659.6 964577.9 965115.9
BIC 4271611.3 4259951.1 4269921.7 4273528.9
RMSE 0.32 0.32 0.32 0.32
FE: pin X X X X
FE: year X X X X
FE: Triad X
FE: clean_name X
FE: Township X
FE: FIRM_PAN X

Appendix C: Full Q2 regression coefficients

Dissertation page 77 directs readers to Appendix C for the coefficients underlying the reported marginal effects. This section exposes the saved main-model objects as well as the existing formatted summary. The models use logged sale price, property and sale-year fixed effects, and 10-digit-parcel clustered standard errors. They progressively interact post-release timing and high Flood Factor status with SFHA categories and assessment triads.

The source is scripts/q2_dissertation_scripts/q2_dissertation_03_main_marginaleffects.R. Its current text excludes changing-SFHA properties, but the saved fully interacted model still contains some Changes SFHA coefficients. The saved objects and current estimation script therefore do not describe identical model versions. These outputs are presented as saved evidence, not as a verified reproduction of dissertation Table XXVI. Reconcile the intended version before replacing results.

Existing formatted summary

The saved summary below was produced with a coefficient-omission rule for flagged terms. The unfiltered estimated-coefficient table that follows makes those terms visible for inspection.

Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index

All estimated slope coefficients

Each row below is an estimated slope coefficient from a saved main model, before subgroup marginal-effect aggregation. Absorbed fixed effects and reference categories are not individual rows. Terms removed for collinearity are listed separately; they are not zero estimates. Very large standard errors or confidence intervals are retained rather than silently hidden.

Model Term removed for collinearity
Post × HighFF None recorded in saved object
Post × HighFF × SFHA change_typeAlways SFHA
Post × HighFF × SFHA change_typeAlways SFHA:high_ff_scoreTRUE
Post × HighFF × SFHA change_typeChanges SFHA:high_ff_scoreTRUE
Post × HighFF × Triad × SFHA change_typeAlways SFHA
Post × HighFF × Triad × SFHA TriadNorth
Post × HighFF × Triad × SFHA TriadSouth
Post × HighFF × Triad × SFHA change_typeAlways SFHA:high_ff_scoreTRUE
Post × HighFF × Triad × SFHA change_typeChanges SFHA:high_ff_scoreTRUE
Post × HighFF × Triad × SFHA change_typeAlways SFHA:TriadNorth
Post × HighFF × Triad × SFHA change_typeAlways SFHA:TriadSouth
Post × HighFF × Triad × SFHA change_typeChanges SFHA:TriadSouth
Post × HighFF × Triad × SFHA high_ff_scoreTRUE:TriadNorth
Post × HighFF × Triad × SFHA change_typeAlways SFHA:high_ff_scoreTRUE:TriadNorth
Post × HighFF × Triad × SFHA change_typeAlways SFHA:high_ff_scoreTRUE:TriadSouth
Post × HighFF × Triad × SFHA change_typeChanges SFHA:high_ff_scoreTRUE:TriadSouth
Post × HighFF × Triad × SFHA change_typeChanges SFHA:eventTRUE:TriadSouth
Post × HighFF × Triad × SFHA change_typeChanges SFHA:high_ff_scoreTRUE:eventTRUE:TriadSouth
Post × HighFF × Triad TriadNorth
Post × HighFF × Triad TriadSouth
Post × HighFF × Triad high_ff_scoreTRUE:TriadNorth

From coefficients to subgroup contrasts

For a linear log-price model, let x(post, h, s, r) be the model row for post-release status, high-score status, SFHA group, and region. The post-versus-pre contrast for a subgroup is [x(1,h,s,r) - x(0,h,s,r)]' beta. The high-versus-low difference in that post/pre contrast subtracts the same expression evaluated at h = 0 from the expression at h = 1. These are different estimands: a post/pre change for high-risk properties is not automatically the high-minus-low difference-in-differences effect.

For a contrast vector c, its standard error uses the full covariance matrix, sqrt(c' V c), including covariances among interaction coefficients. The existing analysis computes contrasts with marginaleffects::avg_comparisons(). Saved estimates are in log points unless explicitly transformed; 100 * (exp(estimate) - 1) gives the corresponding percentage conversion, not estimate itself. Nonlinear transformation and averaging should not be interchanged without specifying the target estimand.

