Bayesian Spatial Synthetic Control in Python: California's Proposition 99

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/python_sc_bayes_spatial/data/"
use "${BASE}python_sc_bayes_spatial_source_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/python_sc_bayes_spatial/data/"
df = pd.read_stata(BASE + "python_sc_bayes_spatial_source_data.dta")

# load every dataset at once
files = ["python_sc_bayes_spatial_source_data", "python_sc_bayes_spatial_spatial_edges", "python_sc_bayes_spatial_spatial_summary"]
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 + "python_sc_bayes_spatial_source_data.dta", "python_sc_bayes_spatial_source_data.dta")
df, meta = pyreadstat.read_dta("python_sc_bayes_spatial_source_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/python_sc_bayes_spatial/data/"
df <- read_dta(paste0(BASE, "python_sc_bayes_spatial_source_data.dta"))

Overview & sources

Companion data for a Python tutorial that fits three nested synthetic-control estimators to California's Proposition 99: the classical simplex of Abadie, Diamond & Hainmueller (2010), a Bayesian horseshoe prior on unconstrained weights, and the Bayesian spatial (SAR) model of Sakaguchi & Tagawa (2026). The outcome panel is the same real tobacco data used throughout the synthetic-control literature: per-capita cigarette sales (cigsale) and real retail price (retprice) for 39 US states over 1970–2000, as distributed in the scspill package. California is the one treated unit; the other 38 states form the donor pool.

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Bayesian Spatial Synthetic Control in Python: California's Proposition 99 with scspill and mlsynth [Data set]. https://carlos-mendez.org/post/python_sc_bayes_spatial/

Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for comparative case studies: Estimating the effect of California's tobacco control program. Journal of the American Statistical Association, 105(490), 493–505. https://doi.org/10.1198/jasa.2009.ap08746
Sakaguchi, S., & Tagawa, H. (2026). Identification and Bayesian Inference for Synthetic Control Methods with Spillover Effects. The Econometrics Journal. https://doi.org/10.1093/ectj/utag006

BibTeX

@misc{mendez2026pythonscbayesspatial,
  author       = {Mendez, Carlos},
  title        = {Bayesian Spatial Synthetic Control in Python: California's Proposition 99 with scspill and mlsynth},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/python_sc_bayes_spatial/}},
  note         = {Data set}
}

@article{abadie2010synthetic,
  author  = {Abadie, Alberto and Diamond, Alexis and Hainmueller, Jens},
  title   = {Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California's Tobacco Control Program},
  journal = {Journal of the American Statistical Association},
  volume  = {105}, number = {490}, pages = {493--505}, year = {2010},
  doi     = {10.1198/jasa.2009.ap08746}
}
@article{sakaguchi2026spillover,
  author  = {Sakaguchi, Shosei and Tagawa, Hisahiro},
  title   = {Identification and {Bayesian} Inference for Synthetic Control Methods with Spillover Effects},
  journal = {The Econometrics Journal},
  year    = {2026},
  doi     = {10.1093/ectj/utag006}
}

Variable explorer search & filter all 13 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

Three nested estimators are fitted to these files. Each keeps everything the previous one assumed except a single restriction:

• Stage 1 — classical SCM. α̂ = argmin ‖Y₁,pre − Y_c,pre · α‖² subject to α_j ≥ 0 and Σ α_j = 1 — the simplex. Fitted with mlsynth.VanillaSC.
• Stage 2 — Bayesian horseshoe SCM. α_j | λ_j ~ N(0, λ_j²) with λ_j | τ ~ C⁺(0, τ) — the hard constraint replaced by a shrinkage prior that prefers zero without forbidding anything. Fitted with mlsynth.BSCM, and available for free as the ρ = 0 case of Stage 3.
• Stage 3 — Bayesian spatial SCM. Y_c,t = ρ (w · Y₁,t + W · Y_c,t) + X_t β + u_t — a spatial autoregressive layer on the donor outcomes, using the two files documented here. Dropping SUTVA makes a second estimand estimable: the spillover ξ_c,t = Y_c,t − Y_c,t(0) received by each donor. Fitted with scspill.SCSPILL(method="sar").

The treated dummy is 1 if state == "California" and year ≥ 1988 else 0 — the replication package's convention. Everything else in these files is observed data or a deterministic function of the TIGER/Line boundaries.

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

• Narrow predictor set. The shipped panel carries only cigsale and retprice — not the log income, youth-share or beer-sales predictors Abadie et al. (2010) matched on. This is the main reason the classical ATT here (≈ −18.4) is smaller in magnitude than the ≈ −27 usually quoted from the original paper.
• Treatment timing is 1988, not 1989. Proposition 99 passed in November 1988 and the tax took effect on 1 January 1989. Abadie et al. and mlsynth's own example treat 1989 as the first treated year; the scspill replication convention used here codes 1988. Neither is wrong, but they cannot be mixed — comparisons against other sources should check which convention that source used.
• Contiguity is a modelling choice, not a measurement. The edge list is rook contiguity from a 2024 shapefile applied to a 1970–2000 panel. Borders did not move, but 'neighbour' could reasonably have meant trade intensity, commuting flows or inverse distance instead — and the estimated spillover intensity would differ. The post's section 17 discusses when each is appropriate.
• Eleven states are absent. Alaska, Arizona, Florida, Hawaii, Maryland, Massachusetts, Michigan, New Jersey, New York, Oregon and Washington are excluded from the donor pool for having run their own large tobacco-control programmes. Two of them — Oregon and Arizona — border California. Their exclusion is why Nevada is California's only contiguous donor, and therefore why the spatial parameter is identified off a single channel.
• The weights are unnormalised here. Both files store raw 0/1 contiguity. Row-normalisation happens inside the estimator, not in the data.