v. 0.9.0. "Frank Exchange of Views"

Changelog: version 0.9.0. (2025-08-14)

Evolution of code by language since v0.8.3

Changelog summary -- how the content of the repository evolved for each programming language (Python, R and Stata) between the previous release v0.8.3 (25 Nov 2022) and the latest release v0.9.0.

Overall, the transition to seaborn and pyfixest drove most of the Python‑side evolution, while the R side adopted fixest/marginaleffects. Stata materials remained largely stable, reflecting a focus on modernizing the Python and R components for reproducibility and ease of use. No Julia yet.

Python

Python notebooks underwent substantial refactoring and feature additions:

• Seaborn as the plotting backbone: All chapters migrated from the old plotnine library to seaborn. This included developing a custom da_theme, adding functions for time‑series plots (tsplots), and standardizing default figure sizes. The change eliminated the dependency on plotnine and simplified the plotting pipeline.
• Regression engine upgrade: Regression examples were moved from statsmodels to the pyfixest package. Code was refactored to accommodate the new API, and formulas were updated to match the textbook notation. Throughout the migration several minor bug fixes were applied, and later updates bumped pyfixest to version 0.30.2.
• New model‑interpretation tools: A LIME explainer was introduced to help interpret machine‑learning models. Other helper functions were added to improve variable importance and spline calculations.
• Environment and dependency clean‑up: The Python environment was modernized with new conda/macOS/Windows YAML files, support for Python 3.12, and removal of obsolete packages such as plotnine and shap. prophet, lime and other dependencies were updated.
• Testing and reproducibility: Scripts were added to automate environment creation and test notebooks. OSF paths/links were integrated, and default data‑loading paths were standardized. Numerous notebooks were tidied up, including fixes to bar‑plot axes and cleaning of gender/earnings data.

R

Changes in R code were more targeted but still significant:

• Adoption of fixest and marginaleffects: Examples previously using base R lm() were rewritten to use feols() from the fixest package. The marginaleffects package was added for computing marginal effects, and formulas were aligned with the notation in the textbook. Minor adjustments were made to ensure compatibility with matchit and dplyr syntax.
• SHAP support (experimental): An early experiment added SHAP value calculations for R models; although later the focus shifted back to Python, the code remains available for reference.
• General maintenance: A few bug fixes and readability improvements were made across R scripts, but no major structural changes occurred.

Stata

Stata materials saw minimal changes during this release cycle:

• Code stability: Most .do files remained unchanged. A small number of scripts were updated to improve labels or path handling; for example, the football‑manager‑success chapter received a minor update to correct a plot option.
• Consistency with new data paths: Where necessary, OSF links and standardized data paths were incorporated to ensure that Stata examples work seamlessly across operating systems.