Six Ways to Evaluate a Policy: 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/r_causalpolicy_workshop/data/"
use "${BASE}proposition99.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_causalpolicy_workshop/data/"
df = pd.read_stata(BASE + "proposition99.dta")

# load every dataset at once
files = ["proposition99", "data_california", "data_imputed"]
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 + "proposition99.dta", "proposition99.dta")
df, meta = pyreadstat.read_dta("proposition99.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_causalpolicy_workshop/data/"
df <- read_dta(paste0(BASE, "proposition99.dta"))

Overview & sources

Companion data for a hands-on R tutorial that evaluates California's 1989 Proposition 99 cigarette tax by applying six estimator families — naive pre-post, difference-in-differences, two interrupted time-series variants (linear growth curve and an AICc-selected ARIMA), regression discontinuity on time, synthetic control (tidysynth), and Bayesian structural time series (CausalImpact) — to one shared panel and asking how much they disagree. The data are a balanced annual panel of 39 U.S. states over 1970–2000 (1,209 state-year rows) with per-capita cigarette pack sales as the outcome and four covariates, prepared from the canonical Abadie, Diamond & Hainmueller (2010) Proposition 99 dataset distributed by the causalpolicy.nl workshop. Every causal method targets the average treatment effect on the treated (ATT) for California; the entire pipeline is open and reproducible.

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Six Ways to Evaluate a Policy using R: Comparative Case Studies of Proposition 99 [Data set]. https://carlos-mendez.org/post/r_causalpolicy_workshop/

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.

BibTeX

@misc{mendez2026rcausalpolicyworkshop,
  author       = {Mendez, Carlos},
  title        = {Six Ways to Evaluate a Policy using R: Comparative Case Studies of Proposition 99},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/r_causalpolicy_workshop/}},
  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}
}

Variable explorer search & filter all 8 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

Every method targets the average treatment effect on the treated (ATT) for California over 1989–2000, each building the missing counterfactual Ŷ₁ₜ(0) — California's sales without Proposition 99 — a different way.

• Naive pre-post: Ŷ₁ₜ(0) = Ȳ₁,pre — California's own pre-period mean.
• Difference-in-Differences: Ŷ₁ₜ(0) = Ȳ₁,pre + (Ȳ₀,post − Ȳ₀,pre) — add Nevada's pre-to-post change.
• ITS (growth curve): Ŷ₁ₜ(0) = α̂ + β̂·t — extrapolate California's pre-period linear fit.
• ITS (ARIMA): Ŷ₁ₜ(0) = forecast from an AICc-selected ARIMA(1,2,0) fitted on 1970–1988.
• RDD on time: segmented regression cigsale ~ year0 + prepost + year0:prepost; the level break prepost coefficient is the headline effect.
• Synthetic Control: Ŷ₁ₜ(0) = Σ wᵢ* Yᵢₜ — convex weighted blend of donor states minimising pre-1988 RMSE.
• CausalImpact (BSTS): y₁ₜ = μₜ + βᵀxₜ + εₜ — a local-level trend plus a regression on donor-state series and covariates, fit on the pre-period and projected forward.

Covariate imputation (for data_imputed): missing lnincome, beer and age15to24 values are filled with one round of random-forest multiple imputation, mice(m = 1, method = "rf"), under a global set.seed(42). Only the three NA-bearing covariates change; the fully-observed cigsale and retprice are byte-identical to the source panel.

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

• Single imputation. data_imputed uses one random-forest draw (m = 1) for tutorial speed; with multiple imputation (m > 1) or a different model the CausalImpact estimate can move by 1–3 packs.
• Covariate missingness is heavy. In the source panel beer is missing 54.8% of rows, age15to24 32.3%, and lnincome 16.1%; cigsale and retprice are fully observed. The imputed values are model-based fills, not measurements.
• RDD is segmented regression. The workshop labels the year-as-running-variable specification 'RDD'; with calendar time as the running variable it reduces to a piecewise (segmented) regression, not the classical sharp RDD on a continuous score.
• Counterfactual, not data, drives the estimate. Five of six methods converge on a 13–20 pack reduction while DiD-vs-Nevada collapses to −5.7 packs (p = 0.31) and ARIMA-based ITS flips to +4.5 packs — the disagreement is the lesson, so triangulate across estimators.