Difference-in-Differences for Regional Data

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_did2/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_did2/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_did2/data/"
df <- read_dta(paste0(BASE, "raw_data.dta"))

Overview & sources

Companion data for a hands-on R tutorial that asks whether the Affordable Care Act's staggered Medicaid expansion reduced adult mortality, and uses the question to show how population weighting changes the target parameter when the units (U.S. counties) differ in size by orders of magnitude. Following Baker, Callaway, Cunningham, Goodman-Bacon and Sant'Anna's (2025) Practitioner's Guide, the post runs an eight-stage DiD pipeline — 2×2 cell means, three equivalent TWFE specifications, covariate-adjusted OR/IPW/DRDID, a 2×T event study, the Callaway–Sant'Anna staggered ATT(g,t) design, and a Rambachan–Roth HonestDiD sensitivity analysis — computing every estimate both unweighted and weighted by 2013 adult population. The headline 2×2 ATT(2014) flips sign with weighting, from +0.122 deaths per 100,000 unweighted to −2.563 weighted, while no 95% confidence interval at any stage comfortably excludes zero. The two estimands answer different questions — the typical treated county versus the typical treated adult.

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Difference-in-Differences for Regional Data: Did Medicaid Expansion Reduce Mortality? [Data set]. https://carlos-mendez.org/post/r_did2/

Baker, A., Callaway, B., Cunningham, S., Goodman-Bacon, A., & Sant'Anna, P. H. C. (2025). Difference-in-Differences Designs: A Practitioner's Guide. arXiv:2503.13323. Callaway, B., & Sant'Anna, P. H. C. (2021). Difference-in-Differences with multiple time periods. Journal of Econometrics, 225(2), 200-230. Sant'Anna, P. H. C., & Zhao, J. (2020). Doubly robust difference-in-differences estimators. Journal of Econometrics, 219(1), 101-122. Rambachan, A., & Roth, J. (2023). A More Credible Approach to Parallel Trends. Review of Economic Studies, 90(5), 2555-2591.

BibTeX

@misc{mendez2026rdid2,
  author       = {Mendez, Carlos},
  title        = {Difference-in-Differences for Regional Data: Did Medicaid Expansion Reduce Mortality?},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/r_did2/}},
  note         = {Data set}
}

@article{baker2025did,
  author  = {Baker, Andrew and Callaway, Brantly and Cunningham, Scott and Goodman-Bacon, Andrew and Sant'Anna, Pedro H. C.},
  title   = {Difference-in-Differences Designs: A Practitioner's Guide},
  journal = {arXiv preprint arXiv:2503.13323}, year = {2025}
}
@article{callaway2021did,
  author  = {Callaway, Brantly and Sant'Anna, Pedro H. C.},
  title   = {Difference-in-Differences with multiple time periods},
  journal = {Journal of Econometrics}, volume = {225}, number = {2},
  pages   = {200--230}, year = {2021}
}
@article{santanna2020drdid,
  author  = {Sant'Anna, Pedro H. C. and Zhao, Jun},
  title   = {Doubly robust difference-in-differences estimators},
  journal = {Journal of Econometrics}, volume = {219}, number = {1},
  pages   = {101--122}, year = {2020}
}
@article{rambachan2023honest,
  author  = {Rambachan, Ashesh and Roth, Jonathan},
  title   = {A More Credible Approach to Parallel Trends},
  journal = {Review of Economic Studies}, volume = {90}, number = {5},
  pages   = {2555--2591}, year = {2023}
}

Variable explorer search & filter all 30 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 outcome is the CDC crude mortality rate crude_rate_20_64 (deaths per 100,000 adults aged 20–64). Every estimate is computed twice: unweighted (each county counts equally — the ATT for the typical treated county) and population-weighted by the fixed 2013 adult population set_wt (each adult counts equally — the ATT for the typical treated adult).

• 2×2 cell-means DiD (ATT(2014)): (Ȳ_T,post − Ȳ_T,pre) − (Ȳ_C,post − Ȳ_C,pre) — the treated group's 2013→2014 change minus the control group's change.
• TWFE 2×2: Y_it = β₀ + β₁·1{D=1} + β₂·1{t=2014} + β^(2×2)·(D×Post) + ε_it; on a balanced 2×2 panel the Levels, two-way-FE, and long-difference forms recover the same β^(2×2).
• Normalized difference (covariate balance): (X̄_T − X̄_C) / √[(S²_T + S²_C)/2]; |value| > 0.25 flags imbalance (Imbens & Rubin 2015).
• Doubly-robust DiD (DRDID): (1/n) Σ (ŵ_{D=1} − ŵ_{D=0})(ΔY_i − μ̂_{Δ,D=0}(X_i)) — consistent if either the outcome model or the propensity model is correct (Sant'Anna & Zhao 2020).
• Group-time ATT (Callaway–Sant'Anna): ATT(g,t) = E[Y_it(g) − Y_it(∞) | G_i = g], aggregated by cohort and by event time.
• HonestDiD relative magnitudes: bound the post-period parallel-trends violation at a multiple M̄ of the worst observed pre-period violation; the breakdown value is the smallest M̄ that overturns the sign.

Constructed columns in data_prepared (built by the post's R script from raw_data): perc_white = population_20_64_white / population_20_64 · 100 (and likewise perc_hispanic, perc_female); unemp_rate rescaled to percent (×100); median_income rescaled to thousands of US$ (÷1000); set_wt = each county's 2013 population_20_64, held constant across years; treat_year = yaca if it falls in 2014–2019, else 0 (the did never-treated convention); Treat_2014 = 1 if yaca == 2014; Post = 1 if year ≥ 2014.

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

• Pedagogical, not definitive. The authors of the replicated guide flag this case as illustrative: "The results are pedagogical in spirit and do not represent the best possible estimates of Medicaid's effect on adult mortality."
• Underpowered. None of the six 2x2 covariate-adjusted 95% confidence intervals excludes zero; the data cannot settle the policy question.
• Weighting changes the estimand. Unweighted (ATT for the typical treated county) and population-weighted (ATT for the typical treated adult) answer different causal questions; the 2x2 ATT(2014) flips sign between them (+0.122 vs -2.563).
• Crude, not age-adjusted. The outcome is the CDC crude death rate for ages 20-64, not an age-adjusted rate, so compositional differences across cohorts are not removed.
• Small cohorts are noisy. The 2015, 2016, and 2019 expansion cohorts are small (171 / 93 / 140 counties); cohort-specific estimates carry wide confidence intervals.
• Tutorial bootstrap. The Callaway-Sant'Anna bootstrap uses 2,000 iterations for speed (the reference scripts use 25,000), affecting the third significant figure of each CI.