5. Appendix¶

Appendix A¶

Variable Dictionary: Outcome and Structural Controls¶

Variable	Type	Construction
`resale_price`	Continuous	Raw transaction field (SGD)
`log_resale_price`	Continuous	`np.log(resale_price)` — dependent variable in all regressions
`floor_area_sqm`	Continuous	Raw transaction field
`flat_type`	Categorical	Ordinal encoding: 1-room=1, 2-room=2, 3-room=3, 4-room=4, 5-room=5, Executive=6, Multi-Generation=7; used as `C(flat_type)`
`storey_mid`	Continuous	`(storey_low + storey_high) / 2`
`remaining_lease_year`	Continuous	Raw field; imputed as `99 − (transaction_year − lease_commence_date)` where missing
`town`	Categorical	Raw transaction field; used as `C(town)`

School Exposure Variables¶

Variable	Type	Construction
`top_20_percent_count_0_1km`	Integer	Count of top-quintile SRS schools within 0–1 km (BallTree spatial join)
`top_20_percent_count_1_2km`	Integer	Count of top-quintile SRS schools within 1–2 km
`bottom_20_percent_count_0_1km`	Integer	Count of bottom-quintile SRS schools within 0–1 km
`bottom_20_percent_count_1_2km`	Integer	Count of bottom-quintile SRS schools within 1–2 km
`has_top20_0_1km_only`	Binary	`1` if `top_20_percent_count_0_1km ≥ 1` AND `top_20_percent_count_1_2km = 0`
`has_top20_1_2km_only`	Binary	`1` if `top_20_percent_count_0_1km = 0` AND `top_20_percent_count_1_2km ≥ 1`
`has_top20_both`	Binary	`1` if `top_20_percent_count_0_1km ≥ 1` AND `top_20_percent_count_1_2km ≥ 1`
(baseline)	—	All three binary indicators = 0; no top-20% school within 2 km

RDD Variables¶

Variable	Type	Construction
`nearest_top_20_percent_school_distance_m`	Continuous	BallTree distance to nearest top-quintile school (metres); stored as `distance_to_school_m` in `RDD.PY`
`running_m`	Continuous	`distance_to_school_m − 1000`
`inside_1km`	Binary	`1` if `running_m < 0`
`triangular_weight`	Continuous	`max(1 − \|running_m\| / bandwidth, 0)`
`matched_school_name`	String	Name of the nearest top-quintile school; used as cluster variable in RDD

Amenity Control Variables¶

Distance Variable	Count Variable	Facility
`log_nearest_mrt_distance_m`	`mrt_count_within_500m`	MRT/LRT stations
`log_nearest_supermarket_distance_m`	`supermarket_count_within_500m`	Supermarkets
`log_nearest_shopping_mall_distance_m`	`shopping_mall_count_within_500m`	Shopping malls
`log_nearest_bus_stop_distance_m`	`bus_stop_count_within_500m`	Bus stops
`log_nearest_park_distance_m`	`park_count_within_500m`	Parks
`log_nearest_hawker_distance_m`	`hawker_count_within_500m`	Hawker centres

Appendix B¶

Model Specification Strings: Estimating Equation¶

log_resale_price ~
    has_top20_0_1km_only
  + has_top20_1_2km_only
  + has_top20_both
  + floor_area_sqm
  + C(flat_type)
  + storey_mid
  + remaining_lease_year
  + C(town)
  + C(p1_cycle_year)
  + log_nearest_supermarket_distance_m + supermarket_count_within_500m
  + log_nearest_shopping_mall_distance_m + shopping_mall_count_within_500m
  + log_nearest_mrt_distance_m + mrt_count_within_500m
  + log_nearest_bus_stop_distance_m + bus_stop_count_within_500m
  + log_nearest_park_distance_m + park_count_within_500m
  + log_nearest_hawker_distance_m + hawker_count_within_500m

Estimator: OLS (statsmodels smf.ols)
Inference: cov_type='cluster', groups=df['address']

Appendix C¶

Model Specification Strings: Model Design¶

Two count-based town-slope models are estimated:

Model A:
log_resale_price ~
    0
  + C(town)
  + C(town):top_20_percent_count_0_1km
  + floor_area_sqm
  + C(flat_type)
  + storey_mid
  + remaining_lease_year
  + C(p1_cycle_year)
  + [same 12 facility controls as the main hedonic OLS]

Model B:
log_resale_price ~
    0
  + C(town)
  + C(town):top_20_percent_count_0_2km
  + floor_area_sqm
  + C(flat_type)
  + storey_mid
  + remaining_lease_year
  + C(p1_cycle_year)
  + [same 12 facility controls as the main hedonic OLS]

Estimator: OLS (statsmodels smf.ols)
Inference: cov_type='cluster', groups=df['address']

Appendix D¶

Model Specification Strings: RDD via rdrobust¶

Y = log_resale_price
X = nearest_top_20_percent_school_distance_m

Estimator: local-linear RDD via rdrobust
Kernel: triangular
Main bandwidth: CCT / MSE-optimal bandwidth selected by rdrobust
Inference: robust bias-corrected rdrobust output with nearest-neighbour variance estimation