factor-factory

What It Is

Domain-agnostic Python framework for panel data analysis and causal inference. Two stable contracts — the Panel (unit × period × treatment) and the Engine (a Protocol for a single estimator family) — with 20 engine families behind the contract.

Why It Matters

Research code usually grows a custom did_helpers.py per project. Those helpers drift between papers, miss diagnostics, and silently disagree when you cross-check estimators. factor-factory centralizes the econometric engine surface: the same Panel goes into TWFE, Callaway-Sant'Anna, Sun-Abraham, and Borusyak-Jaravel-Spiess with one call, reports returned as frozen dataclasses with ATT / SE / 95% CI / p / n / diagnostics.

What Ships in 1.0

• DiD — TWFE, Callaway-Sant'Anna, Sun-Abraham, Borusyak-Jaravel-Spiess; event-study plots
• RDD — rd_robust wrapping the NSF-funded Calonico-Cattaneo-Farrell-Titiunik rdrobust
• SCM — vanilla + augmented synthetic control
• Spatial — Moran's I, LISA, spatial-lag, spatial-error
• Inequality — Theil decomposition, Gini, Oaxaca-Blinder
• Changepoint — PELT, binary segmentation, sliding window
• STL — seasonal-trend-loess with significance tests
• Survival, mediation, event-study (Fama-French), DML for heterogeneous effects
• Panel contract — Panel.from_records / TreatmentEvent / Panel.validate()
• First-class jellycell integration — factor_factory.jellycell.cells / figure / notebooks / tearsheets

Ecosystem Fit

Domain-neutral core. nyc311 >= 1.0 ships a PanelDataset.to_factor_factory_panel() adapter and a 17-family as_factor_factory_estimate() dispatcher. subway-access >= 0.5 uses factor-factory for its engine-audit appendix. The blaise-website packages/python-showcase/ routes every causal claim through factor-factory.