Difference-in-Differences with Geocoded Microdata: When Distance Defines Treatment

Downloads

Each dataset is available as a labeled Stata .dta and its source file.

⇩ Download all data (ZIP)stata_codebook.do

Run stata_codebook.do in Stata once to attach long-form per-variable notes to the .dta files.

Load directly in code

Every file loads straight from GitHub (raw URLs). Swap the file name to load any dataset.

Stata

* Stata 14+ : `use` reads an https URL directly
global BASE "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/r_did_ring/data/"
use "${BASE}raw_data.dta", clear
describe
notes

Python

!pip install -q pyreadstat
import pandas as pd
BASE = "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/r_did_ring/data/"
df = pd.read_stata(BASE + "raw_data.dta")

# load every dataset at once
files = ["raw_data", "data_prepared"]
data = {f: pd.read_stata(BASE + f + ".dta") for f in files}

# pyreadstat (richest metadata) reads LOCAL files -> download first
import pyreadstat, urllib.request
urllib.request.urlretrieve(BASE + "raw_data.dta", "raw_data.dta")
df, meta = pyreadstat.read_dta("raw_data.dta")

Copy and paste this snippet in Google Colab app. https://colab.research.google.com/notebooks/empty.ipynb

R

# R : haven::read_dta auto-downloads an https URL
library(haven)
BASE <- "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/r_did_ring/data/"
df <- read_dta(paste0(BASE, "raw_data.dta"))

Overview & sources

Companion data for a hands-on R tutorial on the spatial (ring) difference-in-differences design, where the treatment is a point in space and distance to that point becomes the running variable that defines who is treated and who is control. The post reproduces and extends Linden & Rockoff's (2008) study of how a registered sex offender's arrival affects nearby home prices, comparing a parametric ring estimator (a one-line feols() regression of first-differenced log prices on a treated-ring indicator) against the data-driven nonparametric estimator of Butts (2023), which uses binsreg to trace a whole treatment-effect curve over distance. At the canonical 0.1-mile cutoff the parametric ring DiD returns −5.78 %, but moving the inner ring from 0.05 to 0.15 mile swings the estimate from −6.40 % to −4.21 % (a 52 % relative spread); the nonparametric estimator finds the effect concentrated within the first 300 feet (bin 1 at −20.6 %) and crossing zero at d ≈ 0.094 mile.

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Difference-in-Differences with Geocoded Microdata: When Distance Defines Treatment [Data set]. https://carlos-mendez.org/post/r_did_ring/

Linden, L., & Rockoff, J. E. (2008). Estimates of the Impact of Crime Risk on Property Values from Megan's Laws. American Economic Review, 98(3), 1103–1127.

BibTeX

@misc{mendez2026rdidring,
  author       = {Mendez, Carlos},
  title        = {Difference-in-Differences with Geocoded Microdata: When Distance Defines Treatment},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/r_did_ring/}},
  note         = {Data set}
}

@article{linden2008estimates,
  author  = {Linden, Leigh and Rockoff, Jonah E.},
  title   = {Estimates of the Impact of Crime Risk on Property Values from {Megan's} Laws},
  journal = {American Economic Review},
  volume  = {98}, number = {3}, pages = {1103--1127}, year = {2008}
}

Variable explorer search & filter all 55 variables

Type to filter by name or label, or use the chips to filter by type. Each row shows a mini distribution. Click a header to sort.

Cross-file variable index

Which file each variable appears in (● = present).

Construction & formulas

The ring DiD reuses the textbook 2×2 difference-in-differences logic, but the treated and control groups are defined by distance to a treatment point rather than by policy assignment. The estimand is the average treatment effect among the treated, τ = E[ΔY | treated] − E[ΔY | control].

• Treated ring (close_offender): homes within the inner radius d ≤ 0.1 mile of the offender's eventual address (Butts's 𝒟_t).
• Control ring (donut): homes between the inner cutoff and the outer boundary, 0.1 < d ≤ 0.3 mile (𝒟_c).
• Parametric estimator: feols(delta_log_price ~ close_post_move | srn_year, cluster = "neighborhood") — the coefficient on the treated×post interaction is the ring-DiD ATT, with neighborhood-clustered SEs. On Linden–Rockoff: −0.0595 log-points = −5.78 %.
• Nonparametric estimator (binsreg): partition distance into L quantile-spaced bins, fit a flat constant in each, difference post−pre, and anchor the right-most bin at zero — yielding a treatment-effect curve over distance (23 bins on the real data; bin 1 at −20.6 %).
• Sample-weighted inner-ring ATT: average the nonparametric τ̂(d) over observations inside 0.1 mile (not over bins): −0.132 log-points = −12.4 %.

Distance bookkeeping: the raw distance column is rescaled to miles inside the analysis sample (distance / 3); dist_post = distance × 10 × close_post_move; offdays = sale_date − offender_address_date (days relative to arrival); srn_year = paste(srn, sale_year) is the neighborhood×year fixed-effect stratum; log_price = log(amt_Price).

The datasets

Switch datasets with the tabs. Each shows the full variable dictionary plus a sortable statistics table with mini distributions and data coverage.

Expand all datasets expand to search (Ctrl/⌘+F) or print across all datasets

Known limitations & caveats

• Replication data. All values are the Linden & Rockoff (2008) home-sale records as cleaned and distributed in Butts's (2023) replication archive; this is real empirical data, not a simulation.
• Ring choice is part of the estimand. The parametric headline depends on the inner-ring cutoff: moving it from 0.05 to 0.15 mile shifts the ATT from −6.40 % to −4.21 % (a 52 % relative spread driven by the researcher's choice, not by sampling noise).
• Local parallel trends + no anticipation. Identification rests on inner- and outer-ring prices sharing the same counterfactual trend, and on buyers not pricing in the offender's arrival before it becomes public; neither is formally testable in this cross-section.
• distance == 0 means 'not in any offender area'. In the raw file, distance is zero for sales outside every offender's catchment; the analysis sample keeps only offender == 1 rows (within 1/3 mile), where distance is strictly positive.
• Fixed-effect singletons are dropped at estimation. The parametric models discard srn_year strata with a single observation (≈63–73 obs), so reported regression N (≈7,534–9,029) is below the 9,092-row analysis sample.