The existing marginal-effect sections display the saved contrasts. Their nearby code records each call’s variables, by, and any newdata grid. These details determine the comparison and weighting and should be retained when linking a dissertation table to its underlying model.

Main Marginal Effects

Not what I want.

me_way_event_sequential <- avg_comparisons(
  q2_main_models[["Post × HighFF × Triad × SFHA"]],
  variables = list(event = c(FALSE, TRUE))
)  # 0.0403
term contrast estimate std.error conf.low conf.high p.value
event TRUE - FALSE 0.04 0.00 0.03 0.05 0.00

What I want.

me_event_by_high_ff <- avg_comparisons(
  q2_main_models[["Post × HighFF × Triad × SFHA"]],
  variables = "event",
  by = "high_ff_score"
)
term contrast high_ff_score estimate std.error conf.low conf.high p.value
event TRUE - FALSE FALSE 0.06 0.01 0.04 0.07 0.00
event TRUE - FALSE TRUE -0.05 0.01 -0.06 -0.03 0.00
Table has no data
me_way_event_sequential_simp <- avg_comparisons(
  q2_main_models[["Post × HighFF"]],
  variables = list(event = c(FALSE, TRUE))
) # 0.0431
term contrast estimate std.error conf.low conf.high p.value
event TRUE - FALSE 0.04 0.00 0.03 0.05 0.00
me_way_event_ff_sfha <- avg_comparisons(
  q2_main_models[["Post × HighFF × Triad × SFHA"]],
  variables = "event",
  by = c("change_type", "high_ff_score")
)
term contrast change_type high_ff_score estimate std.error conf.low conf.high p.value
event TRUE - FALSE Never SFHA FALSE 0.05 0.01 0.04 0.06 0.00
event TRUE - FALSE Never SFHA TRUE -0.04 0.01 -0.06 -0.03 0.00
event TRUE - FALSE Always SFHA FALSE 0.11 0.02 0.06 0.15 0.00
event TRUE - FALSE Always SFHA TRUE -0.06 0.02 -0.09 -0.03 0.00
event TRUE - FALSE Changes SFHA FALSE 0.23 0.04 0.16 0.31 0.00
event TRUE - FALSE Changes SFHA TRUE 0.12 0.09 -0.06 0.28 0.18

The original Wald call below is retained as non-executing source pending clarification of the intended hypothesis. It passes a marginal-contrast object to a model-coefficient test. No replacement p-value is reported; see the code-lineage notes for the moderator question.

These outputs estimate whether the estimated effect of high Flood Factor exposure differs across FEMA SFHA status categories.

me_event_by_change_type <- avg_comparisons(
  q2_main_models[["Post × HighFF × Triad × SFHA"]],
  variables = list(event = c(FALSE, TRUE)),
  by = "change_type"
)
term contrast change_type estimate std.error conf.low conf.high p.value
event TRUE - FALSE Never SFHA 0.04 0.00 0.03 0.05 0.00
event TRUE - FALSE Always SFHA 0.00 0.01 -0.02 0.03 0.72
event TRUE - FALSE Changes SFHA 0.22 0.04 0.15 0.29 0.00

This is the targeted contrast from the effective-map model for properties categorized as Never SFHA, allowing high and low Flood Factor properties to be represented in the comparison.

me_never_sfha_event <- avg_comparisons(
  q2_main_models[["Post × HighFF × Triad × SFHA"]],
  variables = "event",
  newdata = datagrid(
    change_type = "Never SFHA",
    high_ff_score = c(TRUE, FALSE)
  ),
  by = NULL
)
term contrast change_type Triad high_ff_score event estimate std.error conf.low conf.high p.value
event TRUE - FALSE Never SFHA City FALSE FALSE 0.01 0.01 -0.00 0.03 0.12
event TRUE - FALSE Never SFHA City TRUE FALSE -0.11 0.01 -0.13 -0.08 0.00
Table has no data
me_all_effective_grid <- comparisons(
  q2_main_models[["Post × HighFF × SFHA"]],
  newdata = datagrid(
    event = unique,
    high_ff_score = unique,
    change_type = unique
  )
)
term contrast change_type high_ff_score event estimate std.error conf.low conf.high p.value
change_type Always SFHA - Never SFHA Never SFHA FALSE FALSE 0.00 NA NA NA NA
change_type Always SFHA - Never SFHA Always SFHA FALSE FALSE 0.00 NA NA NA NA
change_type Always SFHA - Never SFHA Changes SFHA FALSE FALSE 0.00 NA NA NA NA
change_type Always SFHA - Never SFHA Never SFHA TRUE FALSE 0.00 NA NA NA NA
change_type Always SFHA - Never SFHA Always SFHA TRUE FALSE 0.00 NA NA NA NA
change_type Always SFHA - Never SFHA Changes SFHA TRUE FALSE 0.00 NA NA NA NA
change_type Always SFHA - Never SFHA Never SFHA FALSE TRUE 0.04 0.04 -0.04 0.12 0.33
change_type Always SFHA - Never SFHA Always SFHA FALSE TRUE 0.04 0.04 -0.04 0.12 0.33
change_type Always SFHA - Never SFHA Changes SFHA FALSE TRUE 0.04 0.04 -0.04 0.12 0.33
change_type Always SFHA - Never SFHA Never SFHA TRUE TRUE -0.02 0.04 -0.11 0.06 0.58
change_type Always SFHA - Never SFHA Always SFHA TRUE TRUE -0.02 0.04 -0.11 0.06 0.58
change_type Always SFHA - Never SFHA Changes SFHA TRUE TRUE -0.02 0.04 -0.11 0.06 0.58
change_type Changes SFHA - Never SFHA Never SFHA FALSE FALSE -0.52 8385.62 -16436.03 16434.98 1.00
change_type Changes SFHA - Never SFHA Always SFHA FALSE FALSE -0.52 8385.62 -16436.03 16434.98 1.00
change_type Changes SFHA - Never SFHA Changes SFHA FALSE FALSE -0.52 8385.62 -16436.03 16434.98 1.00
change_type Changes SFHA - Never SFHA Never SFHA TRUE FALSE -0.52 8385.61 -16436.03 16434.98 1.00
change_type Changes SFHA - Never SFHA Always SFHA TRUE FALSE -0.52 8385.61 -16436.03 16434.98 1.00
change_type Changes SFHA - Never SFHA Changes SFHA TRUE FALSE -0.52 8385.61 -16436.03 16434.98 1.00
change_type Changes SFHA - Never SFHA Never SFHA FALSE TRUE -0.34 8385.62 -16435.85 16435.17 1.00
change_type Changes SFHA - Never SFHA Always SFHA FALSE TRUE -0.34 8385.62 -16435.85 16435.17 1.00
change_type Changes SFHA - Never SFHA Changes SFHA FALSE TRUE -0.34 8385.62 -16435.85 16435.17 1.00
change_type Changes SFHA - Never SFHA Never SFHA TRUE TRUE -0.36 8385.62 -16435.87 16435.16 1.00
change_type Changes SFHA - Never SFHA Always SFHA TRUE TRUE -0.36 8385.62 -16435.87 16435.16 1.00
change_type Changes SFHA - Never SFHA Changes SFHA TRUE TRUE -0.36 8385.62 -16435.87 16435.16 1.00
event TRUE - FALSE Never SFHA FALSE FALSE 0.06 0.01 0.05 0.07 0.00
event TRUE - FALSE Always SFHA FALSE FALSE 0.10 0.04 0.02 0.18 0.02
event TRUE - FALSE Changes SFHA FALSE FALSE 0.24 0.04 0.16 0.32 0.00
event TRUE - FALSE Never SFHA TRUE FALSE -0.04 0.02 -0.08 -0.01 0.02
event TRUE - FALSE Always SFHA TRUE FALSE -0.07 0.05 -0.16 0.02 0.14
event TRUE - FALSE Changes SFHA TRUE FALSE 0.12 0.09 -0.05 0.29 0.16
event TRUE - FALSE Never SFHA FALSE TRUE 0.06 0.01 0.05 0.07 0.00
event TRUE - FALSE Always SFHA FALSE TRUE 0.10 0.04 0.02 0.18 0.02
event TRUE - FALSE Changes SFHA FALSE TRUE 0.24 0.04 0.16 0.32 0.00
event TRUE - FALSE Never SFHA TRUE TRUE -0.04 0.02 -0.08 -0.01 0.02
event TRUE - FALSE Always SFHA TRUE TRUE -0.07 0.05 -0.16 0.02 0.14
event TRUE - FALSE Changes SFHA TRUE TRUE 0.12 0.09 -0.05 0.29 0.16
high_ff_score TRUE - FALSE Never SFHA FALSE FALSE 0.18 22938.20 -44957.86 44958.22 1.00
high_ff_score TRUE - FALSE Always SFHA FALSE FALSE 0.18 22938.20 -44957.86 44958.22 1.00
high_ff_score TRUE - FALSE Changes SFHA FALSE FALSE 0.18 22938.20 -44957.87 44958.22 1.00
high_ff_score TRUE - FALSE Never SFHA TRUE FALSE 0.18 22938.20 -44957.86 44958.22 1.00
high_ff_score TRUE - FALSE Always SFHA TRUE FALSE 0.18 22938.20 -44957.86 44958.22 1.00
high_ff_score TRUE - FALSE Changes SFHA TRUE FALSE 0.18 22938.20 -44957.87 44958.22 1.00
high_ff_score TRUE - FALSE Never SFHA FALSE TRUE 0.07 22938.21 -44958.00 44958.15 1.00
high_ff_score TRUE - FALSE Always SFHA FALSE TRUE 0.01 22938.22 -44958.07 44958.08 1.00
high_ff_score TRUE - FALSE Changes SFHA FALSE TRUE 0.06 22938.21 -44958.01 44958.13 1.00
high_ff_score TRUE - FALSE Never SFHA TRUE TRUE 0.07 22938.21 -44958.00 44958.15 1.00
high_ff_score TRUE - FALSE Always SFHA TRUE TRUE 0.01 22938.22 -44958.07 44958.08 1.00
high_ff_score TRUE - FALSE Changes SFHA TRUE TRUE 0.06 22938.21 -44958.01 44958.13 1.00
me_high_ff_post_by_change_type <- comparisons(
  q2_main_models[["Post × HighFF × Triad × SFHA"]],
  variables = "event",
  newdata = datagrid(
    high_ff_score = TRUE,
    change_type = c("Never SFHA", "Always SFHA", "Changes SFHA"),
    Triad = c("South", "North", "City")
  ),
  vcov = ~pin10
)
term contrast change_type Triad high_ff_score event estimate std.error conf.low conf.high p.value
event TRUE - FALSE Never SFHA City TRUE FALSE -0.11 0.01 -0.13 -0.08 0.00
event TRUE - FALSE Never SFHA North TRUE FALSE -0.00 0.02 -0.03 0.03 0.92
event TRUE - FALSE Never SFHA South TRUE FALSE 0.19 0.01 0.16 0.22 0.00
event TRUE - FALSE Always SFHA City TRUE FALSE -0.34 0.03 -0.40 -0.29 0.00
event TRUE - FALSE Always SFHA North TRUE FALSE 0.06 0.02 0.03 0.10 0.00
event TRUE - FALSE Always SFHA South TRUE FALSE 0.22 0.02 0.18 0.27 0.00
event TRUE - FALSE Changes SFHA City TRUE FALSE -0.18 0.09 -0.36 -0.01 0.04
event TRUE - FALSE Changes SFHA North TRUE FALSE -0.08 0.09 -0.26 0.09 0.36
event TRUE - FALSE Changes SFHA South TRUE FALSE 0.12 0.09 -0.06 0.28 0.18

Q2 Robustness Checks

Calculated with ETWFE Package

These models use the etwfe package. The saved outputs below are produced by scripts/q2_dissertation_02_etwfe_models.R.

This version does not include SFHA status or Triad in the equation.

Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index
Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index

Models using Preliminary FIRMs

These models are pretty crappy and not used in the dissertation document, but they do make me feel better about the models using effective FIRM status for the dissertation. (Having no results is a good thing when there shouldn’t be results, yay.)

Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index
Error in `dt_options$value[[which(dt_options$parameter == option)]]`:
! attempt to select less than one element in get1index

Diagnostics and Data Checks

observations parcels pin10s min_sale_year max_sale_year post_release_sales high_ff_sales
668596.00 291321.00 189559.00 2010.00 2025.00 233933.00 96395.00
group observations
group_C_A 490915.00
group_C_B 162730.00
group_C_C 14188.00
group_C_D 595.00
group_T_A 13.00
group_T_B 155.00
group observations
group_C_A 511511.00
group_C_B 162598.00
group_C_C 14288.00
group_C_D 595.00
group_T_A 13.00
group_T_B 155.00

Diagnostics: Dropped Terms

These tables show terms removed from the presentation tables because their standard errors were missing or larger than the threshold used in the scripts. In this fixed-effects setup, these terms are usually either not identified within PIN over time or are extremely sparse combinations.

term estimate std.error p.value
change_typeChanges SFHA -0.52 8385.62 1.00
high_ff_scoreTRUE 0.18 22938.21 1.00
change_typeChanges SFHA -0.70 6824.69 1.00
high_ff_scoreTRUE 0.14 18241.24 1.00
high_ff_scoreTRUE:TriadSouth -0.66 23358.89 1.00
term estimate std.error p.value
prelim_sfha_categoryRemoved from prelim SFHA 0.63 4990.68 1.00
high_ff_scoreTRUE -0.82 15138.30 1.00
prelim_sfha_categoryRemoved from prelim SFHA 0.64 5963.56 1.00
high_ff_scoreTRUE -2.21 14596.62 1.00
high_ff_scoreTRUE:TriadSouth 0.47 26163.32 1.00
term estimate std.error p.value
change_type_prelimChanges SFHA 0.42 5458.37 1.00
change_type_prelimChanges SFHA -1.13 16485.64 1.00
change_type_prelimChanges SFHA:TriadSouth 1.57 13553.70 1.00
term estimate std.error p.value
high_ff_scoreTRUE -0.52 15358.63 1.00
change_typeChanges SFHA -0.52 8385.62 1.00
high_ff_scoreTRUE 0.18 22938.21 1.00
change_typeChanges SFHA -0.70 6824.69 1.00
high_ff_scoreTRUE 0.14 18241.24 1.00
high_ff_scoreTRUE:TriadSouth -0.66 23358.89 1.00
high_ff_scoreTRUE -0.60 13894.31 1.00
high_ff_scoreTRUE:TriadSouth -1.18 23855.31 1.00
term estimate std.error p.value
change_typeChanges SFHA -0.56 5003.58 1.00
change_typeChanges SFHA -0.49 4986.58 1.00
change_typeNever SFHA:highff_postFALSE 0.48 5710.80 1.00
change_typeChanges SFHA:highff_postFALSE -0.44 7243.08 1.00
change_typeNever SFHA:highff_postTRUE 0.32 5710.80 1.00
change_typeChanges SFHA:highff_postTRUE -0.70 7243.08 1.00
highff_postFALSE:TriadNorth -1.21 12558.94 1.00
highff_postTRUE:TriadNorth -1.06 12558.95 1.00
highff_postFALSE:TriadSouth -1.58 30815.59 1.00
highff_postTRUE:TriadSouth -1.24 30815.60 1.00
term estimate std.error p.value
prelim_sfha_categoryAlways SFHA 0.12 9661.66 1.00
eventTRUE -0.04 7667.75 1.00
prelim_sfha_categoryAlways SFHA:eventTRUE 0.12 7667.75 1.00
prelim_sfha_categoryNever SFHA:eventTRUE 0.11 7667.75 1.00
prelim_sfha_categoryRemoved from prelim SFHA:eventTRUE 0.06 7667.75 1.00
high_ff_scoreTRUE:eventTRUE -0.11 12036.35 1.00
prelim_sfha_categoryAlways SFHA:high_ff_scoreTRUE:eventTRUE 0.12 12036.36 1.00
prelim_sfha_categoryNever SFHA:high_ff_scoreTRUE:eventTRUE 0.10 12036.35 1.00
prelim_sfha_categoryRemoved from prelim SFHA:high_ff_scoreTRUE:eventTRUE -0.40 12036.36 1.00
prelim_sfha_categoryRemoved from prelim SFHA -0.22 6614.76 1.00
high_ff_scoreTRUE -0.27 9536.53 1.00
prelim_sfha_categoryNever SFHA:high_ff_scoreTRUE -0.15 11216.34 1.00
term estimate std.error p.value
prelim_sfha_categoryAlways SFHA 0.29 8010.39 1.00
prelim_sfha_categoryRemoved from prelim SFHA -0.30 6658.25 1.00

Ordinal & Continuous FF Risk Variable

See the binary, ordinal, and continuous Flood Factor checks.

Appendix B: Q2 research-design assumptions

This section supports dissertation page 69 and Appendix B. It connects saved plots and diagnostic results to the existing slope and placebo tests below. These outputs concern Flood Factor information; the sample and specification notes are part of the interpretation, especially when comparing historical diagnostics with the main models.

Countywide and regional trends

These saved plots supply the trend visualizations referenced from dissertation page 69 and Appendix B. They were generated by scripts/q2_assumption_tests/01_create_q2_plot_objects.R from outputs/q2_models/bldg_sfha_q2_prepped.RDS. That is a separate saved input from the q2_marginaleffects dataset used by the slope tests below; equality of their samples and data vintages has not been established.

The first four figures group prices by high Flood Factor status (score at least 5), countywide and by assessment triad, in levels and logs. The next four group prices by effective SFHA status. These are descriptive group trends, not regression-adjusted effects. A change in the mix of sold properties can change a group’s mean. Read the embedded legends and time windows; do not infer a causal information effect from a post-release divergence alone.

Error in `geom_line()`:
! Problem while converting geom to grob.
ℹ Error occurred in the 1st layer.
Caused by error in `draw_panel()`:
! `geom_line()` can't have varying colour, linewidth, and/or alpha along
  the line when linetype isn't solid.

Alternative outcome transformations

The levels and log plots above are complemented by inverse-price plots for Flood Factor and SFHA groups, and inverse-hyperbolic-sine plots for SFHA groups. Countywide and triad panels are shown for each. These are existing saved outputs, not newly fitted models or a verified reconstruction of the printed figure numbering. Inverse price is decreasing in price; inverse hyperbolic sine is increasing. Different vertical scales and changing group composition matter when comparing the shapes. Visual similarity does not establish parallel trends for every transformation.

Error in `geom_line()`:
! Problem while converting geom to grob.
ℹ Error occurred in the 1st layer.
Caused by error in `draw_panel()`:
! `geom_line()` can't have varying colour, linewidth, and/or alpha along
  the line when linetype isn't solid.

Pre-event slope tests

The first tables examine linear slope differences by SFHA and high Flood Factor status in sales before 2020, with parcel-clustered standard errors but no property fixed effects. Subsequent specifications use the recorded release date and include property/year fixed effects or split the comparison by triad, as labeled. These are different specifications and pre-period definitions. A non-significant interaction does not establish equivalence of trends; inspect its magnitude and precision as well as the p-value.

Pre-event slope tests
SFHA × time HighFF × time SFHA×HighFF × time
(Intercept) -79.755*** -79.809*** -79.488***
(1.379) (1.480) (1.481)
sale_year 0.046*** 0.046*** 0.045***
(0.001) (0.001) (0.001)
in_eff_sfhaTRUE 8.126 -23.496
(11.878) (14.752)
sale_year × in_eff_sfhaTRUE -0.004 0.011
(0.006) (0.007)
high_ff_scoreTRUE 0.057 -3.149
(3.934) (3.997)
sale_year × high_ff_scoreTRUE 0.000 0.002
(0.002) (0.002)
in_eff_sfhaTRUE × high_ff_scoreTRUE 54.719**
(20.624)
sale_year × in_eff_sfhaTRUE × high_ff_scoreTRUE -0.027**
(0.010)
Num.Obs. 415163 415039 415039
R2 0.021 0.023 0.026
RMSE 0.88 0.88 0.87
Interaction terms assess slope differences; non-rejection is not proof of parallel trends.

Has TWFE:

Pre-event slope tests
SFHA × time HighFF × time SFHA×HighFF × time
prelim_sfha_categoryRemoved from prelim SFHA -0.102 -0.399
(14406.255) (14379.640)
sale_year × prelim_sfha_categoryAdded to prelim SFHA -0.008*** -0.009***
(0.001) (0.001)
sale_year × prelim_sfha_categoryAlways SFHA -0.011*** -0.004
(0.003) (0.005)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA -0.037** -0.043**
(0.012) (0.014)
sale_year × high_ff_scoreTRUE -0.010*** -0.010***
(0.001) (0.001)
sale_year × prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE -0.004
(0.006)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE 0.068+
(0.039)
Num.Obs. 324657 324536 324536
R2 0.855 0.855 0.855
R2 Within 0.000 0.001 0.001
RMSE 0.33 0.33 0.33
FE: pin X X X
FE: sale_year X X X
Interaction terms assess slope differences; non-rejection is not proof of parallel trends.

Models (by Triad)

Pre-event slope tests by Triad
North City South
(Intercept) -78.917*** -84.827*** -81.719***
(2.356) (2.226) (1.841)
sale_year 0.045*** 0.048*** 0.046***
(0.001) (0.001) (0.001)
prelim_sfha_categoryAdded to prelim SFHA 59.347 -67.465***
(275.889) (1.841)
prelim_sfha_categoryAlways SFHA -83.918*** 146.034** 11.138
(16.006) (53.143) (11.910)
prelim_sfha_categoryRemoved from prelim SFHA -828.008*** 39.442
(2.356) (35.527)
high_ff_scoreTRUE -21.903* 5.049 -2.566
(9.091) (4.545) (6.211)
sale_year × prelim_sfha_categoryAdded to prelim SFHA -0.030 0.033***
(0.137) (0.001)
sale_year × prelim_sfha_categoryAlways SFHA 0.041*** -0.072** -0.006
(0.008) (0.026) (0.006)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA 0.411*** -0.020
(0.001) (0.018)
sale_year × high_ff_scoreTRUE 0.011* -0.002 0.001
(0.004) (0.002) (0.003)
prelim_sfha_categoryAdded to prelim SFHA × high_ff_scoreTRUE 0.415
(35.639)
prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE 111.004*** -149.973** -27.589+
(25.620) (55.851) (16.736)
prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE 807.137*** -171.455*
(46.397) (70.946)
sale_year × prelim_sfha_categoryAdded to prelim SFHA × high_ff_scoreTRUE -0.002
(0.017)
sale_year × prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE -0.055*** 0.074** 0.014+
(0.013) (0.028) (0.008)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE -0.401*** 0.085*
(0.023) (0.035)
Num.Obs. 99215 230396 104739
R2 0.045 0.032 0.034
RMSE 0.73 0.90 0.77

Does not have TWFE:

Pre-event slope tests by Triad
(1)
(Intercept) -84.827***
(2.226)
sale_year 0.048***
(0.001)
prelim_sfha_categoryAdded to prelim SFHA -67.465***
(1.841)
prelim_sfha_categoryAlways SFHA 146.034**
(53.145)
prelim_sfha_categoryRemoved from prelim SFHA 39.442
(2697.185)
high_ff_scoreTRUE 5.049
(4.545)
TriadNorth 5.911+
(3.241)
TriadSouth 3.108
(2.889)
sale_year × prelim_sfha_categoryAdded to prelim SFHA 0.033***
(0.001)
sale_year × prelim_sfha_categoryAlways SFHA -0.072**
(0.026)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA 0.068
(5.412)
sale_year × high_ff_scoreTRUE -0.002
(0.002)
prelim_sfha_categoryAdded to prelim SFHA × high_ff_scoreTRUE -0.692
(0.520)
prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE -149.973**
(55.852)
prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE -171.455
(8493.346)
sale_year × TriadNorth -0.003+
(0.002)
sale_year × TriadSouth -0.002
(0.001)
prelim_sfha_categoryAdded to prelim SFHA × TriadNorth 126.813
(275.884)
prelim_sfha_categoryAlways SFHA × TriadNorth -229.952***
(55.502)
prelim_sfha_categoryRemoved from prelim SFHA × TriadNorth -867.450
(20231.662)
prelim_sfha_categoryAlways SFHA × TriadSouth -134.896*
(54.463)
high_ff_scoreTRUE × TriadNorth -26.952**
(10.164)
high_ff_scoreTRUE × TriadSouth -7.615
(7.696)
sale_year × prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE 0.074**
(0.028)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE 0.085
(4.219)
sale_year × prelim_sfha_categoryAdded to prelim SFHA × TriadNorth -0.063
(0.137)
sale_year × prelim_sfha_categoryAlways SFHA × TriadNorth 0.114***
(0.027)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA × TriadNorth 0.329
(12.003)
sale_year × prelim_sfha_categoryAlways SFHA × TriadSouth 0.067*
(0.027)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA × TriadSouth -0.109
(6.394)
sale_year × high_ff_scoreTRUE × TriadNorth 0.013**
(0.005)
sale_year × high_ff_scoreTRUE × TriadSouth 0.004
(0.004)
prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE × TriadNorth 260.976***
(61.447)
prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE × TriadNorth 978.591
(22663.419)
prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE × TriadSouth 122.383*
(58.306)
sale_year × prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE × TriadNorth -0.129***
(0.030)
sale_year × prelim_sfha_categoryRemoved from prelim SFHA × high_ff_scoreTRUE × TriadNorth -0.487
(11.244)
sale_year × prelim_sfha_categoryAlways SFHA × high_ff_scoreTRUE × TriadSouth -0.061*
(0.029)
Num.Obs. 434350
R2 -0.449
RMSE 1.06

Placebo analysis: existing 2017 cutoff

Dissertation Appendix B, Table XXV (p. 106), prints 659,942 observations for the actual event. We treat that count as a reporting error. The corresponding model with SFHA interactions uses 666,566 observations on the saved dissertation data, consistent with the sample size in the main text. Its high Flood Factor × post coefficient rounds to the printed −0.103.

Use each model’s actual estimation sample. The search for the origin of 659,942 is closed. If a future dissertation model independently yields exactly 659,942 estimation observations, flag it as a possible source of the printed count and record its specification and input file. A count match alone leaves the provenance uncertain.

The first two models restrict sales to 2019 or earlier and define post-placebo as sale_year > 2017: 2018-2019 versus earlier years. The next two show actual-release specifications using the broader prepared sample. Their time windows and some fixed effects differ, so they are reference models rather than an exact one-variable timing substitution. This makes the existing implementation explicit without silently changing the dissertation’s placebo design.

Placebo test (cut at 2017)
(1) (2) (3) (4)
high_ff_scoreTRUE -0.329 -0.784 -0.231
(7498.187) (8798.696) (10589.770)
high_ff_scoreTRUE × event_pbTRUE -0.065*** -0.060***
(0.010) (0.010)
change_typeAlways SFHA -0.705 -2.495
(9197.351) (11848.211)
change_typeChanges SFHA 0.308
(4813.836)
event_pbTRUE 0.195
(5021.943)
change_typeAlways SFHA × event_pbTRUE -0.021
(0.030)
change_typeChanges SFHA × event_pbTRUE -0.011
(0.131)
high_ff_scoreTRUE × change_typeAlways SFHA × event_pbTRUE -0.034
(0.044)
eventTRUE 0.023*** 0.022***
(0.006) (0.006)
high_ff_scoreTRUE × eventTRUE -0.106*** -0.103***
(0.011) (0.012)
change_typeAlways SFHA × eventTRUE 0.040
(0.042)
change_typeChanges SFHA × eventTRUE 0.176***
(0.037)
high_ff_scoreTRUE × change_typeAlways SFHA × eventTRUE -0.063
(0.060)
high_ff_scoreTRUE × change_typeChanges SFHA × eventTRUE -0.023
(0.096)
Num.Obs. 298674 298674 666566 666566
R2 0.856 0.856 0.860 0.860
R2 Within 0.000 0.001 0.002 0.002
RMSE 0.33 0.33 0.32 0.32
FE: pin X X X X
FE: sale_year X X X X
FE: qtr X X X

Distributional diagnostics and anticipation

The saved distributional diagnostics come from scripts/Q2_FloodFactor_didFF_test.R. That script uses the older data/processed/df_prep_buildings.rds input, applies its missing-data and parcel filters, averages observations within property-year, and assigns the treated group to 2020 for a property with a high Flood Factor score and an observed post-release sale. The last condition makes group assignment depend on observed sales as well as risk status. This historical construction must be reviewed before treating these outputs as a definitive check of the current main model.

The logged and unlogged saved objects both report p-values of 1. The displays below load those results without rerunning the exploratory script. They are not evidence that the current main-model sample passes every identifying assumption: failure to reject a distributional diagnostic does not prove mean parallel trends, absence of anticipation, or causal identification.


Logged sale price: historical Q2 diagnostic
Saved p-value: 1
NULL

Sale price: historical Q2 diagnostic
Saved p-value: 1
NULL

What the anticipation checks can establish

The pre-event slope tests and the pre-2020 placebo sample address patterns before the recorded Flood Factor release. The existing placebo code defines its post period as years greater than 2017, so 2018-2019 are post-placebo and 2017 itself remains pre-placebo. A significant placebo interaction may reflect differential trends, other events, or anticipation; it does not distinguish among them. A non-significant result cannot rule out advance information or limited power. The assumed information date and sample construction must be considered alongside these diagnostics.