Comparative Causal Metrics

An Introduction to Regional Impact Evaluation

Author

Carlos Mendez

Published

May 17, 2026 (in progress)

Preface

About this book

Comparative Causal Metrics is an introduction to regional impact evaluation using modern causal-inference methods implemented entirely in R and rendered with Quarto. The book is organized in two parts, each anchored by a running case study and a different family of estimators.

Part I — One region, one shock

Chapters 1–8 evaluate California’s 1989 Proposition 99 cigarette tax, the canonical single-treated-unit policy evaluation. Every method in this half builds a counterfactual for California from a different data source — the state’s own past, a neighbour, a weighted blend of donor states, or a Bayesian time-series model — and reports the average treatment effect on the treated (ATT) between 1989 and 2000.

Interrupted time series — extrapolating California’s pre-period trajectory forward.
Regression discontinuity in time — fitting a level-and-slope break at the policy date.
Basic difference-in-differences — subtracting a single control state’s pre-to-post change.
Classical synthetic control — building a weighted donor blend that matches the pre-period.
Structural Bayesian time series — state-space counterfactuals with posterior credible intervals.
Bayesian spatial synthetic control — a horseshoe-prior donor blend with spatial spillovers onto neighbours.
Synthetic control with prediction intervals — frequentist forecast-error decomposition via the scpi package.

Part II — Many regions, staggered adoption

Chapters 9–11 evaluate the Callaway-Sant’Anna minimum-wage county panel, where thousands of counties adopted different state minimum-wage levels at different years. The single-treated-unit toolkit no longer applies: the estimand becomes a family of cohort-specific ATTs indexed by adoption year, and the methods exploit the panel’s panel-data structure rather than a single time series.

Staggered difference-in-differences — group-time ATT(g, t) estimators that avoid two-way fixed-effects bias.
Matrix completion and interactive fixed effects — factor models that relax parallel trends.
Generalized synthetic control — projecting treated units onto factors estimated from never-treated controls.

Each chapter pairs the conceptual logic of the method with a worked R example using publicly available data, so readers can reproduce every estimate end-to-end.

How to read this book

The chapters are designed to be read sequentially but each is self-contained. Code is folded by default — click </> Code in any chapter to reveal the underlying R. The full source is available on GitHub, and the entire book is available as a PDF download in the navbar.

For readers who want to render a single chapter without cloning the repo, every numbered chapter offers a self-contained .zip bundle (chapter source, dataset, and a minimal _quarto.yml) at the bottom of the same </> Code dropdown. Unzip and run quarto render <chapter>.qmd — no renv setup required.

Acknowledgments

This book builds on the infrastructure of the companion project Mastering Causal Metrics and on a long tradition of applied econometric scholarship.

# Preface {.unnumbered} ## About this book *Comparative Causal Metrics* is an introduction to **regional impact evaluation** using modern causal-inference methods implemented entirely in **R** and rendered with **Quarto**. The book is organized in two parts, each anchored by a running case study and a different family of estimators. ### Part I — One region, one shock Chapters 1–8 evaluate **California's 1989 Proposition 99 cigarette tax**, the canonical single-treated-unit policy evaluation. Every method in this half builds a counterfactual for California from a different data source — the state's own past, a neighbour, a weighted blend of donor states, or a Bayesian time-series model — and reports the average treatment effect on the treated (ATT) between 1989 and 2000. - **Interrupted time series** — extrapolating California's pre-period trajectory forward. - **Regression discontinuity in time** — fitting a level-and-slope break at the policy date. - **Basic difference-in-differences** — subtracting a single control state's pre-to-post change. - **Classical synthetic control** — building a weighted donor blend that matches the pre-period. - **Structural Bayesian time series** — state-space counterfactuals with posterior credible intervals. - **Bayesian spatial synthetic control** — a horseshoe-prior donor blend with spatial spillovers onto neighbours. - **Synthetic control with prediction intervals** — frequentist forecast-error decomposition via the `scpi` package. ### Part II — Many regions, staggered adoption Chapters 9–11 evaluate the **Callaway-Sant'Anna minimum-wage county panel**, where thousands of counties adopted different state minimum-wage levels at different years. The single-treated-unit toolkit no longer applies: the estimand becomes a family of cohort-specific ATTs indexed by adoption year, and the methods exploit the panel's panel-data structure rather than a single time series. - **Staggered difference-in-differences** — group-time ATT(g, t) estimators that avoid two-way fixed-effects bias. - **Matrix completion and interactive fixed effects** — factor models that relax parallel trends. - **Generalized synthetic control** — projecting treated units onto factors estimated from never-treated controls. Each chapter pairs the conceptual logic of the method with a worked R example using publicly available data, so readers can reproduce every estimate end-to-end. ## How to read this book The chapters are designed to be read sequentially but each is self-contained. Code is folded by default — click **`</> Code`** in any chapter to reveal the underlying R. The full source is available on [GitHub](https://github.com/cmg777/ccm), and the entire book is available as a PDF download in the navbar. For readers who want to render a single chapter without cloning the repo, every numbered chapter offers a self-contained `.zip` bundle (chapter source, dataset, and a minimal `_quarto.yml`) at the bottom of the same **`</> Code`** dropdown. Unzip and run `quarto render <chapter>.qmd` — no `renv` setup required. ## Acknowledgments This book builds on the infrastructure of the companion project [*Mastering Causal Metrics*](https://github.com/cmg777/intro2causal) and on a long tradition of applied econometric scholarship